A Treatise On Many-Valued Logics

See
discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/259645593
A Treatise on Many-Valued Logics

Book January 2001
CITATIONS
READS
398
91
1 author:
Siegfried Gottwald
University of Leipzig
180 PUBLICATIONS 2,218 CITATIONS
SEE PROFILE
Available from: Siegfried Gottwald

Retrieved on: 30 July 2016
Siegfried Gottwald
A Treatise
on
Many { Valued Logics
December 21, 2000
Preface
In recent years there has been a growing interest in many-valued logic, which
to a large extent is based on applications, intended as well as already realized
ones. These applications range from the eld of computer science, e.g. in
the areas of automated theorem proving, approximate reasoning, multi-agent
systems, switching theory, and program verication, through the eld of pure
mathematics, e.g. in independence or consistency proofs, in generalized set
theories, or in the theory of particular algebraic structures, into the eld of
humanities, e.g. linguistics and philosophy.
The present book rst covers the theoretical foundations from the basic
notions up to most recent results. And it also covers a broad class of such
applications, some of them explicitly in Part IV, other ones more implicitly
within the more theoretical parts. So, e.g., the approaches from Chapter 7 on
proof theoric matters are directly linked to actual investigations into systems
of automated theorem proving. Of course, completeness was neither intended,
nor is it possible.
This book presupposes some acquaintance with elementary notions from
classical logic and set theory. Furthermore some knowledge about algebraic
structures, particularly lattices, is needed for understanding some of the material. In general, all the essential details of the basic material are explained.
Hence an advanced undergraduate level should suce to grasp the main
points and the basic facts. Of course, some of the more advanced, in particular
the applicational, topics need additional knowledge for their understanding.
Nevertheless, the text always intends to make the main points clear for those
readers who have only limited knowledge of the particular facts and methods.
Therefore the book addresses a rather broad public: logicians as well as theoretical computer scientists and mathematicians, philosophers with interests
in logic and their applications, and also linguists.
Part I explains the basic notions, starting from the formalized languages
for the propositional and the rst-order cases, proceeding to the notions of
logical validity and entailment. A chapter is devoted to a concise history of
many-valued logic.
Part II starts with an extended discussion of a variety of dierent candidates for propositional connectives in the many-valued realm. Then it continues with a thorough presentation of a method of axiomatization for a large
VI
class of nitely-valued systems, including axiomatizations of the entailment

relation. As a further important topic it discusses proof theoretic matters,
mainly tableau and sequent calculi, which are of importance e.g. for applications in the computer science eld. Additionally it touches the general topics
of functional completeness and decidability, and looks separately at systems
with \multi-dimensional" truth degree sets.
Part III is devoted to the most important systems of many-valued logic,
explaining those systems which historically have been of prime importance
for the development of the eld, but presenting also most recent systems
which arose out of tendencies towards unication of older, seemingly dierent
approaches. In a separate chapter the important topic of many-valued identity
relations is discussed.
Part IV treats in an extended manner applications of many-valued logic in
as diverse elds as fuzzy logic and articial intelligence, theoretical linguistics,
logic itself, and mathematics.
The topics of this book have been the basic material for dierent courses
and seminars in logic, mathematics, and computer science. Besides the chapters which provide the basic material, there is a broad range of particular
topics which can be the core material for particular themes.
I thank all the friends and colleagues who forced me to start to write
this book. The European COST Action no. 15 on \Many-Valued Logic
for Computer Science Applications" provided extremely valuable support
through the contacts it supported within the last years. People who in uenced the present project through this action include, but are not restricted to (in alphabetical order) Matthias Baaz (Vienna), Walter Carnielli
(Campinas), Costas Drossos (Patras), Francesc Esteva (Barcelona), Lluis
Godo (Barcelona), Rainer Hahnle (Karlsruhe), Petr Hajek (Prague), Ulrich
Hohle (Wuppertal), Erich-Peter Klement (Linz), Radko Mesiar (Bratislava),
Daniele Mundici (Milan), Ewa Orlowska (Warsaw). In the initial phase of
the writing process, a sabbatical term from my home university, and a longer
stay at Indiana University, Bloomington, made possible by the hospitality of
Michael Dunn, contributed essentially to the nal outcome, and are gratefully
acknowledged. And for LATEX matters I have to thank Richard Zach (Vienna)
and Heinz Voigt (Leipzig) who provided me with some nice macros.
Last, but not least I'm grateful to my wife for all the continuing understanding and support she gave me. And my thanks go to the editor and the
publisher who included this book into their \Studies in Logic and Computation" series.
Leipzig/Limbach, Summer 2000
Siegfried Gottwald
Contents
Part I. Basic Notions

1. General Background : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 3
1.1 Classical and Many-Valued Logic . . . . . . . . . . . . . . . . . . . . . . . . .
1.2 Preliminary Notions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3
6
2. The Formalized Language and its Interpretations : : : : : : : : : 15

2.1
2.2
2.3
2.4
2.5
Propositional Syntax . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Propositional Semantics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
First-Order Syntax . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Many-Valued Predicates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
First-Order Semantics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
15
17
21
24
26
3. Logical Validity and Entailment : : : : : : : : : : : : : : : : : : : : : : : : : : 29

3.1
3.2
3.3
3.4
Designated Truth Degrees . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

The Propositional Situation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
The First-Order Situation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Elementary Model Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
29
31
38
40
4. Outline of the History of Many-Valued Logic : : : : : : : : : : : : : 55

Part II. General Theory
5. Particular Connectives and Truth Degree Sets : : : : : : : : : : : : 63
5.1
5.2
5.3
5.4
5.5
Conjunction Connectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
Negation Connectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
Disjunction Connectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
Implication Connectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
The J{Connectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
6. Axiomatizability : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 107
6.1 The Axiomatizability Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . 107

6.2 Axiomatizing Propositional Systems . . . . . . . . . . . . . . . . . . . . . . 108
6.3 Axiomatizing First-Order Systems . . . . . . . . . . . . . . . . . . . . . . . . 120
VIII
Contents
6.4 Axiomatizing the Entailment Relation . . . . . . . . . . . . . . . . . . . . 128
7. Sequent and Tableau Calculi : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 137
7.1 Tableau Calculi for Many-Valued Logic . . . . . . . . . . . . . . . . . . . . 138

7.2 Sequent Calculi for Many-Valued Logic . . . . . . . . . . . . . . . . . . . . 149
8. Some Further Topics : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 161
8.1 Functional Completeness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161

8.2 Decidability of Propositional Systems . . . . . . . . . . . . . . . . . . . . . 171
8.3 Product Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
Part III. Particular Systems of Many-Valued Logic

9. The Lukasiewicz Systems : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 179
9.1 The Propositional Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179

9.1.1 Important tautologies of the Lukasiewicz systems . . . . . 181
9.1.2 Characterizing the number of truth degrees . . . . . . . . . . 185
9.1.3 Axiomatizability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
9.1.4 Decidability of the system L1 . . . . . . . . . . . . . . . . . . . . . . 199
9.1.5 Representability of truth degree functions . . . . . . . . . . . 201
9.2 Algebraic Structures for Lukasiewicz Systems . . . . . . . . . . . . . . 214
9.2.1 MV-algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
9.2.2 MV-algebras and axiomatizations of the L-systems . . . . 234
9.2.3 Wajsberg algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
9.2.4 Lukasiewicz algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
9.3 The First-Order Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
9.3.1 Important logically valid formulas . . . . . . . . . . . . . . . . . . 250
9.3.2 Theoretical results for the L-systems . . . . . . . . . . . . . . . . 253
9.3.3 The innitely many-valued L-system . . . . . . . . . . . . . . . . 259
10. The Godel Systems : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 267

11. Product Logic : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 291
11.1 The Propositional System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291

11.2 The First-Order System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 308
12. The Post Systems : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 313
12.1 The Original Presentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313

12.2 The Present Form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 318
13. t-Norm Based Systems : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 327

Contents
IX
14. Axiomatizing t-Norm Based Logics : : : : : : : : : : : : : : : : : : : : : : : 345

14.1.1 Some particular cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345
14.1.2 A global approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 346
14.1.3 Monoidal logic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 352
14.1.4 Monoidal t-norm logic . . . . . . . . . . . . . . . . . . . . . . . . . . . . 362
14.1.5 Basic t-norm logic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 367
14.1.6 Completeness under continuous t-norms . . . . . . . . . . . . . 370
15. Some Three- and Four-Valued Systems : : : : : : : : : : : : : : : : : : : 385
15.1 Three-Valued Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 385

15.2 Four-Valued Systems. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 393
16. Systems with Graded Identity : : : : : : : : : : : : : : : : : : : : : : : : : : : : 401

16.1
16.2
16.3
16.4
Graded Identity Relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 401

Identity: the Absolute Point of View . . . . . . . . . . . . . . . . . . . . . . 403
Identity: the Liberal Point of View. . . . . . . . . . . . . . . . . . . . . . . . 406
Identity and Extent of Existence . . . . . . . . . . . . . . . . . . . . . . . . . 413
Part IV. Applications of Many-Valued Logic

17. The Problem of Applications : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 419
18. Fuzzy Sets, Vague Notions, and Many-Valued Logic : : : : : : 423
18.1 Vagueness of Notions and Fuzzy Sets. . . . . . . . . . . . . . . . . . . . . . 423

18.2 Basic Theory of Fuzzy Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 425
18.2.1 Elementary set algebraic operations. . . . . . . . . . . . . . . . . 426
18.2.2 Graded inclusion of fuzzy sets . . . . . . . . . . . . . . . . . . . . . . 428
18.2.3 Particular fuzzy sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 431
18.2.4 Generalized set algebraic operations . . . . . . . . . . . . . . . . 433
18.2.5 Fuzzy cartesian products . . . . . . . . . . . . . . . . . . . . . . . . . . 435
18.2.6 The extension principle . . . . . . . . . . . . . . . . . . . . . . . . . . . 437
18.3 Fuzzy Relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 438
18.4 The Full Image Under a Relation . . . . . . . . . . . . . . . . . . . . . . . . . 442
18.5 Special Types of Fuzzy Relations . . . . . . . . . . . . . . . . . . . . . . . . . 445
18.5.1 Fuzzy equivalence relations . . . . . . . . . . . . . . . . . . . . . . . . 446
18.5.2 Fuzzy partitions of fuzzy sets . . . . . . . . . . . . . . . . . . . . . . 448
18.5.3 Transitive hulls . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 452
18.5.4 Fuzzy ordering relations . . . . . . . . . . . . . . . . . . . . . . . . . . . 454
18.6 Graded Properties of Fuzzy Relations . . . . . . . . . . . . . . . . . . . . . 460
Contents
19. Fuzzy Logic : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 471

19.1
19.2
19.3
19.4
19.5
Many-Valued Logic with Graded Consequences . . . . . . . . . . . . . 472

The Semantic Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 473
The Syntactic Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 475
Axiomatizing Fuzzy Logic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 477
Partial Soundness of Inference Rules . . . . . . . . . . . . . . . . . . . . . . 480
19.5.1 Formalizing the problem . . . . . . . . . . . . . . . . . . . . . . . . . . 480
19.5.2 Partially sound rules in many-valued and fuzzy logics . 482
19.6 Some Theoretical Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 484
19.7 The Algebraic Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 486
20. Treating Presuppositions with Many-Valued Logic : : : : : : : : 493
20.1 The Phenomenon of Presuppositions . . . . . . . . . . . . . . . . . . . . . . 493

20.2 Three-Valued Approaches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 496
20.3 Four-Valued Approaches. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 499
21. Truth Degrees and Alethic Modalities : : : : : : : : : : : : : : : : : : : : 503
21.1 Interpreting Modal Logic as Many-Valued Logic . . . . . . . . . . . . 503

21.2 Graded Modalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 512
22. Approximating Intuitionistic and Other Logics : : : : : : : : : : : 525
22.1 Many-Valued Approaches toward Intuitionistic Logic . . . . . . . . 525

22.2 Approximating Logics by Many-Valued Logics . . . . . . . . . . . . . 527
23. Independence Proofs : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 535
23.1 The Propositional Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 535

23.2 The First-Order Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 538
24. Consistency Considerations for Set Theory : : : : : : : : : : : : : : : 557

References : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 567
Subject Index : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 595
Index of Names : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 601
Index of Symbols : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 603
Part I
Basic Notions
1. General Background
1.1 Classical and Many-Valued Logic

Classical logic is based upon two fundamental principles, the principles of
truth functionality and of bivalence. The principle of compositionality is the
assumption that the truth value of a compound sentence depends only on
the truth values of the components, which in particular means that it should
be irrelevant for the truth degree of the compound sentence whether its components share any common meaning or not. In rst-order logic the principle
of compositionality is accompanied by the principle of extensionality which
adds the assumption that each concept is completely determined by the class
of all objects which fall under this concept, i.e. by its extension. The principle
of bivalence is the assumption that there are exactly two truth values > and
? { denoting \true" and \false", respectively { and that each sentence has
exactly one of these truth values.
The idea that these basic principles hold true in each situation has always
been a point of debate in logic and philosophy. The true point of such discussions, however, cannot be to support a point of view that classical logic in
its modern form as formalized logic is not a suitable tool { to do something
like this seems complete nonsense according to the success classical logic has
shown in scientic discourse up to now. But there is a serious problem behind
this argument that relates to successful applications, and that is the question of the range of classical logic. In other words it is the problem whether
classical logic is the only, the \true" logic for scientic discourse, or whether
there are other, non-classical logics which are more than purely theoretical
speculations { and, if this is the case, what are the ranges of applicability of
such non-classical logics. This problem is strongly tied with the philosophical
problem whether the traditional, Cantorian-Platonic point of view toward
(mathematical) objects covers the whole range of (serious) possibilities, or
whether it should be enriched by approaches toward \non-standard" objects.
Having in mind the logical treatment of statements concerning future
events, or having in mind the logical treatment of questions, then it is not at
all obvious whether these topics are properly covered by the basic principles
of classical logic { or whether one should allow when treating them, e.g., for
truth value gaps, i.e. for (well-formed) \sentences" which are neither true nor
false. Or having in mind formulations given with the help of modalities like:
necessarily, 2 (3 + 4) equals 6 + 8,
necessarily, John Lennon died December 8, 1980,
then intuitively the rst one of these sentences is true, the second one, however, should be false { and hence the principle of extensionality is not satised
here because both sentences can be understood in such a way that a unary
sentence-forming operator \necessarily (it is the case that)" is applied to the
true(!) sentences \2 (3 + 4) = 6 + 8" and \John Lennon died December 8,
1980" but yielding sentences of dierent truth values in both cases.
Many-valued logic, the topic of the present book, assumes { as classical
logic does { the principle of compositionality (or: truth functionality) in its
propositional as well as its rst-order versions. And it accepts for its rstorder systems also a suitable principle of extensionality. It avoids, however,
the principle of bivalence. Therefore it deviates from classical logic in one of
the essential basic principles, and thus belongs to the eld of non-classical1 or
\philosophical" logics. The fact that the principle of bivalence is omitted does,
however, not mean that the truth values disappear at all. Like classical logic,
also many-valued logic refers to the basic understanding that each sentence
should have a truth value { with the main dierence that in many-valued
logic one allows for more than two truth values. Philosophically it is not
obvious how to intuitively interpret these (additional) truth values. To avoid
any confusion with the case of classical logic one prefers in many-valued logic
to speak of truth degrees and to use the word \truth values" only for classical
logic. The lack of an obvious \naive" ontological understanding of the truth
degrees of many-valued logic has two quite distinct aspects:
the negative one that there is no natural understanding of the (meaning of
the) truth degrees,
and the positive one that one is free to interpret the truth degrees for
each application of many-valued logic in a particular way suitable for this
application.
This situation distinguishes many-valued from classical logic. Not totally,
however, as also classical logic allows for interpretations of their set of truth
values in manners quite distinct from the standard interpretation: e.g. as
codes for the electrical conductivity of circuits in switching theory. The lack
of a standard interpretation of the truth degrees in many-valued logic therefore forces one to connect each application of many-valued logic with an
interpretation of the truth degrees. Later on, mainly in Part IV, we will meet
this situation with particular applications { even with applications which take
their truth degrees as related to (systems of) truth values of classical logic.
Such kinds of \reductions" of truth degrees to standard truth values have
also been considered e.g. by D. Scott [492, 493], A. Urquhart [551], G.
Malinowski [346] and M. Byrd [82].
1 The reader interested in an extended survey of the most important non-classical
logics may consult [178] or also [315].
1.1 Classical and Many-Valued Logic
Deleting the principle of bivalence does for many-valued logic, however,

not mean that particular systems of many-valued logic would not have a
xed number of truth degrees, it simply means that there may be systems
of many-valued logic with three, four, ve, : : : or even innitely many truth
degrees { but in each case the set W of truth degrees is supposed to be
xed in advance.2 This \freedom" one has in many-valued logic regarding
the number of truth degrees forces, on the other hand, the general theory to
become essentially independent of the particular choice of the set of truth
degrees. For the notion of logical validity which we intend to introduce for
well-formed formulas in systems of many-valued logic this means, e.g., that
this notion has to be dened without reference to the number of truth degrees
and, hence, can be used in the cases of e.g. three, twelve or innitely many
truth degrees in the same manner.
It should be noted that particular systems of many-valued logic may not
only be characterized by their number of truth degrees but also by additional
structural properties of the set of truth degrees: in the set of truth degrees
f0; 31 ; 23 ; 1g e.g. any two truth degrees are comparable in magnitude, in the
set of truth degrees f(0; 0); (0; 1); (1; 0); (1; 1)g, on the other hand, there are
incomparable truth degrees (with respect to some kind of natural ordering)
{ both sets of truth degrees, however, have four elements.
The only (tacit) assumption furthermore shall be that (i) each sentence
shall have exactly one truth degree, and that (ii) there exist always at least
two truth degrees { usually denoted 0 and 1 { which, in absence of all other
truth degrees (and perhaps of some logical constants like connectives) behave
exactly like the truth values >; ? of classical (two-valued) logic. In this sense,
our systems of many-valued logic will \extend" and \generalize" classical
logic.3
The larger sets of truth degrees in many-valued logic, as compared with
classical logic, allow for systems of many-valued logic to have a larger amount
of operations with truth degrees available. The syntactic counterpart of this
semantic situation will be that systems of many-valued logic usually have
\more" connectives than classical logic does { in the sense that e.g. on the
propositional level one often has in systems of many-valued logic dierent,
but (essentially) equally acceptable versions of the usual connectives of classical logic, like negation, conjunction, disjunction, or implication. In each of
these cases the dierences usually are realized only for truth degrees dierent from 0; 1, such that these \generalizations" coincide for degrees 0; 1 with
2 There have occasionally even been considered systems with a set of truth degrees
which may vary in some sense, cf. [365], but they have (up to now) not gained
much importance.
3 Many-valued logic like (almost) all the other non-classical logics does not intend,
however, to compete with classical logic for \superiority". Many-valued logic
simply is a generalization which, as a generalization, extends classical logic, has
its own interest as a particular system of logic, and proves to be a suitable tool
for dierent applications.
their classical counterparts. On the other hand there may also exist \essentially new" connectives which do not have classical counterparts because they
\map" the set f0; 1g of (pseudo)classical truth degrees into the set W nf0; 1g
of \new" truth degrees.
Unfortunately there exist up to now no general criteria which characterize
a connective (or its corresponding truth degree function) of a system of manyvalued logic as a many-valued negation, a many-valued conjunction etc. As
a partial solution to this problem we later on consider normal conditions as
well as standard conditions which in any case single out some properties of
some types of connectives of classical logic as characteristic ones for these
connectives. A general solution, however, is not provided by this approach
{ and it may be doubted whether such a general solution is possible by
purely logical criteria. Fortunately this is not really a serious restriction for
applications of many-valued logic, because usually in each application one has
an extralogical basis to decide which connective of a system of many-valued
logic is to play the role of which connective of classical logic.
As in logic as a whole, also in many-valued logic there is the well-known
duality between the syntactic and the semantic approaches toward particular
systems of logic. On the basis of a predetermined formalized language that
means, one either may x a system of many-valued logic by determining syntactically a logical calculus to derive some class of formulas of that language,
or one may x semantically an interpretational basis by determining a schema
for evaluating formulas of that language and for distinguishing some class of
\outstanding" formulas. For each one of these basic approaches, ideally one
should have the dual one characterizing the same class of formulas.
In the present book we shall prefer { for propositional as well as for rst
order logics { the semantic approach over the syntactic one, whenever possible
or reasonable. For us, syntactic calculi almost always shall be introduced to
formalize a predetermined semantic basis.
An important part of the following considerations also shall be discussions of applications of systems of many-valued logic. This is of particular
importance because (i) there does not exist any suitable \ontological" basis
for many-valued logic in general, and (ii) the historical roots of many-valued
logic were to be found in intended applications which, nally, did not prove
to be acceptable. And, of course do we like to convince the reader that manyvalued logic is a suitable topic inside the whole eld of logic.
1.2 Preliminary Notions

The presentation of many-valued logic in this book assumes that the reader
knows classical, i.e. two-valued (rst-order) logic PL2 to an extent to which it
is usually developed in courses for philosophers or computer scientists. Our
metalanguage, which we use to discuss systems of many-valued logic (together with their formalized languages) is English combined with a portion
of (formalized) classical logic PL2 , which usually is used in a semiformalized manner. This means that we often use in the metalanguage the symbols :; ^ ; _ ; ); ,; 8 ; 9 for negation, conjunction, disjunction, implication,
biimplication, universal quantication, and existential quantication, respectively, as usual in PL2 . Furthermore, \i" means if and only if, and =def is
denitional equality.
Formulas of a formal language are words over some predetermined alphabet, i.e. strings of basic symbols of this alphabet. For such words \" means
their (syntactic) identity as strings.
We furthermore assume that the reader has some knowledge of the basic
set theoretic notions. As usual, [; \; n denote union, intersection, and dierence of sets, respectively, ; is the empty set, and (a; b) the ordered pair of a; b
characterized by the basic property
(a; b) = (c; d) , a = c ^ b = d :
This notion is extended to the notions of triple, quadruple, . . . , n-tuple recursively via the equations
(a1 ; a2 ; a3 ) =def ((a1 ; a2 ); a3 ) ;
..
.
(a1 ; : : : ; an+1 ) =def ((a1 ; : : : ; an ); an+1 ) :
By the class term fx j H (x)g we denote the class of all objects x which have
some property H (x), and by fx 2 A j H (x)g we denote the class of all the
objects x with property H (x) which additionally belong to A.
The set with just the elements a1 ; a2 ; : : : ; an is denoted fa1 ; a2 ; : : : ; an g,
and the cartesian product A B is the set of all ordered pairs with rst
component from A and second component from B :
A B =def f(a; b) j a 2 A ^ b 2 B g :
(1.1)
n
Finally, A denotes the set of all n-tuples of elements of A and IP (A) the
powerset of A, i.e. set of all subsets of A.
In the case that each element of a set A is also an element of a set B , A
is a subset of B : A B . And the power or cardinality of a set A is denoted:
card A.
Of further particular set theoretic notions we sometimes refer to the set
A B of all functions f : A ! B from A into B , and occasionally also use the
fact that (set theoretically) such a function f : A ! B is a set of ordered
pairs (a; b) 2 A B , i.e. f A B , with (a; b) 2 f indicating that b = f (a).
Furthermore we have to use the generalizations of union [ and intersection
\ of two sets to families Ai ; i 2 I , of sets dened by
[
\
Ai = fx j 9 i 2 I (x 2 Ai )g ;
Ai = fx j 8 i 2 I (x 2 Ai )g :
i2I
i2I
T
Sometimes we also write fAi j i 2 I g for i2I Ai in case of this generalized

intersection and use a similar notation for generalized union.
Here we have used restricted quantications 8 i 2 I and 9 i 2 I , dened as
usual as
8 i 2 IH =def 8 x(i 2 I ) H ) ; 9 i 2 IH =def 9 x(i 2 I ^ H ) : (1.2)
Such restricted quantications shall be used later on also with other restrictions instead of \i 2 I " and have to be understood in a similar way.
In most applications, relations are understood as binary relations, i.e.
as sets of ordered pairs. In general, a relation is an n-ary relation for some
n 2 N , and set-theoretically understood as a set of n-tuples. Also in this book
relations are understood as binary ones if no other arity is indicated.
If R is a (binary) relation and (a; b) 2 R, this fact is expressed by saying
that R holds true for (a; b), or by writing aRb. An equivalence relation S
in a set A is such a relation S A A, which is re exive, symmetric, and
transitive, i.e. for which aSa holds true for each a 2 A, for which always with
aSb also bSa holds true, and for which always with aSb and bSc also aSc holds
true. For each equivalence relation S in A, to each a 2 A there corresponds
its equivalence class [a]S = fb 2 A j aSbg. Any two such equivalence classes
are disjoint, i.e. have ; as their intersection i they are dierent.
A relation R is a partial ordering in a set A i it is re exive and transitive,
and if from aRb and bRa it always follows that a = b. If R is a partial ordering
in A then A is partially ordered by R. To indicate that R is a partial ordering
one often writes a 6R b instead of aRb.
For any partial ordering 6R in A, elements a; b 2 A are R-comparable i
a 6R b or b 6R a holds true. As usual, in the case a 6R b the element a is
called R-smaller or equal b, and a <R b means: a 6R b and a 6= b. An element
a 2 A is an R-maximal (or: R-minimal ) element of A i b 6R a (or: a 6R b)
holds true for each b 2 A which is R-comparable with a. And a 2 A is an
upper (or: lower ) bound of B A i b 6R a (or: a 6R b) holds true for each
b 2 B . The subset B A itself is an R-chain i any two elements of B are
R-comparable.
An important result for partially ordered sets, Zorn's Lemma, states
that a partially ordered set A has an R-maximal element if each R-chain of
A has an upper bound in A. For the existence of such an R-maximal element
it even suces that each well-ordered4 R-chain of A has an upper bound in
A.
The partially ordered sets in which each subset fa; bg has a smallest upper
bound a t b as well as a greatest lower bound a u b are the lattices. The operations t; u dened in each lattice in the way just mentioned can themselves be
used to characterize lattices: a set A together with two (binary) operations
t; u is a lattice i both operations are commutative and associative, and satisfy the absorption laws a = a u (a t b) and a = a t (a u b) for all a; b 2 A.
4 An R-chain is well-ordered i each of its nonempty subsets has an R-minimal
element.
Thus, lattices are particular algebraic structures, i.e. sets equipped with operations and (possibly also) relations. One has always a 6 b , a u b = a and
a 6 b , a t b = b as connections between ordering and algebraic operations
in lattices. Such a lattice A is called distributive i for all a; b; c 2 A the equation au(btc) = (aub)t(auc) as well as5 the equation at(buc) = (atb)u(atc)
hold true. A lattice has 1 as a unit element (or: universal upper bound ) i
a 6 1 always holds true, and it has 0 as a zero element (or: universal lower
bound ) i 0 6 a always holds true.6
If a distributive lattice A with unit element 1 and zero element 0 is
equipped with a further unary operation c which satises the two complementation laws : a u ac = 0 and a t ac = 1 for all a 2 A, then it is called a
Boolean algebra.
Lattices as well as Boolean algebras are particular cases of algebraic
structures, i.e. of sets equipped with some operations and possibly also with
some relations. Such an algebraic structure is denoted
A = hA; op1; : : : ; opn; R1; : : : ; Rk i
to indicate that opj are the operations and that Rl are the relations considered
in the carrier jAj = A. The sequence h%1 ; : : : ; %n ; 1 ; : : : ; k i of all the arities
%i of the operations opi and i of the relations Ri of A often is called the
similarity type of the algebraic structure A.
For algebraic structures Ai , i = 1; : : : ; m, of the same similarity type
there is an important
generalization of the cartesian product (1.1): the direct
Q
product A = m
A
i=1 i . It is again an algebraic structure which has as its
carrier the set A1 Am of all m-tuples a = (a1 ; : : : ; am) with always
ai 2 Ai . And the operations and relations in this direct product are determined \componentwise" from the corresponding operations and relations in
the \factor structures" Ai , which for a binary operation opj e.g. means
opj (a; b) = (op1j (a1 ; b1 ); : : : ; opm
(1.3)
j (am ; bm )) ;
and which means in the case of e.g. a ternary relation Rk
Rk (a; b; c) , Rki (ai ; bi ; ci ) for all i = 1; : : : ; m :
(1.4)
This approach presupposes that all the structures Ai have the same number of operations opij as well as the same number of relations Rki , and that
furthermore \corresponding" operations (or relations) in dierent structures
have the same arity { with \corresponding" understood as having the same
lower index. In brief: it supposes that all the structures Ai have the same
similarity type.
5 It is an interesting result of lattice theory that the universal validity of one of
these two equations in the lattice A suces that the other one also holds true in
this lattice.
6 In the case that they exist, unit and zero elements of a lattice are even uniquely
determined.
10
One even considers direct products of any, not necessarily nite, family
of algebraic structures Ai , i 2 I and I any set. In this case the carrier of
the direct product becomes S
the set of all choice functions f of this family,
i.e. of all functions f : I ! i2I Ai which Q
satisfy f (i) 2 Ai for each i 2 I .
And the operations and relations in A = i2I Ai in this case are dened
\argumentwise" similar
Q to (1.3), (1.4).
For subsets B i2I Ai of direct products a particular
type of mapping
Q
is of importance: the projection mappings i : j i2I Ai j ! Ai into the
(carriers of the) factors characterized for each i 2 I by the equation
Y
i (f ) =def f (i) for all f 2 j Ai j :
i2I
By a slight abuse of language, these projection mappings i with 1 i n

shall sometimes also be used for sets of n-tuples, simply by identifying ntuples with functions over f1; 2; : : :; ng.
A function f : A ! B is called an injection i it is 1-1, i.e. i for all
a; b 2 A with a 6= b one has also f (a) 6= f (b). And such a function f : A ! B
is called an bijection i it is an injection and maps onto B , which additionally
means that for each b 2 B there is some a 2 A with b = f (a).
A 1-1 mapping ' between two algebraic structures A; B of the same
similarity type is an isomorphism i it is a bijection between the carriers
A; B , and i it satises for each pair of operations opAi , opBi and each pair of
relations RjA , RjA of A; B, with arities n; m respectively, the conditions
'(opAi (x1 ; : : : ; xn )) = opBi ('(x1 ); : : : ; '(xn )) ;
(1.5)
A
B
Rj (x1 ; : : : ; xm ) , Rj ('(x1 ); : : : ; '(xm )) ;
(1.6)
for all x1 ; : : : ; xmaxfn;mg 2 A.
Parallel to the property of being isomorphic, one has to take care of
the fact that some \kinds" of algebraic structures may be, and often are,
in some suitable sense equivalently, characterized as structures of dierent
similarity types. The notion of equivalence which is needed in this context,
called denitional equivalence, is the following one: two algebraic structures
A and B are denitionally equivalent i they have the same carrier jAj = jBj,
and i all the basic operations and relations of B can be dened in A and
their basic properties proven, as well as vice versa for the denability of A
in B.
A real function f : A ! B , i.e. a function with A; B R, is continuous at
a point a 2 A i for each sequence (xi )i0 of points xi 2 A which converges
toward a, i.e. which satises a = limi!1 xi , one also has that the values
f (xi ) converge toward f (a): limi!1 f (xi ) = f (a). And it is left continuous
at a 2 A i limi!1 f (xi ) = f (a) holds true for all sequences (xi )i0 with
a = limi!1 xi and always a > xi 2 A. Furthermore such a real function
f : A ! B is continuous, or left continuous, i it is continuous, or left
continuous, at each point a 2 A.
11
For functions f : A ! B one has to distinguish their values f (a) for a 2 A

from their sets of values over some set D A, denoted f [D] and dened as
f [D] = ff (x) j x 2 Dg.
A group G is an algebraic structure G = hG; G ; ei which satises for all
a 2 G the conditions
9 x(a G x = e) ; 9 y(y G a = e) ; a G e = a:
(1.7)
Such a group is called abelian i the group operation is commutative, i.e. i
one has additionally satised
8 x; y(x G y = y G x) :
Each such abelian group G = hG; G ; ei has a uniquely determined inverse
operation ;G 1 of the group operation G characterized by the condition
a G (b ;G 1 a) = b for all a; b 2 G :
An abelian group G is called (partially ) ordered, or a po-group, i it is additionally equipped with some (partial) ordering G and satises the monotonicity condition
a G b ) a G c G b G c for all a; b; c 2 G :
And an ordered abelian group G = hG; G ; e; G i is called o-group (or: `group ) i G is a linear ordering (or: a lattice ordering).
For an isomorphism ' between some `-groups G1 = hG1 ; 1 ; e1 ; 1i and
G2 = hG2 ; 2 ; e2 ; 2 i the general conditions (1.5), (1.6) reduce to
'(x 1 y) = '(x) 2 '(y) ; '(e1 ) = e2 ;
(1.8)
x 1 y , '(x) 2 '(y) ;
(1.9)
for all x; y 2 G1 .
Standard examples for `-groups are the o-group R = hR; +; 0; i of the
reals under addition, and the o-group R+ = hR+ ; ; 1; i of the positive real
numbers R+ = fx 2 R j x > 0g under multiplication, both with their usual
ordering. Both o-groups are isomorphic, with an isomorphism from R onto
R+ given by the mapping exp : x 7! ex.
For some considerations on `-groups, but also on other types of algebraic
structures later on, the notion of a partial isomorphism between two `-groups,
or between two algebraic structures A1 , A2 of the same similarity type,
becomes important. Such a partial isomorphism is a 1-1 mapping into A2
whose domain dom( ) is a subset of A1 , and which satises the isomorphism
conditions (1.5), (1.6) for all x; y 2 dom( ), i.e. which satisfy (1.8), (1.9) for
the case of the `-groups.
And one says that an algebraic structure A1 is locally embeddable into
another one A2 (with the same similarity type as A1 ) i for each nite
set D A1 there exists a partial isomorphism from A1 into A2 with
D dom( ).
12
The crucial result which shall become essential later on in Section 9.2 and
Chapter 11 is the following theorem for o-groups, proved e.g. in [236, 249, 252].
Theorem 1.2.1. Each o-group G = hG; ; G ; ei is locally embeddable into
R.
It is sometimes preferable to state such a result on local embeddability
of an algebraic structure A1 into A2 , or of a whole class K of algebraic
structures into some particular structure A, equivalently in another, more
\logical" form. For a whole class K of algebraic structures and some particular
structure A then one has to refer to the (common) language of the structures
from K = K [fAg. This is a rst-order language LK (for classical rst-order
logic) which has a (n-ary) operation symbol fk for each (n-ary) operation
opA of any structure A 2 K , and a (n-ary) relation symbol Rj for each (nary) relation RjA of any structure A 2 K . And one has to refer to universal
sentences, i.e. to sentences of the form G 8 x1 ; : : : ; xn H (x1 ; : : : ; xn ) with a
quantier-free matrix H (x1 ; : : : ; xn ).
Theorem 1.2.2. The following facts are equivalent for each class K of algebraic structures of the same similarity type (with only nitely many operations
and relations), and for each particular structure A 2 K from this class:
(i) Each algebraic structure A 2 K of the class K is locally embeddable
into the structure A.
(ii) If a universal sentence of the (common) language LK of the class K
of algebraic structures is valid in the particular structure A 2 K, then it is
also valid in each algebraic structure A 2 K.
Proof: To prove this equivalence assume rst (i), i.e. that for each algebraic structure A 2 K with carrier jAj = A and each nite set D A there
exists a partial isomorphism from A into A with D dom( ). Suppose
that the universal sentence G 8 x1 ; : : : ; xn H (x1 ; : : : ; xn ) is not valid in all
algebraic structures A 2 K. Let A1 2 K be an algebraic structure in which G
is not valid. Then there exists an A1 -assignment such that H (x1 ; : : : ; xn )
is false under . Let D = f (x1 ); : : : ; (xn )g and be a partial isomorphism
from A1 into the carrier jAj with D dom( ). Then any jAj-assignment
with (xi ) = ( (xi )) for all i = 1; : : : ; n makes H (x1 ; : : : ; xn ) false in A.
Therefore G is also not valid in A. Hence one has (ii).
Now assume (ii), i.e. that each universal sentence of the language LK of
the class K, which is valid in the particular structure A 2 K, is also valid
in each algebraic structure of K. Let A 2 K be any structure and D A
nite, say D = fa1 ; : : : ; amg. Enrich the language LK with constants ai to
denote ai , i = 1; : : : ; m. Consider the conjunction H of all the sentences of
the forms
fk (ai ; : : : ; ain ) = aj ;
:(fk (ai ; : : : ; ain ) = aj ) ;
1
13
Rj (ai ; : : : ; ain ) ;
:Rj (ai ; : : : ; ain ) ;
for any function symbol fk and any relation symbol Rj of the language LK ,
which are true in D, i.e. true in A. Substitute in H each of these constants
ai by a variable xi . The result shall be the w H (x1 ; : : : ; xm ). Now form
the sentence G 9 x1 ; : : : ; xm H (x1 ; : : : ; xm ). Obviously, G is valid in A,
which means that its negation G 8 x1 ; : : : ; xm :H (x1 ; : : : ; xm ) is not valid
in A. But G is a universal sentence, hence it is also not valid in the particular
structure A. Therefore G is valid in A. Let be some jAj-assignment which
makes H (x1 ; : : : ; xm ) true. By choice of H one then has that the mapping
f : ai 7! (xi ) with domain dom(f ) = D is a partial isomorphism from A
into A. Hence (i) holds true.
2
From an intuitive point of view one thus can understand the fact that all
the algebraic structures of a whole class K can be locally embedded into some
structure A in such a way that A is a kind of prototype for all the structures of
K, at least as far as facts are concerned which can be formulated by universal
sentences.
For the particular case of o-group one immediately has the following corollary from Theorem 1.2.1.
1
Corollary 1.2.1. If a universal sentence of the language of `-groups is valid

in the particular o-group R of the reals, then it is valid in each o-group.
14
2. The Formalized Language and its

Interpretations
In many-valued logic like in all other elds of modern formal logic, either
propositional or rst-order or even higher order, one has a common type of
approach: the main topics of consideration can be singled out most clearly on
the basis of a formalized language which comprises all { and essentially only
{ the necessary means of expressibility one has to have for studying these
topics. And this formalized language has a syntactic, as well as a semantic
aspect.
2.1 Propositional Syntax

For the syntactic considerations we assume that for each of the particular
systems S of propositional many-valued logic, which shall be discussed in the
sequel, its formalized language LS has to include the following items:
(a) an (countably) innite set V0 of propositional variables,
(b) a { usually nite { set J S of connectives, i.e. of propositional operators,
each of which has a (xed) arity 1,
(c) the brackets $" , $" and the comma \," as punctuation symbols which
support the unique readability of the (well-formed) formulas,
(d) and, occasionally, a set KS of truth degree constants.
As our propositional variables we take the letter \p" with any positive
number of primes \ 0 " as upper indices:
V0 = fp0; p00 ; p000 ; : : :g;
(2.1)
for simplicity of notation we also write p(n) instead of p00:::0 with n primes.
However, as usual, further on we only rarely shall have to x truly particular formulas of the formalized languages LS . In most cases, instead, it shall
suce to know the form of such formulas suciently well. And for this it usually suces to indicate that at some places of some (string of the alphabet
for S which represents some) formula some propositional variable appears.
To have a exible notation in these cases, we use the symbols
p; q; r; p1 ; q1 ; r1 ; p2 ; q2 ; r2 ; : : :
(2.2)
16
2. The Formalized Language and its Interpretations
to represent propositional variables.1

For completeness it should be noticed that the symbols mentioned in
(2.2) do not belong to our formalized language of propositional many-valued
logic. Instead they are used in the meta-language we use to speak about
our (formalized) propositional language { therefore these symbols often are
called metavariables or even: metavariables for propositional variables.2 All
the formulas which we later on shall write down and which are to contain
such metavariables from the list (2.2) therefore only indicate (well-formed)
formulas of our formalized propositional language.
The body of (primitive) symbols of our (formalized) language of propositional many-valued logic given by (a), (b), (c), and (d), the alphabet of this
language, determines { via concatenation { among others also those words of
this language which count as meaningful and are called well-formed formulas.3 As usual in logic, this set of ws of LS is the smallest set which
1. contains each propositional variable and each truth degree constant, together called atomic formulas,
2. contains with each n-ary connective ' 2 J S and all ws H1 ; H2 ; : : : ; Hn
also the string \'(H1 ; H2 ; : : : ; Hn )" which is called a compound formula.
By abuse of language, we often shall denote this set of ws of a formal
language LS by the same symbol \LS ". It will always be clear from the
context which type of use is the actual one.
As a standard corollary one gets that each non-atomic w results from
atomic formulas by nitely many applications of construction step 2, and
that one can prove statements for all ws by structural induction on this
construction process as in classical logic.
Additionally we suppose that the alphabets of our systems of propositional many-valued logic are chosen in such a way that each w H , built up
according to 1. and 2., either is atomic or compound. And if H is a compound formula '(H1 ; H2 ; : : : ; Hn ) its \constituents" '; H1 ; : : : ; Hn should be
uniquely determined by H .
In more mathematical terms this means that we suppose that our (propositional) alphabets form free semigroups with respect to concatenation. Additionally this means that the sets LS of ws forms the free algebra w.r.t. the
connectives of the language, now read as operations inside the set LS . And
this free algebra is freely generated by the set of propositional variables and
truth degree constants, i.e. by the set of atomic formulas.
1 In each formula, of course each of these symbols represents at all places of its
appearance the same propositional variable; dierent such symbols, however, are
not forced to represent dierent propositional variables.
2 We have used the \trick", to generate the innite set of all propositional variables
out of the two symbols \p" and 0 as indicated in (2.1), to have the positive integers
free for their use as lower indices with these metavariables.
3 As shorthand notation we shall use w for \well-formed formula" and ws for
\well-formed formulas".
2.2 Propositional Semantics
17
Given a compound formula H '(H1 ; H2 ; : : : ; Hn ) the { in this case

uniquely determined { connective ' usually is called the principal (or also:
main ) connective of H .
In the case of a binary connective ' we usually prefer the inx notation
over the prex notation as introduced in 2., i.e. we usually shall write
(H1 'H2 ) for '(H1 ; H2 ) ;
(2.3)
with the further agreement that in case of the inx notation the outer pair
of brackets may be deleted without causing any misunderstanding. Furthermore, in all the cases where the alphabet of LS has analogues for the classical
connectives :; ^ ; _ ; ) for negation, conjunction, disjunction, and implication, respectively, always the (analogue of the) negation connective shall have
preference over the (analogues of the) connectives for conjunction and disjunction, and these connectives shall have preference over the (analogue of
the) implication connective. As a result of these agreements one can avoid
parentheses in the actual writings of formulas which often is crucial for the
simplicity of reading.
This agreement for negation connectives we shall extend to any unary
connective and shall write
H1 for (H1 )
(2.4)
whenever possible without causing any misunderstanding.
The subformulas of any w H shall be H itself, and in the case of a
compound formula H '(H1 ; H2 ; : : : ; Hn ) also all the subformulas of the
ws H1 ; H2 ; : : : ; Hn .

Because many-valued logic accepts the principle of compositionality one has
truth functional behavior of the connectives, and as in classical logic the
propositional variables (formally) are taken as variables for truth degrees: in
each particular application of propositional many-valued logic it is only the
truth degree of the particular sentences appearing in place of the propositional
variables which counts for the discussion of the truth degree of the whole
(particular) sentence. Therefore propositional many-valued logic, like classical
propositional logic, presupposes that the connectives always are completely
determined by their corresponding truth degree functions { and therefore
also restricts its eld of considerations to the truth degree behavior of its ws
(avoiding any other reference to a possibly dierent \meaning" not connected
with truth degrees).
There are two parallel approaches toward the semantics of a system S
of many-valued logic, i.e. toward the relationship between the ws of its
language LS and their corresponding truth degrees:
18
either there exists some particular truth degree structure MS for S, the
(standard) logical matrix, which comprises all the necessary semantical

data which are needed to interpret the ws of the formalized language LS ,
or there exists some characteristic class KS of algebraic structures (of coin-
ciding similarity types) which all together act as logical matrices for S \of
equal right", and which all have to oer for LS all the necessary semantical
data which are needed to interpret the ws.
For each system S of propositional many-valued logic with a standard
logical matrix one denotes the set of truth degrees of S by W S . And one
has to have for each of its connectives ' a truth degree function verS' in W S
with the same arity as '. Accordingly one has to have for each truth degree
constant t of S a truth degree tS denoted by t. The (algebraic) structure
D
;

; E
MS = W S ; verS' '2J S ; tS KS
(2.5)
consisting of the truth degree set, the family of truth degree functions, and
the family of truth degrees denoted by the truth degree constants, then is
the logical matrix for S.4
Then each valuation : V0 ! W S of the propositional variables with
truth degrees assigns to each w H a truth degree, determined inductively5
by:
8
(p);
if H is the propositional variable p,
>
> S
>
if H is the truth degree constant t,
<t ;
ValS (H; ) = > verS' (ValS (H1 ; ); : : : ; ValS (Hn ; ));
(2.6)
>
if
H

'
(
H
;
:
:
:
;
H
)
for
the
n
-ary
>
1
n
:
connective '.
Because of the conditions we imposed on our alphabets this inductive denition uniquely determines the truth degree ValS (H; ) for each w H and
each valuation .6
As in classical propositional logic one now is able to prove inductively
that each truth degree ValS (H; ) depends only on the truth degrees which
assigns to the propositional variables of H .
Theorem 2.2.1. For each w H and all valuations ; : V0 ! W S which
satisfy (p) = (p) for each propositional variable p which occurs in H one
has
4 Besides this standard \matrix" semantics for S also Kripke-style semantics and
relational semantics are possible, cf. e.g. [314]. They are, however, not as popular
as the matrix semantics.
5 As usual, the induction here is over the complexity of the compound formulas
with the case of the atomic formulas as basis.
6 In the case that the semantics of S is not determined by some standard logical
matrix, but by a whole class KS of algebraic structures,
then each mapping of the
set V0 into the carrier of any structure A 2 KS counts as a valuation. Denition
(2.6) in this case applies accordingly in each of these structures from KS .
19
ValS (H; ) = ValS (H; ):

Proof: In the case that H is an atomic formula this is a straightforward consequence of denition (2.6). And for compound formulas H
'(H1 ; : : : ; Hn ) and valuations ; with (p) = (p) for all propositional
variables p which occur in H one has
ValS (Hi ; ) = ValS (Hi ; )
for all i = 1; : : : ; n by the induction assumption and the fact that all propositional variables which occur in one of the ws Hi also occur in H . Therefore
one has
ValS (H; ) = verS' (ValS (H1 ; ); : : : ; ValS (Hn ; )
= verS' (ValS (H1 ; ); : : : ; ValS (Hn ; )
= ValS (H; )
and the theorem is proved.
2
The basic reading of this theorem is that for ws H , and their set var(H )
of propositional variables which occur in them, the truth degree ValS (H; ) of
H under some valuation really depends only on the restriction var(H ).
In this sense H represents an m-ary truth degree function.
With this \truth degree behavior" of ws an important equivalence relation is connected: ws H; G of the language LS of some system S of propositional many-valued logic are semantically equivalent i one has
ValS (H; ) = ValS (G; ) for all valuations .
(2.7)
A simple calculation shows that this indeed is an equivalence relation in the
class of all ws of S. And this equivalence relation additionally satises the
following theorem on the substitutivity of equivalents.
Theorem 2.2.2 (Replacement Theorem). Let H 0; H 00 be some semantically equivalent ws, and let the w G be built up from the w H by exchanging some of the subformulas of H of form H 0 by the w H 00 . Then also H; G
are semantically equivalent.
Proof: The proof is by induction on the structure of the w H . Suppose
that the ws H 0 ; H 00 ; G satisfy the assumptions of the theorem.
In the case that H is an atomic formula, H is the only subformula of H .
If H 0 is dierent from H , then H 0 is not a subformula of H and G has to
coincide with H . However, if H 0 is H , then G is H or H 00 . In any case, H
and G are semantically equivalent.
Now assume that H '(H1 ; : : : ; Hn ) is a compound formula and that
the claim of the theorem holds true for the subformulas H1 ; : : : ; Hn of H . If
H 0 coincides with H or is not a subformula of H , then G is H or H 00 and semantically equivalent with H . Otherwise H 0 is subformula of some of the ws
H1 ; : : : ; Hm , and there exist ws H100 ; : : : ; Hn00 such that G '(H100 ; : : : ; Hn00 )
20
and such that each one of the ws Hi00 , 1 i n, can be built up

from Hi by exchanging some of the subformulas H 0 of Hi by H 00 . Therefore Hi ; Hi00 are semantically equivalent for all 1 i n, which means
ValS (Hi ; ) = ValS (Hi00 ; ) for each valuation : V0 ! W S . Hence one has
for all such valuations
ValS (H; ) = verS' (ValS (H1 ; ); : : : ; ValS (Hm ; ))
= verS' (ValS (H100 ; ); : : : ; ValS (Hm00 ; ))
= ValS (G; );
i.e. the ws H; G are semantically equivalent.
2
This truth degree function, which the w H represents, shall be denoted
by wH . Its value is for each n-tuple (t1 ; : : : ; tn ) of truth degrees given by
wH (t1 ; : : : ; tn ) =def ValS (H; )
(2.8)
for any valuation satisfying (pi ) = ti for all i = 1; : : : ; n.
Because of Theorem 2.2.1, condition (2.8) really characterizes a truth
degree function which is called truth degree function of H .
Sometimes it is suitable to allow additional, dummy arguments for the
truth degree function of a w H . Let the n propositional variables which occur in H appear among the propositional variables p1 ; : : : ; pk , as pi ; : : : ; pin .
Then an extended truth degree function w~H shall be determined by the equation
w~H (x1 ; : : : ; xk ) =def wH (xi ; : : : ; xin ):
(2.9)
In general, from the semantic point of view a system S of propositional
many-valued logic is uniquely determined by the totality of the following
data:
its set W S of truth degrees,
its set J S of connectives together with the truth degree function verS' for
each ' 2 J S ,
its set KS of truth degree constants together with the truth degree tS for
each t 2 KS .
These data, together with the set DS of designated truth degrees (and possibly
also the set of antidesignated truth degrees), as introduced in the following
Chapter 3, usually are collected into the logical matrix for S.
On the other hand, the set W S together with the family of the truth degree functions verS' of all the connectives of ' 2 J S form again an algebraic
structure: the truth degree structure. And this relationship between systems
of many-valued logic and their truth degree structures can also be reversed:
one can start from some algebraic structure A with a family op1 ; : : : ; opn of
operations as truth degree structure, and then form a corresponding formalized language LA which has for each one of the operations opi a corresponding
1
2.3 First-Order Syntax
21
connective 'i (of the same arity) such that opi is the truth degree function of
'i . We shall not follow this more abstract approach in general. But sometimes
we shall have to refer to it, e.g. in Chapters 14 and 21.
To simplify notation we shall omit the super-/subscripts S referring to a
particular system of propositional many-valued logic in the cases where this
system is clearly determined by the context.
The particular, basic means of expression provided by the alphabet of a
logical system S often are used to dene further connectives or truth degree
constants. We shall do the same later on. Regarding the (formalized) language
LS of this system one has two possibilities in such a situation, explained for
the case of a newly dened connective :
(i) either has to be added to the alphabet of the system S as a new (sign
for some) connective and the notion of w has to be adapted accordingly,
or
(ii) one allows to write down formally all the ws containing the new connective but reads them only as shortenings at the level of metalanguage
for those ws of the \pure" language of S which contain instead of the
new connective the \ -free" w dening .
We shall prefer version (ii) because on the one hand we already use metavariables (for propositional variables) and newly dened connectives or truth
degree constants act in the same way, and because on the other hand inductive proofs on the structure of ws remain as before (but would become more
complicated for version (i)).
In what follows, ws shall be indicated by upper case Latin letters:
H; G; A, B , C; : : :, possibly also with subscripts. And upper case Greek letters
like ; ; : : : shall denote sets of ws.

The languages LS of systems S of rst-order many-valued logic follow the
same pattern as the languages of systems of classical rst-order logic. For the
predicate symbols of these languages one thus has two dierent approaches:
one either may be interested in a rather uniform approach and decide to supply for any possible application and for each possible arity n 1 of predicate
symbols \suciently many" symbols, which means to supply countably many
of them { or one may choose, application dependent, just as many predicate
symbols (of suitable arity) as one needs to represent the basic notions for
that application. We shall choose this second type of approach, also having
in mind that in this case it usually is simpler to completely describe the
particular interpretations one is working with.
Therefore the alphabet of each system S of rst-order many-valued logic
shall be based on some countably innite set V of individual variables and
be determined by
22
its (nonempty) family J S of (basic) propositional connectives,

its (possibly empty) set KS of truth degree constants,
its set QS of quantiers, and for each one of them its number of argument
places for individual variables and its number of argument places for ws,7
its family P S of predicate symbols together with an arity function assigning

to each predicate symbol its arity,
the brackets $" , $" and the comma \," as technical signs which support
the uniqueness of the reading of (well-formed) formulas,
and possibly its set C S of individual constants and its set OS of function
symbols.
As individual variables we take the letter \x" with any number of primes \ 0 "
as upper indices:
V = fx0 ; x00 ; x000 ; : : :g:
(2.10)
For simplicity of notation we sometimes also write x(n) instead of x00:::0 with
n primes.
However, as usually, further on we only rarely shall have to x truly
particular formulas of the formalized languages LS . In most cases, instead,
it shall suce to know the form of such formulas suciently well. And for
this it usually suces to indicate that at some places of some (string of the
alphabet for S which represents some) formula some individual variable or
some predicate symbol occurs. To have a exible notation in these cases, we
use the symbols
x; y; z; x1 ; y1 ; z1; x2 ; y2 ; z2; : : :
(2.11)
to represent individual variables, i.e. we use these symbols as metavariables
for individual variables.8
The individual symbols are the individual variables together with the individual constants and shall be indicated by the symbols
a1 ; a2 ; a3 ; : : :
(2.12)
which sometimes, however, are restricted to indicate individual constants
only. The correct reading shall always be clear from the actual context.
7 This is a very general notion of quantier which was introduced in [465]. Quite
often one restricts the considerations to quantiers which have only one argument
place for an individual variable, the quantied variable, and one argument place
for a w, the scope of that quantier. But already in two-valued rst-order logic
either restriction can be lifted: so e.g. in innitary rst-order logic one allows
innitely long ws together with quantiers 8; 9 which may have countably many
argument places for individual variables, and the usual rst-order logic sometimes
is enriched with an equi-cardinality quantier I with two argument places for ws
such that the w Ix(H (x); G(x)) is understood as \there are just as many x such
that H (x) as there are x such that G(x)".
8 In each formula, of course each of these metavariables represents at all places of its
appearance the same individual variable; dierent such metavariables, however,
are not forced to represent dierent individual variables.
23
It then is a routine matter to dene the class of well-formed formulas and

to distinguish free and bounded occurrences of individual variables within
well-formed formulas (ws). The class of all ws of the language LS of S is
(in the case OS = ; that there are no function symbols in the alphabet) the
smallest set of strings over the alphabet of S which
1. contains each truth degree constant and each atomic w, i.e. each string
P (a1 ; : : : ; an ) which is built up from an n-ary predicate symbol P 2 P S
with its argument places lled with individual symbols a1 ; : : : ; an ;
2. contains for each connective ' 2 J S of arity n and all ws H1 ; : : : ; Hn
of LS also the string '(H1 ; : : : ; Hn );
3. contains for each quantier Q with k argument places for individual
variables and n argument places for ws and any individual variables
x1 ; : : : ; xk and ws H1 ; : : : ; Hn of LS also the string
(Qx1 ; : : : ; xk )(H1 ; : : : ; Hn ):
(2.13)
Such a quantier Q which has k argument places for individual variables and
n argument places for ws is said to be of arity (k; n). As in classical logic, the
quantication (Qx1 ; : : : ; xk ) in the w (2.13) is binding the individual variables x1 ; : : : ; xk , and all the occurrences of the individual variables x1 ; : : : ; xk
in the scope (H1 ; : : : ; Hn ) of the quantication (Qx1 ; : : : ; xk ) in the w (2.13)
are bounded occurrences. In general, an occurrence of an individual variable
x within some w H is a bounded occurrence i x occurs within the scope of
a quantication which is binding this variable, i.e. where x lls in one of the
argument places for individual variables of the quantier. On the contrary,
an occurrence of a variable x within a w H is a free occurrence i this is
not a bounded occurrence and is not an occurrence in an argument place for
individual variables of some quantier. And the sentences are those ws in
which there are no free occurrences of individual variables.
Our languages LS allow also that there are function symbols in the alphabet. In this case one furthermore assumes that there is an equality sign = in
the alphabet. As in classical logic, each such function symbol has some xed
arity. And with the function symbols terms of LS can be formed. The set of
terms of LS again is the smallest set of strings over the alphabet which
contains all the individual symbols;
contains with each n-ary function symbol F 2 OS and all terms 1 ; : : : ; n
also the string F (1 ; : : : ; n ).
In such a situation with OS 6= ; the notion of atomic w has to be generalized
a bit: an atomic w then is each string which one gets by lling the argument
places of some predicate symbol with terms, and also each string of the form
1 = 2 for terms 1 ; 2 .
The relationship between systems S with and without function symbols
hence is completely the same in many-valued logic as is in classical logic.
Therefore we shall not treat systems with function symbols particularly, but
24
mainly restrict to systems without function symbols. The only aspect which
needs emphasis is the understanding of the equality symbol. And this shall
be discussed later on in Chapter 16.
We adopt for the rst-order systems the notational conventions (2.3), (2.4)
for the use of connectives. And we add for quantiers Q with one argument
place for variables and also one argument place for ws the agreement to
write
QxH instead of (Qx)(H ):
(2.14)
And also the possibilities to omit parentheses which are oered by the preferences among (dierent kinds of) connectives is extended to the rst-order
setting: quantications as in (2.13) always shall have preference over (propositional) connectives. Hence QxH ^ G e.g. means (QxH ) ^ G, and the scope
of a quantication is the shortest string following this quantication which is
itself a w.
2.4 Many-Valued Predicates

The predicates of classical rst order logic formalize the intuitive notion of
\notion". For the formal treatment of the semantics of rst-order logic each
such predicate is considered w.r.t. some predetermined, nonempty universe
of discourse { and identied by the extent of the notion this predicate is formalizing. The extension principle of classical logic states that each predicate,
i.e. each notion, is uniquely determined by its extent. Again in more formal,
set theoretic terms: the extent of some n-ary predicate, i.e. of some notion
which represents a relationship between n objects, is a set of n-tuples of the
respective universe of discourse, collecting all those groupings of n objects
which fall under this notion.
Accordingly also rst-order many-valued logic always starts from the assumption that each particular reading of its formulas is tied with some predetermined, nonempty universe of discourse. There are, as in the case of classical
logic, no restrictions concerning the nature of the objects of these universes
of discourse. Therefore all the considerations of rst-order many-valued logic
apply to all possible universes of discourse. As in the propositional case one
has the principle of compositionality for the propositional connectives. Additionally, as does classical rst-order logic, also rst-order many-valued logic
accepts a principle of extensionality. And this principle of extensionality says
for the predicates of rst-order many-valued logic that they are uniquely
characterized by their extents.
Of course, rst-order many-valued logic omits the principle of bivalence,
in the very same manner as propositional many-valued logic does. And this
means in the rst-order setting additionally that objects or n-tuples of objects
(of the respective universe of discourse) fall under some given n-ary predicate
P to some truth degree (which may be dierent to the classical truth values
2.4 Many-Valued Predicates
25
>; ?). In the case that W is this set of truth degrees, P is also called a
W -valued predicate.
But what kind of object should, formally, be the extent of such a W -valued
predicate? It cannot be a set or a class in the usual understanding of these

terms which one knows from classical logic or (usual) set theory. A nice and
natural way out of this problem is oered by the mathematical modeling of
vague notions which started, following a series of less in uential forerunners,
in about 1965 mainly from an applicational background in systems science.
Such vague notions are notions like: hot water, gifted student, high mountain,
pretty woman, tall man etc. They all have (intuitively) the crucial property
that there are \borderline cases" for their applicability, i.e. objects for which
one is unable to decide whether such a notion applies to them, or does not
apply { or for which such a notion does apply only \to some degree". The
basic idea for the mathematical modeling of such vague notions has be to
introduce, for their extents, a kind of generalized sets { called fuzzy sets in
[582], cf. Chapter 18 { which allow for degrees of membership.
The relationship to the extents of \many-valued" predicates, i.e. predicates of rst-order many-valued logic then is almost obvious: having given
such a unary many-valued predicate P and some object a from its universe
of discourse, the truth degree to which P applies to a, i.e. the truth degree
of P (a), has to be the (membership) degree to which a is a member of the
extent of P . The extent of P hence becomes a fuzzy set over, or a fuzzy subset
of the universe of discourse for P . And for each n-ary many-valued predicate
Q its extent, obviously, becomes a fuzzy set of n-tuples of elements of the
universe of discourse for Q. On the other hand one gets with the language of
rst-order many-valued logic a very suitable tool for the development of the
theory of fuzzy sets, cf. Chapter 18.
Another possibility to make precise what the extents of (n-ary) manyvalued predicates P should be, is to introduce such an extent as a whole
family of classes: for each truth degree t (with the possible exception of
just one of these truth degrees) one class which collects just all those (ntuples of) elements of the universe of discourse for P to which P applies with
degree t. However, this second type of approach is essentially equivalent to
the previously mentioned one via fuzzy sets, because each fuzzy set can be
represented by such a family of usual sets, cf. e.g. [31].
But independent of these mathematical details, in any case the extent
of some many-valued predicate P w.r.t. some universe of discourse A is determined \up to isomorphy" by the assignment of the truth degrees to the
objects of A, or the n-tuples of objects of A, i.e. by a function (of suitable
arity) from A into the truth degree set.
26
2.5 First-Order Semantics

Associated with each system S of rst-order many-valued logic one has the
following data:
the language LS including the alphabet and the rules for forming ws,
the set W S of truth degrees, which again preferably shall be chosen as in
(5.2) or as in (5.3), (5.4), together with the subset DS of designated truth
degrees (or together with disjoint subsets DS+ ; DS; of designated and of
antidesignated truth degrees),
the family of truth degree functions verS' for the connectives ' 2 J S ,
for each truth degree constant t of S its corresponding truth degree tS 2
W S,
and nally also
the family of (classes of) generalized truth degree functions which interpret
the quantiers.
The generalized truth degree function VerSQ which is associated with a
quantier Q of arity (k; n) has, over a given universe of discourse A, to map
each function f : Ak ! (W S )n to some truth degree. And additionally this
generalized truth degree function VerSQ has to satisfy VerSQ (f ) = VerSQ(g) for
all functions f : Ak ! (W S )n and g : Bk ! (W S )n for which there exists a
bijection h : A ! B with property
f (a1 ; : : : ; ak ) = g(h(a1 ); : : : ; h(ak ))
for all a1 ; : : : ; ak 2 A. The intuition behind this choice of VerSQ is (i) that this
generalized truth degree function should not distinguish between elements of
A, i.e. should treat them all in the same way, and (ii) that VerSQ also should
act in the same way over dierent, but equipollent universes of discourse.
Two particularly important quantiers are immediate generalizations of
the universal and the existential quantier of classical logic, also denoted \8"
and \9". Both have arity (1; 1) and are characterized by the generalized
truth degree functions om and ex dened for all functions f : A ! W S by
VerS8 (f ) = om(f ) =def inf ff (a) j a 2 Ag ;
(2.15)
VerS9 (f ) = ex(f ) =def supff (a) j a 2 Ag :
(2.16)
Obviously both these generalized truth degree functions satisfy the above
mentioned conditions, and they either suppose W S = W or at least that W S
bears the structure of a complete lattice.
From the data which have to be associated with some xed system S
of rst-order many-valued logic it thus only remains open which objects are
denoted by the individual symbols, and which are the many-valued predicates
which are denoted by the predicate symbols of S. And all these data may vary
from one particular case to another one, as in classical logic { and are xed
in any particular case by some interpretation A for S.
2.5 First-Order Semantics
27
Such a particular interpretation A therefore collects

a nonempty set A = jAj of individuals, the universe (of discourse) of the
interpretation,
for each individual constant a of LS a corresponding individual aA 2 A,
for each predicate symbol P of LS with arity n a corresponding n-ary
W S -valued relation P A in A.
As a means to discuss the truth degree behavior of the ws in some xed
interpretation A for a system S of rst-order many-valued logic we need a
generalization of the satisfaction relation of classical logic. The satisfaction
relation (A; ) j=S H , H any well formed formula and : V ! A any
assignment (of values to variables), now becomes a graded relationship and
hence preferably is dened as a function ValSA (H; ) mapping well-formed
formulas of LS and assignments to truth degrees. As usual, the denition is
by induction on the complexity of the w H . In the case that H is a truth
degree constant t one has
ValSA (t; ) =def tA :
(2.17)
For atomic ws H P (a1 ; : : : ; an ) one has
ValSA (P (a1 ; : : : ; an ); ) =def P A ( : a1 ; : : : ; : an );
(2.18)
using for the individual symbols a1 ; : : : ; an the notation

individual variable
: ai =def aA(a; i ); ifif aai isis an
(2.19)
i an individual constant.
i
For compound formulas H '(H1 ; : : : ; Hn ), built up from ws H1 ; : : : ; Hn
using some n-ary connective ', one has as in the propositional case
;

ValSA ('(H1 ; : : : ; Hn ); ) =def verS' ValSA (H1 ; ); : : : ; ValSA (Hn ; ) :(2.20)
And for the last possibility that H has been formed by quantication, i.e.
that one has H (Qx1 ; : : : ; xk )(H1 ; : : : ; Hn ) for some quantier Q 2 QS
of arity (k; n), suitable individual variables x1 ; : : : ; xk , and ws H1 ; : : : ; Hn ,
one rst has to introduce a modication of an assignments determined by
some n-tuple y1 ; : : : ; yn of individual variables and some n-tuple b1 ; : : : ; bn of
elements of the universe of discourse as

yi for all i = 1; : : : ; n
[y1 =b1; : : : ; yn =bn](x) = b (x); ifif xx 6=
= yi for some i = 1; : : : ; n. (2.21)
i
Then one is able to built for each assignment and each quantied w
H (Qx1 ; : : : ; xk )(H1 ; : : : ; Hn ) another mapping AH : jAjk ! (W S )n characterized for all b1; : : : ; bn 2 jAj by the equation
;

AH (b1 ; : : : ; bn) =def ValSA (H1 ; ); : : : ; ValSA (Hn ; )
with additionally = [x1 =b1; : : : ; xn =bn].
Having all these additional tools at hand one simply has
28
ValSA ((Qx1 ; : : : ; xk )(H1 ; : : : ; Hn ); ) =def VerSQ(AH ):

(2.22)
These details are essentially standard. Moreover, the quantier case can
be considerably simplied for the above mentioned many-valued versions 8; 9
of a universal and an existential quantier: one nds
ValSA (8xH; ) = inf fValSA (H; [x=b]) j b 2 jAjg;
(2.23)
S
S
ValA (9xH; ) = supfValA (H; [x=b]) j b 2 jAjg
(2.24)
by straightforward calculations.9
9 The reader should check the details for himself.
3. Logical Validity and Entailment
3.1 Designated Truth Degrees

Despite the diculties with a uniform intuitive understanding of the meaning
of all the truth degrees, which we mentioned in Section 1.1, or only of the
\additional" truth degrees which are added to the classical ones, there is a
kind of common agreement that truth degrees realize some sort of grading of
\truth", and perhaps also some sort of grading of \falsity".
Therefore usually one assumes that the classical truth values >; ? (or
some \isomorphic" copies of them, coded by 1 and 0) appear among the
truth degrees of any suitable system S of many-valued logic, and are their
lower and upper bounds:
f0; 1g W S and 0 u 1 for each u 2 W S .
(3.1)
This agreement, however, does not solve the problem which truth degrees
code \degrees of truth" and which other ones, possibly, code \degrees of
falsity". It only tells that degree 1 codes truth, and degree 0 codes falsity.
For the remaining \additional" truth degrees from W S nf0; 1g some further
decision along those lines seems to be advisable.
Quite often one is mainly interested in the truth degrees which, besides
degree 1, code the intuitive notion of truth. Formally this means, that with
each system S of many-valued one connects not only his set W S of truth
degrees but also some set DS of designated truth degrees. Of course, usually
one supposes
1 2 DS W S together with 0 2= DS :
(3.2)
Such a choice of designated truth degrees is of fundamental importance for
the generalization of the notions of logical validity and of logical consequence.
A further generalization, although not so common as the use of a set of
designated truth degrees, intends to have truth degrees which correspond to
> as well as truth degrees which correspond to ?. Accordingly, then, one has
to distinguish designated truth degrees and antidesignated 1 ones, i.e. one has
to determine two disjoint sets DS+ ; DS; such that
1 The terminology is not unique here: sometimes one denotes these designated and
antidesignated truth degrees also as positively and negatively designated ones.

This use of language has, however, the disadvantage that the objects called here
30
DS+ [ DS; W S and DS+ \ DS; = ; ;
usually together with the additional assumptions

1 2 DS+ and 0 2 DS; :
Regarding these sets DS , respectively these pairs DS+ ; DS; of sets of designated and antidesignated truth degrees there is still another agreement
concerning their choice which comes into consideration in the cases where
the whole truth degree set W S has some internal ordering 6: with a 2 DS
and a 6 b 2 W S also b 2 DS should hold true, respectively with a 2 DS+ and
c 2 DS; for all b; d 2 W S with d 6 c and a 6 b also b 2 DS+ and d 2 DS;
should hold true.
Even in the case that DS+ 6= ; and DS; 6= ; it is, however, not necessarily DS+ [ DS; = W S , which means that together with designated and
antidesignated truth degrees also undesignated truth degrees may exist. This
possibility indicates two essentially dierent positions regarding the designation of truth degrees. The rst one assumes only a binary division of the set
of truth degrees and can proceed by simply marking a set of designated truth
degrees, treating the undesignated ones like antidesignated ones. The second
position assumes a tripartition and marks some truth degrees as designated,
some others as antidesignated, and has besides these both types also some
undesignated truth degrees, usually presupposed to lie in between the designated and the antidesignated ones. (This second position, therefore, usually
also presupposes some ordering of the truth degrees.)
In particular systems of many-valued logic both these positions are not
always clearly distinguished, and preference is given to the particular denitions in these systems over this general classication. Moreover, the rst
position usually has de facto preference over the second one. Therefore in
what follows we shall essentially consider only this rst position.
With designated truth degrees at hand it is rather easy to establish \correlations" between connectives or truth degree functions of systems of manyvalued logic and of classical logic, mainly in the sense to classify connectives
in many-valued logic as being \generalizations" of classical negation, conjunction etc.
Let us suppose that non; et; vel; seq shall be some truth degree functions
of connectives in many-valued logic which shall count as generalizations of
negation, conjunction, disjunction, and implication, respectively. We shall say
that these truth degree functions satisfy the standard conditions i there hold
true:
(N) non(x) is a designated truth degree i x is not a designated truth degree;
(C) et(x; y) is a designated truth degree i x; y both are designated truth
degrees;
designated truth degrees, then sometimes have to appear as designated truth
degrees, and sometimes as positively designated truth degrees { depending on
whether there exist antidesignated truth degrees or not.
3.2 The Propositional Situation
31
(D) vel(x; y) is a non-designated truth degree i x; y both are non-designated

truth degrees;
(I) seq(x; y) is a non-designated truth degree i x is a designated truth degree
and y is a non-designated one.
Here it is assumed that there exist only designated truth degrees coding
\degrees of truth", and that the non-designated truth degrees code \degrees
of falsity", cf. [465].
Reference to these standard conditions presupposes that one is interested
to relate a given connective or truth degree function of a system of manyvalued logic with a xed classical connective or truth function. There is,
however, also another possibility to relate some of the connectives or truth
degree functions of many-valued logic with classical counterparts. We shall
say that a connective ' of many-valued logic or its corresponding truth degree
function verS' satisfy the normal condition i the truth degree function verS'
has for all the cases where all of its arguments are from f0; 1g also its values
from this subset f0; 1g of the truth degree set. To put it in more mathematical
terms: to satisfy the normal condition means for a connective ' that the
restriction of the truth degree function verS' to f0; 1g maps into f0; 1g, i.e. is
itself a truth function of classical logic.2
The requirements set by the normal condition on the one hand, and by
the standard condition on the other, are mutually independent: we later on
shall have to consider connectives which satisfy the normal condition but
not the standard condition, as well as connectives which satisfy the standard condition but do not satisfy the normal condition. Furthermore it is
rather obvious that one can construct examples of connectives or { better {
truth degree functions which either satisfy the normal as well as the standard
condition, or which satisfy neither.
Having the set DS+ of designated truth degrees of a system S of many-valued

logic one has an almost standard way to introduce the notions of logical
validity and of (semantical) entailment for this system S. A w H of the
language LS of a system S of propositional many-valued logic is S-logically
valid 3 or an S-tautology i its truth degree always is a designated one, i.e. i
ValS (H; ) 2 DS for all valuations : V0 ! W S :
(3.3)
In other words, H is S-logically valid i the truth degree function wH realized
by H maps into DS . The set of all S-logically valid ws shall be denoted by
tautS .
2 Of course this means that one has to identify the classical truth values >; ? with
their counterparts 1; 0 from many-valued logic.
3
In short, logically valid in the cases where the referential system S is clear from
the context.
32
As in classical logic, the set tautS is closed under substitutions.

Theorem 3.2.1. Suppose that H is S-logically valid and that H 0 results from
H by substituting (simultaneously) for the propositional variables p1 ; : : : ; pk
of H the ws H1 ; : : : ; Hk 2 LS . Then also H 0 is S-logically valid.
Proof: Let : V0 ! W S be a valuation and consider the new valuation
dened by

S (H ; ); if p p for some i = 1; : : : ; k
i
i
(p) = Val
(p)
otherwise.
By construction of H 0 and one immediately proves inductively on the
structure of H that
ValS (H 0 ; ) = ValS (H; ) :
Therefore one has ValS (H 0 ; ) 2 DS for all valuations and thus H 0 2 tautS .
If one is interested to generalize the notion of contradiction or logical

falsity in a similar way, one has two distinct possibilities:
1. In the case that the given system S of propositional many-valued logic
has a suitable negation connective :, one can take as logical falsities all
those ws H for which :H is a logical truth.
2. In the case that the given system S of propositional many-valued logic
has antidesignated truth degrees, one can take as logical falsities all those
ws H which assume only antidesignated truth degrees, i.e. all those H
whose associated truth degree function wH maps into DS; .
Both these possibilities can apply for the same system of many-valued logic.
They need not, however, characterize the same classes of ws. But with a
negation connective in possibility (1) which satises the standard condition
(N), and with no undesignated truth degrees for possibility (2), i.e. with
DS+ [ DS; = W S , both possibilities coincide.
A valuation : V0 ! W S which gives a w H a designated truth degree
ValS (H; ) 2 DS is called a model of H . The model class ModS (H ) of a w
H is the class of all models of H :
ModS (H ) =def f j ValS (H; ) 2 DS g :
(3.4)
For sets of ws their model class correspondingly shall be the set of all
valuations which give to all the ws of a designated truth degree:
ModS ( ) =def f j 8 H 2 (ValS (H; ) 2 DS )g :
(3.5)
As an elementary corollary one has that the logical validities are just those
ws H whose model class is the class of all valuations.
By the way we mention that a valuation which is not a model of a
w H or of a set of ws , i.e. a valuation with ValS (H; ) 2= DS , or with
33
ValS (H ; ) 2= DS for some H 2 , is often also called a countermodel of H

or of .
Sometimes it is also suitable to consider for any truth degree t 2 W S the
notion of t-model of by imposing the condition that each H 2 has
truth degree t under :
ValS (H; ) = t for each H 2 :
(3.6)
And in the case of the presence of some (partial) ordering of the truth
degree set also the notion of a ( t)-model of becomes important: is
a ( t)-model of i each H 2 has truth degree t under . In the case
of a singleton = fH g one, of course, speaks of a t-model (or: ( t)-model)
of H .
A third class of ws and sets of ws will be those ones which are satisable,
in a suitable sense of this word. A w H shall be called S-satisable i there
exists some valuation which gives H a designated truth degree, i.e. whose
model class is nonempty. Accordingly we dene for sets of ws:
S-satisable =def ModS ( ) 6= ; :
(3.7)
Equally important as the notion of being logically valid is for ws the
notion of being entailed by or being a consequence of a set of premises, in
many-valued logic as in classical logic. And again the reference to designated
truth degrees is of crucial importance for dening this notion, which in manyvalued logic again in a natural way generalizes its classical counterpart.
For a system S of propositional many-valued logic let H be a w and a
set of ws of the language of S. Then H is called a consequence of , or one
says that H is entailed by , i each assignment of truth degrees which gives
designated truth degrees to all ws of also gives a designated truth degree
to H . Of course, this entailment relation depends on the particular system
S of many-valued logic under consideration. We shall write j=S H for the
fact that entails H , and shall write 6j=S H i this is not the case.4 Using
the notion of model as introduced in (3.4) we formally dene
j=S H =def ModS ( ) ModS (H ) :
(3.8)
For the particular case = ; one simplies notation as usual and agrees that
j=S H means ; j=S H :
(3.9)
The fact that ModS (;) is the class of all valuations gives as an immediate
corollary for each w H that
j=S H i H 2 tautS ;
(3.10)
which { like in classical logic { means that the logically valid ws are just
those ws which are \generally" entailed, i.e. entailed by the empty set of
premises.
4 The entailment relation j=S thus itself is a two-valued relation. Later on in Chapter 19 we shall discuss also graded notions of entailment.
34
As a side remark it should be mentioned that the notion of ( t)-model

allows also for another denition of the entailment relation, which shall be
denoted by j=wS and dened by the condition
j=wS H =def each ( t)-model of is a ( t)-model of H . (3.11)
We will, however, not look at this version of entailment in greater detail.
Using the notation CnS ( ) for the set of all consequences of , i.e. dening
CnS ( ) =def fH j j=S H g
(3.12)
causes also for systems of propositional many-valued logic that this consequence operation CnS becomes a closure operation, i.e. has the following basic
properties.
Theorem 3.2.2. For all sets ; of ws of any propositional system S of
many-valued logic there hold true
(a) CnS ( ) ;
embedding property,
S
S
(b) ) Cn ( ) Cn () ;
monotonicity,
(c) CnS (CnS ( )) = CnS ( ) ;
idempotency.
Proof: Claim (a) results immediately from denitions (3.5) and (3.7). To
establish claim (b) it suces to remind ourselves that in the case each
model of is also a model of , i.e. that in this case ModS () ModS ( )
holds true. Because then for H 2 CnS ( ) one has j=S H and hence also
ModS (H ) ModS ( ) ModS (), therefore j=S H holds true, i.e. H 2
CnS (). And to establish claim (c) it suces to prove CnS (CnS ( )) CnS ( ),
because the converse inclusion is a simple consequence of (a). For any H 2
CnS ( ), however, one has CnS ( ) j=S H and hence also ModS (CnS ( ))
ModS (H ) according to (3.8). Therefore, for proving j=S H it suces to
prove ModS ( ) ModS (CnS ( )). But for any valuation 2 ModS ( ) and
any H 2 CnS ( ) one has 2 ModS (H ) according to (3.7) and (3.8). Hence
ModS ( ) ModS (CnS ( )) and thus also (c).
2
Another important result which generalizes from classical to many-valued
logic, at least to all systems with (only) nitely many truth degrees,5 is the
following compactness theorem.
Theorem 3.2.3 (Compactness Theorem). Let S be a system of propositional many-valued logic with a nite set W S of truth degrees, and let be
any set of ws of LS . Then is S-satisable, i.e. has a model i each nite
subset of is S-satisable.
Proof: Obviously, each model of is also a model of each subset of ,
hence also of each nite subset of .
5 Indeed, this compactness theorem does not hold true for all systems of (propo-
sitional) many-valued logic.
35
Therefore we now suppose that each nite subset of has a model. To

nd a model of , we try to construct by successively extending \partial"
valuations suitably chosen over initial segments of the sequence of all propositional variables. For this let Pn = fp0 ; p00 ; : : : ; p(n) g for all integers n 1,
and additionally P0 = ;.
By induction on n we shall prove the claim that there exists a partial
valuation n : Pn ! W S such that:
(a) For each r < n the valuation r is the restriction n Pr , i.e. one has
n (q) = r (q) for all q 2 Pr .
(b) For each nite 0 there exists a model n of 0 which coincides on
Pn with n , i.e. which satises n (q) = n (q) for all q 2 Pn .
For n = 0 nothing has to be proved because condition (a) becomes empty, and
condition (b) is satised because each nite 0 has a model. Therefore
we now suppose that our claim holds true for n = k and have to prove that
it holds true also for n = k + 1. We simply put
k+1 (q) = k (q) for all q 2 Pk
which makes (a) true for n = k + 1. The crucial point now is to dene
k+1 (p(k+1) ).
Because S is supposed to have only nitely many truth degrees, i.e. to be
an m-valued system, we may { without loss of generality6 { suppose to have
W S = f mj;1 j 0 j < mg. Then for the propositional variable p(k+1) there
are two possibilities:
(i) There exists a nite set 0 with (p(k+1) ) = 0 for each model of
0 which coincides on Pk with k .
(ii) For each nite sets 0 there exists a model which coincides on Pk
with k , and for which (p(k+1) ) 2 f m1;1 ; m2;1 ; : : : ; 1g holds true.
In the case (i) each nite set 0 has a model which coincides on Pk with
k and which satises (p(k+1) ) = 0. If this is not the case, there would exist
a nite set 0 0 such that (p(k+1) ) > 0 would hold true for every model
of 0 0 which coincides on Pk with k . But then the nite subset 0 [ 0 0
of could not have a model which coincides on Pk with k contrary to our
inductive assumption that (b) holds true for n = k. Hence we can put
k+1 (p(k+1) ) =def 0 in case (i)
and thus satisfy condition (b) also for n = k + 1 in this case (i).
In the case (ii) we shall consider the following two subcases:
(iii) There exists a nite set 1 with (p(k+1) ) = m1;1 for each model
of 1 which coincides on Pk with k .
(iv) For each nite sets 1 there exists a model which coincides on
Pk with k , and for which (p(k+1) ) 2 f m2;1 ; m3;1 ; : : : ; 1g holds true.
6 A simple bijection always suces to realize this situation.
36
As before in case (i), also now in case (iii) one easily proves that each nite
set 0 has a model which coincides on Pk with k and which satises
(p(k+1) ) = m1;1 . Because otherwise there would exist a nite set 1 0
with (p(k+1) ) > m1;1 for each model of 1 0 , and this again would result
in a contradiction as before. Hence we can put
k+1 (p(k+1) ) =def m 1; 1 in case (iii)
and thus satisfy condition (b) also for n = k + 1 in this case (iii).
In the case (iv) again we shall subdivide the possible situations into two
further cases constructed in analogy to the cases (iii) and (iv). And this whole
procedure has to be continued7 up to a step where one reaches the situation
that one has to discuss as subcases:
;2
(2m-3) There exists a nite set m;2 with (p(k+1) ) = m
m;1 for each
model of m;2 which coincides on Pk with k .
(2m-2) For each nite set m;2 there exists a model which coincides
on Pk with k , and for which (p(k+1) ) = 1 holds true.
It is now obvious for the case (2m ; 2), and provable as was before in the
cases (i) and (iii) also now for the case (2m ; 3), that one can put

m;2
case (2m ; 3)
(
k
+1)
k+1 (p ) =def 1m;;1 ; in
in case (2m ; 2)
and thus satisfy condition (b) also for n = k + 1 in both these nal cases.
Thus we have that for each integer n there exist valuations n : Pn ! W S
which satisfy the conditions (a) and (b). And with reference to these partial
valuations we dene the valuation : V0 ! W S by
(p(n) ) =def n (p(n) ); n = 1; 2; 3; : : : :
Our proof now is nished if we show that is a model of .
Let therefore H 2 , and k be chosen in such a way that all the propositional variables which occur in H belong to Pk . Now consider a model of the
nite set fH g which coincides on Pk with k . Then one has ValS (H; ) 2 DS ,
and furthermore also ValS (H; ) = ValS (H; ) because and coincide for
all propositional variables of H . Therefore also is a model of H , and even
2 ModS ( ) because H was arbitrarily chosen from .
2
A slight partial generalization of this compactness theorem is also of interest, and helpful for the proof of the next theorem.
Theorem 3.2.4. Let S be a nitely many-valued propositional system and
G 2 LS a w of its language. Assume that for a set LS of ws each of
its nite subsets has a model which at the same time is not a model of G.
Then itself has a model which is not a model of G.
7 Indeed, what we are doing here is a second \local" inductive argument inter-
twined with our \global" induction procedure.
37
The proof of this theorem can be given almost word by word like the
proof of the Compactness Theorem, the main changes have only to be that
one always considers instead of models of or of 0 such models of
these sets of ws which at the same time are countermodels for G. For the
proof that the valuation constructed in the proof is also a countermodel
for G one simply has to refer to the fact that in G there appear only nitely
many propositional variables.8
As a side remark it should be noted that Theorem 3.2.4 is a strengthening
of the (essential half of the) Compactness Theorem in all those cases where
LS contains a w which is not logically valid.
Theorem 3.2.5 (Finiteness Theorem for Entailment). Let S be some
propositional system of many-valued logic with a nite set of truth degrees,
and let j=S H hold true for some w H and some set of ws. Then there
exists a nite set such that j=S H holds true.
Proof: Let 0 be a nite subsetS of . In the case that 0 S6j=S H holds true
there exists a valuation 2 Mod ( 0 ) such that 2= Mod (H ). Therefore,
in the case that 0 6j=S H holds true for every nite 0 , by Theorem
3.2.4 one gets the existence of a 2 ModS ( ) with 2= ModS (H ), i.e. one
gets 6j=S H . By contraposition, the theorem follows.
2
For the situation that the Compactness Theorem as well as the Finiteness
Theorem for Entailment hold true in the above mentioned general forms it
is crucial to consider systems of many-valued logic with only nitely many
truth degrees. With a counterexample we show that both theorems do not
hold true in general for systems of many-valued logic with innitely many
truth degrees.
Consider a propositional system S^ with the innite set of truth degrees

W ^S = f0g [ 21n j n 0
which furthermore has a binary connective *, a unary connective ], a truth
degree constant 0, and a set DS^ of designated truth degrees with 0 2= DS^ . The
truth degree constant 0 shall denote the truth degree 0, and the truth degree
functions of the connectives shall satisfy the conditions
^
verS*
(x; y) 2 DS^ i x y;
(3.13)
^
S
1
ver] (x) = 2 x:
(3.14)
For each w H of S^ we denote by ]n H the n-fold iteration of the ]-application:
this means ]0 H =def H and ]k+1 H =def ](]k H ). And we consider
= fp0 * ]n p00 ) j n 0g;
G (p0 * 0):
8 The interested reader is recommended to work out the details as an exercise to
test his/her understanding of the whole proof procedure.
38
It is easy to recognise that each 2 ModS^ ( ) has to satisfy (p0 ) = 0.

Therefore one immediately has j=S^ G. But for any nite 0 and
any 2 ModS^ ( 0 ) one gets only the inequality (p0 ) ( 12 )m (p00 ) for the
maximal \exponent" m with (p0 * ]m p00 ) 2 0 . Therefore there exist models
2 ModS^ ( 0 ) satisfying (p0 ) 6= 0 and hence 2= ModS^ (G). Hence one has
6j=S^ G for each nite subset 0 of .
Thus Theorem 3.2.5 does not hold true for S^, and therefore also the Compactness Theorem fails for S^.
3.3 The First-Order Situation

Proceeding as usual in the rst-order situation, a rst-order w H is called
S-valid (or simply valid if the reference to the particular system S is obvious)
in an interpretation A i H has a designated truth degree for each suitable
assignment of objects of jAj to the individual variables of LS in A. The w
H is called S-logically valid (or again simply logically valid ) i H is valid in
any S-interpretation. A w H is called S-satisable (or simply satisable ) in
an interpretation A i there exists an assignment : V ! jAj such that H
has a designated truth degree for this assignment . And nally the w H
is called S-satisable (or again simply satisable ) i H is satisable in some
suitable S-interpretation.
For each w H of LS one has that its truth degree ValSA (H; ) under
some assignment only depends on the values which gives to the free
variables of H : one has ValSA (H; ) = ValSA (H; ) for all assignments such
that (x) = (x) for all variables x which occur free in H .9 As an immediate
corollary one has that sentences G of LS have a truth degree w.r.t. each
interpretation A because one has ValSA (G; ) = ValSA (G; ) for all assignments
; . Therefore one writes for sentences G simply
ValSA (G) instead of ValSA (G; ) :
(3.15)
These results open the way to introduce, similar to the approaches in (2.8),
(2.9), for each w H with its free variables among x1 ; : : : ; xn and each interpretation A an n-ary function w~H which maps the universe of discourse
A of A into the truth degree set W S . In this sense each w H determines a
many-valued predicate w.r.t. each interpretation A. Again the details are left
to the reader.
Well-formed formulas H; G are semantically equivalent in an interpretation A i one has ValSA (G; ) = ValSA (G; ) for all assignments ; : V ! A.
And the ws H; G are semantically equivalent (in an unrestricted sense) i
they are semantically equivalent in each interpretation for S.
9 The proof proceeds on the complexity of H just as for the propositional case in
the proof of Theorem 2.2.1. The details are left for the reader as an exercise.
3.3 The First-Order Situation
39
Both relations, the semantical equivalence in some xed interpretation,

and the (unrestricted) semantical equivalence are equivalence relations in the
set of all ws of LS .
As in the propositional case with Replacement Theorem 2.2, one has also
in the rst-order case suitable theorems on the substitutivity of equivalents
(in some xed interpretation as well as for the unrestricted case). Such a
theorem reads e.g. for the unrestricted case as follows:
Theorem 3.3.1 (Replacement Theorem). Let H 0; H 00 be semantically
equivalent ws with the same free variables, and let the w G be built up
from the w H by exchanging some of the subformulas of form H 0 of H by
the w H 00 . Then also H; G are semantically equivalent.
The proof is quite similar to the corresponding proof we have given for
Theorem 2.2, and hence shall not be given in detail. The fact that one assumes
that H 0 ; H 00 should have exactly the same free variables allows to exchange
H 0 and H 00 also without any restriction inside the scopes of quantications.
The result could be weakened, but this is not really important here.
A particularly important possibility to transform some w of a rst-order
language into another, semantically equivalent one, is the renaming of bound
variables. Because of Theorem 3.3.1 it suces to discuss this for quantied
ws. Denote for individual symbols a1 ; : : : ; ak by H [x1 =a1 xk =ak ] the w
which results from H by substituting simultaneously for the individual variables x1 ; : : : ; xk at all places of their free occurrence the individual symbols
a1 ; : : : ; ak , respectively. Then one has under the general assumption (2.22)
for the truth degree behavior of the quantiers, that for all variables y1 : : : ; yk
which do not occur in any one of the ws H1 ; : : : ; Hn the ws
(Qx1 ; : : : ; xk )(H1 ; : : : ; Hn ) and (Qy1 ; : : : ; yk )(G1 ; : : : ; Gn ) (3.16)
are semantically equivalent using the shorthand notations
Gi Hi [x1 =y1 xk =yk ]
for all i = 1; : : : ; n. The main reason for this result is that for each interpretation A and each A-assignment one can prove inductively on the complexity of the ws H1 ; : : : ; Hn that the functions (AQx ;:::;xk )(H ;:::;Hn ) and
(AQy ;:::;yk )(G ;:::;Gn) have the same range.
Thus, all the usual basic model theoretic notions are available also for
systems of many-valued rst-order logic. It is, however, sometimes advisable
to split classical notions into distinct ones for the many-valued case. To give
an idea how this can be done, we consider the notion of model.
In the case that a w H is valid in some interpretation A this interpretation is an S-model of H , or simply a model of H if there is no misunderstanding caused by this lack of reference to the underlying system. And the
model class ModS (H ) of H is the class of all S-models of H . As usual, an
interpretation A is an S-model of a set of sentences i each H 2 is valid
1
40
in A. By ModS ( ) the class of all S-models of shall be denoted. And a

w H is entailed by some set of ws of LS , or is a consequence of , i
ValSA (H; ) is a designated truth degree for each assignment for which also
ValSA (G; ) is a designated truth degree for all G 2 .
It is, however, suitable to consider even for any truth degree t 2 W S the
notions of t-model and of ( t)-model A of by imposing the condition
that H 2 always has truth degree t or always has truth degree t in A,
respectively. In the case of a singleton = fH g one of course speaks of a
t-model (( t)-model) of H .
One writes j=S H to indicate that H is entailed by (in S). Accordingly, j=S H denotes that H is logically valid in the system S, and A j=S H
(or: A j=S ) means that A is a model10 of H (or: of ). Then one has
j=S H i ; j=S H;
A j=S H i A 2 ModS (H ) ;
and hence also ModS ( ) = fA j A j=S g. For sentences H0 and sets 0 of
sentences of LS one has even
0 j=S H0 i ModS (0 ) ModS (H0 ) ;
H0 satisable i Mod(H0 ) 6= ;:
As done in the previous section for the propositional case, also in the rstorder case it is of interest to connect with each set of ws the set CnS ( )
of all the ws which are entailed by , its consequence hull :
CnS ( ) =def fH j j=S H g :
(3.17)
It is a routine matter to prove as in Theorem 3.2.2 that for all sets ; of
ws of S there hold true
(a) CnS ( ) ;
embedding property,
S
S
(b) ) Cn ( ) Cn () ;
monotonicity,
S
S
S
(c) Cn (Cn ( )) = Cn ( ) ;
idempotency.
Writing also (A; ) j=S H (or: (A; ) j=S ) i the truth degree ValS (H; )
is a designated one (or: i all the truth degrees ValS (H; ) for H 2 are
designated ones), then one has a further characterization of entailment:
j=S H i: (A; ) j=S ) (A; ) j=S H for all A and all .
3.4 Elementary Model Theory

Further on we shall have to consider for suitable systems S of rst-order
many-valued logic whole families of S-interpretations Ai , i 2 I . The index set
10 Unfortunately the use of the symbol j= is a little bit ambiguous. But this is the
usual way to use this symbol, and it usually causes no misunderstanding because
it always becomes clear from the context whether on the left hand side of j= one
refers to a (set of) w or to an interpretation.
41
I shall be not restricted to specic cases; it shall be allowed to be innite, even
uncountable. Our rst goal shall be to discuss a generalization of the direct

product of algebraic structures, suitable to be applied to S-interpretations in
the sense that the result shall again be an S-interpretation. We have to start
with the introduction of an auxiliary, but important notion.
Connected with the index set I we have its power set IPI = fX j X I g,
i.e. the set of all subsets of I . Any nonempty system F IPI , i.e. any set of
subsets of I is a lter on I i there hold true for all X; Y 2 IPI :
(FI 1) if X; Y 2 F then also X \ Y 2 F ,
(FI 2) if X 2 F and X Y then also Y 2 F .
The power set IPI itself is a lter on I . All the other lters on I dierent
from IPI are the proper lters on I . For each subset ; 6= Z I of I the set
F Z =def fX j Z X I g
(3.18)
is a proper lter on I , the principal lter of (or: generated by) Z . And the
inclusion-maximal proper lters are the ultralters.
Examples of ultralters are all the main lters which are generated by
singletons. Consider some k 2 I and let F (k) = F fkg the main lter of fkg.
Then one has k 2= Y for each Y I with Y 2= F (k) . Assume that F is another
lter such that F (k) F and let be X 2 F n F (k) . Then one has k 2= X
and hence k 2 I n X 2 F (k) . Therefore one has X; I n X 2 F and thus also
; = X \ (I n X ) 2 F , which means F = IPI , i.e. F is not a proper lter.
Hence F (k) is inclusion-maximal among proper lters, i.e. an ultralter.
It is easy to see that the intersection of any number of lters on I is again
a lter on I . On the other hand, the union of two lters F 1 ; F 2 on I is not
necessarily again a lter on I . Considering, however,
F 1 + F 2 =def fX I j 9 Y1 2 F 1 9 Y2 2 F 2 (Y1 \ Y2 X )g (3.19)
gives a lter F 1 + F 2 on I : obviously condition (FI 2) is satised; and for
any X; X 0 2 F 1 + F 2 there exist Y1 ; Y10 2 F 1 and Y2 ; Y20 2 F 2 such that
Y1 \ Y2 X and Y10 \ Y20 X 0, and hence such that Yi \ Yi0 2 F i for i = 1; 2,
which means that one has X \ X 0 2 F 1 + F 2 because of (Y1 \ Y10 ) \ (Y2 \ Y20 )
X \ X 0, i.e. also condition (FI 1) is satised.
It is also a routine matter to see that for each lter F on I and each
subset X0 of I the set
F = fX I j 9 Y 2 F (Y \ X0 X )g
(3.20)

is the smallest lter F F which extends F and contains X0 .
Both these constructions are particular cases of the following more general
result.
Lemma 3.4.1. For any set X IPI of subsets of I the set
hX i =def fY I j Y X1 \ : : : \ Xk for suitable X1 ; : : : ; Xk 2 X g
is the smallest lter which extends X .
42
Proof: One immediately sees that hX i is a lter. And it is also obvious

X hX i holds true. So let F be some lter with X F and let
Y 2 hX i. Then there exist X1 ; : : : ; Xk 2 X such that Y X1 \ : : : \ Xk .
Because of (FI 2) and of X1 \ : : : \ Xk 2 F one hence has Y 2 F . Therefore
hX i F follows from X F .
2
that
Thus each subset of IPI can be extended to some lter on I . But also
each proper lter on I can be extended to some ultralter on I .
Theorem 3.4.1. For each proper lter F on I there exists an ultralter U
on I which extends F .
Proof: Let FF be the class of all proper lters on I which extend F .
Then one has FF 6= ; because of F 2 FF . It is easy to see that the union of
each inclusion-chain of lters from FF is again a lter in FF . Thus FF satises
the assumptions of Zorn's lemma. Therefore FF has a maximal element U
which has to be an ultralter with property F U according to the choice
of FF .
2
For the applications of ultralters it is suitable to have also the following
characterization of ultralters at hand.
Theorem 3.4.2. A lter F on I is an ultralter on I i for each X I
either X 2 F or (I n X ) 2 F holds true, but not both.
Proof: Let F rst be an ultralter on I . In the case that one would have
Y0 2 F and (I n Y0 ) 2 F for some Y0 I , also ; = Y0 \ (I n Y0 ) 2 F would
hold true and hence F = IPI , i.e. F could not be an ultralter. And in the
case that there would be X0 2= F and (I n X0 ) 2= F for some X0 I , there
exists according to (3.18) and (3.19) a lter F which extends F and contains
X0 . Then one has ; 2= F because otherwise there would exist some X 2 F
such that X \ X0 = ;, i.e. such that X (I n X0 ) and hence (I n X0 ) 2 F ,
contradicting (I n X0 ) 2= F . But then F is a proper lter with F F , and
F cannot be an ultralter. Again a contradiction. Hence either X 2 F or
(I n X ) 2 F for each X I .
Now, conversely, let F be some lter such that either X 2 F or (I nX ) 2 F
holds true for each X I . Assume that F is not an ultralter. Then there
exists some proper lter U F and some X0 2 U n F . For this set X0 I
one has X0 2= F and hence (I n X0 ) 2 F , i.e. (I n X0 ) 2 U . But then one has
; = X0 \ (I n X0 ) 2 U , contradicting the fact that U is a proper lter. Thus
F has to be an ultralter.
2
Consider now a family Ai , i 2 I , of S-interpretations. From the universes
Ai one forms their cartesian product
[
X Ai = ff : I ! Ai j f (i) 2 Ai for all i 2 I g:
i2I
i2I
For each lter F on I one denes a binary relation F in Xi2I Ai via
f F g =def fi 2 I j f (i) = g(i)g 2 F

for all functions f; g 2 Xi2I Ai .
43
(3.21)
This relation F is an equivalence relation: it is re exive because I 2 F

holds true for each lter; it is obviously symmetric; and it is transitive because
of (FI 1). The relation F is the F -equivalence or lter equivalence for F .
Let be F -Prodi2I Ai the set of all equivalence classes [f ]F of Xi2I Ai
w.r.t. the lter equivalence (3.21) for F . We intend to build up from the
family Ai , i 2 I , of S-interpretations a new S-interpretation B which has
F -Prodi2I Ai as its universe of discourse. For notational convenience we shall
consider a function f 2 Xi2I Ai , with f (i) = ai for each i 2 I , as the family
of its values, and indicate this fact as f = (ai )i2I .
For any individual constant a of LS and the function (aAi )i2I , which maps
each i 2 I to the interpretation of the constant a in the interpretation Ai , let
the equivalence class be
aB = [(aAi )i2I ]F :
(3.22)
S
For any n-ary predicate symbol P of LS , each truth degree t 2 W , each
assignment : V ! F -Prodi2I Ai , and all individual symbols a1 ; : : : ; an of
LS let
ValSB (P (a1 ; : : : ; an ); ) = t =def
fi 2 I j ValSAi (P (a1 ; : : : ; an ); i ) = tg 2 F
(3.23)
with additionally for each individual variable x 2 V
(x) = [(i (x))i2I ]F
for suitable assignments i : V ! Ai .
It is immediately clear that (3.22) is a suitable denition of aB . For the
acceptability of (3.23), however, it has to be shown that for each n-ary predicate symbol P and each assignment the truth degree ValSB (P (a1 ; : : : ; an ); )
is uniquely determined.
It is obvious that for each assignment : V ! F -Prodi2I Ai its value
(x) for any variable x 2 V is an equivalence class [gx ]F of some function
gx 2 Xi2I Ai . Using the above mentioned notational convention one thus has
(x) = [(gx (i))i2I ]F . So one can dene for each i 2 I a function i : V ! Ai
by i (x) = gx(i) for each x 2 V . This gives immediately (x) = [(i (x))i2I ]F
as needed for (3.23).
The functions gx are not uniquely determined by (x), however, because
they are only representatives of equivalence classes. Hence one has to consider
also other representatives fx such that (x) = [gx]F = [fx]F holds true for
all x 2 V , as well as gx F fx. Now functions i : V ! Ai can be formed out
of the functions fx as before the functions i : V ! Ai out of the functions
gx . By the choice of gx ; fx one has fi 2 I j gx (i) = fx(i)g 2 F for each x 2 V
and hence also
fi 2 I j i (x) = i (x)g 2 F :
44
For each individual constant a one has, using the notation introduced in
(2.19), i : a = aAi = i : a for all i 2 I . Hence one has
fi 2 I j i : a = i : ag 2 F
(3.24)
for each individual symbol a of LS .
And P Ai (i : a1 ; : : : ; i : an ) = t results for each n-ary predicate symbol
P from the assumptions that P Ai (i : a1 ; : : : ; i : an ) = t holds true and
i : ak = i : ak for each k = 1; : : : ; n. Therefore one has
fi 2 I j ValSAi (P (a1 ; : : : ; an ); i ) = tg \
fi 2 I j ValS
n
\
k=1
fi 2 I j i : a = i : ag
Ai (P (a1 ; : : : ; an ); i ) = tg
because of (2.18), and hence by (3.24) also

fi 2 I j ValSAi (P (a1 ; : : : ; an ); i ) = tg 2 F
) fi 2 I j ValSAi (P (a1 ; : : : ; an ); i ) = tg 2 F
for each t 2 W S . By symmetry one gets also the converse of this implication.
So one has shown for the deniendum of (3.23), i.e. for the left hand side of
this denition, its independence from the choice of the representatives of the
equivalence classes (x).
For the suitability of denition (3.23) it remains to show that there do not
exist dierent truth degrees s; t 2 W S such that ValSB (P (a1 ; : : : ; an ); ) = t
together with ValSB (P (a1 ; : : : ; an); ) = s results from (3.23), and that there
always exists some t 2 W S such that fi 2 I j ValSB (P (a1 ; : : : ; an ); ) = tg 2 F
holds true. Let
Yt = fi 2 I j ValSAi (P (a1 ; : : : ; an ); i ) = tg
for t 2 W S . Then one has for any s; t 2 W S with s 6= t immediately Ys \Yt = ;.
Thus having Ys 2 F and Yt 2 F for s 6= t would mean to have ; 2 F , and
hence F = IPI . Therefore we suppose that the lter F in denition (3.23)
has to be a proper lter. If Yt 2= F holds true for each t 2 W S , and F is an
ultralter, then one would haveS(I n Yt ) 2 F for each t 2 WSS according to
Theorem 3.4.2, andShence also t6=s2W S Ys 2 F because of r2W S Yr = I ,
and thus (I n Yt ) t6=s2W S Ys . Assuming furthermore the niteness of W S ,
one has

\ [
;=
Ys 2 F
t2W S t6=s2W S
contradicting the fact that F should be an ultralter.

Therefore (3.23) is a suitable denition in the case that F is an ultralter
and that S is a system of rst-order many-valued logic with only nitely many
truth degrees.
For innitely many-valued systems S with truth degree set W S = W1 ,
and also with truth degree set W S = W0 , denition (3.23) has slightly to be
45
changed. This change has to refer to the notion of neighborhood of a truth

degree. A set V W S is a neighborhood of t 2 W S i there exists a real
number " > 0 such that fs 2 W S j t ; " < s < t + " g V holds true. Then
the substitute for (3.23) becomes
ValSB (P (a1 ; : : : ; an ); ) = t =def
fi 2 I j ValSAi (P (a1 ; : : : ; an ); i ) 2 V g 2 F
(3.25)
for each neighborhood V of t :
As previously done w.r.t. the denition (3.23), also now the independence
of this denition from the choice of the representatives of the equivalence
classes (x) is proved for each x 2 V . Because there exist for any two distinct
truth degrees s; t always neighborhoods Vs of s and Vt of t such that Vs \Vt = ;
holds true, denition (3.25) determines w.r.t. some proper lter F uniquely
the truth degree ValSB (P (a1 ; : : : ; an ); ), as long as (3.25) determines such
a truth degree at all. And one has to suppose that F is an ultralter to
secure the existence of this truth degree. Otherwise there would exist for
each t 2 W S a neighborhood Vt of t such that
Yt = fi 2 I j ValSAi (P (a1 ; : : : ; an ); i ) 2 Vt g 2= F :
The family of all these neighborhoods Vt , t 2 W S , would then be an open
covering of W S and had to have, according to the Heine-Borel theorem,11
a nite subfamily which already had to be an open covering of W S . Therefore
there would exist truth degrees t1 ; : : : ; tk 2 W S such that Ytj 2= F for each
S
j = 1; : : : ; k and also kj=1 Ytj = I . And from this situation the contradiction
; 2 F would follow.
So we have for each one of the truth degree sets Wm for m 2 as well as
for W1 ; W0 the situation that for each family Ai , i 2 I , of S-interpretations
and each ultralter F on I by the denitions (3.22) and (3.23), resp. (3.22)
and (3.25), a further S-interpretation is determined which has F -Prodi2I Ai
as its universe of discourse. This S-interpretation is the ultraproduct of the
family Ai , i 2 I , w.r.t. the ultralter F .
For rst-order systems S with only nitely many truth degrees denitions
(3.23) and (3.25) coincide. This is easily seen: one has only to have in mind
that in the case W S = Wm e.g. the choice " = 21m says that each t 2 Wm has
Vt = ftg as a neighborhood. And furthermore one gets by (3.22) and (3.25)
a suitable denition of an ultraproduct for all those truth degree sets W S
which are compact Hausdorff spaces,12 cf. [95]. We are not going, however,
to treat such a general situation here.
11 This theorem of classical analysis states for any closed and bounded set
D of
reals, that each covering of D which consists of open sets has already a nite
subcovering of D.
12 A Hausdorff space is a particular type of topological space, characterized by the
condition that any two dierent points of this space have disjoint neighborhoods.
46
For the ultraproduct B of a family Ai , i 2 I , of S-interpretations one has

according to (3.23) the equivalence
ValSB (P (a1 ; : : : ; an ); ) = t
, fi 2 I j ValSAi (P (a1 ; : : : ; an ); i ) = tg 2 F
which can equivalently be written as

i 2 I j ValSAi (P (a1 ; : : : ; an ); i ) = ValSB (P (a1 ; : : : ; an ); ) 2 F :
More generally, one has as an immediate corollary for each w H , any t 2 W S ,
and all assignments : V ! F -Prodi2I Ai the equivalence of the condition
ValSB (H; ) = t , fi 2 I j ValSAi (H; i ) = tg 2 F
(3.26)
and of the condition
fi 2 I j ValSAi (H; i ) = ValSB (H; )g 2 F :
(3.27)
Theorem 3.4.3. Let S be a nitely many-valued rst-order system and B
the ultraproduct of the family Ai , i 2 I , of S-interpretations w.r.t. an ultralter F on I . Then one has
fi 2 I j ValSAi (H; i ) = ValSB (H; )g 2 F
for each assignment : V ! F -Prodi2I Ai , each of the corresponding assignments i : V ! Ai w.r.t. the \factors" Ai determined by
(x) = [(i (x))i2I ]F for each x 2 V ;
and for each w H of LS .
Proof: The proof proceeds by induction on the complexity of the w H .
Suppose that the assumptions of this theorem are satised.
For atomic ws H then the claim results immediately from denition
(3.23) and the previous remarks. So let H be any w and suppose that (3.27)
holds true for all of its subformulas.
In the case that H is of the form '(H1 ; : : : ; Hn ) for some n-ary connective
' 2 J S one has
ValSB (H; ) = verS' (ValSB (H1 ; ); : : : ; ValSB (Hn ; )) ;
ValSAi (H; i ) = verS' (ValSAi (H1 ; ); : : : ; ValSAi (Hn ; ))
for each i 2 I according to (2.20). Furthermore one has
fi 2 I j ValSAi (Hj ; i ) = ValSB (Hj ; )g 2 F
(3.28)
for each j = 1; : : : ; n according to the hypothesis for the subformulas of H .
Thus one gets from (FI 1) also
n
\
j =1
fi 2 I j ValSAi (Hj ; i ) = ValSB (Hj ; )g 2 F :
(3.29)
47
And from the obvious relationship

n
\
j =1
fi 2 I j ValSAi (Hj ; i ) = ValSB (Hj ; )g

fi 2 I j ValSAi (H; i ) = ValSB (H; )g
one gets (3.27) for the present choice of H by (3.29) and (FI 2).
In the case that H is of the form Q(x1 ; : : : ; xk )(H1 ; : : : ; Hn ) for some
(k; n)-ary quantier Q 2 QS one has, according to (2.22), to consider for each
assignment : V ! F -Prodi2I Ai its associated function HB : jBjk ! (W S )n
with B the ultraproduct of the family Ai , i 2 I , of S-interpretations. Then
one has
ValSB (H; ) = VerSQ(HB ) ;
(3.30)
S
S
A
i
ValAi (H; i ) = VerQ((i )H )
(3.31)
for each i 2 I . Here again one has
;

(i )AHi (b1i ; : : : ; bki ) = ValSAi (H1 ; i ); : : : ; ValSAi (Hn ; i )
for all b1i ; : : : ; bki 2 Ai and the assignments i = i [x1 =b1i ; : : : ; xk =bki ], and
also
;

HB (c1 ; : : : ; ck ) = ValSB (H1 ; ); : : : ; ValSB (Hn ; )
for all cj = [(bji )i2I ]F 2 jBj with j = 1; : : : ; k, and for the \common" assignment = [x1 =c1 ; : : : ; xk =ck ]. Thus the choice of and all the i , i 2 I ,
yields
(x) = [(i (x))i2I ]F for each x 2 V ,
and therefore according to the hypothesis (3.28) for each j = 1; : : : ; n also
fi 2 I j ValSAi (Hj ; i ) = ValSB (Hj ; )g 2 F :
(3.32)
Because the values of the generalized truth degree function VerSQ depend only
on the ranges of the arguments one has also
n
\
j =1
fi 2 I j ValSAi (Hj ; i ) = ValSB (Hj ; )g

fi 2 I j VerSQ ((i )AHi ) = VerSQ (HB )g ;
and hence gets from (3.32) and (FI 2) according to (3.30), (3.31)
fi 2 I j ValSAi (H; i ) = ValSB (H; )g 2 F ;
and thus (3.27) for the actual choice of H . So the proof is nished.
2
Theorem 3.4.3 can also be extended to innitely many-valued rst-order
systems S under the additional assumptions that the truth degree functions
48
which correspond to the connectives and quantiers of S have suitable continuity properties. This was done in [95] for arbitrary compact Hausdorff
spaces as truth degree sets, i.e. particularly for W S = W1 and W S = W0 .
We will not discuss the details here; the interested reader should consult [95].
Corollary 3.4.1. For nitely many-valued rst-order systems S and ultraproducts B of families Ai , i 2 I , of S-interpretations w.r.t. an ultralter F
on I one has
(B; ) j=S H , fi 2 I j (Ai ; i ) j=S H g 2 F
for each w H of LS and each assignment : V ! jBj.
Proof: Suppose rst (B; ) j=S H . Then there exists a designated truth
degree t0 2 DS such that ValSB (H; ) = t0 . Hence one has because of Theorem
3.4.3 and of (3.26)
fi 2 I j ValSAi (H; i ) = t0 g 2 F :
From (FI 2) one then gets immediately
fi 2 I j ValSAi (H; i ) = t0 g fi 2 I j (Ai ; i ) j=S H g 2 F :
For the converse implication now suppose fi 2 I j (Ai ; i ) j=S H g 2 F ,
and let be DS = fs1 ; : : : ; sk g. Then it holds true
k
[
fi 2 I j ValSAi (H; i ) = sl g 2 F :
l=1
As in the proof of the soundness of denition (3.23) it follows immediately

from this fact that at least one of the sets whose union belongs to F has
itself to be an element of F , because otherwise ; 2 F would result. Therefore
there exists some t 2 DS such that
fi 2 I j ValSAi (H; i ) = tg 2 F ;
and hence ValSB (H; ) = t according to Theorem 3.4.3. But then one also has
(B; ) j=S H according to the denition of the satisfaction relation.
2
The ultraproduct construction is the essential tool to prove a version of
the compactness theorem also for rst-order systems of many-valued logic.
An additional notion is needed, however, to give the compactness result.
A graded set of sentences of LS is a set of ordered pairs (H; t) of
sentences H of LS and truth degrees t 2 W S which is a function (in the
set theoretical sense), i.e. which contains for each sentence H at most one
ordered pair (H; s) with rst component H . And a model of a graded set
of sentences is such an S-interpretation A which for each ordered pair (H; t)
of is a ( t)-model of H . Such a model of a graded set hence gives to
each w H , which is the rst component of an ordered pair from , as truth
degree the second component of this ordered pair. We will continue to write
A j= also for the case that A is a model of a graded set of sentences.
49
Theorem 3.4.4 (Compactness Theorem). Suppose that S is some sys-
tem of nitely many-valued rst-order logic. Then a graded set of sentences

of S has a model i each nite subset of has a model.
Proof: Obviously, if a graded set of sentences of S has a model then
also each nite subset of has a model. So it remains to be shown that a
graded set of sentences of S has a model under the assumption that each
nite subset of has a model.
Suppose that for each nite subset of the graded set of S-sentences the
S-interpretation A is a model of . Let I be the set of all nite subsets of .
And for each 2 I let be + = f 2 I j g. And let Z = f+ j 2 Ig.
Then one has Z IP I and can build up the lter hZ i. This lter hZ i is a
proper lter, i.e. ; 2= hZ i holds true because one has for any family of nitely
many +1 ; : : : ; +k 2 Z
1 [ [ k 2 +1 \ \ +k 6= ; :
Finally let F be an ultralter on I which extends hZ i, and which exists
according to Theorem 3.4.1.
Let B be the ultraproduct of the family A , 2 I , of S-interpretations
w.r.t. the ultralter F . We shall prove that B is a model of . Therefore let
(H; t) 2 and consider (H;t) = f(H; t)g 2 I . It is +(H;t) 2 F and
+(H;t) = f 2 I j (H;t) g f 2 I j A ( t)-model of H g :

Thus one has immediately for each assignment : V ! jBj
f 2 I j ValSA (H; ) tg 2 F ;
(3.33)
and hence because of Theorem 3.4.3 and the equivalence of (3.27) and (3.26)
also ValSB (H; ) t. Therefore B is a ( t)-model of H and hence a model
of .
2
For this Compactness Theorem the assumption that S should have only
nitely many truth degrees is needed in this proof only for the reference to
Theorem 3.4.3 or to Corollary 3.4.1. Both results, however, can be generalized to suitable innite sets of truth degrees. In the same way also the
Compactness Theorem holds true for the same class of innite truth degree
sets, cf. [95].
Corollary 3.4.2. Let S be a system of nitely many-valued rst-order logic,
t 2 W S , and a set of S-sentences. Then one has
(a) has a t-model i each nite subset of has a t-model;
(b) has a model i each nite subset of has a model.
Proof: In the case (a) start from , use the graded set = f(H; t) j H 2
g and apply the previous theorem to it. In the case (b) one has to modify in
50
a similar way into a graded set of ws with only designated truth degrees
as \grades".
2
As in the propositional case it is often welcome to have also a slightly
more general form of the compactness theorem available.
Theorem 3.4.5. Let S be a system of nitely many-valued rst-order logic
and and graded sets of sentences of S. Then has a model which at
the same time is a model of i each nite subset of has a model which
at the same time is a model of .
Proof: The previous proof of Theorem 3.4.4 has to be changed only insofar as one has to consider for each nite only such an S-interpretation
A which is a model of and at the same time a model of .
2
This slightly generalized form of the compactness theorem enables a simple proof of the fact that there exists for each S-interpretation A w.r.t. some
nitely many-valued rst-order system S an S-interpretation B which has a
larger universe of discourse and is still model of the same sets of ws as A.
To state this result we introduce a further model theoretic notion.
Let A; B be S-interpretations. Then A is an elementary substructure of
B i there hold true:
(ES1) jAj jBj;
(ES2) ValSA (H; ) = ValSB (H; )
for each w H of LS and each
assignment : V ! jAj.
The fact that A is an elementary substructure of B is denoted: A B. If A
is an elementary substructure of B one can also say that B is an elementary
extension of A. And by the cardinality of the language of S one understands
the cardinality of the set of all symbols of LS : this is always an innite
cardinality because the set V of individual variables is an innite set.
Theorem 3.4.6. Let S be a system of nitely many-valued rst-order logic
and the cardinality of the language of S. Let also A be an S-interpretation
with an innite universe of discourse. Then there exists for each cardinality
which is not smaller as and as card jAj an elementary extension B of A
such that card jBj.
Proof: In the case = card jAj choose simply B = A. So suppose now
that > card jAj holds. Let = card jAj. By extending the language of S the
system S shall be extended rst to a system S of rst-order many-valued
logic and then to a further system S . The system S is built up from S
by adding to the alphabet of S a set fa j < g of pairwise dierent new
individual constants and a new binary predicate symbol U . And the system
S shall result from S by adding to its alphabet another set fc j < g of
pairwise dierent new individual constants.
51
The S-interpretation A shall be extended to an S -interpretation A with

the universe of discourse jAj in such a way that the interpretation of the new
individual constants is provided by a bijection from fa j < g onto jAj, and
that the interpretation of U is given by
n
ValSA (U (a1 ; a2 ); ) =def 1; if : a1 6= : a2
0 otherwise
for each assignment : V ! jAj and all a1 ; a2 2 jAj. Furthermore let be
the set of all sentences of the language of S , let sG = ValSA (G) for each
G 2 , and let = f(G; sG ) j G 2 g.
Consider the graded set of sentences = f(U (c ; c ); 1) j < < g
of the system S . Each nite subset has a model with universe of
discourse jAj which also is a model of : start from the S -interpretation
A and map the nitely many individual constants c which occur in ws
from one-one to elements of jAj = jA j. Interpret all the other individual
constants c in the same way as the constant aa0 . This gives a model of .
According to Theorem 3.4.5 now has a model B which at the same time
is a model of . Because in each model of the individual constants c
denote for < pairwise dierent elements of the universe of discourse, one
has card jB j.
Starting from the S -interpretation B one can nd an S-interpretation B
by simply \forgetting" the interpretations which B makes for the predicate
symbol U and all the individual constants which had been added to the
language LS . However, before one goes on to forget, one should embed jAj
into jB j. And this can simply be done by exchanging for each individual
constant a , < , the element of jB j denoted by a with the element of
jAj denoted by a . In the same way all the B -interpretations of the other
symbols of the language of S have to be adapted to this embedding of jAj
into jB j.
In summary one thus gets card jB j = card jBj and jAj jBj. And
what remains to be shown is that one has ValSA (H; ) = ValSB (H; ) for all
assignments : V ! jAj and all ws H of LS . To reach this goal we build
from H and another w H of LS by substituting for each free individual
variables x of H (at all the places where x occurs free in H ) the individual
constant a , < suitably chosen, which denotes in A the element (x).
This gives ValSA (H; ) = ValSA (H ; ) as well as ValSB (H; ) = ValSB (H ; ).
And because of H 2 one has also ValSA (H ; ) = ValSB (H ; ), which
nishes the proof.
2
Besides the possibility to have elementary extensions of predetermined
cardinality for given interpretations, i.e. besides the possibility to \enlarge"
given interpretations, also a converse method which reduces the size of given
interpretations is of interest. Therefore we ask also for the elementary substructures of predetermined smaller size.
52
Theorem 3.4.7. Let S be a system of nitely many-valued rst-order logic,
the cardinality of the language of S, and the cardinality of the set W S

of truth degrees. Let also A be an S-interpretation with an innite universe
of discourse and C jAj. Then there exists for each cardinality card jAj
with additionally ; and card C an elementary substructure B of A
such that = card jBj.
Proof: In the case = card jAj choose simply B = A. So suppose now
that < card jAj holds. Let v be a well-ordering of the set of all nite
sequences of elements of jAj, with the elements of jAj identied with the
sequences of length one (and the additional property that shorter sequences
shall also be v-smaller). Let B0 jAj such that C B0 and = card jB0 j.
Assume furthermore that B0 contains all the elements of jAj which are the
A-denotations of individual constants of LS . Starting from B0 , an increasing
sequence
B0 B1 B2 : : : of subsets of jAj is formed such that their union
B = S1n=1 Bn is the universe of discourse of the elementary substructure B.
For each n 0 let Bn+1 be the subset of jAj which contains all the
elements of Bn, and which contains additionally for each k-tuple (H1 ; : : : ; Hk )
of ws of LS with free variables among x1 ; : : : ; xl , and for each (l ; r)-tuple
(br+1 ; : : : ; bl ) of elements of Bn and each k-tuple s of truth degrees, which
can be realised as
;
s = ValSA (H1 ; [xr+1 =br+1; : : : ; xl =bl]); : : : ;

ValSA (H1 ; [xr+1 =br+1; : : : ; xl =bl])
for some assignment : V ! jAj, all terms b1 ; : : : ; br of the v-smallest
sequence (b1 ; : : : ; br ) of elements of jAj for which one has
;

s = ValSA (H1 ; [x1 =b1; : : : ; xl =bl ]); : : : ; ValSA (H1 ; [x1 =b1; : : : ; xl =bl])
for some (and hence for each) assignment : V ! jAj. Here k; r are arbitrary
integers.
Because of @0 the cardinality of the alphabet of S is also the cardinality of the set of ws of LS . Hence from = card Bn and the construction
of Bn+1 one gets immediately card Bn+1 , and also
card Bn+1 +
1
X
r;l;k=1
rl
r k l;r k + @0 + = :
Hence one has = card Bn+1 . So one even has = card Bn for all n 0, and
therefore also
card B @0 = :
Now B shall be chosen as the restriction of A to B, i.e. such that jBj = B,
such that P B = P A \ Bn for each n-ary predicate symbol P , and such that
aB = aA for each individual constant a. Because of card B = and jBj jAj
it remains to be shown that one has
53
ValSB (H; ) = ValSA (H; )

(3.34)
for each w H of LS and each assignment : V ! jBj. And this is shown by
induction on the complexity of H .
For atomic ws H obviously (3.34) holds by construction of B. So let
(3.34) hold true for all subformulas of some compound formula H . In the
case that one has H '(H1 ; : : : ; Hn ) for some n-ary connective ' 2 J S , one
gets (3.34) immediately from (2.20) and this assumption. Hence nally let
H (Qx1 ; : : : ; xk )(H1 ; : : : ; Hn ) for some (k; n)-ary quantier of S. According
to (2.22) then one has
ValSB (H; ) = VerSQ (HB )
for all : V ! jBj. Let the free variables of H occur among xk+1 ; : : : ; xl .
Then the free variables of all the ws H1 ; : : : ; Hn occur among x1 ; : : : ; xl .
Suppose that r is chosen such that (xk+1 ); : : : ; (xl ) 2 Br . By the choice
of Br+1 each element of the range of HA is already a value of HB for some
argument from Br+1. Hence one has
ValSB (H; ) = VerSQ (HB ) = VerSQ (HB ) = ValSA (H; ) ;
because the ranges of HB and HA coincide. Therefore (3.34) is proved and B
is an elementary substructure of A.
2
From these two results one immediately gets two further interesting results which in their version for classical logic usually are referred to as the
downward and upward Lo wenheim-Skolem theorems.
Theorem 3.4.8. Let S be a system of rst-order many-valued logic, the

cardinality of the language of S, and = card W S . Suppose that is a graded
set of ws of LS which has an innite model of cardinality . Then there hold
true:
(a) has models of cardinality for each cardinal such that ; .
(b) If S has only nitely many truth degrees then has models of cardinality
for each cardinal such that .
Proof: Suppose that all the assumptions are satised and that A is an
innite model of . If is a cardinal with property ; then there
exists, according to Theorem 3.4.7, an elementary substructure B A of
cardinality . By denition of , B has to be a model of . So one has (a).
Assume furthermore that W S is nite. Then there exists according to
Theorem 3.4.6 an elementary extension A B such that card B. In the
case = card B, (b) is proved because B has to be a model of too. And
in the case < card B one proceeds as in the proof of part (a) to get for the
model B of an elementary substructure B B of cardinality . Then
also B is a model of because of B B. Hence also (b) is proved. 2
54
4. Outline of the History of Many-Valued

Logic
Many-valued logic, as a separate part of logic, was created by the works of

J. Lukasiewicz [334] and E.L. Post [426] in the beginning of the 1920s.
Admittedly, both authors were not the rst ones who did not assume the
principle of bivalence, but earlier attempts to do logic without this principle
of bivalence did not prove to be in uential1 .
The prehistory of many-valued logic, however, may be traced back to
Aristotle2 who e.g. in his De Interpretatione, chapt. 9, discussed the problem of future contingents. This is the problem to determine \today" the
truth value of a proposition which asserts some future event. This problem
was stimulating even for J. Lukasiewicz [335], and it is closely tied with
the philosophical problems of determinism and the understanding of modalities. The link is provided by the interpretation that the classication of some
future event as (actually) \possible" or \undetermined" may well be seen as
the acceptance of a third \truth value" besides > and ?. Surely, this reading
is not the necessary one. Nevertheless, it may be an interesting one, cf. e.g.
[429]. And, interestingly, the ancient philosophical school of Epicureans which
tended toward indeterminism rejected the principle of bivalence, whereas the
school of the Stoics did accept it { and strongly advocated determinism.3
The same problem of contingentia futura was also the source for several
extended discussions during the Middle Ages, cf. e.g. [34, 310, 336, 364, 401,
448], without getting resolved. And in the phase of the general revival of
investigations into the eld of logic during the second half of the 19th century
the idea of neglecting { at least partly { the principle of bivalence appeared
(partly without clear mentioning of this fact) with H. McColl [355], cf. also
[332], and Ch.S. Peirce, cf. [418, vol. 4] as well as [160, 546], but also to A.
Meinong, whose work even was in uential on J. Lukasiewicz, cf. [506].
However, these authors did not really develop systems of many-valued
logic. They often mentioned only the idea that propositions may have other
1 Even the previous paper [333] of J. Lukasiewicz which also admitted generalized
truth \degrees" besides the traditional truth values >; ? did not in uence the
development of logic toward many-valued logic in any perceivable manner.
2 The interested reader may consult e.g. [335, 338, 448, 413].
3 Actually, the topic of the relationship between many-valuedness and determinism
is discussed only occasionally, e.g. in [543].
56
4. Outline of the History of Many-Valued Logic
values besides \true" and \false". And sometimes their systems are also not
truth functional, cf. [507].
Sometimes also the \non-aristotelian" systems which the Russian logician
N.A. Vasilev [556, 557] introduced have been considered as forerunners of
systems of many-valued logic. It seems, however, in accordance with the work
of A.I. Arruda [8], more appropriate to consider them as early approaches
toward systems of paraconsistent logics, i.e. as systems in which one may
have \contradictory" theorems of the form H; :H without having from this
all w deducible (which is the situation of classical logic).
The real starting phase of many-valued logic was the time interval from
1920 till about 1930, and the main force of development was the Polish school
of logic under J. Lukasiewicz. The papers [335, 339] as well as the in uential textbook [330] explain the core ideas as well as the background of
philosophical ideas and the main technical results proved up to this time.4
They also stimulated further research into the topic.
In [335], as already presented in [334], Lukasiewicz intended to give a
modal reading to his many-valued propositional logic, claiming that only the
three-valued and the innite valued case (with the set of all rationals between
0 and 1 as truth degree set) are really of interest for applications. In [339]
however, all nitely many-valued propositional systems and the just mentioned innitely many-valued one are discussed, always based on a negation
and an implication connective as primitive ones characterized semantically
by their truth degree functions.
Basic theoretical results for systems of many-valued logics which followed
this initial phase of \Polish" many-valued logic have been e.g.
M. Wajsberg's [560] axiomatization of the three-valued (propositional)
system L3 of Lukasiewicz, i.e. the propositional system with Lukasiewicz's implication and negation connectives as primitive connectives,
the extension of Lukasiewicz's system L3 to a functionally complete one
and its axiomatization by J. Slupecki [509],
the work of K. Go del [199] and S. Jaskowski [285] which claried the
mutual relations of intuitionistic and many-valued logic in the sense that
it was proved that there does not exist a single (propositional) manyvalued system whose set of logically valid formulas coincides with the set
of logically valid formulas of intuitionistic (propositional) logic,
the application of systems of three-valued logic to the problems of logical
antinomies by D.A. Bocvar [66, 67] with the third truth value read as
\senseless",
4 As a side remark it has to be mentioned that P. Bernays [52] used more than
the usual two truth values of classical logic to study independence problems for
systems of axioms for systems of classical propositional calculus. But in his case
these multiple values were only formal tools for his unprovability results.
57
the application of systems of three-valued logic to problems of partially
dened functions by S.C. Kleene [304, 305] with the third truth value
read as \undened".
Furthermore, during the 1940s basic approaches were generalized, and essential results were proved by J.B. Rosser and A.R. Turquette in a series
of papers. These authors collected later on most of this material in their
monograph [465] which, besides the Lukasiewicz papers of 1930, was the
standard reference for years. In this monograph strong emphasis is on the development of \Hilbert type" axiomatic calculi for systems of many-valued
logic.
Since that time a lot of further work has been done. One of the important
problems in this work has been the search for applications as indicated in
the basic monograph [465], and again e.g. stressed in [494]. Even today this
problem is not completely solved, and by its very nature may never be completely solved, but the approaches toward it have produced a lot of interesting
applications { in as dierent elds as, e.g., computer science and linguistic
semantics. Some of the more important applications shall be explained later
on in Part IV.
In a comparable manner also the stock of theoretical results has grown.
We will not start to discuss details of these developments at this point, instead
a lot of such results shall be presented later on { but without intending or
being able to reach any sucient kind of completeness in view of the wealth
of results. And for applications, the topics which have been treated cover
applications inside logic like the considerations on independence matters in
Chapter 23, or the problems of the possibility or impossibility of adequate
interpretations or approximations of dierent logical systems by systems of
many-valued logic e.g. for modal (Chapter 21) and for intuitionistic logic
(Chapter 22), as well as the topic of graded notions of entailment (Chapter
19). And they also cover applications outside the eld of logic, like e.g. applications to the theory of fuzzy sets (cf. Chapter 18), or some generalizations
of traditional switching theory as based on classical logic toward a switching
theory based on many-valued logic as discussed e.g. in [86, 451, 289, 391, 143].
But such applications also cover applications of logical calculi from the eld
of many-valued logic inside mathematics like a possible reduction of problems from 0-1-optimisation to problems in switching theory (with don't-care
conditions) via some three-valued calculus in [30], and allow on the other
hand to apply methods from integer programming to many-valued logic, as
explained e.g. in [238, 239]. And such applications cover also problems of the
generation of functions via superposition of basic functions which correspond
to the problem of functional completeness in many-valued logic (cf. Chapter
8.1), which have been treated mainly by algebraic means as explained e.g. in
[283, 284, 424, 461].
Another eld which connects many-valued logic with mathematics and
mathematical applications is the study of particular algebraic structures
58
which are strongly tied with particular systems of many-valued logic like
e.g. MV-algebras (cf. Section 9.2) and Post algebras (cf. Chapter 12) which
form an bridge to connect many-valued logic with even further areas of applications.
From the point of view of computer science the syntactic aspects of suitable logical calculi for systems of many-valued logic have some priority over
the semantic aspects. Accordingly proof theoretic considerations have grown
in recent times. Besides the Hilbert type axiomatic systems of [465], already
in 1955 Schro ter [485] developed a Gentzen type sequent calculus. And
tableau calculi have been studied since the mid-1970s e.g. in [516, 522] and
became more popular only recently, essentially through [83, 85] and [237].
These tableau calculi are particularly important for approaches toward automated deduction systems, as explained e.g. in [21, 237].
Besides these relationships there are further approaches of how to exploit
ideas from the eld of many-valued logic, particularly interpretations of truth
degrees, which relate many-valued logic to as dierent elds of applications
as the treatment of presuppositions in natural language semantics (cf. Chapter 20) or as the discussion of problems in diagnostics e.g. in pedagogical
situations as in [316, 515].
Another bridge to real world applications of (ideas from) many-valued
logic is provided by the theory of fuzzy sets with their graded notion of membership which is a suitable tool for modeling vague notions (cf. Chapter 18).
And this reading of the truth degrees as membership degrees seems to be
more appropriate than probabilistic readings which had been favored in the
earlier years of many-valued logic e.g. by Lukasiewicz [333], Reichenbach
[445], or Zawirski [592]. Moreover, results by Gaines [179] indicate that
the probabilistic understanding of (truth) degrees on the one hand, and the
\standard" many-valued logic as based on the principle of compositionality
on the other hand may be dierent generalizations of a common core. The
interrelationships between both circles of ideas are, however, not clearly understood up to now. And also the \true" relationship between many-valued
logic and quantum logic, as supported e.g. by Reichenbach [446], still is
an essentially open problem, despite the fact that quantum logic usually is
treated as a logic with orthomodular lattices as truth degree structures.
The intuition on which the notion of fuzzy set rests, i.e. the grading of
membership in sets, and hence also the grading of the predicate of \being a
member of some (fuzzy) set" was of in uence for the development of manyvalued logic too. The crucial link is the idea that one could extend the usual
notions of entailment and derivability, which refer to some (usual) set of
premises, in such a way that they allow also for fuzzy sets of premises. These
extensions of the usual notions of entailment and derivability work most
59
naturally within the realm of many-valued logic. And they lead to what is
usually called fuzzy logic (in the narrower sense5 ).
Fuzzy logic (in the narrower sense) was rst studied by the Czech mathematician J. Pavelka in [414]. He was concerned with fuzzy propositional
logic and intended to be as general as possible regarding the truth degree sets.
He presented axiomatizations for the fuzzy logics and succeeded, however,
with the proof of a completeness theorem only for the case that this truth
degree set is (isomorphic to) the truth degree set of one of the Lukasiewicz
systems.
The extension to the rst-order case was mainly done in papers of another
Czech mathematician V. Novak and is reported e.g. in [396, 397]. He used
truth degree constants for all the reals of the unit interval. A quite natural
link between many-valued logic and fuzzy logic (in the narrower sense) is
provided by the problem of partial soundness of inference rules. It was rst
discussed by this author in [220].
An important restriction to \rational fuzzy logic", i.e. to fuzzy logic (in the
narrower sense) with truth degree constants only for the rational numbers of
the truth degree set [0; 1] together with simplications of the axiomatization
and with interesting further results on this fuzzy logic was more recently
given by a third Czech mathematician P. Hajek, e.g. in [243, 246], cf. also
Chapter 19.
Presentations of many-valued logic as a separate topic { either as monographs or as textbooks { for long did exist only scarcely. The oldest monograph [465] on our topic by J.B. Rosser and A.R. Turquette was published in 1952 and became the standard reference for years. It concentrated
on the logical calculi for systems of many-valued logic. Its most in uential
successor with a much more extended philosophical content was the Rescher
monograph [448] of 1969 which combines the presentation of dierent systems
of many-valued logic with extended discussions on intuitive backgrounds and
motivations.
A very concise textbook presentation was given by Ackermann [1].
Mainly philosophical discussions are the cores of the books by Zinov'ev
[601] and Dumitriu [130]. And highly in uential, mainly among the com5 The notion \fuzzy logic" unfortunately has quite dierent meanings. Historically
the rst understanding of \fuzzy logic" was as a reference to some systems of

many-valued logic underlying the considerations on fuzzy sets, cf. Chapter 18.
Later the use changed toward an understanding which tended to identify \fuzzy
logic" with approaches which are also treated under the heading \approximate
reasoning", cf. e.g. [31]. This is also actually one of the main ways to understand
the terminus \fuzzy logic". There are, however, two further ones: the one we
are going to discuss here and later on in Chapter 19 { and an understanding
which tends to denote as \fuzzy logic" all investigations (at least in the eld of
Articial Intelligence or Soft Computing) which use fuzzy sets in a serious way.
These last two kinds of understanding usually are distinguished as fuzzy logic
\in the narrower sense" and \in the wider sense".
60
puter science community, was the collection [451]. A rather critical survey is
presented in [552].6
After a rather long break the German language forerunner [218] of this
book opened a more recent series of books on our topic which was continued
e.g. with the books [73, 237, 347], and the quite recent ones [104, 246, 292]
which re ect the actually growing interest in the whole eld, as do survey
papers like [21, 221, 241, 242, 411].
Nevertheless the research literature has been wide spread. The older one,
fortunately, was to a large extent collected in [448] and, largely complete up
to 1974, in [574]. Additionally [193] has to be mentioned. The most up to
date source, however, is the large logic bibliography [386].
6 This survey contains, however, some smaller inaccuracies which have been dis-
cussed in [577].
Part II
General Theory
61
5. Particular Connectives and Truth Degree

Sets
Historically, the development of many-valued logic started with particular

systems of many-valued logic which each had their own particular connectives and sets of truth degrees. And despite the possibilities of quite general
approaches toward many-valued logic also actually some of these particular
connectives and sets of truth degrees are of central importance { and are
supposed to be well known at least as basic examples.
In the case of nite truth degree sets these are either initial parts of the
sequence of integers e.g. of the form
f1; 2; : : :; mg
(5.1)
for some given number m, or these are sets of rationals between 0 and 1 of
the kind
(5.2)
Wm = f0; m1;1 ; m2;1 ; : : : ; mm;;12 ; 1g.
In the case of innite truth degree sets this is sometimes the set N of nonnegative integers, but it is essentially common usage to consider either the
countably innite set W0 of all rationals from the unit interval
W0 = f nk j 0 k n; n 6= 0g = fx 2 Q j 0 x 1g
(5.3)
or the full real unit interval as an uncountably innite truth degree set:
W1 = [0; 1] = fx 2 R j 0 x 1g :
(5.4)
We shall disregard here the nite truth degree sets (5.1) as well as the
innite truth degree set N also because they do not directly, or even not at
all, satisfy our assumption (3.1). In the case of a nite set of truth degrees
this is, however, completely insignicant also from the viewpoint of ordering
because the sets f1; 2; : : :; mg and Wm have not only the same number of
elements, but are even order isomorphic. Such an order isomorphism between
f1; 2; : : :; mg and Wm is realized by the mapping
x 7! mx ;; 11
(5.5)
which at the same time can be considered as a numbering of the truth degrees
in Wm in increasing order. And if one is interested, one also has an equally
natural order reversing bijection realized by the mapping
64
5. Particular Connectives and Truth Degree Sets
;x
x 7! m
m;1
(5.6)
which at the same time can be considered as a numbering of the truth degrees
in Wm in decreasing order. And each one of these mappings can, in an obvious
way, be used to relate the truth degree functions of a propositional system
of many-valued logic with truth degree set f1; 2; : : :; mg uniquely with the
truth degree functions of another system with truth degree set Wm .
Correspondingly one is able to map any other nite truth degree set W S
bijectively to a suitable truth degree set Wm . If W S is linearly ordered this
can even be an order isomorphism; if W S is only partially ordered this bijection may still preserve the (positive) ordering relationships one has between
elements of W S .
For the truth degree set N there is, however, not such a simple relationship
neither to W0 nor to W1 . On the one hand there exists not even a bijection
between N and W1 , but only injections. On the other hand there exist bijections from N onto W0 , but no such bijection can be an order isomorphism.
The reason for this is that the natural ordering of N is a discrete one, i.e. each
integer has an immediate successor and, except 0, also an immediate predecessor, the natural ordering of W0 , however, is a dense one, i.e. in between
any two rationals from W0 there exists a third one.
If necessary, one can instead of N consider either the set f0; 21 ; 43 ; 45 ; : : : ; 1g
or the set f0; : : : ; 14 ; 31 ; 12 ; 1g which both are subsets of W0 and which \to
a large extent" realize the ordering of the integers within the restrictions
imposed by (5.1). Usually, however, it is not necessary to consider these sets
of truth degrees.
As a further simplication of our notation we shall use 1 ; : : : ; m as an
\downward" enumeration of our truth degrees, i.e. we put
m ; i for all i = 1; : : : ; m
i =def m
;1
getting e.g. 1 = 1 and m = 0 as well as W S = f1 ; : : : ; m g.
(5.7)
Regarding the \standard" innite truth degree sets W0 and W1 , it proves

later on to be of minor importance that both sets have dierent numbers of
elements, in the strong set theoretic sense that there does not exist a bijection
between both. Instead it is of greater importance that within W1 each subset
has a supremum and an inmum which is not the case within W0 .
Most prominent examples of truth degree functions for systems S of propositional many-valued logic derive from the standard connectives in classical
logic and generalize them. To describe these functions { but also any other
truth degree function { one has dierent methods. In the case that the set
W S is nite and not too large, such truth degree functions may be determined
using tables { called truth degree tables { collecting all their values. However,
such tables can become quite large depending on the number of truth degrees
and the arity of the function: having e.g. a quaternary truth degree function
and nine truth degrees yields 94 = 6561 entries for such a table. And in the
5.1 Conjunction Connectives
65
general case of m truth degrees and a k-ary truth degree function

one has to
consider a table with mk entries { having a total number of mmk k-ary truth
degree functions in this situation.
Therefore it is preferable to characterize truth degree functions by suitable
{ e.g. arithmetical { formulas. And this is the most preferable method to
determine truth degree functions in the cases of innite truth degree sets.
For the particular truth degree functions we are going to consider here we
shall { if suitable { use both ways of presentation. But for the characterization
of truth degree functions via truth tables we conne ourselves to ve truth
degrees, i.e. to the truth degree set
W5 = f0; 41 ; 21 ; 34 ; 1g :
Usually this shall be sucient to understand the general behavior of the truth
degree functions.
As in classical logic, also many-valued logic has a kind of standard repertoire of (types of) connectives which are considered as important, partly for
historical reasons. These basic types of connectives include (versions of) negations, conjunctions, disjunctions, and implications. The historically earliest
examples of such connectives always have been considered as interesting particular cases from a much wider variety of possible candidates { a point of
view which forces one to discuss criteria which could be used to characterize
such varieties of possible candidates for e.g. conjunctions, implications, negations etc. The standard as well as the normality conditions mentioned earlier
give particular cases of such criteria. However, other properties like functional
equations for the corresponding truth degree functions may do such a job as
well, as was e.g. considered in [118, 474].
Despite the fact that there is no true ranking among the dierent types
of connectives, quite recent developments support the viewpoint that (some
kinds of) conjunction connectives could be of rather fundamental importance.
Thus we rst consider conjunction connectives.

The most basic example for a truth degree function of a conjunction connective comes from the earlier papers [334, 339] of Lukasiewicz and is also
used by Go del [199]. It is the (binary) function et1 with the denition
et1 (u; v) =def minfu; vg :
(5.8)
Another example which also (implicitely) appeared already in the initial papers by Lukasiewicz has the more complicated denition
et2 (u; v) =def maxf0; u + v ; 1g
(5.9)
66
and is known as Lukasiewicz (arithmetic) conjunction or bounded product,

but sometimes also called bold(face) conjunction.1 Both these denitions are
independent from the number of truth degrees of the particular system of
many-valued logic they belong to. In nitely many-valued systems with truth
degree set W5 one could describe these truth degree functions also by the
tables of Fig. 5.1.
3
4
0 0 0 0 0
0 0 0 0 0
1 0 0 0 0
0 14 41 14 14
41
1
1
1
1
1
0 4 2 2 2
23 0 0 01 41
1
1
3
3
0 4 2 4 4
4 0 0 4 2
1 0 14 21 34 1
1 0 14 12 34
Fig. 5.1. Truth table characterizations of conjunction connectives
et1
0
1
14
32
4
1
4
1
2
3
4
et2
1
4
1
2
1
41
23
4
Another type of truth degree function for a conjunction connective which can
only be used for some suitable truth degree sets is the function
et3 (u; v) =def u v :
(5.10)
This truth degree function has seriously been considered only recently. This
may, at least partly, be caused by the fact that among the usual truth degree
sets, which we mentioned earlier in this chapter, besides the traditional truth
value set W2 = f0; 1g only the innite sets W0 and W1 are closed under et3 ,
i.e. under the product. And at least in the earlier periods of many-valued logic
there was a tendency to prefer truth degree functions which can equally well
be considered in the nitely as well as in the innitely many-valued systems.
These three truth degree functions are, however, only particular cases of
a much wider variety of possible candidates. Obviously, it is not of much
interest to extend this list of examples, instead one likes to have some (few)
leading principles which reasonable candidates for conjunction connectives
should satisfy. Actually, there is a wide agreement that such principles should
be the commutativity and associativity of the truth degree functions together
with some suitable monotonicity and borderline conditions. The type of truth
degree functions thus determined are called t-norms.2
1 The motivations for these names are historical in the rst case, systematical in
the second, and essentially accidental in the third one which refers to a notation
sometimes used for the corresponding connective. The qualication \arithmetic"
depends on the fact that in the Lukasiewicz systems as introduced in Chapter 9
also the connective (5.8) is present.
2 The notion of t-norm, a usual shortening of the longer \triangular norm", rst
appeared in the context of probabilistic metric spaces in the work of Menger
[359] in 1942, cf. also his [360], as well as in the work of Schweizer and Sklar
67
Denition 5.1.1. A binary operation t in the real unit interval [0; 1] is a
t-norm i it is
(T1) associative, i.e. satises t(t(u; v); w) = t(u; t(v; w)) for all u; v; w 2
[0; 1];
(T2) commutative, i.e. satises t(u; v) = t(v; u) for all u; v 2 [0; 1];
(T3) non-decreasing in the rst { and hence in each { argument, i.e. satises
u v ) t(u; w) t(v; w) for all u; v; w 2 [0; 1];
(T4) has 1 as neutral element, i.e. satises t(u; 1) = u for each u 2 [0; 1].
As a corollary of this denition one immediately has that for any t-norm
t it holds true
t(u; 0) = 0 for each u 2 [0; 1]
(5.11)
because of t(u; 0) = t(0; u) t(0; 1) = 0.
Because of the associativity condition (T1) in iterated applications of a
t-norm, parentheses can be omitted and one can (by some abuse of language)
even write t(u; v; w) instead of either t(t(u; v); w) or t(u; t(v; w)). Occasionally we shall use this liberalized notation in the general form t(u1 ; : : : ; un )
with the understanding that one refers to a recursive denition with the
crucial condition t(u1 ; : : : ; un) = t(t(u1 ; : : : ; un;1 ); un ).
In algebraic terms, each t-norm represents a semigroup operation3 in the
unit interval [0; 1] with neutral element 1 and annihilator 0, and together
with the usual ordering it makes [0; 1] an ordered semigroup. And because
of (T1) and (T4), for each t-norm t the algebraic structure h[0; 1]; t; 1; i is
even a commutative (or: abelian ) ordered monoid.
Semigroups may have idempotents, i.e. elements a with t(a; a) = a. Because for each t-norm based semigroup h[0; 1]; ti the elements 0; 1 are obviously idempotents, all the others are called non-trivial idempotents. For
short, the idempotents of such a semigroup h[0; 1]; ti are usually called the
idempotents of the t-norm t, or: t-idempotents ).
It is a standard result from elementary analysis that the set of all idempotents of such a t-norm based semigroup h[0; 1]; ti is a closed subset of the unit
interval [0; 1], provided the basic t-norm is continuous. And this means, again
by a standard result from the elementary topology of the real line, that the
complement of the set of all idempotents of such a t-norm based semigroup
[487, 489, 490], and it was also studied in the context of functional equations in
investigations of Frank [176], who even considered a wider class of functions. A
recent monograph devoted to t-norms and their applications is [309], other books
which provide essential results on t-norms are [81, 170]. A short rst introduction
may also be [306, 308].
3 We prefer here, as later on with the t-conorms, the functional prex notation t(u; v), but the inx notation u t v which is more common in the algebraic context is equally usual for t-norms, sometimes even preferable, as the
reader can recognize in rewriting associativity of a t-norm t as the condition
(u t v) t w = u t (v t w).
68
with a continuous t-norm is a countable union of disjoint, i.e. non-overlapping

open intervals.
It should be mentioned in passim that this is no longer true for t-norms
which are not continuous, as may be seen from the t-norm

fu; vg ; if u < v
t(u; v) =def min
(5.12)
0;
otherwise,
also called nilpotent minimum, which has as its corresponding set of idempotents the set f0g [ ( 12 ; 1].
The class of all t-norms is quite large. Therefore subclasses of it are taken
into account for dierent reasons. One approach toward interesting subclasses
of t-norms is to impose some continuity properties: from this point of view
the most interesting subclasses are formed by the left continuous as well as
by the continuous t-norms.
For the continuous t-norms it proves to be equivalent whether one considers them as continuous functions of two variables, or whether one considers
them as continuous functions in each one of their arguments, with the other
argument as a parameter, as is shown in the following result.4 By analogy,
therefore, we denote a t-norm t as left continuous i it is a left continuous function in each of its arguments, i.e. i it satises the sup-preservation
properties
sup t(ui ; v) = t( sup ui ; v) ;
sup t(u; vi ) = t(u; sup vi )
i!1
i!1
i!1
i!1
for all (non-decreasing) sequences (ui )i0 and (vi )i0 , and for all u; v from
the unit interval. Of course, because of the commutativity (T2) it is again
sucient for the left continuity of t to have only one of these sup-preservation
conditions satised.
Proposition 5.1.1. A t-norm t is a continuous function of two variables i
for each b 2 [0; 1] the unary function tjb characterized by the equation
tjb(x) = t(x; b)
(5.13)
is a continuous (unary) function.
Proof: If t is continuous as a function of two variables, then by this
continuity one immediately has that all the unary functions tjb are continuous.
So assume that all the unary functions tjb for b 2 [0; 1] are continuous
functions. By commutativity (T2) one gets that also all those unary functions
taj for a 2 [0; 1] are continuous which are characterized by the equations
taj(x) = t(a; x).
Consider some point (u ; v ) of the unit square, and let (ui ; vi ) be a
sequence of points of the unit square which converges toward (u ; v ). Then
also the sequences (ui )i0 and (vi )i0 converge toward u and v , respectively.
4 For the proof, we closely follow the proof procedure of [308].
69
Furthermore there exist non-decreasing sequences (u0i )i0 and (vi0 )i0 , as
well as non-increasing sequences (u00i )i0 and (vi00 )i0 , such that
0
00
0
00
u = ilim
and v = ilim
!1 ui = ilim
!1 ui
!1 vi = ilim
!1 vi ;
and such that
u0n un u00n
and
vn0 vn vn00
(5.14)
hold true for all indices n 2 N .

Now let " > 0 be some real number. Remember that tu j is continuous.
Then there exists some integer m 2 N such that
t(u; v ) ; " < t(u; vm0 ) t(u; vn) t(u; vm00 ) < t(u; v) + "
for all n m, n 2 N , because of (5.14) and the fact that t satises the
monotonicity condition (T3).
However, also the functions tjvm0 and tjvm00 are continuous. Therefore there
exists some integer k 2 N such that
t(u; vn0 ) ; " < t(u0j ; vn0 ) t(u; vn) t(u00j ; vn00) < t(u; vn00) + "
for all j k. Additionally it is
t(u0j ; vm0 ) t(uj ; vn) t(u00j ; vm00 )
for all n m. Then one has at all
t(u; v ) ; 2 " < t(un; vm ) < t(u; v) + 2 "
for all j; n maxfm; kg, and thus particularly
t(u; v ) ; 2 " < t(ui ; vi ) < t(u; v) + 2 "
for all i maxfm; kg. And this means t(u ; v ) = limi!1 t(ui ; vi ), i.e. t is
continuous at the point (u ; v ). Because (u ; v ) was chosen arbitrarily from
the unit interval, t is continuous in [0; 1]2 , i.e. continuous as a function of two
variables.
2
Another interesting subclass is formed by the Archimedean t-norms which
are characterized by the condition
t Archimedean i for all u; v 2 (0; 1) there exists some (5.15)
n 2 N such that u(tn) < v
with the power notation u(tn) dened by
8
1;
if n = 0,
>
<
u;
if n = 1,
u(tn) =def > n times
(5.16)
: z }| {
t(u; u; : : : ; u); if n > 1.
Sometimes it is useful to refer to a slightly dierent characterization which
is an immediate corollary.
70
Corollary 5.1.1. A t-norm t is Archimedean i for each u 6= 1 the sequence
(ut(n) )n0 converges toward 0.

For continuous t-norms there is a much simpler characterization of Archimedeanicity which for this class of t-norms often is even taken as the denition of that property.
Proposition 5.1.2. A continuous t-norm t is Archimedean i t(u; u) < u
holds true for all u 2 (0; 1).5
Proof: Let t be a continuous t-norm. Even without the continuity assumption, one has t(u; u) u for each u 2 [0; 1] because of (T3), (T4).
If one has t(a; a) = a for some a 2 (0; 1) then it is a(tn) = a 6< a for
all n 2 N and hence t not Archimedean. Hence Archimedeanicity implies
t(u; u) < u for all u 2 (0; 1).
So suppose that t(u; u) < u holds true for all u 2 (0; 1). Let u; v 2 (0; 1). In
the case u v one has u(2)
there does
not
tn) < v. For the case v < u assume that
n)
(
(
n
)
(2
exist some n 2 N with ut < v. Then one always has v < ut , and (ut n )n0
)
is a decreasing sequence which has to have a limit: w = limn!1 u(2
t . For
this limit w one has 0 < v w < u < 1 and by continuity of t immediately
(2n ) (2m )
(2n m ) = lim u(2k ) ;
t(w; w) = nlim
!1 mlim
!1 t(ut ; ut ) = n;mlim
!1 ut
k!1 t
+
i.e. t(w; w) = w, a contradiction. Hence t is Archimedean.

2
It is obvious from the previous remarks concerning idempotents that one
immediately has the following
Corollary 5.1.2. A continuous t-norm t is Archimedean i it does not have
non-trivial idempotents.
It is interesting to notice that the dierence between left continuity and
continuity disappears for Archimedean t-norms, as shown in [312].
Proposition 5.1.3. Each left continuous Archimedean t-norm is continuous.
Proof: Let t be an Archimedean t-norm which is left continuous. To get
the continuity of t one has, according to Proposition 5.1.1, to show that each
one of the functions tjb for b 2 [0; 1] is continuous. And this is already shown
if one proves that all these functions tjb are right continuous.
So assume that one such function tjb is not right continuous. Then there
exist some u^ 2 [0; 1) and a non-increasing sequence (ui )i0 with u^ = limi!1 ui
such that tjb (û) 6= limi!1 tjb (ui ), i.e. such that
5 The assumption of this Proposition can be weakened: it suces to consider a
t-norm t such that the unary function f (u) = t(u; u) is right continuous (in the
standard sense).
t(b; u^) < ilim

!1 t(b; ui ) :
71
(5.17)
Here one obviously has 0 6= b 6= 1 and u^ 6= 1. But it is even u^ 6= 0, because

otherwise in the case u^ = 0 one has for u+ = limi!1 t(b; ui ) rst t(b; u+) =
0 < u+, and thus for each um < u+ even
u+ t(b; um) t(b; u+) u+ :
Therefore one then has
(n)
u+ = t(b; u+) = t(b; t(b; u+)) = t(b(2)
t ; u+ ) = : : : = t(bt ; u+)
for each 1 n 2 N . However, by Corollary 5.1.1 there exists some m 2 N
such that b(tm) < u+. Hence one gets
u+ = t(b(tm) ; u+) t(u+ ; u+) < u+
by Archimedeanicity, a contradiction.
So we have 0 < b; u+ < 1, and furthermore
0 t(b; u^) < u+ = ilim
!1 t(b; ui ) b
for the previously mentioned sequence (ui )i0 . Now consider also an increasing sequence (zn )n0 converging to 1. Suppose that for each n 2 N integers
kn ; ln 2 N are chosen such that
(zn )(tkn +1) b < (zn )(tkn ) ;
(zn )(tln +1) u^ < (zn )(tln ) :
Then on the one hand one immediately has
(zn )(tkn +ln +2) t(b; u^) ;
and hence also
lim sup (zn )(tkn +ln +2) t(b; u^) :
(5.18)
And on the other hand, again because of Corollary 5.1.1, there exists some
m 2 N with um < (zn )(tln ) . Hence one has
u+ t(b; um) < (zn )(tkn +ln ) ;
and thus also
t(u+; (zn)2t ) (zn)(tkn+ln+2) :
Because one additionally has limn!1 (zn )(2)
t = limn!1 t(zn ; zn ) = 1 from
the left continuity of t, this last inequality gives
(kn +ln +2) ;
+
2
u+ = nlim
!1 t(u ; (zn )t ) lim inf (zn )t
which contradicts (5.18) because of t(b; u^) < u+.

2
Archimedean t-norms t satisfy an interesting strengthening of the characteristic inequality t(u; u) < u for all u 2 (0; 1):
72
t Archimedean ) t(u; v) < minfu; vg :

(5.19)
Obviously, one has t(u; v) minfu; vg in any case. However, from e.g.
t(u; v) = v one immediately gets t(u(tn); v) = v for all n, contradicting the
Archimedeanicity property of t.
And also another fact is interesting and important: if one never has
t(u; u) < u for some u 2 (0; 1), then this characterizes the t-norm et1.
Proposition 5.1.4. The only t-norm which has all u 2 [0; 1] as idempotents
is the t-norm t = min.

Proof: Of course, for t = min each u 2 [0; 1] is an idempotent. On the
other hand, if each u 2 [0; 1] is a t-idempotent, one has for all u; v 2 [0; 1]
with u v:
u = t(u; u) t(u; v) u :
This means
t(u; v) = u = minfu; vg for u v 2 [0; 1].
Similarly one gets the corresponding result for the case v u.
2
Two further important classes of t-norms are the strictly monotone tnorms, i.e. those t-norms t which as binary functions over [0; 1]2 are strictly
increasing functions over (0; 1)2 , that means which satisfy:
u < v ) t(u; w) < t(v; w) ;
(5.20)
and the continuous ones among them which are called strict.6
Strictly monotone t-norms t have the nice property that they do not have
zero divisors, i.e. for all u 6= 0 6= v one has also t(u; v) 6= 0. For let u 6= 0 6= v,
then there exists some 0 < w < u and therefore one has 0 t(w; v) < t(u; v).
Obviously there are t-norms which have zero divisors. The Lukasiewicz
t-norm tL = et2 is an example: because of et2 (u; v) = maxf0; u + v ; 1g a
pair (u0 ; v0 ) with 0 < u0 ; v0 1 is a pair of zero divisors of tL i u0 + v0 1.
On the other hand, the t-norm tP = et3 is an example of a t-norm which
does not have zero divisors.
Furthermore, the (pointwise) ordering relation of t-norms gives additional
information concerning t-norms with zero divisors. One easily recognizes that
for all t-norms t1 ; t2
t1 6 t2 and t2 has zero divisors ) t1 has zero divisors (5.21)
holds true. Therefore one has also
t1 6 t2 and t1 is without zero divisors
) t2 is without zero divisors :
(5.22)
6 One has to be careful with terminology in this case: being strict means for a
t-norm to be continuous and strictly monotonous.
73
All together that gives for a t-norm t:

t 6 tL ) t has zero divisors ;
(5.23)
tP 6 t ) t is without zero divisors :
(5.24)
A nice algebraic characterization is available for the strictly monotone
t-norms:
a t-norms t is strictly monotone i it satises the cancellation law,
i.e. i always t(u; w) = t(v; w) for w > 0 implies u = v.
All the above mentioned truth degree functions etk , k = 1; : : : ; 3, are
continuous t-norms. The t-norms et2 and et3 are even Archimedean ones,
but et1 isn't Archimedean. And et3 is the only strict t-norm among them.
Two other interesting t-norms are the sometimes so-called drastic product
t0 dened by

fu; vg; if maxfu; vg = 1
t0 (u; v) =def min
(5.25)
0
otherwise,
and the nilpotent minimum t1 of [169] dened by
n
fu; vg; if u + v > 1
t1 (u; v) =def min
(5.26)
0
otherwise.
The drastic product t0 is Archimedean but not continuous, and the nilpotent
minimum t1 is neither Archimedean nor continuous, but left continuous.
Obviously, one has additionally for any t-norm t the inequality
t0 6 t 6 et1 = min
(5.27)
with the ordering 6 of the (binary) functions taken pointwise.
Further examples of t-norms are provided by a variety of { usually oneparametric { families. Hamacher [255] e.g. considers the one-parametric
family tH; with parameter 0 dened as
(5.28)
tH; (u; v) =def + (1 ; )(uvu + v ; uv)
with the additional commitment to take tH; (u; v) = 0 for = u = v = 0.
These t-norms contain for = 1 the (algebraic) product et3 and converge for
! 1 toward the drastic product t0 .
A problem on functional equations was the source for Frank [176] to
consider the family

u ; 1)(sv ; 1)
(
s
tF;s(u; v) =def logs 1 +
(5.29)
s;1
of continuous Archimedean t-norms with parameter s > 0; s 6= 1 and the
particular further cases
tF;0(u; v) =def lims!0 u tF;s v = minfu; vg ;
tF;1(u; v) =def lims!1 u tF;s v = u v ;
tF;1(u; v) =def lims!1 tF;s(u; v) = maxf0; u + v ; 1g ;
74
which extend the denition (5.29).

Yager [578] introduced still another family tY;p with parameter p > 0 by
tY;p(u; v) =def 1 ; minf1; ((1 ; u)p + (1 ; v)p )1=pg :
(5.30)
For p ! 0 these t-norms converge toward the drastic product t0 , and for
p ! 1 they converge toward the minimum. Additionally, for p = 1 one has
the t-norm et2 .
A fourth family tW; with parameter > ;1 was introduced by Weber [566]. He dened
(5.31)
tW; (u; v) =def max0; u + v1;+1+ uv :
The parameter = 0 here gives the bounded product et2 . For ! ;1 the
t-norms tW; converge toward the drastic product t0 . And for ! 1, nally,
the t-norms tW; converge toward the (algebraic) product et3 .
To dene parameterized families of t-norms via (real) parameters is one
approach toward subclasses of the class of t-norms. Another way is to refer
to some method of generation of t-norms out of simpler ones or out of some
type of generating functions, and to consider e.g. parameterized families of
generating functions. This idea turns out to work especially well for the classes
of continuous and of Archimedean t-norms.
Moreover, the class of Archimedean t-norms is of central importance if
one is interested in continuous t-norms. To understand why, we rst have to
introduce an additional mathematical notion.
Denition 5.1.2. Suppose that ([ai ; bi])i2I is a countable family of nonoverlapping proper7 subintervals of the unitPinterval [0; 1] and let (ti )i2I be
a family of t-norms. Then the ordinal sum i2I ([ai ; bi ]; ti ) of the combined
family (([ai ; bi ]; ti ))i2I is the binary function T : [0; 1]2 ! [0; 1] characterised
by

ak + (bk ; ak ) tk ( buk;;aakk ; bvk;;aakk ); if u; v 2 [ak ; bk ] (5.32)
T (u; v) = min
fu; vg
otherwise.
It is an easy exercise to see that for such an ordinal sum T all the points
ai ; bi for any i 2 I are idempotents of T .
Furthermore one immediately recognizes that each one of the mappings
k : x 7!
x ; ak
bk ; ak
(5.33)
is an (order) isomorphism from the interval [ak ; bk ] onto the unit interval
[0; 1]. And the inverse mapping of this isomorphism is given by
;1
(5.34)
k : x 7! ak + (bk ; ak ) x ;
7 Because we are considering closed intervals here, this means that the intersection
of two such intervals is at most a singleton, and that no interval is itself a

singleton.
75
and is dened on [0; 1]. Using this notation, formula (5.32) can be rewritten
as
;1 ;

k tk ( k (u); k (v )) ; if (u; v ) 2 [ak ; bk ]
T (u; v) = min
(5.35)
fu; vg
otherwise.
Often it is helpful to visualize the construction of an ordinal sum. For
a simple case which shows some of the interval summands this is done in
Fig. 5.2. So it becomes an easy exercise to show that a t-norm t which can be
1
min
q
tk
tk+1
t3
q
t2
min
t1
0
Fig. 5.2.
The basic construction of an ordinal sum
represented by such an ordinal sum has zero divisors i this ordinal sum has
a \rst" summand ([0; bi ]; ti ) with bi > 0, like t1 in Fig. 5.2, which itself
is determined by a t-norm which has zero divisors.
The result which explains the importance of the continuous Archimedean
t-norms is the following representation theorem rst proved (in a more general
context) in [383].8
Theorem 5.1.1. For each continuous t-norm t one of the following cases
appears:
1. t = et1 = min ,
2. t is Archimedean ,
0
8 It is interesting to notice that this theorem needs neither the commutativity
assumption (T2) nor the monotonicity assumption (T3) for t-norms but holds
true for an even larger class of binary operations in [0; 1], cf. e.g. [5].
76
3. t is the ordinal sum of a nonempty family (([ai ; bi ]; ti ))i2I where all ti

are continuous Archimedean t-norms.
Proof: Let t be a continuous t-norm and Jt = fd j t(d; d) = dg the set
of its idempotents. In the case of Jt = f0; 1g, i.e. that t has only the trivial
idempotents, one has by Corollary 5.1.2 that t is Archimedean.
So suppose that t has non-trivial idempotents, and let d 2 Jt be any
idempotent of t. Then one has for all x d y immediately
d = t(d; d) t(d; y) d ;
hence d = t(d; y), i.e. d acts like an annihilating element for all y d. On the
other hand from the monotonicity of t one has t(d; x) x. But one even has
t(x; d) = x, because otherwise from the continuity of t there would follow
the existence of some x < z < d with x = t(z; d), and one would have a
contradiction from
t(x; d) = t(t(z; d); d)) = t(z; t(d; d)) = t(z; d) = x :
Thus d acts like a unit element for all x d.
As a result, for any two dierent idempotents d1 ; d2 2 Jt , d1 < d2 , the
restriction of t to the square [d1 ; d2 ]2 itself \essentially" is a (continuous)
t-norm, in the sense that the isomorphic copy t(d ;d ) dened by
t(d ;d )(x; y) = t(d1 + x(d2 ; d1); d1 + y(d2 ; d1 )) ; d1
1
d2 ; d1
is a t-norm. And because one has
t(x; y) = d1 + (d2 ; d1
) t(d1 ;d2 )
x ; d1 ; y ; d1
d2 ; d1 d2 ; d1
for all x; y 2 [d1 ; d2 ], this means also that the restriction t [d1 ; d2 ]2 acts like
a summand in an ordinal sum for t.
Of course, if there do not exist (non-trivial) idempotents of t between d1
and d2 , then t(d ;d ) is an Archimedean t-norm.
Furthermore one gets easily for each d1 x d2 and all y < d1 the
inequality t(x; y) t(d1 ; y) = y, and for all d2 < z the inequality t(x; z )
t(x; d2 ) = x, which means for all x; y 2 [0; 1] simply
t(x; y) = minfx; yg for all x 2 [d1 ; d2] and y 2= [d1; d2 ] :
(5.36)
Here d1 = d2 is allowed and yields
t(x; d) = minfx; dg for all x 2 [0; 1] and all d 2 Jt .
(5.37)
So it remains to consider the structure of the set Jt of all t-idempotents.
If Jt is nite, then let I = Jt n f1g, and for each d 2 I also d+ 2 Jt
its immediate successor, i.e. the (uniquely determined) d < c 2 Jt such that
there does not exist
some h 2 Jt with d < h < c. In this case t obviously is
P
the ordinal sum d2I ([d; d+ ]; t(d;d ) ).
1
77
In any case Jt is a closed set, i.e. a limit point of idempotents of t is

itself an idempotent of t. To see this let u be a limit point of Jt , and let
u = limn!1 dn for some sequence (dn )n0 of t-idempotents. One can assume
that this sequence (dn )n0 is either non-decreasing, or non-increasing. If it is
non-decreasing, then one always has dn u, and thus from the continuity of
t and t(dn ; u) = dn:
t(u; u) = nlim
!1 t(dn ; u) = nlim
!1 dn = u :
And if this sequence is non-increasing, then one always has dn u, and thus
from the continuity of t and t(dn ; u) = u:
t(u; u) = nlim
!1 t(dn ; u) = nlim
!1 u = u :
Hence u 2 Jt , and Jt is a closed set.

Hence, if Jt is innite, then each limit points of the set Jt is either isolated,
or belongs to some interval [a; b] [0; 1] of idempotents of t with a 6= b.
BecauseSJt is closed, the complement [0; 1]nJt is open, and hence the union
[0; 1] n Jt = (ai ; bi ) of countably many nonempty open intervals (ai ; bi ). This
ordinal sum may be the sum of an empty family, in this case one has that each
u 2 [0; 1] is an idempotent, and hence has t = min according to Proposition
5.1.4.
2
Having in mind the fact that the ordinal sum of the empty family of continuous Archimedean t-norms is the t-norm et1 = min, and that the ordinal
sum of a family of continuous Archimedean t-norms which consists of just
one member t is nothing but t itself, this theorem could be formulated in a
more compact form:
Corollary 5.1.3. Each continuous t-norm t is the ordinal sum of a family
(([ai ; bi ]; ti ))i2I where all ti are continuous Archimedean t-norms.
For continuous Archimedean t-norms one has another important type of
representation theorem.9 It refers to the pseudo-inverse f (;1) of a function
f : [0; 1] ! [0; 1] which for each non-increasing f is dened10 for all z 2
[0; 1] by
f (;1) (z ) =def supfx 2 [0; 1] j f (x) > z g ;
and which can for strictly decreasing and continuous functions f with property f (1) = 0 be more simply characterized for all z 2 [0; 1] by
9 This theorem can be deduced from results on topological semigroups, cf.
[152, 383, 410]. But there is also a proof available which uses only (mathematically) more elementary means, cf. [331]. In any case, however, these proofs use
mathematical methods beyond the scope of this book and shall therefore not be
given here. The interested reader may e.g. consult [309].
10 Because f is dened over [0; 1], maps into the nonnegative reals, and may be
strongly decreasing, its range has to be (a subset of) the set R0 of all nonnegative reals completed by a \limit point" +1.
78
f (;1) (z ) = f ;1 (minfz; f (0)g) =
f ;1 (z ); if z 2 [0; f (0)]
0;
if z 2 (f (0); 1].
(5.38)
Theorem 5.1.2. To each continuous and Archimedean t-norm t there exists

a continuous and strictly decreasing function f : [0; 1] ! [0; 1] with f (1) = 0
such that for all u; v 2 [0; 1]
t(u; v) = f (;1);f (u) + f (v) :
(5.39)
Proof: We rst dene over [0; 1] two families of functions sn for n 2 N

and rn for n 1. The denition of the functions sn is simple and reads:
sn (x) =def x(tn) for each x 2 [0; 1].
(5.40)
Obviously, each one of these functions sn is continuous because it is the
n-fold iteration of a continuous operation. Furthermore each function sn is
non-decreasing because t is non-decreasing (in both arguments). Each sn with
n > 0 maps onto [0; 1] because of sn (0) = 0 and sn (1) = 1. And one has
0 < sn (x) ) sn+1 (x) < sn (x)
(5.41)
because of sn+1 (x) = t(x; sn (x)) < minfx; sn (x)g sn (x) according to
(5.19).
The denition of the functions rn is a bit more dicult and reads
rn (x) =def maxfy 2 [0; 1] j sn(y) = xg for each x 2 [0; 1]. (5.42)
This denition needs as its justication the proof that such a maximum
always exists. However, as sn maps onto [0; 1] for n 1, one always has
Rn (x) =def fy j sn (y) = xg 6= ;. Consider lx = sup Rn (x) and a convergent
sequence (ym )m0 from Rn (x) with limit lx. Then one always has sn (ym ) = x
and even x = limm!1 sn (ym ) = sn (lx ) by continuity of sn . Hence lx 2 Rn (x)
and lx = max Rn (x). Finally it is easy to recognize that one has for each
n 1:
sn (rn (x)) = x for each x 2 [0; 1].
(5.43)
Looking at the interval Kn = [rn (0); 1] one has that the function sn Kn
is an order-preserving, i.e. strictly increasing 1-1 mapping onto [0; 1]. This
essentially follows from the observation that one has
rn (0) x < y 1 ) sn (x) < sn (y) :
(5.44)
To prove (5.44), suppose rn (0) x < y 1. By (T3) one then immediately
has sn (x) sn (y). In the case of sn (x) = sn (y) observe rst that there would
exist some z0 2 (0; 1) with x = t(y; z0 ) because the mapping z 7! t(y; z ) is
a continuous function from [0; 1] onto [0; y]. And one has in this case that
sn (x) > 0: in the opposite situation of sn (x) = sn (y) = 0 rst x; y 2 Rn (0)
results, and then a contradiction from Rn (0) \ Kn = frn (0)g 6= fx; yg. Then
in the case sn (x) = sn (y) a contradiction results from
79
0 < sn (x) = sn (y) = sn;1 (x) t x = sn;1 (x) t (y t z0 )

= (sn;1 (x) t y) t z0 (sn;1 (y) t y) t z0 = sn (y) t z0 < sn (y) :
Here one has the last inequality because one has either sn (y) = 1 and thus
sn (y) t z0 = z0 < 1 = sn (y), or one has sn (y) < 1 and then sn (y) t z0 < sn (y)
by (5.19).
By (5.44), the continuity of sn and hence also of the restriction sn Kn ,
and by (5.43) one has that rn : [0; 1] ! Kn is the inverse function of sn Kn ,
and hence also order-preserving, i.e. strictly increasing, and continuous.
The family of functions rn , n 1, has another interesting property which
shall be used later on. For all x 2 (0; 1) one has
(rn (x))n1 is strictly increasing with nlim
(5.45)
!1 rn (x) = 1 :
Let x 2 (0; 1). In the case that the sequence (rn (x))n1 would not be strictly
increasing, there would exist some index m with 0 < rm+1 (x) rm (x) < 1,
and hence with 0 < sm (rm+1 (x)) sm (rm (x)) = x. Thus one would have
x = sm+1 (rm+1 (x)) < sm (rm (x)) = x from (5.44) and (5.41), a contradiction. Hence (rn (x))n1 is strictly increasing and has thus a limit % with
rn (x) < % always. Thus one has x sn (%) for any n, and thus % = 1 by
Archimedeanicity.
A last important property of the combination of the functions sm and rn
has to be mentioned for later use, which holds for all x 2 [0; 1]:
sm (rn (x)) = skm (rkn (x)) for all suitable k; m; n 2 N .
(5.46)
If this property (5.46) does not hold, there would exist 1 k; m; n 2 N
and some x 2 (0; 1) such that sm (rn (x)) 6= skm (rkn (x)). For y = rn (x) and
z = rkn (x) this means sn (y) = x and skn (z ) = x, i.e. y; sk (z ) 2 Rn (x),
and hence y sk (z ) and thus sm (y) skm (z ). However, sm (y) 6= skm (z )
by assumption, and therefore sm (y) > skm (z ) and hence y > sk (z ). The
continuous function sk has 1 = sk (1) and sk (z ) among its values, hence
also y, i.e. there exists some w 2 (z; 1) with y = sk (w). But then one has
skn (w) = sn (y) = x, hence w 2 Rkn (x) and w > z , contradicting the choice
z = rkn (x). Thus (5.46) is established.
Now we dene over the set Q 0 of the non-negative rational numbers a
function g : Q 0 ! [0; 1], using some xed c0 2 (0; 1), by
g ( mn ) =def sm (rn (c0 )) for mn 2 Q 0 .
(5.47)
Because of (5.46) this is really the denition of a function over Q 0 . And this
function g has to be considered now.
We rst observe that g is non-increasing. For, having x; y 2 Q 0 with
representations x = md ; y = nd with a common denominator, one gets from
x < y immediately m < n = m + k, and hence
g (x) = g ( md ) = sm (rd (c0 )) sk (sm (rd (c0 ))) = sn (rd (c0 )) = g (y) :
Even more, g is strictly decreasing as long as it assumes positive values, i.e.
80
x; y 2 Q 0 ^ x < y ^ 0 < g (x) ) g (x) > g(y) :

(5.48)
n
m

To see this let x = d and y = d with m < n. From 0 < g (x) = sm (rd (c0 ))
one then gets g (y) = sn (rd (c0 )) sm+1 (rd (c0 )) < sm (rd (c0 )) = g (x),
i.e. (5.48). Furthermore g satises the functional equation
t(g(x); g (y)) = g(x + y) for all x; y 2 Q 0 .
(5.49)
Let x = md and y = nd as before. Then one has
t(g(x); g (y)) = t(sm (rd(c0 )); sn(rd (c0)))
= sm+n (rd (c0 )) = g ( md+n ) = g (x + y) :
And g is a continuous function over Q 0 . This follows from the fact that one
has for each 0 6= x 2 Q 0 :

1
lim g (x ; n1 ) = g (x) = nlim
(5.50)
n!1
!1 g (x + n ) ;
and that one has furthermore according to (5.45)

1
nlim
!1 g (0 + n ) = nlim
!1 s1 (rn (c0 )) = nlim
!1 rn (c0 ) = 1 = g (0) :
For (5.50) one has on the one hand for each x 2 Q 0

1

1
!1 t(g (x); g ( n ))
nlim
!1 g (x + n ) = nlim
1

= t(g (x); nlim
!1 g ( n )) = t(g (x); 1) = g (x) ;
and for each 0 6= x 2 Q 0 on the other hand

1
1
lim g (x ; n1 ) = t(nlim
!1 g (x ; n ); nlim
!1 g ( n ))
n!1

1 1
= nlim
!1 t(g (x ; n ); g ( n )) = nlim
!1 g (x) = g (x)
because of the continuity of t.

Because of this continuity, g can uniquely be extended to a continuous
function g : R0 ! [0; 1] over the set R0 of all non-negative reals. And also g
is non-increasing, and has the properties (5.48) and (5.49).
In particular one has g(0) = 1, and that g is strictly decreasing on the
interval L = [0; b] for b = minfx 2 R0 j g(x) = 0g. Here we have to allow for
b = 1 for the case fx 2 R0 j g(x) = 0g = ;. Therefore we consider now the
\completion" R0 =def R0 [ f1g instead of R0 . Then the restriction g L is
a 1-1 and order-reversing continuous mapping from L onto [0; 1], which has
an inverse mapping f : [0; 1] ! L which is also a 1-1 and order-reversing
continuous mapping.
From the functional relationship g(x + y) = t(g(x); g(y)) for all x; y 2 R0
one has for all u; v 2 [0; 1] with a representation as u = g(x); v = g(y)
immediately
t(u; v) = t(g(x); g(y)) = g(x + y) = g(f (u) + f (v)) ;
(5.51)
81
because one has in this case also x = f (u) and y = f (v). Now it is a routine matter to check that g is the pseudo-inverse of f . Hence (5.51) is the
representation of t which had to be shown to exist.
2
Conversely, each binary function t is a continuous Archimedean t-norm if
it is given via (5.39) using a generating function f with these properties of
continuity and monotonic decreasingness. Each such function f which represents a t-norm t according to (5.39) is called an additive generator of t.
It is interesting to notice that such an additive generator f of a t-norm
t re ects some important properties of t in a simple manner. So one has
e.g. the following characterization of the (pointwise) ordering of continuous

Archimedean t-norms, cf. [490], which can also be extended to continuous
t-norms in general by reference to Theorem 5.1.1, cf. [307].
Proposition 5.1.5. Let t1; t2 be Archimedean t-norms with additive generators f1; f2 , respectively. Then one has t1 6 t2 i the combined function
h = f1 f2;1 with domain [0; f2(0)] is subadditive, i.e. always satises
h(x + y) h(x) + h(y) :
We are, however, more interested in a rather direct characterization of tnorms with zero divisors.
Proposition 5.1.6. Suppose that t is a continuous and Archimedean t-norm,
and generated via (5.39) by the generating function f . Then one has:
(i) t is strict i f (0) = 1;
(ii) t has zero divisors i f (0) < 1.
Proof: (i) Suppose that t is strict. Then one has f (0) = 1. For, assume
that on the contrary one would have f (0) = k 2 R0 . Then consider some
0 < x < 1. In this case one has
0 < t( 21 x; 12 ) < t(x; 21 ) = f (;1) (f (x) + f ( 21 )) :
However, f is continuous, and therefore in the limit one has f (x) ! f (0)
for x ! 0. And therefore one has f (x) + f ( 12 ) > f (0) = k for each x which
satises jf (x) ; f (0)j < 21 , hence t(x; 12 ) = 0 according to (5.38), and thus a
contradiction, if k < 1.
Now suppose conversely f (0) = 1. In this case one obviously has always
t(x; z) = f ;1(f (x) + f (z)). Let x < y. Then one has f (x) > f (y), and
therefore also
f (t(x; z )) = f (x) + f (z ) > f (y) + f (z ) = f (t(y; z ))
and hence t(x; z ) < t(y; z ). Thus t is strictly monotonous and hence strict,
because continuous.
(ii) If t has zero divisors then t is not strict. Hence one has f (0) 6= 1, i.e.
f (0) < 1. So suppose conversely f (0) < 1. Then one has 0 = f (1) < f (0)
82
and, because f is continuous, there exists some 0 < x < 1 with 21 f (0) <
f (x) < f (0). Then one has from (5.38) and (5.39) immediately t(x; x) = 0.
Hence t has zero divisors.

2
Because the generating function f , which exists for each continuous and
Archimedean t-norm, either satises f (0) = 1 or f (0) < 1, one immediately
has the following result.
Corollary 5.1.4. Each continuous and Archimedean t-norm is either strict
or has zero divisors.
And this result can be extended to another interesting representation
theorem which relates the strict t-norms with the usual arithmetic product
tP = et3, and the t-norms with zero divisors with the Lukasiewicz t-norm
tL = et2, in the sense that these two particular t-norms are \prototypes" of
continuous Archimedean t-norms.
To state this result we need the notion of an automorphism of the unit
interval. By such an automorphism we mean an automorphism of the lattice h[0; 1]; min; max; 0; 1i. Equivalently, such an automorphism of the unit
interval is the same as an order automorphism of the ordered set h[0; 1]; ; i,
and also the same as a continuous, strictly increasing surjection of [0; 1] onto
itself.11
Then one has the following theorem.
Theorem 5.1.3. (i) A continuous t-norm t is strict i there exists an automorphism ' of the unit interval with
t(u; v) = ';1 ('(u) '(v)) = ';1 (et3('(u); '(v))) :
(5.52)
(ii) A continuous Archimedean t-norm t has zero divisors i there exists an
automorphism ' of the unit interval with
t(u; v) = ';1 (maxf0; '(u) + '(v) ; 1g) = ';1(et2('(u); '(v))) : (5.53)
Proof: (i) It is easy to check that a function t determined via (5.52) by
an automorphism ' of the unit interval has to be a continuous t-norm which
additionally is strictly monotonous.
So suppose that the continuous t-norm t is strict. Then, according to
Theorem 5.1.2, t is also Archimedean, because one has t(x; x) < t(x; 1) = x
for each x 2 (0; 1). Therefore t can be represented in the form (5.39) with a
generating function f . And this function has, besides f (1) = 0, the property
f (0) = 1 according to Proposition 5.1.6 (i).
Now dene on [0; 1] the function ' by '(x) = e;f (x). Then it is a routine
matter to check that this function ' is an automorphism of the unit interval.
Furthermore one has always f (x) = ; log '(x) and also f (;1) (x) = f ;1(x) =
';1 (e;x). And all together this nally gives
11 It is a routine exercise in elementary analysis to prove that an order automor-
phism of h[0; 1]; ; i has to be a continuous function.
83
t(x; y) = f ;1(f (x) + f (y)) = ';1(elog '(x)+log '(y))
= ';1 (elog '(x) elog '(y) ) = ';1 ('(x) '(y)) :

(ii) Again it is easy to check that a function t determined via (5.53) by an
automorphism ' of the unit interval has to be a continuous and Archimedean
t-norm which has zero divisors.
Hence assume that the continuous
t-norm
t is Archimedean and has zero
;

divisors. Let t(u; v) = f (;1) f (u) + f (v) and b = f (0). Then one has 0 6=
b 2 R0 . Consider the function h : [0; 1] ! [0; 1] dened by
h(x) = 1 ; 1b f (x) for all x 2 [0; 1];
which is continuous and strictly increasing with h(0) = 0 and h(1) = 1. Thus
it is an automorphism of the unit interval. Its inverse function h;1 : [0; 1] !
[0; 1] is given by
h;1 (x) = f ;1(b (1 ; x)) = f (;1) (b (1 ; x)) :
Now one has for all u; v 2 [0; 1]:

;

h(t(u; v)) = h f (;1) f (u) + f (v)

= h f (;1) b (1 ; h(u) + 1 ; h(v))

= h f (;1) b (1 ; et2 (h(u); h(v)))

= h(h;1 (et2 (h(u); h(v)))) = et2 (h(u); h(v)) ;
and therefore always t(u; v) = h;1 (et2 (h(u); h(v))).
Corollary 5.1.5. If t is a continuous Archimedean t-norm then the (lattice)

ordered semigroup h[0; 1]; t; i is either isomorphic with the Lukasiewicz
semigroup h[0; 1]; et2 ; i, or with the product semigroup h[0; 1]; et3 ; i.
Proof: Formulas (5.52) and (5.53) can obviously be rewritten as
'(t(u; v)) = eti ('(u); '(v))

with i = 1; 2. And this is just the standard condition for an isomorphism
between h[0; 1]; t; i and h[0; 1]; eti ; i, because ' is already supposed to be
an order isomorphism over [0; 1].
2
There are more such representation results for dierent classes of t-norms,
cf. e.g. [170, 309], which shall not be discussed here. We mention instead a
slight variation of the formulation of Theorem 5.1.1 which refers to this last
theorem. To do this we need also a slight generalization of the notion of
ordinal sum.
Denition 5.1.3. Suppose that ([ai ; bi])i2I is a countable family of nonoverlapping proper subintervals of the unit interval [0; 1], let (ti )i2I be a
family of t-norms, and let ('i )i2I be a family of mappings such that each
84
'i is an order
P isomorphism from [ai ; bi ] onto [0; 1]. Then the generalized ordinal sum i2I ([ai ; bi ]; ti ; 'i ) of the combined family (([ai ; bi ]; ti ; 'i ))i2I is
the binary function T : [0; 1]2 ! [0; 1] characterised by

'k ;1 (tk ('k (u); 'k (v)); if u; v 2 [ak ; bk ]
T (u; v) = min
(5.54)
fu; vg
otherwise.
Then one immediately gets from Theorem 5.1.1 together with Corollary
5.1.5 the following result.
Theorem 5.1.4. Each continuous t-norm t is the generalized ordinal sum
of (isomorphic) copies of the Lukasiewicz and the product t-norms.
5.2 Negation Connectives

Starting again from the historical origins, one has that the papers [334, 426]
of Lukasiewicz and Post gave two particular truth degree functions for
negation connectives. Lukasiewicz used uniformly
non1 (u) =def 1 ; u
(5.55)
for any one of our standard truth degree sets. The truth table representation
of this function is provided by one of the tables of Fig. 5.3.
Post restricted himself to the nite valued cases and considered within
the truth degree set Wm the truth degree function

1;
if u = 0
(5.56)
non2 (u) =def u ; 1 ; if u 6= 0,
m;1
which has, notwithstanding this seemingly complicated denition, a very simple truth table representation as also shown in Fig. 5.3. This table shows
clearly that non2 realizes a cyclic, \left rotating" permutation of the truth
degrees and thus indicates why it is rather natural to consider non2 only
over nite truth degree sets. In any case it seems quite dicult to generalize
u
non1 (u) 1
1
4
3
4
1
2
1
2
3
4
1
4
1
4
0
non2 (u) 1 0
Fig. 5.3. Truth table characterization of negation connectives
1
2
1
4
3
4
1
2
1
3
4
non2 to a function dened over W0 or even over W1 . However, taking e.g.

W = f2;k j k 2 N g[f0g as truth degree set would allow for a generalization
which could e.g. be dened as

1;
if u = 0

non2 (u) =def 1 u otherwise.
2
(5.57)
85
In [199] Go del added as another truth degree function for a negation
connective the function
n
non0 (u) =def 1; if u = 0
(5.58)
0 otherwise.
The actual discussion, however, considers Post's cyclic operation non2 as a
non-standard, even exotic example and tends to exclude it from more general
considerations. Accordingly the following denition actually is prevalent.
Denition 5.2.1. A function n : [0; 1] ! [0; 1] is called negation function
i n is non-increasing and satises n(0) = 1 and n(1) = 0. A negation
function n is called strict i it is even strictly decreasing and continuous.
And it is called strong i it is strict and an involution, i.e. satises also the
condition
n(n(u)) = u for all u 2 [0; 1]:
(5.59)
It is interesting to remark that the inverse function n;1 of each strict
negation function n is again a strict negation function. And strong negation
functions even coincide with their inverse functions.
Regarding the previous examples, non2 is not a negation function at all in
this sense assuming m > 2, but non0 is a negation function, and non1 is even
a strong negation function. Another example of a negation function which is
not a strict one is characterized by the equation

<1
non (u) =def 10;; ifif uu =
(5.60)
1.
and is obviously a kind of dual to non0 .
These two negation functions non0 ; non are extreme examples in the sense
that for each negation function n it holds true
non0 6 n 6 non :
(5.61)
An example of a strict negation function which is not also a strong one is
provided by the function non3 with
non3 (u) =def 1 ; u2 :
(5.62)
Other examples of strong negation functions are all the functions of the family
u
(5.63)
nS;(u) =def 11+;u
with parameter > ;1, which Sugeno [517] introduced under the name of
-complement.
For strict as well as for strong negations one has nice representation theorems, cf. [4, 149, 168, 395, 544].
86
Theorem 5.2.1. (i) A function n : [0; 1] ! [0; 1] is a strong negation func-
tion i there exists an automorphism ' of the unit interval such that
n(u) = ';1(1 ; '(u)) for all u 2 [0; 1]:
(5.64)
(ii) For any two strict negation functions n1 ; n2 there exist automorphisms
'; of the unit interval such that one has for each x 2 [0; 1]:
n2(x) = (n1('(x))) :
(5.65)
(iii) A function n : [0; 1] ! [0; 1] is a strict negation function i there exist
automorphisms '; of the unit interval such that
n(u) = (1 ; '(u)) for all u 2 [0; 1]:
(5.66)
Proof: (i) It is easy to see that a function n dened by (5.64) is an
involution with the properties n(0) = 1 and n(1) = 0. And it is also strictly
decreasing because one gets from x < y rst '(x) < '(y), and then 1 ; '(x) >
1 ; '(y) and hence n(x) > n(y), because also ';1 is an automorphism of
the unit interval. Hence n is a strong negation function.
Suppose conversely that n is a strong negation function. Now we use that
each strict negation function n has exactly one xed point12 s 2 (0; 1)
with characteristic property n (s ) = s . So let s be the xed point of n,
and consider an increasing bijection h : [0; s] ! [0; 21 ]. Then one has h(0) = 0
and h(s) = 12 . Dene a function ' : [0; 1] ! [0; 1] by

; s]
'(x) = h1 (;x)h; (n(x)); ifif xx 22 [0
(5.67)
(s; 1].
Obviously this function is continuous and satises '(0) = h(0) = 0 as well as
'(1) = 1 ; h(n(1)) = 1 ; h(0) = 1, and it is strictly increasing over [0; s] as
well as over (s; 1]. So let x s < y. Then one has '(x) = h(x) h(s) = 21 ,
and also '(y) = 1 ; h(n(y)) > 1 ; h(n(s)) = 1 ; h(s) = 21 because of
n(y) < n(s) = s and h(n(y)) < h(n(s)), hence '(x) < '(y). Therefore ' is
an automorphism of the unit interval.
Routine calculations show that one has for each x 2 [0; 1]
;1
1
;
1
' (x) = nh (h(;x1)(1; ; x)); ifif xx 22 ([01;; 21].]
2
1
;
1
This gives for each x 2 [0; 2 ] now ' (1 ; '(x)) = n(h;1 (1 ; (1 ; '(x))) =
n(h;1('(x)) =;1n(h;1 (h(x)) = ;n1(x). And it gives;1for each x 2 ( 21 ; 1] on the
other hand ' (1 ; '(x)) = h (1 ; '(x)) = h (h(n(x))) = n(x). Thus
one has at all the representation (5.64).
(ii) Consider the xed points ni (si ) = si for i = 1; 2 and let t = ss . Dene
mappings '; by
2
1
12 Obviously the continuity gives the existence, and the (strict) decreasingness gives
the uniqueness of the xed point.
87
x;
if x s
'(x) =def nt ;1 ( n (x) ); if x > s2 ,
1
2
t

tx;
if
x

(x) =def n (tn ;1 (x)); if x > ss1 .
2 1
1
Obviously '; are continuous with '(0) = (0) = 0 and '(1) = (1) = 1,
and they are strictly increasing. Hence '; are automorphisms of [0; 1].
In the case x < s2 one has n1 ( xt ) > n1 (s1 ) = s1 and thus
x

;1 n1 x
=
n
t
n
= n 2 ( x) :
(n1 ('(x))) =
2
1
t
t

In the case x > s2 one has n t(x) < s1 and thus

n
(
x
)
2
= n2 (x) :
(n1 ('(x))) =
t
2
Therefore one has (5.65), because n2 as well as n1 ' are continuous

functions.
(iii) One has non1 as a strict negation function. Thus, having given any
strict negation function n, one has to choose n1 = non1 and n2 = n to get
claim (iii) out of claim (ii).
2
Strong as well as strict negation functions are hence \variants" of the
negation function non1 .
As a corollary this gives a representation of strong negation functions via
some generating function as in Theorem 5.1.2.
Corollary 5.2.1. A function n : [0; 1] ! [0; 1] is a strong negation function
i there exists a strictly increasing bijection f : [0; 1] ! [0; 1] with f (0) = 0
such that one has
n(u) = f ;1(f (1) ; f (u)) :
(5.68)
Proof: Assume rst that a function n : [0; 1] ! [0; 1] is determined via
(5.68) from some strictly increasing bijection f : [0; 1] ! [0; 1] with f (0) = 0.
Then one has obviously n(0) = 1 and n(1) = 0. And it is easy to see that
with f also the inverse function f ;1 has to be a strictly increasing bijection.
But this means that the function g(u) = f (1) ; f (u), and hence also n, is
strictly decreasing. Because f has to be continuous, also n is continuous, and
hence a strict negation function. But it is also an involution because of
n(n(u)) = f ;1;f (1) ; f (n(u)) = f ;1;f (1) ; (f (1) ; f (u)) = u :
Let on the other hand n : [0; 1] ! [0; 1] be a strong negation function.
Then one has n(u) = ';1 (1 ; '(u)) according to Theorem 5.2.1(i). Additionally one has '(1) = 1 for this representation considering the function '
as dened in (5.67). This means that (5.64) now becomes the form (5.68). 2
Combining t-norms and negation functions allows one to prove some further interesting representation theorem for a class of particular t-norms,
cf. [409, 170].
88
Theorem 5.2.2. Suppose that t is a continuous t-norm and n a strict negation function. Then one has
t(u; n(u)) = 0 for all u 2 [0; 1]
(5.69)
i there exists an automorphism ' of the unit interval such that for all u; v 2
[0; 1] one has
t(u; v) = ';1 (et2('(u); '(v))) and n(u) ';1(1 ; '(u)): (5.70)
Proof: It is easy to check that a function t determined via (5.70) by an
automorphism ' of the unit interval and a strict negation function n with
property n(u) ';1 (1 ; '(u)) has to have property (5.69).
So suppose now conversely that (5.69) holds. Then t is Archimedean. For
otherwise there would exist some x 2 (0; 1) such that t(x; x) = x. However,
in the case x n(x) there would be 0 < t(x; x) t(x; n(x)) = 0, a contradiction. And in the case x > n(x) one would have 0 = t(x; n(x)) n(x) <
x = t(x; x), such that, because of the continuity of t, some y x had to exist
with n(x) = t(x; y). But then one would have
n(x) = t(x; y) = t(t(x; x); y) = t(x; t(x; y)) = t(x; n(x)) = 0 ;
again a contradiction.
And obviously t has zero divisors. Because of the Archimedeanicity of t
one hence has, according to Proposition 5.1.3(ii), a representation of t in the
form
t(x; y) = ';1(maxf'(x) + '(y) ; 1; 0g) :
And one now gets from the equivalence of t(x; n(x)) = 0 with '(x) +
'(n(x)) 1 also n(x) ';1 (1 ; '(x)).
2
Therefore, the pair et2 ; non1 represents in a suitable sense the \typical situation" of a continuous t-norm and a strict negation function which together
satisfy a kind of generalized \law of contradiction".
5.3 Disjunction Connectives

With the actual preliminaries two dierent approaches toward truth degree
functions for disjunction connectives are suitable. One of them is to look for
reasonable conditions such truth degree functions should satisfy { and thus
to parallel the approach toward truth degree functions for conjunction and
negation connectives. The other one is to approach a truth degree function
for a disjunction connective from a truth degree function for a conjunction
connective and another one for a negation connective { and to demand that
some analogue of the usual deMorgan connection between these types of
connectives should hold true.
5.3 Disjunction Connectives
89
There is no reasonable preference which one of these approaches has over

the other. Thus we shall look at both of them.
The general properties one likes to have satised by disjunction connectives lead to the following class of t-conorms as truth degree functions for
them.
Denition 5.3.1. A binary operation s in [0; 1] is a t-conorm i it is
(S1) associative, i.e. satises s(s(u; v); w) = s(u; s(v; w)) for all u; v; w 2
[0; 1];
(S2) commutative, i.e. satises s(u; v) = s(v; u) for all u; v 2 [0; 1];
(S3) non-decreasing in the rst { and hence in each { argument, i.e. satises
u v ) s(u; w) s(v; w) for all u; v; w 2 [0; 1];
(S4) has 0 as neutral element, i.e. satises s(u; 0) = u for each u 2 [0; 1].
As a corollary of this denition one has that for any t-conorm s it holds
true
s(u; 1) = 1 for each u 2 [0; 1]
because of s(u; 1) = s(1; u) s(1; 0) = 1.
In algebraic terms, each t-conorm represents a semigroup operation in
the unit interval [0; 1] with neutral element 0 and annihilator 1, and together
with the usual ordering it makes [0; 1] an ordered semigroup.
The other approach toward truth degree functions via a deMorgan connection amounts to introduce some function s : [0; 1]2 ! [0; 1] by dening
with reference to some negation function n and to some truth degree function
t for a conjunction connective, e.g. thus by reference to some t-norm t
s(u; v) =def n(t(n(u); n(v))) for all u; v 2 [0; 1]
(5.71)
or by dening
s(u; v) =def n;1(t(n(u); n(v))) for all u; v 2 [0; 1]
(5.72)
or something similar.
In the case that one assumes that n is a strong negation function, i.e. an
involutive one, then denitions (5.71) and (5.72) become equivalent and the
particular form of the deMorgan connection between the t-norms and the
t-conorms becomes inessential.
Up to now, the standard choice of a strong negation function to connect
t-norms and t-conorms is non1 . Starting from a t-norm t one then combines
with it the t-conorm
st (u; v) =def 1 ; t(1 ; u; 1 ; v) :
(5.73)
Conversely, starting from any t-conorm s the corresponding denition
ts(u; v) =def 1 ; s(1 ; u; 1 ; v) :
(5.74)
90
obviously gives a t-norm ts. Therefore one even has a 1-1 relationship between
t-norms and t-conorms.
The main examples for truth degree functions of disjunction connectives
which up to now have been discussed t into this schema. The most popular
examples are the truth degree functions
vel1 (u; v) =def maxfu; vg ;
(5.75)
vel2 (u; v) =def minf1; u + vg ;
(5.76)
vel3 (u; v) =def u + v ; u v ;
(5.77)
which are chosen in such a way that for i = 1; 2; 3 always veli is related
to eti via (5.73). A truth table representation of the disjunction connectives
vel1 ; vel2 is given in the tables of Fig. 5.4.
vel1
1
4
1
41
41
23
4
1
2
1
21
21
23
4
3
4
3
43
43
43
4
vel2
1
4
1
41
23
4
1
2
1
23
4
3
4
3
4
1
0 0
1
1
1
1
41
41
1
1
1
23
23
1
1
1
1
4
4
1 1 1 1 1 1
1 1 1 1 1
Fig. 5.4. Truth table characterizations of disjunction connectives
1
14
32
4
1
41
23
4
1
1
1
1
1
1
The usual names for these truth degree functions parallel the names for
the conjunction connectives: vel2 is the Lukasiewicz (arithmetic) disjunction or bounded sum, and vel3 is the algebraic sum. With the drastic product
t0 and the nilpotent minimum t1 in the same manner the t-conorms

fu; vg; if minfu; vg = 0
s0(u; v) =def max
(5.78)
1
otherwise,
n
fu; vg; if u + v < 1
s1(u; v) =def max
(5.79)
1
otherwise
are connected and correspondingly named drastic sum and nilpotent maximum.
Both these main approaches, via t-conorms and via one of the denitions
(5.71) or (5.72), coincide however essentially as the following result shows.13
Proposition 5.3.1. Suppose that n is a strong negation function and that
two binary operations t; s in [0; 1] are related such that
s(u; v) = n(t(n(u); n(v))) for all u; v 2 [0; 1]
holds true. Then t is a t-norm i s is a t-conorm.
13 The proof is easy and left as an exercise for the reader.
5.4 Implication Connectives
91
But there are also other, extra-logical aspects which sometimes connect
t-norms and t-conorms, e.g. functional equations as we mentioned already in
connection with the introduction of the Frank family of t-norms tF;s . For
them one has, using sF;s to denote the t-conorm related with tF;s via (5.73),
the following result.14
Theorem 5.3.1. A continuous t-norm t and a continuous t-conorm s satisfy
the functional equation
t(u; v) + s(u; v) = u + v for all u; v 2 [0; 1]
(5.80)
i one of the following conditions is satised:
(i) therePexists some s 2 [0; 1] such that t = tF;s and s = sF;s ,
(ii) t = 2I ([ai ; bi ]; tF;si ) is an ordinal sum of t-norms tF;si from the Frank
family
with always si > 0, and s is determined by the ordinal sum
s = P2I ([ai ; bi]; sF;si ).
And all Archimedean solutions of (5.80) have the form (t; s) = (tF;s ; sF;s ) of
t-norms of the Frank family and their corresponding t-conorms.
Via (5.73) also notions and results, like Archimedeanicity or representation theorems, which hold true for t-norms can be transferred to t-conorms.
For details we refer to [170, 309].

The last kind of truth degree functions which needs to be considered separately are the truth degree functions for implication connectives. For biimplication connectives, the fth type of connective usually taken as (more or
less) basic, then the standard approach also in many-valued logic is to take
them as suitable conjunctions of implications.
Historically the rst example of a truth degree function for an implication connective was given by Lukasiewicz [334]. This function seq 2 can be
characterised15 by the equation
seq 2 (u; v) =def minf1; 1 ; u + vg :
(5.81)
This truth degree function as well as its corresponding implication connective
are usually referred to as Lukasiewicz implication.
Another important example was introduced by Go del [199] and may be
characterized by
14 The proof of this theorem needs a lot of machinery from the theory of functional
equations and does not belong to the realm of logic at all. Therefore we omit it.
The interested reader may consult [176] or [309] for this proof.
15 The choice of the index \2" here has a systematic background which shall become
clear soon, cf. (5.83).
92
n
seq 1 (u; v) =def 1; if u v
(5.82)
v otherwise.
As in the case of the previous example, this truth degree function as well
as its corresponding implication connective are usually referred to as Go del
implication.
From a more general point of view, again one either looks for ways to
reduce or relate implication connectives to other ones, or one asks for general
properties which any (suitable) implication connective should satisfy.
First examining ways to relate implications to other connectives, one of
the standard methods in classical logic is to reduce implication to disjunction
and negation, or to conjunction and negation. And indeed this kind of reducibility holds true for the Lukasiewicz implication seq 2 because one has
for all u; v 2 [0; 1]:
seq 2 (u; v) = vel2 (non1 (u); v) = non1 (et2 (u; non1 (v))) :
(5.83)
These results give the motivation to call vel2 the Lukasiewicz (arithmetic)
disjunction and to call et2 the Lukasiewicz (arithmetic) conjunction.
The last type of reduction is impossible for the Go del implication, even
assuming that the negation function n involved in these formulas is a strong
one. For in such a case seq 1 (n(u); v) had to be a commutative binary operation in u; v which obviously is impossible. However, another type of reduction
is possible and one has
seq 1 (u; v) = supfw j et1 (u; w) vg :
(5.84)
And this reduction is a quite important one. That becomes clear if one
has in mind that Go del's paper [199], which introduced the truth degree
function seq 1 , discussed the possible relationships of many-valued and intuitionistic logic: in Heyting algebras, the characteristic structures for intuitionistic logic, the intuitionistic implication is interpreted as the relative
pseudo-complement and thus has a characterization similar to (5.84) which
can equivalently be written as
w seq 1 (u; v) , et1 (u; w) v :
(5.85)
Algebraically, this construction amounts to having via this adjunction property (5.85) the Go del implication as the relative pseudo-complement of the
lattice h[0; 1]; min; maxi, or amounts to having an adjoint pair et1 ; seq 1 .
It is interesting and important to notice that also for the Lukasiewicz
implication (5.81) one has
seq 2 (u; v) = supfw j et2 (u; w) vg :
(5.86)
Therefore also et2 ; seq 2 form an adjoint pair.
These examples (5.84), (5.86) indicate that the (general) adjunction property
w seq (u; v) , et(u; w) v ;
(5.87)
93
i.e. the fact that et; seq form an adjoint pair, may be a characteristic connection between a conjunction connective et and an implication connective
seq .
Straightforward calculations give, using the adjunction property (5.87),
that the truth degree function seq is isotonic, i.e. non-decreasing in the second
argument, and additionally satises the inequalities
et(u; seq (u; v)) v and v seq (u; et(u; v))
(5.88)
16
which code the soundness of the rule of detachment (for the implication
connective based on the truth degree function seq ) and of a kind of rule
of introduction of conjunction. Particularly the availability of the inference
schema of modus ponens, i.e. of the rule of detachment, for each adjoint pair
et; seq makes this approach highly valuable.
One has even more: the adjunction property (5.87) determines the implication function seq for any conjunction function et completely, and conversely
determines also the conjunction function et for any implication function seq
completely.
Proposition 5.4.1. Suppose that et; seq form an adjoint pair. Then one has
for all u; v 2 [0; 1]
(i) seq (u; v) = supfw j et(u; w) vg ;
(ii) et(u; v) = inf fw j seq (u; w) vg :
Proof: (i) Because of seq (u; v) seq (u; v) one has et(u; seq (u; v)) v,
and therefore seq (u; v) supfw j et(u; w) vg. And because one also has
w seq (u; v) for all w 2 [0; 1] with property et(u; w) v, one easily gets
supfw j et(u; w) vg seq (u; v).
(ii) From et(u; v) et(u; v) one has v seq (u; et(u; v)), and hence
et(u; v) inf fw j seq (u; w) vg. And from seq (u; w) v one gets et(u; v)
w, and therefore also et(u; v) inf fw j seq (u; w) vg.
2
Together, the two types of reduction (5.83) on the one hand and (5.84),
(5.86) resp. (5.87) on the other provide standard examples and motivate the
following denition. Sometimes some further types of implication functions
are discussed which shall also be introduced.
Denition 5.4.1. Let some t-norm t, some t-conorm s and some strong
negation function n be given. Then the R-implication function seq t determined by t is dened as
seq t (u; v) =def supfw j t(u; w) vg ;
(5.89)
the S-implication function seq s;n determined by s and n is dened as
seq s;n (u; v) =def s(n(u); v) ;
(5.90)
16 This way of coding is the same as in classical propositional logic and based on
the understanding that the higher truth degrees are the \better" ones.
94
and the QL-implication function17 seq t;s;n determined by t, s, and n is

dened as
seq t;s;n (u; v) =def s(n(u); t(u; v)) :
(5.91)
Some care is necessary, however, with the denition of the R-implications:
in the full generality as in (5.89) an adjoint pair (t; seq t ) results exactly in
the cases where t is a left continuous t-norm. Therefore, this denition of
R-implication is reasonable only for left-continuous t-norms.
Proposition 5.4.2. For any t-norm t one has that this t-norm t and its
R-implication seq t form an adjoint pair i t is left continuous (in both arguments).
Proof: Because of the commutativity condition (T2) it is sucient to
discuss the left continuity of t only for one of its arguments, say the second
one.
Suppose rst that t is left continuous. Then one has to prove that for the
R-implication (5.89) connected with the t-norm t the adjointness condition
(5.87), i.e.
w seq t (u; v) , t(u; w) v
(5.92)
holds true for all u; v; w 2 [0; 1]. Firstly let t(u; w) v. Then one has w 2
fz j t(u; z ) vg and hence w seq t (u; v). If otherwise w < seq t (u; v) holds
true, then there exists some y > w such that t(u; y) v, which gives t(u; w)
v by (T3). If otherwise, however, w = seq t (u; v) holds true, there is either
w 2 fz j t(u; z ) vg and thus immediately t(u; w) v, or there exists a
non-decreasing sequence (yi )i0 such that always yi < w and t(u; yi ) v
hold true, and such that w = limi!1 yi . Then one has by the left continuity
of t immediately t(u; w) = t(u; limi!1 yi ) = limi!1 t(u; yi ) v.
Now conversely let (t; seq t ) be an adjoint pair. Assume furthermore that
t is not left continuous. Then there exist values u; w 2 [0; 1] and a nondecreasing sequence (xi )i0 such that always 0 xi < w holds true together
with w = limi!1 xi , but such that one has t(u; limi!1 xi ) 6= t(u; w), i.e.
t(u; limi!1 xi ) < t(u; w). Then one always has t(u; xk ) t(u; limi!1 xi ).
Thus for any a with t(u; limi!1 xi ) a < t(u; w) one has t(u; xk ) a, hence
xk seq t (u; a) and therefore also w = limi!1 xi seq t (u; a), because one
also has w = supfxi j i 0g by choice of the sequence (xi )i0 . Thus one has
t(u; w) a by the adjointness condition, and a < t(u; w) by choice of a. A
contradiction. Hence the t-norm t has to be left continuous.
2
Corollary 5.4.1. For left continuous t-norms t; t one has for all u; v 2
[0; 1]
17 This name QL-implication reminds an implication connective used in quantum
logic.
95
(i) seq t (u; v) = maxfw j t(u; w) vg ;

(ii) u v , seq t (u; v) = 1 ;
(iii) t 6 t , seq t > seq t :
Proof: (i) One has t(seq t (u; v)) v by the adjointness condition, and
hence seq t (u; v) 2 fw j t(u; w) vg, which means that the supremum in
(5.89) is the maximum.
(ii) One immediately has u v i t(u; v) 1, and hence i seq t (u; v) = 1
by adjointness.
(iii) From t 6 t one immediately has for all u; v 2 [0; 1] the inclusion
fw j t(u; w) vg fw j t (u; w) vg :
And this just means seq t 6 seq t .
So assume seq t 6 seq t . Assume additionally t 66 t . Then there exist
a; b 2 [0; 1] with t (a; b) < t(a; b). Consider some z 2 [0; 1] such that t(a; b) <
z < t (a; b). Then one has immediately b seq t (a; z ) from (5.87).
But one gets also seq t (a; z ) < b. Here seq t (a; z ) b results from the
representation of seq t given by (i), and the fact that t is non-decreasing (in
its second argument) because of
seq t (a; z ) seq t (a; t (a; b)) b :
And the <-relationship comes from the fact that in the case seq t (a; z ) = b
one would have
t(a; b) = t(a; seq t (a; z)) z :
All together this gives seq t (a; z ) < seq t (a; z ), and hence seq t 66 seq t . 2
For the case of continuous t-norms t their representability as ordinal sums
in one of the forms (5.32) or (5.35) oers even the possibility to determine
seq t (almost) explicitly, as the next result (from [117, 182]) shows.
Proposition
5.4.3. Let T be a continuous t-norm given as an ordinal sum
P
([
a
;
b
]
;
t
), and let for each k 2 I be k the order isomorphism (5:33)
i
i
i
i2I
of [0; 1] onto [ak ; bk ]. Then one has for all u; v 2 [0; 1]:
8
;

< k;1 seq tk ( k (u); k (v )) ;
seq T (u; v) = :
if u > v and u; v 2 [ak ; bk ]
seq 1 (u; v) ;
otherwise.
Proof: Because T is continuous, one has seq T (u; v) = 1 i u v by
Corollary 5.4.1(ii). So assume
furthermore that u > S
v.
S
Firstly let (u; v) 2= i2I [ai ; bi ]2 . In the case u 2= i2I [ai ; bi ] one immediately has
seq T (u; v) = supfz j T (u; z ) vg = supfz j minfu; z g vg = v :
96
And in the case u 2 [ak ; bk ] one has v 2= [ak ; bk ], i.e. v < ak . Now having in
mind that for z 2 [ak ; bk ] there is T (u; z ) 2 [ak ; bk ] and thus y < T (u; z ), one
has
seq T (u; v) = supfz j T (u; z ) vg = supfz 2= [ak ; bk ] j T (u; z ) vg
= supfz 2= [ak ; bk ] j minfu; z g vg = v :
So nally assume u > v and u; v 2 [ak ; bk ] for some suitable k 2 I . In this
case one has with z 2 [ak ; bk ] also T (u; z ) 2 [ak ; bk ], and with z 2= [ak ; bk ] also
T (u; z ) = minfu; z g. This rst gives
seq T (u; v) = supfz j T (u; z ) vg ;
and this supremum obviously is the maximum of the following three terms:
supfz 2 [ak ; bk ] j T (u; z ) vg ;
supfz < ak j T (u; z ) vg ;
supfz > bk j T (u; z ) vg :
Because here the second term is not greater than the rst one, and because
the third of these terms is the supremum of the empty set, and hence = 0, it
simply remains
seq T (u; v) = supfz 2 [ak ; bk ] j T (u; z ) vg :
So one has from (5.35)
;

seq T (u; v) = supfz 2 [ak ; bk ] j k;1 tk ( k (u); k (z )) vg
= supfz 2 [ak ; bk ] j tk ( k (u); k (z )) (v)g
= supf; k;1 ( ) j tk ( k (u); ) (v)g
= k;1 supf j tk ( k (u); ) (v)g
;

= k;1 seq tk ( k (u); k (v)) :
And this proves the whole claim.
2
R-implications as well as S-implications generalize well-known relationships of classical and intuitionistic logic. From this point of view it is dicult
to rank these generalizations. There is, nevertheless, an interesting point of
dierence: R-implications of left-continuous t-norms as implications coming
from an adjoint pair satisfy the generalized modus ponens principle in the
form that one always has
t(u; seq t(u; v)) v;
(5.93)
18
the S-implications in general do not have this property.
18 Also in the case of an S-implication the rule of modus ponens can be adapted
in a way which makes it acceptable. However, to reach this goal, in [553] an

additional \modus ponens function" is introduced.
97
Remark: It is interesting to notice that R-implications did not only appear
in the context of intuitionistic logic, but also in connection with fuzzy relational equations and their solutions. Sanchez [477, 478], interested in biggest
solutions of fuzzy relational equations with max-min composition, i.e. of fuzzy
relational equations of the form19
B = fy k 9x(x " A ^ (x; y) " R)g
(5.94)
in which ^ is a conjunction with truth degree function et1 = min, A; B are
given fuzzy sets and R is a fuzzy relation to be determined, introduced a
so-called -operation which is exactly our operation seq 1 . Generalizing this
max-min composition to max-t composition, t some t-norm which in this
generalization is to be used as the truth degree function for the conjunction
connective ^ in (5.94), W. Pedrycz [416] introduced the notion of -operator
't (connected with t) for a binary operation that has to satisfy the isotonicity condition for the second argument together with further characterizing
conditions which exactly correspond to the properties mentioned in (5.88),
and this author [219] used those -operators as implication connectives in
the context of fuzzy set theory. Straightforward calculations yield also in this
case that a (left continuous) t-norm and its corresponding -operator form an
adjoint pair. Therefore these -operators 't are exactly the R-implications
seq t .
Quite another approach, instead of giving these denitions, is it to have
some list of basic characteristic properties of implication operations i which
collects such basic properties which each suitable implication operation should
have. Unfortunately, however, for implication operators actually there does
not exist a commonly agreed such list. Nevertheless, suitable lists have been
considered, e.g. in [124, 545, 510], and extended by dierent authors, cf. [170].
A reasonably complete collection of such properties consists of the following
ones:
1. Left antitonicity : i is non-increasing in the rst argument.
2. Right isotonicity : i is non-decreasing in the second argument.
3. Left boundary condition : i(0; v) = 1 for all v 2 [0; 1].
4. Right boundary condition : i(u; 1) = 1 for all u 2 [0; 1].
5. Normality condition : i(1; 0) = 0.
6. Degree ranking property : i(u; v) = 1 , u v for all u; v 2 [0; 1].
7. Left neutrality : i(1; v) = v for all v 2 [0; 1].
8. Exchange principle : i(u; i(v; w)) = i(v; i(u; w)) for all u; v; w 2 [0; 1].
9. Law of contraposition : i(u; v) = i(n(v); n(u)) for all u; v 2 [0; 1] w.r.t.
some strict negation function n.
The rst ve of these conditions seem to represent a kind of minimal
requirement for suitable truth degree functions for implication connectives.
Hence, following [170], we give the
19 Here we use notions which later on will be explained in Chapter 18.
98
Denition 5.4.2. A function i : [0; 1]2 ! [0; 1] is an implication function

i it satises the conditions of left antitonicity and right isotonicity together
with the left and right boundary and the normality condition.
Not all the properties listed above are independent. One of the interesting
results in this context is the following
Proposition 5.4.4. Any function i : [0; 1]2 ! [0; 1] which satises the condition of right isotonicity, the exchange principle, and which has the degree
ranking property is an implication function which satises also the left neutrality condition.
Proof: The left and the right boundary condition immediately follow
from the degree ranking property.
For the left neutrality condition it suces to prove the two inequalities
v i(1; v) and i(1; v) v for each v 2 [0; 1]. But one has on the one hand
i(v; i(1; v)) = i(1; i(v; v)) = i(1; 1) = 1, hence v i(1; v). And one has on
the other hand i(i(1; v); v) = 1 from i(1; i(i(1; v); v)) = i(i(1; v); i(1; v)) = 1,
i.e. one has also i(1; v) v.
The normality condition is, of course, an immediate corollary of the left
neutrality condition. Thus it remains to show the left antitonicity. So suppose
u u0 . Then one has u0 i(i(u0 ; v); v) because of i(u0 ; i(i(u0 ; v); v)) =
i(i(u0; v); i(u0; v)) = 1, and hence has
1 = i(u; u0) i(u; i(i(u0 ; v); v)) = i(i(u0 ; v); i(u; v)) ;
which means i(u; v) i(u0 ; v). Therefore also the left antitonicity condition
is proved.
2
Even more interesting is the fact that the R-implications as well as the
S-implications have characterizations in terms of these properties, as the next
results show.
Theorem 5.4.1. A function i : [0; 1]2 ! [0; 1] is an R-implication based
on some suitable left continuous t-norm i it satises the right isotonicity
condition as well as the exchange principle, has the degree ranking property
and is itself a right continuous function in the second argument for any rst
argument from [0; 1].20
Proof: Assume rst that i is the R-implication w.r.t. the left continuous
t-norm t, i.e. that one has
i(u; v) = seq t(u; v) = supfz j t(u; z) vg :
20 A partial strengthening of this result was announced in [117] and states that
if even for each u 2 (0; 1) the function iu dened by iu(y) = i(u; y) is strictly
increasing on [0; u], then i is an R-implication which is based on a continuous

t-norm. A closely related result was given in [366], cf. also [309].
99
Then i satises obviously the right isotonicity condition because of the monotonicity condition (T3). And from (T3) and (T4) one gets that i also has the
degree ranking property. One furthermore has
i (u; i(v; w))=supfz j t(u; z) i(v; w)g = supfz j t(v; t(u; z)) wg
=supfz j t(t(u; v); z ) wg = i (t(u; v); w)
(5.95)
and gets thus from the commutativity of t that i satises the exchange principle.
For the right continuity of each one of the functions x 7! i(u; x) it nally
suces to show that one has i(u; inf j!1 vj ) = inf j!1 i(u; vj ) for all nonincreasing sequences (vj )j0 and each u 2 [0; 1]. That means one has to
prove
supfz j t(u; z ) j!1
inf vj g = j!1
inf i(u; vj ) :
(5.96)
Let a = supfz j t(u; z ) inf j!1 vj g. Then one has t(u; a) inf j!1 vj by
the left continuity of t, i.e. always a i(u; vj ), and thus a inf j!1 i(u; vj ).
In the case of a < inf j!1 i(u; vj ), however, there would exist some a <
b < inf j!1 i(u; vj ) with hence always b < vj . Thus there exists always some
zj > b with t(u; zj ) vj . From t(u; b) t(u; zj ) one then gets t(u; b)
inf j!1 vj , and therefore b a by denition of a. A contradiction. Thus
(5.96) is established, and the rst half of the equivalence claim proved.
Now assume conversely that i has all these properties. Then i is an implication function which satises also the left neutrality condition, cf. Proposition 5.4.4.
Remember Proposition 5.4.1(ii) and consider the function eti over [0; 1]
given by
eti (u; v) =def inf fz j i(u; z ) vg :
(5.97)
By denition, eti is non-decreasing in its second argument, and one has always
eti (1; v) = inf fz j i(1; z ) vg = inf fz j z vg = v :
Thus eti satises condition (T4) for t-norms, and it also satises condition
(T3) if it additionally is commutative. But its commutativity results from
the fact that always
i(u; z) v , i(v; z) u ;
which itself follows via the degree ranking property of i from the fact that i
satises the exchange principle:
i(u; z) v , i(v; i(u; z)) = 1
, i(u; i(v; z )) = 1 , i(v; z ) u :
Thus eti satises also condition (T2). And only the associativity (T1) remains
to be shown to know that eti is a t-norm.
100
Because of the commutativity of eti , the associativity of eti is equivalent

with the validity of eti (w; eti (u; v)) = eti (u; eti (w; v)). And this equation is,
according to denition (5.97), satised if the condition
i(w; z) eti (u; v) , i(u; z) eti(w; v)
(5.98)
always holds true. However, one has from (5.97) by the right continuity assumption for i for all u; v the inequality i(u; eti (u; v)) v, and thus also
i(w; z) eti (u; v) ) i(u; i(w; z)) i(u; eti(u; v)) v :
And from the right isotonicity of i one has also
eti (u; i(u; v)) = inf fz j i(u; z ) i(u; v)g v ;
(5.99)
and thus from i(w; z ) eti (u; v) and the exchange principle:
eti (w; v) eti (w; i(u; i(w; z ))) = eti (w; i(w; i(u; z ))) i(u; z ) :
Thus, the direction ()) of (5.98) is proved. And, by symmetry, the other
direction follows in a similar way.
Hence, eti is a t-norm. To prove its left continuity means to prove for each
non-decreasing sequence (vj )j0 and each u 2 [0; 1] the equality
sup eti (u; vj ) = inf fz j i(u; z ) sup vj g :
(5.100)
j !1
j !1
With a = inf fz j i(u; z ) supj!1 vj g one obviously has eti (u; vj ) a for
each j , and thus also supj!1 eti (u; vj ) a. And one has i(u; a) supj!1 vj
by the right continuity property of i. In the case supj!1 eti (u; vj ) < a,
however, there would exist some supj!1 eti (u; vj ) b < a with property
b eti (u; vj ) = inf fz j i(u; z ) vj g and hence with always i(u; b) vj , i.e.
with i(u; b) supj!1 vj and thus a b. A contradiction. This means that
(5.100) holds and that eti is also left continuous.
What remains to be shown is that i is the R-implication w.r.t. the left
continuous t-norm eti . This means to prove i = i for the function i dened
by the equality
i(u; v) =def supfz j eti(u; z) vg :
(5.101)
However, from (5.95) and the fact that eti is a left continuous t-norm one
has
i(eti(u; z); v) = i(z; i (u; v)) ;
and thus: eti (u; z ) v i i (u; v) z . Together with (5.99) one hence gets
i(u; v) i (u; v) for all u; v. Using again the inequality i(u; eti(u; v)) v
and substituting in it i (u; v) for v gives
i(u; v) i(u; eti(u; i (u; v))) i(u; v) :
And both these last inequalities give i = i .
2
Together with the Propositions 5.4.1 and 5.4.2 one therefore has in these
cases
101
i = seq t for the function t = eti ;

t = eti for the function i = seq t :
(5.102)
(5.103)
As a side remark it should be mentioned that these relationships are already
satised if t is left continuous and non-decreasing in its second argument, and
if i is non-decresing and right continuous in its second argument, cf. [534].
Proposition 5.4.5. An implication function is an S-implication function
w.r.t. some strong negation function n and some suitable t-conorm i it satises the exchange principle, the law of contraposition w.r.t. n, and the left
neutrality condition.
Proof: Let i be the S-implication function w.r.t. the t-conorm s and the
strong negation function n, i.e. let i(u; v) = s(n(u); v) for all u; v 2 [0; 1].
Then one has left neutrality because of i(1; v) = s(0; v) = v. One has the
exchange principle because of associativity and commutativity of s and thus
i(u; i(v; w)) = s(n(u); s(n(v); w)) = s(n(v); s(n(u); w)) = i(v; i(u; w)) ;
and one has the law of contraposition because of
i(u; v) = s(n(u); v) = s(v; n(u)) = s(n(n(v)); n(u)) = i(n(v); n(u)) :
Now suppose conversely that the implication function i satises the exchange principle, the law of contraposition w.r.t. the strong negation function
n, and the left neutrality condition.
Dene a binary operation s in [0; 1] by s(u; v) =def i(i(u; 0); v). By
i(u; 0) = i(1; n(u)) = n(u) one then has s(u; v) = i(n(u); v) and therefore
i(u; v) = i(n(n(u)); v) = s(n(u); v) ;
i.e. i is an S-implication, provided s is an t-conorm. And this remains to be
shown.
However, s satises (S2) because of
s(u; v) = i(n(u); v) = i(n(u); n(n(v))) = s(v; u) ;
it satises (S3) because n is order reversing and i non-increasing in the rst
argument, and it satises (S4) because of s(0; v) = i(1; v) = v. Thus the
associativity condition (S1) remains to be proved:
s(u; s(v; w)) = i(n(u); i(n(v); w)) = i(n(u); i(n(w); v))
= i(n(w); i(n(u); v)) = i(n(i(n(u); v); w)
= s(s(u; v); w) :
For continuous R-implications, furthermore, the Lukasiewicz implication seq 2 again is a kind of \prototype" as the following representation theorem shows, cf. [170].
102
Theorem 5.4.2. A function i : [0; 1]2 ! [0; 1] is continuous and satises
the right isotonicity condition as well as the exchange principle, and has the
degree ranking property i there exists an automorphism ' of the unit interval
such that
i(u; v) = ';1 (seq 2('(u); '(v))) for all u; v 2 [0; 1].
(5.104)
Proof: If there exists an automorphism ' of the unit interval such that
i : [0; 1]2 ! [0; 1] has a representation of the form (5.104) then it is a routine
matter to show that i has all the properties mentioned in the theorem.
Thus suppose conversely that i : [0; 1]2 ! [0; 1] is continuous and satises
the right isotonicity condition as well as the exchange principle, and has the
degree ranking property. Consider again the function
eti (x; y) = inf fz j i(x; z ) yg :
As in the proof of Theorem 5.4.1 one shows that eti is a left-continuous
t-norm.
However, it is a routine matter to show that eti is even continuous, because i is supposed to be continuous. What remains to be shown is the right
continuity of eti . But, if eti is not right continuous, there exist a; b 2 [0; 1]
such that
eti (a; b) < hlim
et (a; b + h) :
!+0 i
So for all h > 0 let:

d0 = inf fz j i(a; z ) bg < d < inf fz j i(a; z ) b + hg :
(5.105)
By continuity of i one thus has i(a; d0 ) b, and there exists for each " > 0
some > 0 such that for each 0 < h0 < there holds eti (a; d0 + h0 ) <
eti (a; d0 ) + " . Of course, here always < d ; d0 can be chosen. And this
contradicts (5.105).
According to the following Theorem 5.4.3 a strong negation function is
given by the function n with characterizing equation
n(x) =def i(x; 0) ;
and one immediately has
eti (x; n(x)) = inf fz j i(x; z ) i(x; 0)g = 0 :
Hence by Theorem 5.2.2 there exists an automorphism ' of the unit interval
with
eti (x; y) = ';1 (et2 ('(x); '(y)))
which immediately gives (5.104), because i is the R-implication w.r.t. the
t-norm eti according to Theorem 5.4.1.
2
103
Corollary 5.4.2. An R-implication i is continuous i there exists an automorphism ' of the unit interval such that
i(u; v) = ';1 (seq 2('(u); '(v))) for all u; v 2 [0; 1].
(5.106)
Proof: The results follows immediately from the previous theorem and
Theorem 5.4.1.
2
Despite the fact that we started from truth degree functions for conjunction and negation connectives, mainly because in these cases one widely
agrees on the basic structural properties of such connectives, also (the truth
degree functions of) implication connectives could form the starting point to
derive other types of connectives.
This is most famous in the case of negations, again following the paradigm
set by intuitionistic logic.
Theorem 5.4.3. Suppose that a function i : [0; 1]2 ! [0; 1] satises the right
isotonicity condition and the exchange principle, and has the degree ranking
property. Then the function n : [0; 1] ! [0; 1] which is dened by the equation
n(u) =def i(u; 0) for all u 2 [0; 1]
(5.107)
has the following properties:
(i) n is a negation function.
(ii) id 6 n n.
(iii) n = n n n.
And if n is also continuous, then it is involutive and the law of contraposition
is satised.
Proof: According to Proposition 5.4.4 one has that the function i is an
implication function.
(i) The degree ranking property gives n(0) = i(0; 0) = 1. And from the
fact that i is an implication function one immediately gets n(1) = i(1; 0) = 0,
and that the function n is non-increasing.
(ii) The inequality x n(n(x)) follows by the degree ranking property
from
i(x; n(n(x)) = i(x; i(i(x; 0); 0)) = i(i(x; 0); i(x; 0)) = 1 :
(iii) Because n is non-increasing one has from (ii) that n(x) n(n(n(x)))
always holds. So the converse inequality remains to be shown. But this follows
via the degree ranking property from
i(n(x); n(n(n(x)))) = i(i(x; 0); i(i(i(x; 0); 0); 0))
= i(i(i(x; 0); 0); i(i(x; 0); 0)) = 1 :
And if n is also continuous then there exists for each x 2 [0; 1] some y 2 [0; 1]
with n(y) = x. Then one always has n(n(x)) = n(n(n(y))) = n(y) = x, i.e.
n is involutive. And one also has
104
i(x; y) = i(x; n(n(y))) = i(x; i(i(y; 0); 0))

= i(i(y; 0); i(x; 0)) = i(n(y); n(x)) ;
i.e. the law of contraposition is satised too.

2
In the particular case of the Lukasiewicz implication, by (5.107) one
gets the negation function non1 :
non1 (u) = seq 2 (u; 0) for all u 2 [0; 1].
(5.108)
And the same type of approach yields via
nont (u) = seq t (u; 0) for all u 2 [0; 1]
(5.109)
a negation function for each (left continuous) t-norm. Furthermore, even the
Lukasiewicz (arithmetic) disjunction and conjunction functions are determined by the Lukasiewicz implication:
vel2 (u; v) = seq 2 (non1 (u); v) for all u; v 2 [0; 1],
et2 (u; v) = non1 (seq 2 (u; non1 (v))) for all u; v 2 [0; 1].
In general, having given some left continuous t-norm t one has its corresponding R-implication seq t given by (5.89) and can introduce a further
unary truth degree function by
nont (u) =def seq t (u; 0)
for all u 2 [0; 1].
(5.110)
These functions nont obviously have the properties
nont (0) = 1 and
nont (1) = 0 ;
and are non-increasing, hence they are negation functions. These functions
nont are the t-norm based negation functions.
For the particular case that the left continuous t-norm t does not have zero
divisors, i.e. that t(u; v) = 0 holds only under the condition u = 0 _ v = 0,
one easily gets

u=0
nont (u) = non0 (u) = 10; ifotherwise,
i.e. the negation functions which are based on (left continuous) t-norms without zero divisors coincide with the Go del negation function.
5.5 The J{Connectives

In classical logic one is able to code the fact that some sentence is true by
claiming that it is equivalent with a denitely true sentence, and the same
can correspondingly be done for false sentences.
A kind of substitute in many-valued logic for this possibility in classical
logic is provided by a family of unary connectives which indicate that the
sentences they are applied to have some predetermined truth degree.
5.5 The J{Connectives
105
These particular connectives are usually denoted Js , s 2 W any truth

degree, and correspond to truth degree functions js dened by
n
js (u) =def 1; if u = s
(5.111)
0 otherwise.
The interest in these connectives additionally comes from a rather uniform
method of adequate axiomatization of a whole class of nitely many-valued
systems, cf. Chapter 6, and also from their relationship with the problem of
functional completeness, cf. Section 8.1.
106
6. Axiomatizability
6.1 The Axiomatizability Problem

Like modern, formal logic in general also many-valued logic seeks to present
its systems not only semantically but also syntactically in the form of logical
calculi. Such a logical calculus is a purely syntactical method to generate
ws of some predetermined language, and even a method which may be
executed \purely mechanically", i.e. which is an algorithm. Our logical calculi
in each case shall be calculi for some xed system S of many-valued logic, and
therefore their predetermined language always shall be the language LS of
this system. And within LS these logical calculi are used to produce syntactic
transformations and derivations of ws.
In order to generate ws of a suitable language for such logical calculi the
following kinds of data should be available:
(K1) some { possibly empty { set of ws, called axioms of the logical calculus,
as a starting point for the particular generation processes of (further)
ws;
(K2) rules which allow the transition from some given ws to other ones {
the derivation rules of this logical calculus;
(K3) perhaps additional restrictions concerning the situations in which some
of these derivation rules are allowed to be applied.
Because logical calculi are supposed to be purely syntactically determined
methods, the rules and restrictions of (K2) and (K3) also have to be of a purely
syntactical nature, i.e. they have to refer to formal, syntactical properties of
the ws involved, but are not allowed to refer in any sense to their \meaning",
their semantical status. Because furthermore logical calculi are supposed to
determine algorithms for the generation of ws, the sets of axioms and of
rules of such a logical calculus have to be decidable sets, and the syntactic
properties of ws referred to in these derivation rules and in the restricting
conditions for their applicability have to be decidable (syntactic) properties
too.
The ws which can be generated by a logical calculus K are called derivable in K or simply K -derivable or also theorems of K . Usually one writes
`K H to indicate that the w H is derivable in the logical calculus K . Ad-
108
6. Axiomatizability
ditionally `K is called the notion of derivability or the derivability relation

determined by the logical calculus K subject to the conditions (K1) to (K3).
As in the case of the problem of decidability, also here one is mainly
interested in the set of logically valid ws of a given propositional system S
of many-valued logic. And for this set one likes to have a logical calculus which
generates as its theorems just these logically valid ws, i.e. which generates as
theorems only logically valid ws, but which generates all of them. Similarly
one could ask for logical calculi which generate (as its theorems) just the set
of all S-satisable ws or just the set of all S-contradictions { but both these
additional problems already in classical logic are mainly neglected, and the
same holds true for the eld of many-valued logic.
If one is not starting from a semantically determined system S of (manyvalued) logic, as done in this book, one just uses a logical calculus K for
a syntactic determination of a logical system. Then, similarly, the problem
arises to nd a suitable semantic which has as a logically valid w just the
K -derivable formulas.
A (semantically determined) system S of propositional many-valued logic
is called axiomatizable i there exists a logical calculus K which generates
just the logically valid ws of S; and such a calculus is said to realize an
adequate axiomatization of the system S. The property that K generates
only S-logically valid ws is the soundness of K , and the property that K
generates all S-logically valid ws is called the completeness of K .
Principally the problem of nding a suitable logical calculus K which provides an adequate axiomatization for some xed system S of propositional
many-valued logic is a particular one for this system. It would be quite welcome, however, to have some more general and uniform way to axiomatize
logical systems of some type.
Additionally there are, in classical logic as well as in the eld of nonclassical logics, rather dierent approaches toward logical calculi. Historically
the rst, and still the best known logical calculi are Hilbert-type calculi
which are determined by some set of ws which act as axioms, i.e. as formulas
which are \given" from the very beginning, together with some set of inference
rules which allow to derive ws out of previously given ones. This type of
calculi is discussed in this section. Later on in Chapter 7 also tableau calculi
as well as Gentzen-type sequent calculi and calculi of natural deduction
shall be considered.
6.2 Axiomatizing Propositional Systems

An approach toward Hilbert-type axiomatizations was oered for a wide
class of systems S of many-valued logic with nite truth degree sets of the
form W S = Wm ; m 2, by J.B. Rosser and A.R. Turquette in [465].
They assume that the set J S of connectives of S shall contain a binary
connective !, denoting a kind of implication connective, and unary connec-
109
tives Js for each s 2 W S , or at least that such connectives are denable from
the primitive connectives of the system S.
These connectives have to satisfy the following conditions:
(RT1) The truth degree function corresponding to the connective ! assumes
a non-designated truth degree just in the case that its rst argument is
a designated truth degree and its second argument is a non-designated
one, i.e. ! satises the standard condition of an implication connective.
(RT2) The unary truth degree function corresponding to the connective Js ,
s 2 W S , assumes a designated truth degree just in the case that their
argument value is s, and a non-designated truth degree in all other cases.
For a compact formulation of the axioms a sort of nite iteration of the
connective ! is needed. Therefore we dene for any well-formed formulas
H1 ; H2 ; : : : ; G of the language LS recursively:
i! (Hi; G) =
k
i! (Hi; G) =
0
=1
+1
=1
def
def
G;

i k
(6.1)

Hk+1 ! ! (Hi ; G) :
=1
(6.2)
Additionally, for each connective ' 2 J S its corresponding truth degree

function shall be ver' .
Now let AxRT be the set of all well-formed formulas falling under one of
the following schemata:
AxRT 1 : A ! (B ! A) ,
AxRT 2 : (A ! (B ! C )) ! (B ! (A ! C )) ,
AxRT 3 : (A ! B ) ! ((B ! C ) ! (A ! C )) ,
AxRT 4 : (Js (A) ! (Js (A) ! B )) ! (Js (A) ! B ) for each s 2 W S ,
m
AxRT 5 : ! i=1 (J mi;; (A) ! B; B ) ,
AxRT 6 : Js (s) for each truth degree s and each truth degree constant
s denoting it,
AxRT 7 : Jt (A) ! A for each designated truth degree t ,
n
AxRT 8 : ! i=1 (Jsi (Ai ); Jt ('(A1 ; : : : ; An ))) for each n-ary connective
' 2 J S , for all s1 ; : : : ; sn 2 W S , and for the particular truth
degree t = ver' (s1 ; : : : ; sn ).
Here A; B; A1 ; : : : ; Am are ws of the language LS . This logical calculus
K mRT has a single rule of inference, the rule of detachment
1
1
(MP)
A; A ! B
B
which allows to infer a w B given two ws of the forms A and A ! B .
110
6. Axiomatizability
There is no extra restriction concerning the applicability of this rule (MP).

A particular w H 2 LS is called derivable in the logical calculus K m
RT or a theorem of K m
,
denoted:
`
H
,
i
there
exists
a
nite
sequence
H
;
RT
1 H2 ; : : : ; Hn
RT
of ws of LS such that
(D1) Hn is the w H ;
(D2) each w Hk of this sequence is either an axiom of K m
RT or the result of
an application of the inference rule MP to preceding ws of this sequence.
And each such sequence is called a derivation or a proof of H (in the logical
calculus K m
RT ).
Related to this notion of derivability `RT obviously axiom schema (AxRT 8)
codes for each connective ' 2 J S its truth degree behavior, i.e. essentially its
truth degree function. In a similar manner axiom schema (AxRT 7) codes that
each w with (provably) designated truth degree is derivable in K m
RT . And
axiom schema (AxRT 6) codes for a truth degree constant t, which denotes the
truth degree t 2 W S , that one can derive that t denotes t.
Theorem 6.2.1 (Soundness Theorem). Suppose that some propositional
system S of many-valued logic satises the conditions (RT1) and (RT2). Then
m
the logical calculus K m
RT is sound w.r.t. this system S, i.e. each K RT -derivable
w is an S-tautology.
Proof: For proving this soundness result it suces to show that all the
m
axioms of K m
RT are S-tautologies and that each inference rule of K RT leads
from S-tautologies only to an S-tautology. And to show this last property
for the rule of detachment (MP) it suces that the following condition holds
true:
(I ) For any ws H; G, if H has a designated truth degree and G has a
non-designated truth degree, then H ! G has a non-designated truth
degree.
Because from (I ) one immediately gets that it is impossible that at the
same time ws H; H ! G have designated truth degrees and G has a nondesignated truth degree, and hence the conclusion which results from an
application of rule (MP) to S-tautologies has to be also an S-tautology.
But condition (I ) holds true in the present case because it is an immediate
corollary of condition (RT1).
Before we test all axioms for being S-tautologies
we remind ourselves that
n
!
all the axioms of K m
are
ws
of
the
form
(
H
i
RT
i=1 ; G), and that according
to denition (6.2) one therefore has for each n 2:
i!n (Hi; G)
=1
n;1
is the w Hn ! ! (Hi ; G) :
=1
According to the standard condition (I) we have supposedn to hold true for
the (implication like) connective !, therefore to give ! i=1 (Hi ; G) a nondesignated truth degree means hat Hn has to have a designated truth degree,
n ;
111
and that ! i=1 (Hi ; G) has to have a non-designated truth degree. Inductively hence one proves the following
n
Claim : To give ! i=1 (Hi ; G) a non-designated truth degree means that all
the ws H1 ; : : : ; Hn have to have designated truth degrees, and that G has
to have a non-designated truth degree.
Now let H be an axiom of the logical calculus K m
RT . If H is an instance
of axiom schema (AxRT 1) then it has the form A ! (B ! A). Hence to give
H a non-designated truth degree means that A; B have to have designated
truth degrees and that A has to have a non-designated truth degree. And
this, of course, is impossible.
If H is an instance of the axiom schema (AxRT 2) then it has the form
(A ! (B ! C )) ! (B ! (A ! C )). To give H a non-designated truth
degree therefore means that the ws A ! (B ! C )); B; A have to have
designated truth degrees, and that C has to have a non-designated truth
degree. But then B ! C has a non-designated truth degree by (I ) and
A ! (B ! C ) too, a contradiction.
In the case that H is an instance of schema (AxRT 3) and should have a
non-designated truth degree one has to have ws A ! B; B ! C and A with
designated truth degrees such that C has a non-designated truth degree. But
then also B has to have a non-designated truth degree and hence also A,
again a contradiction.
If H is an instance of axiom schema (AxRT 4) and should have a nondesignated truth degree, some w Jt (A) ! B has to have a non-designated
truth degree according to the standard condition (I). But then Jt (A) has
to have a designated truth degree and B a non-designated one. Therefore
Jt (A) ! (Jt (A) ! B ) gets a designated truth degree and H too. Hence H
has to be an S-tautology.
If H is an instance of axiom schema (AxRT 5) and has a non-designated
truth degree, then some w B has to have a non-designated truth degree
and at the same time all ws Jt (A) ! B for t 2 W S = Wm have to have
designated truth degrees. However, for any valuation exactly one of the ws
Jt (A) has a designated truth degree, and therefore B has to have a designated
truth degree too, a contradiction.
If H is an instance of one of the axiom schemata (AxRT 6) or (AxRT 7) then
obviously H is an S-tautology.
And, nally, if H is an instance of schema (AxRT 8) then to give H a
non-designated truth degree means that for some ws A1 ; : : : ; Ak and truth
degrees t1 ; : : : ; tk all the ws Jti (Ai ) have to have designated truth degrees,
and the w Js ('(A1 ; : : : ; Ak )) with s = verS'(t1 ; : : : ; tk ) has to have a nondesignated truth degree. But this again is impossible.
Therefore all K m
RT -axioms are S-tautologies and our proof is nished. 2
Theorem 6.2.2 (Completeness Theorem). Suppose that some proposi-
tional system S of many-valued logic satises the conditions (RT1) and (RT2).
112
6. Axiomatizability
Then the logical calculus K m

RT is complete w.r.t. this system S, i.e. each Stautology is K m
RT -derivable.
Proof: We start the proof with some particular properties of the derivability relation `RT which become important for later arguments. A very
simple observation is that by the rule of detachment (MP) one immediately
has for any ws A; B; C :
if `RT A and `RT A ! B , then `RT B .
(6.3)
m
To see this, let the sequence H1 ; : : : ; Hn be a K RT -derivation of A and let
G1 ; : : : ; Gr be a K mRT -derivation of A ! B , then the sequence
H1 ; : : : ; Hn ; G1 ; : : : ; Gr ; B
is a K m
RT -derivation of B whose last w B results from the previous ws
Hn ; Gr by an application of (MP).
Applying (6.3) twice to axiom schema (AxRT 3) gives that the rule of
syllogism holds true in the form:
if `RT A ! B and `RT B ! C , then `RT A ! C
(6.4)
for any ws A; B; C . And from axiom schema (AxRT 2) one gets accordingly
if `RT A ! (B ! C ), then `RT B ! (A ! C ).
(6.5)
And applying this last observation (6.5) to schema (AxRT 3) itself, one gets
by (6.3) immediately
`RT (A ! B ) ! ((C ! A) ! (C ! B )) ;
(6.6)
which means, among others, that it holds true:
if `RT A ! B , then `RT (C ! A) ! (C ! B ).
(6.7)
m
It is, however, in K RT not only possible to put derivably in front of antecedent
and succedent of an implication A ! B a further antecedent C as in (6.6),
(6.7), but instead of C one can take a whole sequence of ws. This means,
formally, that we have the following
Claim 1 : For any ws A; B; H1 ; H2 ; : : : and each integer n it holds
i n
=1
i n
(6.8)
i k
i k
(6.9)
;

`RT (A ! B ) ! ! (Hi ; A) ! ! (Hi ; A) :
=1
In the case n = 1 this is just the previous result (6.6). For n > 1 a simple
proof proceeds inductively with n = 1 as basis case. Thus suppose that (6.8)
holds true for n = k, i.e. suppose
;

`RT (A ! B ) ! ! (Hi ; A) ! ! (Hi ; A) :
Then by (6.6) one gets
=1
=1
i k
i k
k
k
! (Hi ; A)) ! (Hk !
! (Hi ; B )) ;
!
i
i
113
`RT ( ! (Hi ; A) ! ! (Hi ; B ))

=1
=1
! ((Hk+1
+1
=1
and from this and denition (6.2) also
i k
=1
i k
k
k
! (Hi ; A) !
! (Hi ; B )) :
! (
i
i
(6.10)
`RT ( ! (Hi ; A) ! ! (Hi ; B ))

=1
=1
+1
+1
=1
=1
(6.11)
Combining now (6.9) and (6.11) one gets according to (6.4) the result (6.8)
for n = k + 1. Hence Claim 1 is proved for all n 1. It remains to prove this
claim for the case n = 0, i.e. to show
`RT (A ! B ) ! (A ! B ) :
(6.12)
This is, however, a particular case of a more general fact, viz. that
`RT C ! C
(6.13)
holds true for any w C . And to show this we start from an instance of axiom
schema (AxRT 2):
`RT (C ! (D ! C )) ! (D ! (C ! C ))
(6.14)
which has an instance of axiom schema (AxRT 1) as its antecedent. Hence we
can apply (MP) via (6.3) and get
`RT D ! (C ! C ) :
Choosing D as any axiom of K m
RT allows again to apply (6.3) and yields
(6.13). Thus (6.8) is also proved for n = 0.
Before we proceed, a side remark is of interest. The transition we made
from (6.10) to (6.11) via (6.6) shall also be used in other contexts later on.
Then we shall avoid to write down the intermediate formula corresponding
to (6.10) and only refer to applications of remark (6.6) and denition (6.2).
To supply the details then is an easy exercise for then reader.
Having another look at the ws of the kind ! i=1 (Hi ; G) as dened in
n
(6.1), (6.2) one can interpret the \iterated antecedents" ! i=1 (Hi ; : : :) as
a unary operator on ws which itself depends on ws H1 ; : : : ; Hn as parameters. Such operators are to play a crucial role in the nal part of this
proof. Therefore we derive here two further properties of them { on the one
hand showing that parameters in these operators may be \redistributed and
added", and on the other hand showing that some of these operators may be
\distributed" over antecedent and succedent of an implication.
114
6. Axiomatizability
Claim 2 : Suppose that the sequence (A1 ; : : : ; Ar ) of ws is a rearranged subsequence of the sequence (H1 ; : : : ; Hn ) of ws. Then it holds true for any w
G that
i k
=1
j n
(6.15)
i l
=1
i k
(6.16)
i r
j k
(6.17)
i r
k+1
(6.18)
j k
k+1
(6.19)
`RT ! (Ai ; G) ! ! (Hj ; G) :

=1
For (A1 ; : : : ; Ar ) to be a rearranged subsequence of (H1 ; : : : ; Hn ) means

that there exists a permutation of the set f1; : : :; rg, i.e. a 1-1 map of this
set onto itself, such that (A(1) ; : : : ; A(r) ) is a subsequence of (H1 ; : : : ; Hn ).
Of course, in this case r n has to hold true.
Again we shall prove (6.15) by induction over n. In the case n = 0, Claim 2
simply states: `RT G ! G, and thus is just (6.13). Therefore we now assume
that (6.15) holds true for n = k, i.e. we assume
`RT ! (A0i ; G) ! ! (Hj0 ; G)

=1
for any rearranged subsequence (A01 ; : : : ; A0l ) of a sequence (H10 ; : : : ; Hk0 ). So

let (A1 ; : : : ; Ar ) be a rearranged subsequence of the sequence (H1 ; : : : ; Hk+1 )
of length k + 1. We have to show
`RT ! (Ai ; G) ! ! (Hj ; G) :

=1
=1
In the case that (A1 ; : : : ; Ar ) is a rearranged subsequence of (H1 ; : : : ; Hk ),

the assumption (6.16) yields
`RT ! (Ai ; G) ! ! (Hj ; G) ;

=1
=1
and axiom schema (AxRT 1) together with denition (6.2) furthermore
`RT ! (Hj ; G) ! ! (Hj ; G) :

=1
=1
Therefore (6.15) with n = k + 1 now results via (6.4) from (6.18) and (6.19).
Otherwise there exists an index u k + 1 with Ar Hu . And then the
sequence (H1 ; : : : ; Hu;1 ; Hu+1 ; : : : ; Hk+1 ) of length k has (A1 ; : : : ; Ar;1 ) as
a rearranged subsequence. Therefore we have according to (6.16)
ri;

k+1
`RT ! (Ai ; G) ! ! (Hj ; G) :

=1
(6.20)
j =1;j 6=u
Taking now C Ar Hu in (6.7), and choosing in the case u < k + 1 e.g.

the index v = k + 1, or in the case u = k + 1 the index v = 1, yields
i r
;
`RT ! (Ai ; G) ! Hu ! Hv !
=1
! (Hj ; G)
j ;j 6 u;v
k+1
=1 =

(6.21)
115
because of (6.2). And by (6.5) one gets
`RT (Hu ! (Hv !
!
k+1
j =1;j 6=u;v
! (Hv ! (Hu !
(Hj ; G)))
!
k+1
j =1;j 6=u;v
(Hj ; G))) :
(6.22)
From this and (6.16) one has because of (6.2) also

;
`RT Hu !
k
! (Hj ; G) !
! (Hj ; G) ;

j ;j 6 u;v
j ;j 6 v
k+1
+1
=1 =
=1 =
(6.23)
and thus nally because of (6.7) and (6.2)

;
`RT Hv ! (Hu !
k
! (Hj ; G)) !
! (Hj ; G) :

j
j ;j 6 u;v
k+1
+1
=1
=1 =
(6.24)
Now one has to apply the rule of syllogism (6.4) to (6.21) and (6.22), and
then to the result of this application and (6.24). This gives (6.17) also in this
case.
Hence Claim 2 is proved, and we can proceed to the next claim.
Claim 3 : For all ws A; B; H1 ; H2 ; : : : and all truth degrees t1 ; t2 ; : : : one has
for each integer n
i n
`RT ! (Jti (Hi ); A ! B )

=1
i n
i n
;

! ! (Jti (Hi ); A) ! ! (Jti (Hi ); B ) :
=1
=1
(6.25)
Once again the proof shall be given by induction on n. For n = 0 the

claim reduces because of (6.13) to `RT (A ! B ) ! (A ! B ), i.e. to an
instance of (6.13). Therefore we assume (6.25) for n = k and go to prove it
for n = k + 1. As a rst step we apply (6.7) to (6.25) with n = k, choose
particularly C Jtk (Hk+1 ), and get
+1
k+1
;
`RT ! (Jti (Hi ); A ! B ) ! Jtk (Hk+1 ) !
=1
i k
+1
i k

! ( ! (Jti (Hi ); A) ! ! (Jti (Hi ); B ))
=1
=1
(6.26)
according to (6.2). Next it proves to be helpful later on to know that for any
ws H; G1 ; G2 and truth degrees s it holds
`RT (Js (H ) ! (G1 ! G2 ))
! ((Js (H ) ! G1 ) ! (Js (H ) ! G2 )) :
(6.27)
116
6. Axiomatizability
To get this fact we start from an instance of axiom schema (AxRT 2) and thus
have
`RT (Js (H ) ! (G1 ! G2 )) ! (G1 ! (Js (H ) ! G2 )) ;
(6.28)
and because of (6.6) then
`RT (G1 ! (Js (H ) ! G2 ))
! ((Js (H ) ! G1 ) ! (Js (H ) ! (Js (H ) ! G2 ))) :
(6.29)
On the other hand we get from axiom schema (AxRT 4)
`RT (Js (H ) ! (Js (H ) ! G2 )) ! (Js (H ) ! G2 ) ;
(6.30)
and hence by (6.7) also
`RT ((Js (H ) ! G1 ) ! (Js (H ) ! (Js (H ) ! G2 )))
! ((Js (H ) ! G1 ) ! (Js (H ) ! G2 )) :
(6.31)
Now we apply the rule of syllogism (6.4) to (6.28) and (6.29), and again to
the result of this application and (6.13), which gives (6.27).
Writing D for the succedent of the implication whose K m
RT -deducibility
was established in (6.26), an application of (6.27) gives via (6.2)
k+1
k+1
=1
=1
`RT D ! ! (Jti (Hi ); A) ! ! (Jti (Hi ); B ) :
(6.32)
Applying now (6.4) to (6.26) and (6.32) yields (6.25) for the case n = k + 1.
Hence our inductive proof of Claim 2 is complete.
For our forth (and last) claim we suppose that H is a w of LS with its
propositional variables among q1 ; : : : ; qn . Then we know from Chapter 2 that
H describes a truth degree function w~H (x1 ; : : : ; xn ) such that
ValS (H; ) = w~H ( (q1 ); : : : ; (qn ))
(6.33)
holds true for each valuation .
Claim 4 : For any truth degrees t1 ; : : : ; tn one has
i n
`RT ! (Jti (qi ); Jw~H (t ;:::;tn) (H )) :

1
=1
(6.34)
Once again the proof is by induction, but this time by induction on the
complexity of the w H . The basic cases, therefore, are the cases that H is a
truth degree constant or a propositional variable. So rstly let H be the truth
degree constant t denoting the truth degree t 2 W S . Then we have `RT Jt (t)
by axiom schema (AxRT 6). But this is already (6.34) for n = 0. From (6.15)
with r = 0 we get furthermore
i n
`RT Jt (t) ! ! (Jti (qi ); Jt (t)) ;

=1
(6.35)
which via an application of (6.3) gives (6.34) in general for this rst case.
117
As a second basic case let H be the propositional variable qk for some

1 k n. Then one has for the corresponding truth degree function
w~H (x1 ; : : : ; xn ) = xk . And from (6.15) one gets
i n
`RT ! (Jtk (qk ); Jtk (qk ) ! ! (Jti (qi ); Jtk (qk )) :

=1
=1
But because of
i! (Jt (qk)Jt (qk)) (Jt (qk) ! Jt (qk)) ;

1
=1
(6.36)
(6.37)
according to (6.2), the antecedent of the implication considered in (6.36) is

an instance of (6.13). Hence we can apply (6.3) in this situation, and we get
(6.34) for this case.
So nally H shall be a compound formula H '(H1 ; : : : ; Hr ) for some rary connective ' 2 J S . Then also for each one of the ws Hj the propositional
variables of Hj are among q1 ; : : : ; qn . The assumption of our inductive proof
now is that (6.34) holds true for all these ws H1 ; : : : ; Hr . For simplicity of
notation let always hj (x1 ; : : : ; xn ) be the truth degree function described by
Hj . Then obviously
w~H (x1 ; : : : ; xn ) = verS' (h1 (x1 ; : : : ; xn ); : : : ; hr (x1 ; : : : ; xn ))
(6.38)
holds true for all x1 ; : : : ; xn , and from axiom schema (AxRT 8) we get
j r
`RT ! (Jhj (t ;:::;tn ) (Hj ); Jw~H (t ;:::;tn ) (H )) :

1
=1
(6.39)
With the auxiliary notation
i k
Dk ! (Jhj (t ;:::;tn ) (Hj ); Jw~H (t ;:::;tn ) (H ))

=1
for any k r we can write (6.39), according to (6.23), shorter as

`RT Jhr (t ;:::;tn) (Hr ) ! Dr;1 :
By the assumption we made for this induction step we now have
1
i n
`RT ! (Jti (qi ); Jhj (t ;:::;tn ) (Hj ))

=1
(6.40)
(6.41)
(6.42)
for each 1 j r. Therefore we get from (6.8) and (6.41) by detachment

according to (6.3)
i n
i n
`RT ! (Jti (qi ); Jhr (t ;:::;tn) (Hr )) ! ! (Jti (qi ); Dr;1) ;

=1
=1
(6.43)
and furthermore from (6.42), with j = r, and (6.43) again via (6.3) the
derivability statement
118
6. Axiomatizability
i n
`RT ! (Jti (qi ); Dr;1) :
(6.44)
=1
For r = 1 this is already (6.34). And for r > 1 we can write Dr;1 via (6.2)
as implication and can apply (6.25). Another application of (6.3) then yields
because of (6.44)
i n
i n
`RT ! (Jti (qi ); Jhr; (t ;:::;tn) (Hr;1 )) ! ! (Jti (qi ); Dr;2) ;

1
=1
=1
(6.45)
thus (6.42), with j = r ; 1, and detachment according to (6.3) leads to
i n
(6.46)
i n
(6.47)
`RT ! (Jti ; Dr;2) :

=1
We now iterate the transition from (6.44) to (6.46) another r ; 2 times. This
nally gives
`RT ! (Jti (qi ); D0 )

=1
and this is (6.34) because of D0 Jw~H (t ;:::;tn ) (H ).

Thus we have (6.34) also for compound formulas, and therefore Claim 3
is proved.
Now all the preliminaries are ready to nish the whole proof. Therefore
suppose now that H is an S-tautology which contains exactly the propositional variables q1 ; : : : ; qn . Then the truth degree function wH (x1 ; : : : ; xn )
described by H maps into DS , hence according to (AxRT 7) we have for any
truth degrees t1 ; : : : ; tn
`RT JwH (t ;:::;tn) (H ) ! H;
(6.48)
which together with (6.8) via (6.3) yields
1
j n
j n
`RT ! (Jtj (qj ); JwH (t ;:::;tn) (H )) ! ! (Jtj (qj ); H ) :

1
=1
=1
(6.49)
By (6.34) another application of (6.3) gives
j n
`RT ! (Jtj (qj ); H ):

=1
(6.50)
For shortness of notation we introduce as a further abbreviation for any

k n:
j k
Ck =def ! (Jtj (qj ); H ) ;

=1
and write formula (6.50) not as `RT Cn but equivalently as

`RT Jtn (qn ) ! Cn;1 :
From axiom schema (AxRT 5) we get
(6.51)
(6.52)
im
`RT ! (J mi;; (qn ) ! Cn;1 ; Cn;1 )
(6.53)
1
1
=1
119
respectively, according to (6.2),
m;1
`RT (J1 (qn ) ! Cn;1 ) ! ! (J mi;; (qn ) ! Cn;1 ; Cn;1 ) :

1
1
=1
(6.54)
Because tn may be any truth degree in (6.52), we choose tn = 1 and apply

(6.3) to (6.52) and (6.54). Via (6.2) this yields
m;2
`RT (J mm;; (qn ) ! Cn;1 ) ! ! (J mi;; (qn ) ! Cn;1 ; Cn;1 ) :
(6.55)
(6.56)
2
1
=1
1
2
;2
Now we put tn = m
m;1 within (6.52), and apply (6.3) to it and to (6.55). The
result is, again according to (6.2),
m;3
`RT (J mm;; (qn ) ! Cn;1 ) ! ! (J mi;; (qn ) ! Cn;1 ; Cn;1 ) :

3
1
=1
1
1
The transition from (6.54) to (6.55), respectively from (6.55) to (6.56), we

have to iterate m times. The result of all these iterations is
`RT Cn;1 :
(6.57)
Now there is also another, \larger" transition we have to iterate: the
transition from (6.50) to (6.57), i.e. the transition from `RT Cn to `RT Cn;1 .
After n ; 1 further steps of this iteration we get `RT C0 , and this means
nothing but `RT H according to (6.51).
m
Therefore each S-tautology is K m
RT -derivable, i.e. the logical calculus K RT
is complete for S.
2
Looking back at these proofs of the Soundness and the Completeness
Theorem, one recognizes that the proof of the Completeness Theorem does
not use any semantical property of the implication connective !, particularly
it does not use the standard condition (I). It also does not use any semantical
property of the connectives Jt . And for the proof of the Soundness Theorem
it was only important to have satised the conditions:
(E1) The implication connective ! satises condition (I ) from p. 110.
(E2) Each axiom from (AxRT 1),. . . , (AxRT 8) is an S-tautology.
It was solely the proof of (E2), i.e. the proof that each K m
RT -axiom is an
S-tautology, which used the assumptions (RT1), (RT2) concerning semantic
properties of ! and all the Jt . Thus we get even the following result.
Theorem 6.2.3. The logical calculus K mRT provides an adequate axiomatization1 for each propositional system S0 of many-valued logic with m truth0
degrees which has in its alphabet connectives ! and Jt , for each t 2 W S ,
1 This means that soundness and completeness hold true.
120
6. Axiomatizability
which satisfy conditions (E1) and (E2) { or in which such connectives are
denable.
We shall see later on for the Lukasiewicz m-valued propositional systems that the implication of these systems does not satisfy the standard
condition (I), but that nevertheless they can be adequately axiomatized by
the corresponding calculi K m
RT , thanks to Theorem 6.2.3.
This Hilbert-type axiomatization method of Rosser-Turquette hence
is suitable for a large class of nitely many-valued propositional logics. Its
main drawback, however, is that it needs besides the implication connective
! the whole class of all the connectives Jt for each truth degree t. This is
not a severe restriction as long as the set of connectives together with the set
of truth degree constants form a functionally complete system (cf. Chapter
8.1 for this problem of functional completeness). However, a lot of interesting systems of propositional many-valued logic prove to be not functionally
complete. Therefore it would be quite welcome to have (also) a method of
axiomatization which does not presuppose these restrictions.
Such a method was rst described by K. Schro ter [485]. And this
method is also suited to formalize the relation of entailment. We discuss
it later on in Chapter 7.
6.3 Axiomatizing First-Order Systems

In this section we consider only systems S of rst-order many-valued logic
which have a nite set of truth degrees of cardinality m which even shall be of
the form W S = Wm for some m 2. This method of axiomatization extends
the Rosser-Turquette approach toward the axiomatization of propositional systems, given in the previous section, to the rst-order case, and was
also given in [465]. And as for this propositional case one of the crucial points
for the choice of the axioms was the schema AxRT 8, which represents inside
the set of axioms the behavior of the truth degree functions of the connectives of S; also for the rst-order case a quite important point shall be to nd
a way to represent the truth degree behavior of the quantiers of S in the
axioms.
Again we assume that the set J S of connectives of S shall contain an
implication connective !, which satises the standard condition of an implication connective, and also the unary connectives Js for each s 2 W S , or
at least that such connectives are denable from the primitive connectives
of the system S. Furthermore Vwe suppose that there is available a negation
connective and a quantier of arity (1; 1), which shall correspond to the
universal quantier of classical logic.
The representation of the truth degree behavior of the quantiers of S
needs some care and can not be restricted to the consideration of quantied
formulas, instead it has to be part of a more general approach which allows to
121
treat also quantied formulas as subformulas of \larger" ws. It should not

come as a surprise that the representation of the truth degree behavior of the
quantiers is more complicated than the representation of the truth degree
behavior of the connectives has been. For the latter purpose the expressive
power of the language of classical rst-order logic was sucient. We assume
here that this expressive power of the language of classical rst-order logic
also suces to represent the truth degree behavior of the quantiers of S.2
As done earlier, the truth degrees of W S = Wm shall be 1 ; : : : ; m in
decreasing order, i.e. we take i = mm;;1i as in (5.7).3 The goal is to nd
for each w H and each truth degree i a condition which characterizes
the situations under which H assumes truth degree i . Such a condition is
supposed to refer only to the truth degrees of the atomic subformulas of H
and the truth degree constants which occur in H , and it is referred to as i-th
value condition Bi (H ) of H . In correspondence to the inductive denition of
the notion of a w also these value conditions for ws shall be introduced
inductively over the complexity of H .
We simplify matters even more and assume that the language LS does
not have truth degree constants. Connected with each predicate symbol P
of LS there shall be m (pairwise dierent) predicate symbols P1 ; : : : ; Pm of
(the language of) classical rst-order logic PL2 , which have the same arity as
P and are chosen in suchj a way that for dierent predicate symbols P; Q of
LS all the symbols Pi ; Q , for 1 i; j m, shall be pairwise dierent. For
simplicity we suppose furthermore that LS and the language of classical rstorder logic shall have exactly the same individual variables and individual
constants. This is not an essential restriction and has the great advantage
that ongoing changes between ws of S and of PL2 shall not have to change
the individual symbols of the ws. Even more: also the metavariables used
to indicate individual symbols shall be the same for both languages. With all
this in mind it is easy to x the intended meaning of the predicate symbols
P1 ; : : : ; Pm :
Pi (a1 ; : : : ; an ) says that P (a1 ; : : : ; an ) has truth degree i .
Of course, this notation is understood always w.r.t. some xed interpretation
and some xed assignment. Therefore we dene now for each atomic formula
P (a1 ; : : : ; an ) of LS and the corresponding predicate symbols P1 ; : : : ; Pm of
the language of PL2 :
Bi (P (a1 ; : : : ; an )) =def Pi (a1 ; : : : ; an ); i = 1; : : : ; m:
(6.58)
2 This is a restriction having in mind the quite general notion of quantier which
is used for the rst-order many-valued systems in this book. But it corresponds
in some suitable sense to the restrictions in expressive power one has also in
classical rst-order logic if compared with classical second-order or even higherorder logic.
3 Because the number m of truth degrees of S is supposed to be xed it is not
necessary to indicate this number here; the choice of m shall always become
clear from the context.
122
6. Axiomatizability
The value conditions for compound formulas H of the form '(H1 ; : : : ; Hn )

with some connective ' refer to the fact that the truth degree function verS'
can be determined via its truth degree table. This means that there exist for
each truth degree i , which belongs to the range of verS' , nitely many n-tuples
(i j ; : : : ; inj ), 1 j l, of truth degrees such that verS' (s1 ; : : : ; sn ) = i i
(s1 ; : : : ; sn ) is one of these n-tuples (i j ; : : : ; inj ) with 1 j l. Then let
;

Bi (H ) =def Bi (H1 ) ^ Bi (H2 ) ^ ^ Bin (Hn ) _
;

_ Bi l (H1 ) ^ ^^Binl (Hn ) :
(6.59)
What we avoid here is to make visible in the notation the free variables of the
subformulas H1 ; : : : ; Hn .4 And in the case that i is a truth degree which does
not belong to the range of verS' , then let Bi (H ) be some PL2 -contradiction,
e.g.
Bi (H ) =def B1 (H1 ) ^ :B1 (H1 ) :
(6.60)
As an example consider the strong conjunction & of the Lukasiewicz
systems Lm as dened in (9.7) and characterized by the truth degree function
verL& = et2 . Suppose m = 5 such that the truth degree table is given in
Fig. 5.1. Then one may for i = 3, e.g., discuss the truth degree 3 = 21
immediately yielding
;

;

B3 (H1 & H2) = B3 (H1 ) ^ B1 (H2 ) _ B2 (H1 ) ^ B2 (H2 )
;

_ B1 (H1 ) ^ B3 (H2 ) :
What remains to be discussed in general is the case that H is a quantied
formula H (Qx1 ; : : : ; xk )(H1 ; : : : ; Hn ) for some quantier Q of S. The above
mentioned ;additional assumption for the
quantiers of S now reads: a PL2 formula Bi (Qx1 ; : : : ; xk )(H1 ; : : : ; Hn ) exists for each i = 1; : : : ; m which
is built up from subformulas
Bl (Hj ) for 1 l m
;
and 1 j n and
has the property that Bi (Qx1 ; : : : ; xk )(H1 ; : : : ; Hn ) holds true w.r.t. some
interpretation A and some assignment : V ! jAj i
VerSQ ((AQx ;:::;xk )(H ;:::;Hn ) ) = i
(6.61)
holds true in this case. If one is able to give within the PL2 -language a characS
terization of the generalized truth degree function
; VerQ , then this character
ization yields immediately value conditions Bi (Qx1 ; : : : ; xk )(H1 ; : : : ; Hn ) ,
and these conditions result directly from the PL2 -characterisation as the corresponding value conditions for connectives result from the truth tables for
their truth degree functions.
In the case that in the w H (Qx1 ; : : : ; xk )(H1 ; : : : ; Hn ) of LS there
occur only the predicate symbols P1 ; : : : ; Pr , which shall have as their corresponding PL2 -symbols the predicate symbols ji for j = 1; : : : ; r and i =
1; : : : ; m, and that furthermore CH ; DH are the PL2 -formulas
4 For completeness it shall be mentioned that by this denition Bi (H ) is xed only
1
11
21
up to the order of the terms in the disjunctions and the conjunctions. But this
does not matter for the present purposes.

^
^r
^
^m
^
^
: ( i ^ h ) ;
j =1
x1 ;:::;xk j j
^
^r ^
^
_
_m
DH =def j=1 x1 ;:::;xk i=1 ji ;
CH =def
i;h=1
i6=h
123
(6.62)
(6.63)
V
V W
W
in which ; are the nite iterations of conjunction ^ and of disjunction _

of PL2 , in this case it is an immediate corollary of the previous assumption,
concerning the quantiers of S and their representability, that
_
_m
CH ^ DH ) i=1 Bi (H ) ;
(6.64)
CH ^ DH )
^
^m
i;h=1 :(Bi (H ) ^ Bh (H ))
i6=h
(6.65)
hold true classically. At the same time the validity of (6.64) and (6.65) is
a necessary criterion that the formulas B1 (H ); : : : ; Bm (H ) can be the value
conditions for H : because (6.62) says that for i 6= h the PL2 -formulas ji ; jh
never can be valid together, and (6.63) says that always one of the PL2 formulas j1 ; : : : ; jm is valid. And both these conditions have to be satised
for the { elementary { value conditions ji for the atomic ws of S.
For a particular quantier of S, however, condition (6.61) is a necessary
further condition because conditions (6.64) and (6.65) have to hold true for
each quantier of S. Conversely, having given a system of value conditions
B1 (H 0 ); : : : ; Bm (H 0 ) for some quantied w H 0 (Q0 x1 ; : : : ; xk )(P1 ; : : : ; Pn )
with predicate symbols P1 ; : : : ; Pn of S in the scope of the quantier Q0 , then
this system of value conditions can be used to dene a generalized truth
degree function VerSQ0 of this quantier Q0 just in the case that it satises (in
the sense of classical logic) the corresponding versions of (6.64) and (6.65) for
arbitrary choices of the predicate symbols P1 ; : : : ; Pn . In this case one then
has
0
2
VerSQ0 (HA 0 ) = i
i ValPL
(6.66)
A (Bi (H ); ) = >
for the w H 0 (Q0 x1 ; : : : ; xk )(P1 ; : : : ; Pn ).
As an example consider again some system S with W S = W5 which shall
have quantiers Q1 ; Q2 of arity (1; 1) such that one has as in (2.15) and (2.16)
VerSQ (HA 0 ) = om(ValSA (H; [x= : : :]))
(6.67)
S
S
A
VerQ (H 00 ) = ex(ValA (H; [x= : : :]))
(6.68)
for all quantied ws H 0 Q1 xH and H 00 Q2 xH . Here ValSA (H; [x= : : :])
shall be that one function g dened on jAj which for each b 2 jAj has the
value g(b) = ValSA (H; [x=b]). Then one has e.g. for the quantier Q1 :
B1 (Q1 xH ) = 8 xB1 (H ) ;
B2 (Q1 xH ) = 8 x(B1 (H ) _ B2 (H )) ^ 9 xB2 (H ) ;
and for the quantier Q2 e.g.:
B3 (Q1 xH ) = 8 x(B3 (H ) _ B4 (H ) _ B5 (H )) ^ 9 xB3 (H ) :
1
124
6. Axiomatizability
The reader should add the other value conditions for both quantiers for
himself and test whether (6.64) and (6.65) are satised under the assumption
that these conditions (6.64) and (6.65) are satised for the value conditions
Bi (H ). Furthermore he/she should conrm that from the value conditions for
Q1 ; Q2 the denitions (2.15) and (2.16) of VerSQ and VerSQ can be recovered
according to (6.66).
The next goal now is to dene for any w H the PL2 -formulas CH ; DH ,
which have been introduced in (6.62) and (6.63) for quantied formulas
H (Qx1 ; : : : ; xk )(H1 ; : : : ; Hn ) only. This is not dicult: the choice of the
basic predicate symbols ji , with i = 1; : : : ; m; j = 1; : : : ; r, remains as before, and the PL2 -universal quantication 8 x1 ; : : : ; xk always has to be taken
w.r.t. all the individual variables x1 ; : : : ; xk which are involved into some Squantication within H , taking k = 0 if there are no quantications in H .
Lemma 6.3.1. If one extends the denitions (6.62) and (6.63) of CH and
DH to all ws H of S as just mentioned, then (6.64) and (6.65) hold true
for all ws of S.
Proof: The proof, again, proceeds inductively on the complexity of H .
In the case of an atomic formula H P (a1 ; : : : ; an ), with the PL2 -predicate
symbols Pi representing the predicate symbol P of S, one has
^
^m
_
_m
CH i;h :(ji ^ jh ) and DH i=1 ji ;
1
=1
i6=h
and according to (6.58) also

Bi (H ) Pi :
(For simplicity of notation here we avoid to mention the individual symbols
a1 ; : : : ; an which occur in all the argument places of all the n-ary predicate
symbols.) In this case (6.64) and (6.65) obviously hold true.
In the case H '(H1 ; : : : ; Hn ) with ' 2 J S one immediately gets
^
^n
^
^n
CH , j=1 CHj ;
DH , j=1 DHj
(6.69)
from the assumption that (6.64), (6.65) hold for all subformulas H1 ; : : : ; Hn .
Hence one also has
^
^n _
_m
CH ^ DH , j=1 i=1 Bi (Hj ) :
(6.70)
Having now in mind (i) the structure of the formulas Bi (H ) as given in

(6.59), (ii) the fact that ^ ; _ and thus also 8 ; 9 satisfy suitable distributive
laws, and (iii) that one can extend, semantically equivalent, conjunctions and
disjunctions of PL2 by adding some of their terms repeatedly, then one gets
that
^
^n _
_m
_
_m
B
(
H
)
)
B (H )
(6.71)
i
j
j =1 i=1
i=1 i j
125
is PL2 -valid. Both results (6.70), (6.71) yield (6.64) for the present case. And
from (6.69) together with the assumption of the inductive proof one gets that
^
^n ^
^m
:(Bi (Hj ) ^ Bh (Hj ))
CH ^ DH ,
(6.72)
j =1 i;h
=1
i6=h
is PL2 -valid. From this, the implication (6.65) follows by reductio ad absurdum. To see this, assume that CH ^ DH is valid. If in this case the succedent
of (6.65) is not valid, then there would exist indices i 6= h with 1 i; h m
such that
Bi (H ) ^ Bh (H )
would hold true. But then there would exist, according to (6.59), PL2 -valid
terms
Bi j (H1 ) ^ ^ Binj (Hn ) and Bh l (H1 ) ^ ^ Bhnl (Hn )
of the disjunctions Bi (H ) and Bh (H ). But because of CH ^ DH and (6.72)
this would yield i1j = h1l ; : : : ; inj = hnl . Therefore the truth degree function
verS' had to have for the argument (i j ; : : : ; inj ) the values i and h with,
however, i 6= h. A contradiction. Thus also (6.65) holds true in this case.
And in the case that H is a quantied formula of LS , both formulas (6.64)
and (6.65) have to be valid because it was just this property which we assumed
from the very beginning by supposing that the truth degree behavior of the
quantiers of S should be representable by value conditions.
2
The value conditions thus are for those nitely many-valued systems, for
whose quantiers they exist, the conditions which correspond to the propositional formulas (of the languages of many-valued systems) which represent
the truth degree functions of the connectives. Therefore one has the following
result.
Proposition 6.3.1. Let the m-valued rst-order system S be without truth
degree constants, such that the truth degree behavior of all the S-quantiers
is describable by value conditions. Let also DS = f1 ; : : : ; k g. Then a w H
of LS is logically S-valid if the formula
1
CH ^ DH )
_
_k
B (H )
i=1 i
(6.73)
is PL2 -valid.
The proof is obvious because of the meaning of the value conditions
Bi (H ). Nevertheless the result is of interest because the following axiomatization of S shall refer back to the value conditions for the S-quantiers.
As a last preliminary notion for the axiomatization a generalization of
the standard conditions to the rst-order case is needed, cf. [465]. For this
it is possible to restrict the discussion to quantiers of S of arity (1; 1), i.e.
which bind one variable and which have one w within its scope. Such an
126
6. Axiomatizability
S-quantier Q satises the standard condition of a universal quantier for S

i for its value conditions the formula
_
_k
_
_k
8 x i=1 Bi (H ) , i=1 Bi (QxH )
(6.74)
is PL2 -valid, assuming again DS = f1 ; : : : ; k g.
And nally we need for each w H of LS a \retranslation" of its value
conditions Bi (H ) into LS . Here we can use the fact that we supposed in
general that the many-valued rst-order systems S under consideration
shall
V
have in their alphabet connectives !; ; Jt and a quantier whose truth
degree behavior is describable by value conditions. For each value condition
Bi (H ) connected with some w H 2 LS one has a semantically equivalent
version Bi0 (H ) using only negation, implication, and universal quantication
of PL2 . Then let
Bi (H ) = the w of LS which one gets from Bi0 (H ) by exchanging
V
the PL2 -symbols ); :; 8 by their S-counterparts !; ; ,
and writing Ji (P (a1 ; : : : ; an )) instead of Pi (a1 ; : : : ; an )
for each n-ary predicate symbol P of LS .
As in the propositional case, a logical calculus K m
RT for the m-valued system
S shall have its system of axioms be determined via axiom schemata. And
these schemata shall extend the schemata for the propositional part of S.
Hence the rst-order axiom system AxRT shall consist of the propositional
schemata AxRT 1; : : : ; AxRT 8 together with the additional schemata:
V
AxRT 9 :
xA ! B
with B resulting from A by substituting for x (at all places
of its free occurrence) either some individual constant or some
individual variable y such that no place of free occurrence of x
in A belongs to the scope of a quantication of y;
V
V
AxRT 10 : x(A ! B ) ! (A ! xB ) with x not free in A;
AxRT 11 : Bi ((Qx1 ; : : : ; xk )(A1 ; : : : ; An )) ! Ji ((Qx1 ; : : : ; xk )(A1 ; : : : ; An ))
for each quantier Q 2 QS of arity (k; n) and all i = 1; : : : ; m.
Obviously, here schema AxRT 11 plays for the quantiers of S the same role
as schema AxRT 8 does for the connectives of S.
The inference rules of the logical calculus K m
RT shall be the rule of detachment (MP) and the rule of universal quantication
(Gen)
A :
xA
Theorem 6.3.1 (Soundness Theorem). Suppose that the nitely manyvalued rst-order system
S has connectives !; , and Jt for each t 2 W S , as
V
well as a quantier of arity (1; 1) such that the following hold true:
1. each w of S which falls under one of the schemata AxRT 1; : : : ; AxRT 11
is logically S-valid;
127
2. a w H1 ! H2 of S has a non-designated truth degree in the case that

H
has a designated and H2 a non-designated truth degree;
V1
3. xH is logically S-valid if H is logically S-valid.
Then each theorem of the logical calculus K m
RT , i.e. each w of S which is
derivable from AxRT 1; : : : ; AxRT 11 via the inference rules (MP) and (Gen), is
logically S-valid.
Proof: The assumptions made for the present soundness claim just say
that all the axioms of K m
RT are logically S-valid and that both inference rules
(MP) and (Gen) preserve logical S-validity.
2
Corollary 6.3.1. Suppose that the nitely many-valued rst-order system S

has an implication connective ! and a negation connective , which both
satisfy the corresponding standard conditions, has connectives Jt for each
t 2 W S with
V truth degree functions given by (5.111), and has a universal
quantier of arity (1; 1) which satises the standard condition of a universal
quantier. Then within the logical calculus K m
RT for S only logically S-valid
ws are derivable.
Proof: The fact that !; V satisfy the corresponding standard conditions
immediately implies that the conditions (2) and (3) of Theorem 6.3.1 are satised. So it remains to prove that the axiom schemata AxRT 1; : : : ; AxRT 11 determine only logically S-valid ws. For the propositional schemata AxRT 1; : : : ;
AxRT 8 this can easily be shown as in the proof of Theorem 6.2.1. And for the
typically rst-order schemata AxRT 9; : : : ; AxRT 11 the same fact results from
quite straightforward calculations, which are left to the reader.
2
Theorem 6.3.2 (Completeness Theorem). Let the nitely many-valued

rst-order system S have an implication
connective !, a negation connective
, and a universal quantier V of arity (1; 1) which all satisfy the corresponding standard conditions, and let S have connectives Jt for each t 2 W S
with truth degree functions given by (5.111). Then within the logical calculus
K mRT for S all logically S-valid ws are derivable.
The proof of this completeness theorem can, e.g., be given by a suitable
adaptation of Henkin's completeness proof for classical PL2 . This needs the
derivation of a series of particular ws like in the proof of the corresponding
propositional Completeness Theorem 6.2.2. The proof procedure is not very
inspiring in its details and shall therefore not be given here. The interested
reader should consult [465] for a proof.
The whole problem of axiomatizability can, however, also be approached
from another point of view: one can (only) ask for some given system S of
rst-order many-valued logic whether there exist logical calculi which derive exactly the logically S-valid ws, or which derive exactly all the (valid)
128
6. Axiomatizability
entailment relations for S-sentences. This means to ask whether the set of logically S-valid ws or the set of all (valid) entailment relations for S-sentences
are recursively enumerable sets of strings. In this general form Mostowski
[384, 385] discussed the problem and got positive as well as negative results,
depending on the particular choices of the truth degree sets W S , DS , as well
as of the sets J S , QS of connectives and of quantiers of S. This more general
approach has, however, in the case of a positive answer, i.e. in the case that
some of these sets are proved to be recursively enumerable, the disadvantage
that the proof does not yield a method to generate these sets, i.e. does not
provide a corresponding logical calculus. Therefore we also shall not treat
this matter in more detail.
6.4 Axiomatizing the Entailment Relation

The knowledge of the entailment relation for a system S of many-valued logic
is at least as basic as the knowledge of its logically valid ws. Therefore,
extending the axiomatic approaches of the previous sections toward a complete formalization, i.e. adequate axiomatizations of the notion of S-logical
validity, we now are interested in a complete axiomatization of the notion of
entailment for S: given the system S we ask for a logical calculus K with a
corresponding notion `K of K -derivability such that one has for any (suitable)
set of ws of LS that a w H is entailed by the set of premises i H is
K -derivable from this set .
For any logical calculus K which completely formalizes the notion of entailment for a system S, i.e. which satises
`K H i j=S H
(6.75)
for any w H and each set of ws, one says that K adequately axiomatizes
the entailment relation j=S .
Obviously there is a close connection between both kinds of axiomatization: that one of the entailment relation for S and that one of the logical
validity for S. For, having a logical calculus K which satises condition (6.75),
then according to (3.9), (3.10) one has immediately
; `K H i H 2 tautS ;
(6.76)
which means, that in this case the logical calculus K also adequately axiomatizes the class of S-tautologies.
This is the background reason that one adapts the notation we introduced
in (3.9) also for K -derivability and usually writes
`K H instead of ; `K H :
(6.77)
However, some care is necessary with this type of notation. Having a logical
calculus K to formalize the notion of entailment, then \ `K H " as well as
\`K H " are suitably dened notions. Having, however, a logical calculus K 0
129
to formalize the notion of logical validity, one usually has only \`K H " as a
suitable notation and has \ `K H " undened.
Fortunately, for a wide class of logical calculi there is a simple standard
way to extend logical calculi which are intended to derive ws to logical calculi
which allow to derive ws from given sets of premises. Having given a logical
calculus K , which satises the conditions (K1), (K2), (K3) of Section 6.2,
p. 107, and which has its notion of derivability characterized by the further
conditions (D1), (D2) of this section, cf. p. 110, then the extension consists
in the agreement to consider as a K -derivation from a given set of premises
any nite sequence H1 ; H2 ; : : : ; Hn which satises the conditions
(D1) Hn is the w H ;
(D20 ) each w Hk of this sequence is either an axiom of K 0 or the result of
an application of one of the inference rules of K 0 to preceding ws of this
sequence.
This extended notion of K -derivability will be called standard extension
of `K and shall further on be denoted by `?K . This standard extension allows
not only to derive ws \relative" to some given set of premises, it allows
also to \shorten" derivations by using new inference rules, which have to be
\suitable" in some appropriate sense, of course. For reaching this goal, we
call an inference rule
H1 ; H2 ; : : : ; Hn
(6.78)
G
0
admissible, or sound, for the logical calculus K i fH1 ; : : : ; Hn g `?K G holds

true. Intuitively, this admissible inference rule (6.78) symbolizes the whole
derivation which leads from fH1 ; : : : ; Hn g to G.
One gets as an immediate corollary that derivations \within K ", which
also use (for K ) admissible inference rules which do not belong to the (\primary") inference rules of K , can be transformed into K -derivations: one simply has to replace each application of an admissible inference rule (6.78) by
a whole K -derivation it symbolizes.
It is not straightforward at all, however, that the extended notion of
derivability `?K also satises condition (6.75), even if one assumes that K
adequately axiomatizes logical validity.
Fortunately, some few additional assumptions suce to prove that the
standard extension of a logical calculus K adequately axiomatizes the notion of entailment, provided K adequately axiomatizes logical validity. The
additional properties to be considered are:
(FINj= ) For each w H and all sets of ws it holds true
j= H i 9 0 ( 0 nite ^ 0 j= H ) :
(FIN` ) For each w H and all sets of ws it holds true
` H i 9 0 ( 0 nite ^ 0 ` H ) :
130
6. Axiomatizability
(DEDj=) For all ws H1 ; H2 and all sets of ws it holds true

[ fH1 g j= H2 i j= (H1 ! H2 ) :
(DED` ) For all ws H1 ; H2 and all sets of ws it holds true
[ fH1 g ` H2 i ` (H1 ! H2 ) :
Property (FINj= ) is an important property of the logical system under
consideration and of its entailment relation. In the case that (FINj= ) holds
true for a logical system S one says that for S the niteness theorem for
entailment holds true.
Property (FIN` ) is a property of the logical calculus under consideration.
It usually holds true in all the cases where all the inference rules of this logical
calculus have only nitely many premises, i.e. are nitary.
Property (DEDj= ) is essentially a condition which is related to the set J S
of connectives of the logical system S under consideration and to an implication connective (resp. its truth degree function) this set should contain.
Property (DED` ) is again a property of the logical calculus under consideration and amounts to formal properties regarding the derivability of
implications within this calculus. In the case that (DED` ) holds true for a
logical calculus K one says that for K (or for its notion `K of derivability)
the deduction theorem holds true.
Therefore properties (FIN` ) and (DEDj= ) are not too strong and important ones, properties (FINj=) and (DED` ) on the other hand are strong and
important properties. This is the reason why they have their own names.
All together, these four properties allow to characterize such cases in
which the standard extension of a derivability relation provides an adequate
axiomatization of the entailment relation.
Theorem 6.4.1. Suppose that S is a propositional system of many-valued
logic and that K is a logical calculus which adequately axiomatizes the set
of S-tautologies. Suppose furthermore that the entailment relation j=S of S
satises properties (FINj=) and (DEDj=), and that the standard extension
`?K of `K satises properties (FIN` ) and (DEDj=). Then this standard extension satises also condition (6.75), i.e. adequately axiomatizes the entailment
relation of S.
Proof: We have to show that for each w H and each set of ws it
holds true
j=S H i `?K H :
(6.79)
But because of the properties (FINj=) and (FIN` ) it suces to prove (6.79)
for nite sets only. For, assume that (6.79) holds true for all nite sets
of ws and let be any set of ws. From j=S H via (FINj= ) then follows
0 j=S H for some nite set 0 of ws, hence one has 0 `?K H by (6.79), and
therefore also `?K H . Conversely, from `?K H it follow via (FIN` ) that
131
0 `?K H holds true for some nite set 0 of ws, hence one has 0 j=S H by
(6.79), and thus also j=S H .
Therefore suppose that is nite, say = fH1 : : : ; Hn g. Then because
of (DEDj=) one has
j=S H i fH2 ; : : : ; Hn g j=S (H1 ! H ) ;
and can repeat this application of (DEDj= ) to get
j=S H i fH3 ; : : : ; Hn g j=S (H2 ! (H1 ! H )) :
Iteration of this application of (DEDj=) nally leads to
i n
j=S H i j=S ! (Hi ; H )

=1
using the notation dened in (6.1), (6.2). Because K is supposed to adequately

axiomatize tautS and because of (3.10) one has
i n
j=S H i `K ! (Hi ; H ) :
=1
Now one has to apply (DED` ), instead of (DEDj= ), yielding
n;1
j=S H i fHn g `?K ! (Hi ; H ) :

=1
Again iterating this argument n ; 1 times nally yields

j=S H i `?K H ;
and the theorem is proved.
2
Because of this theorem, for any logical system S with an adequately axiomatizable set tautS of logically valid ws, and with an entailment relation
for which the niteness theorem holds true, one usually asks whether the
deduction theorem5 holds true for suitable logical calculi adequately axiomatizing tautS . For the nitely many-valued propositional systems adequately
axiomatized in Section 6.2 it was proved in Theorem 3.2.5 that the niteness theorem for entailment holds true for them. And because of the notion
of K RT -derivability, all the logical calculi K m
RT satisfy the property (FIN` ).
Therefore it remains to discuss the properties (DEDj= ) and (DED` ).
Let us start with (DEDj=). Then one has to show that it holds true
[ fH1g j=S H2 i j=S (H1 ! H2 ) :
Assume rst j=S (H1 ! H2 ), and let 2 ModS ( [fH1 g). Then obviously
2 ModS ( ), and hence 2 ModS (H1 ! H2 ). Therefore ValS (H1 ; ) and
ValS (H1 ! H2 ; ) are designated truth degrees.
5 The interest in the deduction theorem may, on the other hand, be also related to
the fact that there are methods to adequately axiomatize tautS which presuppose
that the deduction theorem holds true.
132
6. Axiomatizability
Suppose now additionally that the implication connective ! satises condition (I*) of Section 6.2, then also ValS (H2 ; ) has to be a designated truth
degree, which means 2 ModS (H2 ) and hence also [ fH1 g j=S H2 .
Now assume, conversely, that [ fH1g j=S H2 holds true and consider
any 2 ModS ( ). Then either one has 2 ModS (H1 ) and hence also
2 ModS (H2 ), or one has 2= ModS (H1 ). Suppose additionally that the
implication connective ! satises the standard condition (I), then in both
these cases one has 2 ModS (H1 ! H2 ). Therefore then j=S (H1 ! H2 ).
At all we have, among others, proved the following result.
Proposition 6.4.1. If the implication connective which appears in (DEDj=)
satises the standard condition (I) of an implication connective then property
(DEDj=) holds true.
So nally property (DED` ) needs to be considered. As a preliminary
result one has the following
Proposition 6.4.2. In each one of the Rosser-Turquette calculi K mRT it
holds true that
(DED` ) i `RT ((A ! B ) ! ((A ! (B ! C )) ! (A ! C ))) :
Proof: First assume that (DED` ) holds true. Then by three applications
of the rule of detachment (MP) one nds
fA; A ! B; A ! (B ! C )g `?RT C
for any ws A; B; C . Because of (DED` ) one nds successively
fA ! B; A ! (B ! C )g `?RT (A ! C ) ;
fA ! B g `?RT ((A ! (B ! C )) ! (A ! C )) ;
and nally
`RT ((A ! B ) ! ((A ! (B ! C )) ! (A ! C ))) :
(6.80)
Now assume that (6.80) holds true. Then, because of the availability of
the rule of detachment (MP), obviously [ fH1 g `?RT H2 is a consequence
of `?RT (H1 ! H2 ). Thus for (DED` ) it remains to show that the derivability `?RT (H1 ! H2 ) is a consequence of [ fH1 g `?RT H2 . And this
shall be done in such a way that we assign to each (relative) K m
RT -derivation
G1 ; G2 ; : : : ; Gk of some w Gk from some set of ws a K mRT -derivation of
G ! Gk from n fGg, G any w. This new derivation is obtained from the
given one by replacing some of the w Gi with (short) sequences according
to the following rules:
1. In the case that Gi is an axiom or a premiss from and that Gi 6 G holds
true, one replaces Gi with the sequence: Gi ; Gi ! (G ! Gi ); G ! Gi .
Here the w Gi ! (G ! Gi ) is an instance of (AxRT 1), and G ! Gi
results from the two previous ws by an application of (MP).
133
2. In the case that Gi is an axiom or Gi 2 and that Gi G holds, one

has to replace Gi with a K m
RT -derivation of the w G ! G, which exists
according to (6.13), and which does not need G as a premiss.
3. In the case that Gi results from an application of (MP), there are indices
j; l < i such that Gj Gl ! Gi . Then Gi has to be replaced by the
sequence
H1 ; : : : ; Hn ; (G ! (Gl ! Gi )) ! (G ! Gi ); G ! Gi
with H1 ; : : : ; Hn a K m
RT -derivation of (G ! (Gl ! Gi )) ! (G ! Gi ).
Such a derivation exists according to (6.80) and does not need any extra
premises.
By induction on i it is then easily proved that under this replacement each
subsequence G1 ; : : : ; Gi of G1 ; : : : ; Gk is transformed into a K m
RT -derivation
of G ! Gi which uses only premises from n fGg. And for i = k we get the
result we were looking for: `?RT (H1 ! H2 ) follows from [fH1g `?RT H2 .
Hence we have reduced the problem of the validity of (DED` ) to the

problem of the validity of (6.80). But because of the Completeness Theorem
6.2.2 the validity of (6.80) is established if one is able to show
((A ! B ) ! ((A ! (B ! C )) ! (A ! C ))) 2 tautS
(6.81)
for the system S of propositional many-valued logic which is axiomatized by
the logical calculus K m
RT .
But assuming that ! satises the standard condition (I) makes it a routine matter to establish (6.81) quite in the same way as one would prove in
classical logic that the w mentioned in (6.81) is a tautology.
Thus we have also the following result.
Proposition 6.4.3. If the implication connective of a system S, which appears in (DED` ), satises the standard condition (I) of an implication connective then the deduction theorem (DED` ) holds true for the logical calculus
K mRT which adequately axiomatizes S.
Both these propositions together establish the following main theorem.
Theorem 6.4.2 (Strong Completeness Theorem). Let S be a propositional system of many-valued logic with m truth degrees which satises the
conditions (RT1), (RT2). Then the logical calculus K m
RT with the standard extension `?RT as its derivability relation provides an adequate axiomatization
of the entailment relation j=S , i.e. then there always holds true
j=S H i `?RT H :
A closer inspection of the proof of Proposition 6.4.2 shows that one really
only needs for the logical calculus under consideration that (MP) and (AxRT 1)
are available { either as a (primitive) rule of inference and an axiom schema,
134
6. Axiomatizability
or as an admissible rule of inference and a derivable schema. Even ` (A ! A)

results then already from the w considered above in (6.80).
In [465] it is additionally proved that one can derive from the w considered in (6.80) via (AxRT 1) and (MP) all the schemata (AxRT 2), . . . , (AxRT 4).
However, substituting the w of (6.80) as a new axiom schema for these
m
schemata in the axioms of K m
RT would give a stronger calculus than K RT , i.e.
a logical calculus which derives more ws.
We shall see later on in Part III that there exist also m-valued propositional systems S which are adequately axiomatized by K m
RT , which however
have an implication connective for which the deduction theorem does not
hold true, and which do not satisfy condition (RT1). The assumptions of
Theorem 6.4.2 hence are sucient ones, but not necessary ones.
One may therefore try to weaken the assumptions made for Theorem 6.4.2.
Furthermore also condition (RT2), demanding the S-denability of all the Jconnectives Jt , t 2 W S , is rather restrictive regarding the logical systems to
which Theorem 6.4.2 can be applied.
Instead of trying to adapt this Hilbert type axiomatization of RosserTurquette better to the systems of many-valued logic, we shall in Chapter
7 discuss other methods of axiomatization, which are many-valued adaptions
of sequent calculi and of tableaux calculi well known from the proof theory
of classical logic.
Compactness for a nitely-valued rst-order system S in the strengthened
form of Corollary 3.4.2 allows to prove the niteness property (FINj= ) for the
entailment relation j=S in just the same way as it was done for the propositional case in the proof of Theorem 3.2.5. Writing `RT for the derivability
relation of the logical calculus K m
RT , i.e. for the derivability relation constituted by the axiom schemata AxRT 1; : : : ; AxRT 11 and the inference rules (MP)
and (Gen), then one immediately gets that `?RT has the niteness property
(FIN` ). For a proof of the deduction theorem (DED` ) for the (extended)
derivability relation `?RT one can proceed as in the proof of Theorem 6.4.2,
but has to add the consideration for the case that the derivation, which has
to be transformed into another one in the actual step which led to the w
V Gi ,
consisted in an application of the rule (Gen). In this case one has Gi xGk
for some k < i and has to replace Gi by the sequence
^
^
^
x(G ! Gk ); x(G ! Gk ) ! (G ! xGk ); G ! Gi
assuming additionally that the variable x does not occur free in the w G.
Therefore we shall restrict our discussion on and our use of (DED` ) to those
cases where one is considering only S-sentences. With this restriction, (DED` )
obviously holds true. Hence we also discuss the semantic deduction property
(DEDj=) only w.r.t. S-sentences. And then, as in Proposition 6.4.1: if the
implication connective which appears in (DEDj=) satises the standard condition (I) of an implication connective, then property (DEDj=) holds true.
In summary, we thus have proved:
135
Theorem 6.4.3 (Strong Completeness Theorem). Let a nitely manyvalued rst-order system S have connectives Jt for each t 2 W S with truth
degree functions given by (5.111), and have an implication
connective !, a
V
negation connective , and a universal quantier of arity (1; 1) which all
satisfy the corresponding standard conditions. Then for all S-sentences H and
all sets of S-sentences it holds true
j=S H i `?RT H ;
with `?RT the standard extension of the derivability relation of the logical
calculus K m
RT constituted by the axiom schemata (AxRT 1); : : : ; (AxRT 11) and
the inference rules (MP) and (Gen).
136
6. Axiomatizability
7. Sequent and Tableau Calculi
For classical logic, logical calculi have been introduced by G. Gentzen in

[184] which are completely dierent from the Hilbert-type axiomatization,
which was the background for the Rosser-Turquette axiomatization of
systems of many-valued logic given in Chapter 6.4. These Gentzen-type calculi work with sequents as the objects which are derived. And as an outcome
of investigations into these calculi a further kind of calculi was developed for
classical logic, essentially by E.W. Beth [54] and J. Hintikka [263]: tableau
calculi.1
For classical logic, sequents have the form ; for (usually nite) lists
;; of ws, connected by the symbol which symbolizes some otherwise
unanalyzed \entailment\ relation. The corresponding semantics counts such
a sequent fA1 ; : : : ; Ak g fB1 ; : : : ; Bn g as valid i the w A1 ^ ^ Ak )
B1 _ _ Bn is valid in the standard sense. This means that such a sequent
; is valid i in the case that all the ws of ; are true under some
valuation , at least one w of has to be true (i.e. not false) under too.
And this validity behavior is equivalent with the property that such a sequent
; is not satisable i there does not exist any valuation which makes
all ws of ; true and all ws of false, i.e. i { with some abuse of language
{ there does not exist a valuation which is a \?-model" of and at the
same time a \>-model" of ; .
The derivations for this type of sequent calculus are nite sequences of
sequents built up from suitably chosen \initial" sequents according to a given
set of inference rules.
Tableau calculi, on the other hand, present their derivations essentially
in the form of (nite) \derivation" trees, i.e. in the form of directed graphs
with (usually nitely many) nodes which have one source node, and which
have such a system of edges that any pair of path-connected nodes has only
one, uniquely determined connecting path.
1 The standard reference for years, and the book that made this approach really
popular, was [511]. Since, these calculi have become quite popular, particularly
for computer science purposes, as shown e.g. in [111].
138
7.1 Tableau Calculi for Many-Valued Logic

Tableau calculi for classical logic proceed usually in such a way that their
method of \proof construction" starts from some given w H , forms in a
systematic way a series of particular tree-like graphs, the tableaux of this
construction, up to some level of completeness of this construction, and usually aim at either a refutation of the w H the particular tableau construction
starts from, or aim at the construction of a valuation which gives H the truth
value >. The nodes of these tree-like graphs become decorated with ws, or
sets of ws, and the edges indicate the succession of the construction steps toward the completed tableau. The most elementary ones of these tableaux are
given by a graph with only one node (and no edges at all) which is decorated
with the \initial" w H .
The kind of realization of this tableau construction which seems to be
most suitable for a generalization to the many-valued setting \decorates" the
nodes not only with (well-formed) formulas, or sets of such ws, but uses
signed formulas, or sets of such signed ws as decorations. By such a signed
formula one understands an ordered pair consisting of a w and some truth
value as \sign". To illustrate this matter with an example, a tableau for the
w
(H1 ^ H2 ) H1 ) ^ (H1 ) H1 _ H2 )
(7.1)
(of classical logic), combined with the sign ?, is shown in Fig. 7.1.
f?g : (H1 ^ H2 ) H1 ) ^ (H1 ) H1 _ H2 )
H
HH

HH

HH

H
f?g : (H1 ^ H2 ) H1 )
f?g : (H1 ) H1 _ H2 )
f>g : H1 ^ H2
f?g : H1
f>g : H1
f?g : H1 _ H2
f>g : H1
f>g : H2
f?g : H1
f?g : H2
Fig. 7.1.
Tableau construction for some w of classical logic
139
This tree has to be imagined as being \grown" downwards out of the

\root node" which is labeled with the given signed formula. The nodes are
represented simply by the (lists of) signed formulas they are labeled with.
Of course, in the cases where at the position of some node a nite list of
signed formulas is given, this situation has to be understood as a shorthand
for some linear subtree (as shown in general on the left-hand side of Fig. 7.3
later on) with nodes decorated with the single signed formulas of this list. The
construction of \new" edges and nodes during the \growing" of the \tableau
tree" indicates the conditions which have to be met to give some w the truth
value, which is indicated by the sign which is attached to it. The splitting
into dierent branches represents the fact that there are dierent possibilities
for giving a w some truth value.
Each branch analyses one possibility to give to the \root formula" the
truth value which is indicated in its sign.
The whole tableau construction usually is a refutation procedure: to show
that the w (7.1) is logically valid one discusses the possibilities to make it
false. The complete tableau of Fig. 7.1 then shows that this is impossible {
because each of its branches is closed, i.e. contains some w which appears
(within this branch) with two dierent signs.
Furthermore, if one would meet in this tableau construction some branch
which is not closed, and which can also not become closed in further construction steps for extended tableaux (with the same signed root formula),
then such an open branch can be used to give in a constructive way the root
formula of the tableau tree the truth value given by the sign of this w. Of
course, this supposes that one has rules which govern the construction of
tableaux.
We shall not explain these tableau calculi for classical logic here in more
detail. The interested reader may consult e.g. [37, 165, 511].
The very same construction idea can be applied for the case that the logical system S under consideration is a system of many-valued logic, and that
the signs are truth degrees out of W S , cf. [83, 85]. But it is even natural to
have as signs sets of truth degrees,2 which e.g. may come from the approach
that the proof that some given w H is S-logically valid is given by a systematic analysis (via tableau constructions) of the cases in which H has some
non-designated truth degree.
The standard understanding of the signs is that they are (names for) nonempty subsets of the power set IP (W S ). And for a signed formula s : H with a
set s of truth degrees as its sign the usual understanding is that this indicates
that the truth degree of H should belong to s. It then is natural to subsume
the case that the signs are truth degrees under this more general approach:
by taking as signs singletons of truth degrees instead of truth degrees.
The tableau constructions in many-valued logic are, as in the case of
classical logic, governed by rules which say how to extend some given tableau
2 This sets-as-signs approach essentially was rst done in a systematic way in [237].
140
at end nodes of its branches, of course depending on the (structure of the)

signed formulas present as labels in the actual branch. These rules depend
essentially on some compound w H they are applied to, and have to say
whether the tableau extension at this particular end node of a branch has to
be made by some \linear" extension of this branch, or by some splitting. The
particular form of these tableau extension rules, particularly the number of
branches into which the branch under consideration has to split, additionally
depends on the sign, this w H is attached with, as well as on the whole set
of truth degrees W S of the system S one is considering.
As long as W S = Wm is a nite set, the branching number of these tableau
trees remains nite, which means that all the tableaux remain nite trees.
To illustrate the form of the tableau extension rules consider as an example some 3-valued system S0 which has a disjunction connective _ with
truth degree function vel2 introduced in (5.76). Then some w H1 _ H2 of
LS has truth degree ValS0 (H1 _ H2 ; ) = 21 under some valuation just
in the cases that one either has ValS0 (H1 ; ) = 0 and ValS0 (H2 ; ) = 21 , or
that one has ValS0 (H1 ; ) = 12 and ValS0 (H2 ; ) = 0. This means that the
tableaux given in Fig. 7.2 would be acceptable tableaux for the signed formulas f 21 g : (H1 _ H2 ) and f0; 12 g : (H1 _ H2 ). The second one of these
tableaux essentially uses that a set of truth degrees as a sign of some w has
to be understood as saying that this w should have one of the truth degrees
of this set.
0
f 12 g : (H1 _ H2 )

f 12 g : H1
f0g : H2
Fig. 7.2.
HH
H
f0; 21 g : (H1 _ H2 )
H
f0g : H1
f 12 g : H2
H
HH

f0; 12 g : H1
f0g : H2
Two examples of tableaux for many-valued logic
f0g : H1
f0; 21 g : H2
Corresponding tableau extension rules have to x which subtree has to

be appended to the end node of the actual branch under consideration in
which the signed formula s : H appears, and x what are the signed formulas
which have to be taken as the decorations of the nodes of the \new" subtree.
Principally there are two main types of simple subtrees which can be considered as building blocks for the tableau extensions: nite chains and nite
branchings. These types of simple trees have the forms which are explained
in Fig. 7.3. Tableau extension rules which allow one to expand some (branch
of a) tableau by one of these simple trees from Fig. 7.3 shall be written down
as
s:H
s1 : H1 s2 : H2 sk : Hk

s:H
b
s1 : H1
s2 : H2
..
.
141
s:H
Q
A Q
A Q

A QQ

A
Q
s1 : H1
s2 : H2
:::
sk : Hk
sk : Hk
Simple trees for the extension of tableaux
b
Fig. 7.3.
for the linear case (on the left-hand side of Fig. 7.3), and as
s:H
s1 : H1 + s2 : H2 + + sk : Hk
for the branching case. These types of extension rules shall be indicated by
the shorthand notations
s:H
s:H
and
(7.2)
fs : H j 1 i kg
+ fs : H j 1 i kg :
i
The key idea behind the examples in Fig. 7.2 is to give a kind of complete, but
irredundant analysis of all the cases which lead to the situation that the root
formula has one of the truth degrees which are given by the corresponding
sign.
In general a tableau extension will proceed by some combination of extensions of these two simple types which is a branching into linear pieces,
that means it is of the combined form
s:H

:
+ fsij : Hij j 1 i kj g 1 j k
Therefore the tableau extension rules depend on the denition of the (truth
degree functions of the) connectives of the particular system S of (propositional) many-valued logic under consideration as this was the case for the
axiom schema (AxRT 8) of the Rosser-Turquette axiomatization explained
in Section 6.2.
A tableau extension rule for an n-ary connective ' of some system S of
many-valued logic has therefore to be a (partial) function from the class of
signed formulas of LS into the class of tableau trees. And each such rule has
the form
s : '(H1 ; : : : ; Hn )

:
(7.3)
+ f^sj : Hij j (j; ^sj ) 2 M g M 2 T ('; s)
Here one has to suppose that T ('; s) is characterized as the class of all nite
sets M of ordered pairs (j; ^sj ), consisting of an integer 1 j n and a sign
^sj , such that
142
M = f(i1 ; ^s1 ); : : : ; (ik ; ^sk )g 2 T ('; s)
holds true i the following clauses are satised:

(TR1) There exists some valuation such that ValS (Hij ; ) 2 ^sij for each
1 j k.
(TR2) One has verS' [s1 ; : : : ; sn ] s for all n-tuples (s1 ; : : : ; sn ) of signs with
sij = ^sj for all 1 j k.
(TR3) For no 1 j0 k there exists some sign s+ij with ^sj s+ij such that
(TR1) and (TR2) are satised too.
(TR4) There is no proper subset M which satises (TR1) and (TR2) too.
This rather complicated characterization codes in (TR3) the previously
mentioned kind of completeness, here taken in the sense that the signs should
be maximal sets of truth degrees satisfying (TR1) and (TR2), and codes in
(TR4) the irredundancy idea, also mentioned previously.
From this characterization of the tableau extension rules for signed compound formulas the following corollary is immediately clear.
Corollary 7.1.1. If for some signed compound formula s : '(H1 ; : : : ; Hn)
there exist signs si for 1 i n such that verS' [s1 ; : : : ; sn ] s, then there
exists a tableau extension rule (7.3) for s : '(H1 ; : : : ; Hn ) such that sij sij
for all 1 j k.
The case that T ('; s) becomes the empty set for some connective ' and
some sign s means, of course, that there is no tableau extension rule of the
form (7.3) for the signed formula s : '(H1 ; : : : ; Hn ).
There is, rst, a possible articial diculty here depending on the available signs: in \pathological" cases it could happen that such a tableau extension rule cannot be formulated, despite the fact that the signed compound
formula s : '(H1 ; : : : ; Hn ) which acts as a premiss of the rule should allow
for an analysis of the cases under which ValS ('(H1 ; : : : ; Hn ); ) 2 s is the
case { simply because there are not suciently many subsets of W S available
as signs. To avoid this diculty we always shall assume that the set of all
available signs is complete in the sense that for each sign s and each n-ary
connective ' of S there exist for each n-tuple (u1 ; : : : ; un ) of truth degrees
with verS' (u1 ; : : : ; un) 2 s signs s1 ; : : : ; sn such that verS' [s1 ; : : : ; sn ] s. Then
it follows that there exists for each signed formula s : '(H1 ; : : : ; Hn ) a tableau
extension rule with s : '(H1 ; : : : ; Hn ) as its premiss.
For complete sets of signs, fortunately, a simple characterization for the
case T ('; s) = ; is available.
Proposition 7.1.1. For a signed formula s : '(H1 ; : : : ; Hn) there does not
exist a tableau extension rule (7.3) i one has s \ rg(verS') = ;.
Proof: If one has s \ rg(verS') = ; then (TR2) cannot be satised, which
means T ('; s) = ;, i.e. that there does not exist a tableau extension rule
(7.3) for s : '(H1 ; : : : ; Hn ).
0
143
If otherwise s \ rg(verS' ) 6= ;, then there exist signs si for 1 i n such

that verS' [s1 ; : : : ; sn ] s, and hence there exists a tableau extension rule (7.3)
for s : '(H1 ; : : : ; Hn ) according to Corollary 7.1.1.
2
It is obvious that one hence has also the following corollary.
Corollary 7.1.2. If there exists a tableau extension rule (7.3) for some
signed formula s : '(H1 ; : : : ; Hn ), and if s is a sign such that s s, then
there exists such a tableau extension rule also for the signed formula s :
'(H1 ; : : : ; Hn ).
Having given some particular system S of many-valued logic together with
all of its tableau extension rules (7.3) for any compound w '(H1 ; : : : ; Hn )
and any (available) sign s W S , then the tableau construction process for
some given signed formula s : H proceeds in a very similar manner to the
case of classical logic:
One takes as a tableau for s : H the tree with only the source node which
is decorated with the signed formula s : H .
Having given some tableau for s : H which is not closed, one allows to
extend any one of its branches, which are not themselves closed, by an
application of some tableau extension rule to some signed compound formula of this branch, which is not marked as \used", and marks this signed
formula in the extended tableau as used too.
One marks a branch of this extended tableau as closed if it has Tnodes
which
are decorated with signed formulas s1 : G; : : : ; sh : G such that hj=1 sj = ;,
or if it contains only unmarked signed formulas for which there do not exist
tableau extension rules (7.3).
One marks the whole tableau as closed in the case that all its branches are
marked as closed.
One calls a branch as well as a tableau open i it is not closed, one calls a
branch complete i all its signed compound formulas for which there exists
a tableau extension rule are marked as used, and one denotes a tableau as
complete i all its branches are complete.
This successive construction of larger and larger tableaux for some given
signed formula s : H ends either if all branches of some tableau for s : H are
closed, or if some complete tableau is reached.
This tableau extension method has a nice property: it preserves satisability. To state the corresponding result in a sound way, we rst have to
extend the notions of model and of satisability to sets of signed formulas
and to tableaux.
Denition 7.1.1. A valuation is a model of a signed formula s : H of LS
i one has ValS (H; ) 2 s, and it is a model of some set M of signed formulas
i it is a model for each signed formula s : H 2 M .
A signed formula, as well as a set of signed formulas, is satisable i
there exists some model of it.
144
A branch of some tableau is called satisable i the set of all signed formulas which occur as decorations of the nodes of this branch is satisable.
And a tableau is called satisable i it has a satisable branch.
Proposition 7.1.2. Each extension of a satisable tableau for some signed
formula s : H which is constructed according to one of the tableau extension
rules (7.3) is again a satisable tableau for s : H .
Proof: It suces to remark that the characterization of the tableau
extension rules (7.3) by the conditions (TR1) to (TR4) was chosen
in such

a way that each tree + f^sj : Hij j (j; ^sj ) 2 M g M 2 T ('; s) with its root
node decorated by s : '(H1 ; : : : ; Hn ) has a satisable branch i the signed
formula s : '(H1 ; : : : ; Hn ) is satisable.
2
Again as in the case of classical logic, the existence of some closed tableau
for a signed formula s : H means that there does not exist a valuation
with ValS (H; ) 2 s, i.e. there does not exist a model of s : H . And if the
previously explained tableau construction process stops with some complete
tableau which is not closed, then one is able to construct from its non-closed
branches valuations with ValS (H; ) 2 s. To reach this goal one has to
consider in every open branch all the signed formulas which are not marked
as used, i.e. all the signed propositional variables. By construction of this
complete tableau there exists a valuation which assigns to each one of these
propositional variables a truth degree which is a member of the corresponding
sign. The formal details of these ideas are given in the rest of this section.
For the tableau construction systems, related with a given system S of
many-valued logic which we consider in the following, assume always that
they rely on a complete set of signs, and that they have all the possible tableau
extension rules available which can be formed according to this previous
denition.
Theorem 7.1.1 (Soundness Theorem). Let S be some system of manyvalued logic with nitely many truth degrees. Then the existence of a closed
tableau for the signed formula (W S n DS ) : H means that the w H of LS is
S-logically valid.
Proof: Obviously no closed branch of a tableau is satisable. Therefore
the existence of a closed tableau for some signed formula s : H says that this
signed formula is not satisable.
So assume that for the signed formula (W S nDS ) : H there exists a closed
tableau. Then this signed formula is not satisable, which means that for
each valuation one has ValS (H; ) 2= W S n DS , i.e. ValS (H; ) is always a
designated truth degree.
2
To get also a corresponding completeness theorem, some additional notions shall be introduced.3
3 We follow here essentially the approach of [237].
145
Denition 7.1.2. A set H of signed formulas of LS is a Hintikka set i

the following conditions are satised:
T
(H1) for all signed atomic formulas s1 : p; : : : ; sk : p 2 H one has ki=1 si 6= ;;
(H2) for each signed compound formula s : '(H1 ; : : : ; Hn ) 2 H there exists
a tableau extension rule
s : '(H1 ; : : : ; Hn )

:
such that for some M 2 T ('; s) one has f^sj : Hij j (j; ^sj ) 2 M g H.
And such a Hintikka set is called saturated i also the following conditions
are satised:
(H3) for each propositional variable p there is a signed formula s : p 2 H;
(H4) for each signed atomic formula s : p 2 H and each sign s s one has
also s : p 2 H;
(H5) if one of the branching sets f^sj : Hij j (j; ^sj ) 2 M g of a tableau extension
rule for some signed compound formula s : '(H1 ; : : : ; Hn ) is a subset of
H, then one has also s : '(H1 ; : : : ; Hn ) 2 H.
There is some (weak) similarity between Hintikka sets and tableaux, as
well as between complete tableaux and saturated Hintikka sets.
Lemma 7.1.1. Let H be some saturated Hintikka set. For each signed formula s : H 2 H and each sign s with s s one then has s : H 2 H.
Proof: We proceed inductively on the complexity of the w H . In the
case that H is a propositional variable this follows for all signs s; s with s s
immediately from the dening condition (H4) of the saturated Hintikka
sets.
So let H '(H1 ; : : : ; Hn ) a compound w with s : H 2 H, and assume
that the claim of this lemma holds for all its constituents H1 ; : : : ; Hn . Let
also s be a sign with s s.
Then by (H2) there exists a tableau extension rule (7.3) for s : H , and
therefore also a tableau extension rule for s : H , such that one has ^sj : Hij 2 H
for all signed formulas of one of the branching sets f^sj : Hij j (j; ^sj ) 2 M g of
this rule for s : H .
But then the corresponding rule (7.3) for s : H has also a branching set
which consists of some of the signed formulas sj : Hij with signs sj ^sj .
Because one has sj : Hij 2 H for all of them, one has also s : H 2 H by (H5).
2
Proposition 7.1.3. Each Hintikka set can be extended to a saturated Hintikka set.
146
Proof: Let H be a Hintikka set, and let s0 (6= ;) be any sign. Add
rst to H signed atomic formulas which have as their second component
propositional variables which are \new" for H, i.e. consider as a rst extension
the set
H0 = H [ fs : p j p 2 V0 n 2 [H] ^ s0 sg
of signed formulas. Continue the extension by forming for each n 2 N the
sets

Hn+1 = Hn [ s : '(H1 ; : : : ; Hn ) j there is a tableau extension
rule (7.3) for s : '(H1 ; : : : ; Hn ) with
f^sj : Hij j (j; ^sj ) 2 M g Hn for some
M 2 T ('; s) ,
S1
and nally let H = i=0 Hi .
From this construction it is immediately clear that H satises conditions
(H1), (H3), (H4), and (H5) of a saturated Hintikka set.
So it remains to consider condition (H2). Let s : '(H1 ; : : : ; Hn ) 2 H. Then
one either has s : '(H1 ; : : : ; Hn ) 2 H, or one has s : '(H1 ; : : : ; Hn ) 2 Hn+1
for some suitable n 2 N . One then has for each tableau extension rule (7.3)
in both cases f^sj : Hij j (j; ^sj ) 2 M g Hn+2 for some M 2 T ('; s). Hence
also (H2) is satised for H.
2
To get a satisability result for Hintikka sets we hence only need the
satisability of saturated Hintikka sets, because each subset of a satisable
set of signed formulas is again satisable.
Proposition 7.1.4. Each saturated Hintikka set is satisable.
Proof: Let H be a saturated Hintikka set. Consider the set
AH = f j (p) 2 s for all s : p 2 Hg
of valuations. Then one has for all signed formulas s : H the property
, ValS (H; ) 2 s for all 2 AH :
(7.4)
s:H 2H
This is obvious for each atomic w H p from conditions (H1), (H3), (H4)
for saturated Hintikka sets.
The general case follows by induction on the complexity of the w H .
So consider a compound w H '(H1 ; : : : ; Hn ) and assume that (7.4) is
satised for all the ws H1 ; : : : ; Hn and any signs.
Then one has for any sign s :
s : H 2 H , there exists a tableau extension rule (7.3) for s : H
such that ^sj : Hij 2 H for all (j; ^sj ) 2 M of some
M 2 T ('; s)
, there exists a Stableau extension rule (7.3) for s : H
such that Val (Hij ; ) 2 ^sj for all 2 AH and all
(j; ^sj ) 2 M of some M 2 T ('; s)
147
, ValS (H; ) 2 s for all 2 AH :
Here the rst two equivalence statements result immediately from (H2), (H5),
and the assumption of the induction. For the last equivalence the direction
()) is obvious. And for (() one considers for each 1 i n -minimal
signs si such that si fValS (Hi ; ) j 2 AH g. Then one has always si 6= ;
and furthermore verS'[s1 ; : : : ; sn ] s by the completeness of the set of all
signs. But this means according to Corollary 7.1.1 that there exists a tableau
extension rule (7.3) for s : H such that ValS (Hij ; ) 2 sj for all 2 AH and
all (j; ^sj ) 2 M for some M 2 T ('; s).
2
Corollary 7.1.3. Each Hintikka set is satisable.

Proof: This result immediately follows because each Hintikka set can
be extended to a saturated one, which is satisable, and because satisability
is hereditary w.r.t. subsets.
2
Finally we need the notion of \signed consistency property", which corresponds to some kind of approximation of a (saturated) Hintikka set.
Denition 7.1.3. A class C of sets of signed formulas of LS is called a
signed consistency property i one has for all K 2 C that
(CP1)
for all signed atomic formulas s1 : p; : : : ; sk : p 2 K the condition
Tk
i=1 si 6= ; is satised;
(CP2) for each signed compound formula s : '(H1 ; : : : ; Hn ) 2 K and each
tableau extension rule
s : '(H1 ; : : : ; Hn )

:
there exists some M 2 T ('; s) with f^sj : Hij j (j; ^sj ) 2 M g [ K 2 C .
Such a signed consistency property C is of nite character i one has for
each set K of signed formulas:
K 2 C i X 2 C for each nite X K :
It is a simple exercise to see that a signed consistency property C of nite
character is also subset closed, i.e. contains with each K 2 C also each subset
of K as an element.
Theorem 7.1.2 (Model Existence Theorem). If C is a signed consistency property of nite character and K 2 C , then K is satisable.
Proof: Because S is supposed to have nitely many truth degrees, there
are only nitely many signs and hence only denumerably many signed formulas. So assume that 1 ; 2 ; : : : is an enumeration of all signed formulas of LS .
Construct a sequence (Ci )i0 of sets of signed formulas by taking C0 = K
and
148

Cn [ fn g 2 C
Cn+1 = CCn [ fn g ; ifotherwise
n
S
for each n 2 N, and let H = 1
i=0 Ci .
This set H is itself a member of C . For, consider any nite set X H

then there exists some n 2 N with X Cn , which by Cn 2 C means X 2 C
because of the subset closedness of C . But now H 2 C comes from the nite
character of C .
Even more is true: H is an -maximal member of C . If there exists some
H B 2 C then there must exist some signed formula n with n 2 B nH, and
hence with Cn H B = B [ fn g. But this means Cn+1 = Cn [ fn g 2 C
because of the subset closedness, and therefore also n 2 H, a contradiction.
Now H 2 C together with (CP1) gives (H1). And the last mentioned maximality of H together with (CP2) gives (H2). Therefore H is a Hintikka
set, and thus satisable. Hence also K H is satisable.
2
Lemma 7.1.2 (Extension Lemma). Each signed consistency property can

be extended to a subset closed one. And each subset closed signed consistency
property can be extended to a signed consistency property of nite character.
Proof: Suppose that C is a signed consistency property. Consider C + =
fX j X K 2 Cg. Of course, C + is a subset closed family of sets. Because
condition (CP1) is transmitted to subsets, each K 0 2 C + has property (CP1).
Similarly condition (CP2) follows for each K 0 2 C + immediately from the
fact that there exists some K 2 C with K 0 K for which condition (CP2) is
satised. This means that C + is also a signed consistency property.
Now suppose that C + is a subset closed signed consistency property. Consider
C = fK j K 0 2 C + for all nite K 0 K g :
Obviously one has C + C because C + is subset closed. And by denition
C is of nite character. Let K 2 C . If one has s1 : p; : : : ; sk : p 2 K for some
signed atomic formulas, then one has fs1 : p; : : : ; sk : pg 2 C + and hence
Tk
i=1 si 6= ;. That means that K satises condition (CP1). And if one has
some signed compound formula s : '(H1 ; : : : ; Hn ) 2 K and the corresponding
tableau extension rule (7.3), then one has from K 2 C + that there exists some
M 2 T ('; s) with f^sj : Hij j (j; ^sj ) 2 M g [ K 2 C +. Because C + is subset
closed, each nite subset of f^sj : Hij j (j; ^sj ) 2 M g [ K belongs to C + , which
means f^sj : Hij j (j; ^sj ) 2 M g [ K 2 C by denition of C . Hence K satises
also condition (CP2), and it is established that also C is a signed consistency
property.
2
Theorem 7.1.3 (Completeness Theorem). Let S be a system of manyvalued logic with nitely many truth degrees. Then there exists a closed tableau
for the signed formula (W S n DS ) : H if the w H of LS is S-logically valid.
7.2 Sequent Calculi for Many-Valued Logic
149
Proof: Suppose that H is S-logically valid, and assume that there does
not exists a closed tableau for the signed formula (W S n DS ) : H .
Then there exists a complete tableau T for the signed formula (W S nDS ) :
H , and this complete tableau has some (complete) open branch.
Consider for each (complete) open branch B of T the set BB of all signed
formulas which occur as decorations of nodes of B . And let B be the class
of all sets BB of signed formulas which are attached to open branches B of
tableaux for (W S n DS ) : H .
Then one always
has for all signed atomic formulas s1 : p; : : : ; sk : p 2 BB
T
the property ki=1 si 6= ;, because the branch B is not closed. This means
that (CP1) is satised for BB .
So consider a signed compound formula s : '(H1 ; : : : ; Hn ) 2 BB . Because
of the maximality of T this signed formula is marked as used in the branch
B . And this means that in an earlier stage of the construction of the maximal
tableau T a tableau extension rule (7.3) for s : '(H1 ; : : : ; Hn ) has been applied. Hence there exists some M 2 T ('; s) with f^sj : Hij j (j; ^sj ) 2 M g BB .
Thus also (CP2) is satised for BB .
This means that B is a signed consistency property. It can, according
to Lemma 7.1.2, be extended to a signed consistency property B of nite
character. From Theorem 7.1.2 it follows that B has a model, which because
of (W S n DS ) : H 2 BB 2 B is also a model of (W S n DS ) : H , contradicting
the fact that H is S-logically valid.
2

An adaptation of the Gentzen sequent calculi to the systems of many-valued
logic with (only) nitely many truth degrees was introduced by K. Schro ter
[485]. It does not have to refer to the conditions (RT1), (RT2) which had been
basic for the Rosser-Turquette axiomatization. For the presentation of
this method we need some additional notions which we introduce rst.
By an S-sequent we understand any nite sequence (1 ; : : : ; m ) of sets
i LS . Such an S-sequent shall be denoted by (1 j : : : jm ). It is called
nite i all sets i are nite. And we call a valuation : V0 ! W S a
sequent model of an S-sequent (1 j : : : jm ) i is a i -model of i for each
i = 1; : : : ; m.
Furthermore we call an S-sequent (1 j : : : jm ) a valid S-sequent, and
shall write Sval(1 j : : : jm ) to denote this fact, i there does not exist a
sequent model of this S-sequent (1 j : : : jm ).
Remark: In principle also another, dual approach is possible here: to call

a valuation a sequent-model of an S-sequent (1 j : : : jm ) i there exists some i m and some H 2 i with ValS (H; ) = i . Then, of course,
(1 j : : : jm ) has to be taken as valid i each valuation is a sequent-model of
150
(1 j : : : jm ). This approach is e.g. chosen in [24, 403, 466, 526]. The present
one, however, is the better suited one if one is interested in discussing sequent
and tableau calculi in parallel.
For a concise notation of some of the following formulas we shall denote
for each truth degree t 2 W S by K6=S t the set of all truth degree constants of
S which do not denote t. And we consider for S-sequents (1 j : : : jm ) the
condition
(Disj) There exist indices i; j m with i 6= j such that
i \ (j [ K6=S i ) 6= ; :
Proposition 7.2.1. Let (1j : : : jm) be an S-sequent. Then one has:
(i) If (1 j : : : jm ) satises condition (Disj) then Sval(1 j : : : jm ) holds.
(ii) If each set i is a set of atomic ws, i.e. of propositional variables or
truth degree constants, then (Disj) holds true i Sval(1 j : : : jm ) holds
true.
Proof: (i) In the case that (Disj) holds true for (1j : : : jm) then either
there exist two sets i ; j of ws with i 6= j together with some w H 2
i \ j , or some of the sets k contain a truth degree constant which denotes
a truth degree dierent from k . In both cases there cannot exist a sequent
model of the sequent (1 j : : : jm ).
(ii) Assume now that each one of the sets i is a set of atomic ws. If
(Disj) holds true then also Sval(1 j : : : jm) holds true according to (i). If,
however, (Disj) does not hold true, then one has i \ j = ; for all i < j m.
Choosing in this case a valuation : V0 ! W S such that (p) = k for all
1 k m and all propositional variables p 2 k means that is a sequent
model of (1 j : : : jm ), i.e. means that Sval(1 j : : : jm ) does not hold true.
Now we have to consider extensions of sequents. To introduce a w H in

position k m into a sequent (1 j : : : jm ) shall mean the transition to the
S-sequent (1 j : : : jk;1 jk [ fH gjk+1j : : : jm ).
Proposition 7.2.2. For each n-ary connective ' 2 J S, any ws H1; : : : ; Hn,
and each 1 k m one can formulate a necessary and sucient condition
G';k characterizing the introducibility of '(H1 ; : : : ; Hn ) in position k in such
a way that the S-sequent, which results from any given S-sequent (1 j : : : jm )
by introducing the w '(H1 ; : : : ; Hn ) in position k, is a valid S-sequent i
certain S-sequents are valid ones which result from (1 j : : : jm ) via introduction of (some of the) ws H1 ; : : : ; Hn in suitable positions.
Proof: Suppose that the n-ary connective ' and the position k are xed.
Then one can describe the truth degree function verS' by a truth degree table
and extract from this table nitely many n-tuples (tk1 ; : : : ; tkn ); k = 1; : : : ; r,
of truth degrees such that for each valuation it holds true:
151
ValS ('(H1 ; : : : ; Hn ); ) = k i
ValS (Hj ; ) = t1j for each j = 1; : : : ; n
or ValS (Hj ; ) = t2j for each j = 1; : : : ; n
..
.
or ValS (Hj ; ) = trj for each j = 1; : : : ; n :
Denoting furthermore a valuation as a (tl1 ; : : : ; tln )-model of (H1 ; : : : ; Hn )
i one has
ValS (Hj ; ) = tlj for each j = 1; : : : ; n ;
then one immediately gets
f'(H1 ; : : : ; Hn )g has a k -model i
(H1 ; : : : ; Hn ) has a (t11 ; : : : ; t1n )-model
or (H1 ; : : : ; Hn ) has a (t21 ; : : : ; t2n )-model
..
.
or (H1 ; : : : ; Hn ) has a (tr1 ; : : : ; trn )-model :
Because one has furthermore for each truth degree t 2 W S and all 1 j m:
m ; j i j = m ; t(m ; 1) ;
t= m
;1
(7.5)
one easily determines for each S-sequent (1 j : : : jm ), each of its sequent
models , and each w H0 the position k0 with the property that is also
a sequent model of the S-sequent which results from (1 j : : : jm ) by introduction of H0 at position k0 : one has only to choose
k0 = m ; (m ; 1) ValS (H0 ; ) :
Thus we may consider for each S-sequent (1 j : : : jm ) and each k = 1; : : : ; n
another S-sequent (1k j : : : jnk ) which is the result of the iterated introduction
into (1 j : : : jm) of
rst H1 in position m ; (m ; 1) tk1 ;
then H2 in position m ; (m ; 1) tk2 ;
..
.
nally Hn in position m ; (m ; 1) tkn :
According to this construction we immediately have
(1 j : : : jk;1 jk [ f'(H1 ; : : : ; Hn )gjk+1 j : : : jm )
has a sequent model
i
(11 j : : : jm1 ) has a sequent model
152
or (12 j : : : jm2 ) has a sequent model

..
.
or (1n j : : : jmn ) has a sequent model :
And therefore we get for the property of being a valid S-sequent
Sval(1 j : : : jk;1 jk [ f'(H1 ; : : : ; Hn )gjk+1 j : : : jm )
(7.6)
i Sval(1j j : : : jnj ) for each j = 1; : : : ; n :
The right hand side of this last equivalence is the condition G';k we were
looking for.
2
For completeness it should be noted that these conditions G';k in general
are not uniquely determined. Nevertheless, to formulate such conditions often
is even rather a routine matter, cf. [476].
Let us illustrate this general consideration with an example for some 3valued system S with truth degree set W3 . We suppose that in S an implication connective ! is present which has the truth degree function seq 1
introduced in (5.82). Then one has the following three introduction rules for
! in the dierent possible places:
(1 j2 ; Aj3 ; A) (10 ; B j20 j30 ) ;
(1 ; 10 ; A ! B j2 ; 20 j3 ; 30 )
(1 j2 j3 ; A) (10 j20 ; B j30 ) ;
(1 ; 10 j2 ; 20 ; A ! B j3 ; 30 )
(1 ; Aj2 ; Aj3 ; B ) (10 ; Aj20 ; B j30 ; B ) :
(1 ; 10 j2 ; 20 j3 ; 30 ; A ! B )
Here we write, as usual in sequent calculi, for [ only a colon.
The validity of S-sequents now is characterized semantically. The next
problem is to nd a suitable logical calculus which adequately formalizes this
notion of validity. This calculus shall be denoted by K m
G indicating that the
propositional system S is supposed to be m-valued. Its purpose is to generate
just the valid S-sequents of S.
Because this logical calculus K m
G has to generate S-sequents, i.e. sequences
of sets of ws of LS of length m, the language of K m
G has to be more expressive
than the language of S. The crucial point has to be that the language for K m
G
has to be able to treat sets of ws of LS , because after being able to do
this it is a routine matter to treat also S-sequents. The extended language
for K m
G thus needs some kind of reference to a suitable set theoretic world.
Fortunately the details of this world are not really important here because
we are going to discuss problems of derivability in K m
G { but need not really
produce particular derivations in K m
G (and only then would we need such set
theoretic details).
153
The axioms of K m
G shall be all those S-sequents (1 j : : : jm ) which satisfy
the condition (Disj). And the inference rules of K m
G shall be the thinning rule
together with an introduction rule for each connective ' 2 J S and each
position k = 1; : : : ; m. With the same notations as used in the previous
proof of Proposition 7.2.2 the introduction rule for the n-ary connective ' in
position k is
(11 j : : : jm1 ); : : : ; (1n j : : : jmn )
(7.7)
(1 j : : : jk [ f'(H1 ; : : : ; Hn )gj : : : jm )
with the premises of this rule determined by the conditions G';k of Proposition 7.2.2. And the thinning rule is
(1 j : : : jm )
(7.8)
(1 [ 1 j : : : jm [ m )
for any S-sequents (1 j : : : jm ); (1 j : : : jm ).
Again, a derivation in the logical calculus K m
G is a nite sequence of Ssequents such that each one of the S-sequents from this sequence is either an
axiom of K m
G or results from previous S-sequents in the sequence by application of one of the inference rules of K m
G . We shall write `G (1 j : : : jm ) to
indicate that the S-sequent (1 j : : : jm ) is derivable in K m
G , i.e. is the last
sequent of some K m
-derivation.
G
Theorem 7.2.1 (Soundness of K mG ). If a S-sequent is K mG -derivable then
it is a valid S-sequent.
Proof: According to Proposition 7.2.1 each axiom of K mG is a valid Ssequent. And the construction of the introduction rules (7.7) together with
the results from the previous proof of Proposition 7.2.2, particularly the necessary and sucient condition (7.6), forces that each of these introduction
rules produces a valid S-sequent if all its premises are valid S-sequents. Thus
it remains to consider the thinning rule. But in the case that (1 j : : : jm) is
a valid S-sequent, also (1 [ 1 j : : : jm [ m ) is a valid S-sequent because
each sequent model of (1 [ 1 j : : : jm [ m ) would also be a sequent model
of (1 j : : : jm ).
2
For the proof of the corresponding completeness theorem for K m
G we need
some further results which also are of interest by themselves.
Proposition 7.2.3. For each S-sequent (1j : : : jm) there hold true:
1. If (1 j : : : jm ) satises condition (Disj) then `G (1 j : : : jm ) holds
true.
2. If each set i is a set of atomic ws, then condition (Disj) holds true i
`G (1 j : : : jm ) holds true.
Proof: (1) Suppose that (1 j : : : jm) satises condition (Disj). If there
exist i 6= j such that i \ j 6= ; then consider some H 2 i \ j , and for
k = 1; : : : ; m the sets:
154

fH g; if k = i or k = j
;
otherwise.
Then the S-sequent (10 j : : : jm0 ) is an axiom which gives `G (1 j : : : jm )
k0 =
via a suitable application of the thinning rule. If there does not exist such a
pair of indices i; j then there exists an index j with j \ K6=S j 6= ;. Then for
k = 1; : : : ; m let:

S
00
k = j \ K6=j ; if k = j
;
otherwise.
00
00
Then again (1 j : : : jm ) is an axiom, and one gets `G (1 j : : : jm ) by an
application of the thinning rule.
(2) Assume that all i are sets of atomic ws. Because of (i) it then suces
to prove that from `G (1 j : : : jm ) it follows that condition (Disj) holds true.
But having `G (1 j : : : jm ) means that either (1 j : : : jm ) is an axiom, or
has been derived only via applications of the thinning rule. However, each
axiom satises (Disj), and the thinning rule (7.8) cannot disturb that (Disj)
holds true. Hence (Disj) holds true for (1 j : : : jm ).
2
By a nite subsequent of an S-sequent (1 j : : : jm ) we understand a nite
S-sequent (1 j : : : jm ) such that k k holds true for each k = 1; : : : ; m.
Theorem 7.2.2 (Compactness Theorem). If each nite subsequent of an
S-sequent (1 j : : : jm) has a sequent model, then also (1 j : : : jm ) has a sequent model.
Proof: The previous proof of the Compactness Theorem 3.2.3 can be repeated almost word by word, exchanging the phrase \set of ws" with the
phrase \S-sequent (1 j : : : jm )", considering nite subsequents of (1 j : : : jm )
instead of nite subsets of , and taking into account sequent models instead
of models. The details are left for the reader.
2
Theorem 7.2.3 (Finiteness Theorem for Entailment). One has for all
S-sequents (1 j : : : jm ):
1. In the case that Sval(1 j : : : jm ) holds there exists a nite subsequent
(1 j : : : jm ) of (1 j : : : jm ) such that one also has Sval(1 j : : : jm ).
2. If the derivability property `G (1 j : : : jm ) holds true then there exists a
nite subsequent (1 j : : : jm ) of the sequent (1 j : : : jm) such that also
`G (1 j : : : jm ) holds true.
Proof: (1) Assume that Sval(1 j : : : jm ) does not hold true for any nite subsequent of (1 j : : : jm ). Then each such nite subsequent (1 j : : : jm )
of (1 j : : : jm) has a sequent model and thus (1 j : : : jm ) has a sequent
model too according to the Compactness Theorem 7.2.2. Therefore Sval(1 j : : : jm )
does not hold true.
(2) Suppose that `G (1 j : : : jm ) holds true and that some K m
G -derivation
of (1 j : : : jm) is given. Let 0 be the set of all those ws of LS which are
155
introduced in the course of this derivation. Obviously 0 is nite. Let 0 be

the set of all truth degree constants of S and of all subformulas of ws from
0 . Then also 0 is nite. Additionally let f be a choice function over the
family of all non-empty intersections i \ j , i.e.
if i \ j 6= ; then f (i; j ) 2 i \ j ;
and let f be undened for i \ j = ;. Then put

i \ j 6= ;
g(i; j ) =def f;f (i; j )g; ifotherwise
for all 1 i; j m. To each S-sequent (1 j : : : jm ) of the K m
G -derivation of
(1 j : : : jm ), which we assume to be given, we relate a transformed S-sequent
(1 j : : : jm ) with
i =def (i \ 0 ) [
m
[
j =1
g(i; j )
for all i = 1; : : : ; m.
Each such transformed sequent is a nite S-sequent. To our K m

G -derivation of
(1 j : : : jm ) we relate the sequence of all the transformed sequents of this
derivation { and this is again an S-derivation. The transformed sequents of
K mG -axioms are again K mG -axioms, and each application of an inference rule in
the original K m
G -derivation transforms into an application of the same inference rule within the transformed derivation. Therefore one has (1 j : : : jm )
as the last member of the transformed derivation and thus `G (1 j : : : jm ).
Theorem 7.2.4 (Completeness Theorem). Every valid S-sequent is a

derivable sequent in the calculus K m
G.
Proof: Suppose Sval(1j : : : jm), and let (1 j : : : jm ) be a nite subsequent of (1 j : : : jm ) with property Sval(1 j : : : jm ). Via (7.6) one can
remove from all members of (the nite sequence) (1 j : : : jm ) successively all
compound ws such that each single removal step creates a nite set of valid
S-sequents. At the end of this procedure a nite set M of valid S-sequents
remains and each S-sequent of this set is a nite S-sequent with all its members sets of atomic ws. According to Theorem 7.2.1 all members of M are
K mG -axioms. Starting from these axioms, all the previous removal steps can
be reversed by successive applications of suitable introduction rules (7.7),
yielding `G (1 j : : : jm ) in the end. And a nal application of the thinning
rule then results in `G (1 j : : : jm ).
2
Hence we know that K m
G generates just the valid S-sequents of S. To realize
also an adequate formalization of the entailment relation j=S we now go on
to characterize this entailment relation via valid S-sequents. To reach this
goal it becomes, however, necessary to distinguish between designated and
156
non-designated truth degrees of S { a distinction we could avoid up to this

point in our discussion.
So let us assume that DS has exactly s members and that there is
DS = fk ; k ; : : : ; ks g :
Furthermore let us understand by a partition of a set (of ws) such a
family (i )in of pairwise Sdisjoint subsets of which covers , i.e. which
satises the condition = ni=1 i .
Theorem 7.2.5. Let H be a w and a set of ws of LS . Then it holds
true that j=S H i for each non-designated truth degree t0 2 W S n DS and
each partition (1 )is of it holds true `G (1 ; : : : ; m ) for the S-sequent
(1 j : : : jm ) with
(
i ; if j = ki for some i s
j = fH g; if j = m ; t0 (m ; 1)
;
otherwise
for each j = 1; : : : ; m.
Proof: According to the denition (3.8) of the entailment relation, j=S
H holds true i each model of is also a model of H , i.e. i each valuation
which assigns a designated truth degree to each G 2 also assigns a
designated truth degree to H . In other words: j=S H holds true i there
does not exist a model of which at the same time is a t-model of H for
some t 2 W S n DS , i.e. i for each non-designated truth degree t 2 W S n DS
it holds true that there does not exist a model of which is a t-model of H .
What we shall do now is to express in the language of S-sequents the
property of a valuation : V0 ! W S to be a model of and a t-model of
fH g. Let be a valuation with this property and DS = fk ; : : : ; ks g. For
each i = 1; : : : ; s consider
i = fH 2 j ValS (H; ) = ki g :
(7.9)
Obviously (i )is is a partition of . Building up an S-sequent (1 j : : : jm )
in putting
(
i ; if j = ki for some i s
j = fH g; if j = m ; t (m ; 1)
(7.10)
;
otherwise
produces an S-sequent (1 j : : : jm ) which has as a sequent model according
to the choice (7.9) of the sets i .
On the other hand, if (i )is is a partition of and if is a sequent
model of the S-sequent (1 j : : : jmS
) built up from this partition according
to (7.10), then is a model of = si=1 i and at the same time a t-model
of fH g.
Thus there does not exist a model of which also is a t-model of H
i for each partition (i )is of its corresponding S-sequent (1 j : : : jm )
1
157
according to (7.10) does not have a sequent model, i.e. is a valid S-sequent.
Because of the completeness and soundness theorems for K m
G our theorem
now is proved.
2
The goal thus is reached to have a characterization of the entailment
relation of S which does { contrary to Theorem 6.4.2 { not rely on the conditions (RT1) and (RT2). However, the characterization given in Theorem 7.2.5
still has a disadvantage: as far as S has more than one designated truth degree, each innite set 0 of ws of LS has innitely many partitions (i )is ,
i.e. 0 j=S H is characterized by the K m
G -derivability of innitely many Ssequents.
A (partial) way out is to refer to the Finiteness Theorem 3.2.4 for entailment j=S and consider instead of 0 j=S H its equivalent j=S H for
some suitable nite 0 . Then, of course, j=S H is characterized by
the K m
G -derivability of nitely many S-sequents. The above mentioned drawback, however, is not really surmounted because there is no eective way, i.e.
no algorithm to determine the crucial nite subset of 0 with property:
0 j=S H i j=S H .
In the case, however, that S has only one designated truth degree, i.e. in
the case DS = f1g = f1 g, this diculty disappears.
Corollary 7.2.1. Suppose that the m-valued propositional system S has only
the single designated truth degree 1. Then it holds true for each w H of LS
and all sets of ws
j=S H i
`G ( jH j;j : : : j;)
and `G ( j;jH j : : : j;)
and `G ( j;j;jH j : : : j;)
..
.
and `G ( j;j : : : j;jH ) :
Proof: In the present situation is the only partition of which has
to be considered for Theorem 7.7. Hence this corollary is an immediate consequence of Theorem 7.7.
2
The sequent notation used up to now is sometimes changed a bit. Particularly in cases where one supposes that the set of designated truth degrees
is the set
DS = f1 ; : : : ; k g
one considers, following [85], instead of the S-sequents (1 j : : : jm ), which
just are sequences of sets of ws, ordered pairs ((1 ; : : : ; k ); (k+1 ; : : : ; m))
of sequences of sets of ws, which then are denoted by
1 ; : : : ; k k+1 ; : : : ; m :
(7.11)
158
Let us call these modied S-sequents for the moment S-sequents . To have
the same notion of validity for both types of sequents it is natural to accept
the following denition.
Denition 7.2.1. An S-sequent 1; : : : ; k k+1; : : : ; m is valid i
their corresponding S-sequent (1 j : : : jm ) is valid, i.e. i Sval(1 j : : : jm )
holds true.
Having in mind the usual notion of validity of sequents for classical logic,
the following characterization for the validity of S-sequents is preferable.
Corollary 7.2.2. An S-sequent 1; : : : ; k k+1 ; : : : ; m is valid i for
each valuation which is a i -model for each set i , 1 i k, of ws there
is some set j , k < j m, of ws such that is not a j -model of j .
And yet another reading of the S-sequents is possible which refers to the
signed formulas used in the previous subsection, now with only truth degrees
as signs: a whole S-sequent S = (1 j : : : jm ) can be interpreted as the set
m
[
S = fi : H j H 2 i g
i=1
of signed formulas. Then it is obviously clear that a valuation is a sequent

model of S i it is a model of the set S of signed formulas. This means for
each S-sequent S = (1 j : : : jm ) also
Sval(S ) , S not S-satisable :
This point of view opens also the way to form sequent calculi for sequents
built up from signed formulas, cf. [22]. These modied calculi have the advantage that their sequents can simply be taken as (nite) sequences4 of signed
formulas.5 And the inference rules in the case of this approach also get a
simpler shape which looks almost standard from the point of view of sequent
calculi for classical logic.
Other kinds of sequent calculi, rst developed outside the eld of manyvalued logic, are of interest in the context of many-valued logics too. An important example form the hypersequent calculi, used in the context of modal
logic in [427], which have hypersequents as their objects, i.e. (nite) sequences
of sequents, and which enrich the usual inference rules of sequent calculi {
which in hypersequent calculi are applied to the components of hypersequents { with some additional rules which are particularly designed to infer
hypersequents from given hypersequents (and which cannot be reduced to
4 The length of these sequences is no longer tied with the number of truth degrees
here. The truth degrees are, instead, represented by the signs.
5 For the case of sequents for classical logic this corresponds to a transformation
of the (classical) sequents ; into the set f? : H j H 2 ; g [ f> : G j G 2 g

of signed formulas.
159
componentwise applications of sequent rules). Again we shall not consider

the details but refer the interested reader e.g. to [12, 13, 17].
Remark: Besides tableau and sequent calculi one has in classical logic a
third type of logical calculi which are closely related to the other two: calculi
of natural deduction. One of the main points of dierence to sequent calculi
is that the calculi of natural deduction have for each connective introduction
as well as an elimination rules, and that they work only with ws and not
with sequents. In their standard form these calculi of natural deduction have
inference rules with single ws as conclusions. It is, however, possible to
generalize this to natural deduction calculi with inference rules which have
sets of ws as their conclusions.
And this type of \multi-conclusion" calculi of natural deduction can be
generalized toward nitely many-valued logics, as sketched in [23]. For this
generalization one has to consider the introduction and elimination rules
depending on the \position" i (corresponding to the truth degree i ) for
which introduction or elimination has to be made. And this forces to have as
premises for such inference rules also some type of S-sequents consisting only
of m ; 1 sets of ws for an m-valued system S. This means that these natural
deduction calculi for many-valued logics do not really become simpler than
the sequent calculi. Therefore we omit further details.
But we mention still another type of calculi which in classical logic are
of particular interest in Computer Science, mainly in the eld of automated
deduction: resolution calculi. Also these systems can be and have been extended to the eld of many-valued logic, e.g. in [15, 18, 19, 190]. Again we
omit the details here.
160
8. Some Further Topics
8.1 Functional Completeness

Each one of our propositional systems S of many-valued logic has in its alphabet a set J S of (primitive) connectives. As in classical logic, from these
(primitive) connectives other ones become denable { and one can either
really add (symbols for) such denable connectives to the alphabet of the
language LS , or one can treat them simply as abbreviations, i.e. one can introduce such additional, dened connectives only in the metalanguage.1 In
any case, one has { or can have, at least { a greater amount of expressibility
as is immediately present in the alphabet of S. The problem thus arises to
characterize for each propositional system of many-valued logic the class of
connectives expressible within this system, i.e. denable in S.
In this section only a particular case of this problem is discussed. Because
of the principle of extensionality, the rst restriction is that we shall have to
ask only for the class of truth degree functions which are denable from the
truth degree functions verS' for all ' 2 J S . Even this is a very complicated
question. Our second restriction hence consists in asking (only) for criteria
under which from J S all (nitary) connectives are denable, i.e. under which
conditions from fverS' j ' 2 J S g all (nitary) truth degree functions over
W S are denable. Therefore, we restrict our discussion to the problem of
functional completeness.2
As in classical logic, a set J of connectives of a logical system S, or
in the case J = J S this system S itself, is called functionally complete i
all nitary connectives over the truth degree set W S can be dened using
only connectives from J . In the same way a set F of (nitary) truth degree
functions over W S is called functionally complete i each (nitary) truth
degree function over W S can be \combined" out of truth degree functions
from F.
1 For the expressive power of the language of S both ways are essentially equivalent.
2 Similar problems can be formulated in the realm of rst-order logic for the (de-
nability of) quantiers. They are, however, not as popular by far as the propositional counterparts we discuss here. Therefore we shall not consider this problem
in the present book. The interested reader may consult e.g. [83].
162
What have to be specied now are the operations which are allowed for
these \combinations" of truth degree functions, i.e. the methods for dening
new connectives out of given ones.
For the propositional languages under consideration, to introduce a new
n-ary connective using a \stock" J of existing3 connectives means to introduce a new compound w of the form \ (p1 ; : : : ; pn )", with pairwise dierent propositional variables p1 ; : : : ; pn , as denitionally equivalent with some
\more complicated" w H (p1 ; : : : ; pn ) which has all of its propositional variables among p1 ; : : : ; pn , and which contains only connectives out of the set
J . And each compound formula of the form \ (H1 ; : : : ; Hn )" is then understood as the result of substituting simultaneously4 the ws H1 ; : : : ; Hn for the
propositional variables p1 ; : : : ; pn, respectively, in the w H (p1 ; : : : ; pn ). Of
course, it is in accordance with this approach that some of the ws H1 ; : : : ; Hn
coincide. And it is also in accordance with this approach that some of the
propositional variables p1 ; : : : ; pn do not occur in H (p1 ; : : : ; pn ). Such \missing" variables then act as ctitious variables in the w (p1 ; : : : ; pn ).
On the level of truth degree functions to these denitional possibilities
for the introduction of new connectives there correspond the following operations:
1. the substitution of truth degree functions into the argument places of
other truth degree functions, i.e. the superposition of truth degree functions;
2. the identication of argument places, i.e. of variables in the describing
terms of truth degree functions;
3. the addition of ctitious variables, i.e. of new but \unimportant" argument places.
For the following discussions it is helpful to consider also some further, improper \denitional" operations which on the level of ws are completely obvious ways to proceed: these are the possibilities to \dene" a propositional
connective by itself,5 to choose the \compound" w H (p1 ; : : : ; pn ) simply as
the atomic w pk for some k = 1; :::; n, and to exchange two of the argument
places of a given connective. Again transformed to the level of truth degree
functions this means to consider also the possibilities:
4. to treat each truth degree function as a superposition of itself;
3 This means, primitive ones which belong to the set J S of connectives, or previously dened ones.
4 This means that the ws have to be substituted in parallel, not successively for
the propositional variables they substitute. The essential dierence is that in

the case of a successive substitution the nal result may depend on the order of
the single substitutions. In the case of simultaneous substitutions, however, the
result is uniquely determined.
5 Of course, this means on the level of truth degree functions as well as on the
level of formulas that no \new" object is dened.
163
5. to accept as primitive each n-ary projection prnk onto the k-th argument
characterized by the equation
prnk (x1 ; : : : ; xn ) =def xk ;
6. to exchange argument places.
Quite naturally one is thus led to all those operations for truth degree functions which one is usually considering in the theory of function algebras.
Indeed, the algebraic theory of function algebras is not forced to restrict itself to the sets of truth degrees we have in mind in this book, and may also
allow for other operations with functions { but this possibly greater generality is not important for the present purposes. The interested reader may e.g.
consult [345, 424] for the general theory. However, to a large amount even this
general theory is restricted to the case that the functions under consideration
map some given nite set into itself. Accordingly we again restrict here to
the consideration of nitely many-valued propositional systems S, assuming
W S = Wm , i.e. assuming m-valuedness of S.
Some further notation needs to be introduced. By Pm(n) the set of all n-ary
functions from Wm into Wm is denoted:
Pm(n) =def ff j f : Wmn ! Wm g :
And Pm shall be the set of all truth degree functions over Wm :
Pm =def
1
[
n=1
Pm(n) :
In accordance with a common way of understanding in algebra, we identify

each constant unary truth degree function with the truth degree which is the
only value of this particular function. In this sense, Wm is a subset of Pm .
For any set F of truth degree functions from Pm , by hFi the function algebra generated by F is denoted, i.e. the -smallest class G Pm of functions
which is closed under all the above mentioned operations (1) to (6) for truth
degree functions, and which is a superset of F: F G. Particularly this means
that with any n-ary function f 2 hFi and all k-ary functions g1 ; : : : ; gn 2 hFi
also all the functions h1 ; : : : ; h6 are members of hFi which are characterized
by the equations:
h1 (x1 ; : : : ; xk ) =def f (g1 (x1 ; : : : ; xk ); : : : ; gn (x1 ; : : : ; xk )) ;
h2 (x2 ; : : : ; xn ) =def f (x2 ; x2 ; x3 ; : : : ; xn ) ;
h3 (x0 ; x1 ; : : : ; xn ) =def f (x1 ; : : : ; xn ) ;
h4 (x1 ; : : : ; xn ) =def f (x1 ; : : : ; xn ) ;
h5 (x1 ; : : : ; xn ) =def prni (x1 ; : : : ; xn ) for all i = 1; : : : ; n ;
h6 (x1 ; : : : ; xn ) =def f (x2 ; x1 ; x3 ; : : : ; xn ) :
Obviously, these six possibilities realize the operations (1) to (6) mentioned
above in the same order.
164
These notations immediately lead to a \simple" characterization of functional completeness because we have for all classes F Pm :
F functionally complete , hFi = Pm :
In any case, Pm is an innite set of functions. Each one of the sets Pm(n)
however is nite. A rst question concerning functionally complete sets F of
functions thus is whether there exist nite functionally complete sets. The
next theorem gives a positive answer.
Theorem 8.1.1. For each k = 1; : : : ; m let ck and jk be the unary functions
characterized by the equations
n
ck (x) =def k ;
jk =def 0; if x = k
(8.1)
1 otherwise.
Then the set of functions
FP = fmax; min; c1 ; : : : ; cm ; j1 ; : : : ; jm g
(8.2)
is functionally complete.
Proof: Using the truth degree enumeration (5.7) one can for each function f 2 Pm(n) and all t1 ; : : : ; tn 2 Wm characterize the value f (t1 ; : : : ; tn ) as
that one among all the values f (t1 ; : : : ; tn;1 ; k ); k = 1; : : : ; m, for which also
k = tn holds true, i.e. for which also jk (tn ) = 0 holds true. This observation
yields immediately

f (x1 ; : : : ; xn ) = min maxff (x1 ; : : : ; xn;1 ; 1 ); j1 (xn )g ;
maxff (x1 ; : : : ; xn;1 ; 2 ); j2 (xn )g ; : : : ;
(8.3)

maxff (x1 ; : : : ; xn;1 ; m ); jm (xn )g :
But each one of the functions f (x1 ; : : : ; xn;1 ; k ) is an (n ; 1)-ary function.
Therefore (8.3) provides a representation of f 2 Pm(n) as a superposition of
functions from the sets Pm(n;1) and FP .
Assuming n 2 allows to continue this type of \reduction" and to represent all the (n ; 1)-ary functions in (8.3) as superpositions of functions
from the sets Pm(n;2) and FP , i.e. to represent f itself as a superposition of
functions from the sets Pm(n;2) and FP . The iteration of this reduction nally
yields a representation of f as a superposition of functions from Pm(1) [ FP .
Hence it only remains to show that each unary function can be represented
as a superposition of functions from FP . So let g 2 Pm(1) . Then again for each
t 2 Wm the value g(t) is that one among all the values g(k ); k = 1; : : : ; m,
for which also jk (t) = 0 holds true. Therefore one has

g(x) = min maxfcf ( ) (x); j1 (x)g; maxfcf ( ) (x); j2 (x)g; : : :

: : : ; maxfcf (m) (x); jm (x)g
1
165
and hence is able to represent f as a superposition of functions from FP . But

this means hFP i = Pm .
2
The functions max; min from this set FP did appear already in Chapter 5
as candidates vel1 ; et1 for truth degree functions of a disjunction and a conjunction connective in many-valued logic. The functions jk , for k = 1; : : : ; m,
of FP are dual to the truth degree functions jk and of course also unary truth
degree functions. So we have:
Corollary 8.1.1. Each (nitary) truth degree function of a nitely manyvalued propositional system S can be represented as a superposition of at most
binary truth degree functions.
In the case that one does not have that W S = Wm for some suitable
m 2 N there nevertheless exists a bijection between W S and some suitable
truth degree set Wm , because W S is supposed to be nite. Also in this case
one can apply Theorem 8.1.1 to infer this corollary via that bijection.
It is interesting to notice that Corollary 8.1.1 holds also true for innitely
many-valued systems, cf. [505]. For innite truth degree sets this set of at
most binary truth degree functions is also innite, for nite truth degree sets
it is also nite, however, and thus much easier surveyed.
The knowledge of particular functionally complete sets of truth degree
functions does not only mean to know interesting examples. It is a cornerstone
for testing also further sets of truth degree functions for their functional
completeness.
Proposition 8.1.1. Suppose that F Pm is a functionally complete set of
(truth degree) functions and that each function f 2 F can be represented
as a superposition of functions from G Pm . Then also G is functionally
complete.
Proof: Let F; G Pm and hFi = Pm as well as F hGi. Because
h: : :i is a closure operator like CnS , i.e. has the properties
of embedding,
monotonicity, and idempotency as mentioned for CnS in Theorem 3.2.2, one
now has Pm = hFi hhGii = hGi, which means that G is functionally
complete. In other words: if one has for any f 2 Pm a representation as a
superposition of functions g 2 F, and if one substitutes for each such g 2 F
a representation of g as a superposition of functions from G, then one gets a
representation of f as a superposition of functions from G.
2
From this point of view it is particularly important to know small sets of
truth degree functions which nevertheless are functionally complete. Results
of Post [426] and Webb [565] provide such examples.
Theorem 8.1.2. Functionally complete sets of truth degree functions are the
sets
fvel1 ; non2 g and fshg
166
with the (generalized Sheffer-) function sh : Wm2 ! Wm characterized by

the equation

1;
if x = y = 0
sh(x; y) =def maxfx; yg ; 1 otherwise.
m;1
Proof: Because of Proposition 8.1.1 the method of proving these claims

shall be rst to prove that all functions of the functionally complete set FP ,
introduced in (8.2), can be represented as superpositions of vel1 and non2 ,
and then to prove that vel1 ; non2 can be represented as superpositions of sh.
Starting from the binary function vel1 = max it is a routine matter
to represent the maximum-operation applied to any nite number of arguments in taking successively maxfx1 ; x2 ; x3 g = vel1 (vel1 (x1 ; x2 ); x3 ) and
maxfx1 ; : : : ; x4 g = vel1 (maxfx1 ; x2 ; x3 g; x4 ) etc. Correspondingly one introduces the nite iterations of the function non2 : Wm ! Wm as:
non12 (x) =def non2 (x);
nonk2+1 (x) =def non2 (nonk2 (x)) :
(8.4)
Using denition (5.56) one can easily derive direct characterizations of all
these functions nonk2 . The most interesting one of these characterizations
results for k = m yielding
1
nonm
2 (x) = pr1 (x) = x ;
and thus saying that nonm
2 is the identity. Furthermore one has immediately
for all x 2 Wm :
c1 (x) = 1 = 1 = maxfnon2 (x); non22 (x); : : : ; nonm2 (x)g :
(8.5)
And the fact that non2 (x) = x ; m;1 holds true for all x 6= m = 0 yields
ck+1 (x) = nonk2 (c1 (x)) for k = 1; : : : ; m ; 1 :
Hence all the unary functions ck , k = 1; : : : ; m, are superpositions of vel1 and
non2 . To get also representations of the unary functions jk we rst observe
;i
that for each 1 i m the equality nonm
2 (j ) = 0 holds true i i = j is
the case. Furthermore one has for each t 2 Wm
;i+1 (t) = 1 , t = i :
nonm
2
At all thus one has for each k = 1; : : : ; m:
;1
m;k+1 (x))) ;
jk (x) = nonm
2 (vel1 (c2 (x); non2
and all unary functions of FP are represented as superpositions of vel1 ; non2 .
Therefore it remains to represent also the binary function min = et1
because this \binary minimum" then can be extended to the minimumoperation for any nite set in the same way as this extension earlier was
achieved for the maximum-operation. But because one has
minfx; yg = et1 (x; y) = non1 (vel1 (non1 (x); non1 (y)))
167
it suces to represent non1 as superposition of vel1 ; non2 . To reach this goal

we rst remark that the function non1 can either be characterized by the
equation non1 (x) = 1 ; x or by the condition
non1 (i ) = m;i+1 for all i = 1; : : : ; m :
And secondly we introduce a family of unary functions gi;k for i; k = 1; : : : ; m
by
gi;k (x) = noni2;1 (vel1 (jk (x); cm;i+1 (x))) :
Straightforward computations lead to the simpler characterizations
n
x = k .
gi;k (x) = 0i ; ifotherwise.
Therefore one has the representation
non1 (x) = maxfgm;1(x); gm;1;2 (x); : : : ; g1;m(x)g
and knows that all the functions gi;j as well as the function max are representable as superpositions of vel1 and non2 .
Hence the set fvel1 ; non2 g is functionally complete. And it remains to
represent vel1 and non2 as superpositions of sh. But one gets by elementary
calculations the representations
non2 (x) = sh(x; x);
;1
m;1
vel1 (x) = sh(nonm
2 (x); non2 (y ))
which prove that also fshg is functionally complete.
2
The idea to discuss the functions which can be represented via superpositions of given functions is not only useful for proving functional completeness,
it can also be used to disprove this property.
Proposition 8.1.2. Suppose that all truth degree functions of some class
F Pm of functions satisfy the normal condition and that m > 2 holds true.
Then F is not functionally complete.
Proof: All the methods for the superposition of functions mentioned in
the basic list of functions h1 ; : : : ; h6 on p. 163 obviously lead from functions
which satisfy the normal condition to functions which satisfy the normal
condition. But for m > 2 there exist truth degree functions which do not
satisfy the normal condition, e.g. non2 or any constant function with function
value dierent from 0 and 1.
2
Each one of the criteria for functional completeness discussed up to now
does refer to particular sets of functions. What is lacking is some kind of
\general" criterion. The most natural criterion, however, is not a \positive"
criterion for functional completeness but a \negative" one for being not functionally complete. It states a necessary and sucient condition, and hence it
does not matter whether it is a \positive" or a \negative" criterion.
168
Theorem 8.1.3. A class of functions G Pm is not functionally complete

i there exists some maximal function algebra6 F such that G F.
Proof: In the case that F is a maximal function algebra and G F holds
true, one has hGi hFi = F Pm and therefore hGi 6= Pm . Hence G is not
functionally complete.
On the other hand, suppose that G is not functionally complete. Then
one has hGi Pm , and it has to be shown that there exists some maximal
function algebra FG with property hGi FG . And this proof shall be given by
reference to Zorn's lemma. Because the class of all function algebras F Pm
is partially ordered by inclusion, what has to be shown is that for each chain
(Fi )i2I of function algebras Fi Pm there exists a function algebra F Pm
such that Fi SF holds true for all i 2 I .
Let F = i2I Fi . Then obviously Fi F holds true for all i 2 I . Now
consider some f 2 hF i. Then f is a superposition of nitely many functions
from F which belong to nitely many of the function algebras Fi and hence to
a maximal one Fi , because (Fi )i2I is a chain. But then f 2 hFi i = Fi F
holds true. Hence F is a function algebra.
And also F Pm holds true. Otherwise one would have F = Pm and
hence sh 2 F . But in this case sh 2 Fk should be true for some k 2 I
yielding Fk = hFk i = Pm and contradicting our assumption that Fk has to
be a function algebra, such that Fk 6= Pm . Therefore F Pm .
Now Zorn's lemma yields the existence of a maximal function algebra F
such that G F Pm .
2
0
Corollary 8.1.2. If a function f 2 Pm does not belong to any maximal

function algebra of Pm , then ff g is functionally complete.
Theorem 8.1.3 may be considered as an acceptable characterization of
all sets F Pm which are not functionally complete, and hence also of all
functionally complete F Pm , if one is able to determine all the maximal
function algebras of Pm . For the case m = 2 this was already done by Post
[425], for m = 3 the problem was solved by Jablonskij [284], and in the
general case nally by Rosenberg [459]. To describe his general result, some
further notions are needed.
A k-ary relation % in Wm , i.e. a set of k-tuples with components from
Wm , is called invariant w.r.t. some n-ary truth degree function f 2 Pm i
for any family of k-tuples
(si1 ; : : : ; sik ) 2 % with i = 1; : : : ; n
the \componentwise" application of f again produces a k-tuple from %:
(f (s11 ; : : : ; sn1 ); f (s12 ; : : : ; sn2 ); : : : ; f (s1k ; : : : ; snk )) 2 % :
6 Maximality of such a function algebra F means that there does not exist some
function algebra G with property hFi G Pm , i.e. is -maximality.
169
The polymorph Pol(%) of such a k-ary relation % shall be the class of all truth
degree functions under which % is invariant.
Pol(%) =def ff 2 Pm j % invariant w.r.t. f g :
The functions f 2 Pol(%) often are called polymorphisms of %.
An equivalence relation % in Wm is nontrivial i neither % is the identity:
% 6= f(x; x) j x 2 Wm g, nor is it the full relation: % 6= Wm Wm . We consider
Em =def set of all nontrivial equivalence relations in Wm .
For a (re exive) partial ordering % in Wm an element a0 2 Wm is the7 %minimum i a0 %b holds true for each b 2 Wm , and an element b0 2 Wm is
the %-maximum i a%b0 holds true for each a 2 Wm . We consider
Om =def set of all those partial orderings % in Wm which have a
%-minimum as well as a %-maximum.
By a permutation of Wm one understands a bijection from Wm onto itself,
often visualized as a kind of \rearrangement" of the elements of Wm . Each
such permutation g of Wm generates an equivalence relation g in Wm by
putting a g b for a; b 2 Wm i there is a nite iteration of g which \transforms" a into b, i.e. i there exists some k 0 such that gk (a) = b holds true
for the k-th iteration gk of g. Then we consider
Pm =def set of all permutations g of Wm for which there exists a
prime number p such that each equivalence class of g
has exactly p elements.
Because each permutation of Wm is a set of ordered pairs, it is also a binary
relation in Wm . Hence Pm is a set of relations.
A k-ary relation % in Wm is central i (i) % 6= Wmk holds true and (ii)
there exists some ; 6= C Wm such that for all a1 ; : : : ; ak 2 Wm and all
1 i < j k there hold true
ai 2 C ) (a1 ; : : : ; ak ) 2 % ;
ai = aj ) (a1 ; : : : ; ak ) 2 % ;
and (iii) % invariant is under any permutation g of the set f1; 2; : : :; kg, i.e.
satises
(a1 ; : : : ; ak ) 2 % ) (ag(1) ; : : : ; ag(k) ) 2 % :
Then let
Cm =def set of all central relations in Wm .
A nonempty family (i )ik , k 1, of equivalence relations in Wm is called
h-regular for some 3 h m i each one of the equivalence relations
7 It is an easy exercise to show that the %-minimum as well as the %-maximum are
uniquely determined in the case that they exist. The reader essentially has to
refer to the antisymmetry of the re exive relation %.
170
1 ; : : : ; k has exactly h equivalence classes and furthermore each family

(Ai )iTk of equivalence classes Ai of i , i = 1; : : : ; k, has a nonempty intersection: ki=1 Ai 6= ;. Each such h-regular family (i )ik of equivalence relations
uniquely determines an h-ary relation % characterized by the condition
(a1 ; : : : ; ah ) 2 % , (a1 ; : : : ; ah ) has at least two i -equivalent components for each 1 i k
for all a1 ; : : : ; ah 2 Wm . Now let
Bm =def set of all relations in Wm which are determined by some
h-regular family of equivalence relations in Wm .
And nally let p be a prime number. Then an abelian p-group is such an
abelian group8 (A;
) for which each \p-th power" ap = a

a, with p
operands a, of any a 2 G satises b
ap = b for all b 2 G. Then let
Gm =def ; if m is not a power of some prime number,
and for the cases that m = pn holds true for some prime number p and some
n 1 let
Gm =def set of all quaternary relations R which can be described
as R = f(a1 ; : : : ; a4 ) j a1
a2 = a3
a4 g for a binary
operation
in Wm such that (Wm ;
) is an abelian
p-group.
Now we are able to state the characterization of the maximal function algebras we look for.
Theorem 8.1.4. A class F Pm of truth degree functions is a maximal
function algebra i it holds true
F = Pol(%) for some % 2 Em [ Om [ Pm [ Cm [ Bm [ Gm .
The proof of this theorem is quite long and complicated and shall not be
given here. The interested reader should consult the original source [459], but
also [424] presents a well-written sketch of the basic ideas of the proof.
Because one has obviously Om 6= ; for each m 2, this last theorem
yields also that in Pm there always exist maximal function algebras.
Corollary 8.1.3. A class F Pm of truth degree functions is functionally
complete i there exists for each relation % 2 Em [ Om [ Pm [ Cm [ Bm [ Gm
some function f 2 F such that % is not invariant under f .
8 In general a group is some set G together with a binary operation

in G which
is associative and such that each equation of the form a

x = b or x
a = b for
a; b 2 G has a solution. Such a group is called abelian i
is also commutative.
8.2 Decidability of Propositional Systems
171
Proof: A set F Pm is functionally complete i it is not a subset of any

maximal function algebra of Pm , i.e. i F 6 Pol(%) for some % 2 Em [ : : : [Gm .
Now the claim results from the denition of Pol(%).
2
It is interesting to notice that Pm has always only nitely many maximal
function algebras despite the fact that there exist innitely many function
algebras F Pm , viz. @0 for m = 2 and 2@ for m 3, cf. [424]. A formula to
determine the number (m) of maximal function algebras in Pm was given in
[460]. It shows that this number is growing rapidly with m and has for small
m e.g. the following values, cf. also [461]:
0
(m) 5 18 82 643 15,182 7,848,984 > 549 109
Proposition 8.1.3. Each functionally complete class F Pm of functions

has a subset F F with at most (m) elements which itself is already
Proof: Let F Pm be a functionally complete class of functions, and
assume that all the maximal function algebras of Pm are enumerated (without
repetitions) as G1 ; : : : ; G(m) . Then one has FnGk 6= ; for each 1 k (m).
Select some fk 2 F n Gk for each such k and consider F = ff1; f2 ; : : : ; f(m) g.
Then by construction F 6 Gk holds true for each 1 k (m), i.e. for each
maximal function algebra. Therefore F is functionally complete according to
Theorem 8.1.3.
2
8.2 Decidability of Propositional Systems

A subset of the set of all ws of a (propositional) system S is called
decidable i there exists an algorithm which allows to test for each w H
(in nitely many \elementary" steps9 ) whether H belongs to or does not
belong to .
Obviously each nite set of ws is decidable. The algorithm which tests
membership of H in simply consists in comparing each one of the nitely
9 Each such \test" has to be realized in nitely many steps. The number of steps
may, however, change depending on the w H to be tested. But each step has
to be \elementary" in the sense that its execution has to be possible \purely
mechanical", i.e. could be done by following clearly determined rules, which are
xed previous to the choice of H , and therefore could be executed even by a
computer.
172
many elements of with H . And to do this it suces to have a complete list

containing just the ws which are elements of .
Usually one considers this problem of decidability only for quite specic
sets of ws, like the set of logically valid ws, the set of logical falsities, or
the set of satisable ws. Among these possibilities the rst and the third
one are more important than the second one. The third one is often called
the satisability problem for S, and the rst one is even identied with the
decidability of S in the sense that decidability of S is usually understood as
decidability of the set of logically valid ws of S.
In classical logic, the method of truth tables is one of the standard algorithms to test a w H of classical logic whether it is logically valid { or
whether it is satisable, or whether it is a contradiction. The testing procedure simply is to write down the complete truth table for H . This is easily
possible because H describes a truth value function of nitely many arguments, and because the whole approach toward classical logic is based on the
principle of extensionality. But also in many-valued logic one has the whole
approach based on the principle of extensionality, and therefore each w H
describes a truth degree function of nitely many variables, and again this
function can be described by reference to the complete truth degree table for
H which is built up as in classical logic by reference to all the propositional
variables of H and all the subformulas of H . Of each of these sorts of objects
one has also in many-valued logic only nitely many for each particular w
H : this is essentially caused by the two facts that each w is built up from
propositional variables in nitely many steps, and that each connective has
nite arity.
The problem whether this method of truth degree tables provides an
algorithm to test a given w H for being logically valid (or for being satisable
or even logically false) therefore only depends on whether it is possible to
write down each such truth degree table in nitely many steps. And this
is possible if S has only nitely many truth degrees: one has to start the
complete truth degree table with columns for each one of the propositional
variables of H , has to continue with columns for the subformulas of H (in
order of increasing complexity, preferably) and has to consider in its rows
all possible coverings of these variables by truth degrees. And this gives only
nitely many possibilities.10 As a side remark let us mention that for these
considerations it is not important whether S has a nite set J S of connectives
or not.
Thus we have proved the following theorem.
Theorem 8.2.1. Each system S of propositional many-valued logic with a
nite set W S of truth degrees is decidable. Moreover, for each such system
also the sets of satisable ws and of logically false ws are decidable.
10 The number of rows of the complete truth degree table for
H is mn provided
W S has m elements and n dierent propositional variables occur in H .
8.3 Product Systems
173
For innitely many-valued systems S of propositional many-valued logic,

i.e. for systems with innite sets W S of truth degrees, it is not possible to
prove such a general result. For those systems it essentially depends on the
particular system S whether S is decidable or not. Both possibilities exist, and
the problem hence has to be discussed separately for each innitely manyvalued system.
8.3 Product Systems

We have mentioned in the very beginning that for applications of systems
S of many-valued logic one always has the problem of the suitable choice
of the truth degree sets W S and of the intuitive interpretation of the truth
degrees. One of the possibilities for such an interpretation is to assume that
the truth degrees combine some number { say k { of dierent, perhaps even
independent, aspects (or: criteria) for the evaluation of propositions. In such
a situation it seems preferable to suppose that each one of these dierent
aspects has its own kind of truth degrees, i.e. its set W (i) , i = 1; : : : ; n, of degrees, and that the \combination" of these dierent types of \aspect related"
degrees consists in taking the cartesian product W S = W (1) W (k)
of the single degree sets. The truth degrees in such a case become k-tuples.
More general even, the indices 1; : : : ; k may be replaced by some (possibly
innite) set I of indices, which even may be partially ordered { indicating
an ordering or ranking between
Q the aspects (or criteria) for evaluation. Then
one has e.g. to take W S = i2I W (i) . But with a partially ordered set I of
indices and particularly W (i) = W for all i 2 I and some partially ordered
set W , also the choice of W S as the set of all monotonic functions f : I ! W
may intuitively be preferable, cf. e.g. [53].
It is then often also suitable to assume that the combination of these truth
degrees proceeds componentwise, i.e. one has for each one of the connectives
' 2 J S , one intends to discuss, a corresponding truth degree function verS'
which also acts componentwise on its arguments via reference to componentrelated further truth degree functions.
Just this situation that truth degrees are k-tuples and that this internal
structure of the truth degrees is of importance for dening and understanding
the behavior of the truth degree functions is indicated in the case that one
speaks of multi-dimensional truth degrees.
For some types of applications this multi-dimensional setting is a most
natural one, cf. [261]. And this setting even grasps an idea that can be traced
back up to the scholastic logician J. Buridan, e.g. to his [78], as explained
in [261]. His main idea was that truth, as a philosophical notion, combines
the dimensions of correspondence to reality, and of semantic (in)competence.
We shall meet this idea of multi-dimensionality again e.g. in the discussions
of the problem of presuppositions in Chapter 20. And it has also been used
174
in discussions about the modeling of diagnostics e.g. in a pedagogical setting,

cf. [316, 515].
This whole idea nds its natural realization if one considers S as a product
of logical systems S1 ; : : : ; Sk which all have the same language LS , i.e. the
same set J S of connectives and which interpret e.g. a binary connective
' 2 J S by the truth degree function
;

verS'(u; v) = verS' (u; v); : : : ; verS' (u; v)
(8.6)
using the truth degree functions verS' of ' in the \factor" systems Si . With
this approach towardQthe connectives of J S the system S is denoted as the
product system S = ki=1 Si . Product systems in this sense have in modern
times (as it seems) rst been used by Jaskowski [285] in investigations toward intuitionistic logic, cf. Chapter 22. They also appear in Lukasiewicz's
later discussion [337] of a four-valued system for modal logic. And their use
has been sketched in [456] for an application of an 8-valued system to (elementary) geometry: the main idea is a common, parallel treatment of geometric
formulas w.r.t. (i) parabolic (i.e. Euclidean) geometry, (ii) hyperbolic (i.e.
Lobatchevskian) geometry, and (iii) elliptic geometry { and the evaluation
of these formulas by triples of truth-values. The set of truth degrees thus
becomes f?; >g3 in this case. And also the choice of the designated truth
degrees here becomes intuitively important:
having only the triple (>; >; >) as designated truth degree makes all those
geometric formulas (formally) valid which are valid for all three sorts of
geometries;
having all the triples (>; x; y) as designated truth degrees which have > as
their rst component makes all those geometric formulas (formally) valid
which are valid in Euclidean geometry.
As a side remark let us mention that this setting is suciently general, also
for the cases that one has in mind to combine in S into some S-connective
e.g. a \conjunction" '1 of S1 with a \disjunction" '2 of S2 . But such an idea
can also be realized in the { seemingly more restricted { setting indicated by
(8.6): some (more or less tricky) renaming and perhaps \doubling of names"
for the connectives of the systems which are to become the factor systems
does the job. How to proceed in such a case should be intuitively clear, and
it is not necessary to describe the details more generally here.
As an example consider the case that for k = 2 one has S = S1 S2 , with
S1 = S1 = C2 the usual system of classical propositional logic in negation,
conjunction, and disjunction with truth-value set f0; 1g. Then S is a fourvalued system with truth degree set f(0; 0); (0; 1); (1; 0); (1; 1)g, i.e. with a
truth degree
set which in a natural way is not linearly ordered. One may
take J S = f; ^; _g and dene the connectives of S e.g. according to (8.6).
This yields for these connectives the truth degree table characterization with
the tables for negation, conjunction, and disjunction as in Fig. 8.1, where
1
8.3 Product Systems
175
we additionally have written ab instead of (a; b) for the ordered pairs of the
truth degree set W S .
00 01 10 11
11 10 01 00
^ 00 01 10 11
_ 00 01 10 11
00 00 00 00 00
00 00 01
01 00 01 00 01
01 01 01
10 00 00 10 10
10 10 11
11 00 01 10 11
11 11 11

Fig. 8.1. Negation, conjunction, and disjunction tables of S
10
11
10
11
11
11
11
11
It is quite natural
to take as the set of designated truth degrees of the
Q
product system ki=1 Si either all those k-tuples which completely consist of
designated truth degrees of the factor systems Si , i.e. to choose
D1S = DS DSk ;
(8.7)
or all those k-tuples which contain at least one component which is a designated degree for its factor system, i.e. to choose
;

D2S = W S n (W S n DS ) (W S n DSk ) :
(8.8)
Similar to the dierent possibilities for introducing connectives into such
product systems which are not determined by connectives of the factors via
(8.6), one also has quite dierent possibilities for the choice of the designated
truth degrees of the product systems. And this choice is not predetermined
by the notion of product system.
For some of the simpler choices of truth degrees, along with version (8.6)
for the denition of the connectives, one gets easily uniform characterizations
e.g. of the logically valid ws of the product system.
Proposition 8.3.1. For each product system S = Qki=1 Si with connectives
determined according to (8.6) one has
T
1. tautS = ki=1 tautS as set of tautologies for the set D1S of designated truth
degrees given
in (8.7);
S
2. tautS = ki=1 tautS as set of tautologies for the set D2S of designated truth
degrees given in (8.8).
The proof is a simple exercise and left to the reader.
Also
other theoretical problems which are related to such product systems
Q
S = ki=1 Si often can be solved by reference to the corresponding problems
for the factor systems. Thus e.g. in both of the cases (8.7) and (8.8) for
1
176
the choice of designated truth degrees of the product system, this system
is decidable if all the factor systems are, and it is axiomatizable if all the
factor systems are,11 because the properties of recursivity and of recursive
enumerability are preserved under nite unions and nite intersections.
11 This does not mean that one knows a straightforward way to get an axiom system
for such a product system out of axiom systems for the factor systems. However,
some kind of Rosser-Turquette-like axiomatization may often be possible, at
least under some mild restrictions concerning the factor systems. The interested
reader may compare [351] for an axiomatization of a four-valued product system.
Part III
Particular Systems of Many-Valued Logic
177
9. The Lukasiewicz Systems
Historically, many-valued logic was rst presented as a system devised to be

dierent from classical logic and aimed to reach some goals for which classical logic seemed to be inadequate, by the Polish philosopher and logician
Lukasiewicz [334] in 1920. This system did have a third truth degree and
was intended to cover, e.g., modalities like the (future) possibilities, cf. Chapter 21. Soon Lukasiewicz generalized this system to the cases of any nite
number of truth degrees and even to innitely many ones { corresponding to
the truth degree set W0 at rst { and developed during the 1920s the theory of these systems. All these systems with their formal details have been
presented in [339], and their philosophical problems have been discussed in
[335].
It is not problematical at all to extend the innitely many-valued system
with truth degree set W0 to a system with truth degree set W1 too. We
shall do this here and denote the Lukasiewicz systems with truth degree
sets Wm for m 2, W0 and W1 by Lm ; L0 ; L1 , respectively. In all the cases
where the particular set of truth degrees is not of importance, we shall simply
write L or even only L to indicate the Lukasiewicz systems.
9.1 The Propositional Systems

The Lukasiewicz propositional systems have been originally formulated in
negation and implication as basic connectives, i.e. their set J L of connectives
contains the connectives
: (negation); !L (implication)
(9.1)
with the corresponding truth degree functions
verL: = non1 ;
verL!L = seq 2
(9.2)
as already introduced in (5.55), (5.81). Therefore one has for any ws G; H
of the language LL of these systems and any valuation : V0 ! W L:
ValL(:H; ) = 1 ; ValL(H; ) ;
ValL(G !L H; ) = minf1; 1 ; ValL (G; ) + ValL (H; )g :
180
In the (original) language of the Lukasiewicz systems there are no truth

degree constants. And the set of designated truth degrees is (usually)
DL = f1g ;
(9.3)
i.e. the truth degree 1 is the only designated one.
The system L2 with truth degree set W2 = f0; 1g is in [339] explicitly
mentioned as one of the Lukasiewicz systems. Obviously, this system L2
is the classical propositional logic formulated as a system in negation and
implication, and it is the full system of classical logic because the set :; ! of
connectives of classical logic is functionally complete.
Because both truth degree functions non1 ; seq 2 satisfy the normal condition, the Lukasiewicz systems L with 6= 2 are not functionally complete
It is common usage to extend the Lukasiewicz systems by further connectives dened for arbitrary well-formed formulas H1 ; H2 by
H1 _ H2 =def (H1 !L H2 ) !L H2 ;
(9.4)
H1 ^ H2 =def :(:H1 _ :H2 ) ;
(9.5)
H1 $L H2 =def (H1 !L H2 ) ^ (H2 !L H1 ) :
(9.6)
These connectives correspond, obviously, to the connectives of disjunction,
conjunction, and biimplication of classical logic. There are also two other
obvious generalizations of classical conjunction and disjunction in the present
setting, given by
H1 & H2 =def :(H1 !L :H2 ) ;
(9.7)
H1 Y H2 =def :H1 !L H2 :
(9.8)
It is an easy exercise in arithmetical calculations to see that these connectives
have well known truth degree functions:
connective
&
truth degree function vel1 et1 et2 vel2
And the truth degree function verL$L for the biimplication connective $L is
characterized by the equation1
verL$L (x; y) = 1 ; jx ; yj :
(9.9)
The truth degree functions show that neither the two candidates ^; & for generalized conjunctions nor the two candidates _; Y for generalized disjunction
1 This has, by the way, the interesting consequence that the truth degree function
determined by :(H1 $L H2 ) is a metric (in the precise mathematical sense of
this word), i.e. measures the distance of the truth degrees of H1 and H2 .
181
connectives form equivalent generalizations. For reasons that shall become

clear later, the connectives &; Y are called strong conjunction and strong disjunction, and ^; _ are the weak conjunction and the weak disjunction.
In the following considerations of ws of the language LL of the Lukasiewicz systems we shall (almost) always add these new connectives to the
alphabet and thus shall have ws which are easier to read than ws written
down only with negation and implication.
These Lukasiewicz connectives can easily be used to express relationships between truth degrees of ws. To discuss the basic ones of these relationships let H; H1 ; H2 ; : : : ; G be ws of the { extended { language LL .
The basic observation is that because of the choice (9.3) of designated truth
degrees one has
j=L H i ValL(H; ) = 1 for all valuations : V0 ! W L
(9.10)
for each w H of LL . As a rst result one has
j=L (H1 !L H2 ) i ValL (H1 ; ) ValL (H2 ; )
(9.11)
for each valuation ,
and moreover for each valuation : V0 ! W L separately
ValL(H1 !L H2 ; ) = 1 i ValL (H1 ; ) ValL (H2 ; ) ;
(9.12)
as in both cases can be established by simple calculations. Similarly one has
for each valuation : V0 ! W L
ValL(H1 $L H2 ; ) = 1 i ValL (H1 ; ) = ValL (H2 ; ) ;
(9.13)
and thus in general
j=L (H1 $L H2 ) i ValL (H1 ; ) = ValL (H2 ; )
(9.14)
for each valuation .
Therefore the semantical equivalence of ws H1 ; H2 can be expressed in the
Lukasiewicz systems like in classical logic as the logical validity of the
corresponding biimplication H1 $L H2 .
9.1.1 Important tautologies of the Lukasiewicz systems

The gain one has from the knowledge of particular tautologies of these Lukasiewicz systems is like the corresponding gain one has in classical logic from
the knowledge of tautologies: this helps for semantically equivalent transformations of ws as just explained, and it helps to nd suitable (derived)
inference rules in the sense that one has
if j=L H1 and j=L H1 !L H2 ; then j=L H2 .
(9.15)
Hence we shall look here, previous to theoretical considerations on these
systems, particularly for tautologies of the L-systems which have the form of
biimplications and of implications.
182
First we observe that all the conjunction and disjunction connectives of

the L-systems behave commutatively and associatively like in classical logic,
i.e. one has, with > as any one of the connectives ^; _; &; Y, the results
(T1) j=L H1 > H2 $L H1 > H2 ,
(T2) j=L H1 > (H2 > H3 ) $L (H1 > H2 ) > H3 .
Here we have adapted the common behavior that the conjunction and disjunction connectives have preference over the implication and biimplication
connectives. And we shall use this convention throughout this book. The
weak connectives ^; _ furthermore are \idempotent", i.e. have the absorption properties
(T3) j=L H ^ H $L H;
j=L H _ H $L H ,
which, however, are lacking for the strong connectives &; Y, for which one
instead only has
(T4) j=L H ^ H !L H;
j=L H _ H !L H .
Here indeed the corresponding L-biimplications are not L-tautologies because
the truth degree of H & H may be really smaller, and the truth degree of
H Y H really larger than the truth degree of H . (One has simply to take
0 < ValL (H; ) < 1 to reach this situation.)
One also has distributivity of ^ over _, and distributivity of _ over ^:
(T5) j=L H1 ^ (H2 _ H3 ) $L (H1 ^ H2 ) _ (H1 ^ H3 ) ,
(T6) j=L H1 _ (H2 ^ H3 ) $L (H1 _ H2 ) ^ (H1 _ H3 ) .
Again, however, these distributivity properties are lacking for the strong connectives. For them one only has the inequality
;

ValL H1 & (H2 Y H3 ) $L (H1 & H2 ) Y (H1 & H3 ); 21
(9.16)
L
for each valuation : V0 ! W , as well as the dual inequality which results
from exchanging & with Y.
To prove this inequality (9.16) one has, according to (9.9), only to prove
that for any truth degrees s; t; r one has
(9.17)
jet2 (s; vel2 (t; r)) ; vel2 (et2 (s; t); et2 (s; r))j 21 :
With the intermediate notation u = jet2 (s; vel2 (t; r)) ; vel2 (et2 (s; t); et2 (s; r))j
we rst claim that ;u 21 holds true for u < 0. Assuming u < 0 one has
0 < minf1; et2 (s; t) + et2 (s; r)g ; maxf0; s + vel2 (t; r) ; 1g = ;u
and hence immediately
0 < minf1; et2 (s; t) + et2 (s; r)g + minf0; 1 ; s ; vel2 (t; r)g = ;u :(9.18)
Thus one has to have
0 < et2 (s; t) + et2 (s; r) + 1 ; s ; vel2 (t; r) :
183
However maxfx1 ; x2 g + maxfy1 ; y2g = maxfx1 + y1; x1 + y2; x2 + y1; x2 + y2g

together with the denitions (5.9) of et2 and (5.76) of vel2 yields via some
elementary transformations, and by putting v = (s + t + r ; 1), the inequality
0 < maxf;s; ;v; t ; 1; ;r; r ; 1; ;t; v ; 1; s ; 1g ;
i.e. (s + t + r ; 1) < 0 or (s + t + r ; 1) > 1. For (s + t + r ; 1) < 0 one has
immediately
et2 (s; t) + et2 (s; r) 0
contradicting (9.18), so this is impossible. And for (s + t + r ; 1) > 1 one has
s + t ; 1 > 1 ; r 0, as well as s + r ; 1 > 1 ; s 0 and t + r > 2 ; s 1,
which together gives
;u = minf1; (s + t ; 1) + (s + r ; 1)g ; s
= minf1 ; s; s + (t + r ; 2)g minf1 ; s; sg 12 :
But also in the case u 0 one has u 12 . To see this, observe that one has
u = maxf0; s + minf1; t + rg ; 1g
; minf1; maxf0; s + t ; 1g + maxf0; s + r ; 1gg
= maxf0; minfs; s + t + r ; 1gg
+ maxf;1; minf0; 1 ; s ; tg + minf0; 1 ; s ; rgg :
Elementary calculations as in the previous case u < 0 lead to
u = max f ; 1; minfs ; 1; : : :g; minf0; : : :g;
minfs; : : : ; (s + t + r ; 1) + (1 ; s ; t) + (1 ; s ; r)gg ;
which means that u now is the maximum of four terms, the rst three of
which are 0, and a fourth one: minfs; : : : ; 1 ; sg 21 . Hence u 12 holds
true in this case, and we have at all juj 12 as needed for (9.17).
Even if one could guess that the L-tautologies mentioned up to now indicate that the strong connectives, particularly the strong conjunction is not
as important as the weak one, this would be misleading. The reference to the
strong conjunction is, e.g., indispensable in the importation/exportation law
which holds true in the L-systems in the form
(T7) j=L (H1 !L (H2 !L H3 )) $L (H1 & H2 !L H3 ) .
The corresponding formula which uses the weak conjunction instead of the
strong one, i.e.
(H1 !L (H2 !L H3 )) $L (H1 ^ H2 !L H3 )
(9.19)
is not a tautology of any one of the systems L with 6= 2.
Statement (T7) may be established by quite direct calculations with the
truth degree functions for the left hand and the right hand side of this biimplication. This is more or less a routine matter and goes along the same lines
184
as the previous calculations in connection with (9.16) and (9.17). Therefore

we shall not give the details here.2
To see that (9.19) is not an L -tautology for 6= 2, consider such a
valuation and such ws Hi , i = 1; 2; 3, such that for the truth degrees
ti = ValL (Hi ; ) one has 1 > t1 t2 > 0 = t3 . In this case the w (9.19) gets
the truth degree t1 < 1 and hence a non-designated one.
Both strong connectives &; Y are tied together via generalized deMorgan
laws:
(T8) j=L :(H1 & H2 ) $L (:H1 Y :H2 ) ,
(T8a) j=L :(H1 Y H2 ) $L (:H1 & :H2 ) ,
which, correspondingly, also hold true for the weak connectives ^; _ because
of denition (9.5) and the fact that the law of double negation holds true as
(T9) j=L ::H $L H
in all the Lukasiewicz systems. The strong conjunction also has to be used
in the following law of syllogism
(T10) j=L (H1 !L H2 ) & (H2 !L H3 ) !L (H1 !L H3 ) .
The most simple relationships between the corresponding strong and weak
connectives are given by
(T11) j=L H1 & H2 !L H1 ^ H2 ,
(T12) j=L H1 _ H2 !L H1 Y H2 ,
much more interesting, however, is the fact that the weak connectives can be
completely characterized by the strong ones together with negation:
(T13) j=L H1 ^ H2 $L H1 & (:H1 Y H2 ) ,
(T14) j=L H1 _ H2 $L H1 Y (:H1 & H2 ) .
The proofs in any case may proceed via calculating the corresponding truth
degree functions represented by the left hand and the right hand formulas in
these ws.
There are some few cases in which the distinction between the weak and
the strong versions of conjunction or disjunction does not matter. The most
interesting ones concern biimplication where one obviously has
(T15) j=L (H1 $L H2 ) $L ((H1 !L H2 ) ^ (H2 !L H1 ))
according to (9.6), but has also
(T16) j=L (H1 $L H2 ) $L ((H1 !L H2 ) & (H2 !L H1 )) .
Finally we look for some more implications which complete the previous
examples (T10) and (T7). Quite simple ones are e.g., because of (9.12), (9.11):
2 The reader, however, is asked to supply for himself these calculations to un-
derstand clearly the arithmetical background of the tautologies we are going to

mention further on.
185
(T17) j=L H !L H ,
(T18) j=L H ^ H !L H ,
the latter one giving via (T11) together with (T10), (T7) and (9.15) also the
weaker result
(T180 ) j=L H & H !L H .
The law of contraposition holds true in the common form
(T19) j=L (H1 !L H2 ) !L (:H2 !L :H1 )
in all the Lukasiewicz systems, as does the introduction of conjunction in
the succedent for strong conjunction:
(T20) j=L H1 !L (H2 !L H1 & H2 ) .
In the same way as (T180 ) was derived from (T18), also now the weaker version of conjunction introduction in the succedent can be derived for the weak
conjunction. And nally we mention the possibility of introducing conjunctively or disjunctively additional ws in the antecedent and the succedent of
an implication, which hold true like in classical logic in the forms
(T21) j=L (H1 !L H2 ) !L (H1 > G !L H2 > G) ,
(T22) j=L (H1 !L H2 ) & (G1 !L G2 ) !L (H1 > G1 !L H2 > G2 ) ,
with > any one of the connectives ^; _; &; Y.
9.1.2 Characterizing the number of truth degrees

In classical propositional logic one often considers the tautologies :(A^ :A)
and A_ :A as formalizations of the principle of excluded contradiction and
the principle of excluded middle, respectively.3 These readings, however, are
crucially tied to the situation that these ws are taken as ws of classical
logic, because one has in all the Lukasiewicz systems on the one hand
6j=L :(H ^ :H ) and 6j=L (H _ :H ) ;
but on the other hand one also has in all these systems
(T23) j=L :(H & :H ) and j=L (H Y :H ) .
And there do not exist any syntactical reasons why the w H Y :H of the
Lukasiewicz systems should not be a formalization of the principle of the
excluded middle if the formula A_ :A of classical logic is such a formalization.
Both these principles should therefore be taken metatheoretically as we
did in Section 1.1.
3 As a consequence, hence both these formalizations together, i.e. their conjunc-
tion, give a formalization of the principle of bivalence.
186
Nevertheless, the classical tautology A_ :A contains information concerning the number of truth values, and states that this number is at most
two according to the meaning of the connectives in classical logic. Therefore
we may ask whether a suitable w, or better: the fact that some suitable w
is a tautology of one of the Lukasiewicz systems, gives information on the
number of truth degrees of this system.
For ws which are able to give such information we essentially have to
rely on the strong connectives &; Y and their nite iterations. Therefore we
introduce for any integer n 1 and any sequence H1 ; H2 ; : : : of ws of LL
the following notations:
1
Y
i=1
1
X
i=1
nY
+1
Hi =def H1 ;
i=1
nX
+1
Hi =def H1 ;
i=1
Hi =def
Hi =def
n
;Y
i=1
X
; n
i=1
Hi & Hn+1 ;
(9.20)
Y Hn+1 :
(9.21)
Hi
It is an easy exercise to prove by induction on n that the truth degrees of

such ws are given by
ValL
ValL
n
;Y
i=1
X
; n
i=1
n
X
i=1
n
X
Hi ; = max 0;
Hi ; = min 1;
P
i=1
ValL (Hi ; ) ; (n ; 1) ;
o
ValL (Hi ; ) ;
(9.22)
(9.23)
with the sum-symbol on the right hand side understood in its original
arithmetical sense. These denitions (9.20) and (9.21) together with the characterizations of the truth degree functions et2 ; vel2 yield that
;Q one has in the
system Lm , in the case that ValL (Hn+1 ; ) 6= 1 and ValL ni=1 Hi ; 6= 0
hold true, that also
ValL
+1
;nY
i=1
Hi ; ValL
n
;Y
i=1
Hi ; ; m;1
holds;Ptrue and that,

again in Lm , in the case that ValL (Hn+1 ; ) 6= 0 and

n
L
Val i=1 Hi ; 6= 1 hold true, also
ValL
+1
X
;n
i=1
Hi ; ValL
n
;X
i=1
Hi ; + m;1
holds true, with k as dened in (5.7), and therefore with m;1 = m1;1
particularly.
Q
If one chooses H1 H2 : : : H then the nite iteration ni=1 H of the
&-conjunction of H with itself does not change the truth degree in the case
that one has ValL (H; ) = 1. In the opposite case ValL (Q
H; ) 6= 1, however,
;1
one reaches after at most m ; 2 iteration steps a w m
i=1 H with truth
187
degree = 0 under . Similarly the truth degree ValL(H; ) = 0 remains stable

under any nite Y-iteration. From a truth degree ValL (P
H; ) 6= 0, however,
;1
one reaches after at most m ; 2 iteration steps a w m
i=1 H with truth
degree = 1 under . At the same time the number m ; 2 of iteration steps
is in both cases the minimal number which guarantees that one reaches the
truth degrees 0 or 1, respectively, ;because
in the cases
ValL (H; ) = 2 or
;Pk+1
Qk+1
L
L
L
Val (G; ) = m;1 one has 0 < Val i=1 H; ; Val i=1 G; < 1 for all
k < m ; 2.
For simplicity of notation we also introduce a notation for the nite iterations of the weak connectives and put
1
^
i=1
1
_
i=1
Hi =def H1 ;
Hi =def H1 ;
n^
+1
i=1
n_
+1
i=1
Then one has immediately

ValL
ValL
n
;^
n
;_
Hi =def
Hi =def
n
;^
i=1
n
;_
i=1
(9.24)
(9.25)
Hi ^ Hn+1 ;
Hi _ Hn+1 :
Hi ; = 1min
ValL (Hi ; ) ;

i

n
i=1
Hi ; = 1max
ValL (Hi ; ) :

i

n
i=1
Theorem 9.1.1. For any integers k; m with m 2 and any propositional
variables p; p1 ; p2 ; : : : ; pk , amongst which p1 ; p2 ; : : : ; pk shall be pairwise different, there hold true:
;
P ;1
(a) m k i j=Lm W:p _ W m
p ,
;1 mi=1 (p $ p ) ,
(b) m < k i j=Lm ;m
j =i+1 i L j
Pm;1 i=1 Q
m;2 p is L -satisable .
(c) k m i
:
p
^
m
j =1
i=1
;1 p and any valuation. First

Proof: (a) Let G1 be the w :p _ Pki=1
L
assume m k. In the case Val (p; ) = 0 one has ValL (:p; ) = 1 and
hence ValL (G1 ; ) = 1. And in the opposite case ValL(p; ) 6= 0 one has
ValL (p; ) m;1 = m1;1 and hence, according to (9.23), immediately
kX
;1
ValL(
i=1
p; ) (k ; 1) ValL (p; ) (m ; 1) m;1 = 1 :
Thus G1 is an Lm -tautology in the case m k. Now suppose k < m. In this

case we have to prove that G1 is not an Lm -tautology. To reach this goal let
be a valuation such that ValL(p; ) = m1;1 . Because of k ; 1 < m ; 1 one
then has
ValL(G1 ; ) maxf(k ; 1) ValL(p; ); 2 g 2 < 1 ;
188
and has establishedWthat W

G is not an Lm -tautology.
;1 1m
(b) For the w m
i=1 j =i+1 (pi $L pj ), for having truth degree 1 under a
valuation it is necessary and sucient that at least two of the propositional
variables p1 ; p2 ; : : : ; pk get the same truthWdegree
under . Therefore k > m
;1 Wm (p $ p ) is an L is necessary and sucient that the w m
m
i=1 j =i+1 i L j
tautology.
;

P ;1
Qm;2
(c) Let G2 be the w m
j =1 :p ^ i=1 p . First let k m. Then choose
;2
a valuation such that ValL(p; ) = 2 = m
m;1 . From (9.22) one then gets
ValL
;2
;kY
i=1
p; (k ; 2) 2 ; (k ; 2) + 1
= 1 ; (k ; 2) m;2 1 ; 2 = m;2
and has therefore

;
ValL :p ^
kY
;2
i=1
p; m;2 :
According to (9.23) one hence has ValL(G2 ; ) = 1, i.e. G2 is Lm -satisable.

Now otherwise let m < k, i.e. let m ; 2 < k ; 2. If one then has ValL(p; ) = 1
for some valuation , then one has ValL (:p; ) = 0 and thus ValL (G2 ; ) = 0.
But if one has ValL (p; ) 6= 1, then one has ValL (p; ) 2 and therefore
ValL
;2
;kY
i=1
p; maxf0; (k ; 2) 2 ; (k ; 2) + 1g
= maxf0; 1 ; mk;;22 g = 0 ;
and thus ValL (G2 ; ) = 0, i.e. G2 is not Lm -satisable in this case.
2
The present proof shows even that the w G2 , considered in part (c) of
this theorem, always has truth degree 0 if one has m < k. Therefore one
immediately has
k < m i j=Lm :
kX
;1;
kY
;2
j =1
i=1
:p ^
p :
(9.26)
Having in mind that one can extend the deMorgan laws (T8), (T8a) for
the strong connectives &; Y to their nite iterations and thus having
;Q

P
(T24) j=L : ;Pni=1 Hi $L Qni=1 (:Hi ) ,
(T24a) j=L : ni=1 Hi $L ni=1 (:Hi ) ,
one can see immediately that the w considered in (9.26) is only an &-iterated
variant of the w considered in part (a) of the last theorem.
The problem of the characterizability of the number of truth degrees of
some particular Lukasiewicz system, which got a partial solution in Theorem 9.1.1, may also be considered from another point of view: as a problem
189
to ask for relationships between the sets tautL of tautologies of dierent

L-systems. Using the slightly simplied notation tautL for the set of all logically valid sentences of the system L , and therefore tautL2 for the set of all
tautologies of classical two-valued propositional logic, one has the following
interesting results.
Theorem 9.1.2. For all m; n 2 N with m; n 2 there hold true:
(a) tautLm tautLn , Wm Wn ,
(b) tautLm tautLn , n ; 1 divides m ; 1 ,
(c) tautLm 6 tautLm+1 and tautLm+2 6 tautLm+1 ,
(d) tautL1 = T
tautL0 ,
L
L
(e) taut1 = 1
m=3 tautm .
Proof: (a) Let Wm Wn. Then each valuation : V0 ! Wn into Wn
is also a valuation : V0 ! Wm into Wm . Because all the truth degree
functions of the L-connectives are dened without reference to the particular
number of truth degrees, one hence has for each w H and each valuation
: V0 ! Wn :
ValLn (H; ) = ValLm (H; ) :
Thus one has for each H 2 tautLm and each valuation : V0 ! Wn immediately ValLn (H; ) = 1. But this means H 2 tautLn . To prove also the converse
implication we assume Wn 6 Wm and show that in this case there exists an
Lm -tautology H which is not an Ln -tautology. We proceed in such a way that
we build up some L-formula H only using the propositional variable p and
such that H has truth degree 1 just for those valuations with (p) = nn;;12 .
As an auxiliary w consider the w
nX
;1
Gn
i=1
(:p) ^ p Y
;
For getting ValL (G

ValL
;1
;nX
i=1
;1
;nX
i=1
i=1
p :
n ; ) = 1 for some valuation one obviously needs to have

nY
;2

;
(:p); = ValL p Y
According to (9.23) one has

ValL
nY
;2
i=1
p; = 1 :
(:p); = minf1; (n ; 1) (1 ; (p))g
and therefore
nX
;1
;
ValL
i=1
(:p); = 1 , (p) nn;;12 :
On the other hand one has according to (9.22) and (5.76)
190
ValL p Y
;
nY
;2
i=1
p;
= minf1; (p) + maxf0; (n ; 2) (p) ; (n ; 2) + 1gg

= minf1; maxf (p); (p) (n ; 1) ; (n ; 2) + 1gg
and therefore
ValL p Y
;
nY
;2
i=1
p; = 1 , (p) nn;;12 :
Combining both conditions gives

ValL(Gn ; ) = 1 , (p) = nn;;12
independent of the range of , i.e. independent of the L-system one is working
with.
P ;1
Now we choose H as the w m
i=1 :Gn and have immediately
ValL(H; ) = 1 , (p) 6= nn;;21 :
However, it is nn;;12 2= Wm , because otherwise one would have also nn;;13 =
et2 ( nn;;21 ; nn;;21 ) 2 Wm and thus nally Wn Wm . Therefore this w H is an
Lm -tautology. Yet, obviously, H is not an Ln -tautology.
(b) Suppose tautLm tautLn . Then one has Wn Wm according to (a) and
particularly n;1 1 2 Wm , and therefore n;1 1 = mm;;k1 for some 1 k m. Thus
m ; 1 = (n ; 1) (m ; k) holds true and n ; 1 divides m ; 1. Let conversely
n ; 1 divide m ; 1. Then there exists some l 1 such that m ; 1 = l (n ; 1),
i.e. such that n;1 1 = m;l 1 and hence l m ; 1. For k = m ; l one thus has
1 k m;1 and also n;1 1 = mm;;k1 2 Wm . This immediately gives Wn Wm
because all the nonzero truth degrees of Wn can be generated from the truth
degree n;1 1 via iterated applications of the truth degree function vel2 . And,
of course, vel2 maps truth degrees of WmPtomtruth
degrees of Wm .
(c) Let us rst consider the w :p _ i=1;1 p. Because of m m this is
an Lm -tautology according to Theorem 9.4 (a). Because of m + 1 6 m it is,
however, not an Lm+1 -tautology according to the same result. Hence one has
tautLm 6 tautLm+1 . So we have to nd a w which is an Lm+2 -tautology but
not an Lm+1 -tautology. For m = 2 consider
G2
3
X
i=1
(p & p) _
3
X
i=1
(:p & :p) :
For each valuation : V0 ! W3 with (p) = 12 one has also (:p) = 21 and
therefore
ValL(p & p; ) = ValL (:p & :p; ) = 0 :
Thus one has ValL(G2 ; ) = 0 which means that G2 is not an L3 -tautology.
But for any valuation : V0 ! W4 one has in the case that (p) 23 holds

;P
191
true surely ValL(p & p; ) 13 , and hence ValL 3i=1 (p & p); = 1. And one
has in the case that
(p) 31 holds true surely ValL (:p & :p; ) 31 , and
;P3
L
hence again Val i=1 (:p & :p); = 1. In any case thus ValL (G2 ; ) = 1,
which means that G2 is an L4 -tautology.
For each m 3 we consider for some propositional variable p the w
Gm
mX
+1; Y
m
i=1 j =1
p _
mX
;1
i=1
:p :
For each valuation : V0 ! Wm+1 with (p) = mm;1 = 1 ; m1 one has

according to (9.23)
ValL
;1
X
;m
i=1
:p; (m ; 1) m1 = 1 ; m1 < 1
and furthermore
ValL
m
;Y
i=1
p; = maxf0; m mm;1 ; m + 1g = 0 ;
hence altogether ValL (Gm ; ) < 1, and Gm is not an Lm+1 -tautology. But for
any valuation : V0 ! Wm+2 one has in the case that (p) = 1 holds true
immediately ValL (Gm ; ) = 1. In the case that (p) = mm+1 holds true one
has according to (9.22)
m
;Y

ValL p; = maxf0; m m
m+1 ; m + 1g = m+1 ;
i=1
and therefore according to (9.23)

ValL
+1 ; Y
m
X
;m
i=1 i=1
p ; (m + 1) m1+1 = 1 :
Hence also in this case one gets ValL (Gm ; ) = 1. So nally let (p) < mm+1 .
Then one has ValL (:p; ) m2+1 , and thus
ValL
;1
X
;m
i=1
:p; (m ; 1) m2+1 1
because of m 3. Hence also in this case one has ValL (Gm ; ) = 1. In

summary, thus, one has that Gm is an Lm+2 -tautology.
(d) Because of W0 W1 one gets tautL1 tautL0 as in the proof for part
(a). So it remains to prove that one also has tautL0 tautL1 , i.e. that each
L0 -tautology is also an L1 -tautology, respectively, that each w which is not
an L1 -tautology also is not an L0 -tautology. So suppose that the w H is
not an L1 -tautology and that
192
Val (H; ) < 1 for some : V0 ! W1 :

Assume that in H at most the propositional variables p1 ; : : : ; pn occur, and
that H is built up only with the connectives : and !L. The truth degree
functions non1 and seq 2 are continuous functions over the real unit interval
[0; 1]. Therefore also the truth degree function w~H (x1 ; : : : ; xn ) described by
H is a continuous function, and this means that the truth degree ValL (H; )
continuously depends on the values (p1 ); : : : ; (pn ). And according to Theorem 2.2.1 it depends only on these values. Therefore there exists in the
n-dimensional unit hypercube [0; 1]n = W1n a whole open neighborhood U0
of the point ((p1 ); : : : ; (pn )), such that for each point (t1 ; : : : ; tn ) 2 U0 and
each valuation : V0 ! W1 with (pi ) = ti for all i = 1; : : : ; n one also has
ValL (H; ) < 1. It is, however, also U0 \ W0n 6= ;. So we can additionally assume (t1 ; : : : ; tn ) 2 U0 \ W0n and even choose a valuation 0 : V0 ! W0 with
0 (pi ) = ti for all i = 1; : : : ; n (and e.g. 0 (p) = 0 for all other propositional
variables). Then we have again ValL (H; 0 ) < 1 and H is not an L0 -tautology.
(e) As in the proof for part (a) one has because of Wm W1 for each
m 2 also tautL1 tautLm . Hence it is
1
\
tautL1
tautLm :
m=2
What remains to be proved is that each w H of LL which is not an L1 tautology is already not an Lm -tautology for some suitable integer m 2.
Hence suppose H 2= tautL1 = tautL0 . Then consider a valuation : V0 ! W0
such that Val (H; ) < 1. In the w H at most the propositional variables
p1 ; : : : ; pn shall occur. Suppose that the rational numbers (p1 ); : : : ; (pn )
are written down as common fractions and that k is their least common
denominator. Then one has (pi ) 2 Wk+1 for each i = 1; : : : ; n. Consider
a new valuation 0 : V0 ! Wk+1 determined by 0 (pi ) = (pi ) for all
i = 1; : : : ; n and by 0 (p) = k1 for all other propositional variables. Then
one obviously has ValL (H; 0 ) = ValL(H; ) < 1. Hence H is not an Lk+1 tautology.
2
Corollary 9.1.1. For all integers m; n > 2 one has

(a) tautL1 tautLm tautL2 ,
(b) m < n ) tautLm 6 tautLn .
Proof: (a) One has tautL1 tautLm because of Theorem 9.1.2(e). In the
case of tautL1 = tautLm one, however, would have tautLm tautLm+1 contrary
to Theorem 9.1.2(c). Therefore tautL1 tautLm . Also the inclusion relation
tautLm tautL2 is an immediate consequence of Theorem 9.1.2(a), and equality
does not hold true here because e.g. the w p _ :p is an L2 -tautology, but it
is not an Lm -tautology for any m > 2.
193
(b) Suppose m < n. If one then would have tautLm tautLn , so one would
also have Wn Wm according to Theorem 9.1.2(a), and this contradicts
m < n.
2
Theorem 9.1.2(e) also indicates that the system L1 cannot be characterized by a condition stating that some particular w of LL has to be a
tautology. Furthermore, this system L1 also cannot be characterized by a
condition stating that some particular w of LL has to be satisable, because
each L1 -satisable w H is already Lm -satisable for some suitable m. For
assume that the w H is L1 -satisable and that : V ! W0 is chosen
such that ValL (H; ) = 1 holds true, then H is built up from nitely many
propositional variables p1 ; : : : ; pn and all the nitely many rational values
(p1 ); : : : ; (pn ) belong already to some suitable truth degree set Wm . And
then, obviously, H is already Lm -satisable.
9.1.3 Axiomatizability
Now we are going to consider axiomatizations of the Lukasiewicz systems.
For the nitely many-valued ones one has available the Gentzen-type, sequent style axiomatization method of Schro ter, cf. Chapter 6.4, which does
not only provide an adequate axiomatization of the (set of) logically valid
formulas of these systems, but also provides an adequate formalization of the
consequence relation of these systems. It is interesting to notice, however, that
for the Lukasiewicz systems also Hilbert-type calculi are available because
of the fact that the axiomatization method of Rosser-Turquette remains
applicable also for these systems, despite the fact that the Lukasiewicz implication !L does not satisfy the standard condition for an implication in any
one of the systems L with 6= 2, i.e. does not satisfy the condition (RT1).
Theorem 9.1.3. The axiom schemata (AxRT 1); : : : ; (AxRT8) constitute together with the rule of detachment (MP) an adequate axiomatization for each
one of the nitely many-valued systems Lm .
Proof: Let K (Lm ) be the logical calculus described in this theorem. We
have to prove soundness and completeness for this calculus. According to
Corollary 6.2.3, however, it suces for this to prove that in each one of the
systems Lm all the connectives Jt can be dened for any t 2 Wm , and that
the conditions (E1), (E2) from p. 119 are satised.
For (E1) we have to prove that for any ws H; G of LS and for each
valuation : V0 ! Wm with ValL (H; ) = 1 and with ValL (G; ) < 1 also
ValL (H !L G; ) < 1 holds true. But this is an immediate consequence of
the denition (5.81) of the truth degree function seq 2 of the Lukasiewicz
implication !L .
To establish condition (E2) we have to show that each axiom of the logical
calculus K (Lm ) is an Lm -tautology. We shall do this e.g. for axiom schema
(AxRT 5), leaving the other cases to the reader. Consider the w
194
m;1
H ! (Jm;i (A) !L B; B )
=0
with ! here taken as the nite !L -iteration in the successive antecedents,

and let : V0 ! Wm be any valuation. It has to be shown that ValL (H; ) = 1
holds true. From the denitions (6.1), (6.2) of the nite !L-iteration in the
antecedent together with the importation/exportation law (T7) and denition (9.20) one gets inductively
j=L H $L
;1
m
Y
i=0
(Jm;i (A) !L B ) !L B ;
and hence the equation

;
ValL(H; ) = seq 2 ValL
;1
Y
;m
i=0
(Jm;i (A) !L B ); ; ValL (B; ) :
But because one has seq 2 (0; t) = 1 for each t 2 Wm , and because conjuncts
with truth degree 1 do not in uence the truth degree neither of the strong
nor of the weak conjunction, one has for s = ValL (A; ):
ValL
;1
Y
;m
i=0
(Jm;i (A) !L B ); = ValL (Js (A) !L B; )

= seq 2 (1; ValL (B; ))
= ValL (B; ) ;
and therefore ValL (H; ) = 1.

By the same type of elementary calculations also for all the other axioms
of K (Lm ) one can establish that they are Lm -tautologies.
So it remains to dene all the connectives Jt for each t 2 Wm . According
to (9.22) one can obviously dene
J1 (H ) =def
mY
;1
i=1
H:
(9.27)
For each truth degree t with 21 t < 1 denote by n(t) the largest integer k
such that k (1 ; t) < 1. Then one has 1 ; n(t) (1 ; t) 1 ; t because of
the maximality of n(t). Then let
Jt (H ) =def J1
as well as
nY
(t)
i=1
H $L H
Jt (H ) =def Jn(t)(1;t)
n(t)
; Y
i=1
for n(t) = 1;t t and t 12
(9.28)
for n(t) > 1;t t and t 12 ;
(9.29)
195
and nally
(9.30)
Jt (H ) =def J1;t (:H ) for t < 12 :
The last case (9.30) is a simple reduction to the previous ones, thus only the
suitability of (9.28) and (9.29) needs to be established. The intention is that
the connectives Jt shall be dened successively for t = 1 ; 2 ; : : : ; m with the
case t = 1 determined by (9.27), i.e. we proceed from t = 1 to smaller and
smaller values of t, as long as t 12 is the case.
Case (9.28) applies i 1 ; n(t) (1 ; t) = 1 ; t holds true, and case (9.29)
applies i 1 ; n(t) (1 ; t) > 1 ; t holds true. Now consider some tQ2 Wm
and let be ValLQ
(H; ) = t. Then build up the \longest"Q&-iteration ki=1 H
(t) H by denition
such that ValL ( ki=1 H;
) 6= 0 holds true. This is just ni=1
Qn(t)
of n(t). Write Gt for i=1 H . Then one has ValL(Gt ; ) 1 ; t. In the case
that ValL (Gt ; ) < 1 ; t holds true, one obviously has ValL(:Gt ; ) > t, which
means that the J-connective which characterizes the truth degree of :Gt is already dened and (9.29) is acceptable. In the other case of ValL (Gt ; ) = 1 ; t,
however, :Gt and H have the same truth degree, and this gives (9.28),
having in mind thatQ for each w A of LS with ValL (A; ) 6= t one has
(t) A; ).
ValL (A; ) 6= ValL (: ni=1
2
It is one of the advantages of this axiomatization method of RosserTurquette that it applies to a large class of nitely many-valued systems. Its main disadvantage is, on the other hand, that the logical calculus
which results from the schemata (AxRT 1); : : : ; (AxRT 8) is quite cumbersome,
particularly if one is interested to prove metatheoretical results concerning
this calculus. Thus it is not surprising that there exist for at least some of
the Lukasiewicz systems also much simpler Hilbert-type axiomatizations
{ simpler in the sense of having fewer (and possibly even simpler) axiom
schemata, not in the sense of making particular derivations easier. A short
and very nice such axiomatization of L3 was, e.g., provided by Wajsberg
[560].
Theorem 9.1.4. An adequate, i.e. sound and complete, axiomatization of
the Lukasiewicz system L3 is given by the rule of detachment w.r.t. the
Lukasiewicz implication !L together with the following axiom schemata:
(L3 1) H1 !L (H2 !L H1 ) ,
(L3 2) (H1 !L H2 ) !L ((H2 !L H3 ) !L (H1 !L H3 )) ,
(L3 3) (:H2 !L :H1 ) !L (H1 !L H2 ) ,
(L3 4) ((H1 !L :H1 ) !L H1 ) !L H1 .
It is an easy exercise to show that all the axioms of this system are L3 tautologies, i.e. that this axiomatization is sound. For the completeness proof
the reader is either referred to Wajsberg's original papers [560, 562] or the
nice proof of the completeness w.r.t. the entailment relation in [203].
196
Besides this Lukasiewicz system L3 with the \smallest" (nontrivial) set

of truth degrees also the corresponding system L1 with the \largest" set of
truth degrees has such a nicely simple axiomatization.
Theorem 9.1.5. A sound and complete axiomatization of the system L1
is given by the rule of detachment w.r.t. the Lukasiewicz implication !L
together with the following axiom schemata:
(L1 1) H1 !L (H2 !L H1 ) ,
(L1 2) (H1 !L H2 ) !L ((H2 !L H3 ) !L (H1 !L H3 )) ,
(L1 3) (:H2 !L :H1 ) !L (H1 !L H2 ) ,
(L1 4) ((H1 !L H2 ) !L H2 ) !L ((H2 !L H1 ) !L H1 ) .
This axiomatization4 was found by Lukasiewicz himself in the 1920s and
its adequacy mentioned without proof in [339]. Again it is an easy exercise
to show that all the axioms of this system are L1 -tautologies, i.e. that this
axiomatization is sound. A completeness proof was announced by Wajsberg
[561], but never published. It was only in the 1950s that Rose/Rosser [457]
and Chang [92] gave completeness proofs.
By the way, Lukasiewicz had added another axiom to his system which
was proved dispensable in [361] and in [90].
The completeness proof of [92] uses some ws inferred by Rose/Rosser,
and introduces an important class of algebraic structures called MV-algebras5
which shall be discussed later on in Section 9.2, where also this completeness
proof is presented.6 There we shall nd also nice adequate axiomatizations
for the other nitely many-valued systems Lm .
The notion of completeness discussed up to now is a semantical one. It is
also possible, however, following an early approach of Post [426], to consider
a kind of syntactical completeness, usually referred to as Post-completeness.
This Post-completeness is, like the semantical completeness, a property of
logical calculi.
Suppose that K Ax is a logical calculus with axiom set Ax for a system S
of (many-valued) propositional logic. Then this logical calculus K Ax is called
Post-complete i for any extension Ax of the set Ax of axioms by a w
H { or by the set of all substitution instances of H , in the case that Ax is
formulated with axiom schemata and without substitution rule { such that
the w H is not K Ax -derivable the extended calculus K Ax allows to derive
all ws of the language LS .
4 Indeed the axioms of Lukasiewicz for L1 as well as those ones of Wajsberg
for L3 were given a little bit dierently: not as schemata, as done here, but as
single axioms with propositional variables in place of the ws H1 ; H2 ; H3 above,
and with the substitution rule as an additional inference rule. But this dierence
does not matter at all for our considerations.
5 Having \MV" as shorthand for \many-valued", but pronouncing this only \emvee".
6 Also in [246] this completeness is proved along the lines of the theory of MValgebras and related structures.
197
For the Lukasiewicz propositional systems and their just-mentioned axiomatizations one has also some kind of Post-completeness, as the following
theorem shows which was proved in [455]. The proof is not dicult, the interested reader should consult the original paper.
Theorem 9.1.6. Suppose that H is any w of LL and that Ax is one of the
axiom systems mentioned in the Theorems 9.1.3, 9.1.4, 9.1.5, and 9.2.13.
Then all ws of LL become derivable { via the rule of detachment (MP) as
the only rule of inference { from the extension of Ax by all the substitution
instances of H i H is not already derivable from Ax and is not an L2 tautology.
Another aspect of completeness is whether the standard extension j=RT
of the derivability relation j=RT of the Rosser-Turquette calculi K (L )
for the Lukasiewicz systems L provide an adequate axiomatization of the
notion of entailment for these systems. According to Chapter 6.4 we attack
this problem by asking whether the properties (FINj= ), (FIN` ), (DEDj= ),
and (DED` ) hold true for the Lukasiewicz systems.
According to Theorem 3.2.5 the semantical niteness property (FINj= )
holds true for each of the nite many-valued systems Lm , m 2. It, however,
fails for the system L1 . To see this, consider the set of ws
0 =
n
nX
i=1
p(n) !L p0 j n 2 [
n
nX
i=1
p(n) j n 2 :
From (9.23) one gets immediately for each n 2

n
X

ValL
p(n) ; = 1 ) (p(n) ) n +1 1 ;
i=1
and also
ValL
n
X
i=1

n :
p(n) !L p0 ; = 1 ) (p0 ) n +
1
From these remarks it easily follows that one has 0 j=L1 p0 , but 0 6j=L1 p0
for each nite set 0 0 .
The syntactic niteness property (FIN` ), on the other hand, obviously
holds true for all the Hilbert-type logical calculi we have considered up to
now because all of them have a nitary notion of derivability, i.e. because
all of them have solely inference rules (indeed only the rule of detachment)
which refer to only nitely many premises.
The semantic deduction property (DEDj= ) again does not hold for any
one of the \truly many-valued" systems L with 6= 2. To see this, consider
the set of w
1 = fJ1 (p0 ) !L J1 (p00 )g
for which one has
198
1 [ fp0 g j=L p00 ; but 1 6j=L (p0 !L p00 ) :

By the choice of 1 it is obvious that 1 [ fp0 g entails p00 . Any valuation ,
however, which satises 0 < (p0 ) < 1 and (p00 ) = 0, is a model of 1 which
also satises ValL (p0 !L p00 ; ) < 1, and hence is not a model of p0 !L p00 .
So it remains to discuss whether our logical calculi for the Lukasiewicz

systems satisfy the syntactic deduction property (DED` ). According to the
remarks immediately following Theorem 6.4.2 we know that (DED` ) holds
true for such a calculus i all ws of the form
(A !L B ) !L ((A !L (B !L C )) !L (A !L C ))
(9.31)
are theorems of, i.e. deducible within, this calculus. Using the adequacy of all
these calculi this means that (DED` ) holds true for such a calculus i all ws
of the form (9.31) are L -tautologies. It is, however, an easy exercise to show
that this is the case only for the system L2 : for the systems Lm with m > 2
one e.g. may choose ws A; B; C and a valuation such that ValL(C; ) = 0
and ValL (A; ) = ValL (B; ) = m1;1 hold true, to give the w (9.31) a truth
degree < 1; and for the system L1 one may proceed similarly.7
In summary, thus, we do not get sucient information from these four
properties to infer the adequacy of the standard extension of the derivability
relations of our calculi for the axiomatization of the entailment relations of
the Lukasiewicz systems. These properties have, however, also their own
specic interest. Therefore we discussed them here.
There are even weakened versions of the deduction properties (DEDj= )
and (DED` ) available, cf. e.g. [232, 423]. The semantic version which holds
true in each one of the Lukasiewicz systems, cf. [423], is the fact that for
each of the nitely many-valued systems Lm
m;1
[ fH1g j=Lm H2 , j=Lm ! (H1 ; H2 )

=1
holds true for all sets of ws and any ws H1 ; H2 2 LL , and that for L1
it holds true similarly
i n
[ fH1g j=L1 H2 , j=L1 ! (H1 ; H2 ) for some n 1 :

=1
A further important result in classical propositional logic is the Interpolation Theorem of Craig which states that for any classical tautology A ! B
there exists a w C , which contains only such propositional variables which
occur in both of A and B , and for which A ! C as well as C ! B are
classical tautologies.
For the Lukasiewicz systems a corresponding interpolation result fails,
as was shown in [318]. To prove this consider the w
(r ^ (r !L p)) _ p !L q _ (q !L p) :
(9.32)
7 However, as shown in [232], the w (9.31) always assumes a truth degree 0:5.
199
It is an L1 -tautology and hence a tautology in each of the Lukasiewicz

systems. To see this consider some valuation : V0 ! W1 . In the case that
(q) (p) holds true one has ValL (q !L p; ) = 1, and hence formula (9.32)
has truth degree 1 too. In the opposite case (q) > (p) one needs
ValL(r ^ (r !L p); ) ValL(q _ (q !L p); )
(9.33)
to give formula (9.32) the truth degree 1. But in the case that (9.33) did not
hold one would have (r) > (q) together with the inequality
ValL(r !L p; ) > ValL (q !L p; ) ;
and hence (r) > (q) together with (r) < (q), a contradiction.
Each \interpolating" w C 2 LL for the implication (9.33) would have
to be a w which is built up solely from the propositional variable p. However, choosing in this case a valuation : V0 ! W1 with (p) = 0 causes
ValL (C; ) 2 f0; 1g. In this situation one has furthermore that the truth degree of the antecedent of formula (9.33) is the same as the truth degree of
r ^ :r, and the truth degree of the succedent of formula (9.33) is the same
as the truth degree of q _ :q. Choosing then additionally 0 < (q); (r) < 1
yields immediately that one either has ValL ((r ^ (r !L p)) _ p !L C; ) 6= 1 or
ValL (C !L q _ (q !L p); ) 6= 1, which in any case means that there cannot
exist any interpolant for formula (9.33).
9.1.4 Decidability of the system L1

According to the discussions in Chapter 8.2 all the nitely many-valued systems Lm , m 2, are decidable. As in the case of the axiomatizability problem,
which was not solved in general by the results of Chapter 6 and the adaptation
of the Rosser-Turquette method of axiomatization to the Lukasiewicz
systems, the system L1 has to be considered separately. This we shall do
here and prove that also this system is decidable.
Theorem 9.1.7. The innitely many-valued Lukasiewicz system L1 is decidable.
Proof: We have to prove that there exists a method, an algorithm, which
produces for any given w H 2 LL after nitely many elementary steps
the answer whether H is an L1 -tautology, or whether H is not such an
L1 -tautology, i.e. whether ValL (H; ) = 1 holds true for any valuation :
V0 L! W1 , or whether there exists some such valuation : V0 ! W1 with
Val (H; ) < 1.
Let such a w H be given. Then H contains only nitely many propositional variables. Let them all appear among p1 ; : : : ; pn. Consider the truth
degree function w~H described by H which, according to (2.9), (2.8), can be
determined from H in nitely many elementary steps of the following kind:
(i) replacing each propositional variable pi of H by the real variable xi , and
200
(ii) replacing successively each connective of H by its corresponding truth

degree function, e.g. starting with the \innermost" connectives. And this
transformation of H into w~H is an algorithm of the kind we like to have.
Therefore it suces to prove that for each w H 2 LL , containing only the
propositional variables p1 ; : : : ; pn , and its corresponding truth degree function
w~H (x1 ; : : : ; xn ) there exists an algorithm to decide whether the sentence
;

8 x1 ; : : : ; xn 0 x1 1 ^ : : : ^ 0 xn 1 ) w~H (x1 ; : : : ; xn ) = 1
holds true within the structure of the real numbers. To put it a little bit
dierently: it suces to prove that the open (i.e. quantier free) formula
0 x1 1 ^ : : : ^ 0 xn 1 ) w~H (x1 ; : : : ; xn ) = 1
(9.34)
is valid in the structure hR; +; ; 0; 1; i, i.e. in the ordered eld of real numbers.
It is a well known result of Tarski [529] that the elementary theory of
the ordered eld hR; +; ; 0; 1; i of real numbers, i.e. the theory of this structure as formalized in the language of rst-order (classical) logic, is decidable,
which means that there exists an algorithm which answers for each statement
made in the language of this elementary theory8 the question whether this
statement is valid for the ordered eld of real numbers or whether it is not,
cf. also [33, 109].
For applying this result to our present problem it suces to show that
the statement (9.34) can equivalently be formulated in the language of the
elementary theory of the structure hR; +; ; 0; 1; i. Because each one of the
double inequalities 0 xi 1 in (9.34) obviously is the conjunction of two
\simple" inequalities 0 xi and xi 1 of this language, it only remains
to show that the truth degree functions non1 and seq 2 of the connectives
:; !L of L1 can be represented in this language of the elementary theory of
hR; +; ; 0; 1; i. But for this one only needs to show that in this language the
operations of taking the dierence and of taking the minimum are expressible,
i.e. that the properties of being the result of a subtraction or of being the
minimum of two numbers can equivalently be represented by formulas of this
language. For the case of the dierence one has
z = x ; y i x = z + y ;
and hence particularly
z = 1 ; y i y + z = 1
and thus the representability of the dierence. And for the minimum operation one has as a rst reduction step that
8 This rst-order language has besides its logical constants, including the equality
sign, and its individual variables in its alphabet individual constants denoting the
numbers 0 and 1, a binary relation symbol denoting the ordering , and function
symbols denoting the two binary operations +; of addition and multiplication.
201
z = minfx; yg i z minfx; yg ^ minfx; yg z
and furthermore
z minfx; yg i z x ^ z y ;
minfx; yg z i x z ^ y z ;
which together also give the representability of the minimum operation.
So all the necessary representations are possible, and the decidability of
the propositional system L1 results from the decidability of the elementary
theory of the ordered algebraic structure hR; +; ; 0; 1; i.
2
9.1.5 Representability of truth degree functions

As mentioned earlier, the fact that the Lukasiewicz connectives satisfy the
normal condition means that the systems L , 6= 2, are not functionally
complete. This situation oers two further problems: (i) to characterize the
truth degree functions which can be represented by ws of one of these systems L , and (ii) to determine additional connectives which when added to
these systems L transform them into functionally complete ones.
The problem of the characterization of the truth degree functions which
can be represented by ws of one of the systems L , 6= 2, can be reduced to
the corresponding problem for the particular system L1 . The reason for this
is that to each truth degree function g over W one has a w H 2 LL such
that g = wH , but H is also a w for the system L1 , and thus describes a truth
degree function g over W1 which obviously has the additional property that
g is the restriction of g to W . Furthermore this truth degree function g is a
continuous function in the real unit interval [0; 1] because it is a superposition
of the continuous functions non1 and seq 2 .
The characterization of the truth degree functions over [0; 1] = W1 which
are L1 -representable was given by McNaughton [357] and shall be described now.9
Lemma 9.1.1. Each n-ary truth degree function f : W1n ! W1 with a
characterizing equation of the form

f (x1 ; : : : ; xn ) = min 1; maxf0; b +
n
X
i=1
ai xi g
with integers b; a1; : : : ; an can be represented as truth degree function w~H of

a suitable w H 2 LL in which at most the propositional variables p1 ; : : : ; pn
occur.
9 This presentation follows the original proof from [357]. Meanwhile also other
proofs have been given, e.g. a very interesting one by Mundici in [389] which
uses methods which are of importance, e.g., for the relations between L1 and
computer science.
202
Proof: A w H 2 LL in which at most the propositional variables

p1 ; : : : ; pn occur and whose corresponding truth degree function according
to (2.9) has the representation

w~H (x1 ; : : : ; xn ) = min 1; maxf0; b +
n
X
i=1
ai x i g
(9.35)
shall be denoted as
Hab ;:::;an (p1 ; : : : ; pn ) :
Hab ;:::;an or as
(9.36)
The strategy for the following proof procedure shall be to reduce the construction of such a w of the form Hab ;:::;an to the construction of \simpler"
ws, which shall mean here to ws which are again of the type Hab ;:::;an , but
of some restricted form, and in particular shall have a smaller upper index.
This reduction is to be nished with formulas of the form Hc0 ;:::;cn which
then can be determined directly.
Two particularPcases can be treated immediately: if one has that formula
(9.35) yields b + ni=1 ai xi 1 for all x1 ; : : : ; xn 2 W1 , then let
Hab ;:::;an =def (p1 Y :p1 ) ;
P
and if one has that (9.35) yields b + ni=1 ai xi 0 for all x1 ; : : : ; xn 2 W1 ,
then let
Hab ;:::;an =def (p1 & :p1 ) :
Therefore we now
P can suppose that there exist values t1 ; : : : ; tn 2 W1 such
that 0 < b + ni=1 ai ti < 1 holds true. Suppose that c1 ; : : : ; cn is such a
rearrangement of the coecients a1 ; : : : ; an that for some suitable 0 k n
it holds true that ci 0 for i = 1; : : : ; k, and cj < 0 for j = k + 1; : : : ; n. Let
s1 ; : : : ; sn be the corresponding rearrangement of the values t1 ; : : : ; tn . Then
one has
1
0<b+
k
X
i=1
n
X
ci si +
j =k+1
and particularly
0<b+
as well as
b+
k
X
i=1
n
X
j =k+1
ci si b +
cj b +
Therefore one has
k
X
i=1
ci < b ;
cj sj ;
k
X
i=1
n
X
j =k+1
n
X
j =k+1
ci
cj sj < 1 :
cj =
n
X
j =k+1
jcj j :
(9.37)
203
And this condition (9.37) on the other

P hand suces that there exist values
t1 ; : :P: ; tn 2 W1 such that 0 < b + ni=1 ai ti < 1 holds true. In the
case that
Pk
n
b + i=1 ai ti P0 always holds true, one has particularly b + i=1 ci 0
and thusPb ; ki=1 ci , i.e. condition (9.37) is not satised. In thePcase that
1 b + ni=1Pai ti always holds true, one has in particular 1 b + nj=k+1 cj
and thus ; nj=k+1 cj < b, i.e. condition (9.37) again is not P
satised. But
because of the continuity of any function g(x1 ; : : : ; xn ) = b + ni=1 ai xi one
has either always g(x1 ; : : : ; xn ) 0 or always 1 g(x1 ; : : : ; xn ), if there do
not exist values t1 ; : : : ; tn 2 W1 with 0 < g(t1 ; : : : ; tn ) < 1.
By rearrangements and renumerations of the coecients a1 ; : : : ; an in
(9.35) it therefore is possible to reach a situation that one has to construct
only ws of the type
Hab ;:::;ak ;;ak ;:::;;an with ai 0 for all i = 1; : : : ; n :
Having additionally in mind that
1
b+
+1
k
X
i=1
ai xi ;
;
n
X
j =k+1
n
X
b;
aj xj =
j =k+1
aj xj +
k
X
i=1
ai xi +
n
X
j =k+1
aj (1 ; xj )
holds true, one can put r = nj=k+1 aj and gets

r (p1 ; : : : ; pk ; :pk+1 ; : : : ; :pn )
Hab ;:::;ak ;;ak ;:::;;an (p1 ; : : : ; pn ) = Hab;;:::;a
n
with again always ai 0 and with b ; r < 0 according to (9.37). Hence one
has only to consider ws of the type
Had ;:::;an (q1 ; : : : ; qn ) with d 0 and a1 ; : : : ; an 0
(9.38)
to get the necessary \simpler" ws of type (9.36).
In the case that one has d = 0 here, no further reduction steps are needed
and Ha0 ;:::;an can be chosen according to (9.45). In the other case that one
has d < 0, the further reduction uses the fact that for each k = 1; : : : ; n the
w
;

Had ;:::;an $L (Had+1;:::;ak ;1;:::;an !L :qk ) !L Had ;:::;ak ;1;:::;an (9.39)
is an L-tautology. And this fact may be established by comparing the truth
degrees for the right hand and the left hand side of this L-biimplication,
assuming that (9.35) holds true for all ws of the form H:::c involved here.
This method leads to some elementary calculations which use a few facts
which hold true for the operations of taking the maximum and taking the
minimum. Mainly these are the facts that one always has
u + maxfs; tg = maxfu + s; u + tg ;
(9.40)
; maxfs; tg = minf;s; ;tg ;
(9.41)
1
+1
204
as well as the relationships which one gets from these ones by exchanging
max and min, and the fact that
s t ) maxft; minfs; ugg = minfs; maxft; ugg
(9.42)
holds true. All these facts can be established easily.
As a rst step we consider the right hand side of the w in (9.39) and
determine the truth degree u of its subformula Had+1
;:::;ak ;1;:::;an !L :qk , using
as a shorthand notation

i 6= k
0
ai = aai ; 1 for
for
i=k .
k
According to (5.81) and (9.35) one gets from (9.40), (9.41)
X
u = minf1; 1 ; minf1; maxf0; d + 1 + a0i xi gg + 1 ; xk g
X
= minf1; maxf1 ; xk ; 2 ; xk ; maxf0; d + 1 + a0i xi ggg ;
and has from (9.42), again using (9.40) and (9.41) together with 1 ; xk 1:
X
u = maxf1 ; xk ; minf1; minf2 ; xk ; 2 ; xk ; d ; 1 ; a0i xi ggg
X
= maxf1 ; xk ; minf1; 1 + (1 ; xk ); 1 ; d ; ai xi gg
= maxf1 ; xk ; minf1; 1 ; d ; S gg
P
P
with always the (arithmetic) sum over i = 1; : : : ; n, and with S for ai xi .
Therefore the truth degree v of the whole right hand side of the w in (9.39)
becomes
X
v = minf1; 1 ; u + minf1; maxf0; d + a0i xi ggg
X
= minf1; minfxk ; maxf0; d + S gg + minf1; maxf0; d + a0i xi ggg
X
= minf1; xk + 1; xk + maxf0; d + a0i xi g; 1 + maxf0; d + S g ;
X
maxf0; d + S g + maxf0; a0i xi gg
= minf1; maxfxk ; d + S g; maxf0; d + S; 2(d + S ) ; xk gg
= minf1; maxf0; xk ; d + S g; maxf0; d + S; 2(d + S ) ; xk gg :
Here one applies (9.40), (9.41) repeatedly and deletes for the minimum (resp.
the maximum) all the terms which are not smaller (resp. not greater) than
other ones which also have to be considered. We now distinguish the cases
that xk d + S holds true, and that xk > d + S holds true. In the case
xk d + S also d + S 2(d + S ) ; xk holds true and thus
v = minf1; maxf0; d + S g; maxf0; 2(d + S ) ; xk gg
= minf1; maxf0; d + S gg :
In the case xk > d + S also d + S > 2(d + S ) ; xk holds true and hence
1
205
v = minf1; maxf0; xk g; maxf0; d + S gg

= minf1; maxf0; d + S gg :
P
Thus in any case one has v = minf1; maxfd + ai xi gg, which means that
the truth degree of the right hand side of the w from (9.39) coincides with
the truth degree of the left hand side of this w. Hence (9.39) holds true.
Because of (9.39) now the w Had ;:::;an of (9.38) can in a semantically
equivalent manner be represented by some w which is built up from subformulas H:::c with jcj < jdj and a subformula H:::d with unchanged upper index
d, but with some smaller lower indices. This subformula again can be subject to the reduction step determined by (9.39), because one has Theorem
2.2.2 available for the system L1 . Iterating this application of (9.39) as a
reduction procedure for suitable k = 1; : : : ; n nally leads to some w, semantically equivalent to Had ;:::;an , which is built up from (possibly negated)
propositional variables, ws of the type H:::c with jcj < jdj, and a subformula
of type
H(1d ;d)r ;:::;(1;d)rn
(9.43)
with 1 ; d > 0 and ri 0 for all i = 1; : : : ; n.
Another type of reduction applies to ws of type (9.43) to reduce their
upper index d in its absolute value. This reduction refers to the fact that
j=L H(1d ;d)r ;:::;(1;d)rn $L Hr0 ;:::;rn & H;d+1
(9.44)
dr ;:::;;drn
holds true. And this fact may again be established by comparing the truth
degrees for the right hand and the left hand side of this L-biimplication. Using
again u for the truth degreePof the right hand side of the w in (9.44) and
the shorthand notation S = ri xi , one has according to (5.9)
u = maxf0; minf1; maxf0; S gg + minf1; maxf0; d + 1 ; dS gg ; 1g ;
thus one has because of (9.40), (9.41) and S 0
u = maxf0; minf1; S g ; 1 + minf1; maxf0; 1 + d(1 ; S )ggg
= maxf0; minf0; S ; 1g + minf1; maxf0; 1 ; d(S ; 1)ggg :
For the truth value v of the left hand side of the w in (9.44) one gets
according to in (9.42)
v = minf1; maxf0; d + (1 ; d)S gg
= maxf0; minf1; S ; d(S ; 1)gg :
In the case that S ; 1 0 holds true one has S 1 and
u = maxf0; minf1; maxf0; 1 ; d(S ; 1)ggg
= minf1; maxf0; 1 ; d(S ; 1)gg = 1 = v ;
and in the case that S ; 1 < 0 holds true one gets in a similar way
1
206
u = maxf0; minfS; maxfS ; 1; 1 + (1 ; d)(S ; 1)ggg

= maxf0; maxfS ; 1; minfS; 1 + (1 ; d)(S ; 1)ggg
= maxf0; S ; 1; S ; d(S ; 1)g
= maxf0; S ; d(S ; 1)g ;
using 1+(1 ; d)(S ; 1) = s ; d(S ; 1) and S = 1+(S ; 1) 1+(1 ; d)(S ; 1)
with d 0. Therefore one has u = v because one has
v = maxf0; S ; d(S ; 1)g
from d(S ; 1) S ; 1 which holds true here. So nally (9.44) is established.
Now one has to apply repeatedly on the one hand the reduction of some
ws of type Had ;:::;an according to (9.39), and on the other hand for the ws of
type (9.43), which result from this rst type of reduction, the other reduction
according to (9.44). Altogether this gives according to Theorem 2.2.2 some
w which is semantically equivalent with Had ;:::;an , and which contains as
subformulas besides propositional variables only ws of the type H:::c with
jcj < jdj and with c 0, i.e. which are themselves of the type (9.38).
These ws can be further reduced to ws of the type (9.38) with upper
indices with smaller absolute value. And this procedure can and has to be
continued up to the point where one has reached a w semantically equivalent with the w (9.36) which contains, besides propositional variables, only
subformulas of the form Hc0 ;:::;cn . Then one can take
1
Hc0 ;:::;cn (q1 ; : : : ; qn ) =def

1
ci
n X
X
i=1 j =1
qi :
(9.45)
Via (9.23) one immediately determines the truth degree w of Hc0 ;:::;cn as
w = minf1;
ci
n X
X
i=1 j =1
= minf1; maxf0;
xi g = minf1;
n
X
n
X
i=1
ci xi g
ci xi gg ;
i=1
because one has always ci 0 for these values in the w of (9.45).
Hence nally we have found for each truth degree function of the kind
(9.35) some w of type (9.36) which represents this truth degree function. 2
Now we are able to characterize the truth degree functions which can be
represented in L1 by some w H 2 LL .
Theorem 9.1.8. Let f : [0; 1]n ! [0; 1] be any n-ary function. The function
f is a truth degree function determined by some sentence of the Lukasiewicz
system L1 i f is continuous and there exist a nite number of polynomials
gi (x1 ; : : : ; xn ) = bi +
n
X
j =1
aij xj ;
i = 1; : : : ; m
(9.46)
207
with integer coecients aij ; bi , such that for all t1 ; : : : ; tn 2 [0; 1] there exists
some 1 k m with
f (t1 ; : : : ; tn ) = gk (t1 ; : : : ; tn ) :
Proof: We rst prove that for each w H 2 LL its corresponding truth
degree function w~H over W1 has these properties. And this shall be done by
induction on the complexity of H .
In the case that H is an atomic w, i.e. that one has H pi for a
propositional variable pi , one immediately has
w~H (x1 ; : : : ; xn ) = xi
such that w~H is continuous, and is the only polynomial g1 of the form (9.46)
which represents w~H .
In the case that H is a negation, i.e. that one has H :H1 and that
w~H is being represented by the polynomials gi0 , i = 1; : : : ; m, one has w~H =
non1 (w~H ) = 1 ; w~H which means that w~H is a continuous function and
represented by the polynomials gi = 1 ; gi0 , i = 1; : : : ; m, which obviously are
of the form (9.46).
It remains the nal case that H is an implication, i.e. that one has H
H1 !L H2 and that the corresponding
truth degree functions w~H and w~H
are represented by the polynomials gij , i = 1; : : : ; mj , for j = 1; 2 respectively.
Then one has
w~H (x1 ; : : : ; xn ) = seq 2 (w~H (x1 ; : : : ; xn ); w~H (x1 ; : : : ; xn ))
= minf1; 1 ; w~H (x1 ; : : : ; xn ) + w~H (x1 ; : : : ; xn )g :
Therefore w~H is continuous, and for any t1 ; : : : ; tn 2 [0; 1] one either has
w~H (t1 ; : : : ; tn ) = 1 ;
or one has for some suitable 1 i m1 ; 1 j m2 :
w~H (t1 ; : : : ; tn ) = 1 ; gi1 (t1 ; : : : ; tn ) + gj2 (t1 ; : : : ; tn ) :
Therefore w~H is being represented by the m1 m2 + 1 polynomials
g1 (x1 ; : : : ; xn ) = 1
gi;j (x1 ; : : : ; xn ) = 1 ; gi1 (x1 ; : : : ; xn ) + gj2 (x1 ; : : : ; xn )
which all are of the form (9.46).
Now assume, conversely, that f : [0; 1]n ! [0; 1] is a continuous function
and that g1 ; : : : ; gm are linear polynomials of the form (9.46) with integer
coecients which represent f . One may additionally suppose that the n-ary
polynomial functions determined by these polynomials g1 ; : : : ; gm are pairwise
dierent. Then each one of the linear equations
gi (x1 ; : : : ; xn ) = gj (x1 ; : : : ; xn ) 1 i; j m and i 6= j
determines some (n ; 1)-dimensional subspace Rij , i.e. a hyperplane, of the
n-dimensional space Rn . All these hyperplanes Rij together partition the
1
208
n into
whole space Rn and thus also the n-dimensional unit hypercube W1
nitely many n-dimensional convex regions. Let D1 ; : : : ; DK be all the closed
n by
polyhedrons which result from this partition of the unit hypercube W1
the subspaces Rij .
Consider any 1 k K and let P = (t1 ; : : : ; tn ) be an inner point of
the polyhedron Dk . Then there exists some 1 l m with f (t1 ; : : : ; tn ) =
gl (t1 ; : : : ; tn ), and one has gl (t1 ; : : : ; tn ) 6= f (t1 ; : : : ; tn ) for all i = 1; : : : ; m
with i 6= l according to the choice of Dk . Let P = (t1 ; : : : ; tn ) be another
inner point of the polyhedron Dk and f (t1 ; : : : ; tn ) = gl (t1 ; : : : ; tn ). Because
of the convexity of Dk the whole line segment PP is lying inside Dk . In the
case of l 6= l there would have to exist some point P0 on this line segment such
that f = gl holds true from P up to P0 and that f = gj for some j 6= l holds
true from P0 up to P . But then one would have gl (P0 ) = f (P0 ) = gj (P0 ) by
the continuity of f and all gi , contradicting the fact that P0 as an inner point
of Dk cannot belong to the subspace Rlj . Therefore f shall be described over
the polyhedron Dk by one and the same polynomial gl , and only by this one.
Thus one can renumber either the polynomials gi or the polyhedrons Dk in
such a way that always f is given over Dk by the polynomial gk . Suppose
for the rest of this proof that this has been done.
The goal now is to prove that for each one of these polyhedrons Dk
there exists some w Gk 2 LL such that f (x1 ; : : : ; xn ) = w~Gk (x1 ; : : : ; xn )
holds true for all points (x1 ; : : : ; xn ) 2 Dk , and such that w~Gk (x01 ; : : : ; x0n )
f (x01 ; : : : ; x0n ) holds true for all points (x01 ; : : : ; x0n ) 2= Dk . Then the whole
proof is nished because for the w
H (p1 ; : : : ; pn)
K
_
Gi (p1 ; : : : ; pn)
i=1
= w~H .
one obviously has f

So consider again one of these polyhedrons Dk and the polynomial gk
with property f = gk over Dk . This polyhedron Dk has common points with
some of the (n ; 1)-dimensional subspaces Rij . Assume that S 1 ; : : : ; S T are
all these \touching" Rij . Each of these subspaces S l contains only boundary
points of Dk and is determined by the equation gi = gj for suitable indices
i; j m, but is also determined by each one of the equations gi ; gj = 0 and
gj ; gi = 0. Because Dk is situated completely on one side of the hyperplane
Sl = Rij , one has either gi(P ) ; gj (P ) 0 for all points P 2 Dk , or one
has gj (P ) ; gi (P ) 0 for all these points. So suppose that for each index
l T a pair of indices il ; jl m is determined such that S l = Ril jl holds
true, together with
gil (P ) ; gjl (P ) 0 for all P 2 Dk .
Then one has additionally gil (Q) ; gjl (Q) 6= 0 for each point Q 2= S l , and
even: gil (Q) ; gjl (Q) > 0 i Q is on the same side of S l as Dk .
Now let 1 a 2 N and consider for each 1 l T the polynomial
209
hal = gk + a (gil ; gjl )

in the indeterminates x1 ; : : : ; xn . Obviously hal is always a linear polynomial
with integer coecients. And for any point P = (x1 ; : : : ; xn ) 2 Rn one has:
hal (P ) = gk (P ) i P 2 S l ;
hal (P ) > gk (P ) i P 2= S l and P; Dk are on the same side of S l ,
hal (P ) < gk (P ) i P 2= S l and P; Dk are on dierent sides of S l .
For each one of these polynomials hal consider a w Hla 2 LL such that
w~Hla = minf1; maxf0; halgg
holds true over W1 . Such a w exists according to Lemma 9.1.1 for each one
of these polynomials. And furthermore let Hk 2 LL such that over W1 it
holds true
w~Hk = minf1; maxf0; gk gg :
Then consider the w
Gak Hk ^
T
^
Hla
i=1
a
and denote by wk its corresponding truth degree function in the indeterminates x1 ; : : : ; xn . Then one immediately has
wka = maxf0; minf1; gk ; ha1 ; : : : ; haT gg
because for any real numbers b1 ; : : : ; bn there hold true

min minf1; maxf0; b1gg; : : : ; minf1; maxf0; bngg; 1
= minfmaxf0; b1g; : : : ; maxf0; bng; 1g
= maxf0; minf1; b1; : : : ; bngg :
Therefore for each inner point P of Dk it holds true
wka (P ) = maxf0; minf1; gk (P )gg
= maxf0; minf1; f (P )gg = f (P ) :
Because of the continuity of the functions wka and f one thus has wka = f
over the whole set Dk .
Thus we nally have to prove that the inequality wka f holds true
outside of Dk . And this will only depend on a suitable choice of the parameter
a. Denote by jQ1 Q2 j the length of the line segment Q1 Q2 connecting some
points Q1 ; Q2 of the n-dimensional unit hypercube [0; 1]n. Then there exists
a real number such that
jf (Q1 ) ; f (Q2 )j < jQ1 Q2 j
(9.47)
n
for all Q1 ; Q2 2 [0; 1] . For, if Q1 ; Q2 are points of the same polyhedron Dk ,
then (9.47) follows from the fact that f has constant partial derivatives over
210
Dk , and that Dk is bounded. If, on the other hand, Q1; Q2 belong to dierent
polyhedrons, then there exists a nite sequence P0 ; : : : ; Pr of collinear points

such that P0 = Q1 and Pk = Q2 , that for each 0 < i < r the point Pi
belongs to one of the hyperplanes Rjl , and that all the line segments Pi Pi+1
for i = 0; : : : ; r ; 1 belong to one of the polyhedrons Dk . Then one has,
according to this construction, immediately
jf (Q1 ) ; f (Q2 )j
r;1
X
i=1
jf (Pi ) ; f (Pi+1 )j
and gets (9.47) in general from the fact that (9.47) holds true over each one
of the polyhedrons Dk .
Again consider such a polyhedron Dk with its faces belonging to the
hyperplanes S 1 ; : : : ; ST . Let S i = S i \ Dk for all i = 1; : : : ; T and consider
some inner point P of Dk . Furthermore let
;i = fQ 2 Rn j PQ \ S i 6= ;g ;
i = ;i \ W1n
for each i = 1; : : : ; T . Then each point of Rn belongs to Dk or to one of the
n belongs to D k or to one
sets ;i , and each point of the unit hypercube W1
of the sets i . Fix also some l T and let for each point Q 2 ;l be Q
the intersection point of PQ and S l . Because one has gil (Q) ; gjl (Q) < 0,
according to the choice of Q, one can make the (negative) value hal (Q) as
small as one likes, i.e. one can make the absolute value jhal (Q)j as large as
one likes, in choosing the parameter a suitably. Therefore one can reach by
suitable choice of a that the dierence quotient
a
a
da (Q) = hl (Q) ; hl (Q )
l
jQQ j
of hal in direction PQ becomes negatively as small as one likes, because one
has hal (Q ) = gk (Q ) true independently of the choice of a. But hal is a linear
polynomial and hence this value along the ray ;;!
PQ is independent of the
choice of Q. The corresponding dierence quotient for f is bounded from
below by ; according to (9.47). Because S l is closed and bounded one can
get

db (Q) < f (Q) ; f (Q )
l
jQQ j
for all Q 2 S l and all Q 2 ;;!

PQ , and some suitably large parameter value b.
a

By hl (Q ) = F (Q ) for each Q 2 S l and each a one then immediately has
hcl (Q) < f (Q) for all Q 2 l
for each c b. But there are only nitely many sets l . Hence it is possible
to maximise over all the parameter values b to get some integer e > 1 such
that
211
hel (Q) < f (Q) for all 1 l T and all Q 2 l .

n with Q 2= D k there is some l T such that Q 2 l
Thus for any Q 2 W1
and
wke (Q) maxf0; hel(Q)g f (Q) :
Therefore one can nally choose for each 1 k m the w Gk one is looking
for as Gk Gek , and the whole proof is nished.
2
For the nitely many-valued Lukasiewicz systems Lm one now gets a
quite simple characterization of the truth degree functions which are Lm representable.
Corollary 9.1.2. Suppose m 3. Then a function f : Wmn ! Wm is the
truth degree function represented by some w H 2 LL in the system Lm i
for each n-tuple (t1 ; : : : ; tn ) of truth degrees from Wm the product (m ; 1)
f (t1 ; : : : ; tn ) is an integer and also a multiple of the greatest common divisor
of all the integers t1 (m ; 1); : : : ; tn (m ; 1); m ; 1.
Proof: Consider some H 2 LL and suppose that H describes in L1
the n-ary truth degree function f . Then one has over Wm that w~H is the
restriction of f to Wm , i.e. one has
w~H (t1 ; : : : ; tn ) = f (t1 ; : : : ; tn ) for all t1 ; : : : ; tn 2 Wm .
According to Theorem 9.1.8P
there are for i = 1; : : : ; m linear polynomials with
integer coecients gi = bi + nj=1 aij xj such that for each n-tuple (s1 ; : : : ; sn )
n there exists some index k with
from W1
f (s1 ; : : : ; sn ) = gk (s1 ; : : : ; sn ) :
Assume particularly (s1 ; : : : ; sn ) 2 Wmn . Then one has
w~H (s1 ; : : : ; sn ) = bi +
Therefore
n
X
j =1
aij sj :
(m ; 1) w~H (s1 ; : : : ; sn ) = bi (m ; 1) +
n
X
j =1
aij sj (m ; 1)
is an integer. Hence also the greatest common divisor of all the n + 1 integers
s1 (m ; 1); : : : ; sn (m ; 1); m ; 1 is a divisor of (m ; 1) w~H (s1 ; : : : ; sn ).
Now conversely let f : Wmn ! Wm be a function with these properties.
Let (t1 ; : : : ; tn ) 2 Wmn and ti = mk;i 1 for each i = 1; : : : ; n. Consider also
s = (m ; 1) f ( mk;1 ; : : : ; mk;n 1 )
and let d be the greatest common divisor of k1 ; : : : ; kn ; m ; 1. Form from
these integers the new integers ki = kdi for i = 1; : : : ; n, as well as s = ds
1
212
and d = d1 (m ; 1). By construction the integers k1 : : : ; kn ; d are relatively

prime, and therefore there exist integers b and ai for 1 i n such that
s = bd +
n
X
i=1
ai ki
holds true. Dividing this equation by d gives

n
s = b + X ai k i :
m;1
m;1
i=1
Therefore we consider the linear polynomial
g(x1 ; : : : ; xn ) = b +
n
X
i=1
ai xI ;
and for each i = 1; : : : ; n additionally the function

gki (x) = maxf0; minf(ki + 1) ; x (m ; 1); ;(ki ; 1) + x (m ; 1)gg :
From all these functions we can dene another n-ary function hsk ;:::;kn as
hsk ;:::;kn (x) = minfg(x); gk (x); : : : ; gkn (x)g
using the shorthand notation x = (x1 ; : : : ; xn ). By construction of g and all
the gki this gives immediately
hsk ;:::;kn ( mk;1 ; : : : ; mk;n 1 ) = ms;1 ;
and furthermore for each n-tuple (s1 ; : : : ; sn ) 2 Wmn dierent from the n-tuple
( mk;1 ; : : : ; mk;n 1 ) even
hsk ;:::;kn (s1 ; : : : ; sn ) = 0 :
Therefore we consider nally the function f characterized by the equation

f (x) = max hsk ;:::;kn (x) j 0 ki < m for i = 1; : : : ; n

and s = (m ; 1) f ( mk;1 ; : : : ; mk;n 1 ) :
Direct checking shows that this function f satises all the assumptions for
Theorem 9.1.8. Therefore f can in the system L1 be represented by some
w G. But f and f coincide over Wm . Hence f is also in the system Lm
represented by the same w G.
2
At rst the result of Corollary 9.1.2 may look a little bit technical. It is,
however, rather simple as the following reformulation shows which can be
inferred easily from Corollary 9.1.2.
Corollary 9.1.3. Suppose m 3. Then a function f : Wmn ! Wm is the
truth degree function represented by some w H 2 LL in the system Lm i
for each Wr Wm the restriction f Wr of f to Wr is a function from Wrn
into Wr .
1
213
Additionally it shall be mentioned that the result of Theorem 9.1.8 can be

used to nd adequate axiomatizations for each one of the nitely many-valued
Lukasiewicz systems Lm , cf. [537].
Before closing this section we come back to the second problem mentioned on p. 201 of how to complete the set fnon1 ; seq 2 g of connectives of
the Lukasiewicz systems in such a way that the resulting systems become
functionally complete. Of course, what has to be added should be connectives
which do not (all) satisfy the normal condition.
Proposition 9.1.1. If one adds to one of the Lukasiewicz systems Lm for
m 3 either a truth degree constant tm denoting the truth degree 2 = mm;;12
or a unary connective Tm with its truth degree function satisfying
m;2
for all x 2 Wm ;
(9.48)
verLTm (x) = 2 = m
;1
then the resulting extension of Lm is functionally complete.
Proof: Because of Theorem 8.1.2 it suces to prove that in both of these
extensions of Lm the unary Sheffer function sh or a w which describes this
truth degree function becomes denable.
Adding the truth degree constant tm allows to dene the mentioned type
of connective Tm as
Tm(H ) =def tm :
(9.49)
Therefore it suces to discuss the case that a connective Tm with truth
degree function (9.48) is available. In this case another binary connective
is dened by
(H1; H2 ) =def (J0 (H1) ^ J0(H2 )) _ ((H1 _ H2) & Tm(H1 ))
for which one gets for any u; v 2 Wm
verL (u; v) = sh(u; v):
(9.50)
L
L
This results in the case that Val (H1 ; ) = Val (H2 ; ) = 0 holds true from
the fact that one then has ValL (J0 (H1 ) ^ J0 (H2 ); ) = 1, using the Lm denable connective J0 , and hence also has ValL ( (H1 ; H2 ); ) = 1. In the
other case one has
ValL( (H1 ; H2 ); ) = et2 (ValL (H1 _ H2 ; ); 2 )
= ValL (H1 _ H2 ; ) ; m;1
and hence (9.50) at all. Thus this extension produces a functionally complete
system.
2
The assumptions of this proposition can be slightly changed e.g. in such a
way that one only needs to have an extension of Lm in which the connective
with truth degree function (9.50) or only the particular truth degree 2
becomes denable. Obviously instead an extension of Lm which allows one to
214
dene the truth degree m;1 = m1;1 would also be functionally complete, cf.
also [151].
For L3 this completability result was already found in [509] and formulated with reference to the additional connective T3 . This paper also gave
an adequate axiomatization of the extended system LSl3 by extending the
previously mentioned axiom system (L3 1); : : : (L3 4) by the axiom schemata
(LSl3 5) T3 (H ) !L :T3 (H ) ,
(LSl3 6) :T3 (H ) !L T3(H ) .
9.2 Algebraic Structures for the Lukasiewicz Systems

In classical logic there are two types of close relationships between propositional systems and Boolean algebras, cf. e.g. [38, 260, 444].
One of these relationships comes from the fact that, assuming that the
basic connectives are those ones for conjunction, disjunction, and negation,
one can in a quite natural way generalize the usual valuations, which are mappings from the set of propositional variables into the set >; ? of truth values,
to mappings into any Boolean algebra { which take the basic operations of
this Boolean algebra as generalizations of the usual truth-value functions
which semantically characterize the connectives. The core result then is that
the class of all logically valid ws of classical propositional logic is the class
of all those ws which have for each Boolean-algebra-valued valuation as
their (truth) value the unit element of this Boolean algebra.
The other type of relationship is to connect with each system of classical
(propositional) logic some particular algebraic structure which is built up
starting from this logical system. This relationship exists on the semantic
level as well as on the syntactic level. The crucial fact, from the semantical
point of view, is that the relation of semantical equivalence is an equivalence
relation in the class of all propositional ws, e.g. for a propositional language
which has a functionally complete set of (primitive) connectives. From the
syntactical point of view, on the other hand, one has that for each sound
and complete logical calculus its relation of \provable equivalence", i.e. the
relation between ws H; G which holds true i ` H ! G as well as ` G ! H
hold for the derivability relation ` of this calculus, is also an equivalence
relation in the class of all ws.
For both these equivalence relations their set of equivalence classes within
the set of all ws can be transformed into an algebraic structure with operations u; t; , which are dened between equivalence classes [H ] and [G] by
reference to the conjunction, disjunction, or negation of the representatives,
respectively, which e.g. means [H ] u [G] = [H ^ G]. The resulting algebraic
structure is a Boolean algebra for both equivalence relations10 and usually
10 And the two resulting Boolean algebras are not only isomorphic, but even
identical.
9.2 Algebraic Structures for Lukasiewicz Systems
215
called the Lindenbaum algebra of (the corresponding system of) classical

propositional logic.
In a similar manner one can also dene for systems of many-valued logic
their corresponding Lindenbaum algebras { either via the semantical approach or via the syntactical one. Of course, dierent systems of many-valued
logic in general shall have dierent algebraic structures as their Lindenbaum
algebras. However, even for the Lukasiewicz systems there has been for a
long time no unique understanding concerning what are the suitable algebraic structures to be connected with these systems. This largely depends
upon the fact that, because of the functional incompleteness of the \standard" Lukasiewicz systems, one did not have for a long time a comparably
well-accepted set of basic connectives as one has in classical logic.
Meanwhile, the MV-algebras are recognized as the most interesting algebraic counterparts of the Lukasiewicz systems. Therefore in the sequel the
main emphasis is laid on these structures. However, also some other algebraic
structures discussed in connection with the Lukasiewicz systems shall be
mentioned, particularly the Wajsberg algebras.
9.2.1 MV-algebras
MV-algebras have been introduced by Chang [91, 92] in investigations toward a completeness proof for the innite-valued Lukasiewicz system. They
play, however, an important rôle in algebraic studies related to all the Lukasiewicz systems, and proved to have interesting relationships to other mathematical structures too.11 For simplicity we use for the operations in MValgebras the same notation as for the corresponding connectives of the language LL .
Basic notions and properties. The following denition is a simplication,
essentially due to Mangani [349], of the original denition of Chang [91],
and refers only to some of the fundamental operations one usually considers
in the Lukasiewicz systems.
Denition 9.2.1. An algebraic structure A = hA; ; :; 0i with similarity
type h2; 1; 0i is an MV-algebra i hA; ; 0i is an abelian monoid with neutral
element 0, and if furthermore for all x; y 2 A there hold true
(i) ::x = x ;
(ii) x :0 = :0 ;
(iii) :(:x y) y = :(:y x) x :
11 The theory of MV-algebras for quite a time seemed to be of particular interest

only for the Lukasiewicz many-valued logics. Only more recently has it be-
come clear that this class of structures is much more important in mathematics.
Accordingly the theory is now studied in detail. The rst extended monograph
devoted to MV-algebras is [104] and covers the majority of the important actual
results in the eld of MV-algebras. A good survey of essential notions and results
is presented in [105].
216
Such an MV-algebra is nontrivial i it contains at least two elements.

Having in mind that in a lattice each inequality a 6 b can equivalently
be written e.g. as the equation a u b = a, an inspection of this denition
shows that all the dening conditions for MV-algebras can be written down
as (universally quantied) Horn formulas of the form
A1 ^ : : : ^ Ak;1 ) Ak
with term equations A1 ; : : : ; Ak . This means, in algebraic terms, that the class
of all MV-algebras is a quasivariety, and hence closed under the formation of

subalgebras and direct products, cf. e.g. [59, 110, 225].
Obviously each structure hW ; vel2 ; non1 ; 0i is an MV-algebra. What is
not explicitly taken into consideration here is e.g. the truth degree function
of the Lukasiewicz implication. But an MV-algebraic equivalent of it, and
also equivalents of the other truth degree functions which are important for
the Lukasiewicz systems, are denable in MV-algebras.12
Denition 9.2.2. Suppose that A = hA; ; :; 0i is an MV-algebra. Then
one denes for any x; y 2 A:
x
y =def :(:x :y) ;
x ! y =def :x y ;
x _ y =def (x
:y) y ;
x ^ y =def (x :y)
y ;
1 =def :0 :
Using these denitions, an MV-algebra is nontrivial i 0 6= 1.
One has some quite elementary corollaries of these denitions. We mention
the following ones.
Corollary 9.2.1. For each MV-algebra A = hA; ; :; 0i there hold true for
all a; b 2 A:
(MV1) a b = b a ,
(MV 1) a
b = b
a ,
(MV2) a (b c) = (a b) c , (MV 2) a
(b
c) = (a
b)
c ,
(MV3) a :a = 1 ,
(MV 3) a
:a = 0 ,
(MV4) a 1 = 1 ,
(MV 4) a
0 = 0 ,
(MV5) a 0 = a ,
(MV 5) a
1 = a ,
(MV6) ::a = a ,
(MV 6) :0 = 1 ,
(MV7) :(a b) = :a
:b ,
(MV 7) :(a
b) = :a :b ,
(MV8) a _ b = b _ a ,
(MV 8) a ^ b = b ^ a .
12 On the other hand, these basic examples indicate that the fundamental types
of truth degree functions, as discussed in Chapter 5, can also be understood as

particular operations in MV-algebras, supporting a point of view which does not
start the introduction of classes of basic connectives for many-valued logics from
the real unit interval, but from (suitable classes of) MV-algebras, cf. e.g. [43].
217
Proof: All these claims either coincide with MV-axioms as given in Definition 9.2.1, or follow immediately from the Denitions 9.2.1 and 9.2.2. So
e.g. (MV3) results from condition (iii) of Denition 9.2.1 by choosing y = :0.
2
Each MV-algebra A is, in a natural way, equipped with an ordering relation 5 dened for all a; b 2 A by
a 5 b =def there exists c 2 A with a c = b :
(9.51)
Before we prove that this is really an ordering relation we rst mention some
equivalent characterizations for this relation.
Corollary 9.2.2. For each MV-algebra A one has for all a; b 2 A
a 5 b , :a b = 1 , a
:b = 0 , b = a (b
:a) , :b 5 :a :
Proof: Assume a 5 b. Then there exists some c 2 A such that :a b =
:a (a c) = 1 c = 1 according to (MV1), (MV2), and (MV4). In the case
:a b = 1 one has a
:b = :(:a b) = :1 = 0. From a
:b = 0 one
gets a (b
:a) = b (a
:b) by Denition 9.2.1(iii), hence a (b
:a) =
b 0 = b. And from b = a (b
:a) one gets a 5 b by Denition (9.51),
using c = (b
:a).
Hence the rst four of these conditions are equivalent. And one has furthermore that a 5 b holds true i :a b = 1, i.e. i 0 = :(:a b) = :b
::a,
and hence i :b 5 :a.
2
Proposition 9.2.1. The relation 5 dened by (9.51) is a lattice ordering in

each MV-algebra A = hA; ; :; 0i and has ^; _ as its corresponding lattice
operations.
Proof: The re exivity of 5 follows from a 0 = a. If one has a 5 b

and b 5 c, then one has a d1 = b and b d2 = c for suitable d1 ; d2 , hence
a d1 d2 = c and thus a 5 c. Hence 5 is transitive. And assuming a 5 b
together with b 5 a gives a
:b = 0 as well as a = b (a
:b), according
to Corollary 9.2.2, thus a = b 0 = b, i.e. 5 is also antisymmetric.
Thus 5 is a partial ordering. What remains to be proved is that a _ b is
the 5-supremum and that a ^ b is the 5-inmum of fa; bg.
Because of a 5 (b
:a) a = (a
:b) b, according to (9.51) and
Denition 9.2.1(iii), one has a 5 a _ b. In the same way one gets b 5 a _ b
and hence supfa; bg 5 a _ b. For the reverse inequality it suces to show
a _ b 5 z , i.e. :(a _ b) z = 1 for each upper bound z for a; b. So suppose
a; b 5 z , i.e. :a z = 1 and z = b (z
:b). Then one has
:((a
:b) b) z = (:(a
:b)
:b) b (z
:b)
= (b
(a
:b)) :(a
:b) (z
:b)
= (b
(a
:b)) :a b (z
:b)
= (b
(a
:b)) :a z = 1 :
218
Hence one has supfa; bg = a _ b.

It remains to show inf fa; bg = a ^ b. And because of (MV7), (MV8) this is
shown if one proves inf fa; bg = : supf:a; :bg. However, one has inf fa; bg 5
a; b and thus :a; :b 5 : inf fa; bg, i.e. supf:a; :bg 5 : inf fa; bg. And from
:a; :b 5 supf:a; :bg one gets : supf:a; :bg 5 a; b and thus : supf:a; :bg 5
inf fa; bg. Hence ^ is the 5-inmum.
2
Corollary 9.2.3. For all elements a; b; c of an MV-algebra A = hA; ; :; 0i
there hold true

(i) a 5 b ) a c 5 b c ^ a
c 5 b
c ;
(9.52)
(ii) a 5 b c , a
:b 5 c and a
b 5 c , a 5 :b c ; (9.53)
(iii) a
b 5 a 5 a b :
(9.54)
Proof: (i) Assume a 5 b. Then one has b = a d for some suitable d,
and hence also b c = a c d, i.e. a c 5 b c. Therefore a
c 5 b
c
follows via the order reversing property of : and (MV 6).
(ii) One has a 5 b c i 0 = a
:(b c) = a
:b
:c, and thus i
a
:b 5 c. The rest follows in a similar way.
(iii) One immediately has a 5 a b from denition (9.51). And from (ii)
one has that a
b 5 a holds i one has a 5 :b a = a :b, and this is
always the case.
2
Hence all the operations ;
; ^; _ are isotonic w.r.t. the ordering 5,
and : is order reversing and thus antitonic. Therefore, also the operation
! is isotonic in its second argument, and antitonic in its rst one. And
the operations
; ! form an adjoint pair. Additionally there are further
associativity and distributivity properties, including the distributivity of the
underlying lattice.
Corollary 9.2.4. For each MV-algebra A = hA; ; :; 0i there hold true for
all a; b 2 A:
(MV9) a _ (b _ c) = (a _ b) _ c ,
(MV 9) a ^ (b ^ c) = (a ^ b) ^ c ,
(MV10) a (b ^ c) = (a b) ^ (a c), (MV 10) a
(b _ c) = (a
b) _ (a
c) ,
(MV11) a _ (b ^ c) = (a _ b) ^ (a _ c) , (MV 11) a ^ (b _ c) = (a ^ b) _ (a ^ c) .
Proof: Conditions (MV9) and (MV9) state only that the lattice operations are associative, and this obviously holds true for each lattice.
For (MV10) one has a (b ^ c) 5 a b; a c by the isotonicity of , hence
also a (b ^ c) 5 (a b) ^ (a c). To get also the reverse inequality it suces
to prove t 5 a (b ^ c) for each t such that t 5 a b and t 5 a c. So suppose
t 5 a b and t 5 a c, i.e. suppose t :a 5 b and t :a 5 c according to
Corollary 9.2.3(ii). But then one has t :a 5 b ^ c and thus t 5 a (b ^ c).
And (MV 10) easily follows from (MV10).
For (MV11) one has a _ (b ^ c) 5 (a _ b) ^ (a _ c) immediately by the
isotonicity of _, which holds true in each lattice. So it remains to prove
219
(a _ b) ^ (a _ c) 5 a _ (b ^ c), and for this it suces to prove t 5 a _ (b ^ c)

for each t with t 5 a _ b; a _ c. However, one has t 5 a _ b i t 5 a (b
:a),
and hence i t
:a 5 b
:a, and has also t 5 a _ b i t
:a 5 c
:a. Thus
one gets t
:a 5 (b
:a) ^ (c
:a) = (b ^ c)
:a from t 5 a _ b; a _ c, and
thus also t 5 a ((b ^ c)
:a), i.e. t 5 a _ (b ^ c), according to Corollary
9.2.3. And (MV 11) follows either similarly, or simply by the fact that any
lattice satises both distributive laws (MV11) and (MV 11) i it satises one
of them.
2
Therefore, each MV-algebra A = hA; ; :; 0i yields in a natural way an
abelian lattice ordered monoid hA;
; 0; 1; 5i with universal lower and upper
bounds.
The original denition of Chang [91] was based on the (primitive) operations ;
; : and the constants 0; 1, and consisted of a much longer list of
conditions: (MV1), . . . , (MV10), together with (MV 1); : : : ; (MV 10). Both
kinds of algebraic structures are thus denitionally equivalent structures.
Proposition 9.2.2. Each algebraic structure A = hA; ;
; :; 0;1i which
satises all the conditions (MV1); : : : ; (MV10) and (MV 1); : : : ; (MV 10), and
has the additional operations _; ^ dened as in Denition 9.2.2, corresponds
to a uniquely determined MV-algebra A = hA; ; :; 0i which has the additional property that the A-operations
; 1, dened according to Denition 9.2.2, coincide with the equally denoted A -operations.
Proof: It has rst to be proved that for each algebraic structure A =
hA; ;
; :; 0; 1i, which satises the conditions (MV1), . . . , (MV10) as well as
(MV 1); : : : ; (MV 10), the structure A = hA; ; :; 0i is an MV-algebra. However, is commutative by (MV1) and associative by (MV2). And (MV5) says
that 0 is neutral w.r.t. the operation . Hence hA; ; 0i is an abelian monoid
with neutral element 0. Condition (i) of Denition 9.2.1 is just (MV6), and
condition (ii) just (MV4) because of (MV 6). Finally, condition (iii) results
via the denition of _ (which is used in A as well as in A ) from (MV8)
together with (MV6), (MV 7).
What now only remains to be shown is that the operations
; 1 of A are
just the same operations one has already available from the structure A .
However, in A one has 1 = :0 by Denition 9.2.2, and in A one has the
same equation according to (MV 6). Furthermore, one has a
b = :(:a :b)
in A by Denition 9.2.2, and gets the same relationship in A from (MV 7)
and (MV6).
2
MV-algebras and L-systems. Now it is easy to prove one of the important

relationships between MV-algebras and the L-systems.
Theorem 9.2.1. The Lindenbaum algebra of each one of the L-systems is
an MV-algebra.
220
Proof: For simplicity we suppose to have the L-systems based on the set
fY; :g of (primitive) L-connectives. Let L be the set of all L-formulas with
Y; : as their basic connectives. We use also all the other connectives !13L, &,
^, _, $L , and suppose that they are introduced by suitable denitions. For
any H; G 2 L let:
H t G =def j=L (H $L G) ;
(9.55)
i.e. let t be the (equivalence) relation of semantical equivalence in L . The
equivalence class of H 2 L under t shall be [H ]. Dene in the quotient set
L = t operations ;
; c by
[H ] [G] =def [H Y G] ;
(9.56)
[H ]
[G] =def [H & G] ;
(9.57)
[H ]c =def [:H ] :
(9.58)
It is a routine matter to check that these denitions do not depend on the

particular choice of the representatives: having ws H; H 0 ; G; G0 2 L such
that H t H 0 and G t G0 , then also (H YG) t (H 0 YG0 ), (H & G) t (H 0 & G0 ),
and :H t :H 0 hold true. Furthermore let 0 =def [p0 &:p0 ]. Then the claim
of the theorem can be restated as:
hL = t; ; c; 0i is an MV-algebra :
(9.59)
Obviously, the operation in L = t is commutative and associative. It has
also 0 as a neutral element because one has j=L H Y (p0 &:p0 ) $L H for each
w H 2 L . Hence it remains to show conditions (i) to (iii) of Denition
9.2.1. However, (i) corresponds just to (T9), (ii) corresponds to the fact that
each w H Y :H is an L-tautology (in each one of the L-systems), and (iii)
corresponds to the fact that j=L H1 Y :(H1 Y :H2 ) $L H2 Y :(H2 Y :H1 )
holds true for all L-systems { which easily follows from the facts (T1), (T14),
(T8a), and (T9).
2
From a more algebraic point of view the notion of ideal is of crucial
importance (also) for MV-algebras.
Denition 9.2.3. A nonempty subset I 6= ; of an MV-algebra A is an ideal
of A i it contains the universal lower bound and is downward closed w.r.t. the
natural ordering 5 and also closed under , i.e. i one has for all a; b 2 A:
a5b2I )a2I;
a; b 2 I ) a b 2 I :
(9.60)
An ideal I of A is proper i one has I 6= A. Such a proper ideal I is prime
i it satises for all a; b 2 A the condition:
a
:b 2 I _ :a
b 2 I :
(9.61)
13 The details of these denitions do not matter here. In general we assume that
the dened connectives are used as abbreviations on the metalanguage level.
221
The maximal ideals are the -maximal elements in the class of all proper
ideals. And the intersection of all the maximal ideals of an MV-algebra A is
the radical Rad(A) of A.
The carrier jAj = A itself is an ideal of A, and often called the trivial
ideal. All the other, non-trivial ideals are the proper ones. Of course, one has
0 2 I for each ideal I of A. And 1 2= I characterizes the proper ones among
the ideals. Furthermore it is easy to see that the intersection of each family
of ideals of A is again an ideal of A, and also that each proper ideal, which
has a prime ideal as its subset, is itself a prime ideal.
If additionally one denes a function d in an MV-algebra A by
d(a; b) =def (a
:b) (:a
b) ;
(9.62)
then one gets the following relationships between ideals and congruence relations14 in MV-algebras, cf. e.g. [91, 104]:
1. If R is a congruence relation of A, then the R-equivalence class of 0 is an
ideal of A.
2. If I is an ideal of A, then RI = f(x; y)k d(x; y) 2 I g is a congruence
relation of A.
3. If R is a congruence relation of A, then the quotient structure A=R is
again an MV-algebra.
These are standard results which hold true for large classes of algebraic structures, viz. for the standard types of universal algebras, cf. e.g. [225, 344]. To
have a simplied notation, in the following we shall write A=I instead of A=RI
for each ideal I of an MV-algebra A.
Lemma 9.2.1. Let A be an MV-algebra. Then there hold true:
(a) A proper ideal I of A is prime i for all a; b 2 A it holds true:
a ^ b 2 I ) a 2 I or b 2 I :
(9.63)
(b) If P is a prime ideal of A then the quotient structure A=P is an MValgebra whose lattice ordering 5P is linear.
(c) For each a 2 A with a 6= 0 there exists a prime ideal P of A with a 2= P .
Proof: (a) Let I be prime and a ^ b 2 I . Assume that a
:b 2 I . (The
case b
:a 2 I may be treated similarly.) Then one has a ^ b 5 a, and hence
a = (a ^ b) (a
:(a ^ b)) according to Corollary 9.2.2. But one has
a
:(a ^ b) = a
(:a (a
:b)) = a ^ (a
:b) 5 a
:b 2 I ;
and therefore a
:(a ^ b) 2 I . Hence one has also
14 As usual in universal algebra, a congruence relation in an MV-algebra A is such
an equivalence relation which is compatible with the operations ; : of A, i.e.

which always changes the result of an application of such an operation into an
equivalent one if one changes the operands into equivalent ones.
222
a = a _ (a ^ b) = (a
:(a ^ b)) (a ^ b) 2 I ;
which means that (9.63) holds true.

Now suppose that (9.63) holds true for the proper ideal I of A. Because
of 0 2 I it suces for (9.61) to prove that
(a
:b) ^ (:a
b) = 0
(9.64)
holds true for any elements a; b of an MV-algebra. To get this equality, rst
recognize that parallel to property (iii) of the MV-algebras, which is just
(MV8), one also has for all elements x; y of any MV-algebra
(x :y)
y = (y :x)
x ;
which is nothing but (MV 8). Applying this relationship repeatedly, and always marking the terms to which it is applied by underlining, one has
(a
:b) ^ (:a
b)
= (a
:b)
((:a
b) :(a
:b))
= a
:b
(b [:a (:a
b)])
= a
[:a (:a
b)]
(:[:a (:a
b)] :b)
= :a
b
(:(:a
b) a)
(:[:a (:a
b)] :b)
= :a
(:(:a
b) a)
[a
:(:a
b)]
(:b [:a (:a
b)])
and hence (9.64) because of (MV 3); (MV 4).
(b) According to the previous remarks one has only to show that the
lattice ordering 5P of A=P is linear, i.e. that any two RP -equivalence classes
[a]P ; [b]P of A=P are 5P -comparable: [a]P 5P [b]P or [b]P 5P [a]P . According
to Corollary 9.2.2 this means that one has to show
[a]P
[b]P c = [0]P or [b]P
[a]P c = [0]P ;
which is equivalent to: a
:b 2 P or :a
b 2 P . But this is just the
characteristic property (9.61) of the prime lters.
(c) Let 0 6= a 2 A. It is immediately clear from Denition 9.2.3 that the
union of each inclusion-ordered chain of ideals of an MV-algebra A (which
do not contain a 2 A) is again an ideal of A (which does not contain a 2 A).
Therefore Zorn's Lemma can be applied to the class Ma of all proper ideals
I of A with property a 2= I , i.e. Ma has an inclusion-maximal member Pa .
By denition of Ma the ideal Pa is proper. It is, in fact, prime.
Assume that Pa is not prime. Then there exist b; c 2 A with b
:c 2= Pa
as well as with :b
c 2= Pa . Now consider the smallest ideal I1 of A such that
Pa I1 and also b
:c 2 I1 hold true.15 Fortunately, I1 can be determined
explicitly. To do this one denes for each x 2 A its integer-multiples nx
recursively by
15 This ideal exists because A is an ideal with these two properties, and the intersection of any family of ideals (with these properties) is again an ideal (with
these properties).
223
0x =def 0;
(k + 1)x =def kx x ;
and considers the subset
I1 =def fy 2 A j y 5 s n(b
:c) for some s 2 Pa ; n 2 N g
of A. It is easy to recognize that I1 is an ideal of A. And obviously one has
Pa I1 and b
:c 2 I1 . Hence I1 I1 . However, also I1 I1 holds true
because for each y 2 I1 one has to have y 2 I1 . By the maximality property
of Pa one has a 2 I1 and hence
a 5 s n(b
:c) for some s 2 Pa and n 2 N :
(9.65)
Considering in a similar way the smallest ideal I2 such that Pa I2 as well
as :b
c 2 I2 hold true, gives
a 5 t m(:b
c) for some t 2 Pa and m 2 N :
(9.66)
All together this gives for u = s t and k = maxfn; mg the relationships
u 2 Pa ; a 5 u k(b
:c); a 5 u k(:b
c)
by monotonicity of and 0 5 x for each x 2 A. Thus one has
a 5 (u k(b
:c)) ^ (u k(:b
c)) = u (k(b
:c) ^ k(:b
c)) :
From (9.64) one has (b
:c) ^ (:b
c) = 0. And in general one has with
x ^ y = 0 also nx ^ ny = 0 for all x; y 2 A and all n 2 N : From x ^ y 5 x one
has x = x (x ^ y) = (x x) ^ (x y) = 2x ^ y, hence 0 = x ^ y = 2x ^ y
and thus 0 = 2x ^ y. In a similar way one gets furthermore 0 = 2x ^ 2y and
even 0 = 2nx ^ 2n y for each n 2 N , and thus nally nx ^ ny 5 2n x ^ 2ny = 0.
Hence one has a 5 u 0 = u 2 Pa and therefore also a 2 Pa , a contradiction. Thus Pa is prime.
2
Among the MV-algebras, the subclass of all the MV-chains, i.e. of all
those MV-algebras for which their natural ordering 5 is linear, is of particular
importance.
To state the next result we have to introduce a further algebraic notion
which generalizes the notion of direct product: the notion of subdirect product. And to do this, we have to refer to the notion of homomorphism for
MV-algebras.16
Denition 9.2.4. For MV-algebras Ai = hAi ; i; :i ; 0ii, i = 1; 2, a homomorphism between them is such a mapping h : A1 ! A2 , for which one has
for all x; y 2 A1 :
(a) h(01 ) = 02 ,
(b) h(x 1 y) = h(x) 2 h(y) ,
(c) h(:1 x) = :2 h(x) .
16 This is the usual notion of homomorphism for algebraic structures. We add the
denition here simply for completeness.
224
Denition 9.2.5. By a subdirect product of some family (Ai )i2I of algebraic structures, here of MV-algebras, one
means such a substructure B, here
Q
a sub-MV-algebra, of the direct product i2I Ai such that for each one of the
projection mappings i for i 2 I the restriction i B is a homomorphism
onto the factor Ai .
Theorem 9.2.2. Each MV-algebra is (isomorphic to) a subdirect product of
MV-chains.
Proof: LetTA be an MV-algebra and PA the class of all its prime ideals.
Then one has PA = f0g by Lemma 9.2.1(c) and thus
\
fRP j P 2 PA g = f(x; y) j d(x; y) = 0g = f(x; x) j x 2 Ag = idA ;
because one has (x
:y) (:x
y) = 0 from d(x; y) = 0, hence x
:y = 0
and :x
y = 0, and thus x 5 y and y 5 x, which means x = y because 5 is
a partial ordering.
By a general result of universal algebra, cf. e.g. [59, 225], it follows from
this property of the class of all congruence relations RP for P 2 PA , that A
is isomorphic to a subdirect product of the family of all MV-algebras A=P
for P 2 PA . Hence the claim follows from Lemma 9.2.1(b).
2
As a side remark it should be mentioned that for nite MV-algebras an
even stronger characterization is possible.
Theorem 9.2.3. An MV-algebra is nite i it is isomorphic to a nite product of nite MV-chains.
Moreover, given a nite MV-algebra, the MV-chains which appear as \factors" in its representation are uniquely determined (up to their order as
factors).
For another interesting result which refers to MV-chains we need the notion of an MV-equation : this is any equation in the rst-order language of
MV-algebras, i.e. in the language with the same similarity type as the MValgebras { and usually also with operation and relation symbols for all the
notions introduced in Denition 9.2.2 { with its individual variables interpreted as elements of an MV-algebra under consideration.
The notion of satisfaction of such an MV-equation in an MV-algebra A
(w.r.t. a given valuation, i.e. a mapping of the set of individual variables
into the carrier A of A) is the usual model theoretical notion of satisfaction.
The same holds true for the notions of validity (of an MV-equation in an
MV-algebra) and of model. We use the symbol j= for these relations as usual.
Then one immediately has the following results.
Lemma 9.2.2. Let A; B be MV-algebras and t; t1 ; t2 terms of the language
of MV-algebras with their variables among x1 ; : : : ; xk .
225
(a) If B is a subalgebra of A and A j= t1 = t2 holds, then one has also

B j= t1 = t2 .
(b) If h is a homomorphism from A to B and a1 ; : : : ; ak any k-tuple of elements of jAj, then one has tB (h(a1 ); : : : ; h(ak )) = h(tA (a1 ; : : : ; ak )) for
the interpretations tA ; tB of t in A; B.
(c) If h is a homomorphism from A onto B and A j= t1 = t2 holds, then one
also has B j= t1 = t2 .
Proof: (a) and (b) follow immediately from the corresponding denitions
in classical logic. And (c) is an immediate corollary of (b).
2
Lemma 9.2.3. For each family (Ai )i2I of MV-algebras and all terms t1; t2
of the language of MV-algebras one has
Y
Ai j= t1 = t2 for all i 2 I )
Ai j= t1 = t2 :
i2I
Proof: Assume that the

variables of the terms t1 ; t2 occur among x1 ; : : : ; xk ,
Q
and let f1 ; : : : ; fk 2 A = i2I Ai . Then one has because of Ai j= t1 = t2 for
each i 2 I :
tA1 (f1 ; : : : ; fk )(i) = tA1 i (f1 (i); : : : ; fk (i))
= tA2 i (f1 (i); : : : ; fk (i)) = tA2 (f1 ; : : : ; fk )(i) :
Q
Thus one has tA1 (f1 ; : : : ; fk ) = tA2 (f1 ; : : : ; fk ), which means i2I Ai j= t1 = t2
by the assumptions about f1 ; : : : ; fk and the variables of t1 ; t2 .
2
Proposition 9.2.3. For each subdirect product B of some family (Ai )i2I of
MV-algebras and all terms t1 ; t2 of the language of MV-algebras one has

B j= t1 = t2 , Ai j= t1 = t2 for all i 2 I :
Proof: If one has Ai j= t1 = t2 for all i 2 I , then one has Qi2I Ai j=
t1 = t2 by Lemma 9.2.3, and hence B j= t1 = t2 by Lemma 9.2.2(a).
Suppose conversely B j= t1 = t2 . Then one has Ai j= t1 = t2 for all i 2 I ,
because of Lemma 9.2.2(c) and the denition of the subdirect product. 2
Theorem 9.2.4. An MV-equation is valid in all MV-algebras i it is valid
in all MV-chains.
Proof: If an MV-equation t1 = t2 is valid in all MV-algebras then it is
a fortiori valid in all MV-chains.
And if an MV-equation is valid in all MV-chains, then it is valid in all
subdirect products of MV-chains according to Proposition 9.2.3. Hence it is
valid in all MV-algebras, which are isomorphic to such a subdirect product
according to Lemma 9.2.2(c).
2
According to Theorem 9.2.4, any MV-equation for which there is an MVcounterexample hence is already not valid in a suitable MV-chain. In other
226
words: the equational theory of all MV-algebras coincides with the equational
theory of all MV-chains.
This result can be further strengthened to the following algebraic completeness theorem using an interesting relationship between MV-algebras and
ordered abelian groups.
To relate MV-chains and o-groups, consider an o-group G = hG; ; o; G i,
and some a 2 G with o G a. Then consider the set
G+a =def fx 2 G j o G x G ag :
Within this set G+a , operations ~; ? shall be dened by
b ~ c =def minfa; b cg ; b? =def a ;1 b ;
with min taken w.r.t. the ordering G . Now let
; (G ; a) =def hG+a ; ~; ? ; oi :
(9.67)
Remark: In the particular case of the ordered group R = hR; +; 0; i of all
reals under addition one gets by this construction the truth degree structure
of the Lukasiewicz system L1 :
; (R; 1) = hW1 ; vel2 ; non1 ; 0i :
On the other hand, if A = hA; ; :; 0i is an MV-algebra, then consider
the set
A+ =def f(m; a) j a 2 A ^ m integerg
of ordered pairs, and in this set the equivalence relation t dened by:
(m; a) t (n; b) =def (:a = b = 0 ^ n = m + 1)
_ (a = :b = 0 ^ n + 1 = m) :
Write hm; ai for the equivalence class of (m; a) under t. Now consider the
set
A =def fhm; ai j a 2 A ^ m integerg ;
and in it an operation A dened by

+ n; a bi;
if a b < 1
hm; ai A hn; bi =def hhm
m + n + 1; a
bi; if a b = 1 ,
as well as a relation dened by
hm; bi hn; ci =def m < n _ (m = n ^ b 5 c)
_ (m = n + 1 ^ b = :c = 0) :
It is a routine matter to check that these denitions are independent of the
particular choice of the representatives of the equivalence classes involved, and
that is a partial ordering, i.e. a re exive, transitive, and antisymmetric
relation. Then let
G (A) =def hA ; A ; i :
227
Proposition 9.2.4. (i) Let A = hA; ; :; 0i be an MV-chain. Then the

structure G (A) = hA ; A ; h0; 0i; i is an o-group, and A and ; (G (A); h0; 1i)
are isomorphic as MV-algebras. Furthermore there exists for each x 2 A
some integer k such that kh0; 1i x (k + 1)h0; 1i.
(ii) Let G = hG; ; o; G i be an o-group. For each o G a then the algebraic structure ; (G ; a) = hG+a ; ~; ? ; oi is an MV-chain, with G as its
MV-ordering. And if there exists furthermore for each y 2 G an integer n
such that na G y G (n + 1)a, then G and ; (G ; a) are isomorphic as ogroups.
Proof: (i) It is immediately clear that the operation A is commutative

and has h0; 0i as a neutral element. Furthermore one has that for each hk; ci 2
A the element h;k ; 1; :ci 2 A is the A -inverse of hk; ci, i.e. has the
property hk; ci A h;k ; 1; :ci = h;k ; 1; :ci A hk; ci = h0; 0i. Hence there
exists for each x 2 A an element x^ 2 A such that x A x^ = x^ A x = h0; 0i.
Assuming the associativity of A this means that condition (1.7) is satised:
for all x; y 2 A one has x A (^x A y) = y as well as (y A x^) A x = y.
Hence one has to show that A is an associative operation. This is not
really dicult, but tedious because it needs the consideration of a series of
dierent cases. We shall not discuss these details, but refer the interested
reader to the paper [92] where the present proof was rst given, and where
the details can be found.
What remains to be shown is now that the ordering is linear, and that
the monotonicity condition
hm; ai hn; bi ) hm; ai A hk; ci hn; bi A hk; ci
(9.68)
is satised.
Let ; 2 A such that = hm; ai and = hn; bi. The case that m =
n + 1 ^ a = :b = 0 gives = hn + 1; 0i and = hn; 1i, i.e. = . Hence one
has for the case 6= as characterization of immediately
< , , m < n _ (m = n ^ a 5 b) :
Therefore the linearity of results from the fact that the usual ordering of
the integers as well as the MV-ordering 5 are linear.
To get the monotonicity condition (9.68) suppose hm; ai hn; bi. First
let b c = 1. In the case m < n one then has m + k < m + k + 1 < n + k + 1
and hence hm; ai A hk; ci hn; bi A hk; ci. In the case m = n and
a 5 b one has for a c < 1 immediately m + k < n + k + 1, and one
has for a c = 1 the inequality a
c 5 b
c by Corollary 9.2.3(i), and
hence again hm; ai A hk; ci hn; bi A hk; ci. And in the case m = n + 1
and a = :b = 0 one has hm; ai = hn + 1; 0i = hn; 1i = hn; bi and hence
hm; ai A hk; ci hn; bi A hk; ci.
Now let b c < 1. In the case m < n one has m + k < n + k for a c < 1
and hence hm; ai A hk; ci hn; bi A hk; ci, and one has a
c 5 c 5 b c
or m + 1 + k < n + k for a c = 1 by Corollary 9.2.3(i), and hence again
228
hm; ai A hk; ci hn; bi A hk; ci. In the case m = n and a 5 b one has
immediately ac 5 bc < 1 and thus again hm; ai A hk; ci hn; bi A hk; ci.
And in the case m = n + 1 and a = :b = 0 one has hm; ai = hn; bi. Thus,
(9.68) is established.
To get an isomorphism from A onto ; (G (A); h0; 1i), consider the mapping ', characterized by '(a) = ho; ai for each a 2 A. That ' really is an
isomorphism follows immediately from the fact that one has
(A )+h0;1i = fh0; xi j a 2 Ag
and from the denitions of A and . Furthermore, an easy induction argument gives
nh0; 1i = nh1; 0i = hn; 0i
for each n 2 N. Therefore one has for each hm; ai 2 A with a 6= 1
hm; 0i hm; ai < hm + 1; 0i :
And each 2 A can be represented as = hm; ai with a 6= 1.
(ii) It is a routine matter to check that ; (G ; a) = hG+a ; ~; ? ; oi for o G a
is an MV-algebra.
For any b; c 2 G+a one gets b A (a ;A1 b) = a G c A (a ;A1 b)
from b G c, and hence that minfa; (a ;A1 b) A cg = a = o? . And this
means b 5 c for the MV-ordering 5 of ; (G ; a). From b G c _ c G b for
any b; c 2 G+a one thus gets b 5 c _ c 5 b, i.e. one gets that ; (G ; a) is an
MV-chain. And in case of b 5 c one cannot have b 6G c, because otherwise
one would have c G b and hence c 5 b, i.e. b = c, contradicting b 6G c.
Hence G and 5 coincide in G+a .
If there exists for each y 2 G some integer ny such that ny a G y <G
(ny + 1)a, then one can dene a function over G by
(y) =def hny ; y ;1 ny ai ;
using the inverse operation ;1 of the group operation . Let z 2 G another
group element and nz a G z <G (nz + 1)a, and consider u = y ;1 ny a
and v = z ;1 nz a. Then one has o G u, o G v as well as y = ny a u,
z = nz a v. Therefore it is
y z = (ny a u) (nz a v) = (ny + nz )a (u v)
as well as
(y) = hny ; ui ;
(z ) = hnz ; vi :
In the case u v G a one has (ny + nz )a G y z G (ny + nz + 1)a and
hence
(y z ) = hny + nz ; u vi = hny + nz ; u vi :
And in the other case u v 6G a one has a G u v G 2a, and therefore
now (ny + nz + 1)a G y z G (ny + nz + 2)a and
229
(y z ) ;1 (ny + nz + 1)a = (u v) ;1 a

= a ;1 ((a ;1 u) (a ;1 v)) = u
v ;
which means nothing but
(y z ) = hny + nz + 1; u
vi :
Thus it is shown that respects the operations in G and ; (G ; a).
The facts that is 1-1, that it maps onto G+a , and that it respects also
the ordering relations in G and ; (G ; a), can be shown in a similar, elementary
way. We leave the details to the reader.
2
Corollary 9.2.5. For each nontrivial MV-chain A there exists some o-group
G (A) = hG; G ; o ; i, isomorphic with G (A), such that A = jAj is a closed
interval G+e in G (A) for some suitable e with o < e, and such that additionally ; (G ; e) = A.
Proof: Because of the isomorphy of the MV-chains A and ; (G (A); h0; 1i)
it is a routine matter to exchange isomorphically in the o-group G (A) all the
elements of its closed interval I = fx j h0; 0i x h0; 1ig by the elements
of A, and to adapt the operations of G (A) to this exchange, such that G (A)
and G (A) become itself isomorphic (as o-groups).
2
Proposition 9.2.5. (i) For each MV-term t there exists some term te of the
language of o-groups, extended by a constant e, such that for each MV-chain
A and each A-assignment one has
(A; ) j= t = 0 , (G (A); ) j= te = o
(9.69)
for the corresponding o-group equation te = o.
(ii) For each MV-equation E there exists an 8 -formula E (x) of the language of ordered (abelian) groups with one free variable x such that an MVchain A is a model of E i its corresponding o-group A is a model of the
8 -sentence E (h0; 1i).
Proof: (i) Suppose that A is an MV-chain. Consider a corresponding
o-group G (A) such that A = jAj is a closed interval G+e in G (A) for some
suitable e, and such that ; (G ; e) = A. Let t be some MV-term. Transform
it according to the following principles:
substitute the constant e for the constant 1;
substitute the constant o for the constant17 0;
substitute the term minfe; g for each subterm of the form ;
substitute the term e ;1 for each subterm of the form :.
17 For simplicity we suppose here that the constants of the languages under con-
sideration are identied with the names of the objects we have used up to now.
230
Now choose e as the unit element in G (A). Then it is a routine matter to

show that one has for each A-assignment :
(A; ) j= t = 0 , (G (A); ) j= te = o :
(ii) Let E be an MV-equation t1 = t2 , and let the MV-terms t1 ; t2 contain
at most the variables x1 ; : : : ; xn , and of course possibly the symbols 0 and
1. Let x be another variable, dierent from x1 ; : : : ; xn . Now transform E
according to the previous principles. However, always write x instead of e.
Then a w E 0 results which does not contain any quantiers, and which is
built up using the function symbols ; ;1; min, the relation symbol , the
constant o and the variables x; x1 ; : : : ; xn . Then let
E (x) def 8 x1 ; : : : ; xn (o x1 x ^ : : : ^ o xn x ) E 0 ) :
It is again a routine matter to check for each MV-chain A, using Proposition
9.2.4, that A is a model of E i A is a model of E (h0; 1i).
2
Theorem 9.2.5 (Algebraic Completeness Theorem). An MV-equation

is valid in all MV-algebras i it is valid in the particular innite Lukasiewicz
MV-algebra hW1 ; vel2 ; non1 ; 0i.
Proof: Remember that one has for the o-group R = hR; +; 0; i the
equality ; (R; 1) = hW1 ; vel2 ; non1 i. Hence it suces to show that an MVequation E is valid in all MV-algebras i it is valid in the MV-chain ; (R; 1).
Of course, if E is valid in all MV-algebras then it is also valid in the
MV-chain ; (R; 1).
So suppose that E is valid in ; (R; 1), i.e. that ; (R; 1) is a model of E .
Then it suces to show that E is valid in all MV-chains. Assume that this is
not the case, i.e. that there exists some MV-chain B which is not a model of
E . Form from E the 8 -formula E (x) according to Proposition 9.2.5(ii). Then
the o-group G (B) is not a model of the 8 -sentence E (h0; 1i), and hence also
not a model of the sentence 8 x(o < x ) E (x)), which is also (equivalent
to) an 8 -sentence, i.e. a universal sentence.
According to Theorem 1.2.1 therefore also the o-group R is not a model
of 8 x(o < x ) E (x)). This means that there exists a real number c > 0
such that E (c) is false in R. But then also E (1) is false in R because
there exists an automorphism of R which maps c to 1. Because according to
Proposition 9.2.4 the o-groups R and G (; (R; 1)) are isomorphic, the sentence
E (h0; 1i) is false in G (; (R; 1)), which means by Proposition 9.2.5 that the
MV-equation E is not valid in the MV-chain ; (R; 1). And this contradicts
the choice of E .
2
Corollary 9.2.6. A w H of the language LL of the Lukasiewicz systems
is an L1 -tautology i H is valid in all MV-chains, and hence in all MValgebras.
231
Proof: It is a routine matter to transform the w H of LL into a term H

of the language of MV-algebras.18 The main idea for this transformation is
simply to change from propositional variables to variables for elements of MValgebras, and to change from connectives to their corresponding operations
in MV-algebras. And the corresponding operation for a connective ' is just
this one which { algebraically { corresponds in some given MV-algebra to the
truth degree function of ' in the MV-algebra hW1 ; vel2 ; non1 ; 0i.
With the w H one now connects the MV-equation H = 1. Then one
has that H is an L1 -tautology i the MV-equation H = 1 is valid in the
MV-algebra hW1 ; vel2 ; non1 ; 0i. According to the previous theorem hence H
is an L1 -tautology i the MV-equation H = 1 is valid in all MV-algebras,
i.e. i it is valid in all MV-chains according to Theorem 9.2.4. And obviously
H is valid in some MV-algebra A i the MV-equation H = 1 is valid in A.
2
The MV-algebras, thus, have for the Lukasiewicz system of innitevalued logic essentially the same rôle as the Boolean algebras have for classical two-valued logic.
As shown in Theorem 9.2.4, the MV-chains are of particular importance.
And for the consideration of MV-chains it is interesting to recognize that the
particular innite Lukasiewicz MV-chain hW1 ; vel2 ; non1 ; 0i is the \prototypical" MV-chain in the sense that all other ones can be locally embedded
into it.
Corollary 9.2.7. Each nontrivial19 MV-chain is locally embeddable into the
innite Lukasiewicz MV-algebra hW1 ; vel2 ; non1 ; 0i.
Proof: According to Theorem 1.2.2 one has e.g. to show that any universal sentence G 8 x1 ; : : : ; xn H (x1 ; : : : ; xn ) of the language of MV-algebras
which is valid in the innite Lukasiewicz chain L1 = hW1 ; vel2 ; non1 ; 0i is
valid in each MV-chain.
Because G is logically equivalent with the slightly more complicated universal sentence
8 x1 ; : : : ; xn (0 x1 1 ^ : : : ^ 0 xn 1 ) H (x1 ; : : : ; xn )) ; (9.70)
we have only to consider universal sentences of the form (9.70).
So consider some nontrivial MV-chain A, and assume that G is not valid
in A. Then there exists some A-assignment such that
(A; ) 6j= (0 x1 1 ^ : : : ^ 0 xn 1 ) H (x1 ; : : : ; xn )) :
18 It is not a serious problem which type of primitive vocabulary is supposed to
be used. If one is interested to get a term in the original, i.e. not denitionally
extended language of MV-algebras, one has to change from H to a w H 0 which
is semantically equivalent with H . And if one likes to start with H one may get
a term which uses some of the \additional" operations from Denition 9.2.2.
19 The case of the trivial MV-chains, which have a singleton as their carrier, is
obvious.
232
The open w H (x1 ; : : : ; xn ) is built up from suitable MV-equations t1 = t2

using propositional connectives.20 Because each such MV-equation t1 = t2 is
equivalent to the MV-equation (t1
:t2 ) _ (t2
:t1 ) = 0, we may assume
that H (x1 ; : : : ; xn ) is built up from simple MV-equations of the form t = 0.
Now Proposition 9.2.5(i) gives
(G (A); ) 6j= (o x1 e ^ : : : ^ o xn e ) He (x1 ; : : : ; xn ))
for the w He(x1 ; : : : ; xn ) of the language of o-groups, which is formed out
of H by transforming each MV-equation t = 0 which occurs in H into the
equation te = o of the language of o-groups, and leaving the propositional
structure unchanged. The crucial point here is that one has (9.69).
Hence one has
G (A) 6j= (o x1 e ^ : : : ^ o xn e ) He (x1 ; : : : ; xn )) ;
and therefore also
R j6 = (o x1 e ^ : : : ^ o xn e ) He (x1 ; : : : ; xn )) ;
Retransforming He into H , this means that one has
(R; ) 6j= (0 x1 1 ^ : : : ^ 0 xn 1 ) H (x1 ; : : : ; xn ))
for some suitable R-assignment .
Now there exists some isomorphism f of R onto itself which maps (the
positive real number21 which interprets) e to 1, and which respects all the
other operations. This nally gives for the R-assignment , given by (x) =
f ( (x)) for all variables x, the result
(L1 ; ) 6j= (0 x1 1 ^ : : : ^ 0 xn 1 ) H (x1 ; : : : ; xn )) ;
and thus also (L1 ; ) 6j= H (x1 ; : : : ; xn ). Hence G is also not valid in the
Lukasiewicz MV-chain L1 .
2
Some further algebraic results are of importance too for dierent applications of MV-algebras and shall be mentioned here without proof. The interested reader may consult e.g. [104].
Denition 9.2.6. An MV-algebra A is simple i it is nontrivial and has
f0g as its only proper ideal, and it is semisimple i Rad(A) = f0g.
There are characterizations of the simple and the semisimple MV-algebras
which are interesting and important from the viewpoints of many-valued logic
and of fuzzy set theory. For the simple MV-algebras one has the following:
20 Of classical propositional logic, of course.
21 If this is not immediately the case, i.e. if e is (pathologically) interpreted as
the real number zero, change from the w H (x1 ; : : : ; xn ) to the equivalent w
H (x1 ; : : : ; xn ) ^ 0 < 1.
233
Theorem 9.2.6. An MV-algebra A is simple i it is isomorphic to a subalgebra of the particular innite Lukasiewicz MV-algebra hW1 ; vel2 ; non1 ; 0i.
This theorem immediately gives a characterization of the truth degree
structures of the nite many-valued Lukasiewicz systems.
Corollary 9.2.8. The nite Lukasiewicz MV-algebras hWn ; vel2; non1; 0i
are, up to isomorphism, the only nite and simple MV-algebras.
And for the semisimple MV-algebras one has the following two characterizations.
Theorem 9.2.7. (i) An MV-algebra A is semisimple i it is a subdirect
product of subalgebras of the innite Lukasiewicz chain hW1 ; vel2 ; non1 ; 0i.
(ii) An MV-algebra A is semisimple i it is isomorphic to an algebra of
[0; 1]-valued continuous functions, i.e. to an algebra of fuzzy subsets of some
nonempty compact Hausdor space.
Relations to other algebraic structures. As algebraic structures, MValgebras are closely related to some other types of structures. The most prominent examples are the bounded commutative BCK-algebras and the abelian
lattice-ordered groups with order unit.
Denition 9.2.7. A bounded commutative BCK-algebra is such an algebraic structure B = hB; ; 0; 1i of similarity type h2; 0; 0i which satises for
all a; b; c 2 B the equations
(a b) c = (a c) b ; a (a b) = b (b a) ;
a a = 0;
a 0 = a; a 1 = 0:
In this case one has the following relationship between both types of structures.
Theorem 9.2.8. (i) If hA; ; :; 0i is an MV-algebra, then the algebraic structure hA; ; 0; :0i with the operation dened as a b =def a:b is a bounded
commutative BCK-algebra.
(ii) If B = hB; ; 0; 1i is a bounded commutative BCK-algebra, then the
structure hB; ; :; 0i with the operations :; dened as :a =def 1 a and
a b =def 1 ((1 a) b) is an MV-algebra.
(iii) These mappings between the classes of MV-algebras and of bounded
commutative BCK-algebras are inverse to one another and hence bijections.
Denition 9.2.8. An algebraic structure hG; +; o; 6; ui is an `-group with
order unit i hG; +; o; 6i is an `-group, such that the order unit o 6 u 2 G
has the property that for each a 2 G there is an integer n 0 with a 6 nu
and also ;a 6 nu for the inverse ;a of a characterized by a + (;a) = o.
That these `-groups with order unit are closely tied to MV-algebras in
general was found by D. Mundici [387] who proved the following result.
234
Theorem 9.2.9. The operator ; dened by (9.67) maps abelian lattice-
ordered groups with strong order unit to MV-algebras and is even a natural equivalence between the categories of abelian lattice-ordered groups with a
strong order unit and of MV-algebras (both with the respective natural homomorphisms as morphisms).
Besides its algebraic content, this last result may be read as stating that
MV-algebras are not only important for Lukasiewicz's many-valued logics
and interesting algebraic objects but also provide an equational formulation
of the theory of magnitudes with an Archimedean unit.
This operator ; is also of interest because of its relations to injective MValgebras which form another algebraically interesting22 class of MV-algebras.
Denition 9.2.9. An MV-algebra A is injective i each MV-homomorphism
h : C ! A from some MV-subalgebra C of an MV-algebra B into A can be
extended to an MV-homomorphism from B into A.
For these injective MV-algebras the following interesting theorem was
proved by Gluschankof [196].
Theorem 9.2.10. An MV-algebra A is injective i it is isomorphic to an
MV-algebra of the form ; (G ) for some divisible abelian lattice-ordered group
G with strong order unit whose underlying lattice is complete. In the particular
case jAj [0; 1] one has even that A is injective i jAj = [0; 1], and also i
the underlying lattice of A is complete.
9.2.2 MV-algebras and axiomatizations of the L-systems

In the previous subsection it was shown that the relation of semantic equivalence leads from the Lukasiewicz systems L to MV-algebras via a suitable
denition of operations between the equivalence classes. Now a second, closely
related way is considered which leads from logical calculi for the L-systems
to MV-algebras.
For the language LL of the L-systems we consider a logical calculus K L
which generates ws of LL and which shall be based on the axiom schemata:
(L1 1) H1 !L (H2 !L H1 ) ,
(L1 2) (H1 !L H2 ) !L ((H2 !L H3 ) !L (H1 !L H3 )) ,
(L1 3) (:H2 !L :H1 ) !L (H1 !L H2 ) ,
(L1 4) ((H1 !L H2 ) !L H2 ) !L ((H2 !L H1 ) !L H1 ) ,
and the rule of detachment (MP) as inference rule; cf. Theorem 9.1.5. By `L
the notion of derivability for the calculus K L is denoted.
22 The interest in this class of MV-algebras is not a purely algebraic one as it
indicates a compactness theorem for fuzzy logic mentioned later on in Section 19.
235
Lemma 9.2.4. In the set LL of all ws let a binary relation be dened by

the stipulation
H
(9.71)
= G =def `L (H !L G) ^ `L (G !L H ) :
This relation
= is an equivalence relation23 in LL . If one furthermore denes
in the quotient LL =
= representativewise operations ; : between equivalence
classes according to
[H ]= [G]= =def [:H !L G]= ;
(9.72)
:[H ]= =def [:H ]= ;
(9.73)
then these operations are suitably dened. With the particular equivalence
class
0 =def [:(p0 !L p0 )]=
(9.74)

nally the algebraic structure L = hLL = =; ; :; 0i becomes an MV-algebra,
the Lindenbaum algebra of the logical calculus K L .
Proof: Let A; B; C be any ws. By denition, = is a symmetric relation.
And it is also transitive because (L1 2) yields via two applications of (MP)
the fundamental property
A !L B; B !L C `L A !L C ;
(9.75)
which shall be used further on in this proof without explicit reference.
For the proof that
= is re exive some more steps are needed. By (L1 4)
one has
`L ((A !L B ) !L B ) !L ((B !L A) !L A)
(9.76)
and thus, by symmetry of the axiom schema (L1 4) w.r.t. H1 ; H2 , even
((A !L B ) !L B )
(9.77)
= ((B !L A) !L A) :
And also this type of argument which leads from (9.76) to (9.77) shall more
often be used further on, and only roughly indicated.
Choosing in the axiom schema (L1 1) the w B !L A for H2 , and the w
A for H1 gives `L A !L ((B !L A) !L A), and gives via (9.77) and (9.75)
also `L A !L ((A !L B ) !L B ). Hence an application of (MP) to (L1 2)
gives `L (((B !L C ) !L C ) !L (A !L C )) !L (B !L (A !L C )), if one
takes B for H1 , (B !L C ) !L C for H2 , and A !L C for H3 . And (L1 2)
gives also
`L (A !L (B !L C )) !L (((B !L C ) !L C ) !L (A !L C )) :
Hence one has via (9.75):
23 Because of its denition (9.71) this relation usually is called the relation of prov-
able equivalence. It is, obviously, the syntactic analogue to the relation of semantic equivalence.
236
`L (A !L (B !L C )) !L (B !L (A !L C )) ;
(9.78)
and thus immediately also

(A !L (B !L C ))
(9.79)
= (B !L (A !L C )) :
By choosing C A in (9.78) and using (L1 1) for an application of (MP),
one has `L (B !L (A !L A)). And choosing B as any axiom gives
`L (A !L A) :
(9.80)
Thus
= is also a re exive relation, and hence an equivalence relation.
From (9.78) and the axiom schema (L1 1) one has immediately
`L (B !L C ) !L ((A !L B ) !L (A !L C )) :
(9.81)
Thus the particular case `L ::A !L (::B !L ::A) of (L1 1) yields via
(L1 3), rst `L ::A !L (:A !L :B ), and then `L ::A !L (B !L A). And
because of (9.78), which holds for any ws A; B; C , one therefore even gets
`L B !L (::A !L A). So, choosing again some axiom for B , one has
`L (::A !L A) :
(9.82)
With ::A for H1 , A for H2 , and :B for H3 this gives via an application of
the rule of detachment `L (A !L :B ) !L (B !L :A), and hence one also
has
(A !L :B )
(9.83)
= (B !L :A) :
Choosing here B :A gives `L (:A !L :A), such that because of (9.80) an
application of (MP) leads to `L (A !L ::A). Hence one has by (9.82)
A
(9.84)
= ::A :
Now we are able to prove that the denitions (9.72), (9.73) are suitable
in the sense that they are independent of the representatives. From axiom
schema (L1 2) one has (A !L B ) `L (B !L C ) !L (A !L C ) and, of course,
also (B !L A) `L (A !L C ) !L (B !L C ). Therefore one even has:
if A
(9.85)
= B then (A !L C )
= (B !L C ) .
And similarly one gets, using (9.81) instead of (L1 2):
if A
(9.86)
= B then (C !L A)
= (C ! L B ) .

By (9.84) and (9.86) one has furthermore (A !L B ) = (A !L ::B ). And
substituting :B for B in (9.83) gives (A !L ::B )
= (:B !L :A). All
together this gives
if A
(9.87)
= B then :A
= :B .
Therefore denition (9.73) is a suitable one.
Now applications of (9.85) and (9.87) give
if H
= H 0 then [H ]= [G]= = [H 0 ]= [G]= ,
237
and applications of (9.85) and (9.86) give

if G
= G0 then [H ]= [G]= = [H ]= [G0 ]= .
Therefore also denition (9.72) is a suitable one.
Thus now the conditions (i), (ii), (iii) of Denition 9.2.1 have to be shown
for L together with the commutativity and associativity of and the property, that one has a 0 = a for all a 2 jLj.
For the commutativity a b = b a for all a; b 2 jLj it is necessary and
sucient to prove
(:A !L B )
= (:B !L A) for all ws A; B of LL .
But one has, using (9.83), (9.84), and (9.85), successively for any ws A; B :
(:A !L B )
= (:A !L ::B )
= (:B !L ::A)
= (:B !L A) :
Hence is commutative.
For the associativity a (b c) = (a b) c it is, because of this commutativity, sucient to prove
a (c b) = c (a b) for all a; b; c 2 jLj .
And this condition is equivalent to have
:A !L (:C !L B ) = :C !L (:A !L B ) for all ws A; B; C of LL .
This is, however, nothing but a particular case of (9.79). Hence is also
associative.
For the condition a 0 = a one has similarly to show that one has for all
ws A of LL
:A !L :(p0 !L p0 ) = A :
But according to (9.83), (9.84), and (9.85) one has
:A !L :(p0 !L p0 ) = (p0 !L p0 ) !L ::A = (p0 !L p0 ) !L A ;
hence it remains to be shown (p0 !L p0 ) !L A
= A.
By denition (9.71) of
= this means that one needs on the one hand
`L A !L ((p0 !L p0 ) !L A) ;
which is an immediate consequence of (L1 1), and needs on the other hand
`L ((p0 !L p0 ) !L A) !L A :
(9.88)
But from (9.80) one has `L ((p0 !L p0 ) !L A) !L ((p0 !L p0 ) !L A) and
hence by (9.78) also
`L (p0 !L p0 ) !L (((p0 !L p0 ) !L A) !L A) :
Thus (9.88) follows from `L (p0 !L p0 ).
Therefore L is an abelian monoid with neutral element 0, and it remains
to discuss the additional conditions (i), (ii), (iii).
238
However, (i) is obvious from (9.84). And for (iii) one has to show according
to the denitions (9.72) and (9.73):
::(::A !L B ) !L B = ::(::B !L A) !L A ;
which because of (9.84), (9.85), and (9.86) immediately follows from (9.76).
And the last condition (ii) will have been shown if we succeed to prove
that one always has
if `L B then A !L B
(9.89)
=B ,
0
0
because (ii) is the particular case B (p !L p ).
To get (9.89) one needs `L B !L (A !L B ), which is obvious because of
(L1 1). And one needs on the other hand `L (A !L B ) !L B , which results
from `L B !L ((A !L B ) !L B ), again a particular case of (L1 1), and from
`L B .
2
Remark: The last proof shows also that, because of (9.84), (9.85), (9.86),
and (9.87), one can at each position inside some w H of LL exchange any
subformula A by ::A, and vice versa exchange any subformula ::A by A,
such that the result of such exchanges always is provably equivalent with H .
Theorem 9.2.11 (Completeness Theorem for L1). Each L1-tautology
is derivable from the axiom schemata (L1 1), . . . , (L1 4) via the inference rule
(MP).
Proof: One has that to each w H of the language LL, whose implicational subformulas C !L B have always negated ws as antecedents, i.e.
are of the form :A !L B , there corresponds uniquely some translation TH
which is a term of the language of MV-algebras, and which is determined by
the recursive clauses:
8
if H is the propositional variable pi
< xi ;

[
T
]

[
T
]
;
TH =def : A = B = if H :A !L B
:[TA ]= ;
if H :A .
By this translation, H is an L1 -tautology i the MV-equation TH = 1 is
valid in the MV-algebra ; (R; 1) = hW1 ; vel2 ; non1 ; 0i.
Because one has in each one of the L-systems
j=L (A !L B ) $ (::A !L B );
(9.90)
each w of the language LL can semantically equivalent be written as a w
in which all subformulas of the form C !L B have a negated antecedent:
C :A. Let us call such a particular w of LL implication normalized.
Therefore to each L1 -tautology there corresponds a suitable valid equation
TH = 1 in ; (R; 1).
We now nd that each valid equation E of the MV-algebra ; (R; 1) is also
a valid equation of the Lindenbaum algebra L of the logical calculus K L ,
introduced in Lemma 9.2.4. For, according to Theorem 9.2.2 one has that L
239
is isomorphic to a subdirect product of MV-chains Ai . And by Proposition

9.2.5 one has that each valid equation E of ; (R; 1) is also a valid equation
in all these \factors" Ai . Therefore E is also a valid equation in the direct
product of all these MV-chains Ai , and hence also in the subdirect product
L, cf. e.g. [96, 225].
So let H be any L1 -tautology. Suppose that H is transformed into a semantically equivalent implication normalized w H . Of course, H is an L1 tautology. According to the previous observations, to H there corresponds
the MV-equation TH = 1 which is valid in ; (R; 1). But then TH = 1 is
also valid in the Lindenbaum algebra L, which means [H ]= = 1, i.e. `L H .
Now all the steps which had lead from the w H to the w H can be reversed, using of course provable equivalence instead of semantical equivalence,
according to the remark preceding this theorem. This gives `L H .
2
Because all those ws are L1 -tautologies which are K L -axioms, i.e. which
fall under one of the axiom schemata (L1 1), . . . , (L1 4), this calculus K L gives
an adequate axiomatization24 of the innite-valued L-system L1 . The next
goal now is to get in a similar way25 also adequate axiomatizations for the
nitely many-valued L-systems Lm . Fortunately, this is possible with only a
small modication of the proof which was given for Theorem 9.2.11, cf. [227].
For each integer m 3 we understand by an MVm -algebra any MValgebra A which additionally satises for each a 2 A the conditions:
(MVm 1) (m ; 1)a a = (m ; 1)a ;
(MVm 2) am;1
a = am;1 ;
and for each integer 1 < k < m ; 1 which does not divide m ; 1 also the
conditions:
(MVm 3) [ka
(:a :(k ; 1)a)]m;1 = 0 ;
(MVm 4) (m ; 1)[ak :a
:ak;1 ] = 1 :
Obviously, here the conditions (MVm 3) and (MVm 4) become empty in the
case m = 3.
We rst look for natural examples for MVm -algebras and prove that the
particular MV-algebra Wn , n 3, is an MVm -algebra i n ; 1 divides m ; 1.
Suppose rst that n ; 1 divides m ; 1. Then one has Wn Wm according
to Theorem 9.1.2(b), which means that Wn is an MVm -algebra if Wm is an
24 This axiomatization does not, however, provide via its standard extension an
adequate axiomatization of the corresponding entailment relation. A geometrical

interpretation of this fact was e.g. given in [388].
25 This means that we are not interested in the Rosser-Turquette type of axiomatization here, but like to have a much shorter list of axiom schemata which,
preferably, should be similar to the list of axiom schemata of the calculus K L .
Depending on the actual proof strategy for the completeness theorem, also other
sets of axioms are suitable. An example is [483] where the proof strategy is not
the algebraic one used here, but refers to maximally consistent sets of ws.
240
MVm -algebra. However, Wm is an MVm -algebra. Surely it is an MV-algebra.

Obviously one has also (m ; 1)a = 1 for each 0 6= a 2 Wm as well as bm;1 = 0
for each 1 6= b 2 Wm , which means that (MVm 1) and (MVm 2) are satised
for Wm . And because Wm is an MV-algebra, one gets (MVm 4) from (MVm 3)
via (MV6), (MV 6). Hence only (MVm 3) remains to be discussed. The crucial
point is that one has
(9.91)
ka
(:a :(k ; 1)a) = 1 , a = 1 :
To see this remember that, similar to (9.22) and (9.23), one has in the present
situation ak = maxf0; k a ; (k ; 1)g and ka = minf1; k ag (with ; ; here
for the usual arithmetical operations), and that in each such MV-algebra Wn
one has: a b = 1 i a = b = 1. Because of
ka + 1 , a k1 and :a :(k ; 1)a = 1 , a k1
one thus has (9.91). Even more: (9.91) holds in each MV-algebra Wn . For
any k 2 N with 1 < k < m ; 1 which is not dividing m ; 1 one, however,
has k1 2= Wm , and therefore also ka
(:a :(k ; 1)a) < 1. And this gives
[ka
(:a :(k ; 1)a)]m;1 = 0 and thus (MVm 3).
So suppose conversely that Wn is an MVm -algebra and that n ; 1 does
not divide m ; 1. Then one has (MVm 3) for Wn and can particularly choose
k = n ; 1. This gives
(n ; 1)a
(:a :(n ; 2)a) 6= 1
for all a 2 Wn , and thus particularly for a = n;1 1 2 Wn , contradicting (9.91).
Hence one has
Wn MVm -algebra , n ; 1 divides m ; 1 .
(9.92)
Lemma 9.2.5. Suppose that I is a prime ideal of an MVm-algebra A. Then
the quotient structure A=I is isomorphic to an MV-algebra Wn such that
n ; 1 divides m ; 1.
Proof: Let A be an MVm-algebra and I one of its prime ideals. It is
immediately clear that the quotient structure A=I has to be an MVm -algebra.
Because I is a prime ideal, A=I is even an MV-chain.
We rst claim that A=I has an atom, i.e. an element a 6= 0 such that for
each b 2 A=I with 0 5 b 5 a either b = 0 or b = a holds. For, if such an
atom does not exist, then for each a 2 A=I with 0 < a < 1 there would exist
some b 2 A=I such that (m ; 1)b 5 a. In the case m = 2 this is obvious. And
for m > 2 we show by induction that for each integer k there exists some
0 6= b 2 A=I such that kb 5 a. As induction hypothesis, assume 0 6= c 2 A=I
and kc 5 a. Because A=I is supposed to contain no atom there exists some
e 2 A=I such that 0 < e < c. And one has d(e; c) 6= 0 for the function d
dened in (9.62). But it is even 2e 5 c or 2d(e; c) 5 c: because one gets from
2e 65 c rst c 5 2e = ee, thus also 1 = :cee = :(c
:e)e, and therefore
241
d(e; c) = :e
c 5 e and also 2d(e; c) 5 e d(e; c) = e :(e
c) = c t e = c.
Thus there exists some b 6= 0 such that (k + 1)b 5 2kb 5 k(2b) 5 kc 5 a. In
the case that A=I does not have an atom thus some a 2 A=I with 0 < a < 1
shall be xed, and an element b 2 A=I chosen such that (m ; 1)b 5 a. Then
the set fx 2 A=I j x 5 (m ; 1)bg is a nontrivial proper ideal of A=I because
of (MVm 1), contradicting the fact that I is a maximal ideal of A and that
therefore A=I can have only the trivial ideals.
Thus the quotient structure A=I contains an atom a. And it is also na = 1
for some integer n m ; 1, because otherwise again fx 2 A=I j x 5 (m ; 1)ag
would be a nontrivial proper ideal of A=I according to (MVm 1). Choose
furthermore n as small as possible. This atom a gives a chain
0 < a < 2a < : : : < (n ; 1)a < na = 1 :
Obviously one has in this chain always ka 5 (k + 1)a. However, if for some
k < n one would have ka = (k +1)a, then one would also have ka = (k +1)a =
ka a = (k +1)a a = (k +2)a = : : : = na = 1, contradicting the minimality
property of n.
Because A=I is a chain w.r.t. its lattice ordering 5, there exists for each
b 2 A=I an integer k < n such that ka < b < (k + 1)a. Then one has
ka
:b = 0 because of ka 5 b, and b
(:a)k+1 = 0 because of b 5 (k +1)a, and
hence b
(:a)k
:a = 0, i.e. d(b; ka) 5 a. However, in the case d(b; ka) = a
one would have b
(:a)k = a and b = b0 = b(ka
:b) = ka(b
(:a)k ) =
ka a = (k + 1)a, a contradiction. Therefore one has d(b; ka) < a, which
means d(b; ka) = 0 and hence b = ka. Thus each b 2 A=I is a multiple of the
atom a.
Obviously one has ka ra = (k + r)a for all integers k; r. But one has
also :(ka) = (n ; k)a for each 0 k n: For k = n this is obvious.
And for k < n one has (n ; k)a ka = 1, and thus :(ka) 5 (n ; k)a.
But :(ka) 5 (n ; k ; 1)a would give 1 = (n ; k ; 1)a ka = (n ; 1)a,
contradicting the choice of n. Hence the mapping ' which is dened by the
equation '(ka) = nk is an isomorphism from A=I onto Wn+1 . Thus Wn+1 is
an MVm -algebra, and therefore one has by (9.92) that n divides m ; 1. 2
Theorem 9.2.12. Each MVm-algebra is isomorphic to a subdirect product

of MV-algebras Wn for which n ; 1 divides m ; 1.
Proof: As in the proof of Theorem 9.2.2 one shows that each MVmalgebra A is a subdirect product of the family of all MVm -algebras A=I for
the prime ideals I of A. Then the claim follows immediately from Lemma
9.2.5.
2
Theorem 9.2.13 (Axiomatizability Theorem for Lm). For each integer
m 3 an adequate axiomatization of the m-valued Lukasiewicz system Lm
is given by the axiom schemata
(Lm 1) H1 !L (H2 !L H1 ) ,
242
(Lm 2) (H1 !L H2 ) !L ((H2 !L H3 ) !L (H1 !L H3 )) ,

(Lm 3) (:H2 !L :H1 ) !L (H1 !L H2 ) ,
(Lm 4) P
((H1 !L H2P
) !L H2 ) !L ((H2 !L H1 ) !L H1 ) ,
m
(Lm 5)
A !L mi=1;1 A ,
i
=1
;
Pm;1 Qk
Pk;1
(Lm 6)
for each 1 < k < m ; 1 for
i=1
j =1 AY :A & i=1 A
which k ;1 does not divide m;1
together with the rule of detachment (MP) w.r.t. the implication !L as inference rule.
Proof: It is easy to check that each of these axioms is an Lm-tautology.
Therefore this system of axioms constitutes together with the inference rule
(MP) a sound logical calculus for Lm . What remains to be proved is that each
Lm -tautology is a theorem of this logical calculus.
As in Lemma 9.2.4 one forms the Lindenbaum algebra L(Lm ) of this
calculus, which now even proves to be an MVm -algebra. First one gets from
Lemma
L(Lm)is an MV-algebra. However, because one also has
P 9.2.4 thatP
m
;
1
`L i=1 A !L mi=1 A for each w A of LL , one has (m ; 1)a 5 ma for
P
Pm;1
each a 2 L(Lm ). Together with the fact `L m
i=1 A !L i=1 A , which
is just a restatement of axiom schema (Lm 4), one thus has (MVm 1). Then
(MVm 2) is an immediate corollary, which one e.g. gets via the deMorgan
laws (MV6); (MV 6). Furthermore one has (MVm 4) immediately from (Lm 6),
and thus gets (MVm 4) again via (MV6), (MV 6).
Suppose now that H0 is a w of LL , which is not derivable from these
axiom schemata (Lm 1); : : : ; (Lm 4) using (MP). Then one has [H0 ]= 6= 1 in
L(Lm). Let ' be an isomorphism from L(Lm ) onto a suitable subdirect product of MVm -algebras Wn ; Wn ; : : :. Then
'(H0 ) = (a1 ; a2 ; : : :)
is a sequence of elements ai 2 Wni : And because of [H0 ]= 6= 1 there exists
an index k such that one has ak 6= 1 as well as ak 2 Wm for the k-th member
of this sequence. Now a valuation ' : V0 ! Wm shall be determined by
' (p) = k-th member of the sequence '([p]= ) .
Then one has ValL (H0 ; ' ) = ak 6= 1, i.e. H0 is not an Lm -tautology. Hence
also completeness is proved.
2
1
9.2.3 Wajsberg algebras

The MV-algebras as introduced in the Denition 9.2.1 in a natural way refer
to a formulation of Lukasiewicz logic in a language with negation : and
strong disjunction Y as basic connectives, completed by the use of the truth
degree constant 0. And in their original, more complicated denition, given
by the characterization of Proposition 9.2.2, one even refers to a formulation
243
of this logic in a language with all the connectives &; Y; ^; _; : as basic ones,
completed by the use of the truth degree constants 0; 1.
The historically rst, and still the most simple axiomatization of the innitely many-valued system L1 which is given, according to Theorem 9.2.11,
by the logical calculus K L with the axiom schemata L1 1, . . . , L1 4, cf. p. 234,
uses instead as basic connectives only the Lukasiewicz implication !L together with the negation :.
So it is natural to ask for algebraic structures which correspond more
directly to the logical calculus K L , e.g. in the sense that the Lindenbaum
algebra of K L is an algebraic structure of that type. Such structures have
been introduced in [171] and considered e.g. in [246, 550].
Denition 9.2.10. An algebraic structure B = hB; ; ; 1i is a Wajsberg
algebra i there hold for all x; y; z 2 B :
1 x = x ;
(9.93)
;
(x y) (y z ) (x z ) = 1 ;
(9.94)
(x y) y = (y x) x ;
(9.95)

(x y ) (y x) = 1 :
(9.96)
If one compares the characterising equations (9.94), . . . , (9.96), one recognizes that they immediately correspond to the axiom schemata L1 2, . . . ,
L1 4. Therefore it is a routine matter to prove the following result.
Corollary 9.2.9. The Lindenbaum algebra of the logical calculus K L is a
Wajsberg algebra.
Proof: Instead of deriving suciently many ws in K L , we refer to the
Completeness Theorem 9.2.11. Then the examples of L-tautologies from Section 9.1 immediately give that by
H1 H2 =def j= H1 $L H2 ;
or equivalently by
H1 H2 =def `KL H1 $L H2
an equivalence relation is dened in the class of all ws of the language of
the L-systems, which is the basic equivalence relation for the Lindenbaum
algebra LL of the logical calculus K L .
Now take 1 as the -equivalence class of any L1 -tautology, i.e. as the
class of all K L -theorems. Then (L1 2) and (L1 4) obviously give (9.94) and
(9.96) in LL .
That (9.93) holds true in LL results from the facts that for each K L theorem H one can derive G from H !L G in K L , and that one can also
derive H !L G from G via (L1 1).
244
Finally one gets (9.95) immediately from (9.4) and the tautology (T1)
mentioned on p. 182.
2
The Wajsberg algebras could have been used in the previous section instead of the MV-algebras. The fact which makes this exchange possible is that
both types of algebraic structures are interdenable, and that there is even a
one-one correspondence between Wajsberg algebras and MV-algebras. However, the MV-algebras are not only the structures which have { historically
{ rst been used for an algebraic completeness proof for L1 , they actually
seem to be the more interesting class of algebraic structures.
Theorem 9.2.14. Let A = hA; ; :; 0i be an MV-algebra and dene
x y =def :x y ;
1 = :0 :
Then hA; ; :; 1i is a Wajsberg algebra.
Proof: First one has (9.93) via (MV5) because of
1 x = :1 x = 0 x = x :
Furthermore one has, using condition (iii) of Denition 9.2.1 and (MV3),
(MV4):
;

(x y) (y z ) (x z )
;

= :(x y) :(y z ) (x z )
;

= :(:x y) :(:y z ) (:x z )
;
;

= :(:x y) :x :(:y z ) z
;
;

= :(::y :x) :x :(:y z ) z
;
;

= :(::x :y) :y :(:z y) y
= : : : :y y = : : : 1 = 1 ;
and therefore (9.94).
Again using condition (iii) of Denition 9.2.1 one gets immediately also
(9.95):
(x y) y = :(:x y) y = :(:y x) x = (y x) x :
And nally condition (9.96) easily follows from
y x = :y x = x :y = ::x y = :x :y
together with the fact that always z z = :z z = 1.
2
Theorem 9.2.15. Let B = hB; ; ; 1i be a Wajsberg algebra and dene

x y =def x y ;
0 =def 1 :
Then hB; ; ; 0i is an MV-algebra.
245
Proof: We start with an auxiliary result and show that a re exive ordering relation is dened in B by
x 5 y =def x y = 1 :
(9.97)
;

This relation is re exive because one has (1 1) (1 x) (1 x) = 1
from (9.94), and therefore also 1 (x x) = 1 from (9.93). Now (9.93) gives
also (x x) = 1.
To see that this relation is transitive, assume x 5 y and y 5 z , i.e. assume
x y = 1 and y z = 1. Then (9.94) gives, together with two applications
of (9.93), that x z = 1, i.e. that y 5 z .
Finally assume for antisymmetry that x 5 y and y 5 x, i.e. that x y = 1
and y x = 1. Then (9.95) gives 1 x = 1 y, which means x = y by
(9.93).
This ordering relation has 1 as its greatest element because one has always
x 1 = 1: and this results from the fact that one has
(x 1) 1 = (1 x) x = x x = 1
from (9.95) and (9.93), i.e. that one has (x 1) 5 1 together with
1 (x 1) = x 1 = x ((x 1) 1)
= (1 x) ((x 1) (1 1)) = 1 ;
i.e. together with 1 5 (x 1).
And the operation is, w.r.t. 5, non-increasing in its rst argument,
i.e. one has in the case of x 5 y, i.e. in the case of x y = 1, immediately
(y z ) (x z ) = 1 from (9.94). Thus
x5y ) yz 5xz
(9.98)
results. Therefore one has also
x (y x) = 1 ;
(9.99)
which can, because of y 5 1, equivalently be written as the inequality
x = 1x 5 y x:
(9.100)
It immediately implies together with (9.96) and (9.93) also
1 5 x 1 5 1 x = x :
(9.101)
Now one can prove that one has the representation
x = x 1 :
(9.102)
On the one hand the inequality x 5 (1 ) x 5 x 1 follows from (9.100)
and (9.95). On the other hand one has from (9.93) and (9.95)
x 1 5 1 x = x ;
and hence from (9.98) and (9.95)
246
x 1 5 (x 1 ) 1 = (1 x ) x :
However, one has in general
u 5 v w ) v 5 uw;
(9.103)
because in the case of u 5 v w one has (v w) w 5 u w and hence
v 5 (w v) v = (v w) w 5 u w.
So in the present case one gets

(1 x ) 5 (x 1 ) x ;
which means x 1 5 x because (9.101) gives 1 x = 1. And this nally
establishes (9.102).
As a further consequence one has now
x = (x 1 ) 1 = (1 x) x = 1 x = x ;
(9.104)
which is just condition (i) of Denition 9.2.1.
Rewriting (9.96) as x y 5 y x makes it easy to see that (9.104) gives
also y x 5 x y , which means that one even has
x y = y x :
(9.105)
And this gives immediately the commutativity of by
x y = x y = x y = y x = y x :
Moreover one easily nds
x 0 = x 1 = 1 x = x ;
x 0 = x 1 = x 1 = 1 x = 1 = 0 ;
and has therefore the neutrality of 0 under as well as condition (ii) of
Denition 9.2.1.
Also condition (iii) of Denition 9.2.1 now is a direct consequence of (9.95)
because one has
(x y) y = (x y) y = (x y) y
= (y x) x = (y x) x = (y x) x :
So it remains to show that is also an associative operation. This needs as
a preliminary step to derive the equation
x (y z ) = y (x z )
(9.106)
for Wajsberg algebras. Having (9.106) available, the associativity simply
results from
(x y) z = x (y z ) = x (z y) = z (x y)
= z (x y) = z (x y) = (x y) z :
To get (9.106), apply rst (9.103) to (9.94) having in mind (9.97), i.e. apply
(9.103) to the inequality u v 5 (v w) (u w). This gives
247
v w 5 (u v ) (u w ) :
From this together with (9.100) and (9.95) one gets

y 5 (z y) y = (y z ) z 5 (x (y z )) (x z ) ;
and again using (9.103) also
x (y z ) 5 y (x z ) :
By symmetry, i.e. by a similar argument simply exchanging the roles of x
and y one has
y (x z ) 5 x (y z ) :
and hence (9.106).
2
Corollary 9.2.10. There is a one-one correspondence between MV-algebras

and Wajsberg algebras determined either by the correspondence from Theorem 9.2.14, or by the correspondence from Theorem 9.2.15.
Proof: It is a routine matter to check that both possible iterations of the
two correspondences, studied in Theorems 9.2.14 and 9.2.15, give the identity
either over the class of MV-algebras, or over the class of Wajsberg algebras.
2
9.2.4 Lukasiewicz algebras

The MV-algebras and the Wajsberg algebras, as well as the MVm -algebras,
give suitable algebraic counterparts for the Lukasiewicz systems. They have
been introduced in [91, 171, 227], respectively. And they generalize the relationships one has between classical logic and Boolean algebras.
Another algebraic approach toward the Lukasiewicz systems of manyvalued logic, which diers considerably from the approach via MV-algebras or
Wajsberg algebras, was initiated by Moisil [371, 372]. He started from the
3-valued system L3 and the observation that, whilst it is impossible to dene
in this system !L from ^; _; :, this becomes possible if one adds a unary
connective M with truth degree function verLM = vel2 , i.e. with verLM (0) = 0
and verLM ( 12 ) = verLM (1) = 1. A denition of !L can then be inferred from the
fact that one immediately gets26
j=L (H !L G) $L :H _ G _ M(:H ^ G):
The generalization to the m-valued case would then consist in the addition
of n ; 1 new unary connectives. The algebraization of this idea resulted in
the following denition.
3
26 It is interesting to remark that on the other hand also this new connective M can
in the three-valued L-system be dened using the Lukasiewicz implication !L ,
e.g. as MH =def (:H !L H ). Thus M is just the modal operator \it is possible
that" which was discussed in the modal reading of the system L3 , cf. Chapter
21.
248
Denition 9.2.11. Given any 2 m 2 N, an m-valued Lukasiewicz al-
gebra is an algebraic structure

A = hA; [; \; ; sm1 ; : : : ; smm;1 ; 0; 1i
of similarity type h2; 2; 1; 1; : : :; 1; 0; 0i such that hA; [; \; 0; 1i is a distributive
lattice with zero and unit, and such that for all x; y 2 A and all 1 i; j
m ; 1 there hold true:
(LA1) ( (x)) = x ;
(LA2) (x [ y) = (x)^ (y) ;
m
m
(LA3) sm
i (x [ y ) = si (x) [ si (y ) ;
m
(LA4) sm
i (x)[ si (x) = 1 ;
m
m
(LA5) sm
j (si (x)) = si (x) ;
m
(LA6) sm
i ( (x)) = sm;i (x) ;
m
m
(LA7) si (x) [ si+1 (x) = sm
i+1 (x) for i < m ; 1 ;
m
m
(LA8) x [ sm;1 (x) = sm;1(x) ;
m
(LA9) (x\ (sm
i (x)) \ si+1 (x)) [ y = y for i < m ; 1 :
Here the conditions (LA8) and (LA9) could equivalently be replaced by
the condition
m
if sm
i (a) = si (b) for all i = 1; : : : ; m ; 1 then a = b .
Simple examples for m-valued Lukasiewicz algebras can be built up in
two dierent ways. The rst one is that one starts from the truth degree
set Wm together with the truth degree functions vel1 ; et1 ; non1 for [; \; ,
respectively. Then one can dene

i + t (m ; 1) m
m
si (t) =def 10; ifotherwise
,
and gets an m-valued Lukasiewicz algebra.
The second way is to start from a Boolean algebra B = hB; [; \; c ; 0; 1i
with lattice ordering 5, and to dene in the set
B [m] =def f(a1 ; : : : ; am;1 ) 2 B m j a1 5 a2 5 : : : 5 am;1 g
(9.107)
c
operations [; \; componentwise from the equally denoted operations of B ,
as well as the operations sm
i by
sm
i (a) =def (ai ; ai ; : : : ; ai )
for all a = (a1 ; : : : ; am;1 ) 2 B [m] and all i = 1; : : : ; m ; 1.
As for MVm -algebras, also for m-valued Lukasiewicz algebras one can
prove that each one of them is (isomorphic to) a subdirect product of a
suitable family of the particular m-valued Lukasiewicz algebra
hWm ; vel1 ; et1 ; non1 ; sm1 ; : : : ; smm;1 ; 0; 1i :
9.3 The First-Order Systems
249
For each m-valued Lukasiewicz algebra A the set B(A) of complemented

elements of A, i.e. the set
B(A) = fa 2 A j a[ a = 1g = fa 2 A j a\ a = 0g
forms a Boolean algebra w.r.t. the operations [; \; of A and the designated elements 0; 1. And from this Boolean algebra B(A) one again gets
as in (9.107) an m-valued Lukasiewicz algebra B(A)[m] , which proves to
be isomorphic to the m-valued Lukasiewicz algebra the whole construction
started from.
However, only for m = 3; 4 are these m-valued Lukasiewicz algebras
really algebraic counterparts to the m-valued Lukasiewicz system Lm . This
is so, because only within the Lukasiewicz algebras W3 and W4 one is able
to dene the corresponding Lukasiewicz implication seq 2 , but not within
the Lukasiewicz algebras Wm with m 5.
The way out, chosen in [101], was to enrich these m-valued Lukasiewicz
algebras with a family (Fi;j )(i;j)2Sm of further binary operations with index
set Sm = f(i; j ) j 3 i m ; 2 ^ 1 j m ; 4g for m 5, and to call
them proper m-valued Lukasiewicz algebras. These additional operations
have to satisfy for each 1 k m ; 1 the condition
8
>
>
<0
;
if k i ; j
;

m
m
m
sm
F
(
x;
y
)
=
k ij
sm;i (x)\ sm;i;1 (x) \ sm;j (y)\ sm
m;j ;1 (y ) ;
>
>
:
if k > i ; j .
These proper m-valued Lukasiewicz algebras are then also for m 5 true
algebraic counterparts of the m-valued L-systems.
We shall not go deeper into the theory of Lukasiewicz algebras here.
The theory of these structures has to a large extent been developed e.g. by
Cignoli [99, 100], and is in its present form given in [74].

The rst-order Lukasiewicz systems L extend their propositional counterparts, i.e. have the same sets W L = W of truth degrees and DL = f1g
of designated truth degrees. Also their sets of connectives and of truth degree constants coincide. Therefore all the rst-order Lukasiewicz systems
L have the same (formalized) language LL .
The crucial new item, besides the standard changes in the formalized
language, hence is the choice of quantiers. Usually these are a universal
quantier 8 and an existential quantier 9, both of arity (1; 1). The corresponding characterizations (2.22) by generalized truth degree functions are
just the characterizations (2.15), (2.16), i.e. one has
250
ValLA (8xH; ) =def inf fValLA (H; [x=b]) j b 2 Ag ;

(9.108)
ValLA (9xH; ) =def supfValLA (H; [x=b]) j b 2 Ag
(9.109)
for each interpretation A and each A-assignment . Obviously 8 is an innitary generalization of the weak conjunction ^, and 9 an innitary generalization of the weak disjunction _.
The fact that the set W0 of all rationals of the unit interval [0; 1] is not
closed under taking suprema or inma of its subsets would cause problems for
the further discussion of semantical topics, e.g. in the sense that one would
have to restrict the acceptable interpretations to such ones for which quantied formulas always would have their truth degree inside W0 , or one would
have quantied formulas with truth degree gaps. These diculties usually are
seen as inconveniences caused by mathematical properties of W0 but not as
facts which express some important meaning. Therefore one usually excludes
the (propositional) Lukasiewicz system L0 from the rst-order extension.
We follow this behavior and therefore restrict the range of the index in this
section to = 2; 3; : : :; 1. As in the propositional case = 2 is not excluded
and yields as L2 classical rst-order logic, and the index m shall range over
all integers m 2, usually.
In the rst-order case we hence consider just one innitely many-valued
Lukasiewicz system: L1 .
It is an immediate corollary from the denitions (9.108), (9.109) of the
quantiers 8; 9 that one has generalizations of the well known deMorgan
laws, connecting those quantiers in classical logic, also for the Lukasiewicz
biimplication $L introduced in (9.6).
Proposition 9.3.1. The following expressions are logically valid:

(i) j=L :8xH (x) $L 9x:H (x) ;
(ii) j=L :9xH (x) $L 8x:H (x) ;
(iii) j=L 8xH (x) $L :9x:H (x) ;
(iv) j=L 9xH (x) $L :8x:H (x) :
Proof: Straightforward from the corresponding denitions.
2
Therefore it always suces to have one of the two quantiers 8; 9 present
in the language of L ; the other one is denable. Occasionally we shall refer
to this fact.
9.3.1 Important logically valid formulas

The way the rst-order Lukasiewicz systems L are constructed by extending their propositional counterparts immediately yields that tautologies
\remain" logically valid:
251
Proposition 9.3.2. Every substitution instance H of an L -tautology, i.e.

every w of the rst-order Lukasiewicz systems L which results from an
(propositional) L -tautology H by substituting ws of the rst-order language
LL for the propositional variables of H , is logically valid for L .
It is an easy exercise to see, on the other hand, that any propositional
schema (formulated with metavariables for ws) which is not a propositional
tautology for L , also does not give a logically valid rst-order schema for
L . One knows that there is some truth degree assignment to the propositional metavariables which gives the whole w a non-designated truth degree;
the same assignment, however, can be realized via some suitable rst-order
interpretation together with some suitable assignment.
All the particular (schemata of) tautologies mentioned in Section 9.1 for
the propositional case are, thus, also examples for logically valid ws for the
(corresponding) rst-order systems L . So we only continue the previous list
with examples of a truly rst-order nature, i.e. essentially involving quantiers. The most basic properties concern the order of the quantiers in the
prex of a w. This order is unimportant for quantiers of the same type
because of
(T25) j=L 8x8yH $L 8y8xH ,
(T26) j=L 9x9yH $L 9y9xH .
This order of the quantiers is important, however, for quantiers of dierent
type. One has
(T27) j=L 9x8yH !L 8y9xH .
The converse implication to this one, however, fails to be logically valid, like
in classical logic.27
Quantiers distribute over (their corresponding) weak conjunctions and
weak disjunctions in the sense that one has
(T28) j=L 8x(H1 ^ H2 ) $L 8xH1 ^ 8xH2 ,
(T29) j=L 9x(H1 _ H2 ) $L 9xH1 _ 9xH2 .
For the strong versions of conjunction and disjunction one has only the subdistributivities
(T30) j=L 8xH1 & 8xH2 !L 8x(H1 & H2 ) ,
(T31) j=L 9x(H1 Y H2 ) !L 9xH1 Y 9xH2 ,
27 The proof of this fact follows here and in later, corresponding situations essen-
tially the very same ideas as in the corresponding classical cases. Therefore the
reader is asked to complete the arguments for himself. Moreover, we always intend to present the strongest result, which means that in all the cases where
some implication is given as logically valid, the corresponding biimplication is
not logically valid, i.e. the converse implication fails. Usually this shall not be
mentioned in every case.
252
with the converse implications failing to be logically valid. Additionally one

has for 2 f^; &g the subdistributivities
(T32) j=L 9x(H1 H2 ) !L 9xH1 9xH2 ,
and for 2 f_; Yg always
(T33) j=L 8xH1 8xH2 !L 8x(H1 H2 ) .
And the universal quantier also subdistributes over Lukasiewicz implication:
(T34) j=L 8x(H1 !L H2 ) !L (8xH1 !L 8xH2) .
Therefore one gets from (T30) and (T10) also the particular rst-order version
of the rule of syllogism
(T35) j=L 8x(H1 !L H2 ) & 8x(H2 !L H3 ) !L (8xH1 !L 8xH3 ) .
Denoting again for some individual symbol a by H [x=a] the w which
results from H by substituting for the individual variable x at all places of
its free occurrence the individual symbol a, one has under the additional
assumption that there is no place of free occurrence of x inside the scope of
a quantication of (the variable) a:28
(T36) j=L 8xH !L H [x=a] ,
(T37) j=L H [x=a] !L 9xH ,
and thus at all also
(T38) j=L 8xH !L 9xH .
Assuming additionally that G is a w of LL which does not contain the
variable x free, then one has as further examples of logically L-valid formulas:
(T39) j=L 8x(H !L G) $L (9xH !L G) ,
(T40) j=L 8x(G !L H ) $L (G !L 8xH ) ,
(T41) j=L 9x(H !L G) $L (8xH !L G) ,
(T42) j=L 9x(G !L H ) $L (G !L 9xH ) .
In each of these cases, the proof that these formulas are logically L-valid
derives from (9.108), (9.109) and well known facts of Lukasiewicz propositional logic.
The laws (T39), . . . , (T42) for quantier shifts allow together with Proposition 9.3.1 and (T9) to nd, in the same way as in classical logic, to each
w H 2 LL a semantically equivalent w in prenex normal form. As usual,
a w H is said to be in prenex (normal) form if it has the form H
28 Having in mind that renaming of bound variables inside a w H always results
in a w H 0 which is semantically equivalent with H and that the equivalence
theorem holds true, this restriction can be avoided assuming that one has a
liberalized notion of substitution which allows the renaming of bound variables,
particularly in all quantications of the variable a.
253
Q1 x1 : : : Qk xk G0 with a quantier free w G0 , i.e. if it has all its quantications in front of a purely propositional matrix.
By the way, the transformations of each w H 2 LL into a semantically
equivalent prenex form remains possible in the very same way also if one enriches the language LL with further connectives, as e.g. discussed later on in
Chapter 13, as long as the truth degree functions for these additional connectives are monotonically non-decreasing in each argument { and continuous in
the case of the system L1 .
9.3.2 Theoretical results for the L-systems

For the nitely many-valued Lukasiewicz systems Lm with m 2 all the
results mentioned in Section 3.4 are true, i.e. one has e.g. that:
the ultraproduct construction is available;
the compactness theorem holds true;
the downward as well as the upward Lo wenheim-Skolem theorems hold
true.
From Section 9.1 it is known that in each one of the Lukasiewicz systems
Lm all the unary connectives Jt for t 2 Wm are denable. The applicability
of the Rosser-Turquette method of axiomatization to these L-systems
therefore depends on the availability of a conjunction, a negation, and a
universal quantier all satisfying the corresponding standard conditions. It is
immediately clear that the universal quantier 8 of the language LL satises
the standard condition because the truth degree 1 is the only designated
truth degree, and this means that the general condition (6.74) reduces here
to the simple condition
B1 (8xH ) , 8 xB1 (H )
which immediately results from denition (9.108). However, for m > 2 neither
satises !L the standard condition for an implication connective, nor : the
standard condition for a negation. Fortunately it is not dicult to dene
other connectives ! and : in Lm which satisfy the corresponding standard
conditions. A suitable implication connective is given by
H1 ! H2 =def J1 (H1 ) !L H2
(9.110)
which according to (9.27) and (T7) is equivalent to the denition
m;1
H1 ! H2 =def ! (H1 ; H2 )
=1
with ! dened as in (6.1), (6.2), here w.r.t. the implication connective !L .

Equally simple, one can dene : as
: H =def :J1 (H ) :
(9.111)
254
Hence each one of the nitely many-valued rst-order systems Lm can be

adequately axiomatized by a corresponding logical calculus K m
RT with the
axiom schemata (AxRT 1); : : : ; (AxRT 11) and the inference rules (MP) and
(Gen).
This axiomatization yields, however, a rather uncomfortable logical calculus with a lot of not quite natural axioms. Therefore, as in the propositional
case, one is interested in simpler axiomatizations. The (probably) rst such
axiomatization was given in [530]. It starts with a suitable adequate axiomatization of the propositional system Lm , e.g. from the axiomatization mentioned in Theorem 9.2.12. The metavariables used there now, of course, have
to be read as metavariables for ws of the rst-order language LL . However,
instead of adding a further list of particularly rst-order axioms (similar to
the schemata (AxRT 9); : : : ; (AxRT 11) for K m
RT ) there is a more extended system of inference rules besides the (propositional) rule of detachment (MP).
The additional rules in [530] are (i) the rules of universal quantication in
the antecedent (Gena ) and of universal quantication in the succedent (Gens )
meaning
G !L H ;
1 !L H2 ;
(
Gens )
(9.112)
(Gena ) 8H
xH1 !L H2
G !L 8xH
as well as (ii) the corresponding rules of existential quantication in the
antecedent (ExQa ) and of existential quantication in the succedent (ExQs )
meaning
H2 ;
!L G
(ExQs ) HH1!!L8xH
(9.113)
(ExQa ) 9H
xH !L G ;
1 L
2
which all use the additional assumption that the w G does not contain the
variable x free. And the additional rules are (iii) two rules (Renb ) for renaming
of bound and (Renf ) for renaming of free variables meaning in the rst case
(Renb ) HH
and in the second case

(Renf ) H [H
x=a]
with H built from H by renaming of bound variables ,
with the variable x not occurring bounded in

H , and a either a constant or a variable which
does not occur bounded in H .
In this way one gets a logical calculus which adequately axiomatizes the
systems Lm , i.e. for which soundness and completeness theorems hold true.
Even more: for the standard extension of the derivability relation of this
calculus, also the entailment theorem holds true, cf. [530].29
29 As for propositional logic, also in the rst-order case this logical calculus can be
simplied for the three-valued Lukasiewicz system, and still the completeness
theorem for the entailment relation proved, as done e.g. in [329].
255
The results of Section 3.4 essentially gave conditions under which suitable
graded sets of ws did have models, or did have together with some model
also other \bigger" or \smaller" ones. Now we are able to describe conditions
under which some given graded set of ws is not satisable in some given
L-interpretation.
In the following let be a graded set of ws of LL with its free variables
among x1 ; : : : ; xn , and let be a set of LL -sentences. Then is locally realized
by i there exists some w H of LL with its free variables among x1 ; : : : ; xn
such that [fH g is satisable and [fH g j=L holds true. And is locally
omitted by i is not locally realized by , i.e. i for each H of LL with
its free variables among x1 ; : : : ; xn from the satisability of [ fH g there
follows [fH g 6j=L . And nally we say that an interpretation A omits the
graded set of ws i is not satisable in A. Of course, all these notions
have to be taken relative to some predetermined logical system S.
Theorem 9.3.1 (Omitting Types Theorem). Consider an L-system Lm.
Suppose that is some Lm -satisable graded set of sentences and that is
some graded set of ws of Lm with its free variables among x1 ; : : : ; xn . Assume also that is locally omitted by . Then there exists some countable
model of which omits .
Proof: Assume for simplicity that the alphabet of Lm has only the connectives !L and : and the quantier 9. Let C = fc1 ; c2 ; : : :g be a countable
set of new individual constants for Lm . Add these constants to the system
Lm and denote the resulting extended system by LC . Let
G0 ; G1 ; G2 ; : : :
be an enumeration of all LC -sentences, and let
x1 ; x2 ; x3 ; : : :
be a sequence of pairwise dierent individual variables. We use this sequence

to construct an increasing sequence of satisable sets of LC -sentences which
shall start with the set T 0 = . Of course, is a set of LC -sentences and
itself LC -satisable. So let any such set T k be given and let
T k = [ fH1; : : : ; Hr g :
Put H H1 ^ ^ Hr and choose l such that the (additional) constants which
occur in H are among c1 ; : : : ; cl . Exchange these constants with the individual
variables x1 ; : : : ; xl , i.e. consider H^ H [c1 =x1 cl =xl ]. (If necessary, this
substitution has to be accompanied by a renaming of bounded variables of
H .) And let H 9x1 9xl H^ .
The set [fH g of L-sentences is satisable by construction. Because
is locally omitted by , this means that one has [ fH g 6j=L , i.e. that
there exists some w G+ (x1 ; : : : ; xn ) 2 such that the set [ fH ; : G+ g
of ws of LL is satisable. Here we use the modied negation connective :
256
introduced in (9.111) which has the property that : G+ has a designated

truth degree if G+ has a non-designated truth degree. Let
G : G+ (cmn+1 ; : : : ; cmn+n ) ;
i.e. substitute in : G+ always the constant cmn+i for the variable xi . Now
consider
T 0k = T k [ fGg :
Obviously, T 0k is a satisable set of LC -sentences. Let additionally
0
T 0k [ fGk g is satisable,
T 00k = TT k0 [[ ff:GkGg;k g ifotherwise
.
k
00
Also T k is a satisable set of LC -sentences. Finally let
8 00
T k [ fH (cr )g; if Gk 9xH (x) and cr is the
>
>
>
<
constant of C with smallest inT k+1 = >
dex which occurs neither in T k
>
>
nor in Gk ,
: 00
Tk
otherwise .
Then also T k+1 is a satisable set of LC -sentences, and additionally a nite
extension of T k , and hence
S also a nite extension of .
Consider now T 1 = 1
k=0 T k . According to the Compactness Theorem
3.4.4 one has that also T 1 is a satisable set of LC -sentences. And additionally one has by construction that either G 2 T 1 or : G 2 T 1 holds for
each LC -sentence G. Therefore consider an innite model B of T 1 and let
A jBj be the set of all elements cB for some constant c 2 C (and also of all
cB for some constant c of the language of Lm ). The LC -interpretation A shall
be the restriction of B to A. Then A is also a model of T 1 : only for ws of
the form 9xH (x) this is not completely clear. In this case, however, there is
H (cr ) 2 T 1 for a suitable constant cr 2 C , and hence also A j=L 9xH (x).
By construction A is a countable model of T 1 , and hence also a model
of which omits . And this property remains unchanged if one \reduces"
A to an Lm -interpretation A0 by forgetting the meaning assignment to the
additional constants from the set C .
2
Up to now we discussed here only nitely many-valued Lukasiewicz
systems. In closing this subsection let us take into consideration also the
system L1 , and let us compare his set of logically L1 -valid ws with the
logically valid formulas of the nitely many-valued systems Lm , i.e. let us
generalize Theorem 9.1.2 to the rst-order case. To do this the following
notation is used:
LVL =def set of all ws of LL which are logically valid for L .
Proposition 9.3.3. For any integers m; n 2 there hold true:
(a) LVL m LVL n , Wm Wn ,
257
T
L
(b) LVL 1 = 1
m=n LV m .
Proof: (a) In the case that one has Wm Wn, each Ln -interpretation
is also an Lm -interpretation, and therefore in this case also each logically
Lm -valid w of LL is a logically Ln -valid w of LL . If one conversely has that
Wm 6 Wn , then one already has tautLm n tautLn 6= ; according to Theorem
L 6= ;, i.e. LVL 6 LVL .
9.5. But then one also has LVL m n LV
m
n
T n
T1
L
L
(b) First observe that one has 1
m=n LV m = m=2 LV m , because there
exists, according to (a), for each 2 k < n an integer r n such that one
has LVL r LVL k . Furthermore one has immediately
LVL 1
1
\
m=2
LVL m
because each Lm -interpretation is also an L1 -interpretation. For proving that

here really equality holds true we show, following [473], that each w of LL
which is not logically valid for L1 is also not logically valid for some suitable
nitely many-valued system Lm .
So let H0 be a w of LL which is not logically L1 -valid. Let also A be
an L1 -interpretation and 0 an A-assignment such that ValLA (H0 ; 0 ) = t0
with t0 < 1. What we go on to discuss is the type of change which happens
for the value t0 in the case that one xes H0 ; 0 , and jAj, and varies the
interpretations P A of the predicate symbols over the universe of discourse
jAj. Additionally let P k be the set of all k-ary W1 -valued predicates over
jAj, i.e. let P k = jAj W1 . Consider over P k the binary function jj::jj into W1
which is for all R1 ; R2 2 P k determined by
jjR1 ; R2 jj =def sup jR1 (b1 ; : : : ; bk ) ; R2 (b1 ; : : : ; bk )j :
(9.114)
b1 ;:::;bk 2A
This denition works not only for all W1 -valued predicates, it works also
for all Wm -valued predicates, and has its values always in Wm . It is well
known from elementary calculus that jjR1 ; R2 jj is a metric in the set P k
(independent of the value set). For each w H of LL , in which at most the
predicate symbols Pi of arity ki occur for i = 1; : : : ; n, let
FH (; R1 ; : : : ; Rn ) =def ValLB (H; )
(9.115)
for each A-assignment , all Rki 2 P ki for i = 1; : : : ; n, and for the L-interpretation B which diers from A only insofar that one has
(Pi )B = Ri for all i = 1; : : : ; n .
Also let (H ) be the number of occurrences of the connective !L in H .
Inductively on the complexity of the w H we prove the following
Claim 1 :
jFH (; R1 ; : : : Ri : : : ; Rn ) ; FH (; R1 ; : : : Ri0 : : : ; Rn )j 2(H ) jjRi ; Ri0 jj
holds true for all i = 1; : : : ; n. To simplify notation we write
258
dmi(H; ) = jFH (; R1 ; : : : Ri : : : ; Rn) ; FH (; R1 ; : : : Ri0 : : : ; Rn)j
without explicit reference of the many-valued predicates R1 ; : : : ; Rn ; Ri0 because they are supposed to be xed within the proof of this Claim 1. For
atomic ws H one immediately has (H ) = 0, and according to (9.115) and
(9.114) thus
dmi(H; ) jjRi ; Ri0 jj = 2(H) jjRi ; Ri0 jj :
Thus suppose now that Claim 1 holds true for all subformulas of (the compound formula) H . In the case H :H1 one has (H ) = (H1 ) and
dmi(H; ) = dmi (H1 ; ) 2(H ) jjRi; Ri0 jj = 2(H) jjRi ; Ri0 jj :
In the case H (H1 !L H2 ) one has (H ) = (H1 ) + (H2 ) + 1 and thus by
elementary estimations
dmi(H; ) dmi (H1; ) + dmi (H2; )
2(H ) jjRi ; Ri0 jj + 2(H ) jjRi ; Ri0 jj
2(H ) jjRi ; Ri0 jj :
And in the nal case H 9xH1 one again has (H ) = (H1 ) and thus
dmi(H; ) = j sup FH ([x=b]; R1 ; : : : Ri : : : ; Rn)
1
b2jAj
; sup FH ( [x=b]; R1 ; : : : Ri0 : : : ; Rn )j

b2jAj
sup dmi (H1 ; [x=b])

b2jAj
sup 2(H ) jjRi ; Ri0 jj = 2(H ) jjRi ; Ri0 jj :

1
b2jAj
Hence Claim 1 is proved because it suces to have the quantier 9 in the

language.
Now choose some " > 0 such that t0 + " < 1 and assume that in the
w H0 there occur only the predicate symbols P1 ; : : : ; Pn . Then Claim 1
immediately yields
if jj(Pi )A ; Ri jj < "(H ) for each i = 1; : : : ; n ,
n2
then jFH (0 ; (P1 )A ; : : : ; (Pn )A ) ; FH (0 ; R1 ; : : : Ri : : : ; Rn )j < " ;
and thus
if jj(Pi )A ; Ri jj < "(H ) for i = 1; : : : ; n; then ValLB (H0 ; 0 ) < 1
n2
for the L-interpretation B which diers from A only insofar that one has
(Pi )B = Ri for all i = 1; : : : ; n. Obviously it is not of interest here how the
predicate symbols of LL which do not occur in H0 are interpreted in B.
For m > 1+ 1" n 2(H ) , i.e. for m1;1 < n2"H = n" 2;(H ) one even can
nd Wm -valued predicates Ri for each i = 1; : : : ; n such that jj(Pi )A ; Ri jj <
0
0)
259
" ;(H0 ) . So consider for such an m0 an Lm -interpretation B such that

0
n 2
(Pi )B = Ri for all i = 1; : : : ; n, such that B has the same universe of
discourse as A, and such that B interprets the individual constants as A

does. Then one immediately has that H0 is not valid in B . Therefore one
gets H0 2= LVL m .
2
Theorem 9.3.3 gives direct generalizations of results concerning logical validity from the propositional to the rst-order case. For results on satisability
the situation is, however, more complicated. There are ws, even sentences
of the language LL of the Lukasiewicz systems which have a L1 -model but
do not have any Ln -model for any 2 n 2 N . An example is provided by the
formula
9xP (x) ^ 8x9y(P (y) P (x)) ;
(9.116)
where P is a unary predicate symbol and dened for any formulas H1 ; H2
by
H1 H2 =def :(H1 & H1) ^ (H1 Y H1 $L H2 ) :
The essential idea here is that an interpretation A is a model of a w H1 H2
i the truth degree of H2 in A (w.r.t. some given assignment) is twice the
truth degree of H1 in A, cf. [434]:
(A; ) j=L H1 H2 i ValLA (H1 ; ) = 12 ValLA (H2 ; )
holds true for all A-assignments .
Having then the w (9.116) valid in some L-interpretation A forces that
the many-valued predicate P A assumes innitely many truth degrees as values: besides the truth degree 1 also all the truth degrees 2;n for all n 1.
0
9.3.3 The innitely many-valued L-system

The innitely many-valued L-system L1 needs to be considered separately
because a lot of the previous results, e.g. those of Section 3.4, have been
proved only for the nitely many-valued L-systems. The crucial point for the
present generalizations is to prove the Compactness Theorem also for L1 .
And this shall be possible, with reference to (3.25) instead of (3.23).
Theorem 9.3.2. Suppose that B is the ultraproduct of a family Ai, i 2 I ,
of L1 -interpretations w.r.t. an ultralter F on I . Then one has
ValLB (H; ) = t i
fi 2 I j ValLAi (H; i ) 2 V g 2 F for each neighborhood V of t
for each assignment : V ! F -Prodi2I Ai , each of the corresponding assignments i : V ! Ai w.r.t. the \factors" Ai determined by
(x) = [(i (x))i2I ]F for each x 2 V ;
each w H of LS , and each truth degree t.
260
Proof: Denition (3.25) provides the starting point for the inductive
proof on the complexity of the w H we are going to give. Hence suppose
that the claim of this theorem holds true for all subformulas of the compound
formula H .
Firstly let H (H1 !L H2 ), as well as ValLB (H; ) = t, i.e.
t = minf1; 1 ; ValLB (H1 ; ) + ValLB (H2 ; )g ;
and let V t be a neighborhood of t (in W1 = [0; 1]). Then there exist neighborhoods Vj of ValLB (Hj ; ) for j = 1; 2 such that
minf1; 1 ; s1 + s2 g 2 V t for all s1 2 V1 ; s2 2 V2 :
Because of fi 2 I j ValLAi (H1 ; i ) 2 Vj g 2 F for j = 1; 2 one has
fi 2 I j ValLAi (H1 ; i ) 2 Vj g \ fi 2 I j ValLAi (H1 ; i ) 2 Vj g 2 F ; (9.117)
and hence also fi 2 I j ValLAi (H; i ) 2 V t g 2 F . The reason is that this set is
a superset of the meet in (9.117), and because lters are closed under taking
supersets. If one has, conversely, that fi 2 I j ValLAi (H; i ) 2 V g 2 F for each
neighborhood V of t then suppose ValLB (H; ) = s 6= t. Then there exist
neighborhoods V s of s and V t of t such that V s \ V t = ;. According to the
choice of s one, however, has fi 2 I j ValLAi (H; i ) 2 V s g 2 F and hence also
; = fi 2 I j ValLAi (H; i ) 2 V s g \ fi 2 I j ValLAi (H; i ) 2 V t g 2 F ;
a contradiction. Therefore ValLB (H; ) 6= t is impossible.
In the case H :H1 the claim of the theorem immediately follows from
the corresponding claim for H1 which is supposed to hold true.
So it remains to consider the case H 9xH1 . Assume rst ValLB (H; ) = t,
i.e. supb2jBj ValLB (H1 ; [x=b]) = t. If in this case there exists some neighborhood V0 of t such that
fi 2 I j ValLAi (H; i ) 2 V0 g 2= F ;
then let c 2 jBj be chosen such that tc = ValLB (H1 ; [x=c]) 2 V0 , and let V 0
be a neighborhood of tc with property V 0 V0 . Then one would have
fi 2 I j ValLAi (H1 ; i [x=ci ]) 2 V 0 g 2 F ;
and hence ; 2 F because of
fi 2 I j ValLAi (H; i ) 2= V0 g \ fi 2 I j ValLAi (H1 ; i [x=ci ]) 2 V 0 g
fi 2 I j ValLAi (H; i ) < ValLAi (H1 ; i [x=ci ])g = ;
according to the choice of the truth degree t and the neighborhoods V0 ; V 0 .
A contradiction. Hence fi 2 I j ValLAi (H; i ) 2 V g 2 F for each neighborhood
V of t, because F is an ultralter.
So assume in this nal case for H that one conversely has for each neighborhood V of t 2 W1 that fi 2 I j ValLAi (H; i ) 2 V g 2 F . If one then has
261
ValLB (H; ) = s 6= t, there exist neighborhoods V s of s and V t of t such that

V s \ V t = ;, which would give rise to ; 2 F as in the case H H1 !L H2 .
Therefore ValLB (H; ) = t also in this case, and the whole proof is nished. 2
Theorem 9.3.3 (L1-Compactness Theorem). A graded set of ws of

LL has an L1 -model i each nite subset of has an L1 -model.
Proof: Repeat the proof procedure previously chosen for Theorem 3.4.4.
As in this proof one gets for each (H; t) 2 :
+(H;t) = f 2 I j (H;t) g f 2 I j A t-model of H g 2 F :
Therefore one has here for each B-assignment and each neighborhood V
of t
f 2 I j ValLA (H; ) 2 V g 2 F ;
and hence ValLAi (H; i ) according to Theorem 9.3.2. Now continue as in the
previous proof.
2
Thus also for the innitely many-valued L-system L1 all the main results
of Section 3.4 hold true which, using Theorem 3.4.4, were proved only for
nitely many-valued systems. Additionally one is able to prove for L1 also
the omitting types theorem together with a series of further model theoretic
results, cf. [224].
It was an unsolved problem for a long time to nd an adequate axiomatization for the rst-order system L1 . The classical monograph [465] discussed
only the adequate axiomatization of nitely many-valued systems. A rst
positive result was proved in [472]: an adequate axiomatization was given for
the \monadic fragment" of L1 , i.e. for the fragment of L1 which has only
unary predicate symbols in its language. Further axiomatizability results have
been given in [256] and [44], cf. also [463], which however did not constitute
{ nitary { logical calculi. We have to introduce some further notions to explain these results and to show in which sense the resulting logical calculi are
not nitary { i.e. not calculi in the strong sense of this word.
A w H of LL is weakly logically valid for L1 i :H is not L1 -satisable,
i.e. i one has ValLA (H; ) > 0 for each L1 -interpretation A and each Aassignment . And such a w H is strongly logically valid i H is valid in all
MV-interpretations for L1 . By an MV-interpretation for L1 one understands
any interpretation B suitable for the language LL whose corresponding truth
degree structure is a complete MV-chain hA; ; :; 0i such that the connectives
!L ; : of LL are interpreted by the truth degree functions30
verMV
! (u; v) =def :u v = u ! v;
verMV
: (u) =def :u ;
30 All the notions and notations used here have been introduced in Section 9.2,
particularly in Denitions 9.2.1 and 9.2.2.
262
and such that the quantier 9 is characterized by the condition

MV
ValMV
B (9xH; ) =def sup ValB (H; [x=b]) :
b2jBj
And a w H of LL is called (> t)-valid for some t 2 W1 i ValLA (H; ) > t

holds for each (usual) L1 -interpretation A and each A-assignment .
Hay [256] and Belluce/Chang [44] present quite similar systems of
axioms for L1 . The corresponding derivability relations shall be denoted by
`H and by `BC , respectively. The system of axioms in [44] is determined
by the propositional schemata (L1 1); : : : ; (L1 4) mentioned in Theorem 9.1.5
together with the additional axiom schemata
(L1 5) 9xH & 9xH $L 9x(H & H ),
(L1 6) 9xH Y 9xH $L 9x(H Y H ),
(L1 7) H [x=y] !L 9xH
for each variable y which does not occur bounded in H ,
(L1 8) 9xH $L 9yH [x=y]
for each variable y which does not occur bounded in H ,
(L1 9) 8x(H !L G) !L (9xH !L G)
for all variables x which do not occur free in G,
(L1 10) 9x(G !L H ) $L (G !L 9xH )
for all variables x which do not occur free in G ;
and the inference rules (MP) and (Gen). For this axiomatization a weak completeness theorem is proved in [44] stating that
H strongly logically valid , `BC H
holds true for each w H of LL . Of course, each w which is strongly logically
valid is also logically L1 -valid. The converse implication does not hold true,
however. Only the weaker result holds that one has for each w H of LL :
H logically L1 -valid ,
HY
n
Y
i=1
H strongly logically valid for some n = 1; 2; : : : :
On the other hand it is proved in [256] that for each n 1 and each w H
of LL it holds true
n
X
H:
H (> 1 )-valid , `
Therefore one also has
i=1
H weakly L1 -valid , `H
n
X
i=1
H for some n = 1; 2; : : : :
263
Furthermore it is proved in [256] that for each n 1 and each w H of LL it

holds true
n
n )-valid , ` H Y Y
H:
H (> n +
H
1
i=1
Hence one has two characterizations of logical L1 -validity as
H logically L1 -valid , `BC H Y
, `H H Y
n
Y
i=1
n
Y
i=1
H for each n = 1; 2; : : :
H for each n = 1; 2; : : :
neither of which solves the axiomatization problem for L1 in the standard

sense because logical L1 -validity (of a w H ) in both cases is equivalent to
the derivability of innitely many ws (determined by H ).
Accepting, however, innitary logical calculi which may e.g. have inference
rules with innitely many premises, then one gets adequate axiomatizations
in this generalized sense. The present results then say that one gets adequate
axiomatizations for the system L1 by enriching the above mentioned axiom
systems with the innitary inference rule
fH Y Qni=1 H j 1 n 2 N g
(9.118)
H
or also, as mentioned e.g. in [270], by enriching it with the innitary inference
rule
f:H !L Qni=1 H j 1 n 2 N g :
(9.119)
H
Denoting by LPC this innitary Lukasiewicz predicate calculus, one even
has the following completeness theorem, cf. [270].
Theorem 9.3.4. For each well-formed formula H of LL there are equivalent
(i) H is LPC -derivable ;
(ii) H is valid in all L1 -interpretations ,
(iii) H is valid in all interpretations whose truth degree structure is
a complete MV-algebra .
The nal, yet negative, solution of the standard axiomatizability problem
for L1 was given in [480] with the proof that the set LVL 1 cannot be the set
of all ws which are generated by some logical calculus (in the usual, nitary,
sense). The crucial idea behind this proof is to relate logical L1 -validity (of a
w H of LL ) with PL2 -satisability (of a suitable transformation of H ) over
nite universes of discourse.
For the explanation of the idea of this proof a further notion is needed: a
w H of LL shall be weakly satisable i there exists some L-interpretation
A and some A-assignment such that ValLA (H; ) > 0 holds.
264
The transformation of ws, mentioned just before, shall become a transformation of PL2 -formulas essentially into L-formulas. A slight modication
of the language LL is also needed here: LL has to be extended with an additional nullary predicate symbol, i.e. with an additional truth degree constant
. The extended language shall be denoted LL , the related extension of L1
by L1 . Suppose furthermore that all PL2 -formulas are formulated only in the
alphabet with the connectives :; ^ ; _ and the quantiers 8 ; 9 , i.e. that they
are \arrow free", and that otherwise LL and the language of PL2 have the
same individual and predicate symbols. Then the -transformation T (H ) of
any PL2 -formula H can be uniquely characterized by the conditions:
T (H ) H
for atomic formulas H ;
T (H1 ^ H2 ) T (H1 ) ^ T (H2 ) ;
T (H1 _ H2 ) T (H1 ) _ T (H2 ) ;
T (:H ) ( Y ) & :T (H ) ;
T (8 xH ) 8xT (H ) ;
T (9 xH ) 9xT (H ) :
Now let G be some PL2 -formula, and let its predicate symbols occur
among the predicate symbols P1 ; : : : ; Pk , with always Pi of arity ni . Consider the w
(Pi ; ) (8)(Pi (x) ^ (( Y ) & :Pi (x)) ^ :(Pi (x) $L ))
of LL with x some ni -tuple of pairwise dierent individual variables, and with
the symbol (8) indicating the universal quantication of all free variables of
the w following this symbol. Use this w to form the further one
K (G; ) ^ :( Y ) ^
V
k
^
i=1
(Pi ; ) :
Here is the nite iteration of the weak conjunction ^ of the Lukasiewicz

systems as introduced in (9.24), p. 187. By straightforward calculations one
gets as a rst lemma
G PL2 -satisable ,
(T (G) & :) ^ K (G; ) weakly L1 -satisable .
The PL2 -formula K (G; ) is even a sentence. According to its construction
one has for any L1 -interpretation B, which satises
ValLB (K (G; )) > 0 ;
(9.120)
L
even 2 for = ValB (), and has furthermore for each B-assignment
also
ValLB (T (G); ) ; or + ValLB (T (G); ) 2 ; :
Using as a further shorthand notation for any ws A; B of LL
265
A B (A ^ B ) _ (:(A _ B ) & ( Y )) ;

as well as for each i = 1; : : : ; k and any variables x; y
Di (x; y; ) 8x1 : : : 8xni
ni
^
(Pi (x1 ; : : : ; xj;1 ; x; xj+1 ; : : : ; xki )

j =1
Pi (x1 ; : : : ; xj;1 ; y; xj+1 ; : : : ; xki )) ;
then one is able to dene a binary relation w in jBj by:

b1 w b2 =def for all B-assignments :
+ ValLB
k
^
i=1
Di (x; y; ); [x=b1 ; y=b2]) 2 ; :
Under the assumption (9.120) this relation w is an equivalence relation in

jBj.
Finally one chooses a unary predicate symbol Q which does not occur in
G and forms the w of LL :
H (G; ; Q) 8x8y
k
; ^
i=1

Di (x; y; ) & : _ Q(x) $L Q(y) :
The main result is that under the assumption that the sentence H (G; ; Q)
is weakly satised in B, the equivalence relation w has only nitely many
equivalence classes in w. From this one gets as a second lemma:
G PL2 -satisable in some nite universe of discourse ,
(T (G) & :) ^ K (G; ) ^ H (G; ; Q) weakly L1 -satisable : (9.121)
If one now assumes that the set LVL 1 of all logically L1 -valid ws is
adequately axiomatizable then there would exist some logical calculus which
generates just the ws of LVL 1 , i.e. { in recursion theoretic terms { LVL 1
would be a recursively enumerable set of ws of LL . But then also the set
LV 1 of all logically L1 -valid ws would be recursively enumerable because
it is a routine matter to code the ws of LL by ws of LL : one may e.g. always
substitute by a w 8xP0 (x) for some xed unary predicate symbol P0 . But
for each w H of LL one has
H not weakly L1 -satisable , (:H ) 2 LV 1 :
Therefore one has that with LV 1 also the set of all not weakly L1 -satisable
ws would be recursively enumerable, and hence according to (9.121) also
the set of all ws of PL2 which are not satisable in some nite universe of
discourse. However, obviously one has
G logically PL2 -valid over nite universes of discourse ,
:G not PL2 -satisable in some nite universe of discourse
266
for each w G of PL2 . Thus one has that with LVL 1 also the set of all
ws of PL2 which are logically valid over nite universes of discourse would
be recursively enumerable. But this is not the case, cf. e.g. [11], and hence
also LVL 1 is not recursively enumerable, i.e. L1 is not axiomatizable in the
standard sense.
Thus, the set LVL 1 is not a decidable subset of the set of all ws of LL .
The recursion theoretic complexity of LVL 1 has been studied in [434], cf. also
[243, 244, 246]. The result is that this set LVL 1 is 2 -complete.31
Even the \simpler" set of all logically L1 -valid ws of the monadic fragment of LL , i.e. the set of all logically L1 -valid ws which contain only unary
predicate symbols, is not a decidable subset of the set of all ws of LL { in
contrast to the situation in classical rst-order logic, cf. [435].
The situation may change, however, if one changes the set DL of designated truth degrees. Mostowski [385] proves e.g. that the choice DL = (r; 1]
for some rational 0 r < 1 makes the set of logically valid ws recursively
enumerable, i.e. adequately axiomatizable. An axiomatization for this case is
given in [42]. The non-axiomatizability result remains valid, however, in the
case of the choice DL = [r; 1], again for some rational 0 < r 1, cf. [246, 385].
31 That this set belongs to the class 2 means that it can be characterized by a 2 -
formula (of the language of arithmetic, using some eective arithmetical coding
of the ws of the language LL ), i.e. by a w in prenex form having a prex of
the form 8 x 9 y, but not by a simpler w. That this set is complete in the class
2 means that each other set in this class can eectively be reduced to it.
10. The Godel Systems

Go del's family of nitely many-valued propositional logics Gm with m 2
was introduced in [199] in the context of investigations aimed to understand
intuitionistic logic. Accordingly they have been formulated as systems in conjunction, disjunction, negation, and implication, i.e. their common language
LG is constituted by the set of basic connectives
J G = f^; _; ; !Gg :
(10.1)
Their truth degree sets are the sets Wm , and DG = f1g is the set of designated
truth degrees. The truth degree functions are for all systems, respectively, the
functions
et1 ; vel1 ; non0 ; seq 1
(10.2)
as introduced in (5.8), (5.75), (5.58), and (5.82). Hence G2 is just the classical
(propositional) logic.
This choice of the truth degree functions is the reason that seq 1 is called
the Go del implication function. Sometimes, however, one deletes the negation connective from this list. But then one usually is adding a truth degree
constant 0 denoting the truth degree 0, a procedure which allows one to dene
the negation connective by
H =def H !G 0 :
(10.3)
But this slightly modied approach is equivalent to the original one because
in the standard vocabulary (10.1) one can introduce the truth degree constant
0 e.g. via the denition 0 =def (p0 !G p0 ) :
It is a routine matter to extend these Go del systems Gm with m 2 to
innitely many-valued systems G0 and G1 with truth degree sets W0 ; W1 ,
respectively. If we do not intend to distinguish between the nitely and the
innitely many-valued systems we shall use G to indicate all of them.
According to Proposition 8.1.2 the Go del systems G with 6= 2 are not
functionally complete, because all the basic connectives satisfy the normal
condition.
268
As a side remark let us mention that, contrary to the situation for the
Lukasiewicz systems, the truth degree functions seq 1 and non0 are discon-
tinuous functions over the real unit interval [0; 1].

Notice as an interesting fact that all the truth degree functions from the
list (10.2) can be characterized in purely order theoretic terms. This is obvious
for et1 = min, vel1 = max, and seq 1 , but also for non0 because one obviously
has

W ; if x = min W
non0 (x) = max
(10.4)
min W ; otherwise .
Therefore one has that the truth degree of a given w H depends not on the
particular choice of some valuation , but on the order type of the sequence of
values ( (p(n) ))n1 , as the following result shows. Even more, dierent, but
order isomorphic truth degree set generate the same sets of G-tautologies.
To discuss this relationship, denote by G(W ) the Go del system with
truth degree set W .
Proposition 10.1.1. Suppose that W ; W [0; 1] are truth degree sets
with 0; 1 2 W \ W , and that f is an order isomorphic embedding of W
into W which respects the universal lower and upper bounds, i.e. that f is
a 1-1 mapping f : W ! W which satises the conditions
x y , f (x) f (y) and f (0) = 0 ; f (1) = 1
(10.5)

for all x; y 2 W . Then one has for each w H of LG and each W -valuation
the property

f (ValG(W ) (H; )) = ValG(W ) (H; f )
(10.6)

for the W -valuation f characterized by f (q) = f ( (q)) for each propositional variable q.
Proof: The result follows by a standard induction1 over the complexity
of the w H , using the order theoretical characterizations of the basic truth
degree functions (10.2).
2
Obviously equation (10.6) immediately allows one to relate tautologies of
the systems G(W ) and G(W ) in a nice manner.
Corollary 10.1.1. For all truth degree sets W ; W [0; 1] with 0; 1 2
W \ W such that W is order isomorphically embeddable into W one
has that each G(W )-tautology is also a G(W )-tautology.
This result extends, for the nitely many-valued systems, to the entailment relations.
1 This proof shows even more: it is not important that the truth degree sets W ,
W are subsets of [0; 1]. It is sucient that they are linearly ordered with
universal lower and upper bounds. But this extended generality is not really
important for the following considerations.
269
Theorem 10.1.1. A nitely many-valued Godel system is completely char-
acterized by its number of truth degrees, i.e. any two n-valued Go del systems
have coinciding entailment relations.
Proof: Suppose that the truth degree sets W ; W [0; 1] with 0; 1 2

W \ W have the same number of elements. Then there exists an order
isomorphism f which maps W onto W and satises (10.5).
Let be a set of ws of LG and H a w. Assume that j=G(W ) H , and
that is a W -valuation which is a model of . Then the W -valuation
dened by (q) = f ;1 ((q))
is also a model of because of = f and

(10.6). Thus one has ValG(W ) (H; ) = 1 and hence also ValG(W ) (H; ) = 1,
again by (10.6).
All together this means j=G(W ) H . Therefore j=G(W ) H follows
from j=G(W ) H . And by symmetry one has also the converse implication.
In particular one has that any two n-valued Go del systems have coinciding sets of tautologies.
As a side remark observe that this result on the entailment relations is, in
a suitable sense, independent of the denition of the entailment relation. More
precisely, it does not matter here whether the entailment relation is dened
as usually in accordance with (3.8), or whether it is dened in a way similar
to that mentioned in (3.11). This last mentioned idea can be particularly
adapted to the present setting if one introduces a graded entailment relation
G by the denition
if ValG (; ) ValG (H; ) (10.7)
ValG ( G H; ) = 1 ; G
Val (H; ) ; otherwise ,
using the shorthand notation ValG (; ) for: inf fValG (G; ) j G 2 g. Then
one has the following result.
Proposition 10.1.2. For each set of ws and each w H one has
j=G H i ValG ( G H; ) = 1 for all valuations .
Proof: (() If one has ValG ( G H;G) = 1 for all valuations , and
if is a model of , then one has inf fVal (G; ) j G 2 g = 1 and hence
ValG (H; ) = 1, i.e. is also a model of H .
()) Now assume conversely that there exists some valuation such that
ValG ( G H; ) = v < 1. Then consider the modied valuation v

( p) v
v (p) = 1 (; p) ; ifotherwise.
(10.8)
We claim that in this case one has even for each w H1 the corresponding
property

G
G
G
Val (H1 ; v ) = Val (H1 ; ) ; if Val (H1 ; ) v
(10.9)
1;
otherwise .
270
This claim can be shown for any truth degree v by an easy induction on the
complexity of H1 . So for a moment let v be an arbitrary truth degree. The
case that H1 is a propositional variable is obvious from (10.8), and the case
that H1 is a negated formula is trivial. The cases that H1 is a conjunction or
a disjunction follow by rather simple calculations. So only the case that H1
is an implication H1 H 0 !G H 00 should be considered in some detail.
For ValG (H1 ; ) = v = 1 one has = v , thus ValG (H1 ; v ) = ValG (H1 ; ).
In the case ValG (H1 ; ) v < 1 one has ValG (H1 ; ) = ValG (H 00 ; ) v,
and therefore ValG (H 00 ; ) = ValG (H 00 ; v ) by induction hypothesis. Hence
one gets ValG (H1 ; ) = ValG (H1 ; v ) from ValG (H 0 ; ) ValG (H 0 ; v ). And
in the case ValG (H1 ; ) > v one either has ValG (H 0 ; ) > ValG (H 00 ; ) =
ValG (H1 ; ) > v, or one has ValG (H 0 ; ) ValG (H 00 ; ). The rst one of
these two possibilities leads to ValG (H 0 ; v ) = ValG (H 00 ; v ) = 1, which
means ValG (H1 ; v ) = 1. The second one yields two further subcases: (i)
that ValG (H 0 ; ) v and hence ValG (H 0 ; v ) = ValG (H 0 ; ) ValG (H 00 ; v ),
and (ii) that ValG (H 0 ; ) > v and hence ValG (H 0 ; v ) = ValG (H 00 ; v ) = 1. In
both subcases thus ValG (H1 ; v ) = 1 results.
Therefore the claim (10.9) is established for any truth degree v. So let 1 >
v = ValG ( G H; ). Then v is a model of which satises ValG (H 0 ; v ) =
v < 1. Therefore also j=G H does not hold true.
2
Despite the fact that the implication connective !G of the Go del systems
satises condition (I ) it is not possible to apply the Rosser-Turquette
method to axiomatize the Go del systems Gm with m > 2 because not all
of the unary connectives Jt for t 2 Wm are denable in these systems. For
consider some truth degree 0 6= t 2 Wm , then in using the truth degree functions (10.2) one is (only) able to generate the truth degrees 0; t; 1. However,
if one exchanges in the superpositions of truth degree functions from the list
(10.2), which generate these truth degrees 0; t; 1, in all of the argument places
the truth degree t with another truth degree 0 6= s 2 Wm , then one generates always s instead of t. Therefore it is impossible to distinguish between
any dierent truth degrees 0 < s; t 2 Wm in the language LG of the Go del
systems.
We come back to the axiomatizability problem for the Go del systems
later on in this section. First we shall consider the sets tautG of tautologies
of the dierent Go del systems G . Because all the basic connectives satisfy
the normal condition each G -tautology is also a tautology of classical logic.2
The tautology ::p ) p of classical logic, however, is not a tautology of, e.g.,
the system G3 .
Theorem 10.1.2. One has for all 2 m 2 N the relationships:
(a) tautG1 tautGm+1 tautGm ,
2 Of course, for the \classical reading" of a w H of the language of G one has to
exchange each one of the basic connectives of the Godel systems by its classical
counterpart.
271
G
(b) tautG1 = tautG0 = 1
m=2 tautm .
Proof: (a) Let n 2 N and consider the w
T
Gn =def
W
n_
;1 _
n
i=1 k=i+1
(pi !G pk ) ^ (pk !G pi )
(10.10)
with for the nite iteration of the disjunction connective _ as already dened in (9.25), and with pairwise dierent propositional variables p1 ; : : : ; pn .
Because one has
;

ValG (p !G q) ^ (q !G p); = 1 , (p) = (q)
in all the Go del systems, Gn is a Gm -tautology i m > n. Obviously Gm has
m truth degrees. Therefore in the case n > m each valuation : V0 ! Wm
gives to at least two of the propositional variables p1 ; : : : ; pn the same truth
degree according to the pigeonhole principle. But then ValG (Gm ; ) = 1. In
the other case n m there exists some valuation : V0 ! Wm such that
(p1 ); : : : ; (pn ) are pairwise dierent truth degrees, hence ValG (Gm ; ) 6= 1.
Thus one has Gm 2 tautGm n tautGm+1 .
Now consider a w H 2 LG with its variables occuring among p1 ; : : : ; pn .
Let : V0 ! W be any valuation and f an injective and order-preserving
mapping from the set f0; 1; (p1); : : : ; (pn )g of truth degrees into Wm which
additionally satises f (0) = 0 and f (1) = 1. Then it is a routine matter to
prove by induction on the complexity of the w H that
ValG (H; 0 ) = f (ValG (H; ))
(10.11)
holds true for each valuation 0 : V0 ! W which satises 0 (pi ) = f ( (pi ))
for all i = 1; : : : ; n.
For H 2 tautGm+1 and : V0 ! Wm there exists an injective orderpreserving mapping f : f0; 1; (p1); : : : ; (pn )g ! Wm+1 , which satises
f (0) = 0 and f (1) = 1. Then one has for the valuation f : V0 ! Wm+1 ,
determined by f (p) = f ( (p)) for all propositional variables p, immediately
ValG (H; ) = f ;1(ValG (H; f )) = f ;1 (1) = 1 :
Therefore H is also a Gm -tautology, and one thus has tautGm+1 tautGm .
This means tautGm+1 tautGm , and it remains to consider tautG1 . However,
because of Wm+1 W1 each G1 -tautology is also a Gm+1 -tautology. Now
tautG1 tautGm+1 follows from what has already been proved.
(b) The fact that tautG1 = tautG0 holds true was already mentioned and can
be proved in a similar way as the corresponding result for the Lukasiewicz
systems was proved in Theorem 9.1.2. From (a) one has immediately tautG1
T
1 tautG . Therefore it remains to prove that a w H which is not a tautG m
1
m=2
tautology is also not a tautGm -tautology for some suitable m 2. Suppose
that in H at most the propositional variables p1 : : : ; pn occur, and let f be
an injective order-preserving mapping from the set f0; 1; (p1); : : : ; (pn )g
272
into Wn+2 , satisfying f (0) = 0 and f (1) = 1. Choosing 0 : V0 ! Wn+2 as

in (10.11) gives immediately ValG (H; 0 ) = f (ValG (H; )) 6= 1. Hence H is
not a tautGn+2 -tautology.
2
It is therefore not important to distinguish the two innitely many-valued
systems G0 and G1 as far as one is only interested in their sets of tautologies.
Hence one often identies the innitely many-valued (propositional) Go del
systems, and speaks of the innitely many-valued Go del system. For the
corresponding entailment relations, however, this identication is impossible
as shall be shown at the end of this section. On the other hand, all the
innitely many-valued Go del systems have the same sets of tautologies.
Proposition 10.1.3. Each innitely many-valued (propositional) Godel
system G(W ) has tautG1 as its set of tautologies.
Proof: Assume rst that W [0; 1] is innite with 0; 1 2 W . Then one
has tautG1 tautG(W ) tautGm for each m 2 according to Corollary 10.1.1,
and therefore tautG1 = tautG(W ) according to the previous theorem.
2
It is now preferable rst to discuss the axiomatizability of the innitely
many-valued Go del system. Such an axiomatization was given by Dummett
[131]; we get his result following a presentation given in [279]. Because of
the historical origin of the Go del systems it should be no surprise that
we shall nd a close relationship to intuitionistic propositional logic. This
intuitionistic propositional logic can in a propositional language with the
standard connectives ^; _; !; e.g. be axiomatized by a logical calculus
having the following system of axiom schemata, cf. e.g. [75, 315, 444]:
(LC1) H1 ! (H1 ^ H1 ) ,
(LC2) (H1 ^ H2 ) ! (H2 ^ H1 ) ,
(LC3) (H1 ! H2 ) ! (H1 ^ H3 ! H2 ^ H3 ) ,
(LC4) ((H1 ! H2 ) ^ (H2 ! H3 )) ! (H1 ! H3 ) ,
(LC5) H1 ! (H2 ! H1 ) ,
(LC6) H1 ^ (H1 ! H2 ) !G H2 ,
(LC7) H1 ! H1 _ H2 ,
(LC8) H1 _ H2 ! H2 _ H1 ,
(LC9) (H1 ! H3 ) ^ (H2 ! H3 ) ! (H1 _ H2 ! H3 )) ,
(LC10) H1 ! (H1 ! H2 ) ,
(LC11) (H1 ! H2 ) ^ (H1 ! H2 ) ! H1 ,
and having as its single inference rule the rule of detachment (MP) w.r.t. the
implication connective !.
Before we proceed toward completeness proofs for the Go del systems
another side remark concerning the standard vocabulary is of interest, having
in mind that for intuitionistic logic it is well known that all the standard
connectives ^; _; !; are denably independent, i.e. that no one of them is
denable from the other ones, cf. e.g. [554]. The situation is not so clear for
273
the Go del systems. It has been shown in [525] that in G1 neither one of
the connectives ^; !G is denable from the other ones. And it has also been
shown in [48] that in G1 the disjunction connective _ is denable from the
connectives ^; !G .
A possible semantical characterization of the tautologies of intuitionistic
logic, i.e. of the theorems deducible in this logical calculus, is that they are
just those ws of this language which are valid in all Heyting algebras.
Denition 10.1.1. A Heyting algebra, also called pseudo-Boolean algebra, is an algebraic structure A = hA; u; t; ; 0; 1i such that
(HA1) hA; t; u; 0; 1i is a distributive lattice with zero and unit elements,
and that u; form an adjoint pair, i.e. that one has for all a; b; c 2 A:
(HA2) a u b 6 c i b 6 (a c) :
Such a Heyting algebra is a Heyting chain i it satises additionally for
all a; b 2 A the linearity condition
(HAlin) (a b) = 1 or (b a) = 1 .
These Heyting algebras provide algebraic interpretations only for the
connectives ^; _; !. They are, however, immediately enriched by another
operation ; which corresponds to the negation , and which is determined
by the denition:
;a =def a 0 for all a 2 A :
(10.12)
Basic properties of this operation are as follows.
Corollary 10.1.2. For each Heyting algebra A = hA; u; t; ; 0; 1i one
has for all a; b 2 A
(i) a a = 1 ;
1 a = a;
(ii)
;0 = 1 ;
;1 = 0 ;
(iii) a b = 1 i a 6 b ;
(iv) (a ;b) = (b ;a) :
Proof: (i) One has from (HA2) immediately: 1 6 a a i a = 1 u a 6 a,
for each a 2 A, and thus always a a = 1. And one has c 6 1 a i
c = c u 1 6 a for each a; c 2 A again from (HA2), and thus always a = 1 a.
(ii) follows easily from (i).
(iii) One has a b = 1 i 1 6 a b, and therefore i a = 1 u a 6 b, in
both cases using (HA2).
(iv) Because of (10.12) one has to prove that for all a; b 2 A the equation
a (b 0) = b (a 0)
(10.13)
holds. And by symmety it suces to prove inequality 6 here instead of equality. But one has from the adjunction property (HA2) immediately
274
a (b 0) 6 b (a 0)
i b u (a (b 0)) 6 a 0
i a u b u (a (b 0)) 6 0
i a u (a (b 0)) 6 b 0
i a (b 0) 6 a (b 0) :
Because this last condition is obviously true, (10.13) is established.

2
That some w H of the language of intuitionistic propositional logic is
valid in some Heyting algebra A now means that one has to interpret the
connectives ^; _; !; as the operations u; t; ; ;, respectively, that one
has to read the propositional variables as variables for elements of A, and
that then H always gets the value 1.
Now there is a rst, algebraic fact which links the intuitionistic logic with
the Go del systems.
Proposition 10.1.4. The nite Heyting chains are, up to isomorphism,
the Go delian truth degree structures hWm ; et1 ; vel1 ; seq 1 ; 0; 1i with m 2.
Proof: It is an easy exercise to show that each one of the truth degree
sets Wm together with the truth degree functions et1 ; vel1 ; seq 1 ; non0 and the
truth degrees 0; 1 forms a nite Heyting chain. So it remains to prove that
each Heyting chain with nitely many elements is isomorphic to one of these
Heyting chains Wm .
Because of Corollary 10.1.2(iii), (HAlin ) just demands the linearity of the
lattice ordering of the Heyting algebra. Heyting chains hence are really
chains in the usual order theoretic sense. Therefore there exists for each
Heyting chain A with m 2 elements an injective and order-preserving
mapping f of A onto Wm which satises f (0) = 0 and f (1) = 1. We show
that this mapping f is even an isomorphism between the Heyting chain
A and the Godelian truth degree structure hWm; et1; vel1; seq 1; 0; 1i. So we
have to prove that f preserves the operational behavior. The way we choose
to reach this goal is based on the fact that f is order-preserving, and consists
in showing that in the Heyting chain A the operations can be dened from
the lattice ordering 6 in just the same way as the truth degree functions in
Wm can be dened from the usual ordering of these truth degrees.
Because of (HAlin) one has a u b = inf fa; bg = minfa; bg as well as a t b =
supfa; bg = maxfa; bg in the lattice hA; t; u; 0; 1i. So u; t have the same
order theoretic characterization as et1 ; vel1 . For a 6 b we already know that
then (a b) = 1 holds true. And in the case a > b one has a 66 (a b)
according to (HA2), and therefore (a b) < a according to (HAlin ). Because
one has always b 6 (a b), according to (HA2), as well as a u (a b) 6 b,
one has at all
b 6 (a b) = a u (a b) = b u (a u (a b)) 6 b
275
and thus (a b) = b in the case a > b. But this is again the same type
of characterization as we have for seq 1 . Hence f is an isomorphism and our
proposition proved.
2
Furthermore it is interesting to remark that in all Heyting chains the
operation is characterizable from the lattice ordering just in the same way
as the truth degree function seq 1 .
Corollary 10.1.3. In each Heyting chain A = hA; u; t; ; 0; 1i one has
for all a; b 2 A and the lattice ordering 6 of A:

a6b
a b = 1b;; ifotherwise.
Proof: Checking the last paragraph of the previous proof shows that the
arguments there used only the fact that one is considering a Heyting chain,
but no niteness assumption.
2
We have to consider also a third type of algebraic structure, which we
shall call G-algebras and which shall be particular Heyting algebras. And
we have to consider lters in these G-algebras. But this notion of lter is also
of interest for Heyting algebras and thus we dene this notion in the more
general setting of Heyting algebras.
Denition 10.1.2. A Heyting algebra A is a G-algebra i it satises the
condition
(HAG ) (a b) t (b a) = 1
for all a; b 2 A. And a nonempty subset F of a Heyting algebra A is a lter
of this Heyting algebra i there hold true for all a; b 2 A:
(F1) if a; b 2 F then also a u b 2 F ,
(F2) if a 2 F and a 6 b then also b 2 F .
By a proper lter one understands any lter F 6= A. And a prime lter is a
proper lter F which satises for all a; b 2 A the further condition
(F3) if a t b 2 F then either a 2 F or b 2 F .
For each lter F of a Heyting algebra A one has 1 2 F . And for each
element a 2 A the set fb 2 A j a 6 bg is a lter of A, the principal lter
of a. For a 6= 0 the principal lter of a is a proper lter. As an immediate
corollary one gets that the union of each -chain of lters, i.e. of each set of
lters linearly ordered by inclusion, is again a lter. And for each lter F of
A an equivalence relation F in A is determined by the denition
a F b =def (a b) 2 F ^ (b a) 2 F :
This relation F is even a congruence relation, i.e. respects the operations
u; t; ; of A. Therefore it is a routine matter to prove that the quotient
276
structure A=F , with its operations u; t; ; dened between equivalence

classes [a]; [b] by
[a] u [b] = [a u b] ; [a] t [b] = [a t b] ; [a] [b] = [a b] ;
is again a Heyting algebra, cf. e.g. [439, 444].
Lemma 10.1.1. (a) Suppose that A is a Heyting algebra, a 2 A, and F a
lter with a 2= F . Then there exists a prime lter F of A such that F F
and a 2= F .
(b) Let A be a Heyting algebra and a; b 2 A satisfy b 66 a. Then there
exists a prime lter F of A such that b 2 F and a 2= F .
(c) If A is a G-algebra and F some prime lter of A, then the quotient
A=F is a Heyting chain.
Proof: (a) Let Fa;F be the set of all proper lters G of A such that
F G and a 2= F . Then one has Fa;F 6= ; because of F 2 Fa;F . And in Fa;b
each -chain has a supremum, its union. Hence from Zorn's lemma there
follows that Fa;F has a maximal element F . We shall prove that F is a
prime lter.
Because of F 2 Fa;F one has F F and a 2= F . Assume that F is
not a prime lter. Then there exist c1 ; c2 2= F such that c1 t c2 2 F . Now
consider the -smallest lters F1 ; F2 F such that ci 2 Fi for i = 1; 2.
These lters exist because the intersection of any family of lters in A is
again a lter in A. But then one has a 2= F1 or a 2= F2 , because otherwise
one would have a 2 F1 as well as a 2 F2 , and hence there would exist
elements d1 ; d2 2 F such that c1 u d1 6 a and c2 u d2 6 a, i.e. such that
(d u c1 ) t (d u c2 ) 6 a for d = d1 u d2 . Therefore one would have a >
d u (c1 t c2 ) 2 F by the distributivity of the underlying lattice of A, and
thus a 2 F , which contradicts a 2= F .
By symmetry let a 2= F1 . Because of b 2 F F1 one then has F1 2 Fa;F ,
contrary to the choice of F . Hence F has to be a prime lter.
(b) Suppose b 66 a. Then the principal lter Fb of b is a lter with a 2= Fb .
Now apply (a).
(c) Suppose that A is a G-algebra and F a prime lter in A. Then one
has, according to (HAG ), in the quotient structure A=F immediately
[1] = [(a b) t (b a)] = [a b] t [b a] 2 F ;
and hence ([a] [b]) = [a b] 2 F or ([b] [a]) = [b a] 2 F , i.e. in
A=F there hold true ([a] [b]) = [1] or ([b] [a]) = [1], and hence [a] 6 [b]
or [b] 6 [a]. Therefore A=F is a Heyting chain.
2
Lemma 10.1.2. A Heyting algebra A is a G-algebra i A is isomorphic

with a subdirect product of Heyting chains.
277
Proof: It is easy to see that each direct product of G-algebras is again a

G-algebra with its operations dened \componentwise" from the corresponding operations in the factor structures. Thus the direct product of Heyting
chains is always a G-algebra. Furthermore one has that each sub-Heytingalgebra of a G-algebra is again a G-algebra. Hence, being isomorphic to a
subalgebra of a direct product of Heyting chains means to be a G-algebra.
So suppose that A
is some G-algebra. Let P be the class of all prime lters
T
of A. Then one has P = f1g according to Lemma 10.1.1(b). Therefore one
gets from a quite general result in universal algebra, cf. e.g. [59, 225], that
A is isomorphic to a subdirect product of all quotient structures A=F for
F 2 P, i.e. according to Lemma 10.1.1 (c) to a subdirect product of Heyting
chains.
2
Let us now come back to the ws of the language Li of intuitionistic logic
as discussed above. The basic connectives ^; _; !; of this language correspond to the basic operations u; t; ; ; of the Heyting algebras, and these
basic operations can be considered as (generalized) truth degree functions
for these connectives. Consider a Heyting algebra A and an A-valuation
: V0 ! A. Then one is able to dene the A-value ValA (H; ) of any w
H 2 Li w.r.t. this valuation by suitably adapting the denition (2.6). Then
H is valid in A i ValA (H; ) = 1 for each such A-valuation .
It is a well known result in intuitionistic logic that a w H 2 Li is derivable
from the axiom schemata (LC1); : : : ; (LC11) via the rule of detachment (MP)
i H is valid in all Heyting algebras, cf. e.g. [439, 444].QAnd, obviously, each
such w H 2 Li is valid in some direct product A = i2I Ai of Heyting
algebras if H is valid in all the factor algebras Ai . Furthermore its validity is
transmitted from any Heyting algebra to all of its subalgebras. Hence each
w H which is valid in all the Heyting algebras Ai is also valid in each
subdirect product of all these Ai .
We enrich the list (LC1); : : : ; (LC11) of axiom schemata with the further
schema
(LCG ) (A ! B ) _ (B ! A)
and dene within the set Li of all intuitionistic ws a binary relation by
H G =def ì (H ! G) ^ ì (G ! H ) ;
with ì for derivability from (LC1); : : : ; (LC11); (LCG ) using the inference rule
(MP). As in the proof of Lemma 9.2.4 one proves that is an equivalence
relation in Li . This proof does not use schema (LCG ). The quotient Li =
becomes a Heyting algebra w.r.t. the operations dened representativewise
from the basic connectives of the language Li of intuitionistic logic. This
Heyting algebra is even a G-algebra according to schema (LCG ), and shall be
denoted by AIG . The details for establishing these claims are mainly routine
calculations and in detail e.g. presented in [439, 444].
278
Proposition 10.1.5. For a w H of the language Li of intuitionistic propo-
sitional logic there are equivalent:

(a) H is derivable from (LC1); : : : ; (LC11) and (LCG ) using (MP) ,
(b) H is valid in each G-algebra ,
(c) H is valid in each Heyting chain ,
(d) H is valid in some innite Heyting chain ,
(e) H is valid in each nite Heyting chain ,
(f ) H is valid in each Heyting chain with at most n + 2 elements, n being
the number of propositional variables which occur in H ,
(g) H is valid in AIG .
Proof: Suppose that (a) holds true. Because (LC1); : : : ; (LC11) are valid
in each Heyting algebra, these ws are also valid in each G-algebra. And
(LCG ) is valid in some Heyting algebra i this Heyting algebra satises
condition (HAG ), i.e. i it is a G-algebra. Now (b) is established because
(MP) leads from ws which are valid in some G-algebra only to a w which
is valid in the same G-algebra.
Obviously (c) follows from (b). And conversely (b) follows from (c) because of Lemma 10.1.2 and the fact that a w, which is valid in all factors of
a subdirect product, is also valid in this subdirect product.
From (b) immediately also follows (g). And according to the denition of
AIG validity in AIG means ì-derivability. Hence (a) follows from (g).
Thus all of (a), (b), (c), (g) are equivalent.
From (c) immediately (d) follows because hW1 ; et1 ; vel1 ; seq 1 ; 0; 1i is an
innite Heyting chain. From (d) one obviously gets (e) because each nite
Heyting chain is subalgebra of each innite Heyting chain: one simply has
to consider some order-preserving injection3 which surely exists. And from
(e) one obviously gets (f) which is only a particular case of (e).
So assume that (f) holds true and that exactly the n propositional variables p1 ; : : : ; pn occur in H . We infer (c). In the case that (c) does not hold
true there would exist some Heyting chain A and some A-valuation with
ValA (H; ) 6= 1. However, the set f0; 1; (p1 ); : : : ; (pn )g is the carrier of
some subalgebra B of A which is a Heyting chain with at most n + 2 elements. And for the B-modication dened by
n
p 2 fp1; : : : ; pn g
(p) = 0(p); ifotherwise
one obviously has ValA (H; ) = ValB (H; ). Hence H would not be valid in
B contrary to the assumption that (f) holds true. Thus (c) follows from (f).
Hence also the equivalence of all of (c), (d), (e), and (f) is proved.
2

Finally we come back to the Go del systems. Let (LC 1); : : : ; (LC 11), and
(LCG ) be the schemata in the language LG which result from the schemata
3 This is the very same approach as was taken in the proof of Theorem 3.4.1(a).
279
(LC1); : : : ; (LC11); (LCG ) if one exchanges always the intuitionistic implication connective ! by the Go delian one !G , and always reads ^; _; in
their Go delian sense.
Theorem 10.1.3 (Completeness Theorem for G1). A w H of LG is a
G1 -tautology i H is derivable from the schemata (LC 1); : : : ; (LC 11), and
(LCG ) via the rule of detachment (MP) for the implication !G .
Proof: If H is a G1 -tautology then H is a Gm-tautology for all m 2
according to Theorem 10.1.2. Therefore H is valid in all nite Heyting
chains by Proposition 10.1.4, and hence derivable from (LC 1); : : : ; (LC 11),
and (LCG ) via the rule of detachment (MP) according to Proposition 10.1.5,
because the exchange of the intuitionistic connectives with the Go delian
ones does not matter for this derivability result.
Suppose conversely that H is derivable from (LC 1); : : : ; (LC 11), and
(LCG ) via the rule of detachment (MP). Then H is valid in all nite Heyting chains by Proposition 10.1.5, and hence a Gm -tautology for all m 2
according to Proposition 10.1.4. Therefore H is also a G1 -tautology because
of Theorem 10.1.2.
2
The axiom schemata (LC1); : : : ; (LC11); (LCG ) together with the rule of
detachment (MP) constitute a logical calculus which usually is denoted LC
following Dummett [131]. By a schematic extension of LC one means any
logical calculus whose theorems, i.e. whose derivable ws, can be generated
by adding some further axiom schemata to LC and deriving in this extended
LC-calculus. For the rest of this section such a schematic extension of LC shall
be denoted by T. And a T-algebra shall be such a Heyting algebra which is
a model of T. Thus the G-algebras become just the LC-algebras.
All the notions which have been introduced for Heyting algebras apply
also to these T-algebras, and the previous results too. Thus one has that each
T-algebra is isomorphic with a subdirect product of Heyting algebras. And
for each w H of the language of T which is not a theorem of T, i.e. not
T-derivable, there exists according to Proposition 10.1.5 and Lemma 10.1.2
some nite Heyting chain, which is even a T-algebra and in which H is
not valid. Therefore in this case there exists according to Proposition 10.1.4
some m 2 such that H is not a Gm -tautology, but that each T-theorem is
a Gm -tautology.
A schematic extension T of LC is called proper i there exists some Ttheorem which is not an LC-theorem. And such a schematic extension T of LC
is called consistent i not all the ws (of the language of T) are T-theorems.
Theorem 10.1.4. Suppose that T is a consistent proper schematic extension
of LC. Then there exists some m 2 such that the set of theorems of T is
the set of all Gm -tautologies.
Proof: Suppose that T is a schematic extension of LC and also consistent. Let K be the set of all integers m such that Wm is a T-algebra (w.r.t.
280
the \standard" operations from the list (10.2)). One has K 6= ; because of
the consistency of T. Hence T-algebras exist. But K is also nite, because
otherwise one would have K = fn j n 2g, caused by the fact that for any
k < n always Wk is isomorphic to a subalgebra of Wn (always w.r.t. the
operations from the list (10.2)) and that thus Wk has to be a T-algebra too
if Wn is. But in the case K = fn j n 2g all T-theorems would have to be
G1 -tautologies according to Theorem 10.1.2, and T could not be a proper
schematic extension of LC.
So consider m = max K . Then Wm is a T-algebra and thus each Ttheorem a Gm -tautology. But if some w H is not a T-theorem then H is
also not a Gn -tautology for some suitable n 2 which additionally has to
satisfy n m according to the choice of m. Therefore this w H is also not a
Gm -tautology. A contradiction. Hence all the T-theorems are Gm -tautologies.
Thus one can adequately axiomatize the set of all Gm -tautologies, m 2,

in such a way that one adds to the calculus LC, i.e. to the axiom schemata
(LC 1); : : : ; (LC 11); (LCG ), an axiom schema which is given by some w H
of LG which is a Gm -tautology but not a Gm+1 -tautology. This crucial property has e.g. the w considered in (10.10). Therefore one gets an adequate
axiomatization of Gm if one adds to the schemata (LC 1); : : : ; (LC 11); (LCG )
the further schema
m m_
+1 ;
_
i=1 k=i+1
(Ai !G Ak ) ^ (Ak !G Ai ) :
There are other possibilities too for adequately axiomatizing Gm . One of them
was given in [319] where the author, recursively for each n 3, introduces a
sequence of ws by
K3 =def (( p1 !G p2 ) !G (((p2 !G p1 ) !G p2 ) !G p2 )) ;
Kn+1 =def (( Kn !G pn ) !G (((pn !G Kn ) !G pn ) !G pn )) ;
using a sequence p1 ; p2 ; p3 ; : : : of pairwise dierent propositional variables.
Inductively one gets for these ws
n
ValG (Kn ; ) = (pn;1 ) ; if 0 < (p1 ) < (p2 ) < : : : < (pn;1 ) < 1
1
otherwise
for each valuation : V0 ! W1 . Hence one has for each m 3 that Km
is a Gm -tautology but not a Gm+1 -tautology. Therefore one can adequately
axiomatize each one of the Go del systems Gm , with m 3, by the axiom
schemata (LC 1); : : : ; (LC 11); (LCG ) together with the further schema
Km =def (( Kn !G An ) !G (((An !G Kn ) !G An ) !G An ))
with additionally K2 =def A1 .
For the case m = 3 one has a further simplication because an adequate
axiomatization of G3 is already provided by the schema K3 together with only
281
the schemata (LC 1); : : : ; (LC 11), i.e. schema (LCG ) becomes super uous in
this case, cf. [319].
Finally, we take a closer look at the innitely many-valued Go del systems. Their sets of tautologies have been discussed in Proposition 10.1.3
and are all the same. Thus we consider now their entailment relations which
prove to be dierent depending on the choice of the truth degree sets, cf. [29].
Again we consider any (innite) truth degree set W [0; 1] subject only to
the restriction 0; 1 2 W .
Remember that for truth degrees u; v 2 W one has a simple possibility
to express the inequality u v: to have seq 1 (u; v) = 1. The same connection between the validity of an implication (under some valuation) and an
inequality between truth degrees can also be expressed as
ValG (H1 !G H2 ; ) = 1 , ValG (H1 ; ) ValG (H2 ; ) :
It is, however, also possible to express the stronger inequality u < v (or
equivalently the inequality u > v) in a similar way. This can, e.g., be done
by observing that one has

if (p) > (q)
ValG ((p !G q) !G q; ) = 1 (; q) ;
otherwise
for each valuation . And this gives for any ws H1 ; H2 the result:
ValG ((H1 !G H2 ) !G H2 ; ) = 1
, ValG (H1 ; ) > ValG (H2 ; ) _ ValG (H2 ; ) = 1 :
This means, e.g., that for each sequence p1 ; p2 ; : : : of (pairwise dierent)
propositional variables the set of ws
" = f(p2 !G p1 ) !G p1 ; (p3 !G p2 ) !G p2 ; : : : g
;

has each valuation as a model for which (p(n) ) n0 is a strongly increasing innite sequence of truth degrees. And it means similarly that the set of
ws
# = f(p1 !G p2 ) !G p2 ; (p2 !G p3 ) !G p3 ; : : : g
;

has each valuation as a model for which (p(n) ) n0 is a strongly decreasing innite sequence of truth degrees.
Considering additionally a further propositional variable q and the set of
ws
= fp1 !G q; p2 !G q; : : : g ;
one immediately sees that one has
" [ 6j=G1 q and # [ 6j=G1 q ;
(10.14)
"
because one easily constructs e.g. models; of [ with supi0 (pi ) (q)
which are not models of q, but have (p(n) ) n0 as a strongly increasing
282
innite sequence of truth degrees. A similar construction gives the second

one of these results.
;

On the other hand, if one has a model of " such that (p(n) ) n0
is not a strongly increasing innite sequence, then there exists some k 2 N
with (pk ) = 1.
This, however, means e.g. for the innite truth degree set

W " = 1 ; n1 j n > 0 [ f1g
that one has
" [ 6j=G(W ") q but # [ j=G(W " ) q :
(10.15)
And it means e.g. for the innite truth degree set

W # = n1 j n > 0 [ f0g
that one has
" [ j=G(W #) q but # [ 6j=G(W # ) q :
(10.16)
Therefore all three innitely many-valued Go del systems G1 , G(W " ),
and G(W # ) have pairwise dierent entailment relations.
Starting from the connection between the tautology-sets for the nitely
many-valued Go del systems Gm and the innitely many-valued system G1 ,
given in Theorem 10.1.2(b), one now is led to the problem whether there
exists a corresponding relationship also for the entailment relations. The following result shows that the entailment relation of G1 is dierent from the
\intersection" of the entailment relations of all the systems Gm , that this \intersection", however, can be represented as the entailment relation of another
innitely many-valued Go del system.
Proposition 10.1.6. An entailment relationship j=Gm H holds true in all
nite Go del systems Gm i one has j=G(W " ) H in the innitely manyvalued Go del system G(W " ).
Proof: Because, according to Theorem 10.1.1, any two nitely manyvalued Go del systems with the same number of truth degrees have coinciding
entailment relations, we can introduce a new family of truth degree sets by

Wm" =def 1 ; n1 j 0 < n m ; 1 [ f1g
for each n 2, and consider the systems G(Wm" ) instead of the systems Gm .
Then each G(Wm" )-valuation is also a G(W " )-valuation. And this means
that each entailment relationship j=G(W ") H which holds for the system
G(W " ) also holds for all the systems G(Wm" ) with m 2.
If, on the other hand, an entailment relationship j=G(W ") H does
not hold true for the system G(W " ), then there exists a G(W " )-valuation
such that ValG ( G(W ") H; ) = v < 1, cf. Proposition 10.1.2. This
283
means v = ValG (H; ), and v = 1 ; m1;1 for some suitable integer m.

Hence one has v = ValG (H; v ) for the G(Wm" )-valuation v introduced in
(10.8). Because of (10.9) one has also v = ValG (H; v ), which together with
ValG (H; v ) ValG (H; ) for all G 2 gives ValG ( G(Wm" ) H; v ) v < 1,
i.e. the entailment relationship j=G(Wm" ) H does not hold true for the system G(Wm" ) too.
2
Therefore the entailment relation of G1 is dierent from the \intersection" of the entailment relations of all the systems Gm . It is, however, possible
to \reduce" also the entailment relation of G1 to an \intersection" of entailment relations of some \simpler" systems.
Proposition 10.1.7. An entailment relationship j=G1 H holds true in the
innite Go del systems G1 i the entailment relationships j=G(W ) H hold
true in all innitely many-valued Go del systems G(W ) with a countable
truth degree set W .
Proof: As in the previous proof one immediately sees that each entailment relationship j=G1 H which holds for the system G1 also holds for
each system G(W ) with W [0; 1].
And in the case that some entailment relationship j=G1 H does not
hold true for the system G1 , there again exists some [0; 1]-valuation with
ValG ( G1 H; v ) < 1. However, is already a W -valuation for some
suitable countable set W [0; 1], e.g. for W = rg( ) [ f0; 1g. Therefore
the entailment relationship j=G(W ) H does not hold true for the system
G(W ).
2
"
In some sense, thus, the entailment relation of the system G(W ) is the
simplest entailment relation for an innitely many-valued Go del system.
The entailment relation of the system G1 may, because of this last proposition, be considered as \more complicated". In any case it is \smaller" in
the precise sense that there are entailment relationships j=G H which hold
true for G(W " ) but not for G1 .
The situation for the entailment relations of innitely many-valued Go del
systems is even more complicated as already mentioned: not only do the systems G1 , G(W " ), and G(W # ) have pairwise dierent entailment relations,
there exist innitely many innitely many-valued Go del systems \in between" G1 and G(W " ) with pairwise dierent entailment relations, cf. [29].
Also the important compactness property varies for dierent innitely
many-valued Go del systems, as was shown in [29]. The main result is:
Theorem 10.1.5. For an innitely many-valued Godel system G(W ) the
compactness theorem holds true i the truth degree set W [0; 1] contains
a densely ordered4 subset.
4 This means a subset W W with at least two elements such that for any
elements a < b 2 W there exists some c 2 W with a < c < b.
284

The propositional Go del systems G have been dened for the set of basic
connectives J G = f^; _; ; !Gg via the characteristic logical matrices
hW ; et1 ; vel1 ; non0 ; seq 1 i :
Their rst-order versions are formed, as was done for the Lukasiewicz systems, by adding the two (standard) quantiers 8; 9 semantically dened as inmum and supremum, respectively, i.e. characterized by the conditions (2.15),
(2.16), which now become
ValGA (8xH; ) =def inf fValGA (H; [x=b]) j b 2 Ag ;
ValGA (9xH; ) =def supfValGA (H; [x=b]) j b 2 Ag :
Also here A means any interpretation and any A-assignment.
Again as in the case of the Lukasiewicz systems, the truth degree set
W0 = [0; 1] \ Q is not well suited because of the non-existence of suprema
and inma for (a lot of) subsets of W0 . And again here the standard solution
is to consider for the innitely many-valued case essentially only the system
G1 with truth degree set W1 .
For the nitely many-valued Go del systems Gm with m 2 all the
results mentioned in Section 3.4 are true, i.e. one has e.g. that:
the ultraproduct construction is available;
the compactness theorem holds true;
the downward as well as the upward Lo wenheim-Skolem theorems hold
true.
As in the propositional case, also for the rst-order systems Gm the
Rosser-Turquette method of axiomatization of Section 6.3 is not available because not all of the connectives Jt for t 2 Wm are denable in these
systems.
However, one can follow the approach of [530] toward the rst-order
Lukasiewicz systems Lm , either directly or a little bit modied as later
on in Section 14.2. This essentially means to take all the (substitution instances of the) tautologies of { or instead an adequate axiom system for {
the corresponding propositional system as axioms together with the rule of
detachment (MP) as inference rule, and to add some suitable further axiom schemata and inference rules. These usually are either the further axiom
schemata
8x(H1 !G H2 ) !G (8xH1 !G H2 ) ;
8x(G !G H ) !G (G !G 8xH ) ;
8x(H !G G) !G (9xH !G G) ;
8x(H1 !G H2 ) !G (H1 !G 8xH2 ) ;
285
8x H (x) !G H [x=t]
for all terms t which are substitutable5 for x in H ,

H [x=t] !G 9x H (x)
together with the rule of generalization (Gen), or this are the rst two of
these schemata together with the inference rules (Gena ), (Gens ), and (ExQa ),
(ExQs ) from (9.112) and (9.113), in both cases combined with the additional
assumption that the variable x does not occur free in G, and of course w.r.t.
the Go del implication !G .
We shall, however, not discuss these details here, instead we consider the
problem of the axiomatizability of the innitely many-valued Go del system
G1 in more detail.
The standard truth degree structure for G1 is the linearly ordered Galgebra
G1 = h[0; 1]; et1 ; vel1 ; seq 1 ; non0 ; 0; 1i
(or the the BL-algebra h[0; 1]; et1 ; vel1 ; et1 ; seq 1 ; 0; 1i in the terminology of the
later Chapter 14, but both structures are denitionally equivalent according
to Corollary 14.1.3).
Before we give the proof of the intended completeness theorem it is helpful
rst to prove some preliminary results.
Lemma 10.2.1. For each countable and linearly ordered G-algebra L there
exists some countable and linearly as well as densely6 ordered G-algebra L
which extends L and preserves all the (innite) inma and suprema which
exist in L.
Proof: By denition, in any densely ordered set hM; 6i there do not
exist any consecutive elements, i.e. no elements a < b for which there do not
exist some x 2 M with a < x < b. The most simple example of a densely
ordered set is the set of rationals with its usual ordering.
If the G-algebra L = hL; \; [; ; ; 0; 1i is already densely ordered, nothing has to be shown, and one simply takes L = L. Otherwise there exist
pairs of consecutive elements in L. The crucial idea now is to make L densely
ordered by lling in each one of the \gaps" between two consecutive elements
in L by a copy of the set of rationals, or { only seemingly a bit simpler { by
a copy of the set (0; 1)Q = Q \ (0; 1) of rationals from the open unit interval.
Therefore one has to look at the set L0 of all a 2 L which are lower members
5 This means essentially that the free occurrences of x in H are not within the
scope of a quantication of some (free) variable of the term t.
6 This means that for each pair of elements a; b of L with a < b there exists some
c 2 jL j with a < c < b.
286
of pairs of consecutive elements, i.e. for which there exists some a < b 2 L
such that a; b are consecutive elements. Then consider the set
[
L =def L f0g [ fag (0; 1)Q
a2L0
and provide it with the lexicographical ordering 5 characterized for any a; b 2

L and any rationals r; s with 0 r; s < 1 by
(a; r) 5 (b; s) , a < b _ (a = b ^ r s) :
Obviously L is linearly and densely ordered by 5, and has (0; 0) as its
universal lower and (1; 0) as its universal upper bound. L is also countable
if L has been countable (or nite). Furthermore 5 is a lattice ordering. Hence
5 determines in L a lattice meet and a lattice join which shall be denoted
by u; t.
It is immediately clear that the mapping f which is dened by
f : x 7! (x; 0) ; dom(f ) = L
is a 1-1 injection of L into L which is order preserving. Therefore f is an
isomorphic embedding of the lattice hL; \; [i into the lattice hL ; u; ti.
To make hL ; u; ti into a G-algebra in such a way that f is also an isomorphic embedding of the G-algebra L = hL; \; [; ; ; 0; 1i into an G-algebra
L , one has only to have in mind that in each Heyting chain, and hence in
each linearly ordered G-algebra the operations and ; are characterizable
from the lattice ordering, cf. Corollary 10.1.3.
So it nally remains to show that f preserves all the (innite) inma and
suprema which exist in L. Thus let A L, and let a = sup A in L. In the case
a 2 A one has f (a) 2 f [A] and thus also f (a) = sup f [A] in L . In the case
a 2= A one immediately has sup f [A] 5 f (a) in L . Assume sup f [A] < f (a)
in L . Then one has sup f [A] = (b; r) with r > 0, because in the case of
sup f [A] = (b; 0) the element b 2 L would be an upper bound of A in L with
b < a, a contradiction. However, if one considers any rational 0 < s < r one
has (b; s) < (b; r) and still f (x) 5 (b; s) for all x 2 A. Again a contradiction.
Therefore one has in L also in this second case f (a) = sup f [A]. This means
that f preserves all the suprema which exist in L. In the same way one shows
that f preserves all the inma which exist in L.
2
Lemma 10.2.2. For each countable and linearly as well as densely ordered
G-algebra L there exists an isomorphism f onto the G-algebra
G1 jQ = h[0; 1] \ Q ; et 1 ; vel1 ; seq 1 ; non0 ; 0; 1i
of all rationals from the unit interval, and this mapping f is an embedding
of L into G1 which preserves all the (innite) inma and suprema which
exist in L .
287
Proof: It is a well known fact from set theory that any two countable
sets are order isomorphic if they are linearly and densely ordered (and if
both have universal upper and lower bounds). Therefore the carrier L of
each countable and linearly as well as densely ordered G-algebra L is order
isomorphic with [0; 1]Q = Q \ [0; 1].
So let f be an order isomorphism from L onto [0; 1]Q. Then f is also an
isomorphism between the linearly ordered G-algebras L and G1 jQ because
all the operations in any Heyting chain, and hence in any linearly ordered
G-algebra are denable from the lattice ordering, cf. Corollary 10.1.3.
Therefore f is even an isomorphic embedding of L into G1 . Furthermore
[0; 1]\Q is a dense subset of [0; 1]. Hence it is a standard set theoretical result7
that inma and suprema of subsets of [0; 1] \ Q coincide in G1 jQ and in G1 .
2
We shall see in Chapter 14 that the system G1 is a particular case,

even a schematic extension, of rst-order basic t-norm logic (and hence also
of monoidal logic and of monoidal t-norm logic). The following approach
toward an adequate axiomatization shall therefore not start from the rstorder system G1 but from its counterpart GBTL
1 in the realm of basic t-norm
logic. This allows one to use freely the results of Section 14.2, particularly
the Extended Completeness Theorem 14.2.4 and the Lo wenheim-Skolem
Theorem 14.2.3.
According to Theorem 14.1.20 this counterpart GBTL
1 of G1 in the realm
of basic logic is the schematic extension of the rst-order basic t-norm logic
BTL by the axiom schema
H1 ! H1 & H1 :
Its corresponding logical calculus, which is given by the calculus K BTL and
the extension of its set of axioms by this schema, shall be denoted K GBTL .
Theorem 10.2.1 (G1-Completeness Theorem). For each consistent set
of ws, and each sentence H the following assertions are equivalent:
(i) H is K GBTL -derivable from .
(ii) H is valid in all G1 -models of .
(iii) H is valid in all safe L-interpretations A with a linearly ordered BLalgebra L which are also models of .
Proof: The equivalence of (i) and (iii) is a particular case of the Extended
Completeness Theorem 14.2.4.
(i) ) (ii) immediately follows from the Soundness Theorem 14.2.1(ii).
For (ii) ) (iii) assume that H is not valid in all safe L-interpretations
over linearly ordered G-algebras which are also models of . So let A be some
7 The details of this proof are left to the reader. Only elementary set theoretical
facts have to be used for a proof.
288
safe L-interpretation over a linearly ordered G-algebra L such that H is not

valid in A.
Then ;H is valid in A. This means that the set [ f;H g of ws has a
model. Therefore [f;H g has also, according to the Lo wenheim-Skolem
Theorem 14.2.3, some countable model A0 which is a safe L-interpretation
with a countable and linearly ordered BL-algebra L. In this L-interpretation
A0 the sentence H is not valid.
According to the previous two lemmata one can embed L into a countable
and linearly as well as densely ordered G-algebra L in such a way that this
embedding preserves all the (innite) inma and suprema which exist in L.
This means, however, that A0 is also a safe L -interpretation, and that for
each A0 -assignment and each w G the truth degree ValGA (G; ) is the
same, independent of whether one considers A0 as an L-interpretation or as
an L -interpretation.
Now it is a routine matter to change from the L -interpretation A0 to an
G1 jQ -interpretation A simply by exchanging isomorphically the truth degree structure. Then A has the same universe of discourse as A0 , and the sets
of valid ws coincide for both structures. Thus also the G1 jQ -interpretation
A is a model of [ f;H g.
And as a nal step one has again to extend the truth degree structure
of A from G1 jQ to G1 via an isomorphic embedding which preserves all
the (innite) inma and suprema which exist in G1 jQ . As before this means
that also the G1 -interpretation A is a model of [ f;H g, i.e. is a model
of which is not a model of H .
2
Hence one has an adequate axiomatization for the rst-order Go del system G1 . And this is a completely dierent situation as for the rst-order
Lukasiewicz system L1 , cf. Section 9.3, and the rst-order product logic
, cf. Section 11.2.
But even this axiomatization result for the rst-order Go del system G1
essentially depends upon the choice of the truth degree set. If one takes
instead of [0; 1] the innite set

W = n +1 1 j n 2 N [ f0g
as set of truth degrees, then the resulting rst-order Go del logic G1 (W )
is not adequately axiomatizable, cf. [27]. This means also that the sets of
logically valid ws do not coincide for the systems G1 = G1 ([0; 1]) and
G1 (W ).
Independently of the choice of an innite truth degree set one has as
a further result that each innitely many-valued Go del system G1 (W ),
with some truth degree set satisfying 0; 1 2 W [0; 1], is not adequately
axiomatizable if it allows one to dene the connective J1 , cf. [16].
And also the result about the dierence of the sets of logically valid ws
for G1 ([0; 1]) and G1 (W ) can be generalized: one can construct an innite sequence of truth degree sets W (k) , k 1, such that any two systems
0
289
G1 (W (k) ) and G1 (W (j) ) with k 6= j have dierent sets of logically valid

ws, cf. [16].
290
11. Product Logic
11.1 The Propositional System

According to Theorem 5.1.1 each continuous t-norm is either the t-norm
tG = et1, or it is an Archimedean t-norm, or the ordinal sum of (continuous)
Archimedean t-norms. And each continous Archimedean t-norm is isomorphic
either to the Lukasiewicz t-norm tL = et2 or to the arithmetic product
tP = et3, cf. Theorem 5.1.3.
There are other reasons too which give a kind of prominence to these three
t-norms tG = et1 ; tL = et2 , and tP = et3 . Among the t-norms, i.e. among a
wide class of truth degree functions for conjunction connectives of systems
of many-valued logic, the Go delian t-norm tG = et1 is the only idempotent one. Furthermore the Lukasiewicz t-norm tL = et2 is prototypical for
a continuous t-norm with zero divisors, and it is also (essentially together
with the Lukasiewicz negation) prototypical for a continuous t-norm (and
a negation function) satisfying a coded form of the \law of excluded middle", cf. Theorem 5.2.2. In a similar sense the arithmetic product tP = et3 is
prototypical for the continuous strict t-norms according to Theorem 5.1.3.
Regarding corresponding logical systems, tG is basic for the Go del systems, and tL is basic for the Lukasiewicz systems. However, up to now we
have not particularly looked at logical systems for which the t-norm tP = et3
is basic.
This mainly re ects the historical development: the Lukasiewicz systems
were introduced in the 1920s, cf. [339], and the Go del systems in the paper
[199] of 1932. A logical system with the arithmetic product tP = et3 as its
basic t-norm, however, was rst investigated in 1996 in [248].
In any case, truly many-valued logical systems which are based on the
t-norm tP = et3 have to be innitely many-valued systems because no nite
subset of [0; 1], besides f0; 1g, is closed under et3 .
The standard approach toward a product-based many-valued logic is,
to take the whole unit interval as truth degree set and to have truth degree
1 as the only designated one:
W = W1 and D = f1g :
The sets J of basic connectives and K of basic truth degree constants are
292
11. Product Logic
J = f; ! g
K = f0 g ;
(11.1)
with the truth degree 0 denoted by 0, and with corresponding truth degree
functions
et3 ; seq ;
as dened in (5.10), and as characterized by the equation

1; if x y
(11.2)
seq (x; y) = y otherwise.
x
It is a simple exercise to see that the implication ! , characterized by the
equation (11.2), is the R-implication to the t-norm tP = et3 . Therefore the
truth degree function seq of ! is an implication function which satises
the right isotonicity condition and the exchange principle, and has the degree
ranking property, cf. Theorem 5.4.1. And the negation is not only the same
negation as one has in the Go del systems, it is also characterized by the
relation
non0 (x) = seq (x; 0) :
These basic connectives and truth degree constants can be used to enrich the
language L with further connectives and truth degree constants, which all
together prove to be useful for giving very readable ws later on. The usual
denitions are for any ws H1 ; H2 of L :
H1 ^ H2 =def H1 (H1 ! H2 ) ;
H1 _ H2 =def ((H1 ! H2 ) ! H2 ) ^ ((H2 ! H1 ) ! H1 ) ;
H1 $ H2 =def (H1 ! H2 ) (H2 ! H1 ) ;
H1 =def H1 ! 0 ;
1 =def p0 ! p0 :
It is a routine matter to check that the truth degree constants 1,0 denote
the truth degrees 1 and 0, and that one has furthermore the truth degree
functions of these additional connectives characterized by
ver^ (x; y) = minfx; yg ;
ver_ (x; y) = maxfx; yg ;
ver$ (x; y) = seq (x; y) seq (y; x)
= minfseq (x; y); seq (y; x)g ;

ver (x) = non0 (x)
for all x; y 2 [0; 1].
The connectives; ^ are two conjunction connectives for the system .
So the situation is similar to the Lukasiewicz systems. However, for the
present purposes it suces to consider only one disjunction connective _. For
avoiding too many parentheses in the ws, the negation shall, as usual,
293
have preference over ; ^; _, and all these connectives shall have preference
over ! ; $.
We shall not discuss particular tautologies here1 but concentrate on the
problem of axiomatization for the system .
A logical calculus K for the axiomatization of , as given in [248], has
an axiom system Ax determined by the following axiom schemata:
(Ax 1) H1 ! (H2 ! H1 ) ;
(Ax 2) (H1 ! H2 ) ! ((H2 ! H3 ) ! (H1 ! H3 )) ;
(Ax 3) 0 ! H1 ;
(Ax 4) H1 H2 ! H2 H1 ;
(Ax 5) H1 (H2 H3 ) ! (H1 H2 ) H3 ;
(Ax 6) (H1 H2 ) H3 ! H1 (H2 H3 ) ;
(Ax 7) ((H1 H2 ) ! H3 ) ! (H1 ! (H2 ! H3 )) ;
(Ax 8) (H1 ! (H2 ! H3 )) ! ((H1 H2 ) ! H3 ) ;
(Ax 9) (H1 ! H2 ) ! (H1 H3 ! H2 H3 ) ;
(Ax 10) H3 ! ((H1 H3 ! H2 H3 ) ! (H1 ! H2 )) ;
(Ax 11) (H1 ! H2 ) ! ((H1 ! H3 ) ! (H1 ! H2 ^ H3 )) ;
(Ax 12) (H1 ! H3 ) ! ((H2 ! H3 ) ! (H1 _ H2 ! H3 ) ;
(Ax 13) (H1 ! H2 ) _ (H2 ! H1 ) ;
(Ax 14) H1 ^ H1 ! 0 ;
and has the rule of detachment (MP) as its only inference rule.
Theorem 11.1.1 (Soundness Theorem). Each theorem of the logical calculus K is a -tautology.
Proof: By straightforward calculations one gets that each axiom which
falls under one of the schemata (Ax 1); : : : ; (Ax 13) is a -tautology, i.e.
assumes the truth degree 1 under each valuation.
And as usual also in most other systems, discussed in this book, the rule
(MP) leads from -tautologies H1 ; H1 ! H2 to a -tautology H2 .
2
If one compares the product logic with the innitely many-valued systems L1 ; G1 of Lukasiewicz and Go del one recognizes that all the axioms of the set Ax , which do not fall under one of the schemata (Ax 10)
or (Ax 14), are also tautologies2 of L1 and G1 . And in the Go del system
G1 even all instances of axiom schema (Ax 14) become G1 -tautologies.
And as for the Lukasiewicz systems, also for product logic the semantic deduction theorem fails. It is easy to recognize that one has H j= H H
for any w H and the standard entailment relation j= of . However, one
1 The system is a particular system with t-norm based connectives and thus
belongs to the type of systems discussed later on in Chapters 13 and 14. Therefore
all the logically valid ws mentioned there are also tautologies of the present
system.
2 Of course, one has to read ; as &; : in L1 and as ^; in G1 , and has to
read ! in both cases as the corresponding implication connective !L, !G .
294
11. Product Logic
has 6j= (H ! H H ) for each w H which is not a -tautology. Even

more is the case: classical logic is the only extension of product logic for
which this deduction theorem holds true, cf. [2].
To approach also the completeness of the logical calculus K we follow
the same algebraic method which was applied already for the systems L1
and G1 : we introduce a suitable class of algebraic structures such that the
property of being a -tautology becomes equivalent to the property of being
valid in all algebraic structures of this class, and such that the Lindenbaum
algebra of the logical calculus K belongs to this class.
The Lindenbaum algebra L of K is, as usual, determined by the
relation of provable equivalence w.r.t. the logical calculus K . To discuss it,
we denote the relation of provability in K by ` .
Lemma 11.1.1. The relation of provable equivalence w.r.t. K , dened
by
H1 H2 i
` (H1 ! H2 ) and ` (H2 ! H1 ) ;
is an equivalence relation in the class of all ws of L .
Proof: By denition, is a symmetric relation. That it is also transitive follows from axiom schema (Ax 2) via the rule (MP).
So it remains to show the re exivity of , i.e. to show ` (H1 ! H1 )
for each w H1 . However, one gets ` (H1 H2 ! H1 ) immediately from
(Ax 8) via (MP), and has thus ` (H2 ! (H1 ! H1 )) from (Ax 8).
Choosing H2 as any axiom then gives ` (H1 ! H1 ).
2
The equivalence classes of in the class of ws of L shall be denoted
by [H ] for each w H . The basic connectives and truth degree constants of
determine the operations in L in the usual way representativewise:
[A] [B ] =def [A B ] ;
[A] ! [B ] =def [A ! B ] ;
0 =def [0] :
What has to be shown is that these are suitable denitions, i.e. that the
operations ; ! are dened independent of the particular choice of the representatives. So suppose A A0 for ws A; A0 of L , i.e. suppose that
` (A ! A0 ) and ` (A0 ! A). Then one gets ` (A B ! A0 B )
and ` (A0 B ! A B ) from (Ax 9) via an application of (MP). This
means [A] = [A0 ] ) [A B ] = [A0 B ]. Similarly one shows that [B ] = [B 0 ]
implies [A B ] = [A B 0 ], and hence that one has
[A] = [A0 ] ^ [B ] = [B 0 ] ) [A B ] = [A0 B 0 ] :
From A A0 one gets via (Ax 2) the fact that [A ! B ] = [A0 ! B ].
Similarly one gets [A ! B ] = [A ! B 0 ] from B B 0 , and hence
[A] = [A0 ] ^ [B ] = [B 0 ] ) [A ! B ] = [A0 ! B 0 ] :
295
As an algebraic structure, the Lindenbaum algebra of K hence is the structure L = hL = ; ; !; 0i.
The algebraic structures we intend to consider in connection with the
product logic shall have the same similarity type as this Lindenbaum
algebra L , are called product algebras and determined in the following
way.
Denition 11.1.1. An algebraic structure A = hA; ; !; 0i with similarity type h2; 2; 0i is a product algebra i it has, together with the additional
operations u; t; ; and an additional constant 1 dened by
x u y = x (x ! y) ;
x t y = ((x ! y) ! y) u ((y ! x) ! x) ;
;x = x ! 0 ;
1 = 0 ! 0;
the properties:
(PrA1) hA; u; t; 0; 1i is a lattice with zero and unit element.
(PrA2) hA; ; 1; 6i is a commutative lattice ordered semigroup with
neutral element 1 w.r.t. the lattice ordering 6 of hA; u; t; 0; 1i.
(PrA3) ; ! form an adjoint pair, i.e. satisfy for all x; y; z 2 A the
condition:
z 6 (x ! y) , x z 6 y :
(PrA4) For all x; y; z 2 A one has
(a) (x ! y) t (y ! x) = 1 ;
(b) ; ; z (x z ! y z ) 6 (x ! y) ;
(c) x u ;x = 0 ;
(d) x (y u z ) = (x y) u (x z ) ;
(e) x (y t z ) = (x y) t (x z ) :
And such a product algebra A is a product chain i the lattice ordering 6 of
A is a linear ordering.
Having in mind that in a lattice each inequality a 6 b can equivalently be
written e.g. as the equation a u b = a, an inspection of this denition shows
that all the dening conditions for product algebras can be written down as
(universally quantied) Horn formulas of the form
A1 ^ : : : ^ Ak;1 ) Ak
with term equations A1 ; : : : ; Ak . This means, in algebraic terms, that the
class of all product algebras is a quasivariety, and hence closed under the
formation of subalgebras and direct products, cf. e.g. [59, 110, 225].
The product algebras t well into the situation of the product logic . It
is easy to recognize that the structure hW1 ; et3 ; seq ; 0i is a product algebra,
296
11. Product Logic
the unit interval product algebra, and is even a product chain. Additionally
one has the following result.
Proposition 11.1.1. The Lindenbaum algebra L = hL = ; ; !; 0i
of the logical calculus K for the product logic is a product algebra.
Proof: The proof of the dening conditions for product algebras, which
has to be given here for the Lindenbaum algebra L , amounts to show the
K -provability of certain ws of L . For this goal some preliminary remarks
are appropriate.
It is obvious from the denition of the -connectives ^; _; , and the
truth degree constant 1, as well as from the denitions of the product algebra
operations, that one has in L for all ws A; B of L :
[A] u [B ] = [A ^ B ] ;
;[A] = [ A] ;
[A] t [B ] = [A _ B ] ;
1 = [1] :
Furthermore one gets from (Ax 2) immediately for any ws A; B; C the rule
of syllogism in the form:
if ` (A ! B ) and ` (B ! C ) , then ` (A ! C ) ;
and has also the following version of a method of proof by cases
if ` ((A ! B ) ! C ) and ` ((B ! A) ! C ) , then ` C ;
which results via (Ax 13) from the fact that
((A ! B ) ! C ) !
! (((B ! A) ! C ) ! (((A ! B ) _ (B ! A)) ! C )
is a particular case of (Ax 12).
If one takes B A for H1 and A B for H2 in (Ax 2), one gets with C
for H3 via (Ax 4)
` (A B ! C ) ! (B A ! C ) ;
and then from (Ax 8) and (Ax 7) the importation/exportation principle
(A ! (B ! C ) (A B ! C ) ;
and by the rule of syllogism also the exchange principle in the form:
` (A ! (B ! C ) ! (B ! (A ! C )) :
As an important application one gets from (Ax 7) the further result
` ((H2 ! H3 ) ! ((H1 ! H2 ) ! (H1 ! H3 ))) :
(11.3)
From importation/exportation one has ` (A B ! A), and gets via
(Ax 4) also ` (A B ! B ). Hence one has by denition of ^ on the one
hand also ` (A ^ B ! A). And from ` ((A ! B ) ! (B ! A)) one
297
thus has ` ((A ! B ) A ! B ), hence ` (A (A ! B ) ! B ) via

(Ax 4), and thus ` (A ^ B ! B ) by denition of ^.
From the instance 0 ! (A ! 0) of (Ax 1) one thus has ` (A !
(0 ! 0) for each w A, which means ` (A ! 1). On the other hand one
gets ` (1 ! A) from ` A via (Ax 1). This means that ` A implies
[A] = [1]. Conversely one gets ` (1 ! A) from [A] = [1], and thus ` A
because of ` 1 (which just means ` (0 ! 0) and holds because of the
re exivity of ). Thus one has
` A , [A] = [1] :
From (Ax 4) one has ` (A B ! B A) and at the same time also
` (B A ! A B ) for any ws A; B , and hence the commutativity of
the operation in L .
Similarly one gets from (Ax 5) and (Ax 6) the associativity of the operation in L .
The commutativity of u follows because one has ` ((A ^ B ! B ) !
((A ^ B ! A) ! (A ^ B ! B ^ A) from (Ax 11), and of course the
K -provability of a corresponding w with A; B exchanged. And this gives
the commutativity of t by denition of t.
For the associativity of u remember that one has ` (H1 ^ H2 ! Hi ) for
all ws H1 ; H2 and i = 1; 2. This immediately gives ` (A ^ (B ^ C )) ! A,
and it gives by iterated application also ` (A ^ (B ^ C )) ! B as well as
` (A ^ (B ^ C )) ! C . Thus one has ` (A ^ (B ^ C )) ! A ^ B via
(Ax 11) and therefore also ` (A ^ (B ^ C )) ! ((A ^ B ) ^ C )). Similarly
one gets ` ((A ^ B ) ^ C ) ! (A ^ (B ^ C )), and hence the associativity of
u.
In a similar way one gets the associativity of t, using (Ax 12) together
with ` (Hi ! H1 _ H2 ), which holds true for all ws H1 ; H2 and i = 1; 2.
Here ` (H1 ! H1 _ H2 ) comes from ` (H1 ! ((H1 ! H2 ) ! H2 ),
which results from ` (H1 ^ H2 ! H2 ) via importation/exportation, and
from ` (H1 ! ((H2 ! H1 ) ! H1 ), which immediately results from
(Ax 1).
By ` (H1 ! H1 _ H2 ) one has particularly ` (A ! A _ (A ^ B )).
And ` (A _ (A ^ B ) ! A) comes via (Ax 12) from ` (A ! A) and
` (A ^ B ! A). In a similar way one gets also ` (A ! A ^ (A _ B ))
and ` (A ^ (A _ B ) ! A). This means that the operations u; t in L also
satisfy the absorption laws [A] = [A] u ([A] t [B ]) and [A] = [A] t ([A] u [B ]).
Therefore hL = ; u; ti is a lattice, and hL = ; i is a commutative
semigroup.
We next discuss the lattice ordering 6 of this lattice, and claim that one
has
[A] 6 [B ] , ` (A ! B ) :
(11.4)
298
11. Product Logic
Because [A] 6 [B ] means [A] = [A] u [B ] = [A ^ B ], and because one has

` (A ^ B ! B ) in any case, one has [A] 6 [B ] , ` (A ! A ^ B ), and
(11.4) reduces to
` (A ! A ^ B ) , ` (A ! B ) :
One has, however, ` ((A ! A) ! ((A ! B ) ! (A ! A ^ B )))
from (Ax 11), and thus gets ` (A ! A ^ B ) from ` (A ! B ). But
assuming ` (A ! A ^ B ) leads immediately to ` (A ! B ) because of
` (A ^ B ! A). And this establishes (11.4).
Now it is clear from (Ax 3) that [0] is the universal lower bound of the
lattice hL = ; u; ti, and [1] is its universal upper bound, because one has
generally ` (A ! 1) as mentioned previously. Hence condition (PrA1) is
established for the Lindenbaum algebra L of K .
Next we show that [1] is a neutral element w.r.t. the operation , i.e. that
one always has [A] = [A] [1]. The provability ` (A 1 ! A) is obvious,
and ` (A ! A 1) results via ` 1 from ` (1 ! (A ! A 1)),
which comes via importation/exportation from ` (1 A ! A 1)).
Thus we are done with the proof that L satises condition (PrA2) if we
can show [A] 6 [B ] ) [A] [C ] 6 [B ] [C ]. By (11.4) this follows, however,
immediately from (Ax 9).
From all these discussions we have now that [C ] 6 [A] ! [B ] holds true i
` (C ! (A ! B )), i.e. i ` (C A ! B ), and thus i [A][C ] 6 [B ].
And this means that also condition (PrA3) is satised for L .
This leaves only to consider condition (PrA4). Here item (a) comes immediately from (Ax 13). Item (b) comes from (Ax 10) via the importation/exportation law and (11.4). And item (c) comes from (Ax 14). Hence
one nally has to discuss the distributivity laws (d) and (e). From the monotonicity of under 6 one immediately has
[A] ([B ] u [C ]) 6 inf f[A] [B ]; [A] [C ]g = ([A] [B ]) u ([A] [C ]) ;
[A] ([B ] t [C ]) > supf[A] [B ]; [A] [C ]g = ([A] [B ]) t ([A] [C ]) :
So it remains for (d) to show ([A] [B ]) u ([A] [C ]) 6 [A] ([B ] u [C ]), or
also ([B ] [A]) u ([C ] [A]) 6 ([B ] u [C ]) [A] by the commutativity of .
According to (11.4) and the denition of u, this means to show ` G for the
w
G (B A ! C A) ! (B A ! (B ^ C ) A) :
This problem shall be approached via the method of cases we have available:
we intend to establish ` ((B ! C ) ! G) and ` ((C ! B ) ! G).
First we have ` ((B ! C ) ! (B ! B ^ C ), because one has
` (H1 ! (H2 ! H2 H1 )) from ` (H1 H2 ! H2 H1 ) by
importation/exportation. And one has
` ((B ! B ^ C ) ! (B A ! (B ^ C ) A)
immediately from (Ax 10). Hence it results
` ((B ! C ) ! (B A ! (B ^ C ) A))
299
(11.5)
by the rule of syllogism. Thus one has also

` ((B ! C ) (B A ! C A) ! (B A ! (B ^ C ) A)) ;
because one has in general ` ((H1 H2 ! H1 ). And this gives immediately
` ((B ! C ) ! G) via importation/exportation.
On the other hand one has ` ((C ! B ) ! (C A ! (C ^ B ) A))
immediately from (11.5), because one can exchange B; C within this formula.
Thus one gets ` ((C ! B ) ! (C A ! (B ^ C ) A), using suitably
the provable commutativity of ^, i.e. the commutativity of u. Then one gets
the second case ` ((C ! B ) ! G) from a suitable application of (11.3).
In summary one thus has ` G. And this means that also (PrA4)(d) is
established.
And for (e) it remains to show [A] ([B ] t [C ]) 6 ([A] [B ]) t ([A] [C ]).
This goal is, by denition and commutativity of t and by the order theoretic
characterization and commutativity of u, however, already reached if one
shows ([B ] t [C ]) [A] 6 (([B ] [A] ! [A] [C ]) ! [A] [C ]). And this
means to have
` ((B _ C ) A ! ((B A ! A C ) ! A C ))
for any ws A; B; C . According to the exchange principle this means to derive
` ((B A ! A C ) ! ((B _ C ) A ! A C )) :
Using importation/exportation and using the notation G A ! A C
one thus has to show
` ((B ! G) ! (B _ C ! G)) :
(11.6)
Yet one has immediately ` (C ! (A ! A C )), which means nothing
but ` (C ! G), and gets therefore (11.6) from (Ax 12) via an application
of the exchange principle. Hence also (PrA4)(e) is established and the whole
proof nished.
2
The product chains have a similar basic rôle among the product algebras,
as MV-chains have among the MV-algebras, and as Heyting chains have
among Heyting algebras. To prove a corresponding result to the former
Theorem 9.2.2, we rst look at lters3 in product algebras.
3 In suitable lattice-ordered algebraic structures, lters and ideals are essentially
dual objects which satisfy corresponding closure conditions and are, respectively,
upward or downward closed. The fact that for MV-algebras the operation {
a kind of \addition" { is a primitive one, makes ideals the most suitable choice
in that case. The fact that for product algebras the operation { a kind of
\multiplication" { is a primitive one, now makes lters the more suitable objects
for the following discussion, as was already the situation with Heyting algebras.
300
11. Product Logic
Denition 11.1.2. A nonempty subset F 6= ; of a product algebra A is a

lter of A i it is upward closed w.r.t. the natural ordering 5 and also closed
under , i.e. i one has for all a; b 2 A:
a 5 b ^ a 2 F ) b 2 F;
a; b 2 F ) a b 2 F :
(11.7)
A lter F of A is proper i one has F =
6 A. Such a proper lter F is prime
i it satises for all a; b 2 A the condition:
(a ! b) 2 F _ (b ! a) 2 F :
(11.8)
For each lter F of a product algebra A = hA; ; !; 0i one has 1 2 F .
And for each element a 2 A the set fb 2 A j a 6 bg is a lter of A, the
principal lter of a. For a =
6 0 the principal lter of a is a proper lter. As
an immediate corollary one gets that the union of each chain of lters, i.e. of
each set of lters linearly ordered by inclusion, is again a lter. And for each
lter F of A an equivalence relation F in A is determined by the denition
a F b =def (a ! b) 2 F ^ (b ! a) 2 F :
; ! of A. Therefore it is a routine matter to prove that the quotient structure A=F , with its operations ; ! dened between equivalence classes [a]; [b]
by
[a] [b] = [a b] ; [a] ! [b] = [a ! b] ;
and with the designated element o = [0], is again a product algebra.
Lemma 11.1.2. Let A be a product algebra and F one of its lters. Assume
that 1 6= a 2 A. Then one has:
(i) If F is a prime lter then the quotient A=F is a product chain.
(ii) For each 1 6= a 2 A there exists a prime lter Fa with a 2= Fa .
Proof: (i) Suppose rst that F is a prime lter, and consider some
[a]; [b] 2 A=F . Then one has (a ! b) 2 F or (b ! a) 2 F . Firstly let
(a ! b) 2 F . Then one has because of 1 2 F :
[a ! b] = [1] = [0 ! 0] = e ;
denoting here by e the universal upper bound of the lattice ordering of A=F .
This gives
[a] u [b] = [a] [a ! b] = [a] e = [a] ;
and hence [a] 6 [b]. Similarly (b ! a) 2 F gives [b] 6 [a]. Hence 6 is a linear
ordering.
(ii) Let 1 6= a 2 A. It is immediately clear from Denition 11.1.2 that the
intersecion of each inclusion-ordered chain of lters of a product algebra A
(which do not contain a 2 A) is again a lter of A (which does not contain
a 2 A). Therefore Zorn's Lemma can be applied to the class Fa of all proper
301
lters F of A with property a 2= F , i.e. Fa has an inclusion-minimal member

Fa . By denition of Fa , the lter Fa is proper. It is, however, prime.
Assume that Fa is not prime. Then there exist b; c 2 A such that (b !
c) 2= Fa as well as (c ! b) 2= Fa . Consider the smallest lters F1 , F2 which
extend Fa and contain either b ! c, or c ! b, respectively. It is easy to
recognize that one has F1 = fx 2 A j 9 y 2 F 9 n 2 N (y (b ! c)n 6 x)g,
with the power notation here indicating the iteration of . One has a similar
representation for F2 . The crucial points are that a lter F which contains
b ! c also contains y (b ! c)n for each y 2 F and each n 2 N , and that
F1 is closed under because of
(y1 (b ! c)n ) (y2 (b ! c)m ) = (y1 y2 ) (b ! c)n+m :
These lters F1 ; F2 have the property that one has either a 2= F1 or a 2= F2 .
Otherwise one would have d (b ! c)n 6 a and d (c ! b)n 6 a for some
suitable n 2 N and some d 2 Fa , which means
d = d e = d ((b ! c) t (c ! b)) = d ((b ! c)n t (c ! b)n )
= (d (b ! c)n ) t (d (c ! b)n ) 6 a ;
using the distributivity of over t, and thus means a 2 Fa , a contradiction.
Hence Fa is a prime lter.
2
Proposition 11.1.2. Each product algebra is (isomorphic to) a subdirect

product of product chains.
Proof: Let A beTa product algebra and FA the class of all its prime
lters. Then one has FA = f1g by Lemma 11.1.2(ii), and thus
\
fF j F 2 FA g = f(x; x) j x 2 Ag = idA ;
because if one has a F b, i.e. (a ! b) 2 F and (b ! a) 2 F for each prime
lter F , then one has (a ! b) = 1 and (b ! a) = 1, hence a u b = a and
b u a = b, i.e. a 6 b and b 6 a, which means a = b.
By a general result of universal algebra, cf. e.g. [59, 225], it follows from
this property of the class of all congruence relations F for F 2 FA , that
A is isomorphic to a subdirect product of the family of all product algebras
A=F for F 2 FA . Hence the claim follows from Lemma 11.1.2(ii).
2
Again thus, as in the case of MV-algebras, the totally ordered product
algebras, i.e. the product chains are of particular importance. And, again as
in this previous case, one has essentially the same close relationship between
product algebras and particular o-groups, as one knows from Section 9.2
between MV-algebras and particular o-groups.
Proposition 11.1.3. For each product chain A = hA; ; !; 0i there exists
an o-group G = hG; G ; o; G i with neutral element o and an \isomorphism"
between the non-positive part N (G ) = fx 2 G j x G og of G and the set
302
11. Product Logic
A n f0g, i.e. an 1-1 mapping ' : N (G ) ! A n f0g such that one has for all
x; y 2 N (G ):
'(o) = 1 ; '(x G y) = '(x) '(y) ; x G y , '(x) 6 '(y) :
Proof: Let A = hA; ; !; 0i be a product chain. Then one has x y 2
A n f0g for all x; y 2 A n f0g, because x y = 0 gives x y 6 0 and thus
y 6 (x ! 0) = ;x by the adjointness condition, which means y = ;x = 0
for x 6= 0 by x u ;x = 0 and the linearity of 6. Hence is an associative
and commutative operation in A n f0g. Therefore the algebraic structure
A; = hA n f0g; ; 1; >i is an ordered commutative semigroup with neutral
and minimal element 1. Furthermore one has that for all x > y 2 Anf0g there
exists some z 2 A n f0g such that x z = y: the element z = x ! y because
of x (x ! y) = x u y = y. And this element z is uniquely determined: using
the facts that one has ;x = 0 and hence ; ; x = 1 for all x 2 A n f0g,
and that in A always u = v is equivalent with u 6 v ^ v 6 u and hence
with (u ! v) = 1 ^ (v ! u) = 1, this uniqueness follows immediately from
condition (b) of (PrA4).
Ordered semigroups with these properties satisfy all the properties which
one needs of the ordered semigroup hN ; +; 1; i of the natural numbers to
make the (set theoretic) construction of the integers. Therefore this construction can (similarly) be carried through also for A; = hA nf0g; ; 1; >i,
cf. e.g. [177]. The result is an o-group G ; = hG; ; o; vi with neutral element o which contains as its non-negative part P (G ; ) = fx 2 G j o v xg an
isomorphic copy of A; .
Suppose that ' is an isomorphism from P (G ; ) onto A n f0g. Then one
has immediately '(o) = 1 and always '(x y) = '(x) '(y). Furthermore
one has always '(x) > '(y) i x v y, and hence '(x) 6 '(y) i y v x.
It is a routine matter to check that with each o-group hM; ; e; i also
the group hM; ; e; i with the inverse ordering is an o-group. Therefore we
consider now the o-group G = hG; ; o; G i with the ordering relation G =
v;1 . Then one has
N (G ) = fx 2 G j x G og = fx 2 G j o v xg = P (G ; ) ;
which means that ' is also an isomorphism from N (G ) onto A nf0g, satisfying
'(x) 6 '(y) i x G y for all x; y 2 N (G ).
2
Again as in this previous MV-case, also here the validity of equations can
be transferred between both types of structures, cf. [246].
Corollary 11.1.1. For each term t of the language of product algebras there
exists some term t of the language of o-groups, such that for each product
chain A and each A-assignment one has
(A; ) j= t1 = t2 , (N (G ); ) j= t1 = t2
(11.9)
for the corresponding o-group G according to Proposition 11.1.3, and the corresponding o-group equation t1 = t2 .
303
The proof proceeds as the corresponding proof for Proposition 9.2.5(i)

and shall be left to the reader.
Theorem 11.1.2 (Algebraic Completeness Theorem). A -equation
is valid in all product algebras i it is valid in the unit interval product algebra
hW1 ; et3 ; seq ; 0i.
Proof: It suces to show that a -equation which is valid in the unit
interval product algebra is valid in all product chains. Because it then is also
valid in all product algebras by Proposition 11.1.2.
So assume that some -equation t1 = t2 is not valid in some product
chain A = hA; ; !; 0i. Suppose that the variables of t1 ; t2 occur among
x1 ; : : : ; xn , and write therefore also ti (x1 ; : : : ; xn ) for ti . Then there exist
elements a1 ; : : : ; an 2 A such that one has t1 (a1 ; : : : ; an ) 6= t2 (a1 ; : : : ; an ) in
A. Consider the nite set D1 consisting of the universal bounds 0; 1 and of all
values of t1 ; t2 and their subterms under some (and hence each) assignment
with (x1 ) = a1 ; : : : ; (xn ) = an . Let D = D1 n f0g.
Now let, as in Proposition 11.1.3, G = hG; ; o; G i be an o-group with
neutral element o such that the non-positive part N (G ) = fx 2 G j x G og of
G is isomorphic to the set A nf0g. Assume that ' is an isomorphism between
both sets. Let D = ';1 [D ].
The set D is a nite subset of G. Hence there exists, according to Theorem
1.2.1 and the isomorphy of the o-groups R and R , some nite subset K of
the o-group R = hR+ ; ; 1; i of the positive reals under multiplication, and
a partial isomorphism which maps D onto K . Because of 1 2 D, and
because each isomorphism between two o-groups maps the neutral element
of the rst onto the neutral element of the second one, one has 1 = (1) 2 K .
Therefore one has K (0; 1] from 0 <G a G 1 for each a 2 D. Extend K to
K = K [ f0g, and extend the partial isomorphism to = [ f(0; 0)g.
Then one has for all x; y; z 2 D1 :
(x y) = (x) (y) ;
(x ! y) = seq ( (x) (y)) ;

x 1 y , (x) 2 (y) ;
This is obvious for all x; y; z 2 D because extends the partial isomorphism
, and because one has in the case x 6 y that (x ! y) = 1 and hence
(x ! y) = (1) = 1 = seq ( (x); (y)) because of (x) (y), and one
has in the case x > y that y = x u y = x (x ! y), and therefore also
(y) = (x) seq ( (x); (y)), which means (x ! y) = seq ( (x); (y))
because of (x) > (y).
So it remains to conrm these properties also for the case that the universal lower bound 0 of A is one of x; y; z . This, however, is a routine matter
having in mind that one has a 0 = a (a u ;a) = (a a) u (a ;a) =
(a a) u 0 = 0 for all a 2 A.
304
11. Product Logic
In summary thus we have an injection of D1 onto a subset K1 [0; 1]

which is an isomorphism w.r.t. the product chain A and the unit interval
product algebra hW1 ; et3 ; seq ; 0i. By choice of D1 and the isomorphism
properties one immediately has that the -equation t1 = t2 is not valid in
K1 , and hence not valid in the unit interval product algebra.
2
The relationship between o-groups, or even `-groups and product algebras
can be studied in more detail, and has its own particular interest. We shall not
enter into details, but only mention the interesting fact that a similar result
to the result of this last theorem can be proved, cf. [106]. To state it consider
for any abelian `-group G = hG; ; o; G i the subset G; = fx 2 G j x G og
together with some element O , and dene operations ; ! in G; [ fO g by

x; y 2 G;
x y =def xO y; ifotherwise,
8
< minfo; y ;1 xg; if x; y 2 G;
x ! y =def : o;
if x = O
O;
if x 2 G; and y = O :
It is a routine matter to check that the structure
P(G ) =def hG; [ fO g; ; !; O i
is a product algebra.
Proposition 11.1.4. A -equation E is valid in all product algebras i it
is valid in the particular product algebra P(Z ) based on the o-group Z =
hZ; +; 0; i of the integers under addition.
For product algebras, as for the case of MV-algebras, one can show that
among all the product chains the unit interval product chain is a \prototypical" product chain.
Corollary 11.1.2. Each product chain is locally embeddable into the unit
interval product chain hW1 ; et3 ; seq ; 0i.
The proof, given in [246], is similar to the corresponding proof of Corollary
9.2.7 for the MV-case, and rests essentially on Proposition 11.1.3, Corollary
11.1.1, and Theorem 11.1.2. The details shall not be presented here.
Theorem 11.1.3 (Completeness Theorem). For the product logic the
following claims are equivalent for each w H of L :
(i) H is a -tautology;
(ii) H is valid in each product algebra;
(iii) H is K -provable.
305
Proof: With each w H of the language L one can connect a term tH

of the language of product algebras by exchanging the connectives ; !
with the symbols ; ! for the basic operations in product algebras, and by
exchanging the truth degree constant 0 with the (name for the) designated
element 0 of a product algebra.
Assume now that the w H of L is a -tautology. Then the -equation
tH = 1(= ;0) is valid in the unit interval product algebra, and hence in each
product algebra according to Theorem 11.1.2. And this just means that H is
valid in each product algebra.
In the case that H is valid in each product algebra, then it is also valid
in the Lindenbaum algebra L of the logical calculus K according to
Proposition 11.1.1. Hence the -equation tH = 1 is valid in L , and this
means that H and 1 are K -provably equivalent. However, 1 is in L the
class of all K -provable ws. Therefore H is K -provable.
And each K -provable w H is a -tautology according to Theorem
11.1.1. Therefore the claims (i), (ii), (iii) are equivalent.
2
Finally we consider an interesting and important relationship between the
(innitely many-valued) Lukasiewicz system L1 and the product logic :
essentially saying that one can \nd L1 within ", cf. [25], but saying also
that the set of -tautologies is \not simpler" than the set of L1 -tautologies.4
The technical notion we need to formulate this result is that of an interpretation of some logical system S1 within another logical system S2 .
Denition 11.1.3. An interpretation of a logical system S1 within a logical
system S2 is a mapping of the set of ws of S1 into the set of ws of S2
such that one has for each w H of S1 :
j=S H ) j=S (H ) :
(11.10)
And such an interpretation is faithful i one even has for each w H of S1 :
j=S H , j=S (H ) :
(11.11)
For completeness it should be mentioned that besides this notion of interpretability of a logical system S1 within another logical system S2 one has
a parallel notion of (faithful) interpretability of a logical calculus K 1 within
another logical calculus K 2 . The corresponding denitions use simply K i derivability instead of Si -validity in the conditions (11.10) and (11.11). And
furthermore both notions of (semantical or syntactical) interpretability can
be extended to formalized theories over given logical systems or calculuses.
We consider the propositional systems L1 and with their basic vocabulary as introduced in (9.1) and (11.1). Remind (2.1), i.e. that they have the
common set V0 = fp0; p00 ; p000 ; : : :g of propositional variables. Let a preliminary translating function , which maps the set LL of p0 -free ws of LL into
L , be inductively dened by the cases:
4 In technical terms this means NP-completeness of the set of (propositional) 1
tautologies, cf. [25, 246].
306
11. Product Logic

8
<
p(n+1) _ p0 ;
(H ) ! p0 ;
: 1
(H1 ) ! (H2 ) ;
if H p(n+1) and n 1
if H :H1
if H H1 !L H2 .

And let a further translation function : LL ! L be dened by
(H ) =def p0 ! (H ) :
(11.12)
Before we prove that there exists a faithful interpretation of L1 within
some algebraic result should be mentioned which proves to be helpful for
the forthcoming proofs. It is also interesting in itself.
Lemma 11.1.3. For each 0 < a < 1 and the operation a with the characterizing equation
u a v =def maxfa; u vg
the algebraic structure L1 = h[0; 1]; et1 ; vel1 ; et2 ; seq 2 ; 0; 1i is isomorphic with
the algebraic structure a = h[a; 1]; et1 ; vel1 ; a; seq ; a; 1i.
Proof: One recognizes immediately that L1 is a lattice equipped with an
additional adjoint pair et2 ; seq 2 , i.e. L1 is a residuated lattice, cf. Denition
14.1.1. We consider the bijective mapping ga : [0; 1] ! [a; 1] dened by
ga (x) =def a1;x ;
(11.13)
and show that it has all the properties of an isomorphism from L1 onto a .
Then it is immediately clear that a is isomorphic with L1 , i.e. also a lattice
equipped with an additional adjoint pair a ; seq .
It is obvious (from elementary properties of the power function) that ga
is a bijection from [0; 1] onto [a; 1] because it is strongly increasing by the
choice of a. Therefore one has also always
ga (et1 (x; y)) = ga(minfx; yg) = minfga (x); ga (y)g = et1 (ga (x); ga (y)) ;
and the corresponding result for vel1 instead of et1 . And one immediately has
ga (0) = a1 = a as well as ga (1) = a0 = 1.
For et2 one has et2 (x; y) = 0 for x + y < 1, i.e. for 1 < (1 ; x) + (1 ; y).
But then a = ga(x) a ga (y) follows from a(1;x)+(1;y) < a1 = a and denition
(11.13). And in the other case 0 x + y ; 1 one has (1 ; x) + (1 ; y) 1,
which means a = a1 a(1;x)+(1;y) and therefore
ga (et2 (x; y)) = a1;(x+y;1) = a(1;x)+(1;y) = ga (x) a ga (y) :
Hence ga (et2 (x; y)) = ga (x) a ga (y) holds in any case.
For the residuation operation seq 2 one has seq 2 (x; y) = 1 in the case
x y. But then one has also ga (x) ga(y) and hence seq (x; y) = 1. And
in the case x > y one has
ga (seq 2 (x; y))=a1;(1;x+y) = a(1;x);(1;y)
=a(1;x)=a(1;y) = seq (ga (x); ga (y)) :
Hence also ga(seq 2 (x; y)) = seq (ga (x); ga (y)) holds in any case.
2
(H ) =def
307
Proposition 11.1.5. The translation function has for all ws H 2 LL the
property
j=L1 H , j= (H ) ;
and it has additionally the property that one has for each set of ws of LL
and each w H 2 LL :
j=L1 H , j= (H )
for the set = f p0g [ f (H ) j H 2 g of ws of L .
Proof: We rst show that for each (p0-free) w H of LL one has j=L1 H
i one has j= (H ).
So consider a valuation : V0 ! [0; 1]. If one has (p0 ) = 0 then one
has immediately Val ( p0 ; ) = 0 and hence Val ( (H ); ) = 1 from
denition (11.12). And if one has (p0 ) = 1, then an easy induction on the
complexity of H shows that one has Val ( (H ); ) = 1, and hence also
Val ( (H ); ) = 1.
Thus consider the case 0 < (p0 ) < 1, and let a = (p0 ). Combine with
two further valuations 0 ; 00 dened by
0 (p0 ) = 00 (p0 ) = (p0 ) ;
and for all propositional variables p(n+1) with n 1, i.e. dierent from p0 by
0 (p(n+1) ) = ga((p(n+1) )) ;
00 (p(n+1) ) = maxf(p(n+1) ); ag :
By induction on the complexity of the w H one gets for all valuations
and all ws H
ga (ValL (H; )) = Val ( (H ); 0 ) ;
(11.14)

00
Val ( (H ); ) = Val ( (H ); ) :
(11.15)
Now assume j=L1 H , and let be any valuation. In the cases (p0 ) = 0
and (p0 ) = 1 one has Val ( (H ); ) = 1 according to our previous remarks.
And in the case 0 < (p0 ) < 1 form the valuation determined by

0
p0
(q) = g;(p1();(p(n+1) )) ; ifif qq
(11.16)
p(n+1) for some n 1.
a
Then one has 00 = 0 , and therefore for v = Val ( p0 ; )
Val ( (H ); ) = seq (v; Val ( (H ); ))
= seq (v; Val ( (H ); 00 ))
= seq (v; Val ( (H ); 0 ))
= seq (v; ga (ValL (H; )))
= seq (v; 1) = 1 :
And this means j= (H ).
308
11. Product Logic
So assume on the other hand j= (H ) for some H 2 LL . Again let
be any valuation. Consider a valuation 0 which coincides with for all
propositional variables p(n+1) with n 1, and which satises 0 < 0 (p0 ) < 1.
Then one has for a = 0 (p0 ):
ValL(H; ) = ValL(H; 0 ) = ga;1 (Val ( (H ); 00 ))
= ga;1(1) = 1 ;
and therefore j=L1 H .
The additional property that one has 2 ModL ( ) i one has 2
Mod ( ) follows in a similar way via (11.14), (11.15), using again (11.16).
Corollary 11.1.3. There exists a faithful interpretation of L1 within .

Proof: Let H be any w of LL. Consider the new w (H ) which is
formed from H by simultaneously substituting p(n+1) for each propositional

variable p(n) which occurs in H . This mapping can be inductively characterized by the conditions
8
< p(n+1) ;
if H p(n)
(H ) =def : :(H1 ) ;
if H :H1
(H1 ) !L (H2 ) ; if H H1 !L H2 .
Because one obviously has that H is an L1 -tautology i (H ) is, and because
(H ) is p0 -free, i.e. (H ) 2 LL , one has
j=L1 H , j=L1 (H ) , j= ((H )) :
Hence is a faithful interpretation of L1 within .
2
Intuitively this result says that the product logic is at least as complicated as the innitely many-valued Lukasiewicz logic L1 .
11.2 The First-Order System

The language L of rst-order product logic is formed in the usual way, as it
was chosen for the Lukasiewicz and for the Go del systems: two quantiers
8; 9 are added to the standard vocabulary and understood as meaning the
inmum and the supremum, respectively. Hence they are characterized by
the conditions (2.15), (2.16), which now become
ValA (8xH; ) =def inf fValA (H; [x=b]) j b 2 Ag ;
ValA (9xH; ) =def supfValA (H; [x=b]) j b 2 Ag :
Here A means any safe L-interpretation over a linearly ordered product algebra L, and any A-assignment.
309
Propositional product logic is semantically characterized either by the

standard logical matrix hW1 ; et3 ; seq ; 0i, or by the class of all product
algebras. For rst-order product logic this result leaves open the possibilities
of a semantical characterization either by the class of all product algebras,
or by this standard logical matrix hW1 ; et3 ; seq ; 0i. Actually it seems to
be an open problem whether the set of ws which are valid in all safe Linterpretations A with a linearly ordered product algebra L coincides with
the set of all those well-formed formulas which are valid in the standard
logical matrix hW1 ; et3 ; seq ; 0i.
Therefore we distinguish two dierent forms of rst-order product logic:
the standard rst-order product logic s shall be characterized by the standard logical matrix L = hW1 ; et3 ; seq ; 0i. This means e.g. that the set
of s -logically valid ws is the set of all those ws which are valid in all
L -interpretations A, and that a w H is s-entailed by a set of ws i
each L -model of is also a model of H . And the weak rst-order product
logic w shall be characterized by the class of all safe L-interpretations A
with a linearly ordered product algebra L. This means similarly that the set
of w -logically valid ws is the set of all those ws which are valid in all
safe L-interpretations A over a linearly ordered product algebra L, and that
a w H is w-entailed by a set of ws i each L-model of , L any linearly
ordered product algebra, is also a model of H .
As it was the case already for the rst-order Go del system G1 in Section 10.2, also the weak rst-order product logic w is a particular case,
even a schematic extension, of rst-order basic t-norm logic (and hence also
of monoidal logic as well as of monoidal t-norm logic) as discussed later
on in Chapter 14. Thus one has available all the results of the later Section 14.2, particularly the Extended Completeness Theorem 14.2.4 and the
Lo wenheim-Skolem Theorem 14.2.3, also for the weak rst-order product
logic w .
According to the Extended Completeness Theorem 14.2.4 and Theorem
14.1.20, the weak rst-order product logic is just the schematic extension
of the rst-order basic t-norm logic BTL by the axiom schemata
H1 ^ H2 ! 0 ;
::H3 ! ((H1 & H3 ! H2 & H3 ) ! (H1 ! H2 )) :
Its corresponding logical calculus, which is given by the calculus K BTL and
the extension of its set of axioms by these schemata, shall be denoted K .
Hence one has now the following completeness theorem for the weak rstorder product logic.
Theorem 11.2.1 (Weak Completeness Theorem). For each w H and
each set of sentences the following assertions are equivalent:
(i) H is K -derivable from the set of premisses;
(ii) H is valid in all safe L-interpretations A with a linearly ordered product
algebra L which are also models of .
310
11. Product Logic
Actually there exist only a few further investigations devoted to rst-order

product logic, essentially explained in [246]. We shall concentrate upon one of
them which indicates that the standard rst-order product logic is at least as
complicated as rst-order Lukasiewicz logic.5 As in the propositional case
this goal is to be reached by giving a faithful interpretation of L1 within s .
Let therefore the rst-order languages LL and L for Lukasiewicz and
for product logic be chosen in such a way that all the predicate symbols of
LL are also predicate symbols of L (of course such that their arities in both
languages coincide). Assume furthermore that the language L contains also
all the individual constants as well as all the truth degree constants of the
language LL .
Suppose, however, that the language L contains a unary predicate symbol D0 and an individual constant d which do not belong to the vocabulary
of LL .
This means that all the terms and atomic ws of LL are also terms and
atomic ws of L .
Because one has from Proposition 9.3.1(iv) the (semantic) denability of
the quantier 9 from 8 and : in L1 , assume additionally that the basic
vocabulary of LL has only the quantier 8.
Now consider (again as in the propositional case) rst a preliminary translating function , which maps LL into L . In this rst-order case it shall
be inductively dened by the cases:
8
P (t1 ; : : : ; tn ) _ D0 (d) ; if H P (t1 ; : : : ; tn ) is atomic
>
<
(H1 ) ! D0 (d) ;
if H :H1
(H ) =def
(H1 ) ! (H2 ) ; if H H1 !L H2
>
:
8x (H1 ) ;
if H 8xH1 .
As in (11.12) again one denes from it the further translation function by
(H ) =def D0(d) ! (H ) :
Proposition 11.2.1. The translation function has for all ws H 2 LL the
property
j=L1 H , j= s (H ) ;
and it has additionally the property that one has for each set of ws of LL
and each w H 2 LL :
j=L1 H , j= s (H )
for the set = f D0(d)g [ f (H ) j H 2 g of ws of L .
Proof: Let 0 < a < 1 and ga : [0; 1] ! [a; 1] the function dened in
(11.13). Take any L1 -interpretation A, and form from it an L -interpretation
A+ for the extended language L , which contains also the unary predicate
5 In fact it is much more complicated as has been proved in [374] by recursion
theoretic methods which go beyond the scope of this book.
311
symbol D and the constant d. Do this by letting the universe of discourse of A

as well as the interpretations cA of the individual constants c of LL unchanged,
by choosing as new interpretation for each n-ary predicate symbol P of LL
the [0; 1]-valued predicate
P A = P A [ (jAjn fag) ;
(11.17)
i.e. by choosing as the truth degree to which some n-ary predicate symbol P
of LL applies in A+ to some n-tuple x of elements of jAj the maximum of a
and the truth degree to which P applies to x in A, and by choosing for the
new symbols the interpretations
DA = f(b; a)g [ f(x; 0) j b 6= x 2 jAjg ;
dA = b
for some (xed) b 2 jAj.
And form also another L -interpretation ga (A) for the extended language
which diers from A+ only in the interpretations (11.17) of the predicate
symbols of LL , which are now taken as
P ga (A) = f(x; ga (u)) j (x; u) 2 P A g [ (jAjn fag) ;
i.e. by taking now as the truth degree to which some n-ary predicate symbol
P of LL applies in ga (A) to some n-tuple x of elements of jAj the ga -value of
the truth degree to which P applies to x in A.
Similarly to (11.14), (11.15) one now gets for all A-assignments and all
ws H (again by induction on their complexity)
ga (ValLA (H; )) = Valga (A) ( (H ); ) ;
ValA ( (H ); ) = ValA ( (H ); ) :
Then the assertions follow as in the proof of Proposition 11.1.5.
2
From this result it is immediately clear that one has also:
Corollary 11.2.1. The mapping is a faithful interpretation of the rstorder system L1 within the standard rst-order product logic s .
+
312
11. Product Logic
12. The Post Systems
Systems of many-valued propositional logic have been considered independent of the Polish logician J. Lukasiewicz and almost at the same time, i.e.
in the beginning 1920s, also by the US-american logician E.L. Post. Contrary to the mainly philosophical motivations of Lukasiewicz, cf. Chapter
21, Post did develop his systems in the context of studies toward classical
propositional logic, e.g. toward functional completeness. Another dierence
to the Lukasiewicz approach was that Post from the very beginning has
discussed any nite set of truth degrees, whereas Lukasiewicz rst only
considered the three-valued case.
12.1 The Original Presentation

We shall denote by Pm for m 2 the Post system with truth degree set
Wm . These Post propositional systems have been originally formulated uniformly in negation and disjunction as basic connectives, i.e. their set J P of
connectives contains the connectives

(negation); _ (disjunction)
with the corresponding truth degree functions
verP = non2 ;
verP_ = vel1
as already introduced in (5.56), (5.75).
For the sets of designated truth degrees a canonical choice does not exist;
already Post [426] has discussed the possibility that there may be chosen
truth degrees dierent from 1 as designated ones. Nevertheless, DP = f1g is
a kind of standard choice.
The set J P = f; _g of connectives of each one of the Post systems Pm
is functionally complete according to Theorem 8.1.2. Therefore each one of
the Post systems Pm , with DP = f1g as the set of designated truth degrees,
covers its corresponding Lukasiewicz system with the same set of truth
degrees { in the sense that the set of Lm -tautologies is a subset of the set of
Pm -tautologies, and that this set of Pm -tautologies does not contain any w
H 2 LL whose Lukasiewicz negation :H is Lm -satisable, of course always
314
via a suitable reading of the Lukasiewicz connectives in the Post systems.

And the same holds true for the corresponding m-valued Go del system Gm .
Because of this functional completeness each one of these systems can be
adequately axiomatized by the Rosser-Turquette approach as explained
in Chapter 6. The crucial point is the choice of a suitable implication connective which has either to satisfy the standard condition (I), i.e. the condition
(RT1), or at least the weakened condition (I ), cf. Corollary 6.2.3. It is interesting to notice that already Post introduced in [426] a candidate for
such an implication connective, without being engaged in the axiomatizability problem. The truth degree function of this Post implication can easily be
described by reference to the truth degree functions seq 1 ; seq 2 and dened
as

2 (u; v ); if u is a designated truth degree
seq Post (u; v) = seq
seq (u; v); if u is a non-designated truth degree.
1
This implication connective does not satisfy the standard condition (I), but it
satises condition (I ) and hence is a suitable implication connective for this
axiomatization. Actually, however, the usual axiomatizations of the Post
systems do not refer to this Rosser-Turquette axiomatization but result
from the theory of algebraic structures, the Post algebras, which correspond
to the Post systems essentially as e.g. the MV-algebras correspond to the
Lukasiewicz systems.
It is interesting to notice that the functional completeness of the systems
Pm can be established without direct reference to Theorem 8.1.2. Following
Post [426] one may start with introducing a unary connective T1 similar to
(8.5) by dening
T1 (H ) =def H _
W
m_
;1
i=1
i H;
(12.1)
using for the nite iteration of the connective _, as already introduced in

(9.25), and the nite iteration n of given by
1 H =def H; k+1 H =def (k H ) ;
(12.2)
n
which has the truth degree function non2 which has already been introduced in (8.4). Obviously each of the unary operations nonn2 causes for any
n = 1; 2; : : : a cyclic exchange within the (ordered) set Wm of truth degrees.
Therefore via some suitable operation from this family each given truth degree can be transformed in any predetermined (other) one. With reference to
the notation k = mm;;k1 introduced in (5.7) one therefore has
nonk2 (i ) = i+k
if one takes the indices 1; 2; : : :; m of the -symbol modulo m. Therefore the
w T1 (H ) has truth degree 1 = 1 for each valuation.
Introducing for each k = 1; : : : ; m a further unary connective Dk by the
denition
315
Dk (H ) =def m;1 (k H _ m;1 (H _ T1(H )))

(12.3)
leads to the corresponding truth degree functions dk characterized by
n
dk (u) = k ; if u = 1
0 otherwise.
Then one is able to represent each unary truth degree function. Let such a
function h : Wm ! Wm be given, and let k1 ; : : : ; km be such a sequence of
truth degrees that one has
h(i ) = ki for i = 1; : : : ; m:
Then the w
Dk (p) _
1
m
_
i=2
Dki (m;i+1 p);
with p some propositional variable, represents the truth degree function h.

Particularly one has
:H =def Dm (p) _
or even simpler
:H =def Dm (p) _
m
_
i=2
m
_
i=2
Dm;i+1 (m;i+1 p);

Di (i p)
(12.4)
as an introduction of the Lukasiewicz negation in the Post systems. By

H1 ^ H2 =def :(:H1 _ :H2 )
(12.5)
one also has the weak conjunction of Lm with its truth degree function et1 .
Altogether this allows one to introduce a w Bi ;:::;in ;k which contains the
propositional variables p1 ; : : : ; pn and which has the property that Bi ;:::;in;k
has truth degree k 2 Wm for any valuation with (pl ) = il for all l =
1; : : : ; n, and which has truth degree m = 0 for all other valuations. One has
e.g. to consider
1
Bi ;:::;in ;k =def
1
n
^
l=1
Dk (m+1;il pl );
V
with 0 read as the empty symbol, i.e. treated as \not existent", and with
for the nite ^-iteration as dened in (9.24). Any n-ary truth degree function
h : Wmn ! Wm then may be represented by the w
m
_
m
_
i1 =1 i2 =1
:::
m
_
in =1
Bi ;:::;in ;h(i ;:::;in ) :

1
Contrary to the situation for the Lukasiewicz systems, for the Post
systems in their original form there exist only very few syntactically oriented
316
studies toward constituting or investigating logical calculi for these systems.

Instead, for the Post systems one mainly was interested in corresponding
algebraic structures, and investigated such structures earlier, and in more
detail, as similar structures for the Lukasiewicz systems. Rosenbloom in
a paper [462] of 1942 was the rst one to do this. His algebraic structures
shall here be called P-algebras for short.
By a P-algebra of order n one understands an algebraic structure hA; +;0 i
of similarity type h2; 1i whose carrier A contains at least two elements and
for which one has true for all a; b; c; c0; c1 ; : : : ; cn;1 2 A:
(P1) a + b = b + a,
(P2) a + (b + c) = (a + b) + c,
(P3) a + a = a,
P
and for which with for the nite iteration of the binary operation +, with
the additional notations
a0 =def a; ak+1 = (ak )0 ;
t(a) =def
nX
;1
k=0
ak ;
d1 (a) =def
;1
n
X
i=1
ai
n;1
dk (a) =def (ak + (a + t1 (a))n;1 )n;1 for 2 k n ; 1

with the shorthand t1 (a) = (t(a))0 , and nally also with
;a =def
nX
;1
k=1
(dk (a))k ;
a b =def ;((;a) + (;b)) ;
one has furthermore satised the conditions:

(P4) t(a) = tn (a),
(P5) a + (b c) = (a +b) (a + c),
0 Pn;1 0 k
Pn;1
k
(P6)
k=0 ck d1 (a) = k=0 ck d1 (a),
P ;1
(P7) a = nk=0
aP
bk ,
;1 tk (a) dn;k (a),
(P8) a = d1 (a) + nk=1
1
k
k
i
writing t (a) = (t(a)) and d1 (a) = (d1 (a))i for short.
This is a rather complicated denition. But this complicated structure is
caused by a strong similarity with the corresponding Post system Pn : here
the denition of t(a) corresponds to the introduction (12.1) of the connective
T1 into Pm , the power notation ak corresponds to the iteration (12.2) of
the Post negation, the denition of dk (a) corresponds to (12.3), and the
introductions of the additional operations ;; correspond to (12.4) and (12.5),
respectively.
Then it is an easy exercise to show that each one of the algebraic structures
hWm ; vel1 ; non2 i is a P-algebra of order m. And additionally one gets that in
317
these cases the operations ;; are just the truth degree operations non1 ; et1 ,
respectively.
In [462] it is proved among other results that any two P-algebras of order
n with only nitely many elements are isomorphic already in the case that
they have the same number of elements. Therefore one has the structures
hWm ; vel1 ; non2 i as the prototypes of nite P-algebras. From [462] one also
knows the following interesting characterization of the functions dk1 which
we mention without proof because otherwise we would have to infer a lot of
further consequences from the axioms (P1); : : : ; (P8) of P-algebras.
Theorem 12.1.1. Suppose that hA; +;0 i is a P-algebra of order n and that
the unary functions 1 ; : : : ; n from A into A are dened over hA; +;0 i, i.e.
described by using only the operations +;0 of this P-algebra, in such a way
that they are pairwise dierent, are also dierent from the unary function
tn;1 , and satisfy
n
X
i=1
i (a) = t(a);
j (a) k (a) = tn;1 (a)
for each a 2 A and all 1 j < k n. Then these functions 1 ; : : : ; n are

just the functions d1 ; d11 ; : : : ; dn1 ;1 , possibly in a dierent order.
Furthermore one has that each P-algebra of order n is a subdirect product
of a suitable number of copies of the particular P-algebra hWn ; vel1 ; non2 i, or
of P-algebras isomorphic with hWn ; vel1 ; non2 i. This was proved by Wade in
[559] within a larger study of problems in ring theory. As an immediate corollary one has that each nite P-algebra of order n is isomorphic to the direct
product of some nite number of copies of the P-algebra hWn ; vel1 ; non2 i. And
this means that each P-algebra of order n is isomorphic to some P-algebra of
n-valued functions with values in the set Wn , such that the operations between these functions are dened \pointwise" by reference to the operations
vel1 and non2 .
These studies, however, use this complicated notion of P-algebra as dened above and therefore do not really simplify the theory of P-algebras. One
of the main reasons for the diculty and complexity of the dening conditions (P1); : : : ; (P8) of P-algebras is the fact that the Post systems as well
as the P-algebras have only two primitive notions, their connectives resp.
their basic operations, but have maximal expressive power in the sense of
being functionally complete. That this choice of the primitive notions really
is the main obstacle toward a simplication became clear as Epstein [141]
in 1960 changed these basic operations and found a much simpler characterization of \essentially" these P-algebras. The reservation \essentially" here
comes from the fact that formally the choice of other basic operations creates
another type of algebraic structures, which we shall denote as Post algebras
following common actual use.
318
12.2 The Present Form

These Post algebras of Epstein [141] are sets equipped with two binary and
n unary operations which additionally have n + 2 designated elements. They
are denitionally equivalent with the P-algebras of the previous section. The
following denition of Post algebras is not Epstein's but a modication due
to Dwinger [139].
Denition 12.2.1. A distributive lattice A = hA; +; ; 0 ; 1 i with a zero and
a unit element is a Post algebra of order n, n 2, i there hold true:
(PA1) There exist elements e0 ; e1 ; : : : ; en;1 2 A with the property
0 = e0 5 e1 5 : : : 5 en;1 = 1
with respect to the lattice ordering 5 of A.
P ;1
(PA2) Each a 2 A has a unique representation a = ni=1
ai ei such that
each of the coecients ai , 1 i < n, has in the lattice A a complement1
and such that there hold true
a1 = a2 = : : : = an;1 :
It is an easy exercise to show that the elements e0 ; e1 ; : : : ; en;1 are

uniquely determined. Hence one also has that a distributive lattice which
is a Post algebra has a uniquely determined order.
In the case of the elementary P-algebras hWn ; vel1 ; non2 i of order n one
now nds that their corresponding Post algebras of order n are the algebraic
structures hWn ; vel1 ; et1 ; 0; 1i, with the designated elements ei = n;i 1 which
obviously satisfy condition (PA1). To recognize that also condition (PA2)
is satised one rst has to have in mind that in this particular case the
only complemented elements are the elements 0; 1, because each pair a; b
of complemented elements has to satisfy the conditions minfa; bg = 0 and
maxfa; bg = 1. That means that one always has ai 2 f0; 1g for the coecients
of the representation
of a 2
has
nX
;1
ai ei = maxfet1 (ai ; ei ) j i = 1; : : : ; n ; 1g
i=1
Wm according to (PA2). Then it is an easy exercise to see that one
a=
i
nX
;1
X
i
0 ej
ei = n ; 1 = 1 ej +
j =1
j =i+1
= vel1 (maxfej j j = 1; : : : ; ig; maxf0 j j = i + 1; : : : ; n ; 1g)
b in the lattice A one denotes any lattice element c such

that b + c = 1 and b c = 0 hold true in this lattice. Then obviously also b
is a complement of c. And elements, which have a complement, are also called
complemented elements.
1 As a complement of
319
as the unique representation according to (PA2) for each element of Wm .

The operations +; of the Post algebras correspond to the equally denoted operations in the P-algebras. The zero element 0 corresponds to tn;1 (a),
the unit element 1 to t(a), always for any element a of some P-algebra under consideration. And each one of the designated elements ek of the Post
algebras corresponds to the constant function tn;1;k (a) in the P-algebras.
However, the coecients of the representation of the element a mentioned
in (P8) do not immediately correspond to the coecients of the representation of a according to (PA2). The distributivity of the lattice hA; +; ; 0 ; 1 i
together with the fact that each one of the coecients a1 ; : : : ; an;1 from
(PA2) has a complement within this lattice, allows one to turn the \monotonic"Prepresentation of a according to (PA2) into a \disjoint" representation
;1 c e with c c = 0 for all 1 i < j < n, and vice versa. It
a = ni=1
i i
i j
is furthermore quite simple to add another coecientPc0 such that one has
;1 c = 1 . Having
c0 ci = 0 for each 1 i < n and additionally also ni=0
i
done this one gets via Theorem 12.1.1 that these coecients c0 ; c1 ; : : : ; cn;1
correspond to the functions d1 ; d11 ; : : : ; dn1 ;1 in the P-algebras. P
;1 a e
This possible turn from a \monotonic
representation" a = ni=1
i i
Pn;1
to a \disjoint representation" a = i=0 ci ei , and vice versa, indicates
that the conditions (PA1), (PA2) do not represent the only possibility to
dene the class of Post algebras. Indeed there exist a series of dierent
denitions which, however, yield denitionally equivalent structures, cf. e.g.
[139, 141, 144, 439, 540, 542].
Proposition 12.2.1. Each Post algebra hA; +; ; 0 ; 1 i of order n is denitionally equivalent with an algebraic structure
hA; +; ; 0 ; 1 ; D0 ; : : : ; Dn; e0 ; : : : ; en;1 i
with unary operations D0 ; : : : ; Dn and distinguished elements e0 ; : : : ; en;1
for which hA; +; ; 0 ; 1 i is a distributive lattice with zero 0 and unit 1, and
for which the following conditions are satised:
P ;1
(PA 1) For each a 2 A one has ni=0
Di (a) = 1 and Di (a) Dj (a) = 0 for
all i 6= j .
(PA 2) One has 0 = e0 < : : : < en;1 = 1 w.r.t. the lattice ordering 5 in
A, and for each a 2 A and each 1 i n ; 1 also
a e1 = 0 ) a = 0 and a + ei = ei+1 ) a = ei+1 :
P ;1
(PA 3) For each a 2 A one has a = ni=0
ei Di (a).
The proof of this proposition is, algebraically, a routine matter and \simply" needs some calculations in the corresponding structures. We shall not
go into the details, the interested reader may e.g. consult [439]. Instead we
mention thePinteresting fact that the uniqueness condition for the represen;1 ai ei , mentioned in (PA2) is, in some generalized form, also
tation a = ni=1
characteristic for Post algebras.
320
The formulation of this result needs two other algebraic notions we rst
have to introduce. By an n-ary polynomial mapping over some algebraic structure A one understands any mapping f : An ! A which is a superposition of
some n-ary projections and of operations of A. And a k-ary algebraic function g over A results from some n-ary polynomial mapping f over A, with
k n, by substituting elements of jAj for n ; k of the arguments of f . Now
one has:
Proposition 12.2.2. A structure A = hA; +; ; 0 ;1 ; D0; : : : ; Dn; e0; : : : ; en;1 i
with unary operations D0; : : : ; Dn and distinguished elements e0 ; : : : ; en;1 is
a Post algebra of order n i every unary algebraic function g over A has a
unique representation
g(x) =
nX
;1
i=0
ai Di (x) :
And for such a representation one has always ai = g(ei ).

The proof again is, algebraically, more or less a routine matter and shall
not be given here, but can be found in [447].
Looking at the most simple examples hWn ; vel1 ; et1 ; 0; 1i of Post algebras
of order n, one gets in the present case immediately that the unary operations
Di have for each a 2 Wn the property

ei
Di (a) = 10;; ifif aa =
< ei .
Now it is possible to establish a direct correlation between P-algebras and
Post algebras in such a way that, on the one hand, one introduces in each
P-algebra the chain e0 5 e1 5 : : : 5 en;1 and proves (PA1) and (PA2)
for it, or that one, on the other hand, introduces into each Post algebra
a unary operation 0 which together with the lattice meet, i.e. with operation +, satises conditions (P1); : : : ; (P8). But to do this would force one to
carry out extended calculations which one can avoid if one rst discusses the
representability of Post algebras.
In any case this means that Post algebras and P-algebras are denitionally equivalent algebraic structures.
Denition 12.2.1, together with the fact that the set of all elements of a
distributive lattice hA; +; ; 0 ; 1 i which have a complement in this lattice form
a Boolean algebra (with respect to the lattice operations), gives immediately
Theorem 12.2.1. A distributive lattice A = hA; +; ; 0 ; 1 i with zero and
unit element is a Post algebra of order n, n 2, i there exists a Boolean
algebra B = hB; [; \; c ; 0 ; 1 i such that the lattice A is isomorphic to the set
f(b1 ; : : : ; bn;1 ) 2 B n;1 j b1 = b2 = : : : = bn;1 g
of (n;1)-tuples of elements together with 0 = (0 ; : : : ; 0 ) and 1 = (1 ; : : : ; 1 ) as
zero and unit element and with operations which are dened componentwise
from the operations of the Boolean algebra B .
321
This result does not yet establish the relationship between P-algebras and
Post algebras we are looking for. This is done, however, by a representation

theorem of Epstein [141] stating that each Post algebra of order n is isomorphic to a suitable Post algebra of n-valued functions, i.e. of functions
of the same kind as used in [559] for the representation of P-algebras. And
within such structures of n-valued functions it is easy to realize the turn
between P-algebras and Post algebras with the same carrier, cf. [141].
Theorem 12.2.1 is well suited to construct examples of Post algebras.
Consider e.g. as B the two-element Boolean algebra W2 = f0; 1g with
operations max; min; non1 . This Boolean algebra yields a Post algebra of
order n which has as its elements all the (n ; 1)-tuples
(0; : : : ; 0); (1; 0; : : : ; 0); (1; 1; 0; : : :; 0); : : : ; (1; : : : ; 1; 0); (1; : : :; 1)
and which obviously is isomorphic to the Post algebra Wn with the operations max; min and the designated elements ei = n;i 1 .
Denoting for those elements b of a Post algebra A of order n, which have
a complement, this complement (which then is even uniquely
determined) by
;1 a e of a 2 A
b allows to get from the unique representation a = Pni=1
i i
according to (PA2) a unary operation by
a =def
nX
;1
i=1
ai ei :
(12.6)
It is easy to prove that this operation has the properties (LA1), (LA2)
of the Lukasiewicz algebras, cf. p. 248. Therefore the lattice A with this
additional operation becomes a deMorgan algebra.
In general one has that each Post algebra of order n is also an n-valued
Lukasiewicz algebra. One has to dene the unary operation as done in
(12.6), and has to take in (a) = an;P
1 for each 1 i < n considering again
;1 a e . And if one denes
the \monotonic representation" a = ni=1
i i
n
n
Fij (x; y) =def n;i (x) n;i;1 (x) nn;j (y) nn;j;1 (y) en;1;i+j
for all suitable indices i; j , one even gets a proper n-valued Lukasiewicz
algebra, cf. [102].
Conversely one gets a Post algebra of order n from some given n-valued
Lukasiewicz algebra if one additionally supposes that there exist in this
n-valued Lukasiewicz algebra elements e0 ; e1 ; : : : ; en;1 with e0 = 0 and
en;1 = 1 such that

n;i
jn (ei ) = 10 ;; ifif jj
<n;i
holds true for all indices i; j , cf. [100, 102].
The P-algebras arose immediately out of the Post systems of manyvalued logic. The Post algebras arose mainly out of algebraic considerations. So one is interested to connect them more directly with systems of
322
many-valued propositional logic than turning them into P-algebras. But to

reach this goal in a nice and simple way requires turning them into another,
less algebraically motivated, characterization which avoids the reference to
the only partially dened complementation used in (PA2) { because this partial operation would have to correspond to a kind of only partially dened
negation, something quite unusual in logical systems.
A characterization of Post algebras of order m which is well suited for the
purpose of being directly linked with logical systems was given by Rousseau
[467] as a slight modication of the characterization of Post algebras of order
n mentioned in Proposition 12.2.1. He enriches the lattice structure of the
Post algebras with the unary operation as mentioned in (12.6), with a
binary relative pseudo-complementation ! dened (using inx notation) via
the adjointness condition
w u ! v i u ^ w v
or directly as
u ! v =def supfz j u ^ z vg;
with m ; 1 further unary operations Di associating P
with each a 2 A all of its
;1 a e according to
coecients in the \monotonic representation" a = ni=1
i i
(PA*3), i.e. satisfying Di (a) = ai , and enriches the lattice structure nally
with all the elements e0 ; e1 ; : : : ; en;1 . The axioms he has to add for this characterization of Post algebras come from the fact that he has to demand that
the particular structure hA; ^; _; !; ; 0 ; 1 i has to be a Heyting algebra,
and that the further dening conditions just correspond to the axioms (P11),
. . . , (P18) we are going to mention for the corresponding logical calculi.
Based on this approach Rasiowa [439] developed the connection between
Post algebras of order n and n-valued systems of propositional logic in detail.
The m-valued logical system based on these Post algebras of order m
has as set of basic connectives the set
JP = f_; ^; !; ; D1 ; : : : ; Dm;1 g
(12.7)
of three binary and m unary connectives together with a set
fe0; : : : ; em;1 g
(12.8)
of truth degree constants, and is based on the axiom schemata (LC1) to
(LC10), cf, p. 272, and additionally has for all i = 1; : : : ; m ; 1 the further
schemata:2
(P11) Di (H1 _ H2 ) $ (Di H1 _ Di H2 );
(P12) Di (H1 ^ H2 ) $ (Di H1 ^ Di H2 );
2 These schemata additionally refer to a biimplication connective
usually as H $ G =def (G ! G) ^ (G ! H ).
$ dened as
323
(P13) Di (H1 ! H2 ) $ ((D1 H1 ! D1 H2 ) ^ (D2 H1 ! D2 H2 ) ^ : : :

^(Di H1 ! Di H2 );
(P14) Di ( H1 ) $ D1 H1 ;
(P15) Di Dj H1 $ Dj H1 ;
(P16) Di ej for i j and Di ej for i > j;
(P17) D1 H1 _ D1 H1 ;
(P18) H1 $ (D1 H1 ^ e1 ) _ (D2 H1 ^ e2 ) _ : : : _ (Dm;1 H1 ^ em;1);
together with the rule
Dm;1 H
and the rule of detachment (w.r.t. the implication !) as inference rules.

The interested reader should consult [439] and the additional literature
mentioned there for more details. One of the main results there is the completeness theorem for this logical calculus of m-valued Post logic w.r.t. the
class of all Post algebras of order m.
What is not treated in [439] are logical calculi for propositional systems
of many-valued logic, which take their starting point from Post algebras as
truth degree structures, and which proceed in the Gentzen-style of sequent
calculi.3 Such calculi have e.g. been developed in [296, 297, 466, 475].
What are also not covered by these considerations are possible innitely
many-valued generalizations of these logical calculi, or of these Post algebras.
Approaches toward this problem started e.g. with papers on generalizations
of the notion of Post algebras like [87, 138, 139, 541]. The most in uential
paper, however, which also discussed the corresponding logical systems was
the paper [438] of Rasiowa in which she introduced Post algebras of order4
!+ and corresponding systems of innitely many-valued (rst-order) logic.
The crucial points for this generalization are that (i) the sequence 0 = e0 5
e1 5 : : : 5 en;1 = 1 of designated elements of a Post algebra of order n
is replaced by a corresponding sequence 0 = e0 5 e1 5 e2 5 : : : 5 e1 = 1
of order type !+, that (ii) the sequence D0 ; : : : ; Dn of unary operations is
replaced by an innite sequence (Dk )0k (of order type !) of unary operations, and that (iii) the other conditions which dene a Post algebra of order
n are suitably5 adapted. The algebraic theory of these Post algebras of order !+ is partly given in [438]. And an interesting representation theorem is
3 For the case of sequent calculi for classical logic the reader may e.g. consult [520]
for a survey.
4 By ! + , or also by ! +1, one denotes the order type of the set of natural numbers
completed with an additional greatest element 1, i.e. of the set f0; 1; 2; : : : ; 1g,
or equivalently the order type of the set of all fractions k=k +1 of the unit interval
together with the number 1 (w.r.t. the natural order of these numbers).
5 It seems that actually there exists no general agreement what \suitably" here
should mean
in detail. Some modication of the denition of Post algebra of
order !+ from [438] has been discussed e.g. in [146].
324
proved in [342] which essentially says that each Post algebras of order !+
can be isomorphically represented by an algebraic structure whose universe
is a subset of a suitable power set.6
On the side of the corresponding propositional calculi one mainly can
proceed as for the nitely many-valued Post systems: this means that one
enriches the propositional language of these systems with innitely many
unary connectives Di for i 1 as well as with innitely many truth degree
constants ej for j 0 and an additional one denoted e1 , and takes as axioms
the schemata (LC1) to (LC10) together with the schemata (P11) to (P17),
and adds as further schemata
(P 18) Di H ^ ei ! H;
(P 19) e1 :
The rules of inference for this logical calculus are the rule of detachment
(w.r.t. the implication !) together with the (innitely many) rules
H
Di H for every i ;
and
Di H
with the last mentioned one of them (unfortunately) an innitary one which
needs innitely many premises.
For these innitely many-valued (propositional) Post systems an adequate semantics is provided by the class of Post algebras of order !+ ,
cf. [438]. But also forcing-type relational semantics7 have been introduced
and allow one to prove a completeness theorem, i.e. provide an adequate
semantics, cf. [343, 404].
Another such innitely many-valued generalization of the standard Post
systems is discussed e.g. in [147, 148], Post algebras of order ! + ! . For
them sequences of designated elements and of unary operations of the form
0 = e0 5 e1 5 e2 5 : : : ; : : : 5 e;2 5 e;1 = 1
D1 ; D2 ; : : : ; : : : ; D;2; D;1 ;
each of order type ! + ! , are characteristic.8
The Post algebras of nite or innite order and the systems of manyvalued logic related with them seem to be of particular importance for investigations in computer science, which rely on many-valued logic as a toolbox,
because these Post systems are functionally complete and well suited to
6 We shall not look at the details here because the construction is not such a simple
one as given in Theorem 12.2.1 for the Post algebras of order n.
7 This is a type of semantics that is strongly related to the possible worlds seman-
tics for intuitionistic logic, cf. e.g. [161].
8 Also further generalizations of the standard Post algebras have been considered
as basic semantic structures, together with suitable logical calculi intended to

serve as tools for logical investigations into approximate reasoning, e.g. in [440,
441, 443, 442].
325
study the representability of truth degree functions on the basis of some predetermined set of basic truth degree functions, as determined e.g. by available
electronic components, cf. [451] for a good introduction and also [86] for some
particular aspects.
If one enriches all the nitely many-valued (propositional) Lukasiewicz
systems Lm with truth degree constants ei , 0 i < m, in such a way that
always ei denotes the truth degree m;i = m;i 1 , then these enriched Lsystems Lm become functionally complete according to Theorem 8.1.1. And
this means that the extended m-valued Lukasiewicz systems Lm and the
m-valued Post logics become interdenable (for each xed number m of
truth degrees). Hence there is in principle no essential dierence between
both types of systems: all that can be expressed in the \Post world" can
also be expressed in the (extended) \Lukasiewicz world", and vice versa.
326
13. t-Norm Based Systems
A class of candidates for conjunction connectives which was introduced in

Chapter 5 was the class of t-norms. Based on such t-norms a whole class of
systems of innitely many-valued logic with truth degree set W1 = [0; 1] can
be introduced, and the previously discussed innitely many-valued systems of
Lukasiewicz and Go del, as well as the product logic prove to be particular
cases of this more general approach.
In each of these cases the truth degree 1 shall be the only designated
one, which e.g. means that the notions of logical validity and entailment are
understood as in those previously discussed systems.
Before we consider the problem of adequate axiomatizations of a large
class of such systems in Chapter 14, we consider here from a more intuitive
point of view the situation in such t-norm based systems. We restrict, however, the considerations to left continuous t-norms, because we like to have
available in each case the R-implication based on the constitutive t-norm.
Already in these cases there exist well structured systems of many-valued
logic.1
For each t-norm t we denote by ^t the conjunction connective of the language

of some system St of many-valued logic that has this t-norm t as its truth
degree function, similarly we use in this language Lt for the t-norm t a corresponding disjunction connective _t with truth degree function st according
to (5.73). And for each left continuous t-norm t we add to the language of the
system St of many-valued logic also the corresponding R-implication connective !t . Furthermore we intend to refer to the usual ordering of the truth
degrees, which is a (lattice) ordering, and prefer to do this via the corresponding (lattice) operations min; max, referred to by the connectives ^; _.
And nally we add a truth degree constant ? to denote the truth degree 0.
Thus the alphabet of (the language Lt of) each of the systems St has as
its sets of basic connectives and of truth degree constants the sets
1 Actually it is, however, an open problem whether there are applications which
need the consideration of systems of this type which are based on a left continuous
t-norm which is not continuous.
328
J t = f^; _; ^t ; _t ; !t g ;
Kt = f??g ;
(13.1)
with the truth degree 0 denoted by ? , and with the connectives interpreted,
respectively, by the truth degree functions

et1 ; vel1 ; t; st ; seq t ;
as dened in (5.8), (5.75), (5.73), (5.89).
In general, we even allow to enrich the language of these systems with as
many new t-norm based conjunction, disjunction and implication connectives
as seem to be appropriate in each particular case (of an intended application).
Usually that will mean to include one or two such t-norm based conjunction
connectives into the considerations and to add one or two further disjunction
and implication connectives based on these t-norms.2
Sometimes, particularly as shown in Chapter 14, one behaves a little bit
dierent and considers as the suitable language for the t-norm based systems
either only the fragment of Lt which does not contain the t-norm based
disjunction connective _t , or the extension of this fragment by a negation
connective ;t , to be introduced later on in this section, cf. Denition 13.1.1.
For any w H and any valuation one thus has e.g. for the conjunction
H1 ^t H2 the condition
ValS (H1 ^t H2 ; ) = t(ValS (H1 ; ); ValS (H2 ; )) :
(13.2)
Correspondingly one has for the semantic interpretation of the implication
connective !t the simple fact that
ValS (H1 !t H2 ; ) = seq t (ValS (H1 ; ); ValS (H2 ; )) :
(13.3)
Sometimes one also uses a biimplication operator $t connected with such a
left continuous t-norm t dened by
H1 $t H2 =def (H1 !t H2 ) ^t (H2 !t H2 ) :
(13.4)
Proposition 13.1.1. For each left continuous t-norm t and all ws H1; H2
one has
j= H1 !t H2 i ValS (H1 ; ) ValS (H2 ; ) for all ;
(13.5)
S
S
j= H1 $t H2 i Val (H1 ; ) = Val (H2 ; ) for all :
(13.6)
Proof: The rst claim is essentially a restatement of the degree ranking
property, the second one a consequence of the rst using that t(u; v) = 1 i
u = v = 1.
2
Therefore all the logically valid implications of our generalized language
are inequalities between truth degrees \in disguise", and the logically valid
biimplications accordingly represent equalities for truth degrees.
2 Our strategy here is opposed to another one which aims to reduce the class of
acceptable truth degree functions, and which in the most extreme case tries to
nd conditions which characterize e.g. only the pair et1 ; vel1 . Such conditions
were rst given in [39], and later on rened partly e.g. in [4, 291]. They have not
been, however, really convincing up to now.
329
Remark: For simplicity of notation we use the arrows ! and $ here also
without subscripts: in that case we assume that a subscript \t1 " has been
deleted that refers to some left continuous t-norm t1 . And we additionally use
the truth degree constant > to denote the truth degree 1. It is not important
here whether the truth degree constant > is added to the primitive signs of
the alphabet, or introduced by a suitable denition, e.g. by
> =def ? !t ? :
Proposition 13.1.2. For any t-norm t the following expressions are logically valid:
(i) j= H ^t > $ H and j= H ^t ? $ ?;
(ii) j= H _t > $ > and j= H _t ? $ H;
(iii) j= H1 ^t H2 $ H2 ^t H1 ;
(iv) j= H1 _t H2 $ H2 _t H1 ;
(v) j= H1 ^t (H2 ^t H3 ) $ (H1 ^t H2 ) ^t H3 ;
(vi) j= H1 _t (H2 _t H3 ) $ (H1 _t H2 ) _t H3 :
Proof: Those statements immediately follow from the properties (T1)
to (T4) of any t-norm and the properties (S1) to (S4) of any t-conorm. 2
Therefore the t-norm based conjunctions and disjunctions are already by
denition { modulo logical validity { commutative as well as associative operations. Distributivity properties between dierent conjunction connectives
and between conjunctions and disjunctions, however, hold true only for particular cases.
For the simplication of the notation we allow, for the rest of this section,
to alternate between the usual prex notation t(u; v) and a related inx
notation u t v, not only for t-norms but also for t-conorms and even for the
truth degree functions of the R-implications. In the last case, however, we
shall not write u seq t v for seq t (u; v), but shall write u 't v for seq t (u; v).
Proposition 13.1.3. For any t-norm t the following equivalences are logically valid:
(i) j= H1 ^t (H2 ^ H3 ) $ (H1 ^t H2 ) ^ (H1 ^t H3 ) ;
(ii) j= H1 ^t (H2 _ H3 ) $ (H1 ^t H2 ) _ (H1 ^t H3 ) ;
(iii) j= H1 _t (H2 ^ H3 ) $ (H1 _t H2 ) ^ (H1 _t H3 ) ;
(iv) j= H1 _t (H2 _ H3 ) $ (H1 _t H2 ) _ (H1 _t H3 ) :
Proof: Denoting by abuse of language again by ^; _ the (binary) operations min; max, one needs to show for these cases that there hold true
(i) u t (v ^ w) = (u t v) ^ (u t w) ;
(ii) u t (v _ w) = (u t v) _ (u t w) ;
330
(iii) u st (v ^ w) = (u st v) ^ (u st w) ;
(iv) u st (v _ w) = (u st v) _ (u st w)
for any truth degrees u; v; w. But these are immediate consequences of the
denitions of t-norms and of t-conorms, particularly of their monotonicity
properties (T3) and (S3).
2
Sometimes, however, one has only some kind of \subdistributivity", i.e.
sometimes there are no logically true biimplications, but only implications.
Proposition 13.1.4. For any t-norm t the following implications are logically valid:
(i) j= (H1 ^ H2 ) ^t (H1 ^ H3 ) ! H1 ^ (H2 ^t H3 ) ;
(ii) j= (H1 _ H2 ) ^t (H1 _ H3 ) ! H1 _ (H2 ^t H3 ) ;
(iii) j= H1 ^ (H2 _t H3 ) ! (H1 ^ H2 ) _t (H1 ^ H3 ) ;
(iv) j= H1 _ (H2 _t H3 ) ! (H1 _ H2 ) _t (H1 _ H3 ) :
Proof: Denoting again by abuse of language by ^; _ also the (binary)
operations min; max, one needs to show for these cases that there hold true
(i) u ^ (v t w) (u ^ v) t (u ^ w) ;
(ii) u _ (v t w) (u _ v) t (u _ w) ;
(iii) u ^ (v st w) (u ^ v) st (u ^ w) ;
(iv) u _ (v st w) (u _ v) st (u _ w) :
But these claims follow by applying the distributivity results (i) { (iv) mentioned in the preceding proof to the right hand side of these subdistributivities
and then using some simple inequalities for t-norms and t-conorms.
2
One may wonder whether these results can be generalized in the way
that the particular t-norm min can also be replaced by another one. This is,
however, in most cases impossible.
Proposition 13.1.5. For any t-norms t; t1 the condition t = min is equivalent with the fact, that any one of the following conditions holds true for all
ws H; H1 ; H2 :
(i) j= H ! H ^t H ;
(ii) j= H ^t (H1 _t H2 ) $ (H ^t H1 ) _t (H ^t H2 ) ;
(iii) j= H _t (H1 ^t H2 ) $ (H _t H1 ) ^t (H _t H2 ) ;
(iv) j= H ^t (H1 ^t H2 ) ! (H ^t H1 ) ^t (H ^t H2 ) ;
(v) j= (H ^t H1 ) _t (H ^t H2 ) ! H ^t (H1 _t H2 ) :
Proof: Obviously any one of these conditions holds true for the case
t = min. Thus it remains to prove that the truth of each of these conditions
implies t = min.
1
331
(a) In this case one has to have for all u 2 [0; 1] the inequality u u t u,
and hence always u t u = u because one always has u t u u t 1 = u by (T3),
(T4). This means that in this case t has to be idempotent. But then one has
for all u; v 2 [0; 1] with u v even
u = utu utv ut1 = u;
which means u t v = minfu; vg in this case. And for v u the same result
holds true by symmetry.
(b) In this case one has to have for all u; v; w 2 [0; 1] the equality
u t (v st w) = (u t v) st (u t w) ;
which gives with v = w = 1 for all u 2 [0; 1] the idempotency u = u t u. Now
t = min follows as in case (a).
(c) Now the distributivity condition from (b) has to be replaced by the
condition
u st (v t w) = (u st v) t (u st w) :
The choice v = w = 0 again reduces this to the condition of idempotency for
the t-norm t. Hence t = min follows as in (a).
(d) Here the distributivity condition from (b) or (c) has to be replaced
by the subdistributivity condition
u t1 (v t w) (u t1 v) t (u t1 w) ;
which has to be satised for all u; v; w 2 [0; 1]. The choice v = w = 1 reduces
this condition to the condition u u t u for all u 2 [0; 1], i.e. to the case (a).
So again t = min follows.
(e) Proceeding as in case (d) here yields the condition u st u u for all
u 2 [0; 1]. As in case (a) one shows that this means the idempotency of the
t-conorm st , which is equivalent to the idempotency of the t-norm t. So also
in this case t = min results.
2
Proposition 13.1.6. For any left continuous t-norm t the following ws are
logically valid:
(i) j= H ^t (H !t G) ! G ;
(ii) j= G ! (H !t H ^t G) ;
(iii) j= H ! (G !t H ) ;
(iv) j= (H1 !t H2 ) ^t G ! (H1 !t H2 ^t G) :
Proof: The claims (i) and (ii) follow immediately from the inequalities
mentioned in (5.88). (iii) is a restatement of the fact that v seq t (u; v) holds
for any R-implication. And for (iv) it is enough to have
(u 't v) t w u 't (v t w)
(13.7)
for any u; v; w 2 [0; 1]. To get this last inequality one starts from
332
(u 't v) t w = w t supfs j u t s vg = supfw t s j u t s vg :

But with u t s v one has u t (w t s) = (u t s) t w v t w and hence
fw t s j u t s vg fz j u t z v t wg :
Therefore one gets

(u 't v) t w supfz j u t z v t wg = u 't (v t w) ;
and thus has (13.7) and hence (iv).
2
The monotonicity properties of the t-norms and of their corresponding
R-implications can directly be rewritten as the implications (which, yet, are
classical ones, i.e. formulated in the metalanguage):
(a) if j= H1 ! H2 then j= H1 ^t G ! H2 ^t G ;
(b) if j= H1 ! H2 then j= (G !t H1 ) ! (G !t H2 ) ;
(c) if j= H1 ! H2 then j= (H2 !t G) ! (H1 !t G) :
But it is also possible to generalize them to inequalities between truth degrees,
that means to logically valid \many-valued" implications.
As in classical logic one has a direct connection between the result of
Proposition 13.1.6(i) and the inference rule of modus ponens: because of
Proposition 13.1.6(i) the rule of detachment
H; H !t G (MP)
is sound. For, having given ws H and H !t G such that j= H as well as

j= H !t G, then by Proposition 13.1.6(i) one has immediately j= G. Hence
any application of the modus ponens as a rule of inference leads from logically
valid formulas again to a logically valid formula.
As a side remark let us mention that each one of the metatheoretical implications (a),. . . ,(c) which we have just mentioned also provides the soundness
proof for some rule of inference; e.g. because of (b) the rule of inference
H1 !t H2
(G !t H1 ) !t (G !t H2 )
is sound. Furthermore, each logically valid many-valued implication of the
form
j= H1 ^t : : : ^t Hn !t H
yields the soundness of the rule of inference
H1 ; : : : ; Hn :
H
Therefore, as corollaries of the following propositions we get the soundness

of a whole series of rules of inference but will not mention this fact in those
future cases separately.
333
logically valid:
(i) j= (H1 !t H2 ) ! (H1 ^t G !t H2 ^t G) ;
(ii) j= (H1 !t H2 ) ! ((G !t H1 ) !t (G !t H2 )) ;
(iii) j= (H1 !t H2 ) ! ((H2 !t G) !t (H1 !t G)) :
Proof: (i) By the monotonicity condition (T3) one has for each t-norm
u t z v ) (u t w) t z = (u t z ) t w v t w
and therefore, using the characterization (5.89) of the R-implications,

u 't v (u t w) 't (v t w) :
But that is precisely the statement in its truth functional form.
(ii) What has to be proved is
u 't v (w 't u) 't (w 't v) ;
which means, again according to (5.89),
supfz j u t z vg supfx j (w 't u) t z w 't vg :
Therefore the main point is to prove the claim that always
u t z v ) (w 't u) t z w 't v
holds true. Thus assume u t z v. Then using (13.7) one gets
(w 't u) t z w 't (u t z ) w 't v ;
and the proof of (ii) is accomplished.
(iii) In the present case one has to prove that always
u 't v (v 't w) 't (u 't w) ;
and gets as in part (ii) via (5.89) as a sucient condition that this inequality
holds true if one has satised the condition
u t z v ) (v 't w) t z u 't w for all u; v; w; z 2 [0; 1] .
(13.8)
Thus assume u t z v. Because of the left continuity of t and this assumption
one has
(v 't w) t z = z t supfx j v t x wg
= supfz t x j v t x wg supfz t x j u t z t x wg ;
and hence
(v 't w) t z supfy j u t y wg = u 't w ;
i.e. the sucient condition (13.8).
2
334
logically valid:
(i) j= (H1 !t (H2 !t G)) $ (H1 ^t H2 !t G) ;
(ii) j= (H1 !t H2 ) ^t (H2 !t H3 ) ! (H1 !t H3 ) ;
(iii) j= (H1 !t G1 ) ^t (H2 !t G2 ) ! (H1 ^t H2 !t G1 ^t G2 ) ;
(iv) j= (H1 !t G1 ) ^ (H2 !t G2 ) ! (H1 ^ H2 !t G1 ^ G2 ) ;
(v) j= (H1 !t G1 ) ^ (H2 !t G2 ) ! (H1 _ H2 !t G1 _ G2 ) :
Proof: (i) One has to prove for all u; v; w 2 [0; 1] that always
u 't (v 't w) = (u t v) 't w ;
which means by (T1) and the denition of the R-implication to show
supfz j u t z v 't wg = supfz j v t u t z wg :
To get this equality it is sucient always to have that
u t z v 't w , v t u t z w :
But in the case u t z v 't w one has by (T3)
v t (u t z ) v t supfs j v t s wg = supfv t s j v t s wg w :
And conversely in the case v t u t z w one has by (5.88)
u t z v 't (v t (u t z )) v 't w :
(ii) It is enough to prove for all u; v; w 2 [0; 1] that always
(u 't v) t (v 't w) (u 't w) :
But using again the representation (5.89) one has
supfz t (v 't w) j u t z vg = supfz t s j u t z v and v t s wg
supfr j u t r wg ;
because one has u t (z t s) w immediately from u t z v and v t s w.
Thus (ii) follows.
(iii) From (13.7) one gets
(u 't v) t (w 't z ) u 't (v t (w 't z ))
u 't (w t (v 't z )) = (u t w) 't (v t z )
as in the proof of (i) and thus also has (iii).
(iv) What we have to prove for any u1 ; u2; v1 ; v2 2 [0; 1] is that
minfu1 't v1 ; u2 't v2 g (minfu1; u2 g) 't (minfv1 ; v2 g) :
Without loss of generality we may assume u1 u2. Then we have to prove
the simpler inequality
minfu1 't v1 ; u2 't v2 g u1 't (minfv1 ; v2 g) :
(13.9)
335
But if v1 v2 , the right hand side of (13.9) becomes u1 't v1 , and thus is
one of the terms over which on the left hand side the minimum has to be
taken, hence (13.9) holds true in this case. And in the case of v2 < v1 one has
u1 t w u2 t w for any w 2 [0; 1], and hence u2 't v2 u1 't v1 . Therefore
(13.9) becomes
u2 't v2 u1 't v2
which obviously holds true because of u1 u2 .
(v) What we now have to prove for any u1 ; u2; v1 ; v2 2 [0; 1] is
minfu1 't v1 ; u2 't v2 g (maxfu1 ; u2g) 't (maxfv1 ; v2 g) :
Without loss of generality we may assume u1 't v1 u2 't v2 . If then for
some w 2 [0; 1] one has u1 t w v1 , so also u2 t w v2 ; because otherwise in
case v2 < u2 t w one would have w > u2 't v2 and thus u1 't v1 > u2 't v2 , a
contradiction.
Therefore we now have minfu1 't v1 ; u2 't v2 g = u1 't v1 and furthermore
u1 't v1 = supfw j u1 t w v1 g

sup w j maxfu1 t w; u2 t wg maxfv1 ; v2 g
= sup w j maxfu1; u2 g t w maxfv1 ; v2 g
= (maxfu1; u2 g) 't (maxfv1 ; v2 g) ;
hence all is proved.
2
The standard connections (5.73), (5.74) between t-norms and t-conorms
can now be restated as general deMorgan laws for ^t and _t with respect
to the (standard) negation : with truth degree function non1 .
(i) j= :(H1 ^t H2 ) $ (:H1 _t :H2 ) ;
(ii) j= :(H1 _t H2 ) $ (:H1 ^t :H2 ) :
Proof: Obvious.
2
The negation functions nont , introduced in (5.109), cf. p. 104, can now
also be used to consider t-norm based negations.
Denition 13.1.1. For any left continuous t-norm t let ;t be that (unary)
negation connective which has nt as its truth function, i.e.
ValS (;t H; ) = nont (ValS (H; ))
for any valuation .
Proposition 13.1.10. For any left continuous t-norm t and each w H it
holds
j= ;t H $ (H !t ?) :
336
Proof: Obvious by comparing the respective truth degree functions. 2
This way of connecting a negation operator with an implication operator

was for example used by Go del [199] in his studies on intuitionistic logic, and
was also present in the work of the Lukasiewicz group in their development
of many-valued logic; cf. [339].
It is interesting to note that for the Lukasiewicz (arithmetic) conjunction & with truth degree function tL = et2 the corresponding negation ;tL
is nothing but the standard negation :. And furthermore in that special case
one also has
j= (H !tL G) $ (:H _tL G) :
Proposition 13.1.11. For any left continuous t-norm t the following ws
are logically valid:
(i) j= H ! ;t ;t H;
(ii) j= ;t H $ ;t ;t ;t H :
Proof: (i) is an immediate corollary of the denition of the R-implications
and the left continuity of the t-norm t.
(ii) Because of (i) one only has to show
((u 't 0) 't 0) 't 0 u 't 0
(13.10)
for all u 2 [0; 1]. But by the denition of the R-implication and the left
continuity of the t-norm t one has
((u 't 0) 't 0) 't 0 = supfw j supfv j (u 't 0) t v 0g t w 0g
= supfw j supfv t w j (u 't 0) t v 0g 0g :
Now suppose that w0 2 [0; 1] is one of the values which belong to the set
whose supremum has to be considered here, i.e. w0 may be chosen such that
supfv t w0 j (u 't 0) t v 0g 0 :
Then for each v 2 [0; 1]
supfz t v j u t z 0g = (u 't 0) t v 0 ) v t w0 0 ;
which means that for all z; v 2 [0; 1] one has
(u t z 0 ) z t v 0) ) v t w0 0 :
(13.11)
Assuming
w0 > u 't 0 = supfz j u t z 0g
then gives w0 > 0 and u t w0 > 0, i.e. (u t w0 0 ) w0 t v 0) holds true
for any v, and hence v t w0 0 for each v by (13.11). For v = 1 this gives
w0 = 1 t w0 0 and thus a contradiction. Hence w0 u 't 0 and thus (13.10)
by the choice of w0 .
2
337
Proposition 13.1.12. For any left continuous t-norm t the following ws

are logically valid:
(i) j= ;t (H ^t ;t H ) ;
(ii) j= H !t (;t H !t G) ;
(iii) j= (H !t ;t G) $ (G !t ;t H ) ;
(iv) j= (H !t G) ! (;t G !t ;t H ) ;
(v) j= (H !t G) ^t (H !t ;t G) ! ;t (H ^t H ) :
Proof: Claim (i) immediately follows from u t (u 't 0) 0, which is an
instance of (5.93). Claim (ii) follows from (i) by Proposition 13.1.8(i) together
with 0 't v = 1 for any v.
Claim (iii) can be split into two problems, the rst one to prove
j= (H !t ;t G) ! (G !t ;t H ) ;
or equivalently to prove
j= (H !t ;t G) !t (G !t ;t H ) ;
(13.12)
and the second one to prove that the reverse implication from (13.12) is valid
too. Because of the symmetry of both problems it is enough only to infer
(13.12). Using Proposition 13.1.8(i) the problem (13.12) becomes to prove
j= (H !t ;t G) ^t G !t ;t H ;
(13.13)
but with Proposition 13.1.6(iv) one has for each valuation
ValS ((H !t ;t G) ^t G; ) ValS (H !t ;t G ^t G; )
ValS (H !t ?; ) = ValS (;t H; )
from Denition 13.1.1, which is (13.13). Hence (13.12) is proved.
Claim (iv) can equivalently be stated in the form (13.12), but with G and
;t G exchanged. Thus the same calculations as in (iii) yield the proof.
Finally (v) is an easy consequence of Proposition 13.1.8(iii) using (i), i.e.
ValS (G ^t ;t G; ) = 0, which holds true for any valuation .
2
Corollary 13.1.1. For the t-norm tL = et2 contraposition holds true in the
more traditional form

j= (H1 !tL H2 ) $ (:H2 !tL :H1 ) ;
and also for each left continuous t-norm t, for which additionally there holds
true j= (;t ;t H $ H ), even
j= (H1 !t H2 ) $ (;t H2 !t ;t H1 )
holds true.
Proof: Because of j= (::H $ H ) both cases can be proved by the same
argument: simply substitute :G resp. ;t G for G in Proposition 13.1.12(iii)
and then delete the double negations.
2
338

We extend now the considerations upon t-norm-based propositional systems
to the realm of rst-order logic. The basic setting again is to have W1 = [0; 1]
as truth degree set and to consider the truth degree 1 as the only designated
one.
Part of this approach is to dene new t-norm based universal quantiers which for interpretations with nite universes reduce to iterations of a
t-norm-based conjunction connective ^t , as well as to extend them to new
t-norm-based existential quantiers, which under the same restrictions concerning the interpretations reduce to an iteration of a t-conorm-based disjunction connective. But at present it is not clear whether such a generalization
produces interesting quantiers. One of the main diculties, it seems, comes
from the fact that all the t-norms besides t = min are not idempotent, and
that therefore the quantiers based on such non-idempotent t-norms behave
quite dierently in dierent interpretations depending on the cardinalities
of the respective universes { the main dierence being between nite and
innite universes of discourse.
In any case, nite iterations of these t-norm-based connectives, mainly
conjunction connectives ^t and disjunction connectives _t , can be used in
the usual way through inductive generalizations. Up to now only in a few applications, hinted at in Chapter 18 and connected with the discussion of fuzzy
sets, the need may arise for innitary generalizations of such propositional
connectives. In those cases another way round shall be taken instead of introducing new quantiers: innitely long expressions (with a denite length)
will then be used. This method proves to be a suitable substitute to be used
instead of giving the above-mentioned generalization of the quantiers.
As essentially sucient for the (present day) applications to fuzzy sets, we
restrict the considerations to those cases which combine the usual quantiers
of the Lukasiewicz rst-order logics as dened in (9.108), (9.109) with tnorm based propositional connectives.
(i) j= 8xH1 (x) ^t 8xH2(x) ! 8x(H1 (x) ^t H2 (x)) ;
(ii) j= 9x(H1 (x) ^t H2 (x)) ! 9xH1 (x) ^t 9xH2 (x) ;
(iii) j= 8xH1 (x) _t 8xH2(x) ! 8x(H1 (x) _t H2 (x)) ;
(iv) j= 9x(H1 (x) _t H2 (x)) ! 9xH1 (x) _t 9xH2 (x) :
Proof: Let A be any interpretation and some A-assignment.
(i) For each a 2 A = jAj one has ValA (8xHi (x); ) ValA (Hi (a); ) for
i = 1; 2 and thus
ValA (8xH1 (x) ^t 8xH2 (x); ) ValA (H1 (a) ^t H2 (a); ) :
339
Hence for any variable \y" which does not occur (either at all or at least not
free) in \8xH1(x) ^t 8xH2 (x)" one has
ValA (8xH1 (x) ^t 8xH2 (x); ) yinf
Val (H (y) ^t H2 (y); ) :
2A A 1
And this is exactly (i).
Claim (ii) is proved along the same lines. However, the starting point in
this case are the inequalities ValA (Hi (a); ) ValA (9xHi (x); ).
Claims (iii) as well as (iv) follow via Corollary 13.1.1 from (ii) and (i),
respectively, together with the Propositions 9.3.1 and 13.1.9 by simple, direct
transformations.
2
logically valid, if additionally G does not contain the variable x free:

(i) j= 8x(G !t H (x)) $ (G !t 8xH (x)) ;
(ii) j= 8x(H (x) !t G) $ (9xH (x) !t G) :
(i) To get the needed equality
ValA (8x(G !t H (x)); ) = ValA (G !t 8xH (x); )
for the truth degrees we show two corresponding inequalities. By the denition (9.108) of the truth degree of a universal quantication as an inmum
one has
ValA (8x(G !t H (x)); ) ValA (G !t H (a); )
for each a 2 A, and thus
j= 8x(G !t H (x)) !t (G !t H (a)) :
Using Proposition 13.1.8(i) gives
j= 8x(G !t H (x)) ^t G !t H (a) ;
and hence
ValA (8x(G !t H (x)) ^t G; ) ValA (H (a); ) ;
which means for any variable \y" which does not occur in the well-formed
formula \8x(G !t H (x)) ^t G":
ValA (8x(G !t H (x)) ^t G; ) yinf
Val (H (y); ) = ValA (8xH (x); ) :
2A A
Therefore
j= 8x(G !t H (x)) ^t G !t 8xH (x) ;
and again by Proposition 13.1.8(i)
j= 8x(G !t H (x)) !t (G !t 8xH (x)) :
340
But that means just

ValA (8x(G !t H (x)); ) ValA (G !t 8xH (x); ) :
To get the reverse inequality one may start from the fact that one has
ValA (G !t 8xH (x); ) ValA (G !t H (a); )
for each a 2 A, and thus
Val (G !t H (y); ) :
ValA (G !t 8xH (x); ) yinf
2A A
But this is already the inequality which is needed.

(ii) As in the proof of claim (i) one has to prove the equation
ValA (8x(H (x) !t G); ) = ValA (9xH (x) !t G; )
and reaches this goal by proving two inequalities. The fact that one has by
denition of the existential quantier ValA (9xH (x); ) ValA (H (a); ) for
each a 2 A and that therefore, because of the left antitonicity of the Rimplication !t , ValA (9xH (x) !t G; ) ValA (H (a) !t G; ) holds true,
immediately leads to
ValA (9xH (x) !t G; ) yinf
2A ValA (H (y) !t G; ) :
Thus the rst inequality is proved because of
inf Val (H (y) !t G; ) = ValA (8x(H (x) !t G); ) :
y2A A
For the other, converse one we use the notations g = ValA (G; ) and
h(x) = ValA (H (x); ), and have according to the denition of R-implication
to prove that
inf (h(x) 't g) (sup h(x)) 't g = supfw j (sup h(x)) t w gg :
x2A
x2A
x2A
But this results from the fact that the leftmost term itself is one of the values
over which the supremum has to be taken in the rightmost term, for one has:
;

(sup h(x)) t xinf
(
h
(
x
)
'
g
)

inf
(sup
h
(
y
))
t
(
h
(
x
)
'
g
)
t
t
2A
x2A
x2A
y2A
= xinf
sup h(y) t (h(x) 't g) sup h(y) t (h(y) 't g) g :
2A
y2A
y2A
Hence the second inequality is proved too.

2
For the case that one has to consider an existentially quantied implication one has only the following weaker result:
(i) j= 9x(G !t H (x)) ! (G !t 9xH (x)) ;
(ii) j= 9x(H (x) !t G) ! (8xH (x) !t G) :
341
Proof: Both results follow by simple estimations of the truth degrees of

the antecedent and the succedent, respectively.
2
To get the logically valid formulas (T41) and (T42) of the Lukasiewicz
systems also in the present more general setting as logically valid ones would,
e.g., need to suppose the continuity of the R-implication for the basic t-norm
t. Already the left continuity in the second argument of this R-implication
would suce for the proof. However, for both cases one actually does not
know suitable necessary and sucient conditions for the t-norm t which force
the corresponding R-implications to have these continuity properties.
(i) j= 9x(G ^t H (x)) $ G ^t 9xH (x) ;
(ii) j= 8x(H1 (x) !t H2 (x)) ^t 9xH1 (x) ! 9xH2 (x) :
With the notation g = ValA (G; ) and h(x) = ValA (H (x); ) as in the
last proof, for (i) one has to prove
sup (g t h(x)) = g t (sup h(x)) ;
x2A
x2A
and this is exactly the condition of left continuity for the t-norm t. Thus (i)
holds true simply by assumption. And for (ii) one has
ValA (8x(H1 (x) !t H2 (x)) ^t 9xH1 (x); )
= ValA (9x(8y(H1 (y) !t H2 (y)) ^t H1 (x)); )
ValA (9x((H1 (x) !t H2 (x)) ^t H1 (x)); )
ValA (9xH2 (x); )
using (i) together with (5.93), and that is all that needs to be shown.
2
The most interesting point with item (i) of the last proposition is that
the left continuity of the t-norm t means exactly the logical validity of that
expression, i.e. means that as in classical logic an existential quantier can
be \moved into" a conjunction (now: t-norm based) in case only one of the
conjuncts contains the quantied variable. From the logical point of view it
is, besides the denability of a corresponding implication operator, the need
for this property which makes the assumption of the left continuity for the
t-norm t often necessary.
What has been lacking until now with regard to such logically valid formulas, which give equivalences describing possibilities for the distribution of
quantiers over suitable conjunctions and implications, is the case of universal quantication of a t-norm based conjunction. Indeed, that needs another
type of assumption.
342
Proposition 13.2.5. For any continuous t-norm t the following w is logi-
cally valid, if additionally G does not contain the variable x free:

j= 8x(G ^t H (x)) $ G ^t 8xH (x) :
Proof: Straightforward calculations as in the last proofs.
2
Therefore the closest analogies with the situation in classical logic are
obtained from the assumption of the continuity of the t-norm t on which the
connectives under consideration are based. But at present it does not seem
to be clear whether this assumption of continuity is too strong, or whether
one should accept it in every case. In other words, it is an open problem
which discontinuous t-norms (if any) may be of special interest in the realm
of many-valued logic, and particularly in the realm of its application to fuzzy
sets theory, cf. Chapter 18.
As the last topic essentially involving the quantiers, let us reconsider the
problem of Proposition 9.3.1 and take a look at the relations between any of
the t-norm based negations ;t and the quantiers.
logically valid:
(i) j= 9x ;t H (x) $ ;t 8xH (x) ;
(ii) j= ;t 9xH (x) $ 8x ;t H (x) :
Proof: Let A be any interpretation and some A-assignment. By Denition 13.1.1 and Proposition 13.2.2(iv) one has
ValA (9x ;t H; ) = ValA (9x(H !t ?); )
= ValA (8xH !t ?; ) = ValA (;t 8xH; )
and hence (i). To get (ii) one has to refer to Proposition 13.2.2(ii) instead. 2
A generalization of the non-idempotent conjunctions ^t and disjunctions
_t to corresponding universal and existential quantiers faces the problem
that there do not exist commonly used innitary generalizations of t-norms
and t-conorms. The well-known exception is the case of the (only) idempotent
t-norm min and of the (only) idempotent t-conorm max: here the operations
inf ; sup are these innitary generalizations. However, also in all the other
cases of t-norms and t-conorms a generalization is possible, guided by the idea
of a \nite approximation" of innitary generalizations { here turned into a
denition, cf. [532, 533]. Of course, the t-norm based quantier is generalizing
a conjunction and hence should be a kind of generalized universal quantier,
and the t-conorm based quantier is generalizing a disjunction and hence
should be a kind of generalized existential quantier. And this means that
these quantiers shall be dened similar to 8; 9 in (2.15), (2.16).
As background one needs only all the nite iterations tn : [0; 1]n ! [0; 1],
for n 1, of a given t-norm t, dened inductively by
t1 (x) =def x ;
tk+1 (x1 ; : : : ; xk+1 ) =def t(tk (x1 ; : : : ; xk ); xk+1 ) :
343
Then one is able to dene in any system S of many-valued logic for each

t-norm t a t-norm based universal quantier 8t , and for each t-conorm s a
t-conorm based existential quantier 9s , both of arity (1; 1), by dening their
generalized truth degree functions VerS8t , VerS9s for each universe of discourse
A and each function f : A ! W S as
VerS8t (f ) =def inf ftn (x1 ; : : : ; xn ) j n 1 ^ x1 ; : : : ; xn 2 Ag ;
VerS9s (f ) =def supftn (x1 ; : : : ; xn ) j n 1 ^ x1 ; : : : ; xn 2 Ag :
It is obvious that the quantiers 8; 9, dened previously in (2.15) and (2.16),
are the particular cases for the t-norm t = min and the t-conorm s = max.
These t-norm and t-conorm based quantiers behave well in the sense that
they satisfy a lot of the properties mentioned in this section for the standard
quantiers 8; 9. They are e.g. mutually denable in the case that s = st is
the t-conorm determined by the t-norm t via (5.73), because one then has
j= :8t x H (x) $ 9st x :H (x)
for all ws H (x).
For the t-norm based universal quantiers one furthermore has e.g.:
j= 8txH1 (x) ^t 8txH2 (x) ! 8t x(H1 (x) ^t H2 (x)) ;
j= 8txH1 (x) _t 8txH2 (x) ! 8t x(H1 (x) _t H2 (x)) ;
j= 8tx (H1 (x) !t H2 (x)) !t (8t x H1 (x) !t 8t x H2 (x)) ;
j= 8tx H (x) !t H (a) ;
and for dierent t-norms on which the universal quantiers may be based
also
t1 5 t2 ) j= 8t x H (x) !t 8t x H2(x) :
For the t-conorm based quantiers one gets similar properties which shall not
be listed here.
Instead we mention that one can \get back" the t-norm t from the universal quantier 8t because one has
t(u; v) = ValSA (8tx P(u;v) (x); )
for all u; v 2 [0; 1], each unary predicate symbol P(u;v) , each S-interpretation
A with a universe of discourse A which contains at least two dierent elements
a; b and which interprets P(u;v) as P(Au;v) = A f1g \ f(a; u); (b; v)g, and for
each A-assignment .
This \rediscovering" of the t-norm t from the quantier 8t is, of course,
only an interesting tool if one is able to characterize the t-norm based universal quantiers without reference to their basic t-norms. This is done in [533],
but shall not be discussed here.
1
344
14. Axiomatizing t-Norm Based Logics
For the innitely many-valued cases with truth degree set W = [0; 1] the
Lukasiewicz system L1 , the Go del system G1 , and the product logic all
are systems of many-valued logic which are (or can be) t-norm based1 in
the sense of Chapter 13. Contrary to the intuitive approach of Chapter 13,
however, for them we discussed adequate axiomatizations. Thus the problem
now is to try to get adequate axiomatizations also for the t-norm based logical
systems of Chapter 13. Because of the diversity of t-norms the most suitable
way to reach this goal seems to be to look for axiomatizations which cover
a wide class of such systems { in the sense that e.g. each particular t-norm
based system then can be axiomatized by some extension of some \core"
system.

14.1.1 Some particular cases
As a rst remark let us mention that the adequate axiomatizations of the

systems L1 and G1 cover an even wider range of t-norm based systems. The
essential result for such an extension is Corollary 5.1.5, which also implies
that all the additional truth degree functions one can dene in [0; 1] using
either the Lukasiewicz t-norm et2 or the product t-norm et3 , can similarly
be dened in the (lattice) ordered semigroups h[0; 1]; t; i with either a continuous Archimedean t-norm with zero divisors, or with a continuous and
strict t-norm. Therefore the isomorphisms given in (the proof of) Theorem
5.1.3 are also isomorphisms of the extended algebraic structures which result
from the adjunction of such additional operations which are dened (from
the basic t-norms) in the same way in all these structures { particularly for
the t-norm based R-implications and negations.
On the other hand one knows from Proposition 5.4.1 that not only a given
t-norm determines (uniquely) its corresponding R-implication operation, but
also that such an R-implication operation determines its corresponding tnorm. And this means, from the point of view of logical systems, that it is
1 For the nitely-valued systems Lm and Gm one additionally has to restrict the
set [0; 1] of truth degrees of the t-norm based systems to a suitable nite subset.
346
inessential whether such a system has in its language the conjunction connective, or the implication connective (or even both) as a primitive one.
So we get the following two results.
Proposition 14.1.1. The logical calculus K for the product logic gives an
adequate axiomatization for each t-norm based system St which is based upon
a strict continuous t-norm t, and formulated in the language with the primitive connectives &t ; !t and the truth degree constant 0.
Proof: Because of Theorem 5.1.3(i) one has that the algebraic structures
h[0; 1]; t; i and h[0; 1]; et3 ; i are isomorphic. Hence the algebraic structures
h[0; 1]; t; seq t ; nont ; i and h[0; 1]; et3 ; seq ; non0 ; i are isomorphic too, and
thus also the structures h[0; 1]; t; seq t ; 0i and h[0; 1]; et3 ; seq ; 0i. And one can
dene in h[0; 1]; t; seq t ; 0i the operation non0 and the (lattice) ordering .
This means all together that the t-norm based system St can be based on
the logical matrix h[0; 1]; t; seq t ; 0i, and that this logical matrix is isomorphic
to the logical matrix for the product logic. Therefore one immediately gets
by routine calculations that the logical systems St and have the same
entailment relation, and hence can be adequately axiomatized by the same
logical calculus.
2
Proposition 14.1.2. The logical calculus K L for the Lukasiewicz system
L1 gives an adequate axiomatization for each t-norm based system St which

is based upon a continuous Archimedean t-norm t with zero divisors, and
formulated in the language with the primitive connectives &t ; !t ; ;t .
Proof: Because of Theorem 5.1.3(ii) one has that the algebraic structures
h[0; 1]; t; i and h[0; 1]; et2 ; i are isomorphic. Now one argues as in the proof
of the preceeding proposition.
2
14.1.2 A global approach

There is, however, still an important dierence between the previous approaches toward all these particular t-norm based systems, and also toward
the Post systems, and the present situation to look for some core system
for the t-norm based systems in general: for all these previously discussed
systems S some single, \standard" semantical matrix was constitutive. And
this semantic approach toward S was always determined, cf. Chapter 2, by
the sets W S and DS of truth degrees and of designated truth degrees, by the
family of truth degree functions verS' corresponding to the basic connectives
' 2 J S , and by the family of truth degrees tS corresponding to the truth
degree constants t 2 KS , i.e. by some standard logical matrix for S.
For some core system for the class of all t-norm based systems in general
(it seems that) one does not have such a standard logical matrix. Therefore
another type of semantical characterization is needed which is provided by
347
some suitable class of algebraic structures, and which thus is similar to the
semantical characterizations of the Lukasiewicz systems by MV-algebras,
of the Go del systems by G-algebras, of the product logic by the product
algebras, and of the Post systems by the Post algebras.
Looking back at the way t-norms have been used in Chapter 13, one
recognizes that there was an important restriction { to left continuous (or
even to continuous) t-norms, because one (usually) likes to have available
an R-implication connective in each one of these t-norm based systems of
many-valued logic.
And this may seem to cause some problem. It is clear that for t-norms
their dening conditions (T1), . . . , (T4) are \algebraical" ones in the sense
that they are either formulated by equations which some operation has to
satisfy, or formulated by an elementary condition which relates a relation
and an operation. The properties of left continuity and of continuity, however,
are more dicult to state: they need reference to the notion of limit. And
this is not (genuinely) an algebraic notion. The problem hence is that one
has to nd { algebraically characterizable { classes of algebraic structures
which \approximate" well the class of algebraic structures determined by
the continuous, or by the left continuous t-norms (with their associated Rimplication functions).
In any case it is structurally important for the t-norm based systems that
one has on the one hand that h[0; 1]; t; 1i is a commutative semigroup with
a neutral element, i.e. is a commutative monoid. On the other hand it is
important that is a (lattice) ordering { and even a complete one, i.e. all
suprema and inma exist inside the lattice h[0; 1]; i { which has a universal
lower bound and a universal upper bound. And it is also important that
both structures \t together" in the sense that the semigroup operation t is
non-decreasing w.r.t. this lattice ordering.
Altogether this means from the structural, \algebraic" point of view that
one has a (complete) commutative lattice-ordered monoid as a truth degree
structure in all these t-norm based systems.
However, in general, one can have in a commutative lattice-ordered monoid
dierent elements as universal upper bound of the lattice, and as neutral element of the monoid. This is not the case in the t-norm based systems, i.e.
for them one even has an integral commutative lattice-ordered monoid as
a truth degree structure. Furthermore one usually likes to have the t-norm
combined with a corresponding R-implication operation, which algebraically
means that the integral commutative lattice-ordered monoid constituted by
the truth degree structure is a residuated one.
348
And this type of algebraic structure, at which we nally arrived, is called

residuated lattice.2 It has been introduced into lattice theory already in [119,
563, 564] as a generalization of a concept from ideal theory.3
Denition 14.1.1. An algebraic structure hL; \; [; ; ; 0; 1i is a residuated lattice i one has that
(i) hL; \; [; 0; 1i is a lattice with zero and unit element (and with lattice ordering 6);
(ii) hL; ; 1; 6i is a commutative lattice-ordered monoid;
(iii) the binary operations and form an adjoint pair, i.e. satisfy for all
x; y; z 2 L the adjointness condition
z 6 (x y ) , x z 6 y :
(14.1)
The \monoidal product" of a residuated lattice hence has all the crucial
properties (T1), . . . , (T4) of a t-norm. Furthermore these residuated lattices
have some quite nice properties which can be proved rather easily.
Corollary 14.1.1. For each residuated lattice hL; \; [; ; ; 0; 1i and for all
a; b; c 2 L one has:4
(i) a 6 b , a b = 1 ;
(ii) a (a b) 6 b and b 6 a a b ;
(iii) is non-decreasing in its second argument;
(iv) a (b c) = a b c ;
(v) a (b [ c) = (a b) [ (a c) ;
Proof: (i) One has a 6 b i a 1 6 b, hence i 1 6 a b by (14.1), and

hence i 1 = a b.
(ii) Both inequalities follow easily from the adjointness condition (14.1)
and the re exivity of the lattice ordering 6.
(iii) By b 6 a a b one has for b 6 c immediately b 6 a a c, and
thus a b 6 a c by (14.1).
(iv) From (ii) and the monotonicity of one gets by the integrality property ab(a (b c)) 6 c and ab(ab c) 6 c, and then by the adjointness condition (14.1) and the commutativity of also a (b c) 6 ab c
and a b c 6 a (b c).
2 As a side remark it should be mentioned that also a slightly more general al-
gebraic structure, subresiduated lattices, has been of interest inside logic, e.g.
in [145]. This consideration, however, does not concern many-valued logics, and
hence will not be discussed here.
3 The ideal theoretic notion was called Ideal-Quotient in German, and appears to
be due to Dedekind, cf. [563, 119].
4 In the following notations we give preference over to avoid parentheses.
349
(v) By the monotonicity of one has a b 6 a (b [ c) as well as a c 6

a (b [ c), and thus (a b) [ (a c) 6 a (b [ c). And from b 6 a a b and the
monotonicity of in the second argument one gets b 6 a (a b) [ (a c).
Similarly one has c 6 a (a b) [ (a c) and hence b [ c 6 a (a b) [ (a c).
Thus (14.1) gives also a (b [ c) 6 (a b) [ (a c).
2
The adjointness condition (iii) excludes now, according to Propositions

5.4.1 and 5.4.2, all those t-norm based systems for which the basic t-norm
is not left continuous. Hence this adjointness condition (iii) is in the present
setting the suitable algebraic equivalent of the analytical notion of left continuity.
And also the continuity of the basic t-norm has an algebraic equivalent
in the setting of residuated lattices: the property of divisibility for the corresponding commutative lattice-ordered monoid. Such a monoid hL; ; 1; 6i
is called divisible i for all a; b 2 L with a 6 b there exists some c 2 L with
a = b c. We shall coin a residuated lattice as divisible i its corresponding
commutative lattice-ordered monoid is divisible.
This property of divisibility has also another interesting characterization
which relates it immediately to the behavior of the rule of detachment in a
system of many-valued logic based on a divisible residuated lattice as a truth
degree structure.
Proposition 14.1.3. A residuated lattice L = hL; \; [; ; ; 0; 1i is divisible
i one has a \ b = a (a b) for all a; b 2 L.
Proof: Following [271], we rst show that one has in each residuated
lattice
a (a b) = b , 9 x(a x = b)
(14.2)
for all a; b 2 L. Of course, in the case a (a b) = b one has 9 x(a x = b). So
suppose a c = b for some c 2 L. If one then would have a (a b) 6= b, this
would mean a (a b) < b = a c because one always has a (a b) b
by the adjointness condition, and this hence would mean c 6 a b (because
otherwise c a b and hence b = a c 6 a (a b) would be the case) and
therefore also a c = c a 66 b by the adjointness condition, a contradiction.
Thus (14.2) is established.
Supposing now the divisibility of hL; \; [; ; ; 0; 1i, then one has for all
b 6 a 2 L from 9 x(b = a x) immediately a (a b) = b = a \ b. And
for a 6 b one has a b = 1 from the adjointness property, and hence also
a (a b) = a 1 = a = a \ b.
Assuming on the other hand that one always has a \ b = a (a b), one
has for all a 6 b 2 L from a = a \ b = b \ a the equation a = b (b a),
and hence 9 x(a = b x).
2
Corollary 14.1.2. In each divisible residuated lattice hL; \; [; ; ; 0; 1i one

has for all a; b; c 2 L:
350
(i) a > b , a (a b) = b ;
(ii) b > a; c ) a (b c) = c (b a) :
Proof: Because of a > b , a \ b = b, claim (i) is an immediate consequence of the last proposition.
So assume b > a; c. Then one gets from an iterated application of (i)
a (b c) = b (b a) (b c) = (b a) c
and thus (ii).
2
Proposition 14.1.4. A residuated lattice h[0; 1]; min; max; t; seq t; 0; 1i, determined by some t-norm t, the t-norm t is continuous.
Proof: Assume rst that t is continuous. Then one has for a b 2 [0; 1]
immediately t(a; seq t ) = t(a; 1) = a = minfa; bg. And one has for b < a:
t(a; seq t(a; b)) = t(a; maxfz j t(a; z) bg)
(14.3)
= maxft(a; z ) j t(a; z ) bg b
already by the left continuity of t. Continuity of t furthermore gives from
0 = t(a; 0) b < a = t(a; 1) the existence of some c 2 [0; 1] with b = t(a; c),
and thus t(a; seq t (a; b)) = b = minfa; bg by (14.3). Hence this residuated
lattice is divisible according to Proposition 14.1.3.
Assume conversely that h[0; 1]; min; max; t; seq t ; 0; 1i is a divisible residuated lattice. Then the adjointness condition forces that t has to be left
continuous, cf. Propositions 5.4.1 and 5.4.2. Hence for the continuity of t
one has to show that t is also right continuous. Suppose that this is not
the case. Then there exist some a; b 2 [0; 1], and in [0; 1] some decreasing
sequence (xi )i0 with limi!1 xi = b such that t(a; b) 6= inf i t(a; xi ), i.e.
such that t(a; b) < inf i t(a; xi ). Consider now some d with t(a; b) < d <
inf i t(a; xi ) a. Then there does not exist some c 2 [0; 1] with d = t(a; c),
because otherwise one would have d = t(a; c) > t(a; b), hence c > b and thus
inf i t(a; xi ) t(a; c) = d from the fact that b = limi!1 xi and there thus
exists some integer k with xk c. This means that the lack of right continuity for t would cause the lack of divisibility for h[0; 1]; min; max; t; seq t ; 0; 1i.
Therefore divisibility implies the continuity of t.
2
It seems that one has with the class of all residuated lattices some kind
of \rst approximation" for an algebraic semantics to cover the class of all
t-norm based systems of many-valued logic.
However, a further restriction is even suitable for this purpose, having in
mind that the t-norm based residuated lattices h[0; 1]; min; max; t; seq t ; 0; 1i
have a linearly ordered set of truth degrees, and hence yield
j= (H1 !t H2 ) _ (H2 !t H1 )
351
for each one of the t-norm based systems St . From the algebraic point of
view this additional property may (partly) be taken into account with the
condition of pre-linearity 5 which demands that
(x y) [ (y x) = 1
(14.4)
is satised for a residuated lattice. It is interesting to notice that (14.4) can
equivalently be characterized in another form, which will become important
later on.
Proposition 14.1.5. In residuated lattices the following two conditions are
equivalent:
(i) (x y) [ (y x) = 1 ;
(ii) ((x y) z ) ((y x) z ) 6 z :
Proof: Let L = hL; \; [; ; ; 0; 1i be any residuated lattice. Because [
is the supremum w.r.t. 6, one immediately has the inequalities
(x y) 6 (x y) [ (y x) ;
(y x) 6 (x y) [ (y x) ;
i.e. one has the equations
;

(x y) (x y) [ (y x) = 1 ;
;

(y x) (x y) [ (y x) = 1 :
From these equations condition (ii) yields (x y) [ (y x) = 1, i.e.
condition (i).
On the other hand, if one has condition (i) satised, then one has also
condition (ii) satised as the following elementary calculations show:
((x y) z ) (y x) z )
;

= ((x y) z ) (y x) z ) (x y) [ (y x)
6 ;((x y) z) (x y) [ ;((y x) z) (y x)
6 z[z = z:
And a last additional property of residuated lattices which shall be considered in the following discussions is the double-negation property
((x 0) 0) = x :
(14.5)
Following [246], a pre-linear divisible residuated lattice shall be called a
BL-algebra.6
5 This is the name chosen for this property in [246]. Instead, in [271] this property
is called the strong algebraic deMorgan law.
6 The \BL"-part of this name comes from the fact that the logic determined by
the class of all BL-algebras (i.e. which has as its logically valid ws just all those
352
In any case it is clear that each one of the t-norm based systems of manyvalued logic has as its truth degree structure h[0; 1]; min; max; t; seq t ; 0; 1i a
pre-linear residuated lattice, if the basic t-norm t is left continuous, or even
a BL-algebra, if t is continuous. This means that all the ws of (the t-normdisjunction free fragment of) the language Lt which are valid in all pre-linear
residuated lattices (or: in all BL-algebras) are also St -logical valid in each
system St which is based on a left continuous (or: continuous) t-norm.
So the problems arise to get syntactic characterizations of the classes of
all ws of the (t-norm-disjunction free) language Lt which are valid in all prelinear (and divisible) residuated lattices, i.e. to get adequate axiomatizations
for these classes of ws.
For the logical systems S to be constituted here, validity of some w H in
a pre-linear (and divisible) residuated lattice L (with universal upper bound
1) means that H has value 1 for all valuations : V0 ! jLj into the carrier
of the algebraic structure L. The valuation function ValS , which was dened
in (2.6), now has to be taken in essentially the same sense { with the only
dierence that the truth degree function verS' of a connective ' has to be
understood as the operation in the algebraic structure L which corresponds
to this connective '.
Such adequate axiomatizations have recently been given for the class of
ws which are valid in all residuated lattices by Ho hle [270, 271], for the class
of ws which are valid in all pre-linear residuated lattices by Esteva/Godo
[150], and for the class of ws which are valid in all BL-algebras by Hajek
[246]. Moreover, it has been proved in [103, 245, 246] that the class of all ws
which are valid in all BL-algebras coincides with the class of all ws which
are logically valid in all t-norm structures h[0; 1]; min; max; t; seq t ; 0; 1i based
on a continuous t-norm. These results shall be sketched later on.
Following these authors, we use the name basic t-norm logic to refer to the
logical system determined by the class of all BL-algebras, the name monoidal
t-norm logic to refer to the logical system determined by the class of all prelinear residuated lattices, and the name monoidal logic to refer to the logical
system determined by the class of all residuated lattices.
14.1.3 Monoidal logic

The axiomatization of Ho hle [271] for the monoidal logic, i.e. for the class of
ws which are valid in all residuated lattices, shall be given here in a language
LML which has as its basic vocabulary the sets
J ML = f!; &; ^; _g ; KML = f0g
ws of the language Lt which are valid in all BL-algebras) is treated in [246] as a
kind of basic t-norm related logic for all the possible systems St of many-valued
logic which are based on a continuous t-norm.
353
of connectives and of truth degree constants, interpreted in each residuated

lattice hL; \; [; ; ; 0; 1i as the operations ; ; \; [ and as the truth degree
0, respectively. This language is usually extended by the denitions
:H =def H ! 0 ; 1 =def :0 ;
(14.6)
and then becomes the full language of the t-norm based systems St .
This monoidal logic has a possible set of axioms AxML determined7 by the
following list of axiom schemata:
(AxML 1) (H1 ! H2 ) ! ((H2 ! H3 ) ! (H1 ! H3 )) ;
(AxML 2) H1 & H2 ! H1 ;
(AxML 3) H1 & H2 ! H2 & H1 ;
(AxML 4) (H1 ! (H2 ! H3 )) ! (H1 & H2 ! H3 ) ;
(AxML 5) (H1 & H2 ! H3 ) ! (H1 ! (H2 ! H3 )) ;
(AxML 6) H1 ^ H2 ! H1 ;
(AxML 7) H1 ^ H2 ! H2 ^ H1 ;
(AxML 8) H1 & (H1 ! H2 ) ! H1 ^ H2 ;
(AxML 9) 0 ! H ;
(AxML 10) H1 ! H1 _ H2 ;
(AxML 11) H2 ! H1 _ H2 ;
(AxML 12) (H1 ! H3 ) ! ((H2 ! H3 ) ! (H1 _ H2 ! H3 )) ;
and has as its (only) inference rule the rule of detachment (w.r.t. the implication connective !).
The logical calculus which is constituted by this axiom system and its
inference rule, and which has the standard notion of derivation, shall be
denoted by K ML .
Proposition 14.1.6. The (extended) logical calculus K ML is sound, i.e. derives only such formulas which are valid in all residuated lattices.
Proof: It is (essentially) a routine matter to check that all the axioms
of the axiom system AxML are valid in each residuated lattice. And obviously
one has from Corollary 14.1.1(ii) that the rule of detachment (MP) leads from
ws which are valid in some residuated lattice L again to a w which is valid
in L.
2
Proposition 14.1.7. The Lindenbaum algebra of the (extended) logical calculus K ML is a residuated lattice.
The proof needs, as the corresponding proofs e.g. for the product logic or
the basic logic, the derivation of a lot of particular theorems of K ML . This
shall not be done here. All the necessary details are given in [271] as well as
in [270, 272, 273].
7 This means that each w counts as an axiom which falls under one of the fol-
lowing schemata.
354
We only discuss some points of deviance from the axiomatization given

in [271, 273].
Proposition 14.1.8. (i) The logical calculus K ML allows one to derive the
ws
(H1 & H2 ) & H3 ! H1 & (H2 & H3) ;
(14.7)
(H1 ! H1 & :H1 ) ! :H1 :
(14.8)
(ii) From (AxML 1) together with (AxML 6); (AxML 7) one can derive the w
H1 ^ H2 ! H2 :
(14.9)
And using this w one gets (AxML 7) using only (AxML 6); (AxML 8).
(iii) From the axiom schemata (AxML 1), . . . , (AxML 7) one can derive the
w
(H1 ! H2 ) ! ((H1 ! H3 ) ! (H1 ! H2 ^ H3 )) :
(14.10)
And using this w one gets (AxML 8) using the axiom schemata (AxML 1), . . . ,
(AxML 6).
(iv) From the axiom schemata (AxML 1), . . . , (AxML 5) and (AxML 8) one
can derive the w
H1 & H2 ! H1 ^ H2 :
(14.11)
Proof: One rst has from (AxML2) and (AxML 5)
`ML H ! (G ! H ) :
(14.12)
And one gets
`ML (H1 ! (H2 ! H3 )) ! (H2 ! (H1 ! H3 ))
(14.13)
from (AxML 3) via importation/exportation (AxML 4); (AxML 5) and chaining
(AxML 1). Hence one has ws similar to the general axioms (AxRT 1) and
(AxRT 2) of the logical calculi K m
RT , which means that one gets
`ML H ! H
(14.14)
as in K m
RT , cf. the proof of Theorem 6.2.2. And applying this last result with
the w (H1 ! H2 ) instead of H additionally gives
`ML H1 & (H1 ! H2 ) ! H2
(14.15)
by an application of importation (AxML 4) and of chaining (AxML 1).
(i) One immediately has
`ML (H1 & H2) & H3 ! (H1 & H2) & H3 ;
and therefore
`ML (H1 & H2) ! (H3 ! (H1 & H2 ) & H3) ;
via exportation (AxML 9) and chaining (AxML 1), as well as
355
`ML H1 ! (H2 ! (H3 ! (H1 & H2 ) & H3 )) :
So (14.13) together with chaining (AxML 1) gives

`ML H1 ! (H3 ! (H2 ! (H1 & H2 ) & H3 )) :
Now two applications of importation (AxML 8), again together with chaining
(AxML 1), give
`ML (H1 & H2) & H3 ! H1 & (H2 & H3 ) :
And one gets via chaining (AxML 1) from the schemata (AxML 6); . . . , (AxML 8)
immediately
`ML H & (H ! 0) ! 0 :
(14.16)
Having in mind that according to (14.13) one can write (AxML 1) as
`ML (H2 ! H3 ) ! ((H1 ! H2 ) ! (H1 ! H3 )) ;
this gives
`ML (H1 ! H2 & :H2 ) ! :H1
which has (14.8) as a particular case.
(ii) One has as a particular case of axiom schema (AxML 1) the derivability
`ML (H1 ^ H2 ! H2 ^ H1 ) ! ((H2 ^ H1 ! H2 ) ! (H1 ^ H2 ! H2 )) :
From this w two applications of the rule of detachment (MP), using (AxML 6),
(AxML 7), give
`ML H1 ^ H2 ! H2 :
On the other hand one has

(H1 ^ H2 ! H2 ) ! ((H1 ^ H2 ! H1 ) ! (H1 ^ H2 ! H2 ^ H1 ))
as a particular case of axiom schema (AxML 8). And from this w one gets
(AxML 7) by two applications of the rule of detachment, using (14.9) and
(AxML 6).
(iii) One rst has the derivability of each w
(A ! B ) ! (A & C ! B & C )
(14.17)
because one gets the ws
A & (A ! B ) ! B ;
B ! (C ! B & C )
via importation/exportation (AxML 4); (AxML 5) from (14.14). By (AxML 1) one
derives
A & (A ! B ) ! (C ! B & C ) ;
356
and from this w again via importation/exportation successively

A ! ((A ! B ) ! (C ! B & C )) ;
A ! (C ! ((A ! B ) ! B & C )) ;
A & C ! ((A ! B ) ! B & C ) :
Now an application of (14.13) nally gives (14.17). Therefore one gets from
H1 & (H1 ! H2 ) ! H2 the schema
H1 & (H1 ! H2 ) & (H2 ! H3 ) ! H2 & (H2 ! H3 ) ;
which gives via (AxML 8) and chaining (AxML 1) all the ws
H1 & (H1 ! H2 ) & (H2 ! H3 ) ! H2 ^ H3 :
Now it is a routine matter to derive (14.10) via importation/exportation.
On the other hand one gets from (14.10) immediately
(H1 &;(H1 ! H2 ) ! H1 ) !

(H1 & (H1 ! H2 ) ! H2 ) ! (H1 & (H1 ! H2 ) ! H1 ^ H2 ) ;
and thus via (AxML 2) and (14.15) also (AxML 8).
(iv) From (14.12) one gets via (14.17) immediately
H1 & H2 ! H1 & (H1 ! H2 ) ;
and therefore (14.11) by chaining from (AxML 8).
2
Remark. The original axiomatization of Hohle [271] was given in a
language without the truth degree constant 0, did have the ws (14.7) and
(14.8) among its axiom schemata, and was given with the schemata (14.9)
and (14.10) instead of (AxML 7), (AxML 8).
It furthermore did have the schema H1 & :H1 ! H2 instead of (AxML 9).
But this schema immediately comes from (14.16) and (AxML 9) via chaining
(AxML 1). And conversely one has (AxML 9) in the original axiomatization of
Ho hle if one uses the denition 0 =def p0 & :p0 .
Hence the axiomatization in [271] is equivalent to our present axiomatization.
Theorem 14.1.1 (Completeness Theorem). For each w H of LML the
following assertions are equivalent:
(i) H is derivable within the logical calculus K ML ;
(ii) H is valid in all residuated lattices.
Proof: The part (i) ) (ii) here is just the soundness of the calculus
K ML . The part (ii) ) (i) on the other hand results from the fact that the
Lindenbaum algebra of K ML is a residuated lattice and has the class of all
K ML -derivable ws as its unit element, i.e. as the universal upper bound of its
lattice part and at the same time as the neutral element of its monoidal part.
357
Therefore each non-K ML -derivable w is not valid in at least one residuated

lattice.
2
This monoidal logic is not only a core system behind all the t-norm based
systems with R-implication, i.e. with a left continuous t-norm as basic one,
it can also be specialized toward quite dierent systems of (non-classical)
logic by adding suitable further axioms. We look at such axioms rst from
an algebraic point of view, having in mind the dierent algebraic structures
which proved to be important in relation with previously discussed systems
of many-valued logic. First we show that the Heyting algebras are denitionally equivalent with suitable residuated lattices.
Proposition 14.1.9. (i) If hL; \; [; ; ; 0; 1i is a residuated lattice satisfying = \, then hL; \; [; ; 0; 1i is a Heyting algebra.
(ii) For each Heyting algebra hL; u; t; ; 0; 1i the enriched algebraic
structure hL; u; t; u; ; 0; 1i is a residuated lattice.
Proof: (ii) is obvious from Denitions 10.1.1 and 14.1.1.
(i) Let hL; \; [; ; ; 0; 1i be a residuated lattice with = \, then the
adjointness condition (14.1) immediately gives condition (HA2) of Denition
10.1.1. Obviously one also has that hL; \; [; 0; 1i is a lattice, and it is distributive according to = \ and Corollary 14.1.1(v). Therefore the structure
hL; \; [; ; ;; 0; 1i is a Heyting algebra.
2
In connection with the Go del systems the Heyting algebras had, however, not been the most important class of algebraic structures. The most important class are the G-algebras. For them this result can easily be adapted:
also the G-algebras are denitionally equivalent with suitable residuated lattices.
Corollary 14.1.3. (i) If hL; \; [; ; ; 0; 1i is a residuated lattice with =
\, which also satises the pre-linearity condition (14.4) for all a; b 2 L, then
hL; \; [; ; 0; 1i is a G-algebra.
(ii) For each G-algebra hL; u; t; ; 0; 1i the enriched algebraic structure
hL; u; t; u; ; 0; 1i is a residuated lattice which satises the pre-linearity condition (14.4).
Proof: Obvious from the preceding proposition.
2
The next topic is to look at MV-algebras, which also prove to be denitionally equivalent to a particular type of residuated lattices.
Proposition 14.1.10. (i) Let hL; \; [; ; ; 0; 1i be a divisible residuated
lattice which additionally satises the condition a = (a 0) 0 for all
a 2 L. If one denes in L a unary operation ; by ;x =def x 0, and a
binary operation by x y =def ;x y, for all x; y 2 L, then hL; ; ;; 0i
is an MV-algebra.
358
(ii) For each MV-algebra hL; ; :; 0i, the MV-operations ^; _;

; !, and
the element 1 = :0 as introduced in Denition 9.2.2, the algebraic structure
hL; ^; _;
; !; 0; 1i is a divisible residuated lattice which satises the condition a = (a ! 0) ! 0 for all a 2 L.
Proof: (i) Let hL; \; [; ; ; 0; 1i be a divisible residuated lattice. From
the equation a = (a 0) 0 one immediately has a = ; ; a for all a 2 L,
i.e. condition (i) of Denition 9.2.1. One also has ;0 = 0 0 = 1, and
therefore always a ;0 = ;a 1 = 1 according to Corollary 14.1.1(i), i.e.
one has condition (ii) of Denition 9.2.1.
To show that hL; ; 0i is an abelian monoid with neutral element we rst
mention that our assumptions give
;(;a ;b)=;(a ;b) = ;(a (b 0))
=;(a b 0) = (a b 0) 0 = a b
for all a; b 2 L. Thus the commutativity and the associativity of follow
from the commutativity and associativity of . And one also has a 0 =
;a 0 = ; ; a = a for each a 2 L, i.e. 0 is a neutral element.
Because of x y = ;(;x) y = ;x y condition (iii) of Denition
9.2.1 nally means (a b) b = (b a) a for all a; b 2 L. To get this,
we rst mention that one has always a b = ;b ;a because of
;b ;a=(a 0) (a 0) = b (a 0) 0
=b ((a 0) 0) = b a :
Furthermore one has always
;(;b \ ;a)=;(;a (;a ;b)) = ;((a 0) (b a))
=(b a) (a 0) 0 = (b a) ((a 0) 0)
=(b a) a ;
and therefore (a b) b = (b a) a by the commutativity of \.
(ii) Let hL; ; :; 0i be an MV-algebra. Then hL; ^; _; 0; 1i is a lattice with
zero and unit elements. Furthermore one has the associativity and commutativity of
from (MV0 2) and (MV0 1). That 1 is neutral w.r.t.
comes from
(MV0 5), and the monotonicity of
under the lattice ordering 5 is stated
in Corollary 9.2.3(i). Therefore hL;
; 1; 5i is a commutative lattice-ordered
monoid. And nally the adjointness condition for
; ! is the second part of
Corollary 9.2.3(ii).
2
The condition (14.4) of pre-linearity is satised in all MV-algebras. This
comes from the fact that one has 0 = (a
:b) ^ (:a
b) according to (9.64),
and hence 1 = (:a b) _ (:b a) = (a ! b) _ (b ! a).
Theorem 14.1.2. Starting from the monoidal logic one gets an adequate
axiomatization of:
(a)
359
the Lukasiewicz system L1 if one adds to AxML the axiom schemata

H1 ^ H2 ! H1 & (H1 ! H2 ) ;
::H1 ! H1 ;
(b) the Go del system G1 if one adds to AxML the axiom schemata
H1 ! H1 & H1 ;
(H1 ! H2 ) _ (H2 ! H1 ) ;
(c) the product logic if one adds to AxML the axiom schemata
H1 ^ H2 ! H1 & (H1 ! H2 ) ;
(H1 ! H2 ) _ (H2 ! H1 ) ;
H1 ^ :H1 ! 0 ;
::H3 ! ((H1 & H3 ! H2 & H3 ) ! (H1 ! H2 )) :
Proof: (a) According to Proposition 14.1.10 are the MV-algebras denably equivalent with the divisible residuated lattices which satisfy the additional condition a = ; ; a. And the class of MV-algebras characterizes the
Lukasiewicz system L1 according to Corollary 9.2.6. Therefore the problem
to extend the axiom system AxML to an axiom system for the Lukasiewicz
system L1 is nothing but the problem to add to AxML such axioms, or axiom
schemata, which force that the models of the extended axiom system are just
all the MV-algebras, i.e. all the divisible residuated lattices which satisfy the
additional condition a = ; ; a.
Looking at the divisibility condition a \ b = a (a b) one recognizes that
the inequality a (a b) 5 a \ b holds in each residuated lattice, because
one has a (a b) 5 a from the monotonicity of , and a (a b) 5 b from
the adjointness condition. This means that divisibility is already forced if one
considers residuated lattices which are models of H1 ^H2 ! H1 & (H1 ! H2 ),
because these models have to satisfy the inequality a \ b 5 a (a b).
And for the condition a = ; ; a, i.e. for the condition a = (a 0) 0
one recognizes that one has in each residuated lattice always a (a b) 5 b
and hence a 5 (a b) b from the adjointness condition. This means that
one always has a 5 ; ; a. Hence a = ; ; a is already forced if one considers
residuated lattices which are models of ::H1 ! H1 .
(b) According to Corollary 14.1.3 are the G-algebras denably equivalent
with the residuated lattices with = \ which satisfy the pre-linearity condition. Again one recognizes easily that the inequality a b 5 a \ b holds
in each residuated lattice. Therefore = \ holds if one forces that one always has a \ b 5 a b. To have this last condition satised it is, however,
sucient to have always a 5 a a, because one then has for all a 5 b
immediately a \ b = a 5 a a 5 a b, and therefore by symmetry also
a \ b = a 5 a a 5 a b in general. This means that the residuated lattices
which are models of H1 ! H1 & H1 satisfy = \.
360
And the pre-linearity condition (x y) [ (y x) = 1 is satised in all

residuated lattices which are models of (H1 ! H2 ) _ (H2 ! H1 ).
(c) According to Denition 11.1.1 the product algebras hA; ; !; 0i are
denably equivalent with those residuated lattices8 which satisfy the condition (PrA4), cf. p. 295, and which are divisible. Here the divisibility comes
from the denition of the operation u.
The additional axiom schemata which extend AxML to an adequate axiomatization of the product logic therefore have to cause that the residuated
lattices which are models of these additional axiom schemata are divisible
and satisfy the pre-linearity condition. This is done by the rst two axiom
schemata of the list under (c). The next two axiom schemata obviously cause
their models to satisfy the conditions (PrA4)(c) and (PrA4)(b).
So it remains to discuss the distributive laws (PrA4)(d) and (PrA4)(e).
The second one of them, i.e. x (y t z ) = (x y) t (x z ), holds in each
residuated lattice. The reason is that one immediately has (x y) t (x z ) 6
x (y t z ) from the monotonicity of . On the other hand one has for
d = (x y) t (x z ) easily x y 6 d and thus y 6 x ! d by the adjointness
condition. By symmetry one also has z 6 x ! d, and hence (y t z ) 6 x ! d.
But this means x (y t z ) 6 d by the adjointness condition. Both inequalities
give the distributive law (PrA4)(e).
And the rst one of these distributive laws x (y u z ) = (x y) u (x z )
follows from the divisibility property. First x (y u z ) 6 (x y) u (x z )
again immediately comes from the monotonicity of . On the other hand one
has by divisibility
(x y) u (x z ) = (x y) (x y ! x z )
= x (y (y ! (x ! x z ))
= x (y u (x ! x z ))
= x (x ! x z ) ((x ! x z ) ! y) :
Additionally one has always a (a ! b) 6 b from the adjointness condition.
And one has that ! is non-increasing in its rst argument, i.e. one has for
all a 6 b immediately a ! d > b ! d for each d: because one rst has
a c 6 b c from a 6 b, and thus for all d also b c 6 d ) a c 6 d,
which gives for all d also c 6 b ! d ) c 6 a ! d. Thus one can continue
the previous transformation of (x y) u (x z ) with
(x y) u (x z ) 6 x z (z ! y)
= x (z u y) = x (y u z ) :
Both inequalities now give also (PrA4)(d).
8 A comparison of the Denitions 11.1.1 and 14.1.1 shows that conditions (PrA1),
. . . , (PrA3) coincide with the conditions which dene the residuated lattices.
361
Therefore these four axiom schemata suce to give an adequate axiomatization of the product logic.
2
A closer inspection of this proof shows that it lists some interesting connections between algebraic and model theoretic properties of residuated lattices which are also of independent interest.
Corollary 14.1.4. For each residuated lattice L = hL; \; [; ; ; 0; 1i one
has the following relationships:
(a) L is divisible i it is a model of H1 ^ H2 ! H1 & (H1 ! H2 ) :
(b) L has the double-negation property i it is a model of ::H1 ! H1 :
(c) L satises the pre-linearity condition i it is a model of
((H1 ! H2 ) ! H3 ) ! (((H2 ! H1 ) ! H3 ) ! H3 ) ;
and i it is a model of (H1 ! H2 ) _ (H2 ! H1 ) :
(d) L has an idempotent semigroup operation i one has \ = , and this
is the case i L is a model of H ! H & H :
Proof: The previous proof gives all these results up to the claim that
the idempotency of the semigroup operation implies its coincidence with
the lattice meet: \ = .
Because one has in general xy 6 x\y, it remains to show that x\y 6 xy
follows from the idempotency of . However, one has again in general the
subdistributivity property
(a b) \ c 6 (a c) \ (b c) :
And this gives together with the idempotency of
x \ y = (x x) \ y 6 (x \ y) (x \ y) 6 x y :
It is interesting to mention that this monoidal logic covers also further

important systems of non-classical logic. The most important among them
is Girard's commutative linear logic [194] which may algebraically be characterized by the class of all those residuated lattices which additionally have
the double-negation property (14.5): a = (a 0) 0. Therefore one has
from part (a) of the preceding proof immediately the following result.
Proposition 14.1.11. The axiom system AxML together with the additional
axiom schema ::H1 ! H1 gives an adequate axiomatization of the commutative linear logic of Girard.
And also classical (propositional) logic can be axiomatized by an extension
of AxML .
362
Proposition 14.1.12. An adequate axiomatization of classical propositional
logic is given by the axiom system AxML together with the additional axiom
schemata
H1 ^ H2 ! H1 & (H1 ! H2 ) ;
::H1 ! H1 ;
H1 ! H1 & H1 :
Proof: The result comes from the fact that also
Boolean algebras
are particular cases of residuated lattices. According to Section 1.2 such a
Boolean algebra hB; \; [; c ; 0; 1i is a distributive complemented lattice with
zero and unit element. It is a routine matter to check that each Boolean
algebra is a Heyting algebra, i.e. a residuated lattice with = \, and with
the residuation operation a ! b = ac [ b. However, a Boolean algebra has
also the divisibility property a \ b = a \ (a ! b) and satises (ac )c = a.
Therefore a Boolean algebra is also an MV-algebra. On the other hand, if a
residuated lattice is both a Heyting algebra and an MV-algebra, then it is a
distributive lattice as follows from the considerations upon product algebras
in the proof of Theorem 14.1.2, and its operation ;a = a 0 has also the
complementation properties: one has immediately
a \ ;a = a \ (a 0) = 0 ;
a [ ;a = a [ (a 0) = (0 a) [ (a 0) = 1 ;
because each MV-algebra satises the pre-linearity condition.
Therefore the Boolean algebras are (denitionally equivalent with) the
residuated lattices which are at the same time Heyting algebras and an
MV-algebras.
2
14.1.4 Monoidal t-norm logic

If one looks at monoidal logic as a logic which approximates the logic of left
continuous t-norms from the semantical point of view, one is forced to look at
the class of characteristic algebraic structures, i.e. at the class of residuated
lattices. And one has to compare this class with the class of all t-norm based
structures, i.e. with the class of all those residuated lattices which have the
real unit interval [0; 1] as their carrier, and which are determined by some left
continuous t-norm as their semigroup operation. One immediately recognizes
that the condition (14.4) of pre-linearity is satised in all these t-norm based
structures. Thus it seems natural to impose additionally the condition of prelinearity onto the class of characteristic algebraic structures, or similarly to
add e.g. the schema (H1 ! H2 ) _ (H2 ! H1 ) to the list of axiom schemata.
Such an approach toward t-norm based logics was given in [150]. It only
slighty diers from the previously mentioned idea: it is not the axiom schema
(H1 ! H2 ) _ (H2 ! H1 ) which is added, but the slightly more sophisticated
one
363
((H1 ! H2 ) ! H3 ) ! (((H2 ! H1 ) ! H3 ) ! H3 ) :
(14.18)
In the realm of the axiom system AxML this schema does not only cover
the simpler one (H1 ! H2 ) _ (H2 ! H1 ), but allows additionally for a
simplication of the whole system AxML : (weak) disjunction _ becomes a
denable connective.
Proposition 14.1.13. From the schema (14.18) one can derive each w
(H1 ! H2 ) _ (H2 ! H1 )
using the axiom schemata (AxML 1); (AxML 3), (AxML 4), (AxML 5) together with
(AxML 11), AxML 12).
Proof: One has because of (AxML10); (AxML11) the derivability of
(H1 ! H2 ) ! (H1 ! H2 ) _ (H2 ! H1 ) ;
(H2 ! H1 ) ! (H1 ! H2 ) _ (H2 ! H1 ) :
Then (14.18) immediately gives (H1 ! H2 ) _ (H2 ! H1 ).
2
The axiomatization of Esteva/Godo [150] for the monoidal t-norm logic,
i.e. for the class of ws which are valid in all residuated lattices which satisfy
the pre-linearity condition (14.4), is given in a language LMTL which has as
its basic vocabulary the sets
J MTL = f!; &; ^g ; KMTL = f0g
of connectives and of truth degree constants, interpreted in each such residuated lattice hL; \; [; ; ; 0; 1i as the operations ; ; \ and as the truth
degree 0, respectively. This language is usually extended by the previously
mentioned denitions (14.6), i.e. by
:H =def H ! 0 ; 1 =def :0
together with the additional one
H1 _ H2 =def ((H1 ! H2 ) ! H2 ) ^ ((H2 ! H1 ) ! H1 ) ;
(14.19)
and then becomes the full language of the t-norm based systems St .
This monoidal t-norm logic has a possible set of axioms AxMTL determined
by the following list of axiom schemata:
(AxMTL 1) (H1 ! H2 ) ! ((H2 ! H3 ) ! (H1 ! H3 )) ;
(AxMTL 2) H1 & H2 ! H1 ;
(AxMTL 3) H1 & H2 ! H2 & H1 ;
(AxMTL 4) (H1 ! (H2 ! H3 )) ! (H1 & H2 ! H3 ) ;
(AxMTL 5) (H1 & H2 ! H3 ) ! (H1 ! (H2 ! H3 )) ;
(AxMTL 6) H1 ^ H2 ! H1 ;
(AxMTL 7) H1 ^ H2 ! H2 ^ H1 ;
(AxMTL 8) H1 & (H1 ! H2 ) ! H1 ^ H2 ;
(AxMTL 9) 0 ! H ;
364
(AxMTL 10) ((H1 ! H2 ) ! H3 ) ! (((H2 ! H1 ) ! H3 ) ! H3 ) ;

denoted by K MTL .
One easily recognizes that the axiom schemata (AxMTL 1), . . . , (AxMTL 9)
coincide with the schemata (AxMTL 1), . . . , (AxMTL 9) of the logical calculus
K ML for monoidal logic. So there are only two points of dierence with the
axiom system AxML of the monoidal logic: (i) one has the new axiom schema
(AxMTL 10) which expresses the condition of pre-linearity, and at the same time
provides some sort of \proof by cases", and one has (ii) all the disjunction
axioms (AxML 10); : : : ; (AxML 12) deleted.
It is possible, however, to derive all these axiom schemata of AxML , which
are now omitted in the axiom system AxMTL , in the logical calculus K MTL .
Proposition 14.1.14. In K MTL one has the following derivabilities:
`MTL H1 ! H1 _ H2 ;
(14.20)
`MTL H2 ! H1 _ H2 ;
(14.21)
`MTL (H1 ! H3 ) ! ((H2 ! H3 ) ! (H1 _ H2 ! H3 )) :
(14.22)
Proof: Because one immediately has that together with (AxMTL 7) also
the converse implication H2 ^ H1 ! H1 ^ H2 is MTL-derivable from (AxMTL 7),
i.e. because (AxMTL 7) means the commutativity of ^, it is straightforward to
get also
`MTL H1 _ H2 ! H2 _ H1
(14.23)
from denition (14.19) and (AxMTL 7).
From (14.14) one gets via importation/exportation the derivability result
`MTL H1 ! ((H1 ! H2 ) ! H2 ) :
And one has as a particular case of (14.12)
`MTL H1 ! ((H2 ! H1 ) ! H1 ) :
Both results together give (14.20) via (14.10) and chaining.
Then (14.23) immediately yields also (14.21). So it remains to consider
(14.22).
One rst gets from denition (14.19) and (14.20)
`MTL H1 _ H2 ! ((H1 ! H2 ) ! H2 ) ;
and then via (14.13) and chaining
`MTL (H1 ! H2 ) ! (H1 _ H2 ! H2 ) :
And one has as a particular case of (AxMTL 1)
365
`MTL (H1 _ H2 ! H2 ) ! ((H2 ! H3 ) ! (H1 _ H2 ! H3 )) :
Hence (14.13) leads to

`MTL (H2 ! H3 ) ! ((H1 _ H2 ! H2 ) ! (H1 _ H2 ! H3 ))
and therefore (H1 ! H2 ) ^ (H2 ! H3 ) ! (H2 ! H3 ), a particular case of
(14.9), together with chaining and importation/exportation to
;

`MTL (H1 ! H2 ) ! (H1 ! H3 ) ^ (H2 ! H3 ) ! (H1 _ H2 ! H3 ) :
In a similar way, however with an additional reference to (14.23), one gets
also
;

`MTL (H2 ! H1 ) ! (H1 ! H3 ) ^ (H2 ! H3 ) ! (H1 _ H2 ! H3 ) :
Therefore AxMTL 10 and two applications of the rule of detachment give
`MTL (H1 ! H3 ) ^ (H2 ! H3 ) ! (H1 _ H2 ! H3 ) :
So one has from (14.11) and chaining nally
`MTL (H1 ! H3 ) & (H2 ! H3 ) ! (H1 _ H2 ! H3 ) ;
which immediately yields (14.22) by an application of exportation (AxMTL 5).
Proposition 14.1.15. The (extended) logical calculus K MTL is sound, i.e.

derives only such formulas which are valid in all residuated lattices which
satisfy the condition of pre-linearity.
Proof: Because of our previous discussion and Proposition 14.1.6 one
has only to check that the crucial new axiom schema (AxMTL 10) has as its
instances only ws which are valid in all residuated lattices which satisfy the
condition of pre-linearity. This is a routine matter because this axiom schema
corresponds to the inequality
((x y) z ) 6 (((y x) z ) z ) ;
i.e. to the inequality
((x y) z ) ((y x) z ) 6 z :
And this inequality is equivalent to (14.4) according to Proposition 14.1.5. 2
Proposition 14.1.16. The Lindenbaum algebra of the (extended) logical
calculus K MTL is a residuated lattice which satises the condition of prelinearity.
The proof is an immediate corollary of the preceeding Propositions 14.1.7
and 14.1.13.
Theorem 14.1.3 (Completeness Theorem). For each w H of LMTL the
366
(i) H is derivable within the logical calculus K MTL ;

(ii) H is valid in all residuated lattices which satisfy the pre-linearity condition.
(iii) H is valid in all linearly ordered residuated lattices which satisfy the
pre-linearity condition.
Proof: The part (i) ) (ii) here is just the soundness of the calculus
K MTL . The part (ii) ) (i) on the other hand results from the fact that the
Lindenbaum algebra of K MTL is a residuated lattice with pre-linearity and
has the class of all K MTL -derivable ws as its unit element. Therefore again
each non-K MTL -derivable w is not valid in at least one residuated lattice
satisfying pre-linearity.
To get also the equivalence with claim (iii) one may prove, as explained
in [150], that each pre-linear residuated lattice is (isomorphic to) a subdirect
product of linearly ordered residuated lattices.9 As done e.g. in the case of the
systems L1 and , the way to reach this representation result is to consider
lter in residuated lattices L. Such a lter F is any set ; 6= F L which
satises the conditions (i): if x; y 2 F then x y 2 F , and (ii): if x 2 F
and x 6 y 2 L then y 2 F . And such a lter is called prime i it satises
additionally the condition (iii): if x; y 2 F then (x y) 2 F or (y x) 2 F .
The crucial steps then are to prove that each residuated lattice has a linearly ordered residuated lattice as its quotient w.r.t. any prime lter equivalence
x F y =def (x y) 2 F ^ (y x) 2 F ;
and that each residuated lattice is isomorphic to a substructure of the direct
product of all its prime lter quotients. We shall not give the details here. 2
Proposition 14.1.17. Starting from the pre-linear monoidal logic one gets
an adequate axiomatization of:

(a) the Lukasiewicz system L1 if one adds to AxML the axiom schemata
H1 ^ H2 ! H1 & (H1 ! H2 ) ;
::H1 ! H1 ;
(b) the Go del system G1 if one adds to AxML the axiom schema
H1 ! H1 & H1 ;
(c) the product logic if one adds to AxML the axiom schemata
H1 ^ H2 ! H1 & (H1 ! H2 ) ;
H1 ^ :H1 ! 0 ;
::H3 ! ((H1 & H3 ! H2 & H3 ) ! (H1 ! H2 )) :
9 Each linearly ordered residuated lattice L satises the pre-linearity condition. By
linearity one has x 5 y or y 5 x for all x; y 2 L, hence x y = 1 or y x = 1,
and thus (x y) [ (y x) = 1.
367
Proof: Having in mind the algebraic semantics for the logical systems
L1 , G1 , and , these claims follow immediately from Theorem 14.1.2 and
Corollary 14.1.4.
2
14.1.5 Basic t-norm logic
The algebraic characterization of the continuity of t-norms, the divisibility
condition a \ b = a (a b), is added to the basic assumptions of the prelinear monoidal t-norm logic in the axiomatization of Hajek [246] for the
basic t-norm logic.10
It is immediately clear that this divisibility condition can be used to
consider also the (weak) conjunction connective ^ as a denable one.
The basic t-norm logic, hence, is the logic which is characterized by the
class of ws which are valid in all BL-algebras. It is given in [246] in a language
LBTL which has as its basic vocabulary the sets
J BTL = f!; &g ; KBTL = f0g
of connectives and of a truth degree constant, interpreted in each BL-algebra
hL; \; [; ; ; 0; 1i as the operations ; and the element 0.
Dening further connectives and a further truth degree constant as
H1 ^ H2 =def H1 & (H1 ! H2 ) ;
H1 _ H2 =def ((H1 ! H2 ) ! H2 ) ^ ((H2 ! H1 ) ! H1 ) ;
:H =def H ! 0 ;
1 =def 0 ! 0 ;
gives the full expressive power of the (suitable version of the) language Lt of
the t-norm based systems St .
Straightforward, but a bit tedious calculations (in BL-algebras) show that
the additional connectives ^; _ just have the BL-algebraic operations \; [ as
their truth degree functions, cf. [246].
A natural way to get an axiomatization of this basic t-norm logic is to
enrich the axiom system AxMTL for the pre-linear monoidal logic by two
schemata
H1 & (H1 ! H2 ) ! H1 ^ H2 ;
H1 ^ H2 ! H1 & (H1 ! H2 ) ;
which just express divisibility. However, doing so would result in an axiom system which has a dened connective appearing in its axiom schemata. Therefore one is more interested to have the basic t-norm logic determined by an
^-free axiom system.
10 This system of many-valued logic originally was called basic logic or also basic
fuzzy logic by Hajek [246, 245]. The terminus \basic logic" is, however, also in
use in a completely dierent sense: as some weakening of the standard system of

intuitionistic logic, e.g. in [468, 469].
368
With this restriction in mind, this basic t-norm logic can be based upon
an axiom system AxBTL which is determined by the following list of axiom
schemata:
(AxBTL 1) (H1 ! H2 ) ! ((H2 ! H3 ) ! (H1 ! H3 )) ;
(AxBTL 2) H1 & H2 ! H1 ;
(AxBTL 3) H1 & H2 ! H2 & H1 ;
(AxBTL 4) (H1 ! (H2 ! H3 )) ! (H1 & H2 ! H3 ) ;
(AxBTL 5) (H1 & H2 ! H3 ) ! (H1 ! (H2 ! H3 )) ;
(AxBTL 6) H1 & (H1 ! H2 ) ! H2 & (H2 ! H1 ) ;
(AxBTL 7) ((H1 ! H2 ) ! H3 ) ! (((H2 ! H1 ) ! H3 ) ! H3 ) ;
(AxBTL 8) 0 ! H1 ;
denoted by K BTL .
It is obvious that its rst ve axiom schemata coincide with the corresponding schemata of the logical calculus K MTL for the pre-linear monoidal
logic. It is also obvious that the axiom schemata (AxMTL 6), . . . , (AxMTL 8)
have to be deleted. Here schema (AxMTL 8) is substituted by the denition of
^. Schema (AxMTL 6) is easily derived from the denition of ^ and the schema
(AxBTL 2). So it remains to consider schema (AxMTL 7). And this is, via the
denition of ^, just (AxBTL 6).
Now it is (again) a standard procedure to extend the logical calculus K BL
to a logical calculus for the full language Lt simply by adding two axiom
schemata instead of each one of these denitions, each one being an implication with one of the two sides of the denitional equality as its antecedent,
and the other side as its succedent. So the denition of ^ e.g. is to be transformed into the two new axiom schemata:
(^1 ) H1 ^ H2 ! H1 & (H1 ! H2 ) ;
(^2 ) H1 & (H1 ! H2 ) ! H1 ^ H2 :
Proposition 14.1.18. The (extended) logical calculus K BL is sound, i.e. derives only such formulas which are valid in all BL-algebras.
Proof: It is (essentially) a routine matter, but a bit tedious as may be
seen in [246], to check that all the axioms of the axiom system AxML are
valid in each residuated lattice. And again one has from Corollary 14.1.1(ii)
that the rule of detachment (MP) leads from ws which are valid in some
BL-algebras L again to a w which is valid in L.
2
Proposition 14.1.19. The Lindenbaum algebra of the (extended) logical

calculus K BL is a BL-algebra.
369
The proof needs, as the corresponding proof e.g. in the case of the product
logic in Chapter 11, the derivation of a lot of particular theorems of K BL . This
shall not be done here. All the details are given in [246]. These show also that
the designated element 1 of this Lindenbaum algebra of K BL is the class of
all K BL -derivable ws.
Theorem 14.1.4 (Completeness Theorem). For each w H of LBL the
(i) H is derivable within the logical calculus K BTL ;
(ii) H is valid in all BL-algebras;
(iii) H is valid in all linearly ordered BL-algebras, i.e. in all BL-chains;
(iv) H is valid in all t-norm structures h[0; 1]; min; max; t; seq t ; 0; 1i based on
a continuous t-norm.
The proof may proceed along a similar way as the corresponding proofs
for the completeness theorems for the innitely many-valued Lukasiewicz
system L1 and the product logic , which did refer either to the class of
all MV-algebras or to the class of all product algebras. The crucial point
is to show that each BL-algebra is (isomorphic to) a subdirect product of
BL-chains, and that each BL-chain is an ordinal sum of rather particular
BL-chains. The details are in [103, 245, 246] and shall only be sketched in
the following subsection.
The basic t-norm logic hence is really the suitable axiomatization for the
common part of all the particular t-norm based logics, as long as the t-norms
are supposed to be continuous.
This completeness theorem for K BTL can be strengthened in an interesting way which refers to extensions of the axiom system AxBTL by further
axiom schemata. Such an extension of AxBTL by some (nite or innite) set
C of axiom schemata
shall be called a schematic extension of AxBTL and denoted by AxCBTL . For each such schematic extension AxCBTL one obviously has
a corresponding extension K BTL (C ) of the logical calculus K BTL .
Theorem 14.1.5 (Extended Completeness Theorem). For each w H
of Lt , each set C of further axiom schemata, and the set C of all instances
of axiom schemata of C , i.e. of all the ws of Lt which fall under one of the
axiom schemata of C , the following assertions are equivalent:
(i) H is derivable within the logical calculus K BTL (C );
(ii) H is valid in all BL-algebras which are also models of C ;
(iii) H is valid in all BL-chains which are also models of C .
Proof: For the soundness part (i) ) (ii) of the proof one has only to
observe that each K BTL (C )-derivable w H has to be valid in all BL-algebras
which are also models of C . For its converse assertion (ii) ) (i) one e.g. has
to show that the Lindenbaum algebra of K BTL (C ) is not only a BL-algebra
but also a model of C .
370
Assertions (ii) ) (iii) is obvious.

And the nal part (iii) ) (ii) may be shown by observing that each
quotient structure B =P of some BL-algebra B which is a model of C w.r.t.
some prime lter P of B is not only a BL-chain, but also a model of C , and
that each subalgebra of a direct product of BL-chains, which are models of
C , is again a model of C .
2
Proposition 14.1.20. Starting from the basic t-norm logic one gets an adequate axiomatization of:
(a) the Lukasiewicz system L1 if one adds to AxBTL the axiom schema
::H1 ! H1 ;
(b) the Go del system G1 if one adds to AxBTL the axiom schema
H1 ! H1 & H1 ;
(c) the product logic if one adds to AxBTL the axiom schemata
H1 ^ :H1 ! 0 ;
::H3 ! ((H1 & H3 ! H2 & H3 ) ! (H1 ! H2 )) :
Proof: Having in mind the algebraic semantics for the logical systems
L1 , G1 , and , these claims follow immediately from Theorem 14.1.2 and
Corollary 14.1.4.
2
14.1.6 Completeness under continuous t-norms
Remembering the few remarks which followed the statement of the Completeness Theorem 14.1.4 rst the notion of prime lter has to be dened
for BL-algebras, which is done in a very similar manner to e.g. the case of
product algebras, cf. Denition 11.1.2.
Denition 14.1.2. Let A = hA; \; [; ; ; 0; 1i be some residuated lattice.
A non-empty subset F A is a lter in A i F is closed under , i.e. one
has a b 2 F for all a; b 2 F , and is upwards closed under the lattice ordering
6, i.e. has b 2 F for all a 2 F; b 2 A with a 6 b. And such a lter is a prime
lter i it is dierent from A and satises also
(a b) 2 F _ (b a) 2 F
for all a; b 2 A.
For each such lter F of a given residuated lattice A an equivalence relation F is determined by
a F b =def (a ! b) 2 F ^ (b ! a) 2 F :
371

\; [; ; of A. Therefore it is a routine matter to prove that the quotient structure A=F , with its operations \; [; ; dened representativewise between equivalence classes under F , and with the designated elements
0 = [0]; 1 = [1], is again a residuated lattice. Even more: if A is a BL-algebra
then also this quotient A=F is a BL-algebra.
Lemma 14.1.1. Let A be a BL-algebra and F one of its lters. Then one
has:
(i) If F is a prime lter then the quotient structure A=F is a BL-chain, i.e.
a linearly ordered BL-algebra.
(ii) For each 1 6= a 2 A there exists a prime lter Fa with a 2= Fa .
Again as in previous cases this gives the crucial subdirect representation
result.
Proposition 14.1.21. Each BL-algebra is (isomorphic to) a subdirect product of BL-chains.
Particularly, therefore, one has that each w of the language LBTL of basic
t-norm logic which is valid in all BL-chains is also valid in all BL-algebras.
In summary these results mean that one has the implications (i) ) (ii),
(ii) ) (iii), and (iii) ) (ii) for the claims in the Completeness Theorem
14.1.4. And (ii) ) (i) follows from Proposition 14.1.19 together with the fact
that the designated element 1 of the Lindenbaum algebra of K BTL is the
class of all K BTL -derivable ws.
So it remains to show that also claim (iv) of this Completeness Theorem
is equivalent to the other ones. Again (iii) ) (iv) is obvious. But the converse implication needs a closer look at the BL-chains. For this, a suitable
adaptation of the notion of ordinal sum of t-norms has to be introduced.
Denition 14.1.3. Let (I; v) be a chain with least element 0 and largest
element 1, and (Ai )i2I a family of BL-chains. Denote for each i 2 I nf1g by
i+ either the upper neighbor11 of i, in the case that it exists, or otherwise i
itself. Suppose furthermore that each of the BL-chains Ai has the form
Ai = hAi ; \i ; [i ; i ; i ; i; i+i ;
and that for i v j one has (Ai nfi; i+g) \ (Aj nfj; j + g) = ;, i.e. that dierent
BL-chains of the familyP(Ai )i2I overlap at most in their universal bounds.
Then the ordinal sum i2I Ai of the family (Ai )i2I shall be the algebraic
structure
X
[
Ai =def h Ai ; \ ; [ ; ; ; 0; 1i
i2I
i2I
11 This means the (in case of existence uniquely determined) j 2 I with i @ j such
that there does not exist any k 2 I with i @ k @ j .
372

S
with the operations dened for all a; b 2 i2I Ai by

b 2 Ai
a \ b =def aa ;\i b ; ifif a;
a 2 Ai , b 2 Aj , and i v j ;

b 2 Ai
a [ b =def ab ;[i b ; ifif a;
a 2 Ai , b 2 Aj , and i v j ;

b 2 Ai
a b =def aa \ibb; ; ifif a;
a 2 Ai , b 2 Aj , and i 6= j ;
8
if a \ b = a
<1;
a b =def : a i b ; if a; b 2 Ai , a 6= b, and a \ b = b
b;
if a 2 Aj n Ai , b 2 Ai n Aj , and i @ j .
S
Particularly this means that one has I i2I APi , and that all the elements i 2 I are -idempotents of the ordinal sum i2I Ai . On the other
hand, for each BL-chain A one can start from some set I of its -idempotents,
and construct from this set a representation of A as an ordinal sum with index set I , cf. [245]. The construction is rather similar to the construction
of ordinal summands in the proof of the Representation Theorem 5.1.1 for
continuous t-norms.
Proposition 14.1.22. The ordinal sum Pi2I Ai of a family (Ai )i2I of BLchains is a BL-chain with all i 2 I as idempotents (under ).
As an immediate corollary of this result one has also that each BL-chain,
which is the ordinal sum of at least two
S BL-chains, is the ordinal sum of
exactly two BL-chains. Furthermore h iS
2I Ai ; \ ; [ i is a lattice, and its
lattice ordering 6 can for any a; b 2 i2I Ai with a 2 Aj ; b 2 Ak be
characterized with reference to the lattice orderings 6i of the summands by
the condition
a 6 b , j @ k _ (j = k ^ a 6j b) :
Denition 14.1.4. A BL-chain is reducible i it is the ordinal sum of two
BL-chains which both have at least two elements; otherwise it is irreducible.
Denition 14.1.5. (i) A pair (X; Y ) of subsets of a BL-chain A satisfying
X [ Y = A is a cut in A i Y is closed under , and one has for x 2 X and
y 2 Y always x 6 y as well as x y = x.
(ii) A BL-chain A is saturated i for each cut (X; Y ) of A there exists
some -idempotent d 2 A such that x 6 d 6 y holds for all x 2 X and y 2 Y .
Proposition 14.1.23. Each BL-chain A can be isomorphically embedded
into a saturated BL-chain A .
The crucial step in the proof is that all the \gaps" in the given BL-chain,
i.e. all the cuts (X; Y ) for which there does not exist some idempotent d with
x 6 d 6 y holds for all x 2 X and y 2 Y , have to be \lled in" with such an
idempotent, cf. [245].
373
Even more can be reached by this construction: only (non-trivial) idempotents are added, and the (carrier of the) old structure A is dense in the
(carrier of the) extended one.
Proposition 14.1.24. Each saturated BL-chain is the ordinal sum of some
family of saturated irreducible BL-chains.
Furthermore it is important that the structure of the irreducible saturated
BL-chains is not too complicated, as shown in [103, 245].
Proposition 14.1.25. An irreducible saturated BL-chain with at least two
elements is a MV-chain if it has some (non-trivial) zero divisor, and it is a
product chain otherwise.
Hence there exist among the non-trivial irreducible saturated BL-chains
only two types. For the trivial BL-chains, i.e. those with only one element,
there is, obviously, no dierence between these two types.12
Now recall that each MV-chain is, according to Corollary 9.2.7, locally
embeddable into the innite Lukasiewicz MV-algebra hW1 ; vel2 ; non1 ; 0i,
and that furthermore, according to Corollary 11.1.2, each product chain is
locally embeddable into the unit interval product algebra hW1 ; et3 ; seq ; 0i.
Then one gets from Proposition 14.1.24 together with the Representation
Theorem 5.1.1 for continuous t-norms immediately the following:
Corollary 14.1.5. Each saturated BL-chain is locally embeddable into some
t-norm structure h[0; 1]; min; max; t; !t ; 0; 1i with a continuous t-norm t.
Having all these results, it is easy to nish the proof of the Completeness
Theorem 14.1.4. Assume that one has At j= H for some w H of the language
LBTL . Consider all t-norm structures At = h[0; 1]; min; max; t; seq t ; 0; 1i based
on a continuous t-norm t. If there would then exist some BL-chain B such that
B 6j= H , one would have some jBj-assignment such that ValB(H; ) < 1.
One can additionally suppose that B is a saturated BL-chain. (Otherwise
B should be extended to some saturated BL-chain B according to Proposition 14.1.23. This obviously gives also B 6j= H .) Assume that B is locally
embeddable into a t-norm structure At = h[0; 1]; min; max; t0 ; !t ; 0; 1i.
Let now x1 ; : : : ; xk be the free variables of H , and C = f (x1 ); : : : ; (xk )g.
Consider some partial isomorphism from C into this t-norm structure At .
Consider in At some [0; 1]-assignment with (xi ) = ( (xi )) for all 1
0
12 Isolated idempotents do not lead to the consideration of trivial BL-chains in the
ordinal sum representation of a BL-chain; neither do limit points of idempotents,

as long as they are not dense in some interval. In this last case, however, they
form a whole interval because the set of all idempotents is a closed subset of the
domain (in the sense of the order topology in it). And in this last mentioned
case one could go even a further step and collect all the idempotents of some
(maximal) interval of idempotents into one summand of the ordinal sum, which
then had to be isomorphic to a Godel chain, i.e. a linearly ordered G-algebra,
cf. [103].
374
i k. Then it is a routine matter to show that ValL (H; ) < 1, i.e. to show
that At 6j= H . A contradiction.
Hence also the implication (iv) ) (iii) between the claims of Theorem
0
14.1.4 is established.
It is interesting to notice that the representation result given in Propositions 14.1.24 and 14.1.25 can additionally be used to show the decidability
of the basic t-norm logic, cf. [26].
Proposition 14.1.26. The set tautBTL of all BL-derivable ws is decidable.
One has to show that tautBTL is a recursive set, i.e. that tautBTL and its
complement (in the class of all ws of the language of basic t-norm logic) are
recursively enumerable.
The recursive enumerability of tautBTL is obvious because it is the set of
all derivable ws of a given logical calculus. So the crucial point is to nd
an eective method to show for a given w of this language that it is not a
BL-tautology.
The design of such a method rests on the fact that for each w H 2=
tautBTL there exists a continuous t-norm tH with only nitely many idempotents13 such that H is not valid in the particular t-norm based structure
AtH = h[0; 1]; min; max; tH ; !tH ; 0; 1i, cf. [26]. This means that AtH can be
represented as a nite ordinal sum of summands which either are isomorphic to the innite Lukasiewicz MV-algebra, or are isomorphic to the unit
interval product algebra.
Then this method demands the solution of some nite system of equations
in these particular summands. For the Lukasiewicz type summands such
a solution corresponds to some suitable mixed integer programming problem, cf. [238], and is therefore eectively possible. And for the product type
summands such a solution can e.g. eectively be determined in a suitable
isomorphic copy: the additive group of all non-negative reals (including the
\number" +1).

It is essentially a routine matter to extend the systems of monoidal logic, of
monoidal t-norm logic, and of basic t-norm logic, which have been introduced
on the propositional level in Section 14.1, to rst-order systems.
Their rst-order languages are constituted in the usual way as explained in
Section 2.3. What represents the peculiarities of these systems are on the one
hand their sets of connectives, of truth degree constants, and of quantiers:
the connectives and truth degree constants are in the rst-order case the same
13 This number of idempotents does not exceed the number of symbols of
more than two.
H by
375
as in the propositional case, and the quantiers are the \standard" ones 8; 9
with the understanding as explained in (2.23), (2.24).
On the other hand the notion of interpretation has to be suitably adapted.
Each interpretation A now is combined with one of the algebraic structures L
which are characteristic for the corresponding system { i.e. is either combined
with a residuated lattice L in the case of the monoidal logic, or is combined
with a pre-linear residuated lattice in the case of the monoidal t-norm logic,
or is combined with a BL-algebra L in the case of the basic t-norm logic. In all
these cases this combination has to be taken in the sense that each predicate
symbol P of the rst-order language Lt is interpreted in A as an L-valued
predicate P A in the universe of discourse A = jAj of A, i.e. is interpreted
as a function (of suitable arity) from A into (the carrier L of) L. And the
inmum and supremum operations in (2.23), (2.24) have to be understood
as operations in the lattice L.
To stress this fact that an interpretation A here is always in this sense
tied with an algebraic structure L, such an interpretation shall be called an
L-interpretation. Such a particular L-interpretation A therefore collects
a nonempty set A = jAj of individuals, the universe (of discourse) of the
interpretation,
for each individual constant a of the language Lt a corresponding individual
aA 2 A,
for each predicate symbol P of Lt with arity n a corresponding n-ary Lvalued predicate P A in A.
The satisfaction relation (A; ) j=S H again is a graded relation, as discussed in Section 2.5, but has as its grades now elements of L. The value function ValSA for a w H w.r.t. some L-interpretation A and some A-assignment
is dened as in (2.17) to (2.22), however, combined with the understanding
that the truth degree function verS' in (2.20) has to be taken as the suitable
L-operation, as mentioned previously.
To avoid problems with the evaluation of quantied ws, which may arise
out of the non-existence of some suprema or some inma within the lattice
L, one either supposes that this lattice has to be complete (in the sense that
all suprema and inma exist), or assumes that one considers only such Linterpretations A for which in L all the necessary suprema and inma exist.14
These particular L-interpretations are called safe ones in [246].
It is interesting to mention that each complete residuated lattice satises
also a generalized distributive law which states that distributes over arbitrary suprema. Even more, this generalized distributive law becomes equivalent to the adjointness condition.
14 For an L-interpretation with a nite universe of discourse this means e.g. that
there is no restriction imposed on the lattice L.
376
Proposition 14.2.1. For every complete abelian lattice-ordered monoid15

hL; ; 6i the following assertions are equivalent:
(a) There exists a binary operation such that (; ) is an adjoint pair.
(b) The operation satises the generalized distributive law
a sup bi = sup(a bi ) :
(14.24)
i2I
i2I
Proof: Suppose (a). Then one has 0 6 a 0 from the adjointness

condition. Therefore (14.24) holds for I = ;. So let I 6= ; and c = supi2I (a
bi ). For all i one gets bi 6 a c from a bi 6 c via the adjointness condition.
Hence one has supi2I bi 6 a c, because this supremum exists. Adjointness
now gives a supi2I bi 6 c. The other inequality supi2I (a bi) 6 a supi2I bi is
obvious from the monotonicity property of and the denition of supremum.
So one has (b).
Now suppose (b). The completeness of the lattice L allows one to dene
a binary operation in L by
a b =def supfx 2 L j a x 6 bg :
Then a c 6 b ) c 6 a b is obvious. And from (14.24) one immediately
has a (a b) 6 b, and therefore c 6 a b gives a c 6 a (a b) 6 b.
Hence (; ) is an adjoint pair.
2
S
And validity of H in some L-interpretation A means ValA (H; ) = 1 for
all A-assignments . As usual, such a complete { or: safe { L-interpretation
A in which H is valid is called an L-model of H . Accordingly, logical validity
of H means validity of H in all complete { or in all safe { L-interpretations
A.
Because the axiomatizations for the monoidal logic ML, the monoidal tnorm logic MTL, and the basic t-norm logic BTL have been given in Section
14.1 via axiom schemata, these schemata can without changes also be used
in the present rst-order setting: the schematic letters Hi in those schemata
have now only to be taken as symbolizing ws of the rst-order language.
Denition 14.2.1. The rst-order versions AxML, AxMTL, and AxBTL of the
axiom systems for the monoidal, the monoidal t-norm, and the basic t-norm
logic are formed by extending both axiom systems AxML , AxMTL , and AxBTL
with the (same) axiom schemata
(Ax 1) 8x H (x) ! H [x=t]
15 This shorthand means an abelian lattice-ordered monoid whose underlying lattice
is complete, and hence has a universal lower bound 0 and a universal upper bound
1.
16 This means essentially that the free occurrences of x in H are not within the
scope of a quantication of some (free) variable of the term t.
(Ax 2)
377
H [x=t] ! 9x H (x)

8x (G ! H ) ! (G ! 8x H )
(Ax 4) 8x (H (x) ! G) ! (9x H (x) ! G)
and by extending AxMTL as well as AxBTL additionally with the schema
(Ax 5) 8x (H (x) _ G) ! 8x H (x) _ G
for all variables x which do not occur free in G.
And in all cases the corresponding rules of inference are the rule of detachment (MP) and the rule of generalization (Gen):
(Ax 3)
H; H ! G (MP) ;
G
H
8x H (Gen) :
The logical calculi which are constituted by the axiom systems AxML ,

AxMTL , and AxBTL together with their corresponding rules of inference shall
be denoted by K , K and K , respectively.
ML
MTL
BTL
Theorem 14.2.1 (Soundness Theorem). (i) Each K ML -derivable formula

H is valid in all L-interpretations w.r.t. some complete residuated lattice L.
(ii) Each K BTL -derivable formula H is valid in all safe L-interpretations
w.r.t. some linearly ordered BL-algebra L.
(iii) Each K MTL -derivable formula H is valid in all safe L-interpretations
w.r.t. some linearly ordered residuated lattice L.

Proof: The inference rule of generalization (Gen) obviously leads from
a w H which is valid in some L-interpretation A again to a w 8xH which
is valid in A. And the same holds for the rule of detachment (MP) because,
according to the adjointness condition, each operation in a residuated
lattice has the degree ranking property.
Having in mind the soundness results for the propositional cases, it only
remains to discuss the \new" axioms (Ax 1) to (Ax 5). However, all the ws
which fall under the schemata (Ax 1) or (Ax 2) are obviously valid because
of the denitions of inmum, supremum and the degree ranking property
of , and because the existence of all the necessary inma and suprema is
supposed.
For the case of axiom schema (Ax 3) it is sucient to show for each
suitable degree structure L the inequality
inf
(a bi ) 6 (a iinf
b)
i2I
2I i
for any a 2 L and any family (bi )i2I of elements of L. It seems to be even
simpler to prove here equality, because one immediately has inf i2I bi 6 bk
and therefore (a inf i2I bi ) 6 (a bi ) from the fact that is nondecreasing in its second argument. Hence (a inf i2I bi ) is a lower bound for
378
all a bi . But (a inf i2I bi ) is the greatest lower bound for all a bi , i.e.
the inmum of all a bi , because one has for each x 2 L with x 6 a bi
immediately a x 6 bi by the adjointness condition, hence a x 6 inf i2I bi
and thus x 6 a inf i2I bi again by adjointness.
For the case of axiom schema (Ax 4) it is sucient to show the inequality
inf
(b a) 6 (sup bi a)
i2I i
i2I
for any a 2 L and any family (bi )i2I of elements of L. Again it seems to be
simpler to prove equality. And this is done just as in the previous case, now
using the fact that is non-increasing in its rst argument.
Finally it is sucient for the case of axiom schema (Ax 5) to show the
inequality
inf
(b [ a) 6 inf
b [ a:
i2I i
i2I i
Also here one better proves equality. Immediately one has inf i2I bi [ a 6 bi [ a
and thus inf i2I bi [ a 6 inf i2I (bi [ a). And if one has x 6 inf i2I (bi [ a), i.e.
if one always has x 6 bi [ a, then one has in the case of a 6 inf i2I bi always
x 6 bi , hence x 6 inf i2I bi , and one has in the opposite case17 of inf i2I bi < a
for some k 2 I the inequality bk < a, and thus x 6 a. Hence one has in both
cases x 6 inf i2I bi [ a.
2
As the next topic consider the Lindenbaum algebras of the logical calculi
K ML , K ML , and K BTL . Without going into details of derivations, it shall be
mentioned that essentially as in classical logic one is able to prove that the
renaming of bound variables leads from a w H to a \variant" H 0 of H which
is provably equivalent with H . The details shall not be discussed here.
Proposition 14.2.2. The Lindenbaum algebra of the logical calculus K ML
is a residuated lattice, the Lindenbaum algebra of K MTL is additionally prelinear, and the Lindenbaum algebra of K BTL is a BL-algebra. In all cases
one has
[ 8x H (x)]] = tinf
[ H [x=t]]] ;
[ 9x H (x)]] = sup [ Ht [x=t]]] ; (14.25)
2T t
t 2T
with T denoting the set of terms, [ H ] the equivalence class of H under provable equivalence, and Ht [x=t] a suitable variant of H which does not contain
any bound variable which appears (free) in the term t.
Proof: That for the logical calculi K ML , K MTL , and K BTL their relation
of provable equivalence between ws of Lt , i.e. their relation
=K given by
H
=K G , `K (H ! G) ^ `K (G ! H )
is an equivalence relation, follows as in the propositional cases.
17 Here the linearity of the ordering 6 is used.
379
The crucial point now is to show that one has [ 8x H ] = inf t2T [ Ht [x=t]]]
for all ws H . However, [ 8x H ] 6 inf t2T [ Ht [x=t]]] is obvious from axiom
schema (Ax 1). So assume that [ G] is a lower bound for all [ Ht [x=t]]], i.e.
assume that one always has
`K (G ! Ht [x=t]) :
Then one has for each variable z which does not occur in G particularly
`K (G ! Hz [x=z ]) :
Via the rule (Gen) and axiom schema (Ax 3) one then has also
`K (G ! 8z Hz [x=z ]) ;
which means [ G] 6 [ 8z Hz [x=z ]]] = [ 8x H ] . Therefore [ 8x H ] is the greatest
lower bound of all Ht [x=t], and the rst equality from (14.25) hence proved.
The second one follows similarly.
2
Theorem 14.2.2 (Completeness Theorem). (i) Each w H which is

valid in all L-interpretations A with a complete residuated lattice L is K ML derivable.
(ii) Each w H which is valid in all safe L-interpretations A with a linearly ordered BL-algebra L is K BTL -derivable.
Proof: For both cases rather dierent proofs have been given. Both shall
be sketched here, but important details will not be presented.18
(i) For the case of the monoidal logic the proof strategy in [271, 273] is
to show that each residuated lattice can be completed, i.e. can be extended
in such away that the underlying lattice becomes a complete one in the extension. And this lattice-extension can be chosen in such a way that the
extended structure is again a residuated lattice, and that the canonical embedding ~ of the \old" one into the extended, i.e. the identity mapping, is a
homomorphism (between residuated lattices).
Having these results, the Lindenbaum algebra LML of the logical calculus
K ML can be completed to a complete residuated
lattice LML.

Now one considers the canonical LML-interpretation C for Lt , which has
the set T of all Lt -terms as its universe of discourse, and which interprets
each n-ary predicate symbol P of the language Lt by
P C = f(~(t1 ); : : : ; ~(tn )) j t1 ; : : : ; tn 2 T ^ `K P (t1 ; : : : ; tn )g ;
and each individual constant of Lt by itself.
It is a routine matter to show that for each Lt -sentence G one has

ValML
C (G) = [ G] . Thus one can for a non-K ML -provable w H with x1 ; : : : ; xk
as all its free variables rst show [ H ] = [ 8x1 : : : 8xk H ] , and then { using
18 Actually the completeness of K MTL is an open problem. It may, however, be
expected that its completeness
can e.g. be proved in a similar way as later on in
this proof is done for K BTL .
380
the homomorphic embeddability of LML into LML { that ValML

C (H ) is not the
unit element of LML . Therefore H is not valid in the LML -interpretation C,
and hence not logically valid for the monoidal logic ML.
The lattice extension which is basic here is the MacNeille-completion,
introduced in [340]. Their main idea is to build for a given lattice-ordered set
hL; 6i and each subset A L the sets B(A), B (A) of all lower resp. upper
bounds of A w.r.t. 6. Then one can prove that D(L) = fB(B(A)) j A Lg
becomes a complete lattice under inclusion as lattice ordering, and that there
exists a canonical embedding of hL; 6i into hD(L); i which maps a 2 L to
B (B (fag)) = fx 2 L j x 6 ag, cf [271].
(ii) The completeness proof for (the logical calculus K BTL for) basic tnorm logic BTL is given in [246] along the lines of Henkin's completeness
proof for (a suitable logical calculus for) classical logic. Here the core idea
is to start from some (consistent19 ) set of ws and a further sentence20
H which is not K BTL -derivable from . Then this set has to be completed
in the sense that one forms an extension such that H is also not
K BTL -derivable from , but such that for each pair of sentences G1 ; G2 of
Lt either G1 ! G2 or G2 ! G1 is K BTL -derivable from . At the same time
one extends the language Lt by a countably innite set of new individual
constants in such a way that each universally quantied w 8x H (x), which
is not K BTL -derivable from , has a \witness" c under the new constants
such that already H (c) is not K BTL -derivable from .
Then one can construct the Lindenbaum algebra LBTL ( ) of the logical
calculus K BTL relative to derivability from , i.e. based on the equivalence
relation
H
= G , `K (H ! G) ^ `K (G ! H ) :
Because of the (\logical") completeness of one gets that LBTL ( ) is a
linearly ordered BL-algebra.
According to this construction one has that the canonical LBTL ( )-interpretation (which is constructed as in the case of the monoidal logic) is a model
of , but not a model of the sentence H , i.e. H is not logically valid in the
basic t-norm logic BTL. The particular case = ; gives the claim of the
present theorem.
2
An analysis of this proof procedure shows that also the following Lo wenheim-Skolem theorem holds for rst-order basic t-norm logic.
Theorem 14.2.3 (Lowenheim-Skolem Theorem). Each consistent set
of sentences has a countable model which is a safe L-interpretation A
with a countable and linearly ordered BL-algebra L.
19 As usual, consistency of some set of ws shall mean that there exists some w
H which is not K BTL -derivable.
20
As previously in the case of monoidal logic one nds that it is sucient to discuss
non-derivable sentences, because one always has with some non-derivable w also
some non-derivable sentence.
381
Proof: By the construction sketched in the previous proof one has that
the canonical LBTL ( )-interpretation is a model of . This model is countable because its universe of discourse is the (countable) set of terms of the
language Lt (with countably many new constants). And LBTL ( ) is a countable and linearly ordered BL-algebra according to the previous theorem. 2
The previous completeness theorem has for the case of the basic logic
been extended in [246] in the same way as for the propositional case, using
the notion of schematic extension as introduced in Section 14.1. And as in
this previous Section 14.1 we write K BTL (C ) for the logical calculus which
results from K BTL by such a schematic extension through a set C of axiom
schemata, and we write C for the set of all instances of axiom schemata of
C.
Theorem 14.2.4 (Extended Completeness Theorem). For each w H
of Lt , each set C of further axiom schemata, and each set of ws of Lt ,
the following assertions are equivalent:
(i) H is K BTL (C )-derivable from the set of premisses;
(ii) H is valid in all safe L-interpretations A with a linearly ordered BLalgebra L which are also models of C [ ;
The proof for this result can be given by a slight modication of the
previously sketched proof for the corresponding simple completeness theorem,
i.e. for Theorem 14.2.2(ii).
The situation for rst-order basic t-norm logic diers, however, from the
corresponding situation for the propositional case insofar as the semantic
characterization of the class of all K BTL -derivable ws is not possible only
with reference to interpretations which are based on t-norm based structures
(with carrier [0; 1]). And this is caused by the fact, proved in [247], that the
class of all ws of the language Lt which are valid in all At -interpretations
over some t-norm based structure At is not recursively enumerable, hence
not the class of K -derivable ws for any logical calculus K with only nitary
inference rules.
Proposition 14.2.3. The class of all ws of the language Lt which are Atlogically valid w.r.t. any t-norm based structure At is not recursively enumerable, but 2 -complete, and a proper superset of the class of all K BTL -derivable
ws.
Proof: It is obvious that the class of all K BTL -derivable ws is a subset
of the class of all ws of the language Lt which are At -logically valid w.r.t.
any t-norm based structure At . And it has to be a proper subset if this last
class is not recursively enumerable.
Because of the 2 -completeness of the set LVL 1 of L1 -logically valid ws
it suces to reduce this set recursively to the set of all ws which are valid
in all t-norm based structures.
382
To reach this goal we connect with each w H of Lt , in which the existential quantier 9 does not occur, two other ws21 which shall be determined
as follows:
(i) the w H # results from the w H by replacing all atomic subformulas
P (t1 ; : : : ; tn ) of H by their double negations ::P (t1 ; : : : ; tn );
(ii) with reference to some unary predicate symbol Q and some individual
constant c neither of which occur in H , the w H is formed as
H H # _ :Q(c) _ ::Q(c) :
With this notation, the following claims prove to be equivalent ones:
(a) H 2 LVL 1 , i.e. H is [0; 1]L-logically valid for the Lukasiewicz BLalgebra [0; 1]L = h[0; 1]; min; max; et2 ; seq 2 ; 0; 1i;
(b) H # is At -logically valid for each t-norm structure At which is determined
by a continuous t-norm which has a rst summand in its ordinal sum
representation which is given by a t-norm which is isomorphic to the
Lukasiewicz t-norm et2 ;
(c) H is L-logically valid for all t-norm based structures L .
It is obvious that one has true the implications (c) ) (b) and (b) ) (a).
So assume that H 2 LVL 1 , and let t be a continuous t-norm which has a
rst summand in its ordinal sum representation according to Theorem 5.1.1,
which is determined by a t-norm which is isomorphic to the Lukasiewicz
t-norm et2 , and which is given on the square [0; a] for some a 1. Let At be
the (safe) BL-algebra determined by t. Then [0; a) [ f1g is the carrier of a
sub-BL-algebra Ba of At , which also is isomorphic to the Lukasiewicz BLalgebra [0; 1]L. Furthermore the mapping f on [0; 1] given by f (x) =def ::x
is a BL-algebra homomorphism from At onto Ba which satises

x 2 [0; a)
f (x) = x1 ; ifotherwise,
and which preserves innite meets.
It is a routine matter to prove, using this preservation of innite meets, by
induction on the complexity of the corresponding ws G, that for all t-norm
based BL-algebras A and all A-interpretations A one has
ValA (G; ) = ValA; (G; )
for all jAj-assignments , and for all A-interpretations A; which dier from
A just in the interpretation of the predicate symbols, and have
P A; (a1 ; : : : ; an) = ::P A (a1 ; : : : ; an )
for each n-ary predicate symbol P and all a1 ; : : : ; an 2 jAj.
21 This will be ws of the same language, or of a slight, inessential extension by a
new individual constant.
383
From the isomorphy of the BL-algebras [0; 1]L and Ba one gets that
each w H which is valid in all [0; 1]L-interpretations is also valid in all
Ba-interpretations. Hence H # is valid in all At-interpretations which are
based on a t-norm structure At with a continuous t-norm which has a rst
summand in its ordinal sum representation isomorphic to the Lukasiewicz
t-norm et2 . And this means that the implication (a) ) (b) is proved.
So consider nally the w H for a given w H . Let t be any continuous
t-norm. If it is not the case that the rst summand in the ordinal sum representation for t is isomorphic to the Lukasiewicz t-norm et2 , i.e. if this rst
summand is isomorphic to the product t-norm et3 or if 0 is the inmum of
positive t-idempotents, then t does not have zero divisors, and the t-based
negation is the Go del negation non0 . Therefore one has in this case that
8x(:Q(x) _ ::Q(x)) is At -logically valid w.r.t. the t-based BL-algebra At .
Hence also H is At -logically valid in this case.
If it is the case that the rst summand in the ordinal sum representation
for t is isomorphic to the Lukasiewicz t-norm et2 , and if the w H # is
At -logically valid for the corresponding t-norm structure At , then also H is
At -logically valid by construction of H . Hence also the implication (b) ) (c)
is proved.
2
384
15. Some Three- and Four-Valued Systems
Each system of many-valued logic which is not only intended to be some particular kind of formalism but supposed to express some meaning is confronted
with the problem to oer a meaning for its truth degrees, or at least for its
\additional" truth degrees dierent from the degrees 1 for \true" and 0 for
\false". In the light of this problem, besides multi-dimensional approaches
as mentioned in Chapter 8.3 three-valued as well as four-valued systems get
particular importance because for them only one or two \additional" truth
degree exist and need interpretation. Hence it should not appear as a surprise
that from the viewpoint of philosophically-oriented applications three- and
four-valued systems assumed a more prominent role than other systems, at
least as other nitely many-valued systems.1
15.1 Three-Valued Systems

Here we shall be interested only in some such 3-valued systems and mainly
restrict the attention to two strongly related systems introduced by Bocvar
[66] and Kleene [304], and a system designed as the \true" logic of the
natural language by Blau [63]. The intentions of these authors, connected
with their systems, have been quite dierent despite some strong similarities
of the systems.
The main problem of Bocvar has been the philosophical and logical
analysis of logical and semantical antinomies as they appear in rst-order
and higher-order logic, often in connection with some lack of care e.g. in
the use of the comprehension principle or of metatheoretical notions, cf. e.g.
[55]. Therefore his preferred interpretation of the additional truth degree 21
was its reading as \meaningless", \paradoxical", or \senseless". The starting
point of Kleene, on the other hand, was a mathematical one and related to
1 Of course, this remark re ects the actual state of the art, and does not mean
that nitely many-valued systems with more truth degrees could not be of interest. And even today there have been considered some systems with more truth
degrees: e.g. an eight-valued geometry in [456], a seven-valued system in [20, 191]
and a six-valued system in [3, 181] both for the discussion of default reasoning.
However, in connection with default reasoning also three-valued approaches are
under consideration, e.g. in [122].
386
his research on partial recursive relations. Such relations sometimes may be

undened. Therefore in his case the intended reading of the additional truth
degree 21 was \undened" or \indetermined". Both systems coincide in their
approach to consider the truth degrees 1; 0 just as the counterparts of the
classical truth-values >; ?. Accordingly for both systems the degree 1 is the
only designated one.
The main intention for the Kleene approach to understand the third
truth degree as \undened" sometimes is also realized in another way: via
truth value gaps, i.e. within a generalization of classical logic which does not
add new \truth values", but which allows that some (atomic) ws { without
free variables { do not have a truth-value at all. In all the cases where such
truth value gaps behave truth-functionally it is essentially a purely formal
change either to have truth value gaps, or to have a third truth degree indicating such gaps. Therefore any truth-functional approach via truth value
gaps can be reformulated in a 3-valued manner with the intended reading
\undened" for the third truth degree, and vice versa. However, there are
interesting approaches which refer to truth-value gaps, e.g. in linguistic applications as discussed in Chapter 20, in which these truth-value gaps do not
behave truth-functionally.
Bocvar subdivides his truth degree functions and thus also his connectives into internal and external ones. The characteristic property of the internal truth degree functions is that they have always a (truly: the) function
value dierent from 0; 1 if a argument value diers from 0; 1. The external
truth degree functions, on the other hand, map into f0; 1g. Hence this subdivision is not a (complete) classication because there obviously exist truth
degree functions which are neither external nor internal ones. The basic connectives of the 3-valued system B 3 of Bocvar are2
:; ^+ ; J0 ; J1
(15.1)
for internal negation, internal conjunction, external negation, and external
assertion. They have, respectively, the truth degree functions
non1 ; et ; j0 ; j1 :
The truth degree functions non1 ; j0 ; j1 have been introduced in (5.55), (5.111),
and the truth degree function et is characterized by the truth degree table
et 0 12 1
0
1
2
0
1
2
1
2
1
2
1
2
1 0
or by the formula
0
1
2
2 We do not follow here the notation of Bocvar but adapt the symbols to our
previous usage.

387
et(u; v); if u; v 2 f0; 1g

1;
if u = 12 or v = 12
2
with et indicating the truth value function of classical conjunction.
All these connectives obviously satisfy the normal condition. This again
means that the 3-valued Bocvar system is not functionally complete.
Further internal versions of disjunction, implication and biimplication can
be dened as
H1 _+ H2 =def :(:H1 ^+ :H2 ) ;
(15.2)
H1 !+ H2 =def (H1 ^+ :H2 );
(15.3)
H1 $+ H2 =def (H1 !+ H2 ) ^+ (H2 !+ H1 ) ;
(15.4)
and get truth degree functions which coincide over f0; 1g with their classical
counterparts and are internal truth degree functions. Corresponding external
versions of these connectives result by a uniform approach which e.g. for
external conjunction and external disjunction reads as
H1 e H2 =def J1 (H1 ) ^+ J1 (H2 ) ;
H1 d H2 =def J1 (H1 ) _+ J1 (H2 ) :
Among the basic connectives (15.1) of the 3-valued Bocvar system, : is
the negation of the 3-valued Lukasiewicz system L3 , and J0 ; J1 are denable
in L3 , cf. (9.26), (9.29). And also the internal conjunction ^+ is L3 -denable,
e.g. as
H1 _+ H2 =def (H1 ^ H2 ) _ (H1 ^ :H1 ) _ (H2 ^ :H2 )
(15.5)
as the corresponding truth degree table shows. Thus B 3 is a subsystem of L3 .
It is a proper subsystem because each w H of B 3 in which only the propositional variables p1 ; : : : ; pn occur can be written semantically equivalent in
the \normal form"
J1 (H ) _+ U(pi ) _+ _+ U(pik )
(15.6)
for some suitable subsequence (i1 ; : : : ; ik ) of (1; : : : ; n), using the shorthand
notation
U(p) =def p ^+ :p;
cf. [158]. This representability result means that the truth degree function
seq 2 and hence also the Lukasiewicz implication !L are not B 3 -denable.
For, having a w H of type (15.6) and some j 2 fi1; : : : ; ik g together with
a valuation with (pj ) = 21 , then one immediately has ValB (H; ) = 12 .
But having for such a w H some j 2= fi1; : : : ; ik g and (pir ) 2 f0; 1g for
all r = 1; : : : ; k, then one also has ValB (H; ) 2 f0; 1g. Assuming now the
representability of p1 !L p2 by such a w H0 of the type (15.6), then the
fact that seq 2 ( 12 ; 0) = 12 holds true would force that U(p1 ) had to appear
as a disjunct in the \normal form representation" (15.6) of H0 , and this
et (u; v) =def
388
would lead to ValB (H0 ; ) = 21 for each valuation with (p1 ) = (p2 ) = 12 ,
contradicting seq 2 ( 12 ; 12 ) = 1.
The axiomatizability problem for B 3 has also been discussed and solved
in [157, 159]. We shall not treat it here. The interested reader can nd this
system of axioms also mentioned in [73].
The 3-valued system K 3 of Kleene has the so-called strong connectives
:; ^; _; !K ; $K
(15.7)
with corresponding truth degree functions, respectively:
non1 ; et1 ; vel1 ; seq 0 ; eq0 :
These truth degree functions have been introduced in (5.55), (5.8), and (5.75)
for the rst three cases, and are otherwise characterised by the truth degree
tables:
seq 0 0 21 1
eq0 0 21 1
1 1 1
0 1 21 0
1
1 1 1
1
1 1 1
2
2 2
2
2 2 2
1
1 0 2 1
1 0 21 1
Also for this system K 3 the connectives :; ^; _ are just the same as in the
Lukasiewicz system L3 , and for the biimplication $K one obviously has
H1 $K H2 semantically equivalent (H1 !K H2 ) ^ (H2 !K H1 ) :
This system K 3 furthermore has weak connectives ^+ ; _+ ; !+ which coincide with the equally denoted connectives of the Bocvar system B 3 , i.e.
which yield ws which have truth degree 21 i one of their constituents has
truth degree 12 . These weak connectives are denable from the connectives of
the list (15.7) because of (15.5), (15.2), and (15.3).
Because the implication !K of the Kleene system is L3 -denable, e.g.
by
H1 !K H2 =def (H1 !L H2 ) ^ (H1 _ :H1 _ H2 _ :H2 )
(15.8)
also the 3-valued Kleene system K 3 is a subsystem of the Lukasiewicz
system L3 . Again, however, the Lukasiewicz implication !L is not denable in the Kleene system and this system therefore is a proper subsystem
of L3 . This undenability follows from the fact that each (binary) connective
M which is denable from the basic connectives in the list1 (15.7)
has a corresponding truth degree function verM with property verM ( 2 ; 12 ) = 21 and hence
cannot be the truth degree function seq 2 .
If one combines the basic connectives (15.1) of the Bocvar system with
the basic connectives (15.7) of the Kleene system, then !L becomes den
able as mentioned by Sestakov
[498]. A possible denition is
H1 !L H2 =def (H1 !K H2 ) _ (J (H1 ) ^ J (H2 ))
0
1
2
1
2
with
389
J (H ) =def :J0 (H ) ^ :J1 (H ) :

Therefore neither are all the connectives of the Kleene system denable in
the Bocvar system, nor conversely are all the connectives of the Bocvar
system denable in the Kleene system.
Besides the 3-valued approach toward partiality, oered by the (strong)
connectives of K 3 , also a kind of \doubled" two-valued approach is possible
which refers to a relation j= of satisfaction, and at the same time to another
relation =j of dissatisfaction, cf. [32]. Such a pair of a satisfaction and a
dissatisfaction relation is called biconsequence relation in [64]. Based on a
pair M = hV1 ; V2 i of sets of propositional variables3 these relations can be
dened for any propositional variable p by
M j= p =def p 2 V1 ; M =j p =def p 2 V2 ;
(15.9)
and inductively generalized for a language with the connectives :; ^ by the
conditions
M j= :A =def M =j A ;
M =j :A =def M j= A ;
M j= A ^ B =def M j= A ^ M j= B ;
M =j A ^ B =def M =j A _ M =j B :
From the additional assumption V1 \ V2 = ; one then immediately gets
for :; ^ the behavior of the strong connectives of K 3 . Hence K 3 can also
be understood as a kind of \unication" of two parallel (pseudo-)classical
approaches.4
Systems of many-valued propositional logic with an intended reading of
the third truth degree as \meaningless", or \undened", or something like,
as in the Bocvar and Kleene systems, and with basic connectives which
essentially can be dened in L3 have also been investigated e.g. in [7, 140,
253, 420, 421, 422, 496]. In [73] a longer, but concise survey of such and
other 3-valued systems is given. Additionally the interested reader should
also consult [197].
As was done with L3 , also the Bocvar system B 3 can be generalized
to systems B m with any nite number m 3 of truth degrees, cf. e.g.
[68, 70, 228, 448].
It is interesting to notice that the intended reading of the third truth
degree as \undened" in the Kleene system, and the connection with the
3 For the comparison with the standard approach toward classical propositional
logic one should have in mind that each truth valuation : V0 ! f>; ?g can
uniquely be characterized by the set fp 2 V0 j (p) = >g of all propositional
variables which are mapped to > by .
4 It should additionally be mentioned that this approach via (15.9) can also be
taken as a starting point for a modal translation of this partial logic and hence
of K 3 , cf. [32].
1
2
390
similar approach via truth value gaps which we mentioned earlier, was a
substantial fact that this system was more recently considered in connection
with partial, i.e. sometimes undened, truth predicates e.g. in [234, 317, 352,
356].
Another approach toward 3-valued systems comes from the consideration
of vague predicates like \hot water", i.e. of predicates which in some cases
neither really apply nor really do not apply to some objects. This eect can
be modeled in dierent ways. A usual one is via fuzzy sets, and this approach
shall be discussed in Chapter 18. But also 3-valued systems provide a (rough)
possibility { with the third truth degree read as \neither completely applies
nor fully does not apply". From this point of view three-valued logic was
studied with a more philosophical attitude quite early in [60, 61] and more
recently e.g. in [311] and later on in [63, 107, 293, 294], and with a more
mathematical attitude e.g. in [183, 301, 302]. (The last two papers are, among
others, surveyed in [213].)
It should additionally be mentioned that the phenomenon of presuppositions was discussed within the realm of truth value gaps, as well as in the
realm of 3-valued { but also of 4-valued { systems, cf. Chapter 20. And also
systems of paraconsistent logic5 have been discussed which can be based on
nitely many truth degrees, cf. e.g. [142, 499, 500].
Both of these aspects, i.e. vague predicates as well as presuppositions,
are covered in the use of three-valued logic for the analysis of natural language given by Blau [63]. His approach is essentially based on discussions in
the realm of the philosophy of language. From these discussions he gets the
fundamental motivations for the intuitive understanding of the three truth
degrees 0; 21 ; 1, which he denotes w; u; f , respectively. One of the core points
for him is that he identies the classical truth value > with the truth degree
1, and that he splits the classical truth value ? into the two degrees 0 and 12 .
He takes the truth degree 0 as a modied version of the (usual) truth value
?, and his intended reading for the truth degree u = 12 is \undetermined",
combined with the understanding that the appearance of this degree is caused
either by the use of vague predicates, or by the use of non-denoting names,
i.e. by reference to unsatised presuppositions.
This is the background for him to consider his three-valued approach
toward everyday language as an extension of the more traditional approaches
of (logically oriented) linguists using classical logic. We shall not enter into
the particularly linguistically oriented discussions of his approach, but only
look at the (formal) system of three-valued logic he uses in his approach.
5 By paraconsistent logics one means such systems of logic which \allow" contra-
dictions in the sense that a w of the form H ^ :H , or a set of ws of the form

fH; :H g does not entail { or make derivable { the whole set of ws. (Which
would be the situation in classical logic.)
391
On the propositional level6 his considerations are based on three primitive connectives, two kinds of negation connectives :; and a conjunction
connective ^, characterized by the truth degree functions
verBl
verBl = non1 j1 ;
verBl^ = et1 = min :
(15.10)
: = non1 ;
All these connectives satisfy the normal condition, hence this system is not
functionally complete. However, these connectives suce to introduce some
further, interesting connectives, e.g. the Go del negation with truth degree
function verBl = non0 as
H =def : :H ;
which additionally is the J-connective J0 , to introduce the standard maxdisjunction _ as
H1 _ H2 =def :(:H1 ^ :H2 ) ;
and to introduce an implication connective !Bl which has its truth degree
function verBl! characterized by the following truth degree table:
verBl! 0 12 1
0
1 1 1
1 1 1
1 0 12 1
This implication connective7 may be dened from the basic connectives by
H1 !Bl H2 =def :H1 _ H2 :
And also all the standard J-connectives become denable from these basic
connectives. The connective J1 was already mentioned, the other two can be
dened e.g. as:
J1 (H ) =def : H ;
J (H ) =def H ^ :H :
1
2
1
2
6 We shall disregard here the rst-order level of his considerations, despite the fact
that these form an integral, and essential part of his whole approach.
7 The main background idea behind the choice of this implication connective is
the authors claim that only this implication connective is suitable for a threevalued modeling of (two-valued) sentences of the form \All A are B " in natural
language, of course read as short form for: all objects which have property A
also have property B . And this claim is essentially based upon the idea that for
the truth of a sentence of this form \all objects which have property A also have
property B " it is completely out of any rational interest to allow the antecedent
\a particular object has property A" to be undetermined. Formally this means
that inside a 1universally quantied sentence the case that the antecedent has
truth degree 2 should not be a reason that the whole sentence may become not
true.
392
For a proof one simply has to determine the corresponding truth degree
tables.
The implication connective !Bl proves to be quite useful for a lot of discussions. We present only one result which shows some \relative" functional
completeness of the set :; ; ^ of connectives.
Proposition 15.1.1. Each three-valued connective which satises the normal condition can be dened from the basic connectives :; ; ^ of Blau's
system.
Proof: Let ' be any n-ary three-valued connective which satises the
normal condition, and let f = verBl' be its truth degree function. Then f can
be represented by its truth degree table looking like
x1 x2 : : : xn f (x1 ; : : : ; xn )
u11
u12 : : : u1n
v1
uk 1
uk2 : : : ukn
vk
..
.
..
.
..
.
..
.
uN 1 uN 2 : : : uNn
vN
with N = 3n . Using this truth degree table one is able to build a formula
D' in the language of the (propositional) Bl-system which denes ' and
contains just the propositional variables p1 ; : : : ; pn . To do this one should
have in mind that always in the case of vk = 12 for some 1 k N there
exists some jk = 1; : : : ; n with ukj = 12 . The formula D' may be a conjunction
D' G1 ^ ^ GN ;
which has as its constituents the formulas

Gk Juk (p1 ) ^ ^ Jukn (pn ) !Bl Hk
with
8
if vk = 1
< J0 (J (J1 (p1 ))) ;
if vk = 21
Hk : pjk ;
J1 (J (J1 (p1 ))) ;
if vk = 0 :
It is a routine matter to check that f is the truth degree function of D'. 2
Besides the formal value of this result, the particular applicational interest
inside the Blau approach toward natural language comes from the idea that
natural language uses only such connectives which satisfy the normal condition, because the appearance of the truth degree u = 12 for some sentence H
(in some particular situation) is simply an unintended mistake caused either
by the use of vague predicates, or by the use of non-denoting names.
1
1
2
1
2
15.2 Four-Valued Systems
393
It is an easy exercise to show additionally that the extension of the system :; ; ^ of basic connectives with the connective T3 as introduced for the
Lukasiewicz systems in (9.49) makes :; ; ^; T3 into a functionally complete system of connectives. (The previous proof can be suitably adapted in
changing the denition of the ws Hk .)

In contrast to the situation with three-valued systems, where one meets a
lot of approaches and interpretations in the literature, only a few approaches
concern four-valued systems and give particular interpretations to the four
truth degrees. One of the rare exceptions is Lukasiewicz [337] who, in his
later years, preferred a four-valued approach via his system L4 toward a modal
reading of the truth degrees over his original three-valued one via L3 in [334],
cf. Chapter 21.
However, instead of this reference to the system L4 with its linearly ordered truth degree set W4 , an approach toward four-valued systems has become prominent more recently which makes essential use of the natural partial
ordering of the truth degree set W4 = f0; 1g2 = f(0; 0); (0; 1); (1; 0); (1; 1)g,
and which connects a very natural interpretation with these degrees.8
This approach was inspired by theoretical work on systems of relevance
logic9 and later on also applied to considerations on how to treat { possibly
inconsistent { information in computers, e.g. in data bases or knowledge
bases, like in [45, 46, 133, 135], but also to discussions related to the liar
paradox, cf. [558].
In data and knowledge bases information is stored, e.g. in the form of
\facts", i.e. sentences which are marked as true or false. This information
may have been collected from dierent sources, and at dierent times. One
kind of use which can be made from such information is that one asks such
data or knowledge bases, or the computers they are stored in, to answer
8 Another interesting use of this four-valued truth degree set is connected with
discussions about the phenomenon of presuppositions, cf. Chapter 20.
9 Relevance logic, which like many-valued logic splits into a variety of dierent
systems, originates from the problem of the \paradoxes of material implication".

As such paradoxes one often counts the facts that in classical logic ws like
H1 ^ :H1 ) H2 and H1 ) H2 _ :H2 are tautologies, or that inference rules
like H1 ^ :H1 ` H2 and H1 ` H2 _ :H2 are sound, despite the fact that the
ws H1 ; H2 need not have to have any relationship to one another. The background of the feeling that something is paradoxical here comes from an intuition
which likes to have a closer connection between the (material) implication )
and some vague notion of \implies" which presupposes that between sentences
A; B; : : : and other ones H; G; : : : which are implied by A; B; : : :, there should be
some substantial connection, e.g. concerning what they speak about, or { more
formally { concerning common subformulas. For more information on relevance
logic the interested reader may consult [6, 134].
394
some suitable questions, e.g. for (the conrmation or refutation of) simple
statements concerning facts. The crucial point is that this information usually
is incomplete { and often even inconsistent. Therefore one should allow a
computer to answer not only \true" or \false", but also \I don't know" {
and even \true and false". Of course, the answer \I don't know" indicates
incompleteness of the information stored in the data or knowledge base, and
the answer \true and false" indicates that this information is inconsistent {
in the simplest case because con icting facts have been stored in the form of
two (simple, i.e. not compound) sentences of the forms H and :H .
For the computer these answers are just \marks" which it has to connect
with sentences (which e.g. formulate the questions it is asked to answer).
And with these four \values" the computer should also be able to \reason"
internally, because one likes (within a bit more sophisticated applications)
that the computer is not only able to repeat something that he was told
before, he should also be able to connect dierent facts by a kind of (internal)
reasoning mechanism.
Of course, as this explanation shows, this is an epistemic understanding of
these four \values", and not an ontological one: the \real world" is, of course,
treated here as well covered by the basic ideas of two-valued logic. This
remark does not say, however, anything about the suitability of a four-valued
approach based on this (epistemic) understanding of the truth degrees.
It is interesting to notice that some (preliminary) form of an ontological
understanding of these four degrees appears in ancient Indian logic, e.g. in
the work of Sanjaya who worked prior to the sixth century B.C. There the
principle of bivalence was rejected in favour of an idea of so-called \four corners", which correspond just to an ontological reading of these four degrees,
cf. [281, 437].
With this (computer science related) interpretation in mind, we are going
to interpret the truth degrees of W4 = f(0; 0); (0; 1); (1; 0); (1; 1)g, if assigned
as values to a w H w.r.t. some valuation , as saying in case Val4 (H; ) =
(a; b) with their rst component a whether H is given (inside the data or
knowledge base) the truth degree >, and with their second component b
whether H is given the truth degree ?: so Val4 (H; ) = (0; 1) signalizes that
H is given the truth value ?, but not the truth value >, and Val4 (H; ) =
(0; 0) signalizes that H is given no truth value at all.
For simpler reading we write T = (1; 0), F = (0; 1), N = (0; 0), B = (1; 1),
and understand the truth degree N (for \none") as indicating \underdetermination" or a gap, i.e. the lack of information on the truth value of H , and the
truth degree B (for \both") as \overdetermination" or a glut, i.e. the presence of contradictory information on the truth value of H { of course w.r.t.
some partial and (possibly even) non-functional f>; ?g-valuation which is
represented by the W4 -valuation .
One has even more than only the truth degree set W4 in this case: one
has a natural (partial) ordering of these degrees, having in mind that they
395
evaluate the (computers) knowledge about the truth value of H . Then surely
the degree T = (1; 0) is ranked in top position, and the degree F = (0; 1) is
ranked in lowest position { because it is most preferable to have a (denitely)
true sentence H , and worse to have a (denitely) false one. The two other
degrees N = (0; 0), B = (1; 1) are ranked somehow \between" the degrees
T; F because they, in some suitable sense, if assigned to H leave open both
possibilities that H may \really" be true or be false. As a result, this provides
the truth degree set W4 with a lattice structure as indicated in Fig. 15.1.
T
= (0; 0)
;
;
@
@
;@
;
@
;
@
@
;
@
;
@;
s
F
Fig. 15.1.
= (1; 0)
@s B = (1; 1)
;
= (0; 1)
Four-valued truth degree lattice W4
In this (Hasse diagramm of the) truth degree lattice W4 the lattice ordering
5 goes \bottom-up", i.e. lattice elements which are on a lower level position
are smaller ones.
With the corresponding lattice operations u; t one has natural candidates
for truth degree functions for a conjunction and a disjunction connective f; g
of a (propositional) system D4 of four-valued logic which is to be based on
the intuitions discussed up to now.
One should test, however, whether these truth degree functions t well
into the intuitive picture. Fortunately they do. Consider e.g. two ws A; B of
a language for D4 and a valuation 1 with ValD (A; 1 ) = T and ValD (B; 1 ) =
B. This means that \classically" A is given the truth value > under 1, and
B is given the truth value > as well as the truth value ?. Therefore to the
classical conjunction of A and B both truth values >; ? should be assigned,
which means that one should have ValD (A f B; 1 ) = B in D4 . Having for
a second example ValD (A; 2 ) = T and ValD (B; 2 ) = N, this means that
there is no information available (for the computer) about the truth value
of B . But then it is reasonable to suppose that (for the computer) also no
396
information is available about the truth value10 of the classical conjunction

of A and B , because one is (in classical logic) not able to infer the truth value
of a conjunction if one only knows that one of the conjuncts is true. And this
means that one should have ValD (A f B; 2 ) = N in D4 . And looking nally
at the situation ValD (A; 3 ) = N and ValD (B; 3 ) = F, the computer should
argue: well, my information tells me that B is false, so I know { also without
information about the truth value of A { that the conjunction of A and B has
to be false. And this means that one should have now ValD (A f B; 3 ) = F.
All the other cases for the truth degree tables of f and g can be treated
in a similar manner.
This discussion provides also the basis for the introduction of a negation
connective + into D4 , with a similar relationship to classical negation, as
f and g have to classical conjunction and disjunction, respectively. This
essentially means that to a w + H the truth value ? (or: >) should be
assigned if to H the truth value > (or: ?) is assigned. So one immediately
has for + the truth degree table given in Fig. 15.2.
H
+H
Fig. 15.2.
Truth table characterization of the D4 -negation
Thus one has for the basic vocabulary of D4 the set of connectives
J D = ff; g; +g :
(15.11)
Straightforward calculations show that for these connectives their truth degree functions w.r.t. the truth degree set W D = W4 have the characterizing
equations
verDf ((x1 ; y1 ); (x2 ; y2 )) = (et(x1 ; x2 ); vel(y1 ; y2 )) ;
(15.12)
verDg ((x1 ; y1 ); (x2 ; y2 )) = (vel(x1 ; x2 ); et(y1 ; y2 )) ;
(15.13)
D
ver+ (x1 ; y1) = (y1 ; x1 ) ;
(15.14)
for all xi ; yi 2 f0; 1g, with et; vel here the truth value functions of classical conjunction and classical disjunction, respectively { but understood as
functions over the set f0; 1g.
From these representations one immediately sees that such a truth degree
function can have the value N only if one of the arguments is N. And similarly
10 Of course, the principle of bivalence tells that this conjunction has to be either
true or false. However, the computer is supposed to infer its answers only from
information explicitely provided, and to give to a question for the truth value of
some sentence H answers meaning: the information I have say that H should be
given this (and that) truth value.
397
such a truth degree function can have the value B only if one of the arguments
is B. On the other hand one has that the set fT; Fg is closed under these
truth degree functions. This means that all these connectives of D4 satisfy
the normal condition { and it means also that these connectives are not
Particularly one has that both three-element sets fT; F; Ng and fT; F; Bg
are closed under these truth degree functions. They are even isomorphic as
algebraic structures with verDf ; verDg ; verD+ as operations { and isomorphic to
the algebraic structure hW3 ; et1 ; vel1 ; non1 i.
Therefore one has that the two three-valued subsystems of D4 , which are
constituted by the restrictions of the truth degrees to fT; F; Ng or fT; F; Bg,
coincide as three-valued systems, and are subsystems of the three-valued
Lukasiewicz system L3 .
From Fig. 15.1 it is immediately clear that the truth degree lattice W4 has
the same structure as the power set lattice of all subsets of the set 2 = f>; ?g
under inclusion as lattice ordering. So one can even, with a small abuse
of language, identify the values T; F; N; B with the sets f>g; f?g; ;; f>; ?g
of classical truth values, respectively. And this, obviously, is a quite natural
reading of these four degrees, having in mind their epistemic understanding.
Therefore each W4 -valuation for this system D4 can also be considered
as a mapping : V0 ! IP (2). Then the truth degree ValD (H; ) of a w
H under a valuation is a subset of f>; ?g such that all the members of
ValD (H; ) are assigned to H under the classical (partial and non-functional)
valuation coded by .
The crucial point now is to dene a suitable entailment relation j=D which
ts well into the intuitive realm discussed previously.
Here one may start from the observation that the basic intention of this 4valued approach can be seen as a kind of \parallel" treatment of assignments
of the truth values >; ? to sentences. Therefore it is again natural to refer
to biconsequences in this case, as previously for the 3-valued case. Also now
the starting point is a pair M = hV1 ; V2 i of sets of propositional variables,
which determine via (15.9) the starting point for the inductive denition
of satisfaction and dissatisfaction of (suitable) ws under M . This idea is
the basis for [64], leading to a proof theoretic study of such biconsequence
relations.
Another very natural approach seems to be, to say again that a set of
ws entails a w H i each model of is also a model of H . However,
what shall we understand by a model of a set of ws? Well, as in Chapter
3, nothing but some valuation which gives to all the ws of a designated
truth degree.
So the problem arises what the designated truth degrees should be. Up
to now there is no agreement on this point. One of the possible approaches
is to take only T as a designated truth degree, i.e. to put DD = fTg.
398
In this case the resulting notion of model ts well into the background
intuition: for our computer a model of a set of formulas should be any (partial
and non-functional) two-valued valuation which makes all the formulas of
denitely true { with \denitely true" understood here as meaning true but
not false or valueless. And this represents the standard notion of a model.
Thus we get in this case according to (3.8):
j=D H , ModD ( ) ModD (H ) ;
(15.15)
with the notion of model class dened as in (3.4).
With applications to relevance logic in mind, Dunn [133] considers instead
both truth degrees T; B as designated. The intuition behind this choice is
that a w which has such a designated truth degree is considered as \at least
true".11 Problems of denability of truth degree functions, i.e. of connectives,
and of relations between truth degrees for this choice of designated truth
degrees have been discussed e.g. in [14, 433].
The models in the sense of [133] for this situation are the (= B)-models
in the general terminology of Chapter 3, with 5 for the lattice ordering of
the truth degree lattice W4 . Let us call them weak models for the moment,
and denote the class of all weak models of some set of ws by ModD0 ( ).
Then Dunn discusses the following notion j=0D of weak entailment :
j=0D H , ModD0 ( ) ModD0 (H ) ;
which is more general than the previous notion from (15.15), as the following
result shows.
Proposition 15.2.1. For each set of ws of D4 and each w H one has
j=0D H ) j=D H :
Proof:
Let 2 ModD (H ). Then one has 2 ModD0 (H ) and therefore
D
also Val (H; ) = B. However, in the case ValD (H; ) = B for a compound
formula H , one of the constituents, i.e. one of the immediate subformulas,
of H has to have truth degree B under the valuation . So it follows inductively that one has to have (p) = B for one of the propositional variables
which occur in H . And this, obviously, holds true also for the case that
H is an atomic w. Therefore ValD(H; ) = B forces 2= ModD ( ). Thus
2 ModD(H ) gives ValD (H; ) > B, i.e. ValD (H; ) = T.
2
Still another notion of entailment j=D is taken into account in [45, 46].
There Belnap considers however only entailment relationships of the form
H1 j=D H . This corresponds for the two former cases to a restriction to nite
sets of ws. This immediately follows from the fact that the characterization (15.12) of the truth degree function of the D4 -conjunction easily yields
for each nite set = fH1 ; : : : ; Hn g of ws and each D4 -valuation :
11 This type of four-valued semantics may be used to give adequate semantical
interpretations for dierent systems of relevance logic. We will not discuss details
here. The interested reader may e.g. consult [431, 450].
2 ModD (fH1 ; : : : ; Hn g) , 2 ModD (H1 f f Hn ) ;

2 ModD0 (fH1 ; : : : ; Hn g) , 2 ModD0 (H1 f f Hn ) :
399
The denition of j=D is the following:

H1 j=D H2 =def ValD (H1 ; ) 5 ValD(H2 ; ) for all valuations .
It is immediately clear that one has
H1 j=D H2 ) H1 j=0D H2
for all ws H1 ; H2 of D4 .
For this notion of entailment Belnap gives in [45, 46] a nite list of
principles claiming that it is a complete list to infer all the valid entailments.
We shall not give this list here, the interested reader may consult [45] or [46].
Instead we take into consideration another aspect which in a natural way
is connected with the epistemic understanding of the truth degree set W4 =
fT; F; N; Bg. The lattice ordering of W4 , symbolized in Fig. 15.1, was based
on the intuition that the \larger" elements are the better ones in the sense
that the best possible case is denite truth, and the worst possible case is
denite falsehood. Hence this lattice ordering 5 is a kind of truth ordering.
This truth ordering can be contrasted with a knowledge ordering which
counts \complete" lack of knowledge of the truth value of some sentence H as
the worst case, and which prefers \more complete" knowledge. In some sense,
this point of view transforms W4 into a set of knowledge degrees. From this
point of view, another lattice ordering v, and hence another lattice structure, becomes interesting which may be symbolized by the Hasse diagram
of Fig. 15.3.
B
= (1; 1)
s
;@
= (1; 0)
;
;
@
@
;
;
@
@
@
@
;
;
@s F = (0; 1)
;
@;
s
Fig. 15.3.
= (0; 0)
Four-valued knowledge degree lattice W4
400
Again here, the lattice ordering v goes \bottom-up". Of course, the diagram
of Fig. 15.3 is just the diagram of Fig. 15.1 \tourned around", which also
means that the knowledge ordering v goes \left-to-right" in the diagramm
of Fig. 15.1.
The crucial point with these two lattice structures, which are in a natural
way imposed upon W4 , is that they are not independent of one another.
From the structural point of view the main relationship is that the lattice
operations of the truth degree lattice should be non-decreasing w.r.t. the
\knowledge ordering" v. Elementary calculations show that this is the case.
From an algebraic point of view the truth degree set W4 with these two
lattice structures becomes a particular case of a bilattice. This is a type of
structure which more recently has been introduced in lattice theory, and
which seems to be of particular interest for applications in logic and computer
science, cf. e.g. [20, 163, 164, 191, 192].
16. Systems with Graded Identity
If a language LS contains functions symbols, and hence has much more exible
possibilities to denote objects than a language without function symbols has,
one also needs the identity sign. And, ideally, the identity symbol should
act as a logical tool like the connectives or the quantiers to distinguish the
fundamental meaning of this identity symbol from other binary predicate
symbols.
This distinction is important for classical logic because one intuitively
assumes that the identity symbol has to denote \true" identity and not only
some binary predicate with some suitable properties { like those ones of equivalence relations, which are some kinds of \generalized" notions of equality.
The intuitive understanding is that also in many-valued (rst-order) logic
the identity predicate should denote some (kind of) particularly important
logical notion of \graded identity". This intuition is, however, itself not completely precise, or there is at least no common agreement about what is \the"
intuition underlying the intended understanding of identity for many-valued
logic.
Formally, again as in classical rst-order logic, also in many-valued rstorder systems, identity as a logical predicate cannot suitably be dened from
the other purely logical notions but has to be added explicitly, usually with
axioms governing its intended interpretation.
16.1 Graded Identity Relations

The crucial point in the discussions about graded identity relations is the
problem whether the suitable understanding of identity should allow for truth
degrees in between the \classical" degrees 0; 1 to really appear as degrees of
identity statements, or whether such identity statements have to have their
truth degrees out of f0; 1g only.
Therefore, we shall consider both variants: an \absolute" point of view
allowing only two-valued identity relations in interpretations of many-valued
rst-order logic with identity, and the more \liberal" point of view compatible also with \truly" graded identity relations. In the latter case there is
some further kind of splitting: one version which simply allows for degrees
of identity in between 0 and 1, but forces \self-identity", i.e. formulas of the
402
form a = a, to have always the highest truth degree, and another version
which also allows \intermediate" degrees of self-identity, and interprets them
as degrees of existence.
This indeterminate situation concerning the intuitive understanding of
identity in the many-valued setting was the main reason that the notion of
identity was treated in rst-order many-valued logic rather scarcely and in
more detail only from about the mid 1970s.
For the simplicity of formulations we shall always use the symbol $ as the
identity symbol for the rst-order systems to be considered, and we introduce,
previous to any formal denition of identity, the following notions.
Denition 16.1.1. For any system S of many-valued rst-order logic, an
identity relation id : jAj ! W S is called crisp i one has rg(id) f0; 1g, and
it is called many-valued otherwise.
The problem of many-valued identity logic was rst considered explicitly
by H. Thiele [530] for the nitely many-valued Lukasiewicz systems Lm
with the following restrictive result.
Proposition 16.1.1. Let an adequate axiomatization AxL of some Lukasiewicz rst-order system Lm be given which satises the completeness theorem.
Assume that the language LL is in the usual way extended with an equality
sign $. If one then extends AxL with the additional axiom schemata1
8x(x $ x);
(16.1)
8x8y(x $ y !L (H !L H [x==t]))
(16.2)
for all ws H of the extended language, then the extended rst-order system
Lm has only models with crisp identity relations.
Proof: Let A be an Lm-interpretation and some A-assignment such
that one has
0 < ValLA (x $ y; [x=b; y=c]) < 1
for some b; c 2 jAj and dierent variables x; y. Then consider the ws
H0 (x)
n
Y
i=1
(x $ x);
H0 [x==y]
n
Y
i=1
(x $ y) :
For a suitably large n one immediately gets

;

ValLA x $ y !L (H0 !L H0 [x==y]); [x=b; y=c] < 1 ;
1 Here
H [x==t] is a slightly modied substitution notation saying that the term

t is substituted for the variable x at some places of its free occurrence, with
substitution understood in the liberalized sense, i.e. having renaming of bounded

variables if t is substituted for x in the scope of a quantication which binds
variables which occur free in the term t.
16.2 Identity: the Absolute Point of View
403
contradicting the fact that A should be a model of (16.2).

2
In some sense, therefore, the Leibniz principles of identity (16.1), (16.2)
trivialize the Lukasiewicz systems Lm with identity.
Without restriction to the Lukasiewicz systems Lm one is, however, able
to axiomatize a whole class of nitely many-valued logics with identity in the
style of the Rosser-Turquette axiomatization as explained in Chapter 6.
To do this for the absolute as well as for the liberal point of view let
us consider some rst-order system S of many-valued logic which has, like
the systems considered in Section 6.3, in its language LS besides the identity
symbol $ connectives Jt for each t 2 W S as well as connectives !; ; u; t
indicating, respectively, an implication, a negation, a conjunction, and a disjunction. These connectives shall satisfy the correspondingVstandard conditions. The language LS shall also have a universal quantier which satises
its standard condition. Let us suppose furthermore that for all the quantiers
of LS their truth degree behavior can be described by value conditions, and
that S has only nitely many truth degrees.
Then we can assume that the \identity free" part of S is adequately axiomatized by the corresponding logical calculus K m
RT with (AxRT 1); : : : ; (AxRT 11)
as its axiom schemata and with (MP) and (Gen) as its inference rules.
Additionally one knows (from the proofs of the Completeness Theorems
6.2.2 and 6.3.2) that the notion of derivability `S constituted for S by the
logical calculus K m
RT has the property
`S -consistent , S-satisable
(16.3)
for each set of ws. Here for a set of ws to be `S -consistent means that
there does not exist some w H such that `?S (H u H ) holds true. And
from Chapter 6.4 it is known that this derivability notion `S satises w.r.t.
the implication connective ! the deduction theorem (DED` ).

First let us consider the absolute point of view. For an axiomatic approach
we start from the assumption, also realized in (16.1), that the w x $ x
should have the best possible truth degree in each case. Thus we like to have
S-logically valid:
V
(AxId 1) : xJ1 (x $ x) .
The idea to discuss the absolute point of view immediately also yields a
second type of conditions that one likes to have satised:
V
(AxId 2) : x; y(J1 (x $ y) t J0 (x $ y)) .
Because, intuitively, the operations at hand in the system S should satisfy the
\usual" principle of substitutivity for identicals, we also accept as a further
condition the schema
404
x1 ; : : : ; xn y1 ; : : : ; yn x1 $ y1 u : : : u xn $ yn !
! F (x1 ; : : : ; xn ) $ F (y1 ; : : : ; yn ) ;
for each function symbol F of LS .
As a rst idea it appears that for each n-ary predicate symbol P one could
(AxId 3) :
similarly intend to have satised the condition

V
x1 ; : : : ; xn y1; : : : ; yn x1 $ y1 u : : : u xn $ yn u

u P (x1 ; : : : ; xn ) ! P (y1 ; : : : ; yn ) :
V
(16.4)
However, a second thought indicates that this may be a bit too weak a condition. The point is that in a situation where S has more than one designated truth degree the S-logical validity of (16.4) would be compatible with
the situation that one could nd for some unary predicate symbol P an Sinterpretation A and objects a1 ; a2 2 jAj such that all of
A j=S J1 (a1 $ a2 ); A j=S J1 (P (a1 ));
A j=S Js (P (a2 ))
would hold true for some designated truth degree s < 1. But such a situation
may contradict the idea that the \complete identity" of a1 and a2 , as coded
by A j=S J1 (a1 $ a2 ), always should come together with the fact that a1 ; a2
have all properties to the same degree. Therefore we strengthen (16.4) to the
schema

V
V
(AxId 4) : x1 ; : : : ; xn y1 ; : : : ; yn x1 $ y1 u : : : u xn $ yn u

u Js (P (x1 ; : : : ; xn )) ! Js (P (y1 ; : : : ; yn ))
for each predicate symbol P of LS and each truth degree s 2 W S .
By AxId the set of all formulas shall be denoted which fall under one of
these axiom schemata (AxId 1); : : : ; (AxId 4).
These ideas for the choice of suitable axioms governing the approach toward the absolute reading of the identity $ now have to be accompanied with
a semantical counterpart.
Denition 16.2.1. An S-interpretation A is $-absolute i it interprets the
identity symbol $ by a crisp identity relation id having for all b; c 2 jAj the
crucial property
n
idA (b; c) = 1; if b = c
(16.5)
0 otherwise .
It is immediately clear that each $-absolute S-interpretation is a model
of AxId and that, therefore, the set AxId of ws of S is also `S -consistent
according to (16.3).
Theorem 16.2.1. A graded set of ws of LS is satisable in some $absolute S-interpretation i the (graded) set [ AxId is S-satisable.
405
Proof: Let B be some $-absolute S-interpretation with (B; ) j=S

for some B-assignment . Then one also has (B; ) j=S [ AxId because
AxId is a set of sentences.
So suppose conversely that (A; ) j=S [ AxId for some S-interpretation
A and some A-assignment . Then A possibly is not an $-absolute S-interpretation. The goal therefore is now to build from A an $-absolute S-interpretation B in which becomes satisable.
Let idA be the interpretation of the identity symbol $ in A. Dene a
binary relation in jAj by
a b =def idA (a; b) = 1
(16.6)
for a; b 2 jAj. Then one immediately has
;

a b , ValSA x $ y; [x=a; y=b] = 1
for any A-assignment and dierent variables x; y. The relation is an
equivalence relation in jAj: it is re exive because A is a model of (AxId 1),
and it is symmetric and transitive because A is a model of (AxId 4). Consider
the class
B = f[b] j b 2 jAjg
(16.7)
of all -equivalence classes [b] for b 2 jAj. This set B shall become the
universe of discourse of the $-absolute S-interpretation B we are looking for.
What has to be explained is how B interprets the non-logical symbols of the
language of S. For individual constants a put
aB =def [aA ] :
(16.8)
For n-ary predicate symbols P , including the identity symbol $, and for all
objects b1 ; : : : ; bn 2 B dene
P B ([b1 ] ; : : : ; [bn ] ) =def P A (b1 ; : : : ; bn );
(16.9)
and for function symbols F of arity n correspondingly
F B ([b1 ] ; : : : ; [bn ] ) =def [F A (b1 ; : : : ; bn)] :
(16.10)
According to the fact that A is a model of (AxId 4) and of (AxId 3), both
denitions (16.9) and (16.10) are suitable, i.e. independent of the choice of
representatives of the equivalence classes involved.
From the fact that A is also a model of (AxId 2) one gets furthermore that
the many-valued predicate idB which interprets $ in B satises condition
(16.5). Hence B is an $-absolute S-interpretation.
What remains to be shown is that is B-satisable. Let for each Aassignment be the B-assignment dened for each variable x by
(x) =def [ (x)] :
Then one gets for each term t, by induction on the complexity of this term,
from the denitions (16.8) and (16.10) that
406
ValSB (t; ) = [ValSA (t; )]

for each A-assignment . Similarly one gets from denition (16.9) that for all
such assignments and all atomic formulas H one has
ValSB (t; ) = ValSA (t; ) :
(16.11)
Again by induction on the complexity of the w H this result (16.11) generalizes for all ws of LS .
2
Let j= H denote the fact that the w H of LS is valid in all $-absolute Sinterpretations. For the standard extension `?S of the K m
RT -derivability relation
`S one then immediately has the following:
Corollary 16.2.1 (Soundness Theorem). For each w H of LS one has
AxId `?S H ) j= H :
And also the corresponding completeness theorem holds true.
Theorem 16.2.2 (Completeness Theorem). For each w H of LS one
has
j= H ) AxId `?S H :
Proof: Suppose that j= H holds true, i.e. that the w H is valid in
all $-absolute S-interpretations. Then the set AxId [ f H g of ws is not
S-satisable because otherwise H would be satisable in some $-absolute
S-interpretation according to Theorem 16.1. Hence the set AxId [ f H g of
ws is `S -inconsistent by (16.3), i.e. there exists some w G of LS such that
AxId [ f H g `S (G u G) holds true. But then one would have
AxId `S H ! (G u G)
by the deduction theorem (DED` ), and therefore also AxId `S H because
the w ( H ! (G u G)) ! H is logically S-valid and hence `S -derivable
2
16.3 Identity: the Liberal Point of View

Many-valued rst-order identity logic which allows also many-valued identity
relations was rst considered only in the mid 1970s, was partly inspired by research on fuzzy set theory, and followed the lines of the Rosser/Turquette
approach, cf. [379]. Therefore also in the present case the same assumptions
concerning LS as in the previous section shall apply. In particular,
LS shall
F
have the connectives u; t as before. We let, additionally, denote the nite
iteration of t.
One of the fundamental ideas (almost always) connected with the notion
of identity is \self-identity" as expressed previously in (AxId 1) and in (16.1).
For the present setting this idea is still accepted and thus a rst axiom is
407
(AxId1) : xJ1 (x $ x) .
Of course, the previous schema (AxId 2) is unacceptable now. However, also
schema (AxId 3) becomes problematic in the liberalized setting, at least in
the presence of more than one designated truth degree. The crucial point in
such a situation is that one could nd for some unary function symbol F an
S-interpretation A and objects a1 ; a2 2 jAj such that
A j=S J1 (a1 $ a2 ); A j=S Js (F (a1 ) $ F (a2 ))
would hold true for some designated truth degree s < 1. But such a situation
may contradict the intuitive idea that a function symbol always should mean
a function { and that a function has to have a uniquely determined value
for each of its arguments. We shall not go further into philosophical details
of \suitable" ways to deal with function symbols in the liberalized setting,
instead for simplicity we suppose that LS does not have function symbols.
This means that there is no reason to adapt schema (AxId 3) for the present
situation.
It remains to discuss how to deal with the behavior of many-valued predicates under the exchange of their arguments with \nearly identicals", i.e. it
remains to discuss how to adapt schema (AxId 4) for the case of many-valued
identity relations. To simplify this discussion the reader should be reminded
that the usual \full" PL2 -version of the Leibniz principle (16.2) follows from
their particular version (16.4), and that the latter one already follows from
the even more specic version
;

(8 ) 8 xi 8 yi xi = yi ^ P (x1 : : : xi : : : xn ) ) P (x1 : : : yi : : : xn ) (16.12)
for all n-ary predicate symbols P and all i = 1; : : : ; n. (Here (8 ) indicates
the universal quantication of all free variables of the following w, and
P (x1 : : : yi : : : xn ) indicates that the i-th argument place of P is lled with
the variable yi .)
A straightforward \translation" of (16.12) into the language of S, in the
sense that one simply exchanges in (16.12) the connectives and quantiers
of PL2 by corresponding connectives and quantiers of S, is subject to the
same reservations which have been discussed immediately following formula
(16.4). Therefore it seems to be appropriate to take into consideration suitable
connectives Jt , as also done in schema (AxId 4). However, more generally as
in the situation of schema (AxId 4), a more preferable intuition tells that
in the case that two objects a; b are \nearly equal" they also should have
some property P to \nearly" the same degree. Formally this could mean to
consider instead of formula (16.12) the formula
V
^ ^
( )
xi
yi Js (xi $ yi ) u Jt P (x1 : : : xi : : : xn )
;

) J(s;t) P (x1 : : : yi : : : xn )
;
(16.13)
408
as basic, with 1 i n and s; t 2 W S and some binary truth degree

function involved here. An approach via (16.13) allows that the truth degrees t and (s; t) are dierent. It requires, however, that the truth degree
of P (x1 : : : yi : : : xn ), with respect to some xed assignment, is uniquely determined by the truth degrees of xi $ yi and of P (x1 : : : xi : : : xn ). Yet, for
s = 0 this nevertheless looks counterintuitive having in mind what is the case
in classical rst-order logic with identity, and seems to be too far reaching.
Instead, it would seem to be more preferable to have the truth degree of
P (x1 : : : yi : : : xn ) xed by the truth degrees of xi $ yi and P (x1 : : : xi : : : xn )
only \to some degree", i.e. determined to be inside some interval. What has
to be determined now are the borders of such an interval, guided by the ideas
that the truth degrees of P (x1 : : : xi : : : xn ) and P (x1 : : : yi : : : xn ) should not
be \too dierent", and that their dierence should also depend on the degree
of equality of xi ; yi , i.e. also on the degree of xi $ yi . To formalize this idea
in the language of S we use the nite iteration of the disjunction connective
t determined by
1
G
i=1
Hi =def H1 ;
kG
+1
i=1
Hi =def
k
G
i=1
Hi t Hk+1 ;
together with the notation i for the truth degrees of W S = Wm introduced

in (5.7). Therefore we take into account instead of (16.13) the formula
^ ^ ^ ;
( ) xi yi Js (xi $ yi ) u Jt (P (x1 : : : xi : : : xn ))
!G
(s;t)
r=(s;t)
Jr (P (x1 : : : yi : : : xn )) ;
(16.14)
with functions ; ! : Wm Wm ! f1; 2; : : :; mg as well as with 1 i n

and s; t 2 W S . The crucial point now is the choice of these functions ; !.
The natural starting point to read ValSA (x $ y; ) = 1 as saying that (x)
and (y) \completely coincide" immediately yields
(1; t) = !(1; t) = m ; t (m ; 1) ;
which means
(1; t) = !(1; t) and (1;t) = t :
Hence (16.13) and (16.14) coincide for s = 1. Additionally one also likes to
have
1 (s; t) !(s; t) m for all s; t 2 W S = Wm .
Furthermore it seems preferable that in the case s = 0 there are no restrictions
in (16.14) concerning the truth degree of P (x1 : : : yi : : : xn ), which means
(0; t) = 1 and !(0; t) = m
409
for each t 2 W S . And nally one should have that the function is nondecreasing in its rst argument, and that the function is non-increasing in
its rst argument:
s1 s2 ) (s1 ; t) (s2 ; t) ^ !(s1 ; t) !(s2 ; t) ;
indicating that a smaller \degree of coincidence" of (the values of) xi ; yi in
(16.14) allows for more possibilities of deviance regarding the truth degrees
of P (x1 : : : xi : : : xn ) and of P (x1 : : : yi : : : xn ).
To meet all these requirements, Morgan [379] chooses the particular
functions
(i ; j ) = maxf1; j ; i + 1g ;
!(i ; j ) = minfm; j + i ; 1g :
Straightforward transformations show that this means to choose
(s;t) = seq 2 (s; t); !(s;t) = et2 (s; t) :
Following this approach we accept as a further axiom schema the schema

;

V
(AxId2) : xi ; yi Js (xi $ yi ) u Jt P (x1 : : : xi : : : xn ) !

(s;t) J ;P (x : : : y : : : x )
! Fseq
r
1
i
n
r=et (s;t)
for each predicate symbol P of LS and all s; t 2 W S .
Essentially, (AxId2) gives for atomic ws information on their truth degree
behavior, i.e. the possible change of their truth degree w.r.t. some predetermined S-interpretation and some assignment, for the case that one exchanges
in one of the argument places the object to which reference is made at this
place. The exchange of such reference objects in dierent argument places
obviously can be understood as a suitable iteration of reference exchanges
in always one place { and thus can be managed using (AxId2). A general
formulation for this more general case, however, would yield rather complicated formulas. Even assuming that one has for S some suitable version of
the law of exportation/importation available, such a more general formulation would result in implications with antecedents which are conjunctions
comprising \many" conjuncts which additionally may to a larger extent be
the same ws { just depending on the fact that one is not allowed to assume
that the conjunction u used in (AxId2) is idempotent. Thus, particularly, we
give in general no condition which is similar to the condition (16.2) for the
Lukasiewicz systems.
An easy and immediate consequence of (AxId2) is a weak form of transitivity of $ itself:
1
z Js (x $ y) u Js (y $ z ) !
^ ^ ^ ;
x y
seqG
1 (s;t)
r=et2 (s;t)
Jr (x $ z ) :
410
However, (AxId2) does not yield a suitable version of symmetry for $. Therefore one assumes as a further axiom schema the schema
V V
(AxId3) : x y(Js (x $ y) ! Js (y $ x)) for all s 2 W S .
The set of all ws falling under one of these schemata (AxId1); : : : ; (AxId3)
shall be denoted by AxId. Obviously, AxId is a set of sentences.
For any one of the systems S of nitely many-valued rst-order logic which
met the assumptions made previously, the intended interpretations now shall
be such ones which are models of AxId. The semantical characterization of
these intended interpretations shall, however, be given independent of the set
AxId of axioms.
Denition 16.3.1. An S-interpretation A is $-normal i the function id :
jAj2 ! W S = Wm , which interprets in A the identity symbol $, satises for
all b1 ; b2 2 jAj the following three conditions:
(Nid 1) : id(b1 ; b2 ) = 1 ,
(Nid 2) : id(b1 ; b2 ) = id(b2 ; b1 ) ,
(Nid 3) : for the CA (P ; b1 ; b2 ) of all pairs of n-tuples, n the arity of P , which
coincide in all coordinates but one which is b1 in a and b2 in c

id(b1 ; b2 ) inf 1 ; jP A (a) ; P A (c)j P 2 P S and (a; c) 2 CA (P ; b1; b2 ) :
Theorem 16.3.1. An S-interpretation A is $-normal i A is a model of
AxId.
Proof: It is immediately clear that all the sentences falling under (AxId1)
are satisable, and hence valid, in A if condition (Nid 1) is satised for A. And
it is equally clear that all the sentences falling under (AxId3) are satisable,
and hence valid, in A if condition (Nid 2) is satised for A.
So suppose that condition (Nid 3) is satised for the S-interpretation A.
Consider the w
Js (xi $ yi ) u Jt (P (x1 : : : xi : : : xn )) ;
(16.15)
and suppose that for some s; t 2 W S this w shall have a designated truth
degree under the A-assignment . Let b1 = (xi ), b2 = (yi ), and furthermore
c1 = ((x1 ); : : : (xi ) : : : ; (xn )), and c2 = ((x1 ); : : : (yi ) : : : ; (xAn )). Then
one has (c1 ; c2) 2 CA (P ; b1 ; b2 ) as well as id(b1 ; b2) = s and P (c1 ) = t.
Hence one gets for t0 = P A (c2 ) from (Nid 3) immediately the inequality s
1 ; jt ; t0 j, i.e.
s ; 1 t0 ; t 1 ; s :
This chain of inequalities can equivalently be written as
s + t ; 1 t0 1 ; s + t
and is therefore, because of 0 t0 1, also equivalent with
et2 (s; t) t0 seq 2 (s; t) :
411
Thus we get that under the A-assignment also the w

seqG
1 (s;t)
r=et2 (s;t)
Jr P (x1 : : : yi : : : xn )
(16.16)
has a designated truth degree. Hence one has that each instance of (AxId2)
is a valid sequence in A.
Now assume conversely that each sentence of AxId is valid in the S-interpretation A. Then one immediately has that conditions (Nid 1) and (Nid 2) are
satised for A. Hence suppose that condition (Nid 3) does not hold true. Then
there exist objects b1 ; b2 2 jAj, some n-ary predicate symbol P of LS , and
some (c1 ; c2) 2 CA (P ; b1 ; b2) such that
id(b1 ; b2 ) > 1 ; jP A (c1 ) ; P A (c2 )j :
(16.17)
Let s = id(b1 ; b2 ) and t = P A (c1 ). Then there exist some index 1 i n,
variables x1 ; : : : ; xn ; yi , and an A-assignment such that b1 = (xi ), b2 =
(yi ), c1 = ( (x1 ); : : : (xi ) : : : ; (xn )), and c2 = ( (x1 ); : : : (yi ) : : : ; (xn )).
The w (16.15) has, according to the choice of s and t, under this assignment
a designated truth degree. But because of (16.15) one has s > 1 ; jt ; t0 j
for the truth degree t0 = P A (c2 ) and thus
t0 < s + t ; 1 or 1 ; s + t < t0 ;
or equivalently
et2 (s; t) 6 t0
or t0 6 seq 2 (s; t) ;
which means that the w (16.16) does not have a designated truth degree
under . Hence some instance of (AxId2), the implication with antecedent
(16.15) and succedent (16.16), has a non-designated truth degree contradicting the assumption that A should be a model of AxId. Hence also condition
(Nid 3) has to be satised.
2
Corollary 16.3.1. AxId is S-satisable and hence also `S-consistent.

Proof: Because of (16.3) it suces to prove that AxId is S-satisable,
i.e. it suces to nd an S-interpretation B which is a model of AxId. Choose
some object b, and let jBj = fbg as well as P B (b; : : : ; b) = 1 for each predicate
symbol P of S, particularly also id(b; b) = 1. It is an easy exercise to see that
B is a model of AxId.
2
Let j=n H denote the fact that the w H of LS is valid in all $-normal Sinterpretations. For the standard extension `?S of the above mentioned derivability relation `S one then immediately has the following:
Corollary 16.3.2 (Soundness Theorem). For each w H of LS one has
AxId `?S H ) j=n H :
412
Proof: It follows from Theorem 6.4.3 that AxId `?S H holds true i one
has AxId j=S H . Therefore the present claim is an immediate consequence of
Theorem 16.3.1.
2
Theorem 16.3.2 (Completeness Theorem). For each w H of LS one
has
j=n H ) AxId `?S H :

Proof: Suppose that j=n H holds true. Then, according to Theorem
16.3.1, the set AxId [ f H g of ws is not S-satisable. Hence, by (16.3),
this set AxId [ f H g of ws is `S -inconsistent. Now argue as in the proof of
Theorem 16.3.
2
Observe, by the way, that with the same ideas one can even prove that
for each (graded) set of ws of LS it holds true
AxId [ `?S H , j=n H ;
with j=n H meaning that H is valid in all those $-normal S-interpretations
which are models of .
Unfortunately, the axioms in AxId look rather complicated. The situation
changes for the Lukasiewicz systems Lm because one is able to formulate
the semantic conditions (Nid 1); : : : ; (Nid 3) much simpler as by the set AxId,
cf. [214]. Obviously the schemata
(IdL 1) : 8x(x $ x) ,
(IdL 2) : 8x; y(x $ y !L y $ x)
code exactly the conditions (Nid 1); (Nid 2). The crucial simplication concerns
the coding of the condition (Nid 3) by some suitable schema of ws. Instead
of (AxId2) already the schema
;

(IdL 3) : (8)8xi ; yi xi $ yi & P (x1 : : : xi : : : xn ) !L P (x1 : : : yi : : : xn )
for each n-ary predicate symbol P of LS and all 1 i n
does the job.
Consider for some L-interpretation A an A-assignment , and let
s = ValLA (xi $ yi ; ) ;
t = ValLA (P (x1 : : : xi : : : xn ); ) ;
t0 = ValLA (P (x1 : : : yi : : : xn ); ) :
Then one gets et2 (s; t) t0 from the validity of (IdL 3) in A, because the
truth degree 1 is the only designated one. If A is also a model of (IdL 3), i.e. if
condition (Nid 2) holds true for A, then one can exchange in (IdL 3) the rôles
of xi ; yi and gets et2 (s; t0 ) t. And this last inequality is equivalent with
t0 seq 2 (s; t), because et2 ; seq 2 form an adjoint pair. Therefore from the
validity of (IdL 3) in models of (IdL 2) one gets the validity of (AxId2).
16.4 Identity and Extent of Existence
413
It is also easy to show that each $-normal L-interpretation is a model

of the set IdL of all ws falling under one of the schemata (IdL 1); : : : ; (IdL 3).
Hence in summary one has the following:
Proposition 16.3.1. An Lm-interpretation A is $-normal i it is a model
of the set IdL of axioms determined by these schemata.
And the axiomatization of Lm according to the slightly modied RosserTurquette approach suitable for Lm again gives that IdL provides an adequate axiomatization for the liberal nitely many-valued rst-order Lukasiewicz identity logics, cf. [214, 215].
In the particular case of pure identity logic, i.e. in the case that the rstorder language LL has $ as its only predicate symbol, condition (IdL 3) collapses to the requirement of &-transitivity:
8x; y; z (x $ y & y $ z !L x $ z ) :
And this situation, obviously, is easily generalized to the case of an arbitrary
(left continuous) t-norm t. In such a case, therefore, re exivity, symmetry, and
t-transitivity form a suitable set of conditions2 characterizing a generalized,
many-valued identity relation. In other words, in this case the fuzzy equivalence relations (over a given universe of discourse) as discussed in Chapter
18 are just the suitable generalized identity relations.

Still another, slightly more general approach toward many-valued identities,
in the liberal understanding, is oered in the context of monoidal rst-order
logic by Ho hle [269, 273, 274, 275, 276], but is also { at least partly { already
present in [504]. The crucial point now is that even the assumption that for
liberal identities $ each formula t = t has to have the truth degree 1, or at
least a designated3 one, is abandoned.
Then the problem arises, how to intuitively interpret a situation that in
an interpretation A for some individual a the formula a $ a has a truth
degree dierent from the universal upper bound 1. A solution is to interpret
each such truth degree of a $ a as extent of existence 4 of the object a. And
2 This situation has, however, not yet been generally axiomatized.
3 Actually, both these possibilities coincide because only the case that the universal
bound of the monoidal truth structure is the single designated truth degree has
been discussed.
4 This is an adaptation and generalization of a corresponding approach of D.
Scott [495] toward intuitionistic logic, which is also the background for [504].
Accordingly the following axiom schemata (AxIE1) to (AxIE7) reduce to his intuitionistic axioms of identity and existence. Intuitively, this extent of existence
may be understood as indicating a kind of partial existence, e.g. a localized or
locally restricted (in a suitable geometrical sense) one.
414
this interpretation becomes even more convincing if it is accompanied by a

reading of the truth degree of a $ b as a degree to which these objects a; b
overlap or match.
The intended monoidal $-interpretations (with truth degrees in a residuated lattice L = hL; \; [; ; ; 0; 1i) are all the L-valued \monoidal" interpretations A enriched with an L-valued equality E : jAj2 ! L characterized
by the properties:
(E1): E (x; y) E (x; x) ^ E (y; y) ,
strictness
(E2): E (x; y) = E (y; x) ,
symmetry
(E3): E (x; y) (E (y; y) ! E (y; z )) E (x; z ) .
transitivity
Such an L-valued equality E is called global i one has E (x; x) = 1 for all
x 2 jAj. In this case the transitivity condition (E3) reduces to
(Egl 3): E (x; y) E (y; z ) ! E (x; z ) ;
because one has 1 x = x in each residuated lattice.5
For the formal system of monoidal rst-order identity logic ML= it is
suitable to assume that the language LML besides the binary identity symbol
$ has a unary symbol e to denote the existence predicate. Given any monoidal
$-interpretation A with L-valued equality E , the standard understanding is
that the function : jAj ! L with always (a) = E (a; a) is the meaning of
the existence symbol e in the interpretation A.
A suitable set AxIEML of axioms for monoidal rst-order identity logic is
then determined by the following axiom schemata:
(AxIE1) : x $ y ! (e(x) ^ e(y)) ,
(AxIE2) : e(x) ! x $ x ,
(AxIE3) : x $ y ! y $ x , V
(AxIE4) : P (x1 ; : : : ; xn ) ! ni=1 e(xi ) for each n-ary predicate symbol P ,
(AxIE5) : (e(yi ) ! yi $ zi ) ! (P (x1 : : : yi : : : xn ) ! P (x1 : : : zi : : : xn ))
for each n-ary predicate
symbol P ,
V
(AxIE6) : e(F (x1 ; : : : ; xn )) ! ni=1 e(xi ) for each n-ary function symbol F ,
(AxIE7) : yi $ zi ! F (x1 : : : yi : : : xn ) $ F (x1 : : : zi : : : xn )
for each n-ary function symbol F .
Suitability of this set of axioms means that one has a corresponding completeness theorem, cf. [273].
Theorem 16.4.1 (Completeness Theorem). The axioms AxIEML provide,
together with the other axioms of rst-order monoidal logic, an adequate axiomatization of rst-order monoidal logic with identity determined semantically by the class of all monoidal $-interpretations.
5 And this means that a global L-valued equality is the same as a fuzzy equivalence
relation in the sense of Denition 18.5.2, with the membership degrees taken from
the residuated lattice L.
415
A kind of \standard" intended application of this monoidal identity logic

are the models of the pure monoidal identity logic, i.e. of that system of rstorder monoidal identity logic which has no \nonlogical" predicate symbols6
in its language. These interpretations A are formed just of some set A, the
universe of discourse of the interpretation, together with an L-valued equality
in A. In a suitable sense these \simple" interpretations are just ordered pairs
(A; E ) of a set a and a L-valued equality E in A. They can be considered as a
natural generalization of fuzzy sets. Accordingly they are treated as L-valued
sets in [272, 276].
These L-valued sets form a category w.r.t. structure preserving maps as
morphisms. Structure preservation for a map f between two L-valued sets
(A1 ; E1 ) and (A2 ; E2 ) means in this case that f is a mapping from A1 into
A2 which satises for all x; y 2 A the conditions
E1 (x; x) = E2 (f (x); f (x)) ;
E1 (x; y) E2 (f (x); f (y)) :
This category seems to be a bit too general to be considered as a category
of generalized sets, cf. [272]. It seems to be preferable to restrict the objects
to the separated L-valued sets (A; E ), which are L-valued sets which satisfy
additionally the condition
if E (x; x) _ E (y; y) E (x; y) then x = y :
The category of separated L-valued sets, with structure preserving maps as
morphisms, has suitable properties which one likes to have satised by a
category of generalized sets.
More important, however, from an applicational point of view is that
the separated L-valued sets form a suitable mathematical background for
a theory of fuzzy control7 e.g. in the way that they allow to understand
the input fuzzy sets of fuzzy control rules as \fuzzy singletons" w.r.t. some
suitable [0; 1]-valued (separated) identity, cf. [273, 274].
6 This means, that this \pure" monoidal identity logic has only the identity symbol
as a predicate symbol.
7 Fuzzy control is a methodology of mathematical modelling which describes con-
trol algorithms with the help of control rules which use essentially vague notions in formulating the conditions under which some particular (sometimes also
vaguely described) control action has to be taken. Formally the vague notions
are represented with reference to fuzzy sets.
416
Part IV
Applications of Many-Valued Logic
417
17. The Problem of Applications
For quite a long time discussions about the use and the applicability of manyvalued logic mainly centered around two aspects: on the one hand on the
problem of the \true" meaning of the truth degrees, and on the other hand on
the problem whether systems of many-valued logic { or rather: which systems
{ would be a suitable substitute for classical logic. From the actual point of
view, the second problem in any case is a quite misleading one. It is neither
the intention of many-valued logic, nor of any other eld of non-classical
or \philosophical" logic, to compete with classical logic within the realm of
every-day applications for being \the" superior kind of logic { instead1 each
one of these logics which usually extend classical logic in dierent ways claims
to be applicable in particular situations. And this is the case for many-valued
logic too. The crucial point thus is which are suitable elds of application.
And this essentially depends on the intuitive understanding of the additional
expressive possibilites which such an extension has relative to classical logic.
For many-valued logic these are the (additional) truth degrees and the
much larger class of connectives and quantiers which become available with
the \larger" sets of truth degrees. However, a widespread feeling for a long
time was, that these additional tools of many-valued logic are by far more
articial tools than are the additional expressive means e.g. of modal or of
deontic logic (which both extend into the eld of non-truth-functional logic).
The criticism of articiality centered largely around the idea that the sets of
truth degrees are extensions of the set f>; ?g of classical truth-values, and
that one therefore would have to interpret the \additional" truth degrees as
truth-values in between > = 1 and ? = 0. Such interpretations, however,
have for long essentially been lacking insofar as there did not exist philosophically satisfactory, and ontologically oriented, interpretations which link
truth degree sets with more than three truth degrees with the intuitive under1 That intuitionistic logic, particularly in the understanding of L.E.J. Brouwer,
who thought of logic as an integral part of mathematics and started his development of intuitionistic mathematics as an approach toward \true" mathematics
which had to be dierent from classical mathematics, was from its historical origin part of an approach that competed with classical logic does not matter here
because intuitionistic logic in its actually prevailing understanding ts into the
scene of non-classical logics which was just described.
420
standing that there may exist some \grading" of truth in between complete
truth and complete falsity.
Accordingly, the problem of suitable applications was one of the important
problems mentioned in the rst basic monograph [465] as an open one.
A partial exception was constituted by three-valued systems, cf. Chapter
15, because in such systems one has dierent possibilities to give an intuitively
very acceptable reading to the third truth degree which is in accordance with
the understanding that > = 1 and ? = 0. Such interpretations of the third
truth degree e.g. as \senseless" or as \undened", however, supported also
a point of view which understood such three-valued systems as two-valued
systems with truth value gaps \in disguise". This did not really support the
acceptance of many-valued logic.
The situation has changed in the last decades with a growing number
of application ideas, and also with the growing awareness of the situation
in product systems, cf. Section 8.3. What is more widely understood in the
meantime is the fact that logical systems have, besides their often intended
ontological background ideas, also a mainly formal, i.e. technical, side which
gives them the character of a tool, and makes them, as abstract formalisms
like abstract algebraic structures, open for widely diering interpretations {
and this fact constitutes a truly welcome situation. Nevertheless, also today it
is important for any intended application to start with a thorough discussion
on the role or the meaning of the truth degrees.
In this part of the book we are going to consider some rather diverse elds
of applications for the ideas of many-valued logic { applications which partly
are quite natural and successful ones, but which sometimes also proved to miss
the point. We mention even some \applications" of the last type, mainly for
historical reasons, but also to indicate the sometimes quite delicate borderline
which separates problems which can successfully be attacked with the help of
many-valued logic and such ones which cannot. This diversity of application
elds underscores the fact that many-valued logic is more a technical tool to
attack problems in a suitable way, than an ontologically ambitious genuinely
philosophical topic. This, however, is neither a aw for many-valued logic,
nor a particularly exceptional situation { the situation with the \possible
worlds" of the Kripke semantics for modal logic is essentially comparable:
also there a solid ontological basis is lacking and they hence have more the
avor of a technical tool quite useful for suitable purposes.
The applications we are going to discuss have either historical or systematic importance. Completeness is not intended here, neither in the diversity of
dierent applications, nor in the particular cases which shall be considered.
Nevertheless we intend to give an impression of how widely scattered the
topics are which can be related to ideas from many-valued logic and benet
from them. The theoretical depth of the ideas and results of many-valued
logic which become relevant for these applications dier to a large extent
from application to application. However, the level of sophistication of the
421
formal apparatus of many-valued logic to be applied is only loosely linked to

the importance which the reference to the world of ideas of many-valued logic
may have for some particular application. Sometimes already the reference
to some elementary basic ideas and facts may provide an important impetus.
Applications of such a \more elementary nature" often provoke remarks
with the tenor that many-valued logic is super uous and could completely be
substituted by classical logic. But this type of remark is nearly as meaningful
as a remark that modern mathematics in principle can do without category
theory, or that modern life can proceed without the use of cars { because one
has legs.
As usual in serious science one can additionally expect that from dierent
types of applications which prot from many-valued logic also conversely one
may have positive in uence inside many-valued logic. This has been the case
in the starting phase of many-valued logic, e.g. in the case of Lukasiewicz
by the intended applications to modal logic, cf. Chapter 21, and is actually
to be recognized e.g. in the relationships of many-valued logic to the theory
of fuzzy sets and to fuzzy logic, cf. Chapters 18 and 19, as well as to computer science.2 The prot many-valued logic earned from the applications
to fuzzy sets consisted e.g. in the awareness of the basic role which t-norms
and t-norm-based connectives have for many-valued logic, cf. e.g. Chapters
5 and 13. And computer science applications e.g. in the eld of automated
theorem proving, had a profound in uence on the (quite recent) development
of approaches toward proof-theory for many-valued logic, cf. Chapter 7.
2 Important topics for the interplay of many-valued logic and computer science
are e.g. many-valued switching theory and the eld of program verication. Both
topics shall not be treated in detail in this book. Particularly for the eld of manyvalued switching and logic design the interested reader is referred to [143, 391].
422
18. Fuzzy Sets, Vague Notions,

and Many-Valued Logic
We recognized already in Section 2.4 that the classes of classical, \two-valued"

set theory form the suitable tool to represent the extensions of predicates of
classical logic. For predicates of many-valued logic the situation was more
dicult. But one of the possibilities to represent the extensions of predicates
of many-valued logic was, as mentioned in Section 2.4, to use the fuzzy sets
as the suitable tool for this purpose. In general, fuzzy sets are also suited
to form the extensions of \vague" predicates which are quite common in
everyday language, and have formed a challenging philosophical problem for
a long time.
18.1 Vagueness of Notions and Fuzzy Sets

It is a well-known property of colloquial language that it allows one to use
notions which do not have in a \natural" way, i.e. without any arbitrariness,
the property that they either apply or do not apply to each one of the objects
they are intended to be applied to. Instead one has the feeling that sometimes
such notions do not have a crisp \borderline" which divides the area where
the notion applies from the area where it does not apply { but instead have
a kind of \transition area" from application to non-application. Such notions
usually are called vague. Simple examples are e.g.:
{ that some car needs little petrol;
{ that two cars on a lane have sucient distance;
{ that some PC has insucient storage capacity;
{ that some number is much larger than 10;
{ that some radiation level is unhealthy;
{ that some house is cheap;
{ that some road is slippery.
It is an easy task to extend this list.
The problem of vague notions has been discussed from the philosophical
point of view occasionally, e.g. in [60, 61, 258, 471], and more recently in
[132, 454]. Traditional, classical logic and mathematics was, however, unable
to cope with such vague notions in a satisfactory manner. Instead the usual
way of working in these elds, and in applying techniques from these elds,
424
18. Fuzzy Sets, Vague Notions, and Many-Valued Logic
has been to convert such vague notions into traditional ones with \clear", i.e.
sharp boundaries { or into crisp notions as one usually says.
The philosophical discussions, however, did not essentially in uence the
formal treatment of vague notions. To a large extent this seems to be caused
by the fact that in the philosophical discussions the vagueness of notions has
essentially been treated as a kind of superciality { and that in parallel the
(mathematically oriented) approach of classical logic has been to consider
crisp notions as the \standard" case.
The situation only changed in the moment where applicational intentions
made vague notions an interesting topic. The pioneering approach came from
L.A. Zadeh who introduced in [582] the notion of fuzzy set and explained in a
series of in uential papers fundamental ideas for their use in systems theory
and knowledge engineering, cf. e.g. [583, 585, 587] or [590, 591]. However,
fuzzy sets are nothing but the extents of vague notions.
A fuzzy set A is characterized by a generalized characteristic function
A : X ! [0; 1], called membership function of A and dened over a universe
of discourse X . This universe of discourse has to be chosen according to the
specic situation in each case. If one intends to emphasize the universe of
discourse X of a fuzzy set A then one also speaks of A as a fuzzy set over
X or a fuzzy subset of X . In most cases, yet, the universe of discourse is
determined by the context and not mentioned separately. The calligraphic
types X ; Y ; : : : shall in this chapter denote universes of discourse for fuzzy
sets.
Obviously, for each usual, i.e. crisp set M its usual characteristic function
M = M is such a membership function. Therefore we consider crisp sets
as special cases of fuzzy sets, viz. those ones with only 0 and 1 as membership degrees. Fuzzy sets A; B are equal i they have the same membership
functions:
A = B , A (x) = B (x) for all x 2 X :
(18.1)
If one intends to describe some xed fuzzy set A over some universe of discourse X then one denes its membership function A either by giving some
formula to describe A , or by a table of the values, or by a picture of the
graph of A .
The essential idea behind this approach is to have with the membership
degree A (a) for each point a 2 X a graduation of its membership with
respect to the fuzzy set A. And this membership degree obviously is just the
degree to which the sentence \a is a member of A" holds true. Hence it is
natural to interpret the membership degrees of fuzzy sets as truth degrees of
the membership predicate in some (suitable system of) many-valued logic S.
To do this in a reasonable way one has to accept some minimal conditions
concerning the language LS .
Disregarding { for simplicity (as is done in the overwhelming majority of
fuzzy sets applications) { fuzzy sets of higher level, i.e. fuzzy subsets of universes of discourse which themselves have fuzzy sets as their members, as well
18.2 Basic Theory of Fuzzy Sets
425
as fuzzy sets of type 2, i.e. fuzzy sets which have as their membership degrees
fuzzy subsets of the unit interval [0; 1], one has, from the set-theoretical point
of view of the usual a cumulative hierarchy of sets, fuzzy sets as (generalized)
sets of rst level over a given set of urelements, the universe of discourse.
Therefore the intended language needs besides a (generalized, i.e. graded)
binary membership predicate " e.g. two types of variables: (i) lower case
Latin letters a; b; c; : : : ; x; y; z for urelements, i.e. for points of the universe of
discourse X , and (ii) upper case Latin letters A; B; C; : : : for fuzzy subsets of
X . And of course it needs some set of connectives and some quantiers { and
thus a suitable notion of well-formed formula.
Having in mind the standard fuzzy sets with membership degrees in the
real unit interval [0; 1] thus forces one to assume that S is an innitely manyvalued logic. And the usual intuitive understanding of the membership degrees furthermore supports the assumption that DS = f1g is the set of designated truth degrees. Following actual tendencies in fuzzy sets theory we
furthermore suppose that the system S is (quite similar to) some t-norm
based system in the (naive) sense of Chapter 13.
From the point of view of many-valued logic, however, nothing forces one
to consider the real unit interval [0; 1] as the only possible set of membership
degrees. Each other truth degree set can be used instead.
It is not necessary to x all the details of the language LS in advance. We
suppose, for simplicity of notation, that from the context it shall always be
clear which objects the individual symbols are to denote.1 Denoting the truth
degree of a well-formed formula H by [ H ] , to identify membership degrees
with suitable truth degrees then means to put
A (x) = [ x " A ] :
(18.2)
This type of interpretation proves quite useful: it opens the doors to clarify
far-reaching analogies between notions and results related to fuzzy sets and
those ones related to usual sets, as shall be shown in this chapter.

Based on the main idea consider the membership degrees of fuzzy sets as truth
degrees of a suitable membership predicate, it e.g. becomes quite natural to
describe fuzzy sets by a (generalized) class term notation, adapting the corresponding notation fx j H (x)g from traditional set theory, and introducing
a corresponding notation for fuzzy sets.
Denition 18.2.1. For each universe of discourse X and each w H (x) of
our set theoretic language of many-valued logic we denote by fx 2 X k H (x)g,
1 Which, formally, means that we assume that a valuation always is determined
by the context and has not explicitly to be mentioned.
426
or simply by fx k H (x)g, that fuzzy subset A of X whose membership function

A is characterized by A (a) = [ H (a)]] for each a 2 X ; i.e. fx 2 X k H (x)g
has the characteristic property

[ a " fx 2 X k H (x)g] = [ H (x=a)]] for all a 2 X :
Furthermore, we sometimes have to use a simple many-valued translation
of usual identity, which is dened as
n
[ a =: b] = 1; if a = b
(18.3)
0 otherwise
and will also be allowed to be used in the case that instead of a; b some terms
(of a well-dened type) appear.
18.2.1 Elementary set algebraic operations

Denition 18.2.2. For every t-norm t, t-conorm st , negation function n,
for their corresponding connectives ^t ; _t , and n , and for any fuzzy sets
A; B let their (t-norm-based) intersection and union, as well as their complement be:
A \t B =def fx k x " A ^t x " B g ;

A [t B =def fx k x " A _t x " B g ;
CCn A =def fx k n (x " A)g :
For the particular negation functions nont , t any left continuous t-norm, and
non1 a separate, simpler notation shall be used:
C t A =def fx k ;t (x " A)g and CCA =def fx k :(x " A)g :
Thus, the generalized abstraction terms fx k H (x)g allow for denitional

formulas which on the formal level are very similar to the formulas which in
classical set theory usually describe the crisp notions corresponding to the
fuzzy ones introduced here.
To simplify notation occasionally we follow the common usage of writing
in the case of the Go delian t-norm t = tG = et1 = min
\ for \et
and
[ for [et :
(18.4)
Also, as usual, by IF (X ) the class
IF (X ) =def X [0; 1]
of all fuzzy subsets of X will be denoted. Here, IF (X ) is the set power (or the
direct power in algebraic terms) of the set of generalized membership (i.e.:
truth) degrees with respect to the universe of discourse as exponent. It is well
known that this point of view can be extended to include the consideration
1
427
of set algebraic operations for fuzzy sets too.2 That means, having given any
families (ti )i2I of t-norms, (stj )j2J of t-conorms, and (nk )k2K of negation
functions, the algebraic structure
F(X ) =
IF (X ); ([ti )i2I ; (\tj )j2J ; (C nk )k2K
(18.5)
X
of fuzzy subsets of X is the direct power D of the structure
D =def
[0; 1]; (ti)i2I ; (stj )j2J ; (nk )k2K
of generalized membership degrees.3
If one works within the structures D and F(X ) one has to use terms
referring to the elements of those structures. In our case those terms for the
structure D are exactly our expressions of the language of many-valued logic
which are built up from variables for truth degrees { or expressions \x " A"
with x a variable for elements of X and (the letter) A a symbol for elements
of IF (X ) { with the help of the connectives (i.e. operation symbols for truth
functions) ^ti ; _tj ; nk ; and the terms for the structure F(X ) are exactly our
expressions of the extended set theoretic language which are built up from
variables and constants for fuzzy sets with the help of the corresponding
operation symbols \ti ; [tj ; C nk introduced in Denition 18.2.2.
By a basic Horn formula we understand a nite disjunction (in the classical metalanguage!) of at most one term equation and negations of term
equations otherwise. Thus simple examples of basic Horn formulas with respect to D are, with p; q as (propositional) variables for generalized truth
degrees, i.e. for elements of [0,1]:
p ^ti q = q ^ti p;
p ^ti q 6= p ^ti (q_tj nk q) ;
and corresponding basic Horn formulas for F(X ) are e.g.
A \ti B = B \ti A ;
A \ti B 6= A \ti (B [tj C nk B ) :
Furthermore, by a Horn formula we mean such a formula which is built up
from basic Horn formulas (with respect to the same structure) using conjunction, existential and universal quantication (of classical metalanguage).
Referring to a well known theorem in classical model theory, cf. e.g. [96],
we are able to state many set algebraic laws for fuzzy sets at once.
2 It is also well known that one cannot only modify here the families of algebraic
operations one is considering in the set [0; 1] of membership degrees, but can
change from [0; 1]-based algebraic structures to more general ones, e.g. to (some
types of) lattices in general as in the L-fuzzy sets of [201], or also to enriched
ones as e.g. in [441].
3 It is, of course, possible to consider other basic structures here, e.g. residuated
lattices based on one given (left continuous) t-norm, which have been the basic
structures for monoidal as well as basic logic in Chapter 14. Additionally one can
enrich the language by reference to modal operators, e.g. having ws of the form
\2H " expressing (in the object language) the fact that j= H , as e.g. in [327].
428
Theorem 18.2.1 (Transfer Theorem). Suppose that H is any Horn formula (with respect to F(X )). Then the structure F(X ) is a model for H , i.e.
H holds true for fuzzy subsets of X , if the Horn formula H^ corresponding to
H but referring to the structure D (i.e. H^ is built up from H by changing \t
to ^t , [t to _t , C n to n , and by an appropriate change of variables too)
holds true in the structure D.

The proof is given in [96, pp. 326 .], uses only methods from classical model
theory, and shall therefore not be given here.
As an immediate consequence of this theorem we get the commutativity and associativity of all unions [t and all intersections \t for fuzzy sets
from the corresponding properties of t-conorms and t-norms, because these
properties can be simply formulated by Horn formulas as e.g.
^;

A \t B = B \t A ;
A;B
^ ;
A;B;C
A \t (B \t C ) = (A \t B ) \t C
for the intersections.

If we add the natural ordering relation to the structure D we get in
the direct power F(X ) an inclusion relation characterized by
A B i [ a " A] [ a " B ] for all a 2 X ;
(18.6)
which already was dened by [582] and which is a partial ordering in the class
IF (X ) of all fuzzy subsets of X .
It is interesting to note that our theorem can be extended to include these
ordering relations too. For this purpose we have to extend what is meant
by a basic Horn formula: this is now not only a disjunction of negated or
unnegated term equations, but also a disjunction of { negated or unnegated
{ term equations or term inequalities 10 20 and 1 2 for terms 10 ; 20
relating to D, and terms 1 ; 2 relating to the structure F(X ) of (18.5). Hence,
for example, the transitivity of the ordering relation can be described by
a Horn formula and therefore our theorem yields
AB ^ BC ) AC

for any fuzzy sets A; B; C .
18.2.2 Graded inclusion of fuzzy sets

Unfortunately, as we have to refer in Theorem 18.2.1 to the classical metalanguage, the binary relation for fuzzy sets is a two-valued relation, i.e. not
itself a graded one. Thus one would only have to compare the fuzzy sets with
their graded membership by a two-valued implication relation. Instead, from
the intuitive idea of an unsharp boundary for fuzzy sets it would be more
natural to have an \unsharp", i.e. graded inclusion relation too for the fuzzy
429
sets. Fortunately, by using our language of many-valued logic it is quite simple and natural to introduce such a graded, i.e. \fuzzied" or many-valued,
inclusion relation again in strong formal analogy with the usual denition of
the inclusion relation.
Denition 18.2.3. For any fuzzy sets A; B and any left continuous t-norm
t let
A jt B =def 8x(x " A !t x " B ) ;
A t B =def A jt B ^t B jt A :
The truth degree [ A jt B ] is a degree of containment of A in B and the
truth degree [ A t B ] is a degree of equality for the fuzzy sets A; B .
As usual we are going to formulate the results in the form of the logical validity of formulas. If it is equivalent to formulate such a result as an
inequality or as an equation for truth degrees we, as already previously in
Chapter 13, shall write ! and $ without index instead of !t and $t in
all those cases where t0 is any t-norm subject only to the restriction of being
left continuous. But we extend this notation and allow also for j and
with as an index { with the same understanding that this is shorthand for
jt and t with some (otherwise unspecied) left continuous t-norm t0.4
The second one of the graded relations dened in Denition 18.2.3 is a
kind of many-valued identity relation and will be considered later on. First
let us note that obviously for each left continuous t-norm t
A B , j= A jt B ;
(18.7)
and therefore also
A = B , j= A t B :
(18.8)
Hence jt is a suitable generalization of ; a fact which once again supports
the point of view that treating fuzziness by (the language of) many-valued
logic is a useful thing. And t is at least nicely related to the usual identity
relation for fuzzy sets and has typical properties of a graded identity relation,
as shall be shown in Proposition 18.2.3. Furthermore, for the many-valued
inclusion relations jt suitable many-valued versions of the fundamental properties of inclusion for crisp sets hold true too.
Proposition 18.2.1. For each left continuous t-norm t and all fuzzy sets
A; B; C one has
(i) j= A jt A ;
(ii) j= A jt B ! A \t C jt B \t C ;
0
4 The notational conventions need some care concerning the use of the symbol \":
without any index it means the syntactic equality for words of formal languages,
with some t-norm or with as an index it means the graded identity for fuzzy
sets.
430
(iii) j= A jt B ^t B jt C !t A jt C ;
(iv) j= A jt B ! C t B jt C t A ;
(v) j= A jt B ! A [ C jt B [ C :
Proof: Claim (i) is obvious. For (ii) we have to use the corresponding
Denitions 18.2.2 and 18.2.3, and the fact that
[ H1 !t H2 ] [ H1 ^t G !t H2 ^t G] ;
which was proved in Proposition 13.1.7(i).
(iii) By Denition 18.2.2 one has
[ A jt B ^t B jt C ]
= [ 8x(x " A !t x " B ) ^t 8x(x " B !t x " C )]]
[ 8x((x " A !t x " B ) ^t (x " B !t x " C ))]] ;
using the fact that according to Proposition 13.2.1(i)
j= 8xH1 ^t 8xH2 ! 8x(H1 ^t H2 )
holds true for all formulas H1 ; H2 and all left continuous t-norms. One furthermore has
j= (H1 !t H2 ) ^t (H2 !t H3 ) !t (H1 !t H3 )
for all formulas H1 ; H2 ; H3 ; cf. Proposition 13.1.8(ii). Thus one has
[ A jt B ^t B jt C ] [ A jt C ]
which immediately yields this claim.
Claim (iv) becomes, by the same kind of argument as in (ii) and (iii), a
straightforward consequence of
[ H1 !t H2 ] [ ;t H2 !t ;t H1 ] ;
which holds for left continuous t because of Proposition 13.1.12(iv).
And claim (v) immediately follows from Proposition 13.1.8(v) using simple properties of universal quantication.
2
Proposition 18.2.2. For each left continuous t-norm t and all fuzzy sets
A; B; C one has
(i) j= A jt B ^ C jt D ! A \ C jt B \ D ;
(ii) j= A jt B ^ C jt D ! A [ C jt B [ D :
Proof: (i) is an immediate consequence of Proposition 13.1.8 (iv), and

(ii) is one of Proposition 13.1.8(v).
2
431
Proposition 18.2.3. Let the t-norm t be left continuous. Then for all fuzzy
sets A; B; C one has

(i) j= A t A;
(ii) j= A t B ! B t A ;
(iii) j= A t B ^t B t C !t A t C ;
(iv) j= A t B ! A \t C t B \t C ;
(v) j= A t B ! C t A t C t B ;
(vi) j= A t B ! A [ C t B [ C :
Proof: Claims (i) and (ii) are obvious. And (iii) is a simple consequence
of the transitivity of jt , cf. Proposition 18.2.1(ii) together with the Denition
18.2.3 of t and of the left continuity of t.
(iv) By denition of t one has
j= (A jt B ^t B jt A) ! (A \t C jt B \t C ^t B \t C jt A \t C )
and hence the claim as a consequence of Proposition 18.2.1(iii), using that
j= (H1 !t G1 ) ^t (H2 !t G2 ) ! (H1 ^t H2 !t G1 ^t G2 )
holds for all formulas H1 ; H2 ; G1 ; G2 ; cf. Proposition 13.1.8(iii).
Finally, claims (v) and (vi) can be established in exactly the same way
from Proposition 18.2.1(iv) and (v), respectively.
2
18.2.3 Particular fuzzy sets

For the formulation of further results we need some additional notions. At
rst these are the empty fuzzy set, fuzzy singletons, and the universal fuzzy
set, and later on the intersection and the union of a (crisp) family of fuzzy
sets.
Denition 18.2.4. For any t-norm t let the universal set X with respect to
the universe of discourse X be
X =def fx 2 X k x =: xg ;
and for each truth degree u 6= 0 the u-universal set with respect to X
X [u] =def fx 2 X k x =: x ^t ug:
The empty fuzzy set ; (with respect to X ) is
; =def C X :
And for each a 2 X and any truth degree u the (fuzzy) u-singleton of a is
hhaiiu =def fx k x =: a ^t ug :
432
By abuse of notation we have, in the denitions of the u-universal set

X [u] and the u-singletons, also taken the truth degree u as the constant of
our language denoting just this degree. And we have chosen the same symbol
for the empty fuzzy set as for the usual, i.e. crisp empty set. This will not
cause any problems and has the advantage of simplicity.
De facto we have that X [1] = X and that each fuzzy singleton hhaiiu is
independent of the t-norm which is used in this denition, because obviously
always
[ b " X [u]] = u;
n
[ b " hhaiiu ] = u; if b = a
0 otherwise
hold true and additionally [ b " ;] = 0. Furthermore, hhaii0 = ; is always the
case. Obviously too one has
j= 8x(x " X ) ;
j= 8x n (x " ;)
for any negation function n.
Proposition 18.2.4. For each fuzzy set A and each left continuous t-norm
t one has
(i) j= A t ; $ 8x ;t (x " A) ;
(ii) j= ;t 9x(x " A) $ A t ; ;
(iii) j= A t X $ 8x(x " A) :
Proof: (i) is obvious from the corresponding denitions and the fact that
each R-implication which is based on a left continuous t-norm has the degree
ranking property according to Theorem 5.4.1. (ii) follows from (i) and
j= ;t 9xH $ 8x ;t H :
Finally, (iii) again is a simple consequence of the corresponding denitions
and the left neutrality (of the truth degree function) of !t .
2
The denitions of ; and X are quite natural ones from the point of view
of set theory. With the same resulting objects we could have chosen a more
algebraic approach: if one includes the truth degrees 0 and 1 as nullary operation in the structure D then also the direct power F(X ) has to be extended
with two nullary operations, and these are exactly ; and X . As Theorem
18.2.1 (on the transfer of the truth of Horn formulas) covers this case too,
we can also get a series of results this way, e.g. for all fuzzy sets A
; A and A X ;
or in our many-valued language
j= ; jt A and j= A jt X
433
for every t-norm t. From the fact that

p ^t (;t p) = ?
holds true in D because of (5.88), we also get in this way immediately
A \t CCt A = ; ;
as those formulas are (basic) Horn formulas. Thus we have in our many
valued language
j= A \t C t A t ; :
But, again, more and stronger results are provable than by this reference only
to the Transfer Theorem 18.2.1.
Proposition 18.2.5. For all left continuous t-norms t, t-conorms st and
fuzzy sets A; B one has
(i) j= A \t B t X ! A t X ^ B t X ;
(ii) j= A t X ^t B t X ! A \t B t X ;
(iii) j= A [t B t ; ! A t ; ^ B t ; ;
(iv) j= A t ; ^t B t ; ! A [t B t ; :
Proof: We prove only (iii) and (iv), because (i) and (ii) can be derived
in a similar manner. The details are left to the reader.
(iii) We immediately have
[ A [t B t ;] = [ 8x ;t (x " A _t x " B )]]
[ 8x ;t (x " A)]] = [ A t ;]
from monotonicity properties of _t and ;t . Hence by commutativity of [t :
[ A [t B t ;] minf[ A t ;] ; [ B t ;] g ;
and hence the result which was stated.
Claim (iv) can be derived similarly using
[ A t ; ^t B t ;] = [ 8x ;t (x " A) ^t 8x ;t (x " B )]]
[ 8x(;t (x " A) ^t ;t (x " B ))]]
[ 8x ;t (x " A _t x " B )]]
which holds true according to Propositions 13.2.1(i) and 13.1.8(v).
2
1
18.2.4 Generalized set algebraic operations

In order to dene the \big", i.e. innitary unions and intersections of families of fuzzy sets we consider a slight extension of our many-valued language:
we introduce innitary conjunctions and disjunctions. The nite iterations of
each one of the conjunction operators ^t and of each one of the disjunction
434
operators _t are routine and will not be discussed here. What is much more
problematic is the innitary version in general: but we avoid almost all diculties in restricting ourselves to the t-norm min and the t-conorm max only.
Having given any (crisp) index set I and for each i 2 I a formula Hi , we
introduce as new formulas for innitary conjunctions and disjunctions
^
^
_
_
H
and
H:
i
i2I
i2I i
Those new formulas may be used in building more complex ones as we have
so far done with the more simple ones. The truth values of those formulas
are given as
^
^
[ i2I Hi ] =def iinf
[H ] ;
2I i
_
_
H ] = sup [ H ] :
i2I i def i2I i
V
V
W
W
From these denitions it is obvious that i2I and i2I will act much like
the quantiers 8 and 9. But we will not prove a detailed list of logically valid
formulas including such innitary parts { if in the following we have to use
some special results we will derive them when needed.
Denition 18.2.5. For each family (Ai )i2I of fuzzy sets let
\
^
^
Ai =def fx k i2I (x " Ai )g ;
i2I
[
i2I
Ai =def fx k
_
_
i2I (x " Ai )g :
Proposition 18.2.6. For all fuzzy sets (Ai )i2I , B, and arbitrary left continuous t-norms t there hold true:
\
(i) j= Ai jt Ak ;
i2I
(ii) j= Ak t
(iii) j=
^
^
(iv) j=
^
^
i2I
Ai ;
(B jt Ai ) $ B jt
i2I
(A j B ) $
i2I i t
i2I
Ai ;
Ai t B ;
i2I
\
\
(v) j= Ai \t B (Ai \t B ) ;
i2I
i2I
\
\
(vi) j= Ai [t B (Ai [t B ) ;
i2I
i2I
[
[
(vii) j= (Ai \t B ) Ai \t B ;
i2I
i2I
j
j
j
(viii) j=
i2I
(Ai [t B ) j
i2I
435
Ai [t B :
Proof: Claims (i) and (ii) are obvious. For (iii) we observe that
^
^
[ i2I (B jt Ai )]] = inf
inf [ x " B !t x " Ai ]
i2I x2X
^
^
= xinf
[
x
"
B
!
(x " Ai )]]
t
i2I
2X
according to Proposition 13.2.2(i). This immediately gives the result.

To prove (iv) via an equation of suitable truth degrees we proceed analogously but with reference to Proposition 13.2.2(ii).
Furthermore, (v) to (viii) all are simple consequences of the monotonicity
of t-norms and t-conorms. The details can be omitted.
2
Proposition 18.2.7. For all fuzzy sets (Ai )i2I , B, and arbitrary continuous
t-norms t there hold true:
\
\
(i) j= Ai \t B (Ai \t B ) ;
(ii) j=
(iii) j=
(iv) j=
i2I
i2I
i2I
i2I
Ai [t B
Ai \t B
Ai [t B
i2I
\
i2I
[
i2I
i2I
(Ai [t B ) ;
(Ai \t B ) ;
(Ai [t B ) :
Proof: By straightforward calculations from the monotonicity of t and

st as mentioned in the last proof. The continuity assumptions guarantee the
equality of truth degrees of the left and right sides of the t -equations. They
can be suitably weakened if necessary.
2
18.2.5 Fuzzy cartesian products

The next topic shall be the consideration of cartesian products of fuzzy sets.
The problem here is { as later on with fuzzy (binary) relations again { that
the cartesian product of two fuzzy subsets of X need not be a fuzzy subset
of X : it depends on X and on whether X is closed with respect to (usual)
ordered pairing.
Fortunately, for the following discussions of fuzzy cartesian products we
do not need to have the universe of discourse X closed under ordered pairing:
we can always work with fuzzy subsets of dierent universes of discourse.
First let us dene the fuzzy cartesian products using the usual notion of
an ordered pair (a; b) of elements a; b of the universe of discourse X .
436
Denition 18.2.6. For all fuzzy sets A; B and each t-norm t let
A t B =def f(x; y) k x " A ^t y " B g :
Obviously we have as an extended meaning of jt and t for cartesian
products:
A t B jt C t D =def
8x8y(x " A ^t y " B !t x " C ^t y " D) ;
A t B t C t D =def
(A t B jt C t D) ^t (C t D jt A t B )
for all fuzzy sets A; B; C; D and all left continuous t-norms t; t1 .
1
(18.9)
(18.10)
Set algebraic equations involving fuzzy cartesian products easily result

from identities for t-norms, t-conorms and negation functions via the Transfer Theorem 18.2.1. To illustrate the method of proof we give some simple
examples.
Proposition 18.2.8. For t-norms t and fuzzy sets A; B; C there hold true
(i) j= A t (B \ C ) (A t B ) \ (A t C ) ;
(ii) j= A t (B [ C ) (A t B ) [ (A t C ) :
Proof: We have by the corresponding denitions
A t (B \ C ) = f(x; y) k x " A ^t (y " B ^ y " C )g
= f(x; y) k (x " A ^t y " B ) ^ (x " A ^t y " C )g
= (A t B ) \ (A t C )
and hence (i). In the same way the proof of (ii) can proceed.
2
We only mention the fact that the fuzzy cartesian product is commutative and associative { via suitable canonical isomorphisms as in the classical
case. Further equalities for fuzzy cartesian products are easily provable by
the method of the last proof. We will not discuss this topic more extensively.
Instead, we look for proofs of some monotonicity properties and some substitutivity results for fuzzy identities.
Proposition 18.2.9. Let t be a left continuous t-norm, and A; B; C as well
as all (Ai )i2I fuzzy sets. Then one has:
(i) j= A jt B ! A t C jt B t C ;
(ii) j= A t B ! A t C t B t C ;
(iii) j= A t ; _ B t ; ! A t B t ; ;
(iv) j= A t B t ; ! A t ; _t B t ; :
With respect to the innitary unions and intersections one has
(v) j=
(vi) j=
i2I
\
i2I
(Ai t B )
Ai t B j
i2I
i2I
437
Ai t B ;
(Ai t B ) ;
and for a continuous t-norm t even

\
\
(vii) j= Ai t B (Ai t B ) :
i2 I
i2I
We omit the proof. But all those statements are similar to earlier ones
relating e.g. to \t instead of t . In every case, too, the proofs needed here are
similar to those earlier proofs which we have presented in detail or sketched
with hints for the essential steps.
The part of set algebra for fuzzy sets that was presented here will be
representative enough to allow the reader to extend it easily by himself in
any of the directions he is interested in. Therefore, we nish our listing of
interesting results of set algebra.
18.2.6 The extension principle

The ordered pair (x; y) is a very special object among those which can be
built up from x; y. More generally we may assume to have given some binary {
or also n-ary, but we disregard this only slightly more general case { function
f on the universe of discourse X . Additionally we assume that each one of
the values f (a; b) can be described by a term of our language. To extend our
class terms to the case of fuzzy sets with objects f (a; b) as \elements" we
dene for all formulas H (x; y):
ff (x; y) k H (x; y)g =def fz k 9x9y(H (x; y) ^t z =: f (x; y))g
(18.11)
for each t-norm t. To avoid any inessential complication let us additionally
assume that always f (x; y) 2 X , i.e. that f is a binary operation in X . Now
we can formulate the famous extension principle of [585] for any t-norm in a
very simple manner.
Denition 18.2.7 (Extension Principle). Suppose that a binary operation in the universe of discourse X is given and some t-norm t chosen.
This operation is extended to a binary operation t for fuzzy sets from IF (X )
by dening
A t B =def fa b k a " A ^t b " B g
for all fuzzy sets A; B 2 IF (X ).
Corollary 18.2.1. Take the operations and t as in Denition 18.2.7 and
consider C = A t B for A; B 2 IF (X ). Then one has
C (z ) = sup t(A (x) B (y)) for all z 2 X :
x;y2X
z=xy
438
Proof: By (18.11) and the denitions of the quantier 9 and the class
term notation for fuzzy sets, the extension principle immediately yields
C (z ) = sup [ x " A ^t y " B ] t [ z =: x y] :
x;y2X
Therefore
C (z ) = sup [ x " A ^t y " B ]
x;y2X
z=xy
because always [ z =: x y] 2: f0; 1g, and only those pairs (x; y) need to be
considered here for which [ z = x y] = 1, i.e. for which z = x y is the case.5
Now
[ x " A ^t y " B ] = t([[x " A] ; [ y " B ] ) = t(A (x); B (y))
gives the result.
2
18.3 Fuzzy Relations

Intuitively, a fuzzy relation is a fuzzy set of ordered pairs, i.e. such a fuzzy
set whose support is a subset of a suitable crisp cartesian product. That
means, we intuitively identify relations with binary ones. This will be enough
generality for the present purposes. Of course, mathematically it is obvious
that in the same way one may consider n-ary fuzzy relations for any n 2.
From the formal point of view we have dierent possibilities to treat fuzzy
relations, e.g. we may
{ assume that our universe of discourse X consists of ordered pairs only, such
that each fuzzy subset of X is a fuzzy relation;
{ suppose that the universe of discourse X is closed with respect to ordered
pairing, i.e., we may suppose that for a; b 2 X we have (a; b) 2 X too;
{ consider parallel with the universe of discourse X the set X X as a
second universe and distinguish fuzzy sets as fuzzy subsets of X from fuzzy
relations as fuzzy subsets of X X .
We will essentially follow this third approach here. The rst one would
allow for a too restricted treatment of fuzzy relations only; and the second
approach in principle would force us to take as X a whole universe for classical
set theory (or \nearly" such a whole universe): a situation which we are
interested to avoid here.
We have to adapt our language slightly to treat fuzzy relations easily.
First we like to have another sort of symbols: those for fuzzy relations. We
now use the upper case Latin letters
5 It is essential for this argumentation that as usual the supremum over the empty
set is taken to be =1.
R; S; T
439
(possibly with subscripts or primes)

as symbols for fuzzy relations. Because we do not have to quantify over fuzzy
relations here, these symbols will act as constants for fuzzy relations. Of
course, as before we will avoid using the letters R; S; T as symbols of fuzzy
sets. But, again as in connection with cartesian products of fuzzy sets, we
extend in a straightforward way our use of generalized { i.e. many-valued {
class terms to the case of fuzzy relations too. And the same will be done with
the notation for the set algebraic operations; i.e. we take e.g.
R \t S =def f(x; y) k (x; y) " R ^t (x; y) " S g ;
and in the same manner put
R jt S =def 8x8y((x; y) " R !t (x; y) " S ) :
It is obvious that all the results on inclusion and intersection of fuzzy
sets, discussed in the last section, hold (with slightly modied proofs only)
for fuzzy relations too.
Often we shall use the same symbols for the fuzzy case which previously
have been used for the crisp case. This will not cause any misunderstandings
or problems: from the context is will always be clear whether the fuzzy or
the crisp version is under consideration.
Denition 18.3.1. For each t-norm t and all fuzzy relations R; S let the
domain, the range, the inverse relation of R and the (relational) product of
R; S be:
dom(R) =def fx k 9y((x; y) " R)g ;
rg(R) =def fy k 9x((x; y) " R)g ;
R;1 =def f(x; y) k (y; x) " Rg ;
R t S =def f(x; y) k 9z ((x; z ) " R ^t (z; y) " S )g :
Proposition 18.3.1. For all fuzzy relations R; S one has
(i) dom(R;1 ) = rg(R) and rg(R;1 ) = dom(R) ;
(ii) j= dom(R t S ) j dom(R) ;
(iii) j= rg(R t S ) j rg(R) ;
(iv) (R;1 );1 = R;
(v) (R t S );1 = S ;1 t R;1 :
Proof: All the results can be proved by routine methods; thus e.g. (v)
essentially follows from the denition of the inverse fuzzy relation R;1 and
the commutativity of ^t by
(R t S );1 = f(x; y) k 9z ((y; z ) " R ^t (z; x) " S )g
= f(x; y) k 9z ((x; z ) " S ;1 ^t (z; y) " R;1)g
= S ;1 t R;1 :
440
And for (ii), for example, we have to show

j= 8x(x " dom(R t S ) ! x " dom(R)) ;
which means to show
j= 8x(9y((x; y) " R t S ) ! 9z ((x; z ) " R)) :
(18.12)
But because of
[ (x; y) " R t S ] = [ 9z ((x; z ) " R ^t (z; y) " S )]]
we have by monotonicity of the t-norms
j= 9y((x; y) " R t S ) ! 9y9z ((x; z ) " R) ;
and hence (18.12) by rst dropping the empty quantication 9y in the succedent and then universally quantifying.
2
Proposition 18.3.2. If t is left continuous, or if X is nite, or also if
R; S; T have only a nite number of membership degrees, then

R t (S t T ) = (R t S ) t T :
Proof: The proof is a routine matter. The essential point is to prove
that
and
9u((x; u) " R ^t 9v((u; v) " S ^t (v; y) " T ))

9v(9u((x; u) " R ^t (u; v) " S ) ^t (v; y) " T )
have the same truth degrees. To get that result we have { besides using the
associativity of ^t { rst to \move the quantier 9v outside" a conjunction
and then to \move 9u inside" a corresponding conjunction. As stated in
Proposition 13.2.4(i) this is possible without a change of truth degrees for a
left continuous t-norm t. Obviously, these changes of quantiers are unproblematic in those cases where the existential quantiers 9u; 9v correspond to
the operations of taking a maximum of truth degrees, which happens if only
a nite number of truth degrees are possible for the formulas in the scopes
of 9u or 9v, and this happens e.g. for nite universes of discourse X , and
for those fuzzy relations R; S; T with only a nite number of membership
degrees really appearing, i.e. for those membership functions R ; S ; T of
fuzzy relations which have nite ranges.
2
Proposition 18.3.3. For each left continuous t-norm t and all fuzzy relations R; S; T one has
(i) j= R t (S \ T ) j (R t S ) \ (R t T ) ;
(ii) j= R t (S [ T ) j (R t S ) [ (R t T ) ;
(iii) j= R jt S ! R t T jt S t T ;
(iv) j= S jt T ! R t S jt R t T :
441
Proof: One immediately has

R t (S \ T )
= f(x; y) k 9z ((x; z ) " R ^t ((z; y) " S ^ (z; y) " T ))g
= f(x; y) k 9z (((x; z ) " R ^t (z; y) " S ) ^ ((x; z ) " R ^t (z; y) " T ))g ;
which gives (i) by the denition of the union of fuzzy relations and the fact
that
[ 9z (H1 ^ H2 )]] [ 9zH1 ^ 9zH2 ]
holds true for all formulas H1 ; H2 .
Claim (ii) is proved by similar calculations now using the fact that
[ 9z (H1 _ H2 )]] [ 9zH1 _ 9zH2 ]
holds true for all formulas H1 ; H2 .
From the remaining two cases we consider only (iii). Claim (iv) has to be
treated similarly. For (iii) one has
j= R jt S !
8x8y8z ((x; y) " R ^t (y; z ) " T !t (x; y) " S ^t (y; z ) " T ) ;
because one has by Proposition 13.1.7(i)
j= (H1 !t H2 ) ! (H1 ^t G !t H2 ^t G) :
Furthermore one has obviously
j= R jt S ! 8x8y8z ((x; y) " R ^t (y; z ) " T !t (x; z ) " S t T ) ;
and therefore, using the fact that for left continuous t one has
j= 8x8y8z ((x; y) " R ^t (y; z ) " T !t (x; z ) " S t T )
! 8x8z (9y((x; y) " R ^t (y; z ) " T ) !t (x; z ) " S t T )
according to the behavior of the universal quantier, one has by transitivity
of the implication connective
j= R jt S ! 8x8z ((x; y) " R t T !t (x; z ) " S t T ) ;
which had to be proved.
2
Corollary 18.3.1. For any left continuous t-norm t the generalized monotonicity of the relational product of fuzzy relations holds true in the general
form
j= R1 jt R2 ^t S1 jt S2 !t R1 t S1 jt R2 t S2 :
442
Proof: To get the result one has to use claims (iii), (iv) of the preceding proposition together with the transitivity of fuzzy inclusion for fuzzy
relations:
j= R jt S ^t S jt T !t R jt T
which is nothing but a special case of the corresponding result for fuzzy sets
in general.
2
18.4 The Full Image Under a Relation

In classical set theory, having given a set of ordered pairs, one always has
two ways of looking at it: one may consider this set as a (binary) relation or
as a (unary) mapping.6 Depending on that point of view, one is interested in
dierent properties of and operations with sets of ordered pairs.
The same situation, of course, appears with fuzzy sets of ordered pairs.
So far we have looked at them as fuzzy relations. For the present book this
perspective shall be the essential one. Nevertheless, there is one notion from
the realm of the mapping-related ones that has to be considered together
with fuzzy relations: the full image of a (fuzzy) set under a (fuzzy) relation.
The background for this interest is the use of the so-called \compositional
rule of inference" in connection with the topic of approximate reasoning in
some AI applications of many-valued logic.
From a more set theoretical point of view the notions and facts from
the mapping-related world are as essential and interesting as those from the
relation-related world. This also holds true in the eld of fuzzy sets. Moreover, in the mapping-related world of fuzzy sets there arise completely new
problems, e.g. at rst giving a satisfactory denition of uniqueness that is
nontrivial in the sense that it refers to an equality predicate that allows for
truth degrees dierent from the degrees 1 and 0.7
Denition 18.4.1. For each t-norm t, every fuzzy relation R 2 IF (X Y )
and all fuzzy sets A 2 IF (X ) and B 2 IF (Y ), let
R00 A =def fy k 9x(x " A ^t (x; y) " R)g ;
and in the case that t is left continuous also
R # B =def fx k 8y((x; y) " R !t y " B )g :
6 We follow the usage to speak of a mapping also in the case that there is no
uniqueness with respect to the second argument, mappings are thus \multiple
valued" functions. We reserve the name of a \function" as usual for the unique
mappings.
7 For work in this direction the interested reader is referred to [210] as one possible
approach.
18.4 The Full Image Under a Relation
443
Obviously, R00 A is a fuzzied version of the full image of a set under a

mapping, i.e. under a binary relation. Simultaneously, R00 A corresponds to
what in [40] was called \conditioned fuzzy set". And R # A corresponds to
the
-operation of [479]. In some (weak) sense the operation R # B is dual
to taking the full image R00 A of A under R.
As a side remark let us mention that this fuzzied full image of a fuzzy
set under a fuzzy relation could be used to dene the range and hence, cf.
Proposition 18.3.1(i), also the domain of a fuzzy relation. The reason is that
rg(R) = R00 (X [1])
(18.13)
holds true for any fuzzy relation R 2 IF (X Y ).
Usually the t-norm used in the denition of R00 A is clear from the context
and the same one which is (essentially) used there. Thus we shall not mention
it explicitly in our notation.
Proposition 18.4.1. For all fuzzy sets A; B 2 IF (X ) and each fuzzy relation
R 2 IF (X Y ) there hold true
(i) j= R00 (A [ B ) R00 A [ R00 B ;
(ii) j= R00 (A \ B ) j R00 A \ R00 B :
In the case that R00 :: is dened using the t-norm tG = min one has also
(iii) j= R00 A [t R00 B j (R [t R)00 (A [t B ) :
Proof: (i) By Denition 18.4.1 and Proposition 13.1.3(ii) one has
R00 A [ R00 B = fy k 9x(x " A ^t (x; y) " R) _ 9x(x " B ^t (x; y) " R)g
= fy k 9x((x " A ^t (x; y) " R) _ (x " B ^t (x; y) " R))g
= fy k 9x((x " A _ x " B ) ^t (x; y) " R)g
= R00 (A [ B ) :
Claim (ii) follows by the same type of calculations but now using Proposition 13.2.1(ii) for the step from the rst to the second line, giving the
(nonfuzzy) inclusion \" instead of equality.
(iii) With reference to Proposition 13.2.1(iv) the calculations to determine
R00 A [t R00 B can start as in case (i) and proceed as in (ii), but now with an
inclusion \" resulting, and have later on to use that one has
j= (H1 ^ H3 ) _t (H2 ^ H3 ) $ (H1 _t H2 ) ^ (H3 _t H3 )
for all ws H1 ; H2 ; H3 . The rest is a routine matter.
2
For the following let us suppose for simplicity that X = Y , and thus that
all fuzzy sets are from IF (X ) and all fuzzy relations from IF (X X ).
Proposition 18.4.2. Let t be left continuous. Then one has for all fuzzy
sets A; B and all fuzzy relations R; S
444
(i)
(ii)
(iii)
(iv)
j= A jt B ! R00 A jt R00 B ;
j= R jt S ! R00 A jt S 00 A ;
j= A jt B ! R # A jt R # B ;
j= R jt S ! S # A jt R # A :
Proof: (i) Because of Proposition 13.1.1 the logical validity of a t-norm

based R-implication is equivalent to an inequality of the corresponding truth
degrees. What has to be shown therefore is that
[ A jt B ] [ R00 A jt R00 B ] :
We start from the right hand side of this inequality and have
[ R00 A jt R00 B ]
= [ 8x(x " R00A !t x " R00 B )]]
= [ 8x(9y(y " A ^t (y; x) " R) !t 9z (z " B ^t (z; x) " R))]] ;
and using Proposition 13.2.2(ii) furthermore
[ R00 A jt R00 B ]
= [ 8x8y((y " A ^t (y; x) " R) !t 9z (z " B ^t (z; x) " R))]] :
Now from the properties of the existential quantier 9 together with the right
isotonicity of the R-implication and Proposition 13.1.7(i) one has
[ R00 A jt R00 B ]
[ 8x8y((y " A ^t (y; x) " R) !t (y " B ^t (y; x) " R))]]
[ 8x8y(y " A !t y " B )]]
which, by dropping the empty quantication \8x" just means
[ R00 A jt R00 B ] [ A jt B ] :
The same kind of calculation yields (ii). We omit the details.
(iii) As before, it has to be shown that the inequality
[ A jt B ] [ R # A jt R # B ]
holds true. But now the corresponding denitions lead to
[ R # A jt R # B ]
= [ 8x(8y((x; y) " R !t y " A) !t 8z ((x; z ) " R !t z " B ))]]
= [ 8x8z (8y((x; y) " R !t y " A) !t ((x; z ) " R !t z " B ))]]
[ 8x8z (((x; z ) " R !t z " A) !t ((x; z ) " R !t z " B ))]]
[ 8x8z (z " A) !t z " B )]] ;
using the antimonotonicity (of the truth degree function) of !t in the rst
argument as well as Proposition 13.1.7(ii).
Finally (iv) can be proved along the same line of arguments. But now
instead of Proposition 13.1.7(ii) reference has to be made to Proposition
18.5 Special Types of Fuzzy Relations
445
13.1.7(iii) which causes the reversal in the order of appearance of R and S

on both sides of the arrow ! in the formula (iv).
2

In order to be able to discuss some special types of fuzzy relations, we rst
introduce some specic properties of fuzzy relations. In this subsection these
shall always be crisp properties, i.e. properties which a fuzzy relation either
has or does not have. Dierent lists of such properties can be found e.g. in
[126, 206, 408], and [88]. As we only consider binary relations, we will change
our notation slightly and write { as usual for crisp binary relations { simply
xRy for (x; y) " R, i.e. we use [ xRy] = R (x; y).
Denition 18.5.1. Let t be a left continuous t-norm and n some negation
function. Then a fuzzy relation R is:
(i) re exive (in X ) i j= 8x(xRx) ;
(ii) irre exive
i j= 8x n (xRx) ;
(iii) t-transitive
i j= 8x8y8z (xRy ^t yRz !t xRz ) ;
(iv) symmetric
i j= 8x8y(xRy !t yRx) ; :
(v) t-antisymmetric i j= 8x8y(xRy ^t yRx !t x = y) ;
(vi) t-asymmetric
i j= 8x8y n (xRy ^t yRx) ;
(vii) st -linear
i j= 8x8y(xRy _t yRx) :
By the usual truth degree conditions for logical validity and by denition
(13.3) of the universal quantier 8 the irre exivity condition becomes simply:
[ xRx] = 1 for each x 2 X . Therefore Denition 18.5.1(ii) of irre exivity
and Denition 18.5.1(vi) of t-asymmetry are independent of the choice of a
particular negation function.
As an additional consequence the symmetry condition (iv) becomes that:
[ xRy !t yRx] = 1 for all x; y 2 X , which means that this condition reads:
[ xRy] [ yRx] for all x; y 2 X . Thus Denition 18.5.1(iv) is also independent
of the choice of a particular left continuous t-norm.
As usual these properties may also be characterized using the set theoretic operations for fuzzy relations. To do this one needs a kind of fuzzied
diagonal. Let the fuzzy diagonal of X be the fuzzy relation
[
X =def hh(a; a)ii1 = f(x; y) 2 X X k x =: yg:
(18.14)
a2X
Then the following results are obvious.

Corollary 18.5.1. For each fuzzy relation R it holds true that
446
(i) R re exive (in X ) i j= X j R;

(ii) R irre exive
i j= R \t X ;;
(iii) R t-transitive
i j= R t R jt R
i j= R t R j R;
(iv) R symmetric
i j= R j R;1 ;
(v) R t-antisymmetric i j= R \t R;1 j X ;
(vi) R t-asymmetric
i j= R \t R;1 ;:
All these properties look quite natural { up to the: t-antisymmetry, which
uses the articially many-valued equality relation =.
And indeed this is a
weak point of this approach toward relation properties for fuzzy relations
because we shall see later on that it forces fuzzy linear orderings to become
crisp orderings only. The way out, as it seems,
has to be to introduce suit: i.e.
able generalizations of the equality relation =,
to introduce many-valued
identity relations in the sense of Chapter 16, cf. e.g. [71, 72].
18.5.1 Fuzzy equivalence relations

Here we look for a fuzzication of equivalence relations, and shall rst consider
fuzzy equivalence relations in the universe of discourse, i.e. fuzzy equivalence
relations in crisp sets. Such fuzzied equivalence relations have been discussed
e.g. in [303, 408, 584] and [89]. We give only the most basic facts.
Denition 18.5.2. A fuzzy equivalence relation R (in X ) is such a fuzzy
relation which is re exive (in X ), symmetric and t-transitive. For each fuzzy
equivalence relation R and each b 2 X the fuzzy R-equivalence class of b is
hbiR =def fx k xRbg :
(18.15)
And X =R denotes the class of all fuzzy equivalence classes of R in X .
By the symmetry of each fuzzy equivalence relation one has hbiR =
fx k bRxg, and using the notion of full image from Denition 18.4.1 this gives
hbiR = R00 hhbii1
for the fuzzy R-equivalence classes.
Proposition 18.5.1. For each fuzzy equivalence relation R, each left continuous t-norm t and all a; b 2 X one has:
(i) j= a " haiR ;
(ii) j= b " haiR $t aRb ;
(iii) j= haiR t hbiR $t aRb ;
(iv) j= 9x(x " haiR \t hbiR ) $t aRb :
447
Proof: Claim (ii) is simply a restatement of the denition of haiR ; and

(i) follows from this denition by re exivity of R.
(iii) From denition of t and from (ii) one immediately has
j= haiR t hbiR !t (aRa $t aRb) ;
and hence by re exivity of R
j= haiR t hbiR !t aRb :
Furthermore one has by transitivity of R and symmetry
j= aRb !t (xRa !t xRb) ;
j= aRb !t (xRb !t xRa) :
Therefore one immediately has
j= aRb !t ((xRa !t xRb) ^ (xRb !t xRa)) :
Because of the linear ordering of the set of truth degrees one, however, has
[ (H !t G) ^ (G !t H )]] = [ (H !t G) ^t (G !t H )]] ;
which in the present case gives
j= aRb !t ((xRa $t xRb)
and hence, e.g. using Proposition 13.2.2(i),
j= aRb !t haiR t hbiR :
Thus (iii) is proved.
(iv) By (i) and (ii) it holds true that
j= aRb !t b " haiR \t hbiR ;
and hence of course as part of (iv) already
j= aRb !t 9x(x " haiR \t hbiR ) :
(18.16)
Otherwise we also have
j= x " haiR \t hbiR !t xRa ^t xRb ;
and thus together with t-transitivity and symmetry of R
j= x " haiR \t hbiR !t aRb
which by the left continuity of t gives
j= 9x(x " haiR \t hbiR !t aRb) :
(18.17)
Thus nally (iv) is proved by (18.16) and (18.17).
2
Our approach to fuzzy equivalence relations { or: (fuzzy) similarity relations using the terminology of Zadeh [584] { is a straightforward generalization of the corresponding approaches of [584] and of [303]. The approach
gives a fuzzication of the notion of equivalence relation in a given universe
of discourse, i.e. in a given crisp set. What classically comes to the same
conclusion is to consider an equivalence relation in some crisp set: one only
has to interpret that crisp set as the universe of discourse.
448
18.5.2 Fuzzy partitions of fuzzy sets

Unfortunately, for fuzzy sets the situation is not so simple. Of course, one
is interested to have not only a theory of fuzzy equivalence relations in a
given universe of discourse, but also a theory of fuzzy equivalence relations
in a given fuzzy set A. Instead, also a notion of fuzzy partition of a fuzzy set
would be welcome, having in mind that classically equivalence relations and
partitions are interchangeable tools. Such an approach toward fuzzy partitions of fuzzy sets was oered in [274], comes from the theory of L-valued
equality relations as explained in Chapter 16, and approaches fuzzy equivalence relations via fuzzy partitions.
Suppose that L = hL; \; [; ; ; 0; 1i is a complete and divisible residuated lattice, in which additionally the operation distributes over arbitrary
meets.
The starting point is that the usual denition of an equivalence class is not
immediately generalized as in (18.15) but in a (seemingly) more complicated
manner. The basic notion which has to be introduced is the notion of an
L-valued singleton.8
Denition 18.5.3. Let (A; E ) be an L-valued set w.r.t. the complete residuated lattice L = hL; \; [; ; ; 0; 1i as truth degree structure. Then an L-fuzzy
set s : A ! L is an L-valued E -singleton i the following conditions are satised for all x; y 2 A:
(S 1) s(x) (E (x; x) E (x; y)) 6 s(y) ;
extensionality
(S 2) s(x) (e(s) s(y)) 6 E (x; y)
singleton condition
for the height e(s) = supfs(z ) j z 2 Ag of the fuzzy set s.
A simple rewriting of these conditions in the formalized language LML of
rst-order monoidal logic, enriched with the graded membership predicate
" , gives them the form:
(S1) j= x " s & (x $ x ! x $ y) ! y " s ;
(S2) j= x " s & (9z (z " s) ! y 2 s) ! x $ y :
To make this relationship again obvious we proceed to write [ x " s] instead of
s(x) also in this monoidal case. And we also extend the graded inclusion relations to the present case, writing [ A jML B ] for inf x2X ([[x " A] [ x " B ] )
for any L-fuzzy sets A; B : X ! L.
Denition 18.5.4. Let L = hL; \; [; ; ; 0; 1i be a complete residuated lattice, and X some ordinary set. Then a family (Ai )i2I of L-fuzzy subsets of
X is an L-valued partition of X i it satises the condition
8 To see why the notion of singleton may come into consideration here one has to
remind that the classical denition [a] = fb j a ' bg of an equivalence class under
an equivalence relation ', as well as its fuzzication (18.15), are completely
similar to the characterization fag = fb j a = bg of the singleton of a in classical
set theory.
sup
x2X

[x "

j 2I
Aj ] [ x " Ai ] [ x " Ak ]
6 e(Ai ) [ Ai jML Ak ]
for all i; k 2 I .
449
(18.18)
^ e(Ak ) [ Ak jML Ai ]
Proposition 18.5.2. Suppose that L = hL; \; [; ; ; 0; 1i is a complete

residuated lattice, X some ordinary set, and (Ai )i2I a non-empty family of
L-fuzzy subsets of X . Then there are equivalent:
(a) (Ai )i2I is an L-valued partition of X .
(b) There exists an L-valued equality E on X such that
(i) each Ai is an L-valued E -singleton ,
[
(ii) E (x; x) = [ x " Ai ] for each x 2 X .
i2I
Proof: Observe rst that because of e(Ai ) > [ x " Ai ] ; [ y " Ai] for all
i 2 I and all x; y 2 X , one has from Corollary 14.1.2(ii):
[ x " Ai ] (e(Ai ) [ y " Ai ] ) = [ y " Ai ] (e(Ai ) [ x " Ai ] ) ;
as well as E (x; x) > [ x " Ai ] and E (x; x) > E (x; y), and therefore also
;

;

[ x " Ai ] E (x; x) E (x; y) = E (x; y) E (x; x) [ x " Ai ] :
And this means that (S1) and (S2) can equivalently be rewritten as
;

(S1) E (x; y) E (x; x) [ x " Ai ] 6 [ y " Ai ] ;
;

(S2) [ y " Ai ] e(Ai ) [ x " Ai ] 6 E (x; y) :
Now assume rst that E is an L-valued equality on X which satises (i)
and (ii). Then one has from (S1), (S2) for all i; k 2 I and all x; y 2 X

[ y " Ai ] e(Ai ) [ x " Ai ] [ x "

;
j 2I
Aj ] [ x " Ak ]
6 E (x; y) (E (x; x) [ x " Ak ] ) 6 [ y " Ak ] ;
and therefore
;
e(Ai ) [ x " Ai ] ;[ x "
This gives immediately

;
e(Ai ) [ x " Ai ] ;[ x "
j 2I
[
j 2I
Aj ] [ x " Ak ]

6 [ y " Ai ] [ y " Ak ] :
Aj ] [ x " Ak ]

6 [ Ai jML Ak ] ;
which together with e(Ai ) > [ x " Ai ] and Corollary 14.1.2(i) yields
450

;
[ x " Ai ] [ x "
j 2I
Aj ] [ x " Ak ]
= e(Ai ) e(Ai ) [ x " Ai ] [ x "

;
6 e(Ai ) [ Ai jML Ak ] ;
and thus
;
sup [ y " Ai ] [ x "
x2X
j 2I
j 2I
Aj ] [ x " Ak ]
Aj ] [ x " Ak ]

6 e(Ai ) [ Ai jML Ak ] :
Because one can, by symmetry, exchange in these arguments always i; k, one

gets via Corollary 14.1.2(ii) the dening condition (18.18) for an L-valued
partition, i.e. one gets (a).
Assume on the other hand that (Ai )i2I is an L-valued partition of X .
Consider the L-valued binary relation E0 in X which for all x; y 2 X is given
by

[
E0 (x; y) =def [ x " Aj ] inf
([
[
x
"
A
]

[
y
"
A
]
)
i
i
i2I
j 2I

^ [y "
j 2I
Aj ] iinf
([[y " Ai ] [ x " Ai ] ) :
2I
This relation is an L-valued equality on X . It obviously satises condition

(E2) from Chapter 16. And it satises also condition (E1) because one immediately has
[
E0 (x; x) = [ x " Aj ] ;
j 2I
which also means that (ii) is satised. So it remains to show (E3) and (i).
But one has, using the rst \conjunct" in the denition of E0 :
E0 (x;
y) (E0 (y; y) E0 (y; z ))

[
6 inf
([
[
x
"
A
]

[
y
"
A
]
)

[
x
"
A
]
i
i
j
i2I

[y "
j 2I
Aj ] [ y "
;
j 2I
j 2I
Aj ] inf
([[y " Ai ] [ z " Ai ] )
i2I

6 [ x " Aj ] inf
([[y " Ai ] [ z " Ai ] ) inf
([[x " Ai ] [ y " Ai ] )
i2I
i2I
j 2I
6 [ x " [ Aj ] inf
([[x " Ai ] [ z " Ai ] ) ;
i2I
[
j 2I
having in mind that

inf ([[x " Ai ] [ y " Ai ] ) iinf
([[y " Ai ] [ z " Ai ] ) 6 iinf
([[x " Ai ] [ z " Ai ] )
i2I
2I
2I
holds true. Similarly one gets from the second \conjunct"
451
E0 (x;
y) (E0 (y; y) E0 (y; z ))

6 [ z " [ Aj ] iinf2I ([[z " Ai] [ x " Ai ] ) :
j 2I
And both inequalities together give (E3).

So it remains to check for each Ai , i 2 I , the conditions (S1) and (S2) of
an L-valued E -singleton. Using (ii) one immediately has
[ x " Ai ] (E0 (x; x) E0 (x; y))

6 [ x " Ai] [ x " [ Aj ] ;[ x " [ Aj ] iinf2I ([[x " Ai ] [ y " Ai ] )
= [x "
j 2I
j 2I
Aj ] inf
([[x " Ai ] [ y " Ai ] ) [ x "
i2I
j 2I
[ x " Ai ] iinf
([[x " Ai ]
2I
j 2I
[ y " Ai] ) 6 [ y " Ai] ;
Aj ] [ x " Ai ]
and thus condition (S1) satised. And furthermore one gets via (18.18)

[
(e(Ai ) [ x " Ai ] ) [ y " Ai ] [ x " Aj ] [ x " Ak ]

j 2I
= (e(Ai ) [ y " Ai ] ) [ x " Ai ] [ x "
j 2I
Aj ] [ x " Ak ]
6 (e(Ai ) [ y " Ai ] ) e(Ai ) [ Ai jML Ak ] 6 [ y " Ak ] ;
and therefore also

(e(Ai ) [ x " Ai ] ) [ y " Ai ]

6 ;[ x " [ Aj ] [ x " Ak ] [ y " Ak ] ^ [ x "
= [x "
j 2I
j 2I
Aj ] ([[x " Ak ] [ y " Ak ] )
j 2I
Aj ]
for all k 2 I . Using the independence from the choice of k of the left-hand side
of this inequality together with the distributivity of over arbitrary meets,
one now gets
[
(e(Ai ) [ x " Ai ] ) [ y " Ai ] 6 [ x " Aj ] jinf
2I ([[x " Aj ] [ y " Aj ] ) ;
j 2I
which together with the interchangeability of x and y here, according to

Corollary 14.1.2(ii), nally also gives (S2) for Ai .
2
In summary this does also mean that we have a suitable denition of a
fuzzy equivalence relation in a fuzzy set: such a fuzzy equivalence relation in
a fuzzy set A 2 IF (X ) is simply an L-valued equality E on the universe of
discourse X which has the additional property
E (x; x) = [ x " A] for all x 2 X .
452
And this is a very natural generalization of the classical situation where also
an equivalence relation is symmetric and transitive, and has as its equivalence
classes \singletons" in the sense of Denition 18.5.3.
18.5.3 Transitive hulls
As the next topic, let us discuss the construction of the transitive hull of
a given fuzzy relation. As with fuzzy equivalence relations, the approach is
mainly along the same lines as in [584], but as before we easily get some more
general results.
Denition 18.5.5. For fuzzy relations R and any given t-norm t we put
recursively for all natural numbers n
R1 =def R;
Rn+1 =def Rn t R ;
and take as the t-transitive hull of R the fuzzy relation
Tt (R) =def
1
[
n=1
Rn :
Proposition 18.5.3. For each fuzzy relation R one has

(i) j= R j Tt (R) ;
(ii) R t-transitive ) R = Tt (R) ;
and if the t-norm t is left continuous also

(iii) Tt (R) is t-transitive ;
(iv) Tt (Tt (R)) = Tt (R) :
Proof: Claim (i) is obvious. Claim (ii) hence is proved if j= Tt(R) j R
is shown for t-transitive fuzzy relations R. But if R is t-transitive, i.e. if
j= R t R j R, we have by R3 = R2 t R and by the t-transitivity of j
immediately j= R3 j R. Inductively we get j= Rn j R for each integer
n 1. Therefore obviously j= Tt (R) j R and (ii) is proved.
For (iii) let m; n be any non-negative integers. Using the left continuity
of t and the denition of Rm in a straightforward way, one gets
[ (x; y) " Rn ^t (y; z ) " Rm] [ (x; z ) " Rn+m] ;
and thus
j= (x; y) " Rn ^t (y; z ) " Rm !t (x; z ) " Tt (R) :
Now taking suprema on all n 1 and on all m 1 gives
j= (x; y) " Tt (R) ^t (y; z ) " Tt(R) !t (x; z ) " Tt (R) :
i.e. the t-transitivity of Tt (R). By (ii) and (iii) now (iv) is obvious.
2
We can prove some relatively weak results on the monotonicity of the Tt operator for the general case of any t-norm. More interesting monotonicity
properties are provable in the case that t = min.
453
Proposition 18.5.4. For fuzzy relations R; S we have:

(i) if S is t-transitive and j= R j S , then j= Tt (R) j S ;
and if t is left continuous also

(ii) if j= R jt S , then j= Tt (R) jt Tt (S ):
Proof: If S is t-transitive then j= S n j S for each n 1 as mentioned
in the foregoing proof. Furthermore
j= R2 j R t S and j= R t S j S 2
by j= R j S and the monotonicity of t . Together that gives j= R2 j S ;
and in the same way more generally j= Rn j S for each n 1. Therefore
obviously j= Tt (R) j S , and hence (i).
By Proposition 18.5.3(i) we have j= R j Tt (S ) from j= R j S . Because
of the t-transitivity of the fuzzy relation Tt (S ) the claim (ii) follows from (i).
This proposition discusses the situation that, either with or without the
t-transitivity of the fuzzy relation S , one has that \R jt S " has truth degree
1. Then the statement is that some other inclusions have truth degree 1 too.
A more general approach would be to have instead, e.g., inequalities between
the degrees to which these fuzzy inclusion relations hold true. Such results
will now be established for a special case.9
Proposition 18.5.5. For the t-norm t = tG = min one has for all fuzzy
relations R; S in the case that S is tG -transitive:
(i) j= R jtG S ! TtG (R) jtG S ;
and without any transitivity restraint for S the generalized monotonicity
(ii) j= R jtG S ! TtG (R) jtG TtG (S ):
Proof: (i) Suppose S to be tG-transitive, i.e. to be min-transitive. Then
one has (always deleting the index tG ):
j= R j S ! R2 j R S ;
j= R j S ! R S j S 2 ;
and hence together with j= S 2 j S also j= R j S ! R2 j S .
Continuing this line of reasoning gives j= R j S ! Rn j S for each
n 1. From this one gets together with
inf [ Rn j S ] = [
n1
1
[
n=1
Rn j S ]
the fact that j= R j S ! T (R) j S .
9 At present it is an open problem whether the results of the next proposition can
be extended to other t-norms besides t = min or not.
454
(ii) Again deleting all indices tG we have from Proposition 18.5.3(i) that
j= R j S ! R j S ^ S j T (S ) holds, and hence j= R j S ! R j T (S ).

Now claim (i) of our proposition gives because of the transitivity of T (S )
j= R j T (S ) ! T (R) j T (S )
and thus j= R j S ! T (R) j T (S ) altogether.
2
18.5.4 Fuzzy ordering relations

As a further topic we discuss fuzzy ordering relations. Following [584] and
generalizing [206] we give the following:
Denition 18.5.6. A fuzzy relation which is re exive and t-transitive is
called fuzzy preordering; and a re exive, t-transitive, and t-antisymmetric
fuzzy relation is called fuzzy partial ordering.
For fuzzy relations as well as for crisp relations it is enough to discuss the
re exive case: in a simple way one can change from re exive fuzzy ordering
relations to corresponding irre exive ones in essentially the same manner as
in classical set theory.
Denition 18.5.7. For each t-norm t and each fuzzy relation R 2 IF (X X )
let
R+ =def R [t X ;
R; =def R \t CCX :
Corollary 18.5.2. For each t-norm t and each fuzzy relation R one has
(i) R; is irre exive
(ii) R+ is re exive.
The proof is obvious.
Proposition 18.5.6. Let R be any fuzzy relation and t any left continuous
t-norm. Then one has
(i) If R is t-transitive then R+ is t-transitive too.
(ii) If R; is t-transitive then R is t-transitive too.
(iii) If R is t-transitive and t-antisymmetric then R; is t-transitive.
(iv) If R is t-antisymmetric then R; is t-asymmetric.
Proof: Straightforward from the corresponding denitions.
2
For fuzzy ordering relations these results give the above-mentioned simple
connection between the re exive and the irre exive case. To state that result
we adopt here the following:
Denition 18.5.8. A fuzzy relation R 2 IF (X X ) is an irre exive fuzzy
partial ordering i R is an irre exive, t-transitive and t-asymmetric relation.
455
Proposition 18.5.7. For each fuzzy relation R and any left continuous t-
norm t one has:

(i) If R is a fuzzy partial ordering then R; is an irre exive fuzzy partial
ordering.
(ii) If R is an irre exive fuzzy partial ordering then R+ is a fuzzy partial
ordering.
The proof is by straightforward calculations and will not be given in detail.
But because of this result we conne ourselves to the case of re exive fuzzy
ordering relations.
Proposition 18.5.8. (i) If R; S are fuzzy partial orderings (preorderings)
then for each t-norm t the fuzzy relation R \t S too is a fuzzy partial ordering
(preordering).
(ii) Let t be left continuous. Consider any sequence (Ri )i< of type ,
any ordinal number, of fuzzy partial orderings (preorderings) such that for all
indices i; j < :
if i < j then j= Ri jt Rj :
Then the fuzzy relation
R =
i=1
Ri
is a fuzzy partial ordering (preordering) too.

Proof: (i) The re exivity of R \t S is obvious. For the t-transitivity we
have
[ x(R \t S )y ^t y(R \t S )z ]
= [ xRy ^t xSy ^t yRz ^t ySz ]
[ xRz ^t xSz ] = [ x(R \t S )z ]
by the t-transitivity of R and S , and thus the t-transitivity of R \t S results.
If both R; S are t-antisymmetric the t-antisymmetry of R \t S follows in the
same way.
(ii) Obviously R is re exive. Furthermore we have by the left continuity
of t for all x; y; z 2 X :
^t yRz
] = (sup [ xRi y] ) t (sup [ yRj z ] )
[ xRy
i<
i<
= sup [ xRi y ^t yRj z ]

i;j<
]
sup [ xRmaxfi;jg z ] = [ xRz
i;j<
and hence the t-transitivity of R. If all fuzzy relations Ri are t-antisymmetric
then the t-antisymmetry of R follows by corresponding calculations.
2
456
Proposition 18.5.9. If R is a fuzzy preordering then the fuzzy relation

Q =def f(x; y) k xRy ^t yRxg
(18.19)
is a fuzzy equivalence relation.

Proof: Re exivity and symmetry of Q are obvious by denition (18.19)
and the re exivity of R. And the t-transitivity of Q is a simple consequence
of the t-transitivity of R.
2
Considering now the (generalized) quotient structure X =Q one recognizes
that the strong analogy with the situation in the classical case breaks down:
in general the fuzzy equivalence relation R does not induce in X =Q a fuzzy
partial ordering.
Proposition 18.5.10. Let R be a fuzzy preordering and dene the fuzzy
equivalence relation Q as in (18.19). Then by
R^ =def f(haiQ ; hbiQ ) k aRbg
a fuzzy relation is dened in the quotient set X =Q which is a fuzzy preordering.
Proof: First we have to show that R^ is suitably dened, i.e. that this
denition does not really depend on the representatives a; b which describe
the fuzzy equivalence classes haiQ ; hbiQ .
Hence consider a; a0 such that haiQ = ha0 iQ . Then by [ a " haiQ ] = 1 one
has [ a " ha0iQ ] = 1, i.e. [ aQa0 ] = 1. Hence immediately [ aRa0 ] = [ a0 Ra] = 1.
Now one has for every x
[ aRx] = [ a0 Ra ^t aRx] [ a0 Rx] = [ aRa0 ^t a0 Rx] [ aRx]
by the t-transitivity of R, and thus [ aRx] = [ a0 Rx] . Therefore
[ haiQ R^ hbiQ ] = [ aRb] = [ a0 Rb] = [ ha0 iQ R^ hbiQ ] :
In the same way the independence of the choice of b is established.
The re exivity of R is obvious. And from the fact that one has for all
fuzzy equivalence classes haiQ ; hbiQ ; hciQ :
[ haiQ R^ hbiQ ^t hbiQ R^hciQ ] = [ aRb ^t bRc] [ aRc] = [ haiQ R^hciQ ] ;
the t-transitivity results.
2
What remains open here is the antisymmetry of R^ . Of course, one has
[ haiQ R^ hbiQ ^t hbiQ R^haiQ ] = [ aRb ^t bRa] ;
and thus, by Proposition 18.5.1(iii), for any left continuous t-norm t
j= haiQ t hbiQ $t aQb ;
i.e. in the present situation
457
j= haiQ t hbiQ $t aRb ^t bRa ;
and hence at all

[ haiQ R^ hbiQ ^t hbiQ R^haiQ ] = [ haiQ t hbiQ ] :
However, this is not the t-antisymmetry in the sense of Denition 18.5.1.
This means that in the setting of fuzzy equivalence classes it seems to
be natural to consider some (further generalized) identity
predicate which
: and
is dierent from the (rather simple) identity predicate =,
to adapt the
notion of antisymmetry to this \new" identity predicate.
If we assume that our universe of discourse X is already a class of fuzzy
sets, which means that our fuzzy subsets of X and our fuzzy relations in
X are fuzzy sets of higher level in the terminology of [209], then it seems
reasonable to discuss the modied version of antisymmetry of such a fuzzy
relation R, characterized through the denition
j= 8x8y(xRy ^t yRx !t x t y) :
If in that case we repeat the construction of R^ from Proposition 18.5.10 we
nd that this fuzzy relation R^ is antisymmetric in this new sense too: we
have
[ haiQ R^ hbiQ ^t hbiQ R^haiQ ] = [ aRb ^t bRa] [ haiQ t hbiQ ]
and hence
j= 8x8y(hxiQ R^ hyiQ ^t hyiQ R^hxiQ !t hxiQ t hyiQ ) :
To illustrate the problem with the actual version of antisymmetry with a
further result, following [71, 72], we show that fuzzy linear orderings are crisp.
For this, besides the denition of linearity for the fuzzy case, two additional
notions are necessary.
Denition 18.5.9. Suppose that is a crisp, and R a fuzzy (binary) relation in X. Then R is -consistent i one has for all x; y; z 2 X :
y z ) [ xRy] [ xRz ] :
And R is a fuzzication of i one has for all x; y 2 X :
x y ) [ xRy] = 1 :
Proposition 18.5.11. Let be a crisp linear ordering in X, and R a stlinear fuzzy ordering. If R is a fuzzication of then R coincides with in
the sense that the membership function of R is the characteristic function of
.
458
Proof: The re exivity of R together with the -consistency of R give

[ xRy] = 1 for all x; y 2 X with x y. But then the t-antisymmetry and the
st -linearity of R imply that one has [ xRy] = 0 for all x; y 2 X with x 6 y. 2
A way out of this problem of antisymmetry is oered, following [71],
by an approach which combines fuzzy preordering relations with a kind
of
: but
generalized antisymmetry which relates not to the identity relation =,
instead to a truly many-valued version of standard equality { now understood
as any fuzzy equivalence relation in the sense of Denition 18.5.2.
To get things working one has, additionally to the notion of antisymmetry,
also to adapt the notion of re exivity in a suitable way.
Denition 18.5.10. Suppose that is a fuzzy equivalence relation, and that
R is a fuzzy relation. Then R is (with respect to a given t-norm t)
(i) -re exive (in X )
i j= 8x(x x !t xRx) ;
(ii) (t; )-antisymmetric i j= 8x8y(xRy ^t yRx !t x y) .
Of course, the
of re exivity
and t-antisymmetry are just
: previous notions
:
the notions of =-re exivity
and of (t; =)-antisymmetry.
Denition 18.5.11. A fuzzy relation R is a (re exive) fuzzy partial ordering
w.r.t. some fuzzy equivalence relation i R is a t-transitive relation which
is -re exive as well as (t; )-antisymmetric.
For this generalized type of fuzzy partial ordering the previously unattainable result becomes provable which related to each fuzzy preordering a fuzzy
partial ordering, simply by chosing the \right" equivalence relation.
To state the results not in a too narrow sense, also an additional notion
for t-norms is needed.
Denition 18.5.12. A t-norm t2 dominates a t-norm t1 i one has for all
u1 ; u2 ; u3; u4 2 [0; 1] the relationship
(u1 t2 u2 ) t1 (u3 t2 u4 ) (u1 t1 u3) t2 (u2 t1 u4) :
(18.20)
Obviously each t-norm dominates itself, and tG = min dominates any
other t-norm. We shall not study this relation of dominance between t-norms
here any further, but mention only an easy consequence which shall be needed
later on.
Corollary 18.5.3. Suppose that the t-norm t2 dominates the t-norm t1, then
one has t1 6 t2 .
Proof: Choose in (18.20) particularly u2 = u3 = 1 then this gives, with
x for u1 and y for u4 , x t1 y x t2 y for all x; y.
2
459
Proposition 18.5.12. Let R 2 IF (X X ) be a fuzzy preordering and t
some t-norm which dominates the t-norm t involved in the t-transitivity condition for R. Then R is a fuzzy partial ordering w.r.t. the equivalence relation
R dened for all x; y 2 X by
x R y =def xRy ^t yRx :
(18.21)
Proof: By denition, R is symmetric, and also R-re exive. So it rst
remains to discuss the t-transitivity of R . But one has
[ x R y ^t y R z ] = [ xRy(^t yRx) ^t (yRz ^t zRy)]]
[ xRy(^t yRz ) ^t (zRy ^t yRx)]]
[ xRz ^t zRx] [ x R z ]
using the dominance assumption and the t-transitivity of R.
Hence R is a fuzzy equivalence relation. Because R is supposed to be
t-transitive, the R-re exivity and the (t; R)-antisymmetry of R remain to
be shown. The R -re exivity, however, is obvious from the denition of R .
And from
[ xRy ^t yRx] [ xRy ^t yRx] = [ x R y]
the (t; R )-antisymmetry of R follows immediately.
2
Caused by the fact that one refers in (18.21) to some t-norm t which
dominates the t-norm t which is the basic one for the t-transitivity, one has
a whole variety of fuzzy equivalence relations which can be referred to for
the discussions on antisymmetry. Fortunately one has some simple bounds
for the equivalence relations of this variety.
Proposition 18.5.13. Let R be a fuzzy preordering and some fuzzy equivalence relation, which both are t-transitive. If R is a fuzzy partial ordering
w.r.t. the equivalence relation then one has for all x; y 2 [0; 1]
[ xRy ^t yRx] [ x y] minf[ xRy] ; [ yRx] g :
(18.22)
Proof: From the (t; )-antisymmetry of R one immediately has the inequality [ xRy ^t yRx] [ x y] . And from the -re exivity of R and the
symmetry of the other inequality [ x y] minf[ xRy] ; [ yRx] g results. 2
In some particular cases this leads even to a uniquely determined fuzzy
equivalence relation. To state the result in not too restricted a form, the
following additional notion shall be used.
Denition 18.5.13. Given any t-conorm st we denote a fuzzy relation R
as st -linear i
j= 8x8y(xRy _t yRx) ;
and we denote R as weakly linear i supp(R) is a linear relation in the usual
sense.
1
460
Proposition 18.5.14. Let R be a fuzzy preordering and either t-transitive

for t = min, or max-linear. Then there exists a unique fuzzy equivalence
relation 0 such that R is a fuzzy partial ordering w.r.t. 0 .
Proof: If the fuzzy preordering R is either t-transitive for t = min, or
max-linear, then in (18.22) the left hand side and the right hand side of the
chain of inequalities coincide, which means that the fuzzy equivalence relation
is uniquely determined.
2
18.6 Graded Properties of Fuzzy Relations

The properties of fuzzy relations which were introduced in the previous Definition 18.5.1, cf. p. 445, are crisp in the sense that a fuzzy relation R either
has one of those properties or does not have that property. Therefore the
situation here is comparable with that we had before with the Zadeh's crisp
inclusion (18.6) for fuzzy sets. And as in that case we were interested in
generalizing that inclusion to a graded one by introducing the fuzzied
inclusions jt in Denition 18.2.3, we now again look for possibilities to generalize the relation properties of Denition 18.5.1 in such a way that fuzzied,
i.e. graded properties result.
The fact that we always intended to write down our notions in the language of (a suitable system of) many-valued logic now opens the door to
get such graded properties in a canonical manner out of those ones introduced in Denition 18.5.1. The crucial point is that Denition 18.5.1 used
the logical validity of suitable ws to dene the relation properties. Therefore
deleting this demand of logical validity and retaining the characterizing ws
immediately opens a way to get graded properties of the type we are looking
for.
In the formal sense these graded properties themselves are to be represented by many-valued predicates of our language of many-valued logic.
Denition 18.6.1. For all fuzzy relations R and left continuous t-norms t
let
Re (R) =def 8x(xRx) ;
Irre t (R) =def 8x ;t (xRx) ;
Transt (R) =def 8x8y8z (xRy ^t yRz !t xRz ) ;
Symmt (R) =def 8x8y(xRy !t yRx) ;
Antisymmt (R) =def 8x8y(xRy ^t yRx !t x =: y) ;
Asymmt (R) =def 8x8y ;t (xRy ^t yRx) :
To get a simpler approach, in comparison with Denition 18.5.1, we here
dispense with the use of a separate negation function n to dene the predicates Irre t and Asymmt . On the other hand this change from crisp to graded
461
properties also causes a (seeming) dependency on the involved t-norm for the
properties of irre exivity and symmetry.
Immediately one has e.g. the result that
j= Transt (R) , R t-transitive ;
and the corresponding results for all the other graded properties with respect
to their crisp versions.
As in the case of the crisp versions of relation properties, a characterization of these graded properties with genuinely set theoretic notions is also
possible.
Proposition 18.6.1. For each fuzzy relation R it holds true that
j= Re (R) $ X j R ;
j= Irre t (R) $ R \t X t ; ;
j= Transt (R) $ R t R jt R ;
j= Symmt (R) $ R jt R;1 ;
j= Antisymmt (R) $ R \t R;1 jt X ;
j= Asymmt (R) $ R \t R;1 t ; :
Proof: Straightforward from the corresponding denitions, using also
Proposition 18.2.4(i) for the cases of the predicates Irre t and Asymmt . 2
In general, this \fuzzication" of properties of fuzzy relations is not completely new. Previously we essentially did the same with Denition 18.2.3
which introduced graded versions jt and t of inclusion and equality for
fuzzy sets. The small dierence is only that there the generalization to graded
relations concerned binary relations, i.e. binary predicates for fuzzy sets, and
now we are concerned with generalized properties, i.e. unary predicates for
fuzzy relations.
As a relatively simple and very well-known example of a special type of
fuzzy relations, which are not completely elementary, let us take a look at
fuzzy ordering relations which { for the case of crisp properties { we already
discussed in the previous section. Now a graded notion of fuzzy partial
ordering shall be introduced. First we focus on the re exive case.
Denition 18.6.2. Let R 2 IF (X X ) be a fuzzy relation and t any left
continuous t-norm. Put
FPOt (R) =def Re (R) ^t Transt (R) ^t Antisymmt (R) :
Obviously, by the left continuity of t and basic properties of t-norms one has
j= FPOt (R) , R fuzzy partial ordering :
Corollary 18.6.1. Let R 2 IF (X X ) be a fuzzy relation and t a left continuous t-norm. Then
j= FPOt (R) $ X jt R ^t R t R jt R ^t R \t R;1 t ; :
462
Proof: Obvious by Proposition 18.6.1.
Let us rst consider the usual duality between re exive and irre exive
ordering relations established by changing from R to R+ = R [t X , and
from R to R; = R nt X = f(x; y) k (x; y) " R ^t ;t (x; y) " X g, i.e. to
R; = R \t CCX , cf. Denition 18.5.7 and Proposition 18.5.7.
Proposition 18.6.2. Let R 2 IF (X X ) be a fuzzy relation and t any left
continuous t-norm. Then one has
(i) j= Re (R [t X ) ;
(ii) j= Irre t (R nt X ) ;
(iii) j= Transt (R) !t Transt (R [t X ) ;
(iv) j= Transt (R nt X ) !t Transt (R) :
Proof: (i) and (ii) are obvious. For (iii) one has to prove the inequality
[ 8x; y; z (xRy ^t yRz !t xRz )]]
[ 8x; y; z (xR+y ^t yR+ z !t xR+ z )]]
for R+ = R [t X . To get this we rst show the inequality
[ aRb ^t bRc !t aRc] [ aR+b ^t bR+c !t aR+ c]
(18.23)
for arbitrary a; b; c 2 X .
In the case that a = b holds true it remains to be shown that
[ aRa ^t aRc !t aRc] [ aR+ a ^t aR+c !t aR+ c] :
But this holds because both of these implications have truth degree 1: in
both of them the truth degree of the antecedent is not greater than the truth
degree of the succedent.
In the case b = c, essentially the same argument gives (18.23). Thus the
case a 6= b 6= c remains to be considered. Now, however, one has
[ aRb] = [ aR+b] ;
[ bRc] = [ bR+c] ;
[ aRc] [ aR+c] ;
and thus again (18.23). Therefore (18.23) holds true. But then also
[ 8x; y; z (xRy ^t yRz !t xRz )]] [ aR+b ^t bR+c !t aR+c]
holds. Here the left-hand side of the inequality is independent of the choice
of a; b; c. That means that even the inmum (with respect to a; b; c) of the
right-hand side is not smaller than the left-hand side, but this is exactly (iii)
according to Denition 18.6.1.
Finally (iv) follows along the same line of argument as claim (iii).
2
The transitivity property cannot be transferred from R to R nt X in a
manner directly similar to the last claim (iii). However, that transfer becomes
possible with an additional assumption.
463

j= Transt (R) ^t Antisymmt (R) ! Transt (R nt X ) :
Proof: As in part (iii) of the last proof it is enough to show that for
R; = R nt X = R \t C X and all a; b; c 2 X that
[ Transt (R) ^t Antisymmt (R)]] [ aR; b ^t bR; c !t aR; c] : (18.24)
If c 6= a holds true, then [ aR;c] = [ aRc] . By j= R; jt R hence
[ aRb ^t bRc !t aRc] [ aR;b ^t bR;c !t aRc] ;
and thus even
[ Transt (R)]] [ aR;b ^t bR;c !t aRc] :
So (18.24) is established in this case.
Now assume c = a. Then [ aR; c] = 0 and (18.24) becomes
[ Transt (R) ^t Antisymmt (R)]] [ ;t (aR; b ^t bR;a)]] :
(18.25)
In the case a = b one has [ aR; b] = 0, and thus [ ;t (aR; b ^t bR;:a)]] = 1,
which immediately gives (18.25). And in the case a 6= b one has [ a = b] = 0
and thus
[ ;t (aR; b ^t bR; a)]] = [ aR; b ^t bR; a !t a =: b] ;
hence
[ Antisymmt (R)]] [ aR; b ^t bR; a !t a =: b]
and therefore (18.25). Thus (18.24) holds true in any case.
2
Similar transfer results hold true regarding antisymmetry and asymmetry.
(i) j= Antisymmt (R) ! Asymmt (R nt X ) ;
(ii) j= Asymmt (R nt X ) $ Antisymmt (R nt X ) ;
(iii) j= Asymmt (R) ! Antisymmt (R [t X ) :
Proof: The same type of estimations as in the proofs of the last two
propositions yield the results.
2
By and large, the basic links for a generalization of the usual duality of
re exive and irre exive partial orderings seem to be established with those
results. What remains to be given is a denition of graded irre exive fuzzy
partial orderings. Caused by the process of grading, this denition is not as
obvious as in the re exive case. The background diculty is the well-known
fact for classical irre exive partial orderings, that they can either be dened
464
by the requirements of irre exivity and transitivity or { equivalently { by

those of irre exivity, transitivity, and asymmetry.
In the present situation therefore we are rst interested whether transitivity together with irre exivity also implies asymmetry in our more general
context. Indeed we have the following:
continuous t-norm. Then it holds true that
j= Transt (R) ^t Irre t (R) ! Asymmt (R) :
Proof: As in the previous proofs it is enough to prove for any a; b 2 X
j= ;t aRa ^t 8z (aRb ^t bRz !t aRz ) !t (aRb !t ;t bRa) : (18.26)
Using Proposition 13.1.8(i) twice and also Proposition 13.2.2(i), one nds
that (18.26) is equivalent to
j= ;t aRa ^t aRb ^t (aRb !t 8z (bRz !t aRz )) !t ;t bRa :
But now using Propositions 13.1.6(i) and 13.1.12(iv) together with the antitonicity (of the truth degree function) of the R-implication !t in the rst
argument, one has for the truth degree of this formula
[ ;t aRa ^t aRb ^t (aRb !t 8z (bRz !t aRz )) !t ;t bRa]
[ ;t aRa ^t 8z (bRz !t aRz ) !t ;t bRa]
= [ 8z (bRz !t aRz ) !t (;t aRa !t ;t bRa)]]
[ 8z (bRz !t aRz ) !t (bRa !t aRa)]]
=1
and thus (18.26).
2
Therefore we decide to base irre exive partial orderings only on transitivity and irre exivity.
continuous t-norm. Then put
FPOt (R) =def Transt (R) ^t Irre t (R) :
With this denition all the notions and preliminary results are completed,
which we need to state the generalized connection between the re exive and
irre exive graded fuzzy partial ordering relations.
continuous t-norm. Then there hold true
(i) j= Transt (R) ^t Antisymmt (R) ! FPOt (R nt X ) ;
(ii) j= FPOt (R) ! FPOt (R nt X ) ;
(iii) j= FPOt (R) ^t Asymmt (R) ! FPOt (R [t X ) :
465
Proof: Because of Proposition 18.6.2(ii), claim (i) is a direct consequence

of Proposition 18.6.3. Of course, (ii) is a simple corollary of (i). And (iii)
results from Proposition 18.6.2(i) together with Proposition 18.6.2(iii) and
Proposition 18.6.4(iii).
2
The last claim of this proposition could be read as indicating that Definition 18.6.3 was the wrong choice. But considering instead of FPOt the
many-valued predicate
FPO
(18.27)
t (R) =def Transt (R) ^t Irre t (R) ^t Asymmt (R)
one indeed would be able to prove
j= FPO
t (R) ! FPOt (R [t X ) ;
yet instead of Proposition 18.6.6(ii) then only
j= FPOt (R) ^t Antisymmt (R) ! FPO
t (R nt X )
would be provable.
Altogether that means that neither Denition 18.6.3 nor (18.27) gives the
simple results one has for crisp partial orderings. Only a restriction of the tnorm under consideration allows for a simplication of Proposition 18.6.6(ii).
Corollary 18.6.2. For each fuzzy relation R 2 IF (X X ) and the t-norm
t = tG = min one has
j= FPOtG (R) ! FPOtG (R [ X ) :
Proof: In this particular case one has
[ FPOtG (R)]] = [ FPOtG (R) ^ FPOtG (R)]]
[ FPOtG (R) ^ AsymmtG (R)]] ;
and thus the result immediately follows from Proposition 18.6.6(ii).
2
Besides the ordering relations the equivalence relations are of fundamental
importance in crisp mathematics. Like ordering relations we already introduced fuzzy equivalence relations in the last section, cf. Denition 18.5.2.
Here we intend to have a graded property of being an equivalence relation.
continuous t-norm. Then dene
Eqrelt (R) =def Re (R) ^t Symmt (R) ^t Transt (R) :
We do not intend to present here an extended theory of graded fuzzy
equivalence relations. That is to the same degree an open problem as a more
elaborated theory of graded fuzzy orderings. But we shall consider one topic,
the fuzzication of the well-known fact that any two distinct equivalence
classes of a (crisp) equivalence relation are disjoint sets. There was a rst
type of fuzzication, for fuzzy relations which \really", i.e. to truth degree 1,
466
are fuzzy equivalence relations, presented in Proposition 18.5.1(iv), a result

that can equivalently be written as
j= haiR \t hbiR t ; $ ;t aRb
or as
j= haiR \t hbiR 6t ; $ aRb :
Further partial fuzzications of this result read as follows.
continuous t-norm. Then there hold true for any a; b 2 X
(i) j= Transt (R) ! (aRb !t haiR jt hbiR ) ;
(ii) j= Re (R) ! (haiR jt hbiR !t aRb) ;
(iii) j= Transt (R) ^t Symmt (R) ! (haiR \t hbiR 6t ; !t aRb) :
Proof: (i) By the corresponding denitions and the Propositions 13.1.8(i)
and 13.1.6(i), one has the following inequalities for truth degrees:
[ Transt (R) ^t aRb] = [ 8x; y; z (xRy ^t yRz !t xRz ) ^t aRb]
[ 8x(aRb ^t (xRa ^t aRb !t xRb)]]
= [ 8x(aRb^t )(aRbtot (xRa !t xRb)))]]
[ 8x(xRa !t xRb))]]
= [ haiR jt hbiR ] :
Hence (i) immediately follows, again using Proposition 13.1.8(i).
(ii) With essentially the same logical background as in the foregoing proof
of claim (i) one gets (ii) by using Proposition 13.1.8(i) together with
[ Re (R) ^t haiR jt hbiR ] [ aRa ^t (aRa !t aRb)]] [ aRb] :
(iii) According to Proposition 18.2.4(ii) one has
[ haiR \t hbiR 6t ;] = [ 9x(x " haiR \t hbiR )]]
= [ 9x(xRa ^t xRb)]] :
Therefore with Propositions 13.2.4(i) and 13.1.6(i) one gets
[ Transt (R) ^t Symmt (R) ^t 9x(xRa ^t xRb)]]
= [ 9x(Transt (R) ^t Symmt (R) ^t xRa ^t xRb)]]
[ 9x(Transt (R) ^t (xRa !t aRx) ^t xRa ^t xRb)]]
[ 9x(Transt (R) ^t aRx ^t xRb)]]
[ 9x((aRx ^t xRb !t aRb) ^t aRx ^t xRb)]]
[ 9x(aRb)]] = [ aRb] :
Now once again an application of Proposition 13.1.8(i) gives the result. 2
467
Corollary 18.6.3. Under the same assumptions as in Proposition 18.6.7

one has
(i)
(ii)
(iii)
(iv)
j= Re (R) ^ Transt (R) ! (haiR jt hbiR $t aRb) ;

j= Transt (R) ^t Symmt (R) ^t ;t aRb ! haiR \t hbiR t ; ;
j= [Transt (R)]2 ! (aRb ^t bRa !t haiR t hbiR ) ;
j= Transt (R) ! (aRb ^ bRa !t haiR jt hbiR ^ hbiR jt haiR ) :
Proof: Claim (i) results via Proposition 13.1.8(iv) from Proposition

18.6.7(i) and 18.6.7(ii), using that [ H1 $t H2 ] = [ (H1 !t H2 )^(H2 !t H1 )]]
always holds.
(ii) We start from the succedent of the implication in the previous Proposition 18.6.7(iii) and have by Propositions 13.1.11(i) and 13.1.8(ii)
j= (haiR \t hbiR 6t ; !t aRb)
! (;t (haiR \t hbiR t ;) !t ;t ;t aRb) ;
and by Proposition 13.1.12(iii) also
j= (;t (haiR \t hbiR t ;) !t ;t ;t aRb)
! (;t aRb !t ;t ;t (haiR \t hbiR t ;)) :
Thus from Propositions 18.2.4(ii) and 18.6.7(iii) by an iterated application
of (the inference rule corresponding to) Proposition 13.1.8(ii) one nds
j= Transt (R) ^t Symmt (R) ! ;t ;t ;t 9x(x " haiR \t hbiR ) :
With Proposition 13.1.11(ii) that means
j= Transt (R) ^t Symmt (R) ! ;t 9x(x " haiR \t hbiR ) ;
which gives nally
j= Transt (R) ^t Symmt (R) ! haiR \t hbiR t ;
by again applying Proposition 18.2.4(ii).
(iii) Let H (a; b) be the formula aRb !t haiR jt hbiR . Taking the formula
Transt (R) ! H (a; b), logically valid by Proposition 18.6.7(i), together with
the formula Transt (R) ! H (b; a) which results from the earlier one by
interchanging a and b, gives via Proposition 13.1.8(iii)
j= [Transt (R)]2 ! (H (a; b) ^t H (b; a)) :
(18.28)
Now both formulas H (a; b) and H (b; a) themselves are implications. Thus
again applying Proposition 13.1.8(iii) to the succedent of (18.28) nally gives
claim (iii) according to the denition of t .
Claim (iv) follows by the same arguments, but referring to Proposition
13.1.8(iv) instead of Proposition 13.1.8(iii), and thus having
j= Transt (R) ! (H (a; b) ^ H (b; a))
instead of (18.28).
2
For the t-norm tG = et1 = min even more becomes provable.
468
Corollary 18.6.4. For each fuzzy relation R 2 IF (X X ) and all a; b 2 X

there hold true
(i) j= TranstG (R) ! (aRb !tG haiR tG hbiR )) ;
(ii) j= TranstG (R) ^ SymmtG !
(haiR \ hbiR tG haiR \ fx k aRbg) :
Proof: Claim (i) results immediately from Corollary 18.6.3(ii) by the
idempotency of the t-norm tG .
(ii) With regard to Proposition 18.6.7(iii) we have the following sequence
of transformations of the truth degree of the respective succedent:
[ haiR \ hbiR 6tG ; !tG aRb]
= [ 9x(x " haiR \ hbiR ) !tG aRb]
= [ 8x(x " haiR ^ hbiR !tG aRb)]]
= [ 8x(x " haiR ^ hbiR !tG aRb ^ x " haiR )]]
whose last transformation rests on simple properties of the min-operator tG
and its corresponding R-implication operator.
Using the fact that for each x
[ aRb] = [ x " fy k aRbg] ;
(18.29)
and changing the bound variable y of the generalized class term fy k aRbg
into x gives
[ haiR \ hbiR 6tG ; !tG aRb] = [ haiR \ hbiR jtG haiR \ fx k aRbg] ;
and thus altogether
j= TranstG (R) ^ Symm; tG (R) !

haiR \ hbiR jtG haiR \ fx k aRbg :
(18.30)
On the other hand Proposition 18.6.7(i) gives the following sequence of transformations of the truth degree of the respective succedent:
[ aRb !tG haiR jtG hbiR ]
= [ aRb !tG 8x(x " haiR !tG x " hbiR )]]
= [ 8x(aRb ^ x " haiR !tG x " hbiR )]]
= [ 8x(aRb ^ x " haiR !tG x " haiR ^ x " hbiR )]] ;
and thus, again using (18.29),
j= TranstG (R) ! (haiR \ fx k aRbg jtG haiR \ hbiR ) :
(18.31)
Combining (18.30) and (18.31) via Proposition 13.1.8(iv) gives the result. 2
In the previous section, we discussed besides fuzzy partial orderings and
fuzzy equivalence relations also the fuzzy transitive hull of a fuzzy relation. So
it remains here to consider also those results on t-transitive hulls which were
469
not presented in a \truly fuzzied" form. This obviously concerns Proposition
18.5.4 and Proposition 18.5.3(ii).
Therefore we generalize the last mentioned result too. We immediately
have
j= R t Tt (R) $ Tt (R) jt R
because of Proposition 18.5.3 (i). Furthermore it was proved
j= Re (R) ! X jt R ;
j= > ! R jt R ;
j= Transt (R) ! R2 jt R :
The second result, however, was previously given as j= R jt R. Here we use
additionally the truth degree constant > . As a result we have by Proposition
13.1.8(iv)
j= Re (R) ^ > ^ Transt (R) ! (X jt R ^ R jt R ^ R2 jt R) ;
and hence because of Proposition 18.2.2(i)
j= Re (R) ^ Transt (R) !
2
[
i=0
R i jt R :
(18.32)
From R3 = R2 t R according to Denition 18.5.5 together with Proposition

18.3.1(i), i.e. the monotonicity of t , we nd
j= Transt (R) ! R3 jt R2
and thus
j= [Transt (R)]2 ! R3 jt R :
(18.33)
Unfortunately this antecedent [Transt (R)]2 cannot be simplied in such a
way that (18.33) holds true with Transt (R) instead of [Transt (R)]2 . Thus
we only get from (18.32) and (18.33)
j= Re (R) ^ [Transt (R)]2 !
3
[
i=0
Ri jt R :
(18.34)
using Proposition 13.1.8(iv) and the fact that

[ Transt (R) ^ [Transt (R)]2 ] = [ [Transt (R)]2 ] :
Of course, this transition from (18.32) to (18.34) can be iterated using
the fact that (18.33) holds true more generally in the form
j= [Transt (R)]n ! Rn+1 jt R :
Therefore one has for each integer n > 1
j= Re (R) ^ [Transt (R)]n !
n[
+1
i=0
Ri jt R :
(18.35)
470
To nally nd the result for Tt (R) we are looking for, we use the symbolic
1
Q
expression [Transt (R)]n with the truth degree10
[
1
Y
i=0
i=0
n
[Transt (R)]n ] =def nlim
!1 [ [Transt (R)] ]
and get via (18.35) according to Proposition 18.2.6(iv) that

1
Y
j= Re (R) ^ [Transt (R)]n ! Tt (R) jt R

and thus
i=0
1
Y
j= Re (R) ^ [Transt (R)]n ! Tt (R) t R :

i=0
1
Q
Unfortunately, caused by the fact that the expression [Transt (R)]n api=0
pears as part of the antecedent, this result seems to be quite weak. Only
in the special case t = et1 does this innite iteration of the t-conjunction
disappear, giving
j= Re (R) ^ Transet (R) ! Tet (R) et R :
1
10 The existence of this limit is not a problem because it is the limit of a non-
increasing sequence of reals from [0; 1], and thus exists.
19. Fuzzy Logic
The present state of development of the eld of fuzzy logic forces one to
start with a terminological side remark concerning an essential ambiguity
in the use of the terminus fuzzy logic. There is a widespread use, mainly
originating from developments in the engineering elds of automatic control
and of knowledge engineering, to understand by the term fuzzy logic any
topic which involves fuzzy sets or some kind of \approximate" reasoning.
Actually, this understanding of the words \fuzzy logic", which is not tied
with the core ideas of formal logic, often is referred to as: fuzzy logic in the
wider sense. This understanding of the terminus \fuzzy logic" supersedes an
older one which understood fuzzy logic to mean any system of many-valued
logic related to fuzzy sets.
Contrary to this usage, fuzzy logic in the narrow sense of the word refers
to a particular extension of many-valued logic, an extension which allows for
fuzzy sets of premises from which (graded) conclusions may be drawn.
It is this usage in the narrow sense which is our topic in the following
chapter. For more details, the reader should consult the books [246, 399].
As systems of logic, both many-valued as well as fuzzy logic, are based on
some suitable formalized language, and have a syntactic as well as a semantic
aspect.
The formal language L which we intend to use shall, as before, be some
language of a suitable system of rst-order many-valued logic. Their particular connectives and quantiers as well as their truth degree constants shall
become specied only if necessary. As we suppose that the membership degrees of the fuzzy sets which shall enter the discussion shall be truth degrees,
this language shall have enough expressive power to handle the situation we
have in mind.
Often we will restrict our considerations to the sentences of L , i.e. to the
ws without free variables. Their set shall be denoted by L0 .
The semantic considerations for fuzzy logic are, like those ones for (usual)
many-valued logic, based (i) in the propositional case on suitable valuations
v : V0 ! W which map the set of propositional variables V0 into the set
W of? membership degrees, and are completed by an entailment operation
CnS : IF (L ) ! IF (L ) over the fuzzy subsets (with membership degrees
in W ) of the class L of ws. And they are (ii) based in the rst order
472
19. Fuzzy Logic
case on interpretations A which map individual constants into the universe

of discourse jAj, and predicate letters (of arity n) to (n-ary) fuzzy relations
in jAj, and which are completed again by an entailment operation CnS ? :
IF (L ) ! IF (L ) over the fuzzy subsets (with membership degrees in W ) of
the class of ws.
The syntactic considerations for many-valued as well as fuzzy logics are
based on suitable calculi xed by their respective (crisp or fuzzy ) sets of
axioms and their crisp sets of inference rules. Inferences in any case are nite
sequences. For fuzzy logic, however, each inference rule R with k premises
splits into two parts R = (R1 ; R2) such that R1 is a (partial) k-ary mapping
from L into L , and R2 is a k-ary mapping from W into W . The idea is
that R2 associates with the degrees, to which actual premises of R are given,
a degree to which the actual conclusion of R is given.
In any case, the main goal of those calculi is the axiomatization of the set
of logically valid formulas, or the axiomatization of the semantic entailment
relation, i.e. of the logical part of elementary theories.
Let us rst look back at many-valued logic. Of course, not every inference
schema is accepted as an inference rule: a necessary restriction is that to
sound inference schemata, i.e. to inference schemata which led \from true
premises not to a false conclusion".
For fuzzy logic a comparable soundness condition is assumed, called \infallibility" in [73] instead of \soundness" as e.g. in [246, 396, 398, 399].
19.1 Many-Valued Logic with Graded Consequences

Fuzzy logic extends the idea of degrees of truth, which is the constitutive one
for many-valued logic, in the direction of assuming that the premises from
which a derivation is going to start, need to be given only \partially", i.e.
only to some degree. This degree, of course, should always be one of the truth
degrees of the particular system of many-valued logic under consideration.
Having in mind the usual treatment of the situation that one derives
formulas from some given set of premises, the common understanding
in particular applications is that this set should only comprehend the
fundamental principles which are of basic importance for this application
{ and should not contain \redundant information" in the sense of (purely)
logically valid sentences. To underline this situation, one says that by some
(formalized) theory 1 T is determined and that is the (or: some possible)
set of nonlogical axioms of this theory T .
1 In the standard case where the system of logic one is referring to is the classical
rst-order logic, such formalized theories usually are called elementary theories.
We shall extend this terminology and use this terminus \elementary theory" also
in the case the actual logical system is some system of many-valued rst-order
logic.
19.2 The Semantic Approach
473
The extension we actually are interested in now amounts to consider elementary theories with fuzzy sets of nonlogical axioms. Such theories shall
be called fuzzy theories. And the part of many-valued logic devoted to the
examination of such fuzzy theories is the fuzzy logic (in the narrow sense).
In fuzzy logic, like in the standard approach toward many-valued logic
and logical entailment, there is the duality of the semantic approach via interpretations (models), the satisfaction relation and the notion of semantical
entailment, and of the syntactic approach via rules of inference and a suitable
notion of proof or derivation.
19.2 The Semantic Approach

As in the case of (pure) many-valued logic, let us rst consider the semantic
approach. Because the language for fuzzy logic is the same as for many-valued
logic, the notion of interpretation is the same as before: based on xed truth
degree functions for the connectives and the quantiers, as well as on a xed
correspondence between the actual truth degree constants of the language
and the truth degrees denoted by them, any interpretation has to provide
the meanings of the predicate symbols and of the individual constants.
The essential point for fuzzy logic thus, regarding the semantic approach
toward the entailment relation, is the denition of what shall be a model A
for a fuzzy set of sentences. Intuitively, the crucial condition (for classical
logic) reads
if H is a sentence of then A is a model of H .
(19.1)
Because now is a fuzzy set of sentences the antecedent of this condition
(19.1) either has to be read as the graded formula H " { or as some statement involving the truth degree [ H " ] of it. The succedent of (19.1), however, is a classical statement, i.e. either true or false.
Thus, we either should transform H " into a classical statement, or we
should transform \A is a model of H " into some graded formula.
We shall follow the second one of these ideas guided by the intuition that,
classically, the truth value of \A is a model of H " is the same as the truth
value of H if evaluated in the interpretation2 A. Writing j= (A; H ) for the
graded version of \A is a model of H " it is natural to dene
[ j= (A; H )]] =def ValSA (H )
(19.2)
in accordance with the notation introduced in (3.15). Then (19.1) becomes
the condition
H " ! j= (A; H )
(19.3)
2 We remind the reader that we suppose that is a (fuzzy) set of sentences
which means that the truth degree of H in A is independent of any further
A-assignment.
474
19. Fuzzy Logic
which has to be \satised", i.e. which has to have truth degree 1, and in
which ! now has to be understood as a suitable implication connective of
many-valued logic.
Having in mind that the degree ranking property is one of the basic properties of implication connectives for systems of many-valued logic, which is
satised e.g. for all R-implications based upon left continuous t-norms, condition (19.3), and hence also (19.1), becomes
[ H " ] [ j= (A; H )]] :
(19.4)
Therefore we give the following:
Denition 19.2.1. An interpretation A for some system S of many-valued
logic is a model of a fuzzy set of sentences of LS , denoted: A j= , i
condition (19:4) is satised for each LS -sentence H .
This denition opens the way to dene the fuzzy set Cn ( ) of all the sentences which are logically entailed by . In classical logic the corresponding
set Cn( ) may be characterized as
\
Cn( ) = fH j A j= H for all models A of g = fTh(A) j A j= g ;
using the notion Th(A) = fH j A j= H g of the theory of a structure A.
For the present situation we only have to adapt this notion of the theory
of a structure in such a way that it becomes a fuzzy set of formulas. And this
is done simply by dening for any interpretation A the fuzzy theory of A as
the fuzzy set of sentences
Th (A) =def fH 2 L0 k j= (A; H )g :
(19.5)
The notion of a model of a fuzzy set of sentences, i.e. A j= , has already
been dened in Section 3.4, there denoted as a model of a graded set of
sentences. However, the graded sets of Section 3.4 obviously are just fuzzy
sets.
Then, similar to the classical case, one denes furthermore the fuzzy set
of all the sentences entailed by a fuzzy set of sentences as
\
Cnj=( ) =def fTh (A) j A j= g
(19.6)
T
with here for the intersection of a (crisp) class of fuzzy sets.
Of course, the degree to which some sentence H is entailed by now is
the membership degree [ H " Cn ( )]], and one obviously has
[ H " Cnj= ( )]] = inf f[ j= (A; H )]] j A j= g :
(19.7)
Further semantic notions, like e.g. the notion of satisability, then can be
introduced along the lines well known from standard many-valued logic. Suitable generalizations of standard results become provable in this more general
setting. As an example one has the following (propositional) compactness
theorem, cf. [548], which holds true for all propositional fuzzy logics with an
injective MV-algebra as set of truth degrees.
19.3 The Syntactic Approach
475
Theorem 19.2.1. A fuzzy set of well-formed (propositional) formulas is

satisable i each nite fuzzy subset3 of is satisable.
As a side remark it should be mentioned that the approach toward the
semantic basis of fuzzy logic which we consider here is a purely mathematical, particularly set theoretical one. There exist also other, more real-world
related approaches toward such a semantic basis, as e.g. in [188, 412], but not
one of them has up to now been commonly accepted as intuitively convincing.
19.3 The Syntactic Approach

The syntactic approach toward many-valued logic with graded consequences,
i.e. toward fuzzy logic (in the narrow sense) has to refer to a suitable notion
of derivation which allows one to derive consequences from fuzzy sets of
premises, and which as usual has to be based on a suitable calculus, i.e.
some algorithmic system with a rule base and some { possibly fuzzy { set
AxFL of axioms, for fuzzy logic.
As is common usage, also in fuzzy logic a derivation shall be a nite sequence of well-formed formulas. But the intention to get the derived formula
of every derivation with some degree forces to combine each derivation with
some degree. Because of the usual understanding that each initial segment
of a derivation again is a derivation (of its last formula), each initial segment
of a derivation in fuzzy logic itself should come with a degree. And the most
simple way to reach this goal seems to be a parallel treatment of formulas
and degrees in derivations in fuzzy logic. Formally this means that the constituents of derivations in fuzzy logic should be ordered pairs of the form
(formula, degree ).
Fortunately this ts very well with the additional assumption that the
sets of premises now should be fuzzy sets of sentences, because fuzzy sets
of formulas { or their membership functions { according to the usual settheoretic understanding of functions are nothing but (crisp) sets of ordered
pairs of just the type (formula, membership degree ), i.e. are just graded sets
in the terminology of Section 3.4.
But then the inference rules should follow the same pattern: they either
have to have as their premises nite sequences of pairs of the form (formula,
degree ) and a consequent of the same form, or they have to consist of two
parts acting in parallel { a rst one that treats the formulas (in the standard
way well known from calculi for standard many-valued logic) and a second
one that treats the corresponding degrees. In any case, however, the result
of an application of such a rule of inference has to be an ordered pair of the
form (formula, degree ).
The consequence of this idea for inference schemata of the standard form
3 This means any fuzzy set
with a nite support.
476
19. Fuzzy Logic
H1 ; : : : ; Hn
H
(19.8)
is that premises and conclusions have to become ordered pairs, i.e. inference
schemata become the modied form
(H1 ; 1 ); : : : ; (Hn ; n ) :
(19.9)
(H; )
Here one usually has i > 0 for 1 i n as well as > 0. And furthermore
= () has to be a function of = (1 ; : : : ; n ).
A modied soundness condition for the inference schema of this form
(19.9) also is easily at hand.
Denition 19.3.1. A modied inference schema of the form (19.9) is sound
i it always holds true that in case one has [ Hi ] i for all 1 i n, then
one also has [ H ] = ().
In classical logic, the soundness condition for (19.8) often is formulated
in model theoretic terms. Then it becomes the condition that each model of
fH1 ; : : : ; Hn g also is a model of H .
This characterization can be adapted to the present case. Its formulation
becomes rather easy if one calls an interpretation A an (1 ; : : : ; n )-model of
a sequence (H1 ; : : : ; Hn ) of formulas i A is an ( i )-model of Hi for each
1 i n.
Corollary 19.3.1. A modied inference schema of the form (19.9) is sound
i every (1 ; : : : ; n )-model of (H1 ; : : : ; Hn ) is a ( )-model of H .
Having a closer look at schema (R^ ) of (19.9) one recognizes that this
schema can be split into two parts, viz. a traditional schema Rsyn of form
(19.8) and a mapping Rsem : 7! (). Writing (R^ ) = (Rsyn ; Rsem ) one has
that the modied schema (19.9) corresponds to an inference schema of the
form
!
H1 ; : : : ; Hn ; 1 ; : : : ; n :
H
(1 ; : : : ; n )
(19.10)
A derivation in fuzzy logic of a sentence H from a fuzzy set of premises

then is a nite sequence
(H1 ; 1 ); (H2 ; 2 ); : : : ; (Hn ; n ) ;
(19.11)
of ordered pairs (Hi ; i ) of the form (formula, degree ) such that
either Hi comes from and it is i = [ Hi " ] ,
or Hi comes from the fuzzy set of axioms AxFL and i = [ Hi " AxFL ] ,
or (Hi ; i ) results from the application of one of the rules of inference to
pairs (Hk ; k ) with 1 k < i.
19.4 Axiomatizing Fuzzy Logic
477
And the degree to which Hn is derived by the derivation (19.11) is n .

In fuzzy logic each derivation of a sentence H from a fuzzy set of
premises provides a degree to which H is derived from by this derivation.
Unlike classical logic, however, dierent derivations of the same sentence H
from the fuzzy set of sentences may provide dierent degrees to which
they are derivations of H . Therefore, a single derivation of H from gives
only a lower bound for what may be called the degree of derivability of the
sentence H from the fuzzy set . And this degree of derivability ` (H; )
itself should be the supremum
` (H; ) =def supfn j (H1 ; 1 ); (H2 ; 2 ); : : : ; (H; n )
(19.12)
is a derivation of H from g :
As usual now, the (fuzzy) syntactic, i.e. derivability based consequence hull
Cn` ( ) of the fuzzy set , i.e. the fuzzy set of theorems of the (fuzzy) elementary theory T with the fuzzy set of (nonlogical) axioms is the fuzzy
set of sentences
Cn` ( ) =def fH 2 L0 k ` (H; )g :
(19.13)

As each system S of (propositional or rst-order) fuzzy logic extends some
system S of many-valued (propositional or rst-order) logic, the problem of
axiomatizability for this system S of fuzzy logic may be expected to be
closely tied to the problem of axiomatizability of the \underlying" system S
of many-valued logic. The crucial point here is that the semantic entailment
operation Cnj= of S extends the semantic entailment relation Cnj= of S in
such a way that for each crisp set of sentences of the (common) language
LS of S and S one has Cnj= ( ) = Cnj= ( ). Therefore, assuming that S
and S have the same set of designated truth degrees, S has the same (crisp)
set Cnj= (;) = Cnj= (;) of logically valid sentences as S, and any adequate
axiomatization of the logically valid sentences of S is at the same time an
adequate axiomatization for the logically valid sentences of S .
Even more holds true. Given an adequate axiomatization of the semantic
entailment operation Cnj= of S, i.e. a notion of S-derivability `S such that
H 2 Cnj=( ) holds true i `S H for each sentence H and all sets of
sentences of S, then one already has an axiomatization of a \fragment" of
Cnj= , viz. of the restriction of Cnj= to crisp argument sets. The only minor
change which has to be made is a kind of rereading of the inference rules
which, for fuzzy logic, have to treat in parallel formulas and degrees. As
\nonfuzzy" many-valued logics can, from the point of view of fuzzy logic, be
considered as treating all the formulas to degree one (in the set of axioms
as well as in any set of premises), the rereading of the inference rules simply
478
19. Fuzzy Logic
amounts to combining with all the premises of these rules and with their
conclusions the truth degree 1.
Thus, (adequate) axiomatizations of the semantic entailment operation
of fuzzy logics may be chosen as extensions of (adequate) axiomatizations of
the semantic entailment operation of the \underlying" many-valued logic in
such a way that further axioms may be added (perhaps to some degree) and
that the rules of inference need to become joined with a treatment of degrees
to parallel the treatment of formulas they already regulate.
In accordance with the actual state of the art in fuzzy logic, however, we
shall not intend to discuss the problem of axiomatizability for large classes of
systems of many-valued logic as these remarks may seem to suggest. Instead,
we restrict our considerations to the case that the \underlying" system of
many-valued logic is the innite valued Lukasiewicz system L1 and take
this case as a kind of prototype of how to approach the problem for other
systems of fuzzy logic too.
For the Lukasiewicz propositional system L1 an adequate axiomatization not only of the set of tautologies but even of the semantic entailment
operation is provided by the axioms (i) to (iv) mentioned in Theorem 9.1.5
together with the rule of detachment w.r.t. !L .
Before we extend this axiomatization to the fuzzy Lukasiewicz propositional logic we should consider a possibility to \code" the membership degree
u = [ H " Cnj=( )]] of H in the fuzzy consequence hull of inside the language
of this fuzzy logic. This goal can be reached because from (19.7) together with
(19.2) one has that
u ValS A (H ) for all models A of
(19.14)
and therefore that
ValS A (u !L H ) = 1 for all models A of ;
(19.15)
if one introduces the truth degree constant u into the language to denote
the truth degree u. That means that from u = [ H " Cnj=( )]] one gets immediately that [ (u !L H ) " Cnj=( )]] = 1. And then one can hope to nd a
suitable axiomatization from this idea, and a further strategy to have only
minor changes in the sets of inference rules compared with axiomatizations
of the \background" many-valued Lukasiewicz logic.
Guided by this observation we enrich the language of our fuzzy Lukasiewicz logic by truth degree constants for every truth degree, writing from
now on u to denote the truth degree u.
Now we are in a position to give an adequate axiomatization of the propositional fuzzy logic based upon the innite valued Lukasiewicz system L1 .
The fuzzy set of axioms AxL1 for this system consists of:
the axioms (i) to (iv) of Theorem 9.1.5 with membership degree 1,
the truth degree constants u with membership degree u for every u 2 [0; 1],
the formulas (u !L v) $L seq 2 (u; v) with membership degree 1 for all
u; v 2 [0; 1],
479
the formulas :u $L non1 (u) with membership degree 1 for every u 2 [0; 1].
And the rules of inference, written down as in (19.9) or (19.10) as rules which
connect ordered pairs of the form (formula, degree ) to nite lists of such pairs,
are the following ones:
the generalized modus ponens, or: the graded rule of detachment, which
allows one to infer the ordered pair (H2 ; et2 (u; v)) from the ordered pairs
(H1 ; u) and (H1 !L H2 ; v), i.e.:
!
u; v
H1 ; H1 !L H2 ;
;
H2
et2 (u; v)
the constant introduction rule which allows one to infer the ordered pair
(u !L H; seq 2 (u; v)) from the pair (H; v), i.e.:
H
v
u !L H ; seq 2 (u; v) :
Both the fuzzy set of axioms as well as the list of inference rules look, after
some inspection, quite natural. The same still holds true for a lot of further
generalized inference rules which prove to be sound in fuzzy logic, cf. [399,
549]. There is, however, notwithstanding the fact that it is quite usual in set
theoretically oriented mathematical logic to use such uncountable languages,
a severe diculty with this approach: it uses truth degree constants for all the
real numbers of the unit interval [0; 1], i.e. for an uncountable set of degrees {
and therefore this extended language cannot be realized in the usual way as
a language whose words are nite strings of symbols of a countable alphabet.
Fortunately there is a way to overcome this diculty: for most applications it is sucient to have as truth degree constants only symbols for the
rational numbers of the real unit interval [0; 1]. And because this is a denumerable set of numbers, the language of this rational fuzzy Lukasiewicz
logic is itself a denumerable language.
What nally shall be discussed is the extension to the case of rst-order
fuzzy Lukasiewicz logic. Completely standard (as for Lukasiewicz logic in
general) is the change in the language if changing from propositional to rstorder logic. The only point for discussion is, again as usual, how to extend
the (fuzzy) set of axioms and the list of inference rules. One possibility is to
add as further axioms
the formulas 8xH (x) !L H (t), where t is any term free for the individual
variable x in H , with membership degree 1,
the formulas 8x(G !L H (x)) !L (G !L 8xH (x)), where G does not
contain the individual variable x free, with membership degree 1,
the formulas (G !L 9xH (x)) !L 9x(G !L H (x)), where G does not
contain the individual variable x free, with membership degree 1,
480
19. Fuzzy Logic
and thus adding (essentially) just the same axioms which can also be used in
extending an axiomatization of classical propositional logic to an axiomatization of classical rst-order logic.
And the list of inference rules has to be extended by adding
the suitable modied rule of generalization which allows one to infer the
ordered pair (8xH; u) from the ordered pair (H; u), i.e.:
!
u
H
8xH ; u :
19.5 Partial Soundness of Inference Rules

Any well-formed formula of many-valued propositional logic as well as any
closed formula of many-valued rst order logic describes some proposition
which may be only partially true (or partially false), i.e. which has a truth
degree (instead of a traditional truth-value).
Additionally, in many-valued rst order logic formulas which contain free
individual variables describe properties which hold for the objects, of which
they are properties, only to some (truth) degree.
Finally, fuzzy logic combines the idea of a logical calculus as a formal
system for deriving formulas from given (sets of) premises with the idea
of fuzzy sets in the way that fuzzy logic allows for only partially \given"
premises, i.e. for fuzzy sets of premises.
Now, looking back at the approaches we mentioned, one recognizes that
many-valued logic deals with partially true sentences, and that fuzzy logic
deals with even partially given (sets of) premises { but in both cases soundness of the rules of inference is taken in some absolute, i.e. two-valued sense.
So the problem arises whether it is possible to have also some notion of
partial soundness for inference schemata. And how such a notion, if it can be
suitably introduced, can be integrated into many-valued as well as into fuzzy
logic.
Regarding possible applications of partially sound rules of inference, it is
quite obvious that an analysis of the heap paradox, or the bold man paradox,
or other types of paradoxes may, instead of referring to the use of implications
with a truth degree a little bit smaller than 1, become based upon the use of
inference schemata which are \not completely sound".
19.5.1 Formalizing the problem

For simplicity we will restrict the considerations here to Lukasiewicz type
many-valued systems: truth degrees a subset of the real interval [0; 1], larger
truth degrees as the \better" ones, and 1 the only designated truth degree.
481
As the inference schema under discussion let us again consider the schema
(19.8):
H1 ; : : : ; Hn
(R)
H
with n premises H1 ; : : : ; Hn and the conclusion H . As soundness condition
for (19.8) we consider the inequality
[ H1 & : : : &Hn] [ H ]
(19.16)
with & for a suitable conjunction connective, e.g. some t-norm based one,
and [ G] for the truth degree of the formula G. We shall prefer (19.16) here
because it seems to be intuitively more convincing as demanding only
if [ H1 ] = : : : = [ Hn ] = 1 then [ H ] = 1 :
(19.17)
A partially sound schema (19.8) of inference, which is not sound in the
usual sense, has to deviate from (19.16) to some degree. And this degree of
deviance we intend to use to \measure" the deviance of schema (19.8) from
soundness, i.e. to dene some \degree of soundness" of schema (19.8).
Referring to a measure of deviance of schema (19.8) from soundness condition (19.16), it is quite natural to look at the value (R) dened as
(R) = sup (maxf0; [ H1 & : : : &Hn ] ; [ H ] g) ;
(19.18)
where the supremum has to be taken with respect to all interpretations and
all valuations of the individual variables. But other approaches as (19.18) can
be discussed too.
Now, with reference to a suitable implication connective ! one should
have
[ G1 ] [ G2 ] i [ G1 ! G2 ] = 1
(19.19)
for any ws G1 ; G2 , and hence
[ H1 & : : : &Hn ! H ] = 1
(19.20)
as an equivalent soundness condition instead of (19.16). This means that one
thus usually has
(R) sound , always [ H1 & : : : &Hn ] [ H ]
, always [ H1 & : : : &Hn ! H ] = 1
, j= (H1 & : : : &Hn ! H ) :
These equivalences give another way to approach partial soundness instead
of (19.18).
Denition 19.5.1. For a schema of inference (R) its degree of soundness
shall be
(R) =def inf([[H1 & : : : &Hn ! H ] )
(19.21)
with the inmum taken over all interpretations and all assignments of the
individual variables.
482
19. Fuzzy Logic
Corollary 19.5.1. With ! the Lukasiewicz implication !L, one has

(R) = 1 ; (R)
and thus (19.21) as a suitable generalization of the idea which led to (19.18).
With (R) for the degree of soundness of inference rule (R) as in (19.8),
one thus always has
(R) [ H1 & : : : &Hn ! H ]
or even, accepting (R) as a truth degree constant of the language to denote
the degree (R) and having (19.19) satised,
[ (R) ! (H1 & : : : &Hn ! H ) ] = 1
which via importation and exportation for the implication, and commutativity for the conjunction operations is equivalent to
[ H1 & : : : &Hn ! ((R) ! H ) ] = 1 :
This is a rst way to \code" partially sound rules: it presupposes only that
one has to have each degree of soundness as a truth degree constant available
within the language.
19.5.2 Partially sound rules in many-valued and fuzzy logics

The use of sound rules of inference within the process of inference of new
formulas from given premises can be seen as a transfer of \condence" in the
premises to a condence in the conclusion.
For classical logic, this type of interpretation looks completely unproblematical: condence can be understood as the assumption of truth. A sound rule
of inference does then rationally transfer this condence from the premises to
the conclusion. For many-valued logic, one way to interpret \condence" in a
formula or proposition H is to translate this into the statement [ H ] = 1, i.e.
again into the assumption of the (complete) truth of H . But it also seems natural to give \condence in H " another reading , meaning [ H ] u for some
truth degree u. In this second sense condence itself is graded in some sense,
and this is in very interesting coincidence with the basic idea of many-valued
logic, i.e. with the graduation of truth.
Soundness condition (19.16) now can be read as allowing only such inference schemata as sound which transfer the \common condence" in the
premises, i.e. the condence in the conjunction of the premises, into a suitable
condence in the conclusion.
This reading of soundness shows why soundness condition (19.16) seems
to be preferable over soundness condition (19.17). It is based on a kind of
identication of truth degrees with degrees of condence (in the sense that
degree u of condence in H means [ H ] u) { and this does not seem completely unreasonable. Furthermore, here this interpretation of truth degrees
483
as (lower bounds of) condence degrees either refers to an identication of
truth degrees with degrees of condence { or it can simpler be seen as an
addition of condence degrees to classical logic.
However, what does this intuitive idea of \condence" contribute to the
discussion of partial soundness? Of course, depending on their degree of
soundness, the condence in the conclusion, given the condences in the
premises, should be smaller than in the case of a completely sound rule of
inference (with the same premises) { becoming smaller as the degree of soundness becomes smaller.
Looking again at the schema (R) of (19.8), and rst assuming full condence in the premises, i.e. assuming [ Hi ] = 1 or, equivalently, j= Hi for
1 i n, it seems reasonable to assume
(R) condence in H ;
i.e. to assume (R) [ H ] .
Thus, from the intuitive point of view an application of a partially sound
rule of inference can be understood via the additional idea of condence
degrees.
But what about repeated applications of partially sound rules of inference? And is it really a convincing idea to identify degrees of condence with
(lower bounds of) the truth degrees?
At least for the second one of these problems a more cautious approach
may help which rst separates condence and truth degrees: a change from
formulas to ordered pairs consisting of a formula and a degree of condence.
The consequence of this idea for inference schemata like (19.8) now is that
premises and conclusions have to become ordered pairs, i.e. (R) changes into
the modied schema (19.9):
(H1 ; 1 ); : : : ; (Hn ; n ) :
(R^ )
(H; )
Choosing here condence degrees, like membership degrees and truth degrees,
from the real unit interval [0; 1] means that in (19.9) usually i > 0 for
1 i n as well as > 0, and that = () has to be a function of
= (1 ; : : : ; n ).
Regarding condition (19.16), and assuming that the schema (19.8) satises
condition (19.16), one may take in the modied schema (R^ ) the conclusion
H as given in the degree = et(n) () with et(u; v) the truth degree function
characterizing the conjunction &, and et(n) the n-ary iteration of et, then of
course this schema (R^ ) is sound . On the other hand, having the inference
schema (R^ ) sound and using i = [ Hi ] for all 1 i n, so in case () =
et(n) () the corresponding \reduced" schema (R) is simply sound.
(n)
But; if (R^ ) is sound and one does not
always have () et (), then
for ? R) = inf ;:::;n ( () ; et(n) () one has always
1 + ? (R) 1 ; et(n) () + () :
(19.22)
1
484
19. Fuzzy Logic
With ? (R) = 1 + ? (R) this can be written as

? (R) seq 2 (et(n) (); ()) :
(19.23)
Therefore, inference schemata of type (19.9) cover the usual sound inference rules as well as the partially sound rules of inference, and the soundness
for such schemata covers usual soundness together with partial soundness.
Moreover, starting a deduction from a set of \premises" using inference
rules of type (19.9) needs premises which have the form of ordered pairs
(formula ; degree ), i.e. one in such a case has to assume
= f(H1 ; 1 ); (H2 ; 2 ); : : :g :
(19.24)

But one then has to consider a membership function : L ! W if one
\completes" the set of (19.24) with all the ordered pairs (H; 0) for which
H 2= fH1; H2 ; : : :g.
With W = [0; 1] or W [0; 1] this means that is a fuzzy subset of the
set L of all well-formed formulas.
Now it is a routine matter to prove that the introduction of partially
sound inference rules in many-valued logic leads to fuzzy logic.
Proposition 19.5.1. Each many-valued elementary theory which one gets
by enriching many-valued logic with some (crisp) set of nonlogical axioms
and some partially sound rules of inference is equivalent to some elementary
theory in the realm of fuzzy logic characterized by a fuzzy set of nonlogical
axioms.
Extending also fuzzy logic with partially sound rules of inference gives
nothing new, as again can be proved without diculties.
Proposition 19.5.2. Each many-valued elementary theory which one gets
by enriching fuzzy logic with some (fuzzy) set of nonlogical axioms and some
partially sound rules of inference is equivalent to some elementary theory
in the realm of fuzzy logic characterized simply by a fuzzy set of nonlogical
axioms and using only sound rules of inference.
19.6 Some Theoretical Results

Notwithstanding the fact that the semantic as well as the syntactic approaches toward graded consequences parallel the usual approaches in classical and in \standard" many-valued logic in a natural way, the syntactic
approach has some particular peculiarity. The crucial point is that, unlike
the situation in classical logic, because of denition (19.12) the knowledge
of a single derivation of a sentence H from (with some degree for this
derivation of H ) does not necessarily give the membership degree of H in
Cn` ( ). And even a nite set of such derivations of H from may not sufce to determine this membership degree. This may happen in the case that
19.6 Some Theoretical Results
485
the many-valued logic, a particular system of fuzzy logic is based upon, itself
is an innitely many-valued system. In this sense, despite the fact that all
the inference rules are supposed to be nitary, i.e. that each one has only a
nite number of rule premises, the present notion of provability in fuzzy logic
is not a nitary notion of provability.
In this particular sense, provability in fuzzy logic is much more complicated a matter as is provability in classical rst-order logic.
One of the most basic problems for graded consequences is the relation
between the semantic and the syntactic approaches. Here, again, the system
of many-valued logic becomes important within which the approach toward
graded consequences is considered. If suitably chosen, as done in the last
section for propositional as well as rst-order Lukasiewicz fuzzy logics, both
these approaches give the same degrees of consequence.
Theorem 19.6.1 (Completeness Theorem). For the approach toward
graded consequences which is based on one of the (propositional or rst-order)
Lukasiewicz systems L1 with truth degree constants for all the real numbers
from [0; 1], one has for any fuzzy set of sentences
Cnj=( ) = Cn` ( ) :
Therefore one has completeness in the strong sense of an adequate axiomatization of the semantic entailment operation.
These last mentioned results have the disadvantage that the language
of fuzzy logic becomes uncountable by the addition of all the truth degree
constants. It is interesting that essentially the same result is already provable
with only truth degree constants for the rationals from [0; 1], as was shown
by Hajek [243]. He proves the following:
Theorem 19.6.2 (Restricted Completeness Theorem). The approach
toward graded consequences which is based on one of the (propositional or
rst-order) Lukasiewicz systems L1 with truth degree constants for all the
rational numbers from [0; 1], gives for any \rational valued" fuzzy set :
L ! [0; 1] \ Q of sentences
Cnj=( ) = Cn` ( ) :
Interestingly, the denitions (19.6), (19.13) of Cnj=; Cn` can remain unchanged in this new situation with the restricted language.
Even in this restricted form with truth degree constants only for the
rationals the provability relation is not a simple one, and the theoremhood
property is even a quite complicated one, as was proved also by Hajek [243,
244, 246], extending results of Ragaz [434] and Hahnle [238].
Theorem 19.6.3. Suppose that T is an axiomatizable rational fuzzy theory,
i.e. that T : L ! [0; 1] \ Q is a recursive function. Then
(i) the (crisp) satisability relation Cn` (T ) = f(H; r) j [ H " Cn` (T )]] = rg
is a 2 -relation,
486
19. Fuzzy Logic
(ii) and already for T = ; all the sets Cn[`r] = fH j [ H " Cn` (;)]] rg,
r 2 Q , are 2 -complete.
In particular, this means that already the set of logically valid formulas of the
rst-order rational fuzzy Lukasiewicz logic L1 (with all the rationals from
[0; 1] as truth degree constants) is 2 -complete.4 And it means also that the
sets of logically valid formulas of this rst-order rational fuzzy Lukasiewicz
logic has the same complexity as the set of logically valid formulas of the
standard rst-order Lukasiewicz logic L1 , which was proved to be 2 complete in [434, 435].
And even in a further point these two systems of many-valued logic are
not very dierent: they prove the same LL -formulas as shown in [251]. This
means that one has:
Theorem 19.6.4. Let be an L-theory, i.e. a (crisp) set of LL-sentences.
Then one has for each w H of the language LL
`L1 H , [ H " Cn` ( )]] = 1 ;
i.e. the rst-order rational fuzzy Lukasiewicz logic is a conservative extension of the standard rst-order Lukasiewicz logic L1 .
It is interesting to notice that this last conservative extension result holds
also true for the \full", i.e. for the rst-order real fuzzy Lukasiewicz logic
(which has truth degree constants for all the reals from [0; 1]), cf. [251], which
is also a conservative extension of the standard rst-order Lukasiewicz logic
L1 .
For proofs and further results, as well as applications in AI related elds
like expert systems or logic programming, the reader may consult e.g. [246]
and [399] and the ongoing research literature.
19.7 The Algebraic Approach

Besides the model theoretic and the syntactic types of approaches toward the
treatment of logical consequence, one has (in classical logic) also the algebraic
one. This algebraic approach essentially consists in a study of the (semantic
as well as syntactic) consequence operations Cn j= ; Cn ` as mappings from the
power set IP (L ) of the set of all ws into itself, which either are dened by
some characteristic properties, or result from some closure system.
The characteristic properties which dene the (classical, i.e. crisp) closure
operators as some mapping C : IP (X ) ! IP (X ), X any set, are the properties
for arbitrary A; B A:
4 For further complexity results concerning propositional as well as rst-order fuzzy
logic the interested reader should consult [246].
(a) A C(A) ;
(b) A B ) C(A) C(B ) ;
(c) C(C(A)) = C(A) ;
487
embedding property ; (19.25)

monotonicity ;
(19.26)
idempotency :
(19.27)
And these properties can immediately be generalized to the fuzzy case. One
has to consider C : IF (X ) ! IF (X ) as a mapping from some class IF (X ) of
fuzzy subsets of a suitable universe of discourse X into itself, and one has to
take in (19.25) and (19.26) the inclusion for fuzzy sets, or also its graded
version j , for the inclusion for crisp sets. With these slight modications,
conditions (19.25), (19.26), and (19.27) dene also the fuzzy closure operators.
The other approach toward an algebraic treatment of logical consequence
works with closure systems. Such a closure system is (classically) a subset C
of some power set IP (X ) such that one has
\
M C ) M 2 C;
(19.28)
T
with ; = X here. The extension to the realm of fuzzy sets, i.e. the denition
of a fuzzy closure system is even simpler Tas for closure operators: one has
simply to assume C IF (X ), and to read in (19.28) as the intersection of
a family of fuzzy sets.
As an immediate consequence of this denition one has that for each fuzzy
closure system C and each 2 [0; 1] the class
C< =def fY < j Y 2 Cg
of all open -cuts Y > = fx j [ x " Y ] > g of the members of C is a crisp
(i.e. classical) closure system.5 Hence each fuzzy closure system determines
a family of crisp closure systems. And as for fuzzy sets and their families
of cuts, also for fuzzy closure systems one is able to determine them from
families of crisp closure systems. The details are given in [56, 185].
It is a nice fact that these fuzzications remain related to each other as
is the case for their crisp counterparts, cf. [56, 57, 185].
Proposition 19.7.1. For each class C IF (X ) of fuzzy sets there is a fuzzy
closure operator CC dened for each A 2 IF (X ) by
\
CC(A) =def fY 2 C j A Y g :
(19.29)
And for each mapping C : IF (X ) ! IF (X ) which has the embedding and
monotonicity properties a fuzzy closure system is determined by
CC =def fY 2 IF (X ) j C(Y ) = Y g :
(19.30)
5 And the same holds true if one considers all the closed
fx j [ x " Y ] g for Y 2 C.
-cuts Y =
488
19. Fuzzy Logic
Proof: For each class C IF (X ) one immediately recognizes that the

mapping CC has the embedding and the monotonicity properties. So it remains to show for any A 2 IF (X )
CC(CC(A)) CC(A)
(19.31)
to get also the idempotency. But one has for each A 2 IF (X ) from (19.29):
\
CC(CC(A)) = fY 2 C j CC(A) Y g :
Hence one has for any B 2 IF (X ) with B CC (CC (A)) also B Y for
each Y 2 C with CC (A) Y . However, for any Y 2 C with A Y one has
also CC (A) Y by denition (19.29), and hence B Y . But this means
B CC (A). And thus one has (19.31).
So assume that C : IF (X ) ! IF (X ) is a mapping which has the embedding
and the monotonicity property. Consider any
M CC = fY 2 IF (X ) j C(Y ) = Y g :
Then one has
\
\
\
\
C( M ) fC(A) j A 2 M g = fC (A) j A 2 M g = M
T
T
by the monotonicity T
of C. Hence one gets C( M ) = M from the embedding property. Thus M 2 CC , i.e. CC is a fuzzy closure system.
2
Again as for the crisp case, the two ways (19.28) and (19.30) to form a
fuzzy closure operator, or a fuzzy closure system, respectively, are inverse to
each other.
Proposition 19.7.2. For each fuzzy closure operator C and each fuzzy closure system C one has
CCC = C ; and CCC = C :
Proof: For the rst claim one has to recognize that one has on the one
hand A C(A) 2 CC and therefore CCC (A) C(A) for each A 2 IF (X ),
and on the other hand C(A) = A B for each A B 2 CC , and thus also
C(A) CCC (A).
For the second claim one observes that one has CC (A) = A for each
A 2 C, which means C CCC . But one has also A 2 C for each A 2 IF (X )
with CC (A) = ATby the fact that each fuzzy closure system is closed under
the intersection for subsets. This means that one has also CCC C. 2
Based on this result one calls CC , for each fuzzy closure operator C, the
fuzzy closure system of C. And similarly one calls CC , for each fuzzy closure
system C, the fuzzy closure operator of C.
Similar to the case of classical logic one also gets for fuzzy logic that the
operation Cnj= , as introduced in (19.6), is a fuzzy closure operator. And, with
some assumptions concerning the inference rules, also Cn` , as introduced in
(19.13), is a fuzzy closure operator.
489
Proposition 19.7.3. The semantic entailment operator Cnj= of fuzzy logic
is a fuzzy closure operator, and has as its fuzzy closure system the class of
all fuzzy theories Th (A) of interpretations A.
Proof: If Cnj= is a fuzzy closure operator,
its corresponding fuzzy closure
system is the class of all fuzzy theories Th (A) of interpretations A according
to Proposition 19.7.2 and the denition (19.5) of Cnj= .
It is obvious that Cnj= has the embedding as well as the monotonicity
property. So it remains to prove
Cnj=(Cnj=( )) Cnj= ( )
for any fuzzy set of sentences. But one has
\
Cnj=(Cnj=( )) = fTh (A) j A j= Cnj= ( )g
\
= fTh (A) j A j= g ;
because one easily shows
A j= Cnj= ( ) , A j= :
Here ()) is obvious because one always has A j= 0 from A j= and 0 .
And (() results from the fact that for A j= one has Cnj=( ) Th (A),
and hence A j= Cnj=( ).
2
Proposition 19.7.4. Let K S be a logical calculus for graded derivations of

some system S of fuzzy logic. Suppose that either (i) there holds a completeness theorem similar to Theorem 19.6.1 for K S w.r.t. the semantics of S, or
that (ii) the system S of fuzzy logic is a nitely many-valued one, or that (iii)
the system S has truth degree set [0; 1] and all its inference rules
H1 ; : : : ; Hn ; 1 ; : : : ; n
H
(1 ; : : : ; n )
are based on (in each argument) left continuous truth degree functions , then
the corresponding syntactic consequence operator Cn` of S is a fuzzy closure
operator.
And if the syntactic consequence operator Cn` of fuzzy logic is a fuzzy
closure operator, it has as its fuzzy closure system the class of all deductively
closed sets Cn` ( ) for some fuzzy elementary theory with fuzzy axiom set .
Proof: If Cn` is a fuzzy closure operator, its corresponding
fuzzy closure system is the class of all deductively closed sets Cn` ( ) for some fuzzy
elementary theory with fuzzy axiom set , according to Proposition 19.7.2
and the denition (19.13) of Cn` .
The case that there holds a completeness theorem for K S , the corresponding operations Cnj= and Cn` coincide, and the result follows from the previous
theorem.
490
19. Fuzzy Logic
For the other cases, as in the previous proof, it is obvious that Cn` has
the embedding and the monotonicity property. So again it remains to prove
Cn` (Cn` ( )) Cn` ( )
(19.32)
for any fuzzy;set of sentences.

Let H = (H1 ; 1 ); (H2 ; 2 ); : : : ; (Hn ; n ); (H; ) be some derivation of
a w H to degree from the fuzzy set Cn` ( ) in the calculus K S . And let the
pair (Hi ; i ) be chosen in this derivation because Hi is one of the premises,
i.e. because of [ Hi " Cn` ( )]] = i .
Then there exists, in the case that ;the system S of fuzzy logic is a nitely
many-valued one, a derivation Hi = (H1i ; i1 ); (H2i ; i2 ); : : : ; (Hi ; i ) of Hi
to degree i from the fuzzy set . Now one can substitute this sequence Hi
for the pair (Hi ; i ) in the derivation H. The result is again a derivation of
the w H to degree from the fuzzy set Cn` ( ) in the calculus K S , which
however does not refer to Hi as a premiss from Cn` ( ), but uses instead
some premises from . And this can be done for each of the ws which are
chosen inside the derivation H because they are premises from Cn` ( ). The
nal result is a derivation H of the w H to degree from the fuzzy set .
Now consider the case that the system S of fuzzy logic has truth degree
set [0; 1], and that all its inference rules are based on left continuous truth
degree functions.
Suppose [ H " Cn` (Cn` ( ))]] = u for some w H . Then there exists for
each real number > 0 some derivation
;

H = (H1 ; 1 ); (H2 ; 2 ); : : : ; (Hn ; n ); (H; )
of H to a degree u ; from the fuzzy set Cn` ( ) in the calculus K S .
Again let some pair (Hi ; i ) of this derivation be chosen because Hi is one of
the premises, i.e. because of [ Hi " Cn` ( )]] = i . Then the derivation degree
of H depends from i in a left continuous manner.
And this means that there exists some 0 > 0 such that the derivation

H becomes
a derivation of H to some degree u ; 2 from the fuzzy

set Cn` ( ) if one substitutes some pair (Hi ; i ) with i i ; 0 for the
pair (Hi ; i ), and derives all the later pairs (Hj ; j ), i j , of this \new "
derivation by reference to the same inference rules as in the derivation H .
one has that for each 0 > 0 there exists a derivation Hi 0 =
; Furthermore
(H1i ; i1 ); (H2i ; i2 ); : : : ; (Hi ; i ) of Hi to a degree i i ;0 from the fuzzy
set . Now substitute as in the previous case inside H such a derivation Hi 0
for the pair (Hi ; i ), and adapt all the degrees j inside the pairs (Hj ; j )
with j i of this derivation H in the way just described. The result is a
derivation of H to some degree u ; 3 from the fuzzy set of premises
Cn` ( ), which however does not refer to the w Hi as a premiss from Cn` ( ).
In this way one can get for each > 0 and each derivation H of H to
degree u ; , which uses some pair (Hi ; i ) because Hi is a premiss from
Cn` ( ), some derivation Hi 0 of H to degree u ; 3 which does not refer
to the premiss H from Cn` ( ), but does instead refer to premises from .
491
If one does these modications of the derivation H for all the pairs
(Hi ; i ) of it which appear because Hi is a premiss from Cn` ( ), a derivation
of H to some degree u ; k results, for some k 2 N , which is a derivation
from the fuzzy set of premises.
Therefore one has even [ H " Cn` ( )]] u, and thus nally (19.32) also in
this case.
2
492
19. Fuzzy Logic
20. Treating Presuppositions

with Many-Valued Logic
20.1 The Phenomenon of Presuppositions

Common language allows one to build sentences which use empty names, i.e.
proper names or denite descriptions, which do not denote some single object
{ neither one nor many { as e.g. in the sentence:
The present king of France was born in Vienna.
(20.1)
Such sentences are indistinguishable, by their syntactic structure, from other
sentences which use names which all denote and thus refer to some (\existing") objects like the very similar sentence:
The present king of Sweden was born in Vienna.
(20.2)
By the way, it is not important that the names which have been used in these
two sentences are not proper names but descriptions. The following discussions remain completely unchanged if one uses instead of the non-denoting
description \the present king of France" e.g. the non-denoting names \Nicolas Bourbaki",1 \Winnie the Pooh", or \Pegasus". It is, however, essential
that one can, similar to the negation of sentence (20.2)
The present king of Sweden was not born in Vienna.
(20.3)
also form the sentence
The present king of France was not born in Vienna.
(20.4)
which is { seemingly(?) { the negation of sentence (20.1).
If one intends to determine truth values on the one hand for the sentences
(20.2), (20.3), and on the other hand for the sentences (20.1) and (20.4), then
there is no problem in the rst case. However, one runs into diculties with
the second case because the usual understanding of everyday language tells
that sentence (20.1) as well as sentence (20.4) has to be false.
The main reason for this dierent behavior of the two syntactically equivalent pairs of sentences is that (20.1) and (20.4) presuppose the existence of
1 Bourbaki is well known as author of important mathematical papers and books.
However, \he" is not any person: the name \Bourbaki" is a pseudonym for a
group { with members varying in time { of mainly French born mathematicians;
cf. e.g. [233].
494
20. Treating Presuppositions with Many-Valued Logic
a present king of France { and that (20.2) and (20.3) correspondingly presuppose the existence of a present king of Sweden. Such presuppositions may
concern the existence of some objects, as in the previous examples. They may,
however, also be related to other \facts" as in sentences like \Jane stopped
smoking", \Bob does not beat his girlfriend any more", \Only Mike did not
do his homework well", or \Pat is even taller than Joe". And these situations occur even more frequently in everyday discourse as the existential
presuppositions we rst considered.2
The general discussion of the complex phenomenon of presuppositions is a
common topic in theoretical linguistics and applied logic. We shall not go into
this topic in its generality here, not only because we are only interested in
so-called \semantical presuppositions" (and disregard completely e.g. pragmatical presuppositions). The interested reader may consult e.g. [35, 315,
354, 402, 502, 513, 555]. Here only a particular problem shall be considered:
the assignment of truth-values to sentences like (20.1) and (20.4).
Intuitively it seems to be without doubt that neither sentence (20.1) nor
sentence (20.4) should be assigned the truth-value >. The reason is that
by common sense such an assignment would in any case be understood as
saying that there exists a present king of France. In the case, however, that
one classies (20.1) as well as (20.4) as \false", i.e. assigns both sentences
the truth degree ?, then one has to accept dierent forms of negations for
sentences, because the syntactic transformations which transforms (20.1) into
(20.4) is the same as the process of negation which transforms (20.2) into
(20.3). And these dierent types of negation come under the same syntactic
form { and can be distinguished only from the knowledge of the facts the
sentences are about. This is a much worse situation because it forces one to
disregard, even in formal syntax for ordinary languages, the usual viewpoint
of context independence in a severe way. It seems that one has mainly, or
even only, two ways out:
(a) Neither (20.1) nor (20.4) get a truth degree assigned, i.e. one allows for
truth-value gaps.
(b) (20.1) as well as (20.4) get assigned truth degrees which dier from the
usual truth values >; ?, i.e. one changes from classical logic into the eld
of many-valued logics for a discussion of the truth behavior of sentences
of common language.
These ways out sometimes are not really dierent: version (a) occasionally
can be understood in the sense of version (b) { one simply has to read a truth
value gap as the assignment of some (third) truth degree dierent from >; ?.
This simple translation, however, works well only in those cases where the
2 The proper treatment of non-denoting terms, i.e. of names or descriptions which
do not refer to some existing object (i.e. to some object of the respective universe
of discourse), is also discussed in the eld of free logic. Even more, free logic is
that part of logic which is designed to handle just such non-denoting terms;
cf. e.g. [47].
20.1 The Phenomenon of Presuppositions
495
truth-value gaps themselves behave truth-functionally, i.e. follow the lines

marked by the principle of compositionality, as e.g. in [324].
But the more common approaches which use truth-value gaps are not
truth functional, and hence cannot be translated in such a simple manner
into three-valued3 (or other many-valued) approaches. Perhaps the most wellknown of them, introduced by B.C. van Fraassen in [172], cf. also [173,
174], uses supervaluations : valuations which are partial mappings4 of the class
of all ws of the language L2 of classical logic into the set f>; ?g of truth
values, and which have the crucial additional property that they map all
logically valid ws of classical logic to >. This means, among others, that
a disjunction A_ :A becomes evaluated by > even in those cases where
A does not have a truth value. The negation operation behaves here over
f>; ?g as negation in classical logic does, and it retains truth-value gaps {
hence it is similar to the negation in the Lukasiewicz system L3 . However,
a disjunction of the form A_ :B may, for a sentence A which does not have
a truth value, have either no truth value or the truth value >, depending of
course on the \truth value" of B .
A deeper study of truth-value gaps and of logics which allow for truthvalue gaps is conducted in the eld of partial logic, cf. e.g. [62, 326, 390].
We shall not go deeper into this topic of linguistic discussions on presuppositions in general, instead in the following we shall only consider the case
that one is interested to understand the phenomenon of presuppositions in
the extensional context of many-valued logic.5
Approaches toward the phenomenon of presuppositions which have used
ideas from the eld of many-valued logic, up to now essentially have referred
to systems with three or with four truth degrees, and with truth degree 1 as
the only designated one.
3 In any case a gap can be formalized as a (kind of) third truth degree besides >
and ?; the crucial point whether this leads into the eld of many-valued logic
is, however, whether this third degree then behaves truth-functionally or not.
This simple fact seems (essentially) to be the main (hidden) background for a
discussion in linguistics on trivalence versus gapped bivalence, cf. e.g. [79, 80,
503].
4 This means, mappings which are not always dened. And the ws for which such
a mapping f is not dened are just the ws which do not have a truth-value under
this supervaluation.
5 As a side remark it shall be mentioned that one sometimes describes the behavior
of the (standard) connectives under supervaluations also in the form of truthvalue tables. This, however, can only be done in a suitable way if one allows that
these truth tables become generalized in the sense that they sometimes do not
provide an uniquely determined \value" for some of the input situations they
are describing.
496
20.2 Three-Valued Approaches

In the three-valued case, i.e. in the case that one considers the truth degree set
W3 with > = 1 and ? = 0 as intended understanding, the additional truth
degree 12 is assigned e.g. to all those simple sentences of subject-predicate
structure like (20.1), (20.2) which use either non-designating names, i.e. which
refer to existential presuppositions which are not satised, or which refer to
other types of presuppositions which are not satised. In this case one gets in
a rather natural way the truth degree function non1 as truth degree function
for the negation. For the example of the sentences (20.1), (20.4) this means
that both of them get assigned the truth degree 12 , and (20.4) remains the
negation of (20.1). However, the problem of the right analysis of the negation
operator is still under discussion among linguists actually, and there are quite
dierent positions taken w.r.t. its solution, cf. e.g. [79, 280, 281, 502, 503].
Further developments of such a three-valued approach toward presuppositional phenomena can either refer to known systems of three-valued logic,
as it was done by Woodruff [575] with reference to the Kleene system as
mentioned in Chapter 15, or can proceed { guided by reference to linguistics
and language understanding { with the development of separate three-valued
systems as done e.g. in [63], as well as in [501, 502], and in [419].
The approach in [575] proceeds e.g. with reference to implication connectives !, with truth degree functions seq 0 as introduced on p. 388, to a
disjunction connective _ characterized by vel1 , to a negation : with truth
degree function non1 , and to an additional unary connective T with truth
degree function
verr (u) =def et2 (u; u)
which proves to be just the connective J1 .
Intuitively it is quite convincing to dene the relation that some sentence
B is a presupposition for a sentence A by
B presupposition for A =def if A is true or false, then B is true (20.5)
as was done by Strawson, cf. e.g. [173, 514]. Denoting the relation \A is
a presupposition for B " by B A, this relation can in this three-valued
system be dened as
B A =def r(rA _ r:A) ! rB
(20.6)
and thus becomes itself a three-valued predicate, i.e. becomes capable of
getting assigned also the third truth degree. But this relation (20.5) can also
be formalized metatheoretically as a two-valued predicate
pres(B; A) =def j= B A
(20.7)
having in mind that the degree 1 is the only designated one.
The approach of Seuren in [501, 502] does not start from the truth
degree set W3 , but from the set f1; 2; 3g of truth degrees which obviously can
20.2 Three-Valued Approaches
497
be mapped onto W3 in dierent ways. The intended interpretation of these

truth degrees is that degree 1 should be read as `true', that degree 2 should be
read as `minimally false' (or just `false' ), and that degree 3 should be read as
`radically false'. The intuition behind these truth degrees is not explained in
much detail in [501, 502], however the main point is that the degree 2 indicates
\normal" falsity as in classical logic, and the degree 3 indicates presupposition
failure.6 The formalized language under consideration has as its connectives
a conjunction ^S , a disjunction _S , and three negations S ; 'S , and :S with
corresponding truth degree functions as given in Fig. 20.1. Only two of these
^S 1 2 3
_S 1 2 3
1 2 3
1 2 3
1 1 1 1
S 2 1 3
2 1 2 2
'S 2 2 1
2 2 3
3 3 3
3 1 2 3
:S 2 1 1
Fig. 20.1. Truth degree tables of Seuren's connectives
1
2
3
negation connectives have to be considered for the present discussion because

one has immediately from these truth degree tables
:S H semantically equivalent S H _S 'S H;
which means that the negation connective :S , which Seuren calls classical negation, can be dened in his system from the minimal negation S
(which preserves presupposition) and the radical negation 'S (which cancels
presupposition) using the disjunction _S .
From our explanations it follows that for a mapping of Seuren's truth
degree set f1; 2; 3g onto our usual truth degree set W3 for three-valued logic
one in any case should have (1) = 1. Hence there remain two possible such
mappings 1 ; 2 characterized by 1 (1) = 2 (1) = 1 and
(20.8)
1 (2) = 0 and 1 (3) = 0:5;
(20.9)
2 (2) = 0:5 and 2 (3) = 0
There is no internal preference between these two mappings 1 ; 2 , i.e.
both seem to be equally acceptable as \translations" onto our usual truth
degree sets. The crucial point is simply that the truth degrees 21 and 0 in
both cases indicate quite dierent interpretations.
Having this in mind it becomes an interesting problem to see how the
connectives of Seuren's approach translate. This translation can either be
realized as a suitable rewriting of the truth degree tables (which is a quite
simple matter), or it can be realized as a representation of the Seuren connectives using connectives of other three-valued systems.
6 This understanding of the truth degrees is in an interesting way rather similar to
the understanding of Blau in his three-valued approach toward natural language
as mentioned in Section 15.1.
498
This last mentioned way shall be preferred here. The necessary calculations are rather elementary and may be an exercise for the reader. The main
result is the following:
Proposition 20.2.1. Consider the language LL of the (three-valued) Lukasiewicz system L3 and add to it the negation of the (three-valued) Post
system and the internal conjunction ^+ of the Bocvar system B 3 .
(i) For the rst \translation" 1 of Seuren's truth degrees one has
H1 ^S H2 semantically equivalent H1 ^+ H2 ;
H1 _S H2 semantically equivalent (H1 Y H2 ) & (H1 ^ H2 );
S H semantically equivalent :H;
'S H semantically equivalent J (H ):
(ii) For the second \translation" 2 of Seuren's truth degrees one has
H1 ^S H2 semantically equivalent H1 ^ H2 ;
H1 _S H2 semantically equivalent H1 _ H2 ;
S H semantically equivalent : H;
'S H semantically equivalent :J0 (H ):
Together with the fact that the truth degree 1 is considered as the only
designated one, also Seuren's approach determines a notion of entailment
j=S which can be used for a metatheoretical denition of the relation of
(semantical) presupposition
B A =def A j=S B and S A j=S B:
And the approach of Peters [419], which mainly consists in a model
theoretic analysis guided by the ideas of possible worlds semantics, refers
on the one hand to the Lukasiewicz negation :, and on the other hand to
non-commutative connectives u; t for conjunction and disjunction, and to an
implication connective ; determined by the truth degree tables of Fig. 20.2.
Again in this context the truth degree 12 indicates a truth-value gap, which
1
2
u 0
1
2
t 0
1
2
0
1
2
1
12
2
1
2
1
2
0
1
2
1
2
1
12
2
1
1
1
2
0
1
1 1 1 1
Fig. 20.2. Truth degree tables of Peters' connectives
; 0
0 0
1
2
1
2
1
2
1
2
1
2
behaves truth-functionally, and the degrees 0 and 1 mean ?; >, respectively.

Already these three, partly competing formalizations indicate that there
is at present no commonly agreed version of three-valued formalization of the
20.3 Four-Valued Approaches
499
presupposition phenomenon. Therefore we shall not go into further details of

these three-valued approaches.

Combining the idea of a many-valued treatment of presuppositions with the
idea that presupposition satisfaction and truth behavior of a sentence are
two, partly independent \dimensions" which have to be taken into account,
was the intuition behind a four-valued approach presented in [49, 50]. We look
at this approach here also because the truth degrees get a very interesting
intuitive interpretation.
The truth degree set of this four-valued approach is the set f0; 1g2 of
ordered pairs. The background idea for this choice of truth degrees is the
observation that for the discussion of sentences like (20.1), (20.3) one has
to have in mind two important aspects: (i) on the one hand the problem
to assign a truth value to such a sentence, and (ii) on the other hand the
problem whether such a sentence refers to some presupposition, and whether
this presupposition7 is satised or not. Furthermore the feeling is that these
two aspects are (more or less) independent of one another. Hence one can
intend to evaluate these two aspects independent and in parallel { which
just gives the idea to have ordered pairs of these both evaluations as truth
degrees of sentences.8 In the actual discussion we decide to encode by the rst
component of such an ordered pair the truth value, and by the second one a
\soundness value" which indicates whether the presupposition which is tied
with the sentence under consideration is satised or not. For the particular
sentences we started with, this means that the \soundness value" evaluates
the satisfaction of the existence presupposition tied with these sentences.
Thus sentences (20.1) and (20.4) have to be evaluated by the pair (0; 0),
sentence (20.2) has to be evaluated by (0; 1), and sentence (20.3) has to be
evaluated by (1; 1). This is in accordance with our idea that some of our
sentences are negations of other ones if one characterizes the negation
which is present in these examples by the following truth degree table:
A (0; 0) (1; 0) (0; 1) (1; 1)
A (0; 0) (0; 0) (1; 1) (0; 1)

7 For simplicity we assume that such a sentence shall refer to at most one presup-
position. In practice this may seem to be too restrictive. However, considering

in the case of a reference to more than one presupposition the conjunction of all
these presuppositions as the \proper" presupposition reduces the more general
case to the present simpler one.
8 Of course, this idea of an parallel and independent evaluation of dierent aspects
which may be related with common language sentences is not bound to the
present particular case; cf. e.g. [261, 350].
500
With this interpretation, as already in the three-valued model, disappears

the problem that one possibly would have to introduce two (or even more)
intentionally dierent negation operations into the common language interpretations.
This four-valued approach is superior to the three-valued in dierent aspects. One point is that the four-valued approach allows a better separation of
the problems of \pure" truth and of the satisfaction of the presuppositions involved in the sentences. Another point is that the four-valued approach makes
it easy (and natural) to discuss also sentences which are true but which have
their (e.g. existential) presuppositions not satised. As an example consider:
Either Nicolas Bourbaki or Henry Poincare was a member of the
Academy of Sciences in Paris at age forty.
The crucial point is that this disjunctive sentence contains the non-designating
name \Nicolas Bourbaki" and nevertheless should be true because the second disjunct is true: Poincare was a member of the Academy of Sciences
in Paris already at age forty.
The understanding of disjunction _ which was just explained with the
last example is covered by the following truth degree table:
A _ B (0; 0) (1; 0) (0; 1) (1; 1)
(0; 0) (0; 0) (1; 0) (0; 0) (1; 0)
(1; 0) (1; 0) (1; 0) (1; 0) (1; 0)
(0; 1) (0; 0) (1; 0) (0; 1) (1; 1)
(1; 1) (1; 0) (1; 0) (1; 1) (1; 1)
This type of disjunction results in the product system C2 C2 if one considers a connective whose truth degree function is in its rst (truth related)
component determined by the disjunction of classical logic, and in its second (presupposition related) component by the conjunction of classical logic.
It is quite reasonable in the present situation to consider such a \strange"
connective in a product system: classical disjunction for the rst component
is nothing but standard for a disjunction in the product system, and conjunction in the second component just re ects the fact that the (existential)
presupposition of a disjunction should be the combination, i.e. conjunction
of the presuppositions of the disjuncts.
Similarly also an implication connective is introduced into this four-valued
product system in [49]: classical implication w.r.t. the rst components, and
classical conjunction for the second components of the truth degrees.
Another extension is also advisable for the application of this four-valued
system: its extension to the rst-order level. Then the language has to be
enriched by (restricted and unrestricted) quantications and should allow also
individual constants. The notion of interpretation which was introduced in
Section 2.5 needs a generalization in the present applicational context and has
to allow for empty sets as universes of discourse. Additionally it may happen
501
that an interpretation xes objects as meaning of individual constants only

for some of the constants of the language. The truth degree of a w then
has to be dened similar to (2.18), (2.20), and (2.22) with the additional
agreement that the \soundness value" has to become 0 in all the cases that
non-designating \names", i.e. non-designating individual constants, occur or
quantications over the empty domain.9
The intuition behind the choice of truth degrees in this four-valued model
makes it suggestive to consider (1; 1) as the only designated truth degree. This
xes a notion of logical validity for formulas of the formal language of the
product system C2 C2 in the realm of the present applications. Besides these
logically valid formulas, also the logically true and the logically sound ones
are of interest, i.e. those formulas whose truth degree has rst component, or
second respectively, equal to 1 for each interpretation and each valuation. In
the same sense a w H may be called true (or: sound ) in some interpretation
A i the truth degree of H has rst (or: second) component = 1 for each
A-assignment. Then one has a very natural possibility for a formal denition
of the presupposition relation in this context:
pres(B; A) =def B is true in each interpretation
(20.10)
in which A is sound.
This again is a metatheoretical approach toward presuppositions like previously in (20.7). It would also be possible to give a denition completely
inside the four-valued system, but we shall not discuss this here. It is more
important to mention that the intuition behind denition (20.10), that B is
a presupposition for A i the truth of B is necessary for the soundness of A,
is a very interesting renement of the previous analysis of the phenomenon
of presupposition as realized in (20.5).
The suitability of this four-valued approach toward presuppositional phenomena and its superiority over three-valued modeling approaches is in [50]
also explained in \only"-constructions like \Bill is driving only BMW cars",
in \even"-constructions like \Even Fred has a cold today", and for clefts like
\It was a bottle of red wine that Bob gave Lisa as a present yesterday".
It is not our goal to discuss whether these modeling approaches via systems of many-valued logic really grasp the core of the matter from the linguists point of view.10 What is interesting for many-valued logic, as for logic
at all, is the fact that the diversity of logical systems oers remarkable possibilities for rather diverse elds of applications outside logics itself and the
more standard application areas like mathematics and computer science.
9 This means either unrestricted quantications in case of an empty universe of
discourse, or restricted quantications with unsatisable restriction.
10 As a side remark it should be mentioned that also a \three-dimensional" eval-
uation has been proposed in [325], cf. also [354], intending that a truth degree
(a; b; c) of some w A should satisfy a + b + c = 1 and be understood in such a
way that a is the degree to which A is true, that b is the degree to which A is
false, and that c is the degree to which A is \senseless".
502
Successful applications of systems of (many-valued) logic in some areas

stimulate, on the other hand, logicians to take a closer look at those systems which are relevant for such applications. For the discussion of the phenomenon of presuppositions it is actually, however, not really clear whether
these modeling approaches via many-valued logic shall nally be accepted in
the linguistic community. Therefore we do not intend to undertake a deeper
study of the three- and four-valued systems which up to now have been used
for these applications from the structural point of view of formal logic.
21. Truth Degrees and Alethic Modalities
It is a remarkable fact that intended applications to the theory of the (alethic)

modalities \necessary" and \possible" have played a decisive role in the approach of Lukasiewicz [334] to found many-valued logic. Besides this initial
stimulation, however, later on investigations into the relationship between
many-valued and modal logic did have only a minor in uence in both these
elds of logic. And to a large extent they created only \negative" results indicating that neither a many-valued understanding of modalities, nor a modal
reading of truth degrees really helps to understand one of these elds from
the viewpoints of the other one.
But also another approach toward a combination of many-valued and
modal logic is possible: to enrich either many-valued logic with notions from
modal logic, or to go the other way round and to enrich modal logic with truth
degrees, i.e. to do something like many-valued modal logic. For a long time,
the latter approach appeared to be a completely theoretical exercise without
real applications. With AI applications in mind, however, this situation has
changed.
21.1 Interpreting Modal Logic as Many-Valued Logic

The initial stimulus for Lukasiewicz toward a many-valued reading of the
(alethic) modalities was the observation that the theory of these modalities
\necessary" and \possible" cannot suitably be given inside classical logic.
Trying to do this Lukasiewicz soon recognized, cf. [335], that the extension
of classical logic with (unary) connectives 2 for \it is necessary" and 3 for \it
is possible", and with logical principles one intuitively likes to have satised
for these modalities, results in contradictory systems. Therefore Lukasiewicz tried to realize the idea that a third truth value could help, e.g. if it is
read as \possibly true" and assigned to propositions which give statements
about future events. He explains his ideas as follows:1
I can assume without contradiction that my presence in Warsaw at
a certain moment of next year, e.g. at noon on 21 December, is at
1 The original source is [335, p.64]. The present quotation is from [338, p.165f.].
504
the present time determined neither positively nor negatively. Hence

it is possible, but not necessary, that I shall be present in Warsaw
at the given time. On this assumption the proposition \I shall be in
Warsaw at noon on 21 December of next year", can at the present
time be neither true nor false. For if it were true now, my future presence in Warsaw would have to be necessary, which is contradictory
to the assumption. If it were false now, on the other hand, my future
presence in Warsaw would have to be impossible, which is also contradictory to the assumption. Therefore the proposition considered
is at the moment neither true nor false and must possess a third
value, dierent from \0" or falsity and \1"' or truth. This value we
can designate by 21 . It represents \the possible", and joins \the true"
and \the false" as a third value.
This discussion nally led Lukasiewicz to his famous truth degree tables
which characterize the truth degree functions of the propositional Lukasiewicz system L3 in negation and implication, as considered in Chapter 9. For
the modal operators 2; 3 he chose the truth degree tables given in Fig. 21.1,
where we, for simplicity of notation, identify the modal operators with their
u 0 21 1
u 0 12 1
2(u) 0 0 1
3(u) 0 1 1
Fig. 21.1. Lukasiewicz's truth tables for the modal operators
truth degree functions. As Tarski has proved, cf. [335], one then has the
following equivalent characterizations available:
3p =def :p !L p ;
2p =def :(p !L :p) :
Soon after the introduction of the system L3 it became clear to Lukasiewicz
that he could similarly introduce the nitely many-valued systems Lm for
m > 3, and even the innitely many-valued ones L0 and L1 , identifying both
because of Theorem 9.1.2(d) and the fact that he discussed only propositional
systems. However he says2 that:
. . . among all the many-valued systems only two can claim any philosophical signicance: the three-valued and the innite-valued ones.
For if values other than \0" and \1" are interpreted as \the possible", only two cases can reasonably be distinguished: either one
assumes that there are no variations in degree of the possible and
consequently arrives at the three-valued system; or one assumes the
2 Again the original source is [335, p.72] and the present quotation from the trans-
lation [338, p. 173].
505
opposite, in which case it would be most natural to suppose (as in

the theory of probabilities) that there are innitely many degrees of
possibility, which leads to the innite-valued propositional calculus.
I believe that the latter system is preferable to all others.
More than twenty years later in [337] he came back to this topic of manyvalued interpretation of the modalities and constructed, again guided by intuitively motivated principles, a system of modal logic which forced him to
explicitly withdraw this conception.3 Instead he now declared that his new
system of modal logic \is a 4-valued system with two values . . . denoting possibility, but nevertheless, both values represent one and the same possibility
in two dierent shapes". The reason for this latter evaluation is that under
a suitable reading of the modal operators the set of theorems of this latter
system is just the set of tautologies of the product system C2 C2 of classical
propositional logic C2 with itself, cf. Section 8.3.
Despite these interesting ideas of Lukasiewicz, which in a quite early
phase of the development of many-valued as well as of modal logics tried
to connect both topics, both elds developed essentially independent of one
another. This indicates that the close relationship between both elds which
was conjectured by Lukasiewicz does not exist. At least one of the crucial
points for this discrepancy seems to be the fact that the modal operators
are not truth functional by their very nature, but many-valued logic is truth
functional. However, also actually the true nature of the relationship between
both areas of logic is not completely clear. This is nothing truly astonishing
and depends on the one hand from the fact that neither for many-valued nor
for modal logic one has exact borderlines which limit their respective areas,
and it depends on the other hand upon the diculty which one has even
to formulate, let alone to prove the impossibility of any positive relationship
between both elds.
But already in 1940 it was proved in [129] that there does not exist for any
one of the (early) main systems S 1; : : : ; S 5 of modal logic4 a corresponding
nitely many-valued system S such that the S-tautologies are just the theorems of the corresponding S -system of modal logic.
The systems S 1; : : : ; S 5 of modal logic can be determined by logical calculi which extend the usual propositional calculus for classical logic. One
possibility is to start within a propositional language with some set V0 of
3 Whether this withdrawal was really justied or not, however, seems to be essentially an open problem, also w.r.t. some more recent results of Woodruff [576]
and Duffy [128] which shall be explained later on in this section.
4 For modern presentations of modal logic the reader is referred to [98, 155, 282]
or to the corresponding chapters of [178, vol.2] and of [315]. The classic text is
[330]. One of the important dierences between both kinds of approaches is that
the older one [330] was based on a kind of implication connective dierent from
the usual \material" implication of classical logic and called strict implication.
The modern approaches, however, take the standard implication of classical logic
as basic connective.
506
propositional variables and the primitive connectives ; u; 3, to dene additional connectives by

H1 t H2 =def ( H1 u H2 ) ;
(21.1)
H1 H2 =def 3 (H1 u H2 ) ;
(21.2)
2H =def 3 H
(21.3)
for any ws H; H1 ; H2 of this language. One has to be careful here: the
implication connective does not correspond to the standard \material"
implication !, which as usual may be dened as
H1 ! H2 =def (H1 u H2 ) :
(21.4)
The connective is the strict implication and a typical modal connective, i.e.
an \intensional" one which cannot be represented by a truth value function
of classical logic. A syntactic determination of the systems S 1; : : : ; S 5 may
start from the axiom schemata
(M1) A u B B u A ;
(M2) A u B A ;
(M3) A A u A ;
(M4) (A u B ) u C A u (B u C ) ;
(M5) (A B ) u (B C ) (A C ) ;
(M6) A 3A ,
which suce for the system S 1, but have to be extended for the other systems
e.g. with one additional axiom schema for each system which may be
for S 2: 3(A u B ) 3A ;
for S 3: (A B ) (3A 3B ) ;
for S 4: 33A 3A ;
for S 5: 3A 23A .
The rules of inference coincide for all these systems and are the version of
the rule of detachment:
(MP) A; AB B
for strict implication, together with a rule for introducing a conjunction:
(Conj) AA;uBB
and with an exchange rule
B A; H
(Exch) A B;H
0
with H 0 formed from H by exchanging the subformula A of H at some places

of its occurrence in H by B .
These logical calculi which dene the systems S 1; : : : ; S 5 show that S i is
a subsystem of S j (in the sense of having a smaller set of theorems) i i j .
507
For a discussion of the problem whether there exists a nitely many-valued

system, which has as its set of tautologies just the set of theorems of S k for
some k = 1; : : : ; 5, it is helpful to have a further connective for the S -systems
available:
H1
H2 =def (H1 H2 ) u (H2 H1 ) :
From (M2), (M3), and (M5) one gets immediately that each w of the form
A A is a theorem of S 1, and hence also of all the other systems S k.
Because one has the inference rule (Conj) this means also that all ws of the
form A A are theorems of S 1. As an easy exercise one additionally shows
that all ws A A t B are theorems of S 1.
Suppose now that there exists some many-valued system S such that the
class of S-tautologies is the class of S k-theorems for some 1 k 5. Then S
has to have connectives $; _ which correspond to
and t. By the principle
of compositionality for S one has for all valuations : V0 ! W S :
(B1) if ValS (H1 ; ) = ValS (H2 ; ) then ValS (H1
H2 ; ) 2 DS ,
(B2) if ValS (H1 ; ) 2 DS or ValS (H2 ; ) 2 DS , then ValS (H1 t H2 ; ) 2 DS .
Consider for S the \same" w as was already discussed in (10.10) for the
Go del systems:
n =def
W
n_
;1 _
n
i=1 k=i+1
(pi $ pk )
with denoting the nite iteration of the S-connective _, and p1 ; : : : ; pn

pairwise dierent propositional variables. If S has only nitely many truth
degrees, say m, then one gets from (B1), (B2) immediately that each w
n with n > m is an S-tautology. If we now succeed in proving that the
modal versions of the ws n are not S 5-theorems, and hence also not S ktheorems for any k = 1; : : : ; 4, then it is clear that the class of S k-theorems,
k = 1; : : : ; 5, is dierent from the class of S-tautologies.
This last mentioned proof shall be given by another type of application
of ideas from many-valued logic which in more detail is explained later on in
Section 23.1: we construct some system of many-valued logic whose class of
tautologies includes (via some rereading of the modal connectives) the class
of S 5-theorems, and which has the property that some of the formulas n
are not tautologies of this system. Let E be some set of n elements and IP (E )
its power set. We take IP (E ) as a set of truth degrees, take E as the only
designated degree, read ; u as complement and meet in IP (E ), and dene
for each X E :

if X 6= ;
ver3(X ) =def E;
;; if X = ; .
Then one has to choose for all X; Y E

ver (X; Y ) = E; if X \ (E n Y ) = ;
; otherwise
508
to get that the above mentioned reading of u is in accordance with (21.2)

and (21.1). Now it is a routine matter to check that all of the modal axioms
(M1), . . . , (M6), as well as the additional axioms characterizing S 2, . . . ,
S5 are tautologies for this system. However no w r with r < 2n is a
tautology of this system. So we have that none of the ws n is an S 5theorem. Therefore neither S 5 nor any one of the systems S 4; : : : ; S 1 can be
interpreted as a nitely many-valued system.
On the other hand it is possible to nd a many-valued system which has
as its set of tautologies the set of S 5-theorems. Of course, this system has to
have innitely many truth degrees. We start with the product system C1 =
Q1
i=0 C2 of innitely many copies of classical propositional logic (formulated
in negation and conjunction only). Let (>; >; >; : : :) be the only designated
truth degree in the set Nf>; ?g of all truth degrees of C1 . The truth degree
function for the modal operator 2 shall be

; >; >; : : :); if u 6= (?; ?; ?; : : :)
ver3(u) =def ((>
?; ?; ?; : : :); if u = (?; ?; ?; : : :) ,
and the truth degree functions for implication and negation shall be dened
componentwise. One gets an intuitive understanding for C1 if one reads
the truth degrees (tk )k0 as \shorthand" notations for two-sided innite sequences (: : : ; u;3 ; u;2 ; u;1; u0 ; u1; u2 ; : : :), with each such sequence coded by
the \simply" innite sequence (u0 ; u;1 ; u1 ; u;2; u2 ; : : :), such that each one
of the truth values uj for ;1 < j < 1 is related with some time moment j
in the sense that
Val (H; ) = (: : : ; u;2 ; u;1 ; u0; u1 ; u2 ; : : :)
says that the w H has always truth-value uj in the time moment j . One
hence has in mind a temporal reading of C1 which is based on the understanding that a w 2H has the designated truth degree (>; >; >; : : :) in this
system i H is true in each time moment.5 It is not hard to see that each
theorem of S 5 is (via the necessary rereading of the connectives) a tautology
of C1 . And this holds also conversely, i.e. the set of S 5-theorems coincides
with the set of C1 -tautologies.6 Thus we have found a reading of the modal
system S 5 as an innitely many-valued system.
Particularly the system S 5 seems to be of interest for a kind of justication
of the rst, three-valued approach toward modal logic given by Lukasiewicz.
The background for this type of \revival" is the possibility of another look
5 Additionally one presupposes here that time has a discrete, linear, and two-
sided unlimited structure, i.e. consists of a series of time points without rst and
last one and without branching which hence can be enumerated with the whole
numbers.
6 For the proof of this converse implication the reader is referred to the literature
on temporal logic, e.g. to [430, 449] or [178, vol.2]. There he may nd also other
results connecting systems of modal with systems of temporal logic, and connecting systems of temporal and of many-valued logics which shall not be discussed
here.
509
at the relationships between the (propositional) Lukasiewicz systems and

(propositional) modal logic: to provide \modal readings" for the ws of the
L-systems. Woodruff [576] was the rst to do this. He gave a translation
m of the ws of LL into ws of LS5 determined for any ws H; G of LL and
all propositional variables p by the clauses:
m(p) =def p ;
m(H !L G) =def (2m(H ) ! m(G)) u (m(H ) ! 3m(G)) ;
m(:H ) =def m(H ) ;
where he refers to the (standard \material") implication ! instead of the
strict implication .7
To see that this translation immediately provides a semantical interpretation of L3 in modal terms of the system S 5, we rst have to look at a semantics for S 5. The basic structures for such a semantics are S 5-Kripke-frames
hW; Ri consisting of a non-empty set W of \worlds" and a binary accessibility
relation R W W which in the case of these S 5-Kripke-frames has to
be an equivalence relation. The valuations then are either all the mappings
v : V0 W ! f>; ?g or, equivalently, all the mappings v~ : V0 ! IP (W )
always w.r.t. any S 5-Kripke-frame hW; Ri.
As usual, these valuations are extended to mappings, which either provide
a truth value v(H; w) for each pair consisting of a w H and a world w, or
provide a subset v~(H ) of W for each w H . The extension of v follows the
standard approach in modal logic:
v( H; w) = > , v(H; w) = ? ;
v(H ! G; w) = > , v(H; w) = ? _ v(G; w) = > ;
v(2H; w) = > , v(H; w0 ) = > for all w0 2 W with wRw0 :
Correspondingly one has for the other type of valuation:
v~( H ) = W n v~(H ) ;
v~(H ! G) =
(W n v~(H )) [ v~(G) = W n (~v (H ) n v~(G)) ;
if v~(H ) = W
v~(2H ) = W;
; otherwise .
Furthermore one has that some w H is valid in an S 5-Kripke-frame hW; Ri
i either one has v(H; w) = > for all w 2 W and all valuations v, or i one
has v~(H ) = W for all valuations v~. And nally H is S 5-valid i it is valid in
all S 5-Kripke-frames.
Both types of valuations are equivalent for S 5. For the present purposes
we shall restrict the further considerations to the second type of valuations
which map into IP (W ) and do not mention the single worlds separately.
7 This depends on the fact that the modern presentations of (S 4 and) S 5 use
another type of axiomatizations as we have given. These modern axiomatizations,

however, do not cover S 1 which actually is considered as a system of minor
importance.
510
Coming back to the translation m, we connect with each S 5-valuation v~

w.r.t. some S 5-Kripke-frame hW; Ri an L3 -valuation v~m dened by
8
< 1; if v~(m(H )) = W
(21.5)
v~m (H ) =def : 0; if v~(m(H )) = ;
1 otherwise
2
for each w H . Then one has the following basic result.
Proposition 21.1.1. For each S5-valuation v~ w.r.t. an S5-Kripke-frame
hW; Ri the mapping v~m is an L3 -valuation. And for each L3 -valuation vL
there exists some S 5-valuation v~ w.r.t. some S 5-Kripke-frame hW; Ri with
card (W ) = 2 such that vL = v~ m .
Proof: Suppose that v~ is an S5-valuation w.r.t. the frame hW; Ri. Then
it is a routine matter to show that v~m is an L3 -valuation, having in mind
that one has v~(H u G) = v~(H ) \ v~(G) as well as: v~(3H ) = W i v~(H ) 6= ;,
and v~(3H ) = ; otherwise.
Now let vL be an L3 -valuation and W0 = fa; bg for some objects a 6= b.
Dene the S 5-valuation v~ w.r.t. the frame hW0 ; Ri by
8
< W0 ; if v (p) = 1
v~ (p) =def : ; ; if v(p) = 0
fag ; if v(p) = 12
for each propositional variable p. Then it is again a routine matter to show
vL = v~ m .
2
Now the main result for this translation business reads:
Theorem 21.1.1. For each w H of LL the following assertions are equivalent:
(a) H is L3 -valid,
(b) m(H ) is S 5-valid,
(c) m(H ) is valid w.r.t. any S 5-Kripke-frame hW; Ri with card (W ) = 2.
Proof: (a) implies (b) by the rst part of the preceding proposition. (b)
implies (c) by denition of S 5-validity. And (c) implies (a) by the second
part of the preceding proposition.
2
One of the diculties of this translation, however, is that it does not
preserve semantic equivalence. This can be seen in looking at conjunctions in
LL . Obviously one has
H ^ G semantically equivalent G ^ H
in each one of the L-systems. In S 5, however, the ws m(H ^G) and m(G^H )
need not be semantically equivalent. It e.g. suces to look at the case that
H p and G :p. Then one gets by straightforward calculations, having in
mind that the non-modal connectives of S 5 behave classically, rst

511
m(p ^ :p) ;2( p ! 3p) ! p u ;( p ! 3p) ! 3p ;

m(:p ^ p) ;2(p ! 3 p) ! p u ;(p ! 3 p) ! 3 p :
From this one gets, using (i) the fact that the modal ws 2(H t 3G) and
2H t 3G always are semantically equivalent in S 5, and that (ii) each modal
formula H ! 3H is S 5-valid, also
m(p ^ :p) sem.eq. (2p t 3p ! p) ;
m(:p ^ p) sem.eq. (2 p t 3 p ! p) :
And having in mind that furthermore each w 2H ! H is S 5-valid, one
nally has
m(p ^ :p) sem.eq. (3p ! p) sem.eq. 3p u p ;
m(:p ^ p) sem.eq. (3 p ! p) sem.eq. 3 p u p ;
and can see that these two modal formula are not semantically equivalent in
the system S 5.
This non-preservation of semantic equivalence is a strong hint toward the
fact that the intended reading of the modalities in Lukasiewicz's understanding seems to be very dierent from the intuitive understanding of the
modalities in the \usual" systems of modal logic.
This translation topic was further studied in [128] where a list8 is given
of eight (modal) interpretations f of L3 in S 5, including the Woodruff
interpretation m, which all satisfy for every w H of LL the conditions:9
H L3 -valid , f (H ) S 5-valid ;
H L3 -contravalid , f (H ) S 5-contravalid :
None of these mappings preserves semantic equivalence. Therefore in [128]
also translations of Slupecki's functionally complete extension LSl
3 into S 5
(enriched with a sentential constant w always interpreted as a designated
world in a Kripke frame with a designated world10) are discussed. And it is
shown that there exist exactly 3042 such interpretations in total, with just
two of them preserving semantic equivalence.
There are further interesting results which relate systems of modal and
of temporal logics, and which can be used to extend such relationships to
8 Even an exhaustive list under some mild restrictive conditions given in detail in
[128].
9 In the following statement we mean by \contravalidity" in the case of the L-
systems, to have always truth degree 0, and in the case of S 5, to be false in all
worlds of all S 5-Kripke-frames, or to have the emptyset as value for all valuations v~. The Woodruff translation has this property automatically because it
translates L-negations as S 5-negations.
10 And only such \pointed" Kripke frames are considered in this case. This is
not a severe restriction because a completeness theorem holds true for S 5 also
w.r.t. the class of all pointed Kripke frames with an equivalence relation as their
accessibility relation.
512
connections between systems of modal and of many-valued logics { resulting

also in further cases in many-valued readings of modal systems. The whole
linkage between these dierent elds of logic does not, however, reduce one
of these elds to another one or subsumes it into another: not all modal
systems have suitable many-valued interpretations, nor has each system of
many-valued logic some suitable modal or temporal reading. Thus all three
areas each have their own avor and merits for applications.
21.2 Graded Modalities

The second type of approach toward a combination of many-valued and modal
logics, which was mentioned in the introductory remarks for this chapter,
intends to allow not only for a grading of truth, but also for a grading of
necessity and possibility. It is a kind of common agreement, actually, that
in such a case the truth degrees and the degrees for these modalities should
coincide.
Again it is preferable here to start from a semantic approach, and not from
a (purely) syntactic one. To discuss this matter more in detail one should have
in mind that the most famous approach toward semantics of (normal) modal
logics is the approach via Kripke frames which consist { in the classical case
{ of a set of so-called possible worlds11 together with a (binary) accessibility
relation between them. And such a Kripke frame becomes a (propositional)
Kripke model i each one of its possible worlds is combined with a truth
value assignment.
A suitable many-valued generalization then should combine for its generalized Kripke frames a set12 of possible worlds with a many-valued accessibility relation. And a generalized Kripke model should combine each one of
these possible worlds with a truth degree assignment.
The formalized (propositional) language LS one is going to use here has,
as a language of modal logic, besides all the particular means of expression
which have been discussed in Chapter 2, in any case two modal operators 2; 3
which (syntactically) behave like unary connectives. Of course, the choice of
its set J S of connectives otherwise is either determined from the system S
11 These possible worlds should not become combined with any kind of ontological
understanding in a deeper philosophical sense of the notion \world". They can

simply be considered as a set of objects, and can be interpreted in quite dierent
ways in particular applications. They can, e.g. be taken as states of knowledge or
information, with the accessibility relation understood as completion of knowledge (without forgetting) or the addition of information, or they can be taken
as possible stages within some { e.g. non-deterministic { computational process,
with the accessibility relation read as a continuation of the computation.
12 One could also try to take instead a fuzzy set of possible worlds. Actually, however, it is not clear whether this would give a generalization with useful applications.
513
one is supposing to be given, or is connected with the (algebraic) structure

of the set of truth degrees as mentioned in Chapter 2.
A possible applicational situation for such a system of many-valued modal
logic was designed by Fitting e.g. in [166, 167],13 based on the assumption
that one has some set A of communicating \agents" (or \experts"), and that
these agents are additionally linked together by a dominance relation , such
that a1 a2 means that agent a1 dominates agent a2 . For simplicity, this
dominance relation shall be re exive. Each one of these agents a is supposed
to be concerned with some Kripke frame. All agents are bound to have the
same set W of possible worlds. They are allowed, however, to have dierent
(classical) accessibility relations Ra in W . Each agent a asserts, i.e. claims as
(classically) true, in each possible world w 2 W some { atomic { sentences
H , and does not assert other ones, i.e. does not claim their truth in w. The
\private" behavior of each one of these agents is supposed to be two-valued.
Therefore all those atomic sentences H which are asserted in w by a are
treated as being evaluated by a with the truth value > in w, written as
va (H; w) = >, and the atomic sentences H which are not asserted by a in w
are treated as being evaluated by a with the truth value ? in w, written as
va (H; w) = ?.
And one assumes that the communication between the agents is in a
natural way linked with their dominance relation. This shall mean that in
the case of dominance a b of some agent a over some other agent b this
dominated agent b asserts an atomic sentence H in a possible world w in the
case that the dominating agent a asserts H in w:
a b ^ va (H; w) = > ) vb (H; w) = > :
(21.6)
In the case, however, that the dominating agent a does not assert H in w,
the dominated agent b is free either to assert H in w or not to assert H in
w. Accordingly one also assumes that in such a dominance case a b an
accessibility relationship w1 Ra w2 which is \true" for the dominating agent a
is also \true" for the dominated agent b:
a b ) Ra Rb :
(21.7)
Furthermore one likes to have a kind of rational behavior for all these agents
regarding the truth values they ascribe to compound formulas. This is not a
problem for compound formulas which are conjunctions or disjunctions. The
approaches
va (H1 ^ H2 ; w) = > , va (H1 ; w) = > ^ va (H2 ; w) = > ;
(21.8)
va (H1 _ H2 ; w) = > , va (H1 ; w) = > _ va (H2 ; w) = >
(21.9)
13 Other systems of many-valued modal logic have been considered e.g. in [268,
380, 381, 382, 392, 393, 394, 405, 406, 407, 484, 497, 524, 531, 536, 568, 581].
However, they often rely only a binary accessibility relation, or do not come with
as convincing an application as the systems in [166, 167]. So we shall not discuss
them in detail here.
514
are in accordance with the dominance ideas: from a b and va (H1 ^ H2 ; w) =

> one has vb (H1 ; w) = vb (H2 ; w) = > by (21.8) and (21.6), and hence also
vb (H1 ^ H2 ; w) = > by (21.8). And from a b and va (H1 _ H2 ; w) = > one
has vb (H1 ; w) = > or vb (H2 ; w) = > by (21.9) and (21.6), and hence also
vb (H1 _ H2 ; w) = > by (21.9).
The matter is more dicult for the cases that the compound formulas are
negations or implications. We discuss primarily the case of the evaluation of
an implication, because it is then (almost) a routine matter to cover the case
of a negation by
va (:H; w) = > , va (H ! ?; w) = > ;
which only presupposes that one has the truth value constant ? available in
the language. So let a b and consider the implication H1 ! H2 . If one
has va (H2 ; w) = >, then one has also vb (H2 ; w) = > from (21.6), and should
therefore have va (H1 ! H2 ; w) = vb (H1 ! H2 ; w) = > if one e.g. evaluates
an implication as true if it has a true succedent. So (21.6) does not create a
problem in this case. But now assume va (H2 ; w) = ?. Then agent b is free
in his evaluation of H2 in world w. If he chooses vb (H2 ; w) = > then he
should also choose vb (H1 ! H2 ; w) = > and cannot get into con ict with
(21.6). However, if he chooses vb (H2 ; w) = >, and if additionally agent a
has va (H1 ; w) = ?, then the dominated agent b may choose vb (H1 ; w) = >
which makes (in world w) for him H1 ! H2 into an implication with a true
antecedent and a false succedent { which should allow him to reject it, i.e. to
evaluate vb (H1 ! H2 ; w) = ?. But, agent a has in the world w the situation
that for him H1 ! H2 is an implication with a false antecedent and a false
succedent. If he now evaluates H1 ! H2 as in classical logic, he has to take
va (H1 ! H2 ; w) = >, which would force b also to have vb (H1 ! H2 ; w) = >
{ or to violate (21.6). This means that we cannot suppose that agent a does
the evaluation of the implications as in classical logic. The crucial point is
that he has to take care of the evaluations which the agents he dominates give
to the antecedent and the succedent of the implication under consideration.
A way out is to regulate the evaluation according to the rule:
va (H1 ! H2 ; w) = > , vb (H1 ; w) = ? _ vb (H2 ; w) = > (21.10)
for all b 2 A with a b
as was done in [166, 167]. Having chosen this approach, one has for the case
of a negation in more detail
va (:H; w) = > , vb (H; w) = ? for all b 2 A with a b : (21.11)
And one immediately observes that one has also
va (H1 ! H2 ; w) = > ^ va (H1 ; w) = > ) va (H2 ; w) = > ;
i.e. each agent can apply in each possible world the rule of detachment.
For the modal operators one meets a similar diculty. The possibility
operator does not lead into diculties: having va (3H; w) = > understood as
515
saying that one has va (H; w0 ) = > for some possible world w0 with wRa w0 ,
forces that also each agent b dominated by a has to evaluate vb (3H; w) = >,
because one has vb (H; w0 ) = > from (21.6) and wRb w0 from (21.7). However,
to accept va (2H; w) = > in accordance with (21.6) and the fact that all
agents should follow the same rules for their \private" evaluations, it is not
enough to have only va (H; w0 ) = > for all worlds w0 with wRa w0 . Instead one
should take care that also for all the agents b dominated by a one has that
for them from vb (2H; w) = > it follows that vb (H; w00 ) = > for all worlds w00
with wRb w00 . Together with Ra Rb in this case, this leads to the following
general evaluation rules:
va (3H; w) = >, va (H; w0 ) = > for some w0 2 W with wRa w0 (21.12)
va (2H; w) = >, vb (H; w ) = > for all b 2 A with a b and (21.13)
all w 2 W with wRb w .
Intuitively the rules (21.10) and (21.13) mean that the communication between the agents has to be more extended as necessary for (21.6) only: each
agent has to know for the actual world w and the actual sentence H under consideration for each one of the agents b, he dominates, the evaluations
which b gives to all the immediate constituents of H in either w, or even in
all worlds which are accessible from w in the sense of b. So each agent has to
have complete knowledge of the evaluations and the accessibility relations of
all the agents he dominates.
The rule (21.10) as well as the rule (21.13) deviate from the evaluation
behavior of classical logic. They correspond, instead, to the behavior in the
setting of intuitionistic logic.
Now one is interested in a suitable \unication" of the evaluations of all
these experts, and of all of their \private" accessibility relations. First it is
rather natural to construct a common graded, i.e. many-valued accessibility
relation R in W in such a way that one takes for any pair (w1 ; w2 ) of possible
worlds as truth degree [ w1 Rw2 ] of the (common) accessibility of w2 from w1
the set of all agents a which have the accessibility w1 Ra w2 within their own
Kripke frame. In the same sense it is natural to take as a unied, common
evaluation of a w H in a world w the set fa 2 A j va (H; w) = >g of all
agents which consider H in w as true.
To form the corresponding many-valued Kripke models one has to add
for every possible world w a WA -valuation w . This is, of course, the same as
if one adds to some many-valued Kripke frame a mapping : W V0 ! WA .
In the present setting such a WA -valuation should be understood as a kind
of superposition of the family of \private" Kripke models of the agents, in
the sense that the reading of the fact that some propositional variable p gets
under in a possible world w some value t 2 WA is that just the agents a 2 t
give in their own (classical) evaluations the truth value > to the variable p
in the world w. More formally hence we put
(w; p) =def fa 2 A j va (p; w) = >g ;
(21.14)
516
and like to have the corresponding property satised also for compound formulas. This means we like to get the truth degrees in general determined by
the property
Val (w; H ; ) = fa 2 A j va (H; w) = >g ;
(21.15)
which should be satised in the many-valued Kripke model under consideration.
On the other hand, having given some WA -valuation : W V0 ! WA ,
this shall mean that an agent a evaluates some w H in a world w under
by > i a 2 Val (w; H ; ).
The present approach supports the viewpoint to take (only) a suitable
subclass WA of the power set IP (A) of the set of agents as the set of truth
degrees. These truth degrees should be those sets of agents which are closed
under the dominance relation, in the sense that always if some agent a is an
element of a truth degree t also all those agents b have to be elements of t
which are dominated by a.
The choice of this set WA of -closed truth degrees has to be accompanied by the choice of suitable operations in WA which correspond to the
logical connectives present in the language, in such a way that (21.15) becomes satised and that the degrees Val (w; H ; ) of compound formulas H
are determined from the corresponding degrees of the constituents of H by
these operations.
Because of (21.8) and (21.9) one has for each agent a and all ws H1 ; H2 :
fa 2 A j va (H1 ^ H2 ; w) = >g
= fa 2 A j va (H1 ; w) = >g \ fa 2 A j va (H2 ; w) = >g ;
fa 2 A j va (H1 _ H2 ; w) = >g
= fa 2 A j va (H1 ; w) = >g [ fa 2 A j va (H2 ; w) = >g ;
and hence has to combine with conjunction and disjunction as corresponding
operations in WA the usual set theoretic operations of intersection and union.
This means that WA should be closed under intersection and union, and hence
become (at least) a lattice, with inclusion as its lattice ordering. Then one
obviously even has
Val (w; H1 ^ H2 ; ) = Val (w; H1 ; ) \ Val (w; H2 ; ) ;
(21.16)
Val (w; H1 _ H2 ; ) = Val (w; H1 ; ) [ Val (w; H2 ; ) :
(21.17)
To get for the implication ! a corresponding operation # with property
(21.10), consider any -closed set X of agents. Let w be any world, and
any valuation.
Suppose rst that X Val (w; H1 ! H2 ; ) and that a 2 X . If one
then also has a 2 Val (w; H1 ; ), i.e. if one has va (H1 ! H2 ; w) = > and
va (H1 ; w) = > under the valuation , then one even has va (H2 ; w) = >
under , which means a 2 Val (w; H2 ; ). Hence one has X \ Val (w; H1 ; )
Val (w; H2 ; ).
517
Suppose on the other hand X \ Val (w; H1 ; ) Val (w; H2 ; ), and let
a 2 X . Then one, of course, has a 2 Val (w; H2 ; ) or a 2= Val(w; H2 ; ). Now
consider any agent b with a b. Then one has also b 2 X because X is closed
under dominance. In the case a 2 Val (w; H2 ; ) one has va (H2 ; w) = > under
, and thus also vb (H2 ; w) = > under by (21.6). Thus one has vb (H1 ; w) =
? _ vb (H2 ; w) = > under in this case. In the other case a 2= Val(w; H2 ; )
there is no restriction from (21.6) for agent b regarding the truth value he may
give to H2 in w under . In the case he has vb (H2 ; w) = > under , one again
has vb (H1 ; w) = ? _ vb (H2 ; w) = > under . And in the case that agent b has
vb (H2 ; w) = ? under , he has to have also vb (H1 ; w) = ? under : otherwise
from b 2 X and vb (H1 ; w) = > under one would have b 2 X \ Val (w; H1 ; ),
hence b 2 Val (w; H2 ; ) by assumption, and thus a contradiction. So one has
also in this nal case vb (H1 ; w) = ? _ vb (H2 ; w) = > under . Thus one has
vb (H1 ; w) = ? _ vb (H2 ; w) = > for each agent b which is dominated by a,
and each valuation . And this means a 2 Val (w; H1 ! H2 ; ).
With the intention in mind that

Val (w; H1 ! H2 ; ) = Val (w; H1 ; ) # Val (w; H2 ; )
(21.18)
should hold true, one therefore has at all for each X 2 WA :
X Val (w; H1 ; ) # Val (w; H2 ; )
, X \ Val (w; H1 ; ) Val (w; H2 ; ) :
(21.19)
And this is just the adjointness condition (5.87), which means that the operation # is the pseudo-complement in the lattice WA . Hence this lattice
is a pseudo-complemented one, i.e. a Heyting algebra. And one gets from
Proposition 5.4.1(i) the representation
Val (w; H1 ! H2 ; ) =
[
fX 2 WA j X \ Val (w; H1 ; ) Val (w; H2 ; )g :
(21.20)
For the negation this means immediately that one has
[
Val (w; :H ; ) = fX 2 WA j X \ Val (w; H1 ; ) = ;g :
(21.21)
To discuss also the modal operators denote rst for all possible worlds
w1 ; w2 by [ w1 Rw2 ] the truth degree of the accessibility of w2 from w1 . Then
one rst has for all agents a, all ws H , all worlds w, and all valuations
a 2 Val (w; 3H ; )
, va (3H; w) = > under
, for some w0 with wRa w0 : va (H; w) = > under
, there exists some w0 with a 2 [ wRw0 ] and a 2 Val (w0 ; H ; )
from (21.12) and (21.15). And this means
[ ;

Val (w; 3H ; ) =
[ wRw0 ] \ Val (w0 ; H ; ) :
(21.22)
w0 2W
518
And one also has from the re exivity a a together with (21.13) and (21.15)
a 2 Val (w; 2H ; ) ^ a 2 [ wRw0 ] ) a 2 Val (w0 ; H ; ) ;
i.e. one has
Val (w; 2H ; ) \ [ wRw0 ] Val(w0 ; H ; ) :
By (21.19) this means
Val (w; 2H ; ) [ wRw0 ] # Val (w0 ; H ; )
for each world w0 , and hence
\ ;

Val (w; 2H ; )
[ wRw0 ] # Val (w0 ; H ; ) :
w 0 2W
In the case that one would not have equality here, there had to exist some
agent a 2 A such that a 2= Val (w; 2H ; ), but
\ ;

a2
[ wRw0 ] # Val (w0 ; H ; ) 2 WA :
w0 2W
This means for each agent b with a b that also

\ ;

b2
[ wRw0 ] # Val (w0 ; H ; ) ;
w0 2W
i.e. that one has [ wRw ] # Val (w ; H ; ) for each world w , and therefore
[
b 2 fX 2 WA j X \ [ wRw ] Val (w ; H ; )g :
Hence there exists some X0 2 WA such that b 2 X0 and X0 \ [ wRw ]
Val (w ; H ; ). This means that for each world w with wRb w , i.e. with
b 2 [ wRw ] one has b 2 Val (w ; H ; ), i.e. vb (H; w ) = > under . But then
one has va (2H; w) = > from (21.13), i.e. a 2 Val (w; 2H ; ), a contradiction.
So one has
\ ;

Val (w; 2H ; ) =
[ wRw0 ] # Val (w0 ; H ; ) :
(21.23)
w0 2W
And this means that in the many-valued Kripke frame under consideration
the determination of the truth degrees of ws of the forms 3H and 2H is
very similar to the two-valued case.
In general, thus, one comes from the approach via a set of agents with
a dominance relation among them to a generalized notion of Kripke frame
and of Kripke model which both have a many-valued accessibility relation,
and which evaluate their propositional variables (and hence their ws) with
truth degrees from a complete Heyting algebra.
It is interesting that one can reverse this relationship, at least as long as
the Heyting algebra involved here is nite, i.e. as long as the set of agents
is nite. Suppose that hW; v; i is a generalized Kripke model with a manyvalued accessibility relation v with truth degrees from a nite Heyting
519
algebra G = hG; u; t; ; 0; 1i, and with a G-valuation : W V0 ! G.

And suppose that this G-valuation is extended to compound formulas by
(21.16), (21.17), (21.20), (21.21), (21.22), and (21.23), of course always with
the operations from the Heyting algebra G in place of the corresponding
operations from the Heyting algebra WA .
Consider as a set of agents (for this situation) the set AG of all prime
lters of G together with the inclusion relation = as dominance relation.
Let each one of these agents a 2 AG have W as its set of possible worlds,
and the related \private" accessibility relation
R~a = f(w1 ; w2 ) j [ w1 v w2 ] 2 ag :
In the same sense each agent a shall have the \private", two-valued evaluation
va under , which is determined by
v~a (H; w) = > , Val (w; H ; ) 2 a :
Then one has immediately
a b ^ v~a (H; w) = > ) v~b (H; w) = >
for all agents a; b, all worlds w, and all ws H , because a b means a b.
So (21.6) is satised in the present situation. Furthermore one has e.g. for a
disjunction
v~a (H1 _ H2 ; w) = > , Val (w; H1 _ H2 ; ) 2 a
, Val (w; H1 ; ) t Val (w; H2 ; ) 2 a
, Val (w; H1 ; ) 2 a _ Val (w; H2 ; ) 2 a
from (21.17) and the fact that a is a prime lter of G. Within an inductive
proof this is just (21.9) for the present situation. Similarly one gets for a
conjunction the result which corresponds to (21.8).
In the case of an implication let v~a (H1 ! H2 ; w) = > under , i.e.
Val (w; H1 ! H2 ; ) 2 a. And let a b, i.e. a b. Then one has Val (w; H1 !
H2 ; ) 2 b. So assume v~b (H1 ; w) 6= ? under , which means v~b (H1 ; w) = >
and hence Val (w; H1 ; ) 2 b. With (21.18) this means
;

Val (w; H1 ; ) u Val (w; H1 ; ) Val (w; H2 ; ) 6 Val (w; H2 ; )
and hence Val (w; H2 ; ) 2 b, i.e. v~b (H2 ; w) = > under , because one has
x u (x y) 6 y generally valid in each Heyting algebra.
If one has, on the other hand, v~b (H1 ; w) = ?__ v~b (H2 ; w) = > under for
each agent b a, then one has Val (w; H1 ; ) 2= b or Val (w; H2 ; ) 2 b for all
prime lters b a. In; this case one has to show v~a (H1 ! H2 ; w) = > under
, i.e. has to show Val (w; H1 ; ) Val (w; H2 ; ) 2 a. Assume instead
;
Val (w; H1 ; ) Val (w; H2 ; ) 2= a. Then a [ fVal (w; H1 ; )g generates a
proper lter F1 in G, namely
F1 = fy j x u Val (w; H1 ; ) 6 y for some x 2 Gg ;
520
because one has always x u Val (w; H1 ; ) 6= 0 for x 2 G. (In the case that
x0 u Val (w; H1 ; ) = 0 for some x0 2 G; one would instead have thatx0 u
Val (w; H1 ; ) 6; Val (w; H2 ; ), i.e. x0 6 Val (w;H1 ; ) Val (w; H2 ; ) and
therefore also Val(w; H1 ; ) Val (w; H2 ; ) 2 a, a contradiction.) And
this lter F1 has the property Val (w; H2 ; ; ) 2= F1 because otherwise there

would exist some y0 2 G such that y0 6 Val (w; H1 ; ) Val (w; H2 ; ) .
From; the adjointness condition (HA2)
of Denition 10.1.1 one then again
had Val (w; H1 ; ) Val(w; H2 ; ) 2 a, a contradiction. From Lemma
10.1.1(a) it now follows that there exists some prime lter b F1 a such
that Val(w; H1 ; ) 2 b and Val (w; H2 ; ) 2= b. But this means v~b (H1 ; w) 6= ?
as well as v~b (H2 ; w) 6= > under , contradicting the global assumption. Hence
v~a (H1 ! H2 ; w) = > under .
So it remains to consider the modal operators. For 3 one to gets (21.12)
from (21.22) by routine calculations . So consider (21.13). Assume rst that
v~a (2H; w) 6= > under , i.e. that Val (w; 2H ; ) 2= a. Then there exists,
to the characterization
(21.23), some world w 2 W such that
;according

[ wRw ] # Val (w ; H ; ) 2= a, because from the niteness of the Heyting
algebra G it follows that the right hand side of (21.23) is a nite intersection,
and because the lter a is closed under nite intersections. Just as above with
the construction of the lter F1 , this implies that there exists a prime lter
b0 a with [ wRw ] 2 b0 and Val (w ; H ; )) 2= b0. But then one has wRb w ,
as well as v~b (H; w ) 6= > under .
Now assume conversely that there exist an agent c a and some possible
world w 2 W such that wRc w and v~c (H; w; ) 6= > under . Then one has
[ wRw ] 2 Tc and Val
(w ; H ; )) 2= c, hence [ wRw ] # Val (w ; H ; ) 2= c.
;
This gives w0 2W [ wRw0 ] # Val(w0 ; H ; ) 2= c, i.e. Val (w; 2H ; ) 2= c. Via
a c one thus has Val (w; 2H ; ) 2= a, i.e. v~a (2H; w) 6= > under .
Therefore one can in a nice, back and forth translatable manner link
within a many-valued setting a natural generalization of the standard Kripke
type semantics with a multi-agent/multi-expert based semantical setting,
which both determine the same system SF of many-valued modal logic. The
nite Heyting algebra G, which is the truth degree structure, is supposed
to be xed in advance.
The notions of logical validity and logical entailment for SF then are
formed as usual in the many-valued and the modal settings: being valid in
some generalized Kripke model W, or having this generalized Kripke model
as a model means for a w H of the corresponding modal language to have a
designated truth degree in each one of the possible worlds of W. Likewise, being logically valid means being valid in each such generalized Kripke model.
And being logically entailed by a set of ws means for H that each model
W of is also a model of H . The standard choice of the designated truth
degrees furthermore is to have either only the unit element 1 of the Heyting
algebra G as designated truth degree, or to have the set DSF of designated
0
521
truth degrees closed under the lattice join of u, i.e. to have DSF as a principal
lter hdi of some element d 2 G.
What remains to be considered are logical calculi which adequately axiomatize these semantic notions. As such logical calculi, there are considered
sequent calculi in [166], and a tableau calculus in [167]. In both cases the
formalized modal language LSF is supposed to have
J SF = f^; _ !; :; 3; 2g
as its set of connectives, and to have a truth degree constant u^ for each truth
degree u 2 G. This makes it easy to code the fact that some w H should
have a truth degree [ H ] > u (under some valuation, and in some possible
world) by demanding that the implication u^ ! H has the truth degree 1,
because the adjointness condition (HA2) immediately gives
u6v , 1=u#v
for all truth degrees u; v 2 G. Likewise an inequality [ H ] 6 u can be coded
by the claim that H ! u^ has the truth degree 1. Therefore, implications get
a crucial role in the rules and axioms of these calculi. And one even adapts
the terminology and calls ws of the kinds u^ ! H and H ! u^ bounding
formulas. Furthermore, [166] as well as [167] use signed formulas. However,
contrary to the situation discussed in Chapter 7, the signs here are neither
the truth degrees, nor general sets of truth degrees: only the set of designated
truth degrees and the set of undesignated truth degrees appear as signs { and
are in this rôle denoted by T and F .
So the tableau calculus of [167] works only with signed bounding formulas
of the forms s : (û ! H ) and s : (H ! u^) with truth degrees u 2 G,
signs s 2 fT; F g, and ws H of the language LSF . Of course, this syntactic
restriction suces to state for each w its SF -logical validity, because one
obviously has that a w H is SF -logically valid i the bounding formula
d^ ! H is SF -logically valid, with d the truth degree for which the principal
lter hdi of G is the set of designated truth degrees.
As usual, this tableau calculus is intended to be used as a refutation
machinery. Therefore the way to check the SF -logical validity of some w H
of LSF is to construct a (complete) tableau for the signed bounding formula
F : (d^ ! H ), and to check whether it is closed.
The closing conditions for the branches of such tableaux state that a
branch is closed if it has among the formulas which decorate its nodes one of
the following kinds of signed bounding formulas:
1. T : (û ! v^) for truth degrees u 66 v;
2. F : (û ! v^) for truth degrees 0 6= u 6 v 6= 1 ;
3. F : (0 ! H ) ;
4. F : (H ! 1) ;
5. T : (û ! H ) and F : (^v ! H ) for truth degrees v 6 u .
522
As explained in Chapter 7 the (sequence of) tableaux for some signed

bounding formula (or some set of such formulas) of LSF are constructed via
tableau extension rules. These read, adopting the notation (7.2) introduced
in Chapter 7, in the present case e.g. for arbitrary ws H of LSF which are
not truth degree constants as:
F : (û ! H )
with u 6= 0 and v1 ; : : : vn all
T : (H ! v^1 ) + + T : (H ! v^n ) the maximal members of G
which are not greater than u;
F : (H ! v^)
with v 6= 1 and u1 ; : : : uk all
T : (u^1 ! H ) + + T : (u^k ! H ) the minimal members of G
which are not smaller than v;
T : (û ! H )
with u 6= 0 and vi any maximal member of G which is not
F : (H ! vi )
greater than u;
T : (H ! v^)
with v 6= 1 and uj any minimal member of G which is not
F : (uj ! H )
smaller than u.
These tableau extension rules are for the cases that the w H is a conjunction
or a disjunction the rules
T : (û ! H1 ^ H2 )
with u 6= 0;
T : (û ! H1 ) T : (û ! H2 )
F : (û ! H1 ^ H2 )
with u 6= 0;
F : (û ! H1 ) + F : (û ! H2 )
T : (H1 _ H2 ! v^)
with v 6= 1;
T : (H ! v^) T : (H ! v^)
1
F : (H1 _ H2 ! v^)
with v 6= 1.
F : (H1 ! v^) + F : (H2 ! v^)
And for the case that the w H is an implication one has the tableau extension
rule
T : (û ! (H1 ! H2 ))
F : (w^ ! H1 ) + T : (w^ ! H2 )
for u 6= 0 and each w 6 u,
together with the further rule

T : (û ! (H1 ! H2 ))
+ fT : (w^ ! H1 ) F : (w^ ! H2 ) j 0 6= w 6 ug
for which one has to suppose u 6= 0.
This set of tableau extension rules can be shown to be complete w.r.t.
the non-modal part of the language LSF , i.e. for each w H in which the
modal operators 3; 2 do not occur one has that H is SF -logically valid i
523
there exists a closed tableau for the signed bounded formula F : (d^ ! H ).
The proof is essential a routine matter and given in [167].
More complicated are the tableau extension rules for modalized ws.14
One has to look only for signed bounding formulas of the two basic types F :
(û ! 2H ) and F : (3H ! v^). For them, however, the tableau extension rules
do not describe simple extensions of given tableaux, instead they describe the
exchange of a whole branch of a given tableau by a new one if such a rule is
applied. We shall not go into these details here, they are given in [167]. We
mention only that the resulting tableau calculus provides an adequate, i.e.
sound and complete axiomatization of the class of all SF -logically valid ws
of the language LSF .
In a manner similar to the use of only two signs in this tableau calculus,
the sequent calculus of [166] has a rather standard notion of sequent: each
sequent has the form ; with two sets ;; of ws, which, however, are
supposed to be sets of implications. This sequent calculus then has the usual
structural rules for thinning and cut :
;
;; ; 0 ; 0
; ; X ;; X
;
together with two particular rules for implications:

;; u^ ! H1 ; u^ ! H1 for each u 2 G ;
; ; H1 ! H2
;; H2 ! v^ ; H1 ! v^ for each u 2 G
; ; H1 ! H2
which both are nitary because G is a nite set.
As axioms this sequent calculus has besides the usual initial sequents
X X
a transitivity axiom
X ! Y; Y ! Z X ! Z
together with further axioms for conjunctions and disjunctions:

A ^ B ! A,
A ! A _ B,
A ^ B ! B,
B ! A _ B,
C ! A; C ! B C ! A ^ B ,
A ! C; B ! C A _ B ! C ,
and the axioms for importation and exportation:
A ^ B ! C A ! (B ! C ) ; A ! (B ! C ) A ^ B ! C :
Additionally one needs some truth degree constant axioms
14 This is a situation which is not caused by the present generalization into the realm
of many-valued logic. Likewise one has similarly complicated tableau extension

rules in tableau calculi for classical modal logics, cf. e.g. [162].
524
u^ ! v^
u^ ! v^
for all u; v 2 G with u 6 v ;

for all u; v 2 G with u 66 v ;
and also conversion axioms which regulate the change between ws and particular bounding formulas:
X 1^ ! X ; 1^ ! X X :
All this together forms the non-modal part of this sequent calculus. For the
modal part one has to add as a further rule the rule of necessitation
A
2A
together with the modal axioms

2(A ! B) ! (2A ! 2B) ;
(û ! 2A) ! 2(û ! A) for each u 2 G ;
(3A ! v^) ! 2(A ! v^) for each v 2 G ;
2(û ! A) ! (û ! 2A) for each u 2 G ;
2(A ! v^) ! (3A ! v^) for each v 2 G :
Again one can show that this sequent calculus provides an adequate, i.e.
sound and complete axiomatization of the class of all SF -logically valid ws
of the language LSF , cf. [166].
22. Approximating Intuitionistic

and Other Logics
It was a reaction to the set-theoretical and logical antinomies as well as to the

semantical ones1 which were found around the turn of the 20th century, and
at the same time an outcome of its own philosophical position, that the Dutch
mathematician L.E.J. Brouwer rst in his Ph.D.-Thesis of 1907 and later
on in the end of the 1910s and beginning of the 1920s, developed what he
called intuitionistic mathematics. This intuitionistic mathematics should be,
he claimed, independent of any logic which precedes it from a systematic point
of view, and should instead have its own inherent logic, cf. [76]. For years the
principles of such an intuitionistic logic remained hidden, and Brouwer did
not intend to make them explicit mainly because he was convinced that this
logic was inaccessible to a suitable formalization and could be recognized only
in the undertaking of intuitionistic mathematics. It was only Heyting [262]
who gave in 1930 a formalization of this intuitionistic logic and thus made it
accessible to a wider public, including classical logicians and mathematicians.
His approach, however, was a syntactic one, i.e. he gave a logical calculus
claiming that it would adequately represent the basic ideas of intuitionistic
logic. What still was lacking was a corresponding semantics, i.e. a circle of
ideas which would allow one to link the formulas of (the language of) this
logic with intuitively understandable meanings and to distinguish a class of
formulas from this point of view as the logically valid ones.
22.1 Many-Valued Approaches toward Intuitionistic

Logic
Important preliminary steps in the development of such a semantics for intuitionistic logic have been the interpretation of intuitionistic logic as a logic of
problem solving by Kolmogorov [313], and the proof of Go del [199] that
the intuitionistic logic is not a system of nitely many-valued logic. This has
been the context in which Go del introduced the particular systems Gn of
many-valued logic which have been the topic of Chapter 10. As was proved
there, cf. Theorems 10.1.2 and 10.1.3, the sets of tautologies of Gn always are
1 Well written, in depth presentations of the topic of these types of antinomies are
given e.g. in [55, 75, 323].
526
22. Approximating Intuitionistic and Other Logics
supersets of the set of theorems of (Heyting's logical calculus for) intuitionistic logic. Because each one of the ws A $ A as well as each one of the ws
A ! A _ B (with A; B ws for intuitionistic logic) is a theorem of intuitionistic logic, cf. e.g. [400, 444, 481], each (propositional) system of many-valued
logic which is intended to have a set of tautologies which coincides (via a
rereading of the connectives) with the set of theorems of intuitionistic logic
satises the conditions (B1) and (B2) of Chapter 21, cf. p. 507. Therefore
such a system has, as long as it is supposed to have only nitely many truth
degrees, to have ws Gn of the form (10.10) among its tautologies { formulas
which, however, are not theorems of intuitionistic logic.2
A few years later in 1936 the Polish logician Jaskowski [285] constructed
an innite sequence of nitely many-valued systems J k , k = 1; 2; : : :, such
that the intersection of the sets of J k -tautologies is just the set of theorems
of intuitionistic logic (via the \usual" rereading of the connectives). The
starting system J 1 in his construction is classical propositional logic C2 . The
construction of the sequence proceeds with reference to some operation ;
which converts a given system with m truth degrees into a system with m +1
truth degrees. Let S be some m-valued system with connectives ; !; u; t
and the only designated truth degree 1. Consider some ! 2= W S . Then W =
W S [ f!g is chosen as truth degree set of S = ; (S) and shall have exactly
the same designated truth degree as W S . For all x 2 W S let:
n
if x = 1
(x) =def !;
x otherwise.
Then is a 1-1 mapping of W S onto the set of non-designated truth degrees of W . It is used to determine the truth degree functions ver of S
which interpret the connectives ; !; u; t in the new system. For the unary
connective one denes

(ver (1)); if x = 1
ver (x) =def ver
( ;1 (x)) otherwise.

For a characterization of the other, binary connectives we use their representation by truth degree tables. The corresponding tables are given in Fig. 22.1.
Using the product of logical systems as introduced in Section 8.3 together
with the ; -operator one gets Jaskowski's sequence via
nY
+1
J n+1 =def ; ( J n):

k=1
(22.1)
T
J
A proof that the intersection 1
n=1 tautn of the sets of tautologies of all the
systems J n is the set of theorems of intuitionistic logic is sketched in [285].
Detailed proofs are given in [458] and [521]. They shall not be explained
here because they are not of importance for the development of many-valued
logic. That also some simplied versions of the Jaskowski construction yield
a corresponding result is explained in [523].
2 This fact follows e.g. from the proof of Theorem 10.1.2 and from Theorem 10.1.3.
22.2 Approximating Logics by Many-Valued Logics

ver!
527
(y)
1
verS! (1; 1) (verS! (1; y))
(x) verS! (x; 1) verS! (x; y)
1
veru
(y )
1
verSu (1; 1)
(verSu(1; y))
(x) (verSu(x; 1)) (verSu (x; y))
vert
(y)
1 verSt(1; 1) verSt (1; y)

(x) verSt (x; 1) (verSt(x; y))
Fig. 22.1. Ja
skowski's tables for the \new" system
Principally one can, as already mentioned in [285], even use the Jaskowski sequence (22.1) to form an innitely many-valued system which has as
its set of tautologies just the set of theorems of intuitionistic logic. However,
the truth degrees of this innitely many-valued system do not have any recognizable relation to the basic ideas of the intuitionistic approach. Therefore
we shall not consider this system here.

Another aspect of the construction of the sequence of the Jaskowski systems
J n in the previous section is, however, of greater importance: one can consider the whole sequence (J n )n>0 of these Jaskowski systems as a series of
nitely many-valued systems which \approximate" the system of intuitionistic logic. And this understanding of the sequence of Jaskowski systems
is neither restricted to these Jaskowski systems nor to the approximation
of the system of intuitionistic logic. Instead one can consider the problem
whether it is possible also for other systems of logic to approximate them by
systems of many-valued logic.
Before this problem can be discussed in some detail, rst these approximation ideas have to be made more precise. Following [28], we shall restrict
the considerations to nitely many-valued systems as approximating ones,
because these systems are (computationally) rather simple, e.g. decidable
ones as explained in Section 8.2.
The logical systems S we intend to approximate shall be propositional
systems which have the usual type of propositional languages L0 , explained
in Section 2 and used throughout this book, and which shall be syntactically
characterized by a Hilbert-type logical calculus K with nitely many ax-
528
iom schemata and nitely many (nitary) inference rules. In short, such a
propositional system S shall be called K -based. For each inference rule
H1 ; : : : ; Hn
H
(22.2)
of K one supposes that each particular application of (22.2) leads from

a substitution instance H1 [p1 =A1 : : : pk =Ak ]; : : : ; Hk [p1 =A1 : : : pk =Ak ] of the
premisses to the corresponding substitution instance H1 [p1 =A1 : : : pk =Ak ] of
the conclusion of (22.2). Here we suppose that the propositional variables
which occur in (formulas of) the rule (22.2) are among p1 ; : : : ; pk , and that
A1 ; : : : ; Ak are ws of S .
Let Th(S ) be the set of all S -derivable ws. What we are interested in
are systems S of many-valued logic which \cover" S in the sense that one
has Th(K ) taut(S). For simplicity we assume that S and all its covers S
have the same language L0 , and that this language L0 has only nitely many
connectives and truth degree constants.
Denition 22.2.1. Let S be some system of many-valued logic, and S some
K -based propositional system for some Hilbert-type calculus K such that S
and S have the same language LS . Then S is a cover of S , or K is strongly
sound for S, i all the axioms of K are S-tautologies and all the inference
rules of K are S-sound.
Here, as usual, we call an inference rule of the form (22.2) sound in a
system S of many-valued logic, or simply S-sound, i in each particular application instance of (22.2) the conclusion of this rule is entailed (in the sense
of S) by the set of its premisses.
Corollary 22.2.1. For each K -based propositional system S and each system S of nitely many-valued logic it is decidable whether S is a cover of S
or not.
Proof: Because S has only nitely many truth degrees, the property of
being an S-tautology is decidable, cf. Section 8.2. And because of this niteness also the S-soundness of the inference rules of K is a decidable property.
It is interesting to notice that this covering property is stronger as the

simple soundness of the calculus K for the system S understood as the fact
that each K -derivable w is an S-tautology.
Proposition 22.2.1. Let S be some system of many-valued logic, and S
some K -based propositional system. Then S is a cover of S i one has
Th(K 0 ) taut(S0 ) for all extensions K 0 of K and taut(S0 ) of taut(S) such
that
(a) S0 is obtained from S by adding new truth degree constants or new connectives (via their truth degree functions) to S;
529
(b) K 0 is obtained from K by adding new axioms which all are S0 -tautologies.
Proof: Suppose rst that S is a cover of the K -based system S. And let
taut(S0 ) and K 0 be extensions of taut(S) and K , respectively, such that (a)
and (b) are satised. Then for any H 2 Th(K 0 ) one has to show H 2 taut(S0 ).
This is obvious for the axioms of K 0 . So assume that it holds for all ws
H 0 which have a K 0 -derivation of length < l, and that H has a K 0 -derivation
H of length l, and is not an axiom of K 0 . Then the last step in H is an
application case of one of the inference rules (22.2) of K . By assumption, all
the premisses of this rule application are S0 -tautologies, which means also
H 2 taut(S0 ) by the S-soundness of rule (22.2).
Now suppose conversely that Th(K 0 ) taut(S0 ) holds for all extensions
0
K of K and S0 of S which satisfy (a) and (b). Because every K -axiom is
K 0 -derivable it is also an S0 -tautology, and hence an S-tautology. So let rule
(22.2) be an inference rule of K , and assume that Hi0 Hi [p1 =A1 : : : pk =Ak ] for
i = 1; : : : ; n are substitution instances of the premisses of rule (22.2). Consider
any model of fH10 ; : : : ; Hn0 g, and let uj = ValS (Aj ; ) for j = 1; : : : ; k. Denote these truth degrees by the constants u^j , and form S0 by just adding these
truth degree constants to S. Consider now the ws Hi00 Hi [p1 =u^1 : : : pk =u^k ]
for i = 1; : : : ; n. From them one derives in K by an application of the rule
(22.2) the w H 00 H [p1 =u^1 : : : pk =u^k ]. So one has H 00 2 taut(S0 ). And this
means by construction of H 00 that ValS (H [p1 =A1 : : : pk =Ak ]; ) is a designated
truth degree. Hence is also a model of H [p1 =A1 : : : pk =Ak ]. Thus the soundness of rule (22.2) is shown, and S is a cover of the K -based system S . 2
It is a routine matter to check that each one of the Go del systems Gm is
a cover of the intuitionistic propositional logic IPC based e.g. on the calculus
given by the axiom schemata (LC1); : : : ; (LC11) of Section 10.1 together with
the rule of detachment (MP) as the only inference rule. Furthermore, all these
systems Gm are obviously also covers of the innitely many-valued Go del
system G1 based on the calculus LC determined by the axiom schemata
(LC 1); : : : ; (LC 11) and (LCG ) together with the inference rule (MP), cf.
Theorem 10.1.3.
Even more, the sequence (Gm )m>2 of all these Go del systems provides,
according to Theorem 10.1.2(a), better and better covers for IPC. However,
this sequence does not \converge" toward intuitionistic propositional logic
IPC because of Theorem 10.1.2(b) and the fact that the set of G1 -tautologies
does not coincide with the set of IPC-theorems, cf. Theorem 10.1.3. Nevertheless, intuitively the crucial properties of this sequence given in Theorem
10.1.2(a), (b) should be considered as stating a kind of \convergence" of this
sequence of covers of G1 toward G1 .
Being interested in optimal, i.e. best possible covers, as well as in approximation processes, we have to make precise an ordering between covers. For
this it suces to have an ordering between systems of many-valued logic,
which refers to their sets of tautologies.
530
Denition 22.2.2. Let S1 and S2 be systems of many-valued logic with the

same language LS . Then let
S1 S2 =def taut(S1 ) taut(S2 ) ;
S1 S2 =def taut(S1 ) taut(S2 ) :
These relations and are partial orderings between systems of manyvalued logic over the xed language L0 . The relation is re exive, and
is its irre exive version. So it is reasonable to consider as optimal covers of
some given K -based propositional system S (with language L0 ) the minimal
covers w.r.t. these orderings. Fortunately these ordering relations are not too
complicated, at least for a xed number m of truth degrees.
Proposition 22.2.2. For each K -based propositional system S (with language L0 ) there exist for each xed number m only nitely many m-valued
covers.
Proof: Let S be given, and m xed. Then the truth degree set can be
assumed to be Wm . For each n-ary connective ' in L0 there exist only nitely
many possible truth degree functions ver' : Wm n ! Wm . Because L0 has
only nitely many connectives, and only nitely many truth degree constants,
only nitely many dierent interpretations of them in Wm are possible. And
Wm has also only nitely many subsets which can act as a set of designated
truth degrees for a cover S.
2
Looking for an optimal cover for some K -based propositional system S
thus can be understood as inspecting all possible covers, and taking some minimal one. For m-valued covers, with m xed, this can be done eectively.
Proposition 22.2.3. For all nitely many-valued systems S1 and S2 it is
decidable whether S1 S2 holds or does not hold.
Proof: Let Si be mi-valued and m = maxfm1; m2g. Consider instead of
property S1 S2 the simpler property that there exists a w A such that
A 2 taut(S2 ) n taut(S1 ) ;
(22.3)
and denote this property for the moment by S1 S2 . Then one immediately
has that S1 S2 holds i S1 S2 holds but not S2 S1 . Therefore the
decidability of S1 S2 follows from the decidability of S1 S2 .
So we have to prove the decidability of S1 S2 . Of course, (22.3) is
decidable for each w A because (22.3) means A 2 taut(S2 ) and A 2= taut(S1 ),
and because these two properties are decidable.
Assume that A 2 taut(S2 ) n taut(S1 ) and that 1 is some Wm -valuation
such that ValS1 (A; 1 ) is an undesignated truth degree of S1 . Assume furthermore that in A occur at most m (dierent) propositional variables. Because
if A would contain more, at least two of them had to have under 1 the same
value, and thus could be identied without destroying either A 2 taut(S2 ) or
1
531
A 2= taut(S1 ). So assume even that all the propositional variables of A occur

among p1 ; : : : ; pm .
Let B1 ; : : : ; Bk be all the subformulas of A. Each one of them determines
an m-ary truth degree function w~B2 j in Wm . We simplify A to A0 in the
way that, if there is a subformula Bi of A and a proper subformula Bj of
Bi such that w~B2 i = w~B2 j as well as ValS1 (Bi ; 1 ) = ValS1 (Bj ; 1 ), then the
subformula Bj is substituted for Bi in A. The resulting w A0 is shorter as
A (in terms of the number of symbols which form the strings A and A0 ).
Suppose that all these possible \simplications" of A have been done, and
that the w A is the result.3 Of course, this construction does not violate
condition (22.3), i.e. one has also A 2 taut(S2 ) and A 2= taut(S1 ). Even 1
remains a countermodel for A in S1 .
Fortunately, the resulting w A cannot be toom complicated. Their complexity degree4 d(A ) has to be smaller than m(m +1) . Let d(A ) = d. Then
there exists a sequence A C0 ; C1 ; : : : ; Cd of subformulas of A such that
each w Ci+1 is an immediate constituent of the compound formula Ci , and
hence a proper subformula of Ci . Each of these ws Ci determines an m-ary
truth degree function w~C2 i in Wmm , and a truth degree ValS1 (Ci ; 1 ) in Wm .
However, there are at most mm such mtruth degree
functions, and at most m
such truth degrees, i.e. at most m mm = m(mm +1) ordered pairs consisting
2
of such a truth degree function

and such a truth degree.
In the case d > m(mm +1) one thus would have two ws Ci and Cj with
i < j , such that w~C2 i = w~C2 j as well as ValS1 (Ci ; 1 ) = ValS1 (Cj ; 1 ). But then
the above mentioned \simplication" of A could be continued by substituting
in A the shortermsubformula Cj for the subformula Ci . A contradiction. So
one has d m(m +1) .
Therefore one has \only" to search all the (nitely many) ws A of a
complexity degree d(A) m(mm +1) to see whether there exists a w which
satises (22.3), or whether such a w does not exist. And this means that
the property S1 S2 is a decidable one.
2
Corollary 22.2.2. For any two nitely many-valued systems S1 and S2 (over
the same language) it is decidable whether they have the same set of tautologies or not, and it is decidable whether S1 S2 holds or not.
Proof: One obviously has taut(S1 ) = taut(S2) i neither S1 S2 nor
S2 S1 holds. And one has also S1 S2 i S1 S2 or taut(S1 ) = taut(S2 ).
So both claims follow from the preceding proposition.
2
3 This construction process has to terminate because the w A becomes shorter
with each step of this construction, and has a nite length.
4 With the complexity degree d(H ) of a w the number of steps is counted which
are necessary to form H , in the sense that propositional variables and truth
degree constants h have complexity degree d(h) = 0, and that a compound
formula H '(H1 ; : : : ; Hn ) has the complexity degree d(H ) = maxfd(Hi ) j 1
i ng + 1.
532
Corollary 22.2.3. For each K -based propositional system S and each m
the optimal m-valued covers can be eectively determined.

Proof: According to Proposition 22.2.2 there exist only nitely many mvalued covers. Among them the -minimal ones can be eectively determined
according to Proposition 22.2.3.
2
What is not yet solved by these results is the problem of an approximation
of some given K -based propositional system S by a whole (innite) sequence
of covers. So we extend the eectiveness result of Proposition 22.2.3 and prove
even the existence, or nonexistence, of suitable approximating sequences.
Denition 22.2.3. Let S be some K -based propositional system. Then a sequence (Si )i>0 of systems of many-valued logic is a sequential approximation
of S i
(i) each system Si is a cover of S ;
(ii) one has Si Sj for all i j ; and
\
(iii) one has Th(K ) = taut(Si ) .
i>0
Such a sequential approximation is a proper one i for all i j the manyvalued system Si has not more truth degrees than Sj .
In this sense the Jaskowski sequence (J n )n>0 is a sequential approximation of IPC, and the sequence of the ;Go del systems (Gm )m>2 is a sequential
approximation of the system IPC + (A ! B ) _ (B ! A) .
Theorem 22.2.1. If the set Th(K ) of ws is not decidable then one cannot
eectively determine some sequential approximation of S .
Proof: Assume that (Si )i>0 is some sequential approximation of S
which is eectively determined. This means that for each i > 0 the system
Si can be eectively determined.
We shall show that in this case the set Th(K ) is decidable. A contradiction.
By denition Th(K ) is a recursively enumerable set of ws because it is the
set of all K -derivable ws, and because the sets of axioms and inference rules
of K are decidable ones.
So the decidability of the set Th(K ) follows from the recursive enumerability of its complement. And this means that the complement K ; of Th(K )
is the domain of some partial recursive function, i.e. the elements of K ; are
just those ones \accepted" by some algorithm. And this algorithm can be the
following:
Let H be some w. Test successively for i = 1; 2; : : : whether H is a
tautology of Si or not.5 Stop if it is not an Si -tautology, and accept H .
5 This test always gives an answer because the nitely many-valued systems Si are
always decidable.
533
In the case H 2= Th(K ) that H is not K -derivable one has H 2= taut(Sj )

for some j , which means that this algorithm accepts H . And in the case
H 2 Th(K ) this algorithm never stops, and hence does not accept H . Thus
this algorithm accepts just the elements of K ;.
2
Theorem 22.2.2. For each decidable K -based propositional system S one
can eectively determine some proper sequential approximation of S .

Proof: Suppose that for each w H of the language L0 of S some
numerical code c(H ) is determined, which e.g. may be6 a rational from the
open interval (0; 1). Use the same notion d(H ) of complexity degree as in the
proof of Proposition 22.2.2, and for each k 2 N let wf k be the (nite) set of
all those ws H with d(H ) < k in which only propositional variables out of
p0 ; : : : ; p(k) occur.
Now let Sk be the many-valued system with language L0 which has as
truth degree set
WkS =def c[wf k ] [ f1g
the set of all codes of ws from wf k together with the additional degree 1,
which has as set of designated truth degrees the set
;

DkS =def c[wf k ] \ c[Th(K )] [ f1g ;
and which assigns to each n-ary connective ' of L0 the truth degree function
characterized by
(
c('(H1 ; : : : ; Hn )) ; if xi = c(Hi ) for 1 i n
S
ver'(x1 ; : : : ; xn ) =
and '(H1 ; : : : ; Hn ) 2 wf k ;
1
otherwise.
The idea behind this construction is that for ws from wf k the system Sk has
as its tautologies just the theorems from Th(K ). So it is easy to recognize that
each system Sk is a cover of S . Moreover one has Si Sj for all i j 2 N
immediately from this construction.
T
This construction also immediately gives Th(K ) i>0 taut(Si ). And
it even yields that each w H 2= Th(K ) is not an Sk -tautology for each
k > c(H ). Therefore the sequence (Si )i>0 is a sequential approximation of
S , which additionally is a proper one.
2
6 The present choice is not essential. Each other encoding does as well. However,
with rationals from the unit interval as code numbers it is simple to embed all
the following sets of truth degrees into truth degree sets of the kind (5.2).
534
23. Independence Proofs
23.1 The Propositional Case

One of the earliest types of applications of ideas from many-valued logic was
their use in approaches toward proofs of underivability results, as e.g. used
systematically by Bernays [52] in studies on a propositional calculus for
classical logic. In these applications one does not need the full theoretical
framework of many-valued logic, the reference to truth degrees and truth
degree functions suces { and suces in a quite formal sense because no
intuitive understanding of the truth degrees is intended. The underivability
often concern some formulas which appear as formulas of axiom systems of
logical calculi or of mathematical theories,1 and for which one is interested
to prove their indispensability w.r.t. these axiom systems.
The core idea of these applications is that the truth degrees are simply
\labels" which become tied with derivable formulas in such a way that some
type of \designated" label is (always) assigned to formulas derivable from
some suitable set of formulas, and another formula H whose underivability from shall be proved is shown (sometimes) not to get a \designated"
label. As these \labels" one takes a suitable set of truth degrees, and the
\designated" labels are of course the designated truth degrees. The understanding of the connectives has to be realized by truth degree functions which
characterize how \labels" are assigned to compound formulas supposing that
the \labels" of the constituents are given. And the choice of the truth degrees
and the truth degree functions for the connectives has to be in accordance
with the idea that the inference rules of the system under consideration have
to become sound rules for the corresponding many-valued system, i.e. have
to lead from formulas which have (always) a designated truth degree again to
a formula which (always) has a designated truth degree. The formulas which
are derivable from then become the valid formulas of the models of . And
the formula which is intended to be shown to be underivable from has to
be shown to be not valid in some model of .
If one is interested to prove e.g. for a (logical) calculus K with nitely
many axioms (or axiom schemata) (ax1), . . . , (axn) that (ax1) is not derivable
1 Mathematical theories need, however, usually (at least) the framework of rst-
order logic. And this situation is discussed in more detail in Section 23.2.
536
from the remaining axioms (ax2), . . . , (axn) then one has to nd such a
system S of many-valued logic, i.e. such a set of truth degrees and such an
interpretation of the connectives which belong to the language of K (and
perhaps such a valuation) that the inference rules of K become sound rules
of S, that the axioms of K become S-valid formulas, and that (ax1) becomes
not S-valid. If one succeeds in nding such a system S then it is obvious that
(ax1) cannot be derivable from the axioms (ax2), . . . , (axn) because it is not
entailed by them, and because by construction one has soundness of K w.r.t.
the system S.
We shall illustrate these general considerations with a particular example
and consider a propositional calculus for classical propositional logic C2 . The
propositional language shall have all the standard connectives :; ^ ; _ ; ); ,
for classical logic. Then a logical calculus K 2 which adequately axiomatizes
C2 is given by the axiom schemata2
(Ax1)
A ) (B ) A) ,
(Ax2)
((A ) B ) ) A) ) A ,
(Ax3)
(A ) B ) ) ((B ) C ) ) (A ) C )) ,
(Ax4)
A^ B ) A ,
(Ax5)
A^ B ) B ,
(Ax6)
(A ) B ) ) ((A ) C ) ) (A ) B ^ C )) ,
(Ax7)
A ) A_ B ,
(Ax8)
B ) A_ B ,
(Ax9)
(A ) C ) ) ((B ) C ) ) (A_ B ) C )) ,
(Ax10) (A , B ) ) (A ) B ) ,
(Ax11) (A , B ) ) (B ) A) ,
(Ax12) (A ) B ) ) ((B ) A) ) (A , B )) ,
(Ax13) (A ) B ) ) (:B ) :A) ,
(Ax14) A ) ::A ,
(Ax15) ::A ) A ,
together with the rule of detachment (MP) as the only inference rule.
This axiom system is independent in the sense that not one of its axiom
schemata is dispensable, i.e. derivable from the other ones. We shall demonstrate our general methodology here only for axiom schema (Ax2). All the
other cases are discussed e.g. in [9].
For this purpose we consider a three-valued system S with truth degree
set W S = W3 = f0; 12 ; 1g. The connectives :; ^ ; _ ; ); , shall in S be interpreted as the corresponding connectives :; ^; _; !L ; $L of the Lukasiewicz
systems. This essentially means that we consider the many-valued system
2 This propositional calculus for classical logic is particularly interesting because it
has a kind of \saturation" property: if one is interested to nd a logical calculus

which adequately axiomatizes classical propositional logic in a language with
only some of the standard connectives then one gets such a calculus from the
present one by deleting all axiom schemata which do not belong to the restricted
language.
23.1 The Propositional Case
537
S = L3 . If we consider any L3 -valuation such that for suitable ws A; B ,

e.g. for propositional variables, one has
ValL(A; ) = 12 and ValL (B; ) = 0 ;
then one also has ValL(A !L B; ) = 21 and ValL((A !L B ) !L A; ) = 1,
and thus nally
ValL(((A !L B ) !L A) !L A; ) = 12 ;
which means that (Ax2) under this reading does not become L3 -logically
valid. However, all the other axioms (Ax1), (Ax3), . . . , (Ax15) become L3 logically valid formulas.
The discussion of the inference rule (MP) does not create any problem
here because one knows from the discussion of the Lukasiewicz systems
in Chapter 9 that (MP) is a sound rule for !L . The axiom schemata (Ax1)
and (Ax3) are, in their intended L3 -reading, even axioms of the adequate L3 axiomatisation given in Theorem 9.1.4 and thus L3 -logically valid. The well
known degree ranking property
if u v then seq 2 (u; v) = 1
of the truth degree function for the Lukasiewicz implication !L yields
together with simple properties of the other truth degree functions of L3 that
all the axiom schemata (Ax4), (Ax5), (Ax7), (Ax8), (Ax10), (Ax11), (Ax14),
and (Ax15) become L3 -logically valid formulas. (Ax13) has been mentioned in
Chapter 9 as an example (T19) of an L-tautology. (Ax6) and (Ax9) essentially
are particular cases of the L-tautology (T22) of Chapter 9, if one additionally
has in mind the importation/exportation law (T7) and the fact that ^; _
have the idempotency property (T3). So it remains to discuss (Ax12), which
means again via (T7) to show that (A !L B ) & (B !L A) !L (A $L B ) is
an L-tautology. But this is a simple consequence of (T16).
Hence all the axioms (Ax1), (Ax3), . . . , (Ax15) become L3 -logically valid
formulas under the many-valued reading, and our underivability proof for
(Ax2) is complete.
For all the remaining axiom schemata the proof of their underivability
from the respective other ones proceeds along the same lines. The crucial
and nontrivial problem in any case is the suitable choice of a many-valued
interpretation, i.e. of truth degrees and truth degree functions. It should be
emphasized that, contrary to the present example, one should not expect that
such a suitable many-valued interpretation has to be a known system of manyvalued logic in any case, or that the chosen truth degrees and truth degree
functions have to have some intuitive interpretation. The whole approach is
a purely technical one { and this is its only weakness from the point of view
of many-valued logic. However, this method is well suited for its particular
purpose, so well suited that one refers to it often as the matrix method for
underivability proofs, because one often denotes a system of truth degrees
and truth degree functions as a logical matrix.
538
For particular formulas we have, by the way, already used this method in
Chapters 21 and 22 for proving the underivability of formulas n within the
modal logic(al calculus) S 5, and for proving the underivability of formulas
Gn in intuitionistic logic.
A slight extension of the matrix method of underivability proofs leads
even to independence proofs in the usual understanding of \independence":
that a formula H is independent of a set of formulas i neither H nor
the negation :H is a consequence of . Here \consequence" may either be
understood in the syntactical sense as derivability within some suitable logical
calculus, or semantically as entailment w.r.t. some suitable system of logic.
For the previously treated example of a classical propositional calculus,
by the way, the underivability proof we sketched for (Ax2) essentially was
already the independence proof for (Ax2) because of the soundness of the
logical calculus K 2 for classical propositional logic and the fact that for each
w H which falls under the schema (Ax2) the w :H is not a classical
tautology.
23.2 The First-Order Case

The matrix method of underivability or independence proofs, which has been
explained in the previous section, is not restricted to the propositional case
but can similarly be used in the framework of rst-order logic. Of course,
the suitable choice of truth degrees and truth degree functions in the rstorder setting has to be accompanied by suitable choices of generalized truth
degree functions for the quantiers and by suitable choices of many-valued
predicates interpreting the predicate symbols of the language. The formal
apparatus, thus, becomes more complicated in the rst-order case, as usually
do the proofs. This is one of the reasons why this matrix method for long
was mainly used in the propositional setting.
The situation changed in the second half of the 1960s as mathematicians
and logicians tried to understand the forcing-method which was invented in
1963 by P.J. Cohen [108] for independence proofs in set theory. The crucial
idea was introduced and explained in [491] and later on published e.g. in
[36, 286, 320, 464]. We shall sketch here this type of approach with particular
emphasis on the principal ideas and aspects which are related to many-valued
logic and the choice of truth degrees. The set-theoretic peculiarities rst are
of minor importance and become important only later on.
The truth degree structures for the following discussions shall be Boolean
algebras.
Denition 23.2.1. An algebraic structure B = hB; +; ; ; 0; 1i with two binary operations +; , one unary operation , and two distinguished elements
0; 1 is a Boolean algebra i hB; +; ; 0; 1i is a distributive lattice with unit
element 1 and zero element 0 which has as a complementation.
539
According to this denition one has that for each Boolean algebra B =
hB; +; ; ; 0; 1i the following laws hold true for all a; b; c 2 B :
(B1) a + b = b + a ,
(B10 ) a b = b a ,
(B2) a + (b + c) = (a + b) + c ,
(B20 ) a (b c) = (a b) c ,
(B3) (a + b) b = b ,
(B30 ) (a b) + b = b ,
(B4) (a + b) c = (a c) + (b c) , (B40 ) (a b) + c = (a + c) (b + c) ,
(B5) a + a = 1 ,
(B50 ) a a = 0 ,
(B6) a + 1 = 1 ,
(B60 ) a 0 = 0 .
The choice of a Boolean algebra as truth degree structure provides with
the operations +; ; natural candidates for truth degree functions for the
disjunction, the conjunction, and the negation connective of classical logic.
By the denitions
a * b =def a + b ;
(23.1)

a
b =def (a + b) (b + a)
(23.2)
one gets also truth degree functions for the implication connective ) and the
biimplication ,. Furthermore 1 is taken as the only designated truth degree.
The lattice ordering of a Boolean algebra B is characterized by
a 6 b i a b = a i a + b = b
(23.3)
and has 0 as universal lower bound and 1 as universal upper bound. Each
pair set fa; bg has a b as its inmum and a + b as its supremum under
6. If
V
in the Boole
an
algebra
B
each
subset
X

B
has
an
inmum
X
and
a
W
supremum3 X then B is complete. In a complete
Boole
an
algebra
one
has
V W
with the (generalized, \innitary") operations ; also natural candidates
for generalized truth degree functions which interpret the quantiers 8 ; 9 of
classical rst-order logic.
Therefore we restrict our considerations in the present section to complete Boolean algebras as truth degree structures. And we understand by a
Boolean interpretation, or B-interpretation for short, of classical rst-order
logic PL2 any interpretation which has as truth degree structure a complete
Boolean algebra. The usual interpretations for PL2 which are used to determine the standard semantic notions for PL2 are particular B-interpretations.
This comes from the fact that the structure hf>; ?g; vel; et; non; ?; >i, which
has the usual truth-value functions for classical disjunction, conjunction, and
negation as its operations, is a complete Boolean algebra. All the other
B-interpretations are properly many-valued interpretations for PL2 , which
nevertheless have the property that they are models of the set of all logically
PL2 -valid formulas.
3 Besides the notation V X for the inmum of X we also use the index notation
V d for Vfd j i 2 I g, and proceed similar for the supremum.
i
i2I i
540
Theorem 23.2.1. Let H be a logically valid w of classical rst-order logic4

PL2 and A some B-interpretation for PL2 . Then H is valid in A.
Proof: According to the usual completeness theorem for classical logic,
H is derivable in any logical calculus which adequately axiomatizes PL2 .
Therefore it is sucient to prove for all axioms of such a logical calculus that
they are valid in each B-interpretation A for PL2 , and to show that all the
inference rules of this calculus are sound for this B-interpretation, i.e. lead
from A-valid formulas again to A-valid formulas.
Such an adequate axiomatization of PL2 is e.g. provided, cf. [10], by the
axiom schemata (Ax1), . . . , (Ax15) of the previous section together with the
inference rules (Gena ), (Gens ), (ExQa ), (ExQs ), (Renb ), and (Renf ), cf. p. 254,
of course with reference to the implication ) and the quantications 8 ; 9 of
classical logic instead of the connectives and quantiers of the L-systems.
Because all the axioms of this axiomatization are implications, it is helpful
to know as a preliminary result that
a 6 b i a + b = 1
(23.4)
holds true for all Boolean algebras B and all a; b 2 B . For, assuming a 6 b
one has a + b = b and hence
a + b = a + (a + b) = (a + a) + b = 1 + b = b :
And assuming conversely a + b = 1 gives
a = a 1 = a (a + b) = (a a ) + (a b) = 0 + (a b) = a b ;
and hence a 6 b because always 0 + c = (0 c) + c = c holds true.5
From this remark (23.4) one immediately gets that all the ws which fall
under one of the axiom schemata (Ax1), (Ax4), (Ax5), (Ax7), (Ax8), (Ax10),
and (Ax11) are valid formulas in each B-interpretation for PL2 . That the same
holds true also for all ws falling under one of the axiom schemata (Ax13),
(Ax14), (Ax15) follows from the fact that
a = a (a + a ) = (a a) + (a a ) = a a
= (a a) + (a a) = (a + a ) a = a
holds true in each Boolean algebra.
4 We have in mind here rst-order logic \without identity". This can be seen
from the denition of B-interpretation as well as in the following proof because

there are no semantical conditions which would make sure that some particular
binary predicate symbol of the language denotes the identity relation. This also
means that even in presence of an identity symbol in the language this symbol
is treated like any other binary predicate symbol and hence cannot count as
\identity symbol".
5 Because for each Boolean algebra B = hB; +; ; ; 0; 1i the algebraic structure
hB; +; ; 0i is an MV-algebra, (23.4) also follows immediate from Corollary 9.2.2.
541
For the discussion of the remaining axiom schemata it is preferable to

have two further results for Boolean algebras available. To get them we rst
show that from a b = 0 and a + b = 1 it follows b = a :
a = a (a + b) = a + (a b) + (a b) = a + b
= (a b) + (a b ) + b = b + (a b )
= b + (a b a) + (a b b) = b :
Having in mind that one always has true
(a + b) a b = (a a b ) + (b a b) = 0 ;
and because of 1 + c = c + c + c = c + c = 1 and a + a = (1 a) + a = a also
always has true
a + a + (a b) = a + (a b) + (a b) + (a b ) = a + a + (a b) = 1 ;
one gets as a rst preliminary result that
(a + b) = a b :
holds true in each Boolean algebra. By similar calculations one also gets as
a second preliminary result that
(a b) = a + b
holds true in each Boolean algebra.
Taking now into account axiom schema (Ax3) together with (23.4), one
gets from the fact that it holds true in each Boolean algebra
(a + b) + ((b + c) + a + c)
= (a b ) + (b c ) + a + c
= (a b ) + (b c ) + (a b) + (a b ) + c
= b + (b c ) + (a b) + (c b) + (c b )
= b + b + (a b) + (c b) = 1 ;
that each w which falls under this axiom schema (Ax3) is valid in each Binterpretation. In the same way it follows from the fact that it holds true in
each Boolean algebra
(a + b) + ((a + c) + a + (b c))
= (a b ) + (a c ) + a + (b c)
= (a (b + c )) + (a + (b c)) = (a (b c) ) + (a + (b c))
= (a + (b c)) + (a + (b c)) = 1 ;
that each w which falls under the axiom schema (Ax6) is valid in each
B-interpretation.
Similar elementary considerations show that each w which falls under
one of the axiom schemata (Ax2), (Ax9), or (Ax12) is valid in each Binterpretation.
542
From (23.4) together with denition (23.1) one immediately gets the
soundness of the rule of detachment (MP) w.r.t. all these B-interpretations,
i.e. that the conclusion (of an application) of (MP) is valid in some Binterpretation A provided the premises are valid in this B-interpretation A.
This soundness property is obviously true for the inference rules (Renb ) and
(Renf ). That also the remaining inference rules (Gena ), (Gens ), (ExQa ), and
(ExQs ) have this soundness property follows from the denitions of inmum
and supremum, and from the facts that there hold true in all Boolean algebras the relationships
if a 6 b then b 6 a ,
if a 6 b then a + c 6 b + c .
These last mentioned facts themselves can easily be inferred from (23.3) and
the results we already mentioned for Boolean algebras.
2
It is interesting to notice that Theorem 23.2.1 can be strengthened considerably: the logically valid ws of classical rst-order logic PL2 are just those
ws of the language of classical rst-order logic which become valid in each
B-interpretation, cf. e.g. [444]. We shall, however, not go into these details
because we are mainly interested in some particular type of B-interpretation.
Our main goal here is the consideration of B-interpretations for an axiomatic system of (classical) set theory usually denoted6 ZFC which was developed in papers of Zermelo, Skolem, and Fraenkel. This system is
considered and used in numerous textbooks. Good introductions are e.g. the
books [125, 175, 286, 320, 321, 527]. The language of ZFC is the language of
rst-order classical logic with identity \=" and one binary predicate symbol
\2" which denotes the membership relation. The \non-logical", i.e. particularly set-theoretic axioms of ZFC are:
(ZF1) Axiom of Extensionality:
8 x 8 y(8 z (z 2 x , z 2 y) ) x = y) ;
(ZF2) Axiom (Schema) of Separation:
8 u 9 v 8 x(x 2 v , x 2 u ^ H (x))
for any variable v which does not occur free in the w H (x),
(ZF3) Axiom (Schema) of Replacement:

8 u 8 x 2 u 9 yH (x; y) ) 9 v 8 x 2 u 9 y 2 vH (x; y)
for any variable v which does not occur free in the w H (x),
(ZF4) Axiom of Union:
;

8 u 9 v 8 x x 2 v , 9 y 2 u(x 2 y) ;
6 This name is the abbreviation for
C
hoice.
ermelo-Fraenkel system with axiom of
(ZF5)
(ZF6)
Axiom of Power Set:

;
543
8 u 9 v 8 x x 2 v , 8 y 2 x(y 2 u) ;
Axiom of Innity:
;

9 u 9 x(x 2 u) ^ 8 x 2 u 9 y 2 u(x 2 y) ;
(ZF7) Axiom of Foundation:
;

8 x 8 y 2 xH (y) ) H (x) ) 8 xH (x)
for any variable v which does not occur free in the w H (x).
The set theoretic system constituted by the axioms (ZF1), . . . , (ZF7) is
denoted ZF and turns into the system ZFC if one adds the axiom of choice:
(AC) Axiom of Choice:
;

8 u 9 v 8 x 2 u 9 =1 y 2 x 9 x 2 u(y 2 z ) ) 9 =1 y 2 x(y 2 v) :
In this last formulation7 we have used besides the restricted quantications
according to (1.2) the \numerical" quantication 9 =1 x 2 a meaning \there
exists exactly one . . . " e.g. dened by the conjunction
9 =1 x 2 aH (x) =def
(23.5)
9 x 2 aH (x) ^ 8 x 2 a 8 y 2 a(H (x) ^ H (y) ) x = y) :
The construction of the universes of discourse for the B-interpretations
for ZFC which we are going to consider needs some further set-theoretical
notion: that of an ordinal. Ordinals are closely related to the well-orderings.
And a well-ordering in a set A is a partial ordering 6R in A such that any
two elements of A are R-comparable and such that any nonempty subset
; 6= B A contains an R-minimal element of B , i.e. contains an element b
such that b 6R c holds true for each c 2 B . Usually ordinals are denoted by
lower case Greek letters ; ; : : :, we follow this usage.
Formally an ordinal can be dened as a set a which is transitive in the
sense that from c 2 b 2 a always c 2 a follows, and which has the property
that for all b; c 2 a either b 2 c or c 2 b or b = c holds true. In the class On
of all ordinals an (irre exive) ordering relation < is dened by
< =def 2 ;
which has the remarkable property that each ordinal becomes the set of all
smaller ones:
= f 2 On j < g :
7 The present formulation of the axiom of choice is a minor (but equivalent) vari-
ation of one of the usual ones saying that for each set u of pairwise disjoint,
nonempty sets there exists a choice set v of u which contains exactly one element from each x 2 u.
544
The re exive \hull" of this relation <, characterized (as usual) by the
condition =def < _ = is even a well-ordering of the class On.
The smallest ordinals w.r.t. this ordering are
0 =def ;; 1 =def f;g = f0g; 2 =def f0; 1g; 3 =def f0; 1; 2g; : : : ;
i.e. the non-negative integers. The set of all these integers is itself an ordinal,
denoted !. It is even the smallest innite ordinal, i.e. the smallest ordinal
which, as a set, is an innite set.
Intuitively, the ordinals are the order types of well-ordered sets. This
means that there exists for each well-ordered set A with well-ordering relation
6R a uniquely determined ordinal which is order isomorphic with A, i.e.
for which there exists a bijection f : A ! such that
a 6R b , f (a) f (b) for all a; b 2 A .
Such an order isomorphism f realizes an enumeration of the elements of A
in increasing order. The \numbers" which are used in this enumeration are
the ordinals which are elements of , i.e. the ordinals smaller than .
It is important that one has an inductive proof procedure available for the
ordinals which is similar to the well-known method of mathematical induction
in the eld of natural numbers, and which provides a means to prove that
some statement H ( ) on (some property H of) ordinals holds true for all
ordinals. The principle on which this proof procedure is based is expressed
by the valid formula

8 8 < H () ) H ( ) ) 8 H ( )
and states that the fact that some statement H ( ) holds true for all ordinals,
can be inferred if one is able to show that the property H ( ) is hereditary
in the sense that H () always holds true if H ( ) holds true for all ordinals
< .
One also has an inductive procedure to dene functions F ( ) on the class
of all ordinals, or particularly sequences of sets M , in such a way that one
can use for the denition of some particular value F (), or some particular
set M , all values F ( ) resp. all sets M with < .
Just this principle of inductive denitions through the ordinals now is
used to dene for a xed Boolean algebra B (with carrier B ) a sequence of
sets VB by

VB =def u 2 dom(u) B j 9 < (dom(u) VB )
(23.6)
for each 2 On. And the \union" of all these sets VB becomes the universe
of discourse V B of the B-interpretation VB we are going to construct:
V B =def fu j 9 (u 2 VB )g :
(23.7)
What remains to be determined for having the B-interpretation VB completed are the B-valued relations which correspond to the predicate symbols
545
=; 2 of the language of ZFC. We shall denote these B-valued relations by

=^ ; 2^ and agree to write
[ a =^ b] for =^ (a; b);
[ a 2^ b] for 2^ (a; b)
for any a; b 2 V B , as well as ValBV(H; f ) for the truth degree of H in the
B-interpretation VB under the assignment f : V ! V B . Therefore one has
ValBV(x = y; f ) = [ f (x) =^ f (y)]];
ValBV (x 2 y; f ) = [ f (x) 2^ f (y)]]
for all variables x; y and each V B -assignment f . Now we nally dene these
two B-valued predicates by
_
[ a 2^ b] =def
(b(u) [ a =^ u] ) ;
(23.8)
[ a =^ b] =def
u2dom(b)
^
(a(u) * [ u 2^ b] )
u2dom(a)
v2dom(b)
(b(v) * [ v 2^ a] )
(23.9)
for all a; b 2 V B .
This is a simultaneous inductive denition of the B-valued predicates
2^ ; =^ . We shall check whether it is an acceptable denition. For each a 2 V B
let its rank be the smallest ordinal such that a 2 VB . Then one immediately
recognizes that the denitions (23.8) and (23.9) refer to the determination
of the truth degrees [ a =^ b] and [ a 2^ b] to elements of V B which have a rank
which is smaller than the maximum of the ranks of a and of b. But this means
that these two simultaneous denitions could be rewritten as denitions which
proceed inductively on the ranks of a; b. Therefore they are acceptable.
As is obvious from the axioms of ZFC, this elementary theory is formalized in the language of classical rst-order logic with identity. We thus are
interested to extend the result of Theorem 23.2.1 to all logically valid formulas of classical rst-order logic with identity. This is done with the following
result.
Proposition 23.2.1. For all a; b; c 2 V B there hold true:
(a) [ a =^ a] = 1 ,
(b) if a 2 dom(b) then b(a) 6 [ a 2^ b] ,
(c) [ a =^ b] = [ b =^ a] ,
(d) [ a =^ b] [ b =^ c] 6 [ a =^ c] ,
(e) [ a =^ c] [ a 2^ b] 6 [ c 2^ b] ,
(f ) [ a =^ c] [ b 2^ a] 6 [ b 2^ c] .
Proof: (a) The proof is given by induction on the rank of a. Thus suppose
that the claim holds true for all c 2 dom(a). Then one has
a(c) = a(c) [ c =^ c] 6 [ c 2^ a]
according to (23.8), and hence
546
[ a =^ a] =
(a(c) * [ c 2^ a] ) = 1
c2dom(a)
by (23.4) and the fact that u u = u holds true in each Boolean algebra.
From (a) and (23.8) one immediately gets (b). And (c) easily follows from
denition (23.9) which is \symmetric" in a and b.
(d) Again the proof is given by induction on the rank of a. Therefore
suppose that one has true for each u 2 dom(a) and all v; w 2 V B
[ u =^ v] [ v =^ w] 6 [ u =^ w] :
(23.10)
Choosing particularly v 2 dom(b) and w 2 dom(c) then gives
[ a=^ b] [ b=^ c] a(u) b(v) c(w) [ u=^ v] [ v=^ w] 6 c(w) [ u=^ w] 6 [ u2^ c] :
Using the fact that one has from (23.9) the inequalities
[ b =^ c] b(v) 6 b(v) (b(v) * [ v 2^ c] ) = b(v) (b(v) + [ v 2^ c] )
= b(v) [ v 2^ c] 6 [ v 2^ c] ;
and taking the supremum on all the w 2 dom(c), one gets via (23.8)
[ a =^ b] [ b =^ c] a(u) b(v) [ u =^ v] 6 [ u 2^ c] :
Taking again a supremum on the left hand side, now over all v 2 dom(b),
gives
[ a =^ b] [ b =^ c] a(u) [ u 2^ b] 6 [ u 2^ c] :
Together with [ a =^ b] a(u) 6 [ u 2^ b] thus one furthermore has
[ a =^ b] [ b =^ c] a(u) 6 [ u 2^ c] :
Having in mind that for all elements r; s; t of a Boolean algebra which satisfy
the inequality r s 6 t one also has the inequality
r 6 r + (r s ) = (r s) + (r s ) + (r s )
= (r s) + s 6 s + t = s * t ;
this gives in the present case
[ a =^ b] [ b =^ c] 6 a(u) * [ u 2^ c] ;
and therefore also
^
[ a =^ b] [ b =^ c] 6
(a(u) * [ u 2^ c] ) :
(23.11)
u2dom(a)
Because of (c) one can furthermore write (23.10) equivalently as

[ w =^ v] [ v =^ u] 6 [ w =^ u] :
Starting from this inequality instead of (23.10), the same procedure as before
now yields
[ a =^ b] [ b =^ c] 6
547
(c(w) * [ w 2^ a] ) ;
w2dom(c)
and this together with (23.11) gives (d) by (23.9).

(e) W
Because of (d), W(c), and the fact that the (generalized) distributivity
law r fsi j i 2 I g = fr si j i 2 I g holds true in each complete Boolean
algebra, one immediately has
_
[ a =^ c] [ a 2^ b] =
([[a =^ c] b(v) [ a =^ v] )
v2dom(b)
_
(b(v) [ c =^ v] ) = [ c 2^ b] :
v2dom(b)
(f) nally results from (e) and (23.8) using similar calculations:
_
[ a =^ c] [ b 2^ a] =
([[a =^ c] a(u) [ b =^ u] )
u2dom(a)
_
([[u 2^ c] [ b =^ u] )
u2dom(a)
6 [ b 2^ c] :
2
Corollary 23.2.1. Each logically valid formula of classical rst-order logic

with identity is valid in every B-interpretation VB .
Proof: Because of Theorem 23.2.1 it only remains to prove that the
particular identity axioms of rst-order classical logic are valid in each Binterpretation. Having in mind that 2 is the only predicate symbol (dierent
from the identity symbol) in the set theoretic language of ZFC, these are e.g.
the following axioms:
8 x(x = x) ;
8 x 8 y 8 z (x = y ^ x = z ) y = z ) ;
8 x 8 y 8 z (x = y ^ x 2 z ) y 2 z ) ;
8 x 8 y 8 z (x = y ^ z 2 x ) z 2 y) :
We are going to discuss only the second one of these axioms. For the other
axioms one has to proceed similarly to show also their VB -validity. For the
VB -validity of the second axiom it suces to show that one has
Val (x = y ^ x = z ) y = z; f ) = 1
for each VB -assignment f . For a = f (x), b = f (y), and c = f (z ) that means
because of (23.4) that one has to show
[ a =^ b] [ a =^ c] 6 [ b =^ c]
548
for any a; b; c 2 V B . This, however, is an immediate consequence of Proposition 23.2.1(c), (d).

2
It is also suitable for our forthcoming discussion of the VB -validity of the
ZFC-axioms to take a closer look at the truth degrees of bounded quantications.
Proposition 23.2.2. Suppose that H is a w of the language of ZFC, that
f is some VB -assignment and that a = f (x). Then one has
_ ;

(a) Val (9 y 2 xH; f ) =
a(u) Val (H; f [y=u]) ;
(b) Val (8 y 2 xH; f ) =
u2dom(a)
^
u2dom(a)
a(u) * Val (H; f [y=u]) :
Proof: It suces to prove e.g. claim (a) because claim (b) follows in a
similar manner. But using the corresponding denitions one gets successively
Val (9 y 2 xH; f ) = Val (9 y(y 2 x ^ H ); f )
_ ;

=
[ v2^ a] Val (H; f [y=v])
=
=
=
=
v 2V B
_
v2V B u2dom(a)
_
a(u)
u2dom(a)
_
a(u) [ u =^ v] Val (H; f [y=v])

_
v 2V B
([[u =^ v] Val (H; f [y=v]))

u2dom(a)
a(u) Val (9 y(y = z ^ H ); f [y=v; z=u])
u2dom(a)
a(u) Val (H; f [y=u]) :
Taking a look back to the denitions (23.8) and (23.9) one sees that they
correspond to the conditions
a 2 b , 9 x 2 b(a = x) ;
a = b , 8 x 2 a(x 2 b) ^ 8 x 2 b(x 2 a) ;
which hold true in the system ZFC, and which can be proved e.g. inductively
on the rank of the sets a; b. However, these characterizations of membership
and equality for sets are not the simplest ones. Nevertheless, it is just these
denitions which make B-interpretations such an interesting tool for investigating (satisability and) independence problems in classical set theory.
Derivations in classical rst-order logic often use, in dealing with existentially quantied formulas, the (sound) rule:
9 xH (x)
H (a)
549
(23.12)
which leads from an existentially quantied w 9 xH (x) to one of the instances H (a) of its matrix. The soundness of this rule comes from the fact
that in classical logic an existentially quantied w 9 xH (x) is true (w.r.t.
some assignment in some interpretation) i one of its instances H (a) is true.
Unfortunately, this rule is not in general sound in the many-valued setting.
And it is already not sound for the Lukasiewicz setting and for the setting
of B-interpretations. The crucial point in both cases, with the truth degrees
of the existentially quantied formulas semantically characterized by taking
the suprema of the truth degrees of all \instances" of the w in the scope
of the existential quantication, comes either from the fact that the truth
degree set is innite, or from the fact that there exist incomparable truth
degrees. In both these cases the situation can be realized that some existentially quantied w has a designated truth degree, and that nevertheless
all the instances of (the matrix of) this w do not have a designated truth
degree. The mathematical background is the simple fact that in these cases
one may nd W
subsets X of the (partially ordered) truth degree set which have
a supremum X 2= X .
Fortunately, however, rule (23.12) is sound for the present type of B-interpretations for ZFC.
Proposition 23.2.3. For each w H of the language of ZFC, for each Binterpretation VB and each VB -assignment f there exists some b 2 V B such
that
Val (9 xH (x); f ) = Val (H; f [x=b]) :
Proof: For simplicity we write [ 9 xH (x)]] for Val(9 xH (x); f ) and [ H (a)]]
for Val (H; f [x=a]). This means also [ x 2 y] = [ x 2^ y] and [ x = y] = [ x =^ y]
which, however, does not cause any problems. Because the carrier B = jBj
of the truth degree structure B of the B-interpretation VB is a set, and the
universe of discourse V B is a proper class, there exists some ordinal and a
sequence (b )< of elements of V B such that
f[ H (b)]] j b 2 V B g = f[ H (b )]] j < g ;
and therefore also
_
[ 9 xH (x)]] = f[ H (b )]] j < g :
Now let
;_

u = [ H (b )]] f[ H (b )]] j < g
for each ordinal < . From the general properties of (complete) Boolean
algebras one then can prove inductively on that one always has
_
u 6 [ H (b )]] = fu j g :
(23.13)
550
These truth
degrees u now are used to dene a function b whose domain
S
dom(b) = < dom(b ) is the union of the domains of all the b 2 V B (which
are functions by denition) and which is characterized for each z 2 dom(b)
by:
_
b(z ) =def fu [ z 2^ b ] j g :
By construction we then have dom(b) VB for some suitably large ordinal
, and hence b 2 V B . Therefore one has
[ H (b)]] 6 [ 9 xH (x)]] ;
and the present proof shall be nished if we succeed in proving also the reverse
inequality.
To reach this goal we rst prove as an intermediate result that
u 6 [ b =^ b ]
(23.14)
holds true for each < . For each z 2 dom(b) one has
u b(z ) 6 u u [ z 2^ b ] 6 [ z 2^ b ]
by denition of b and hence also, as in the proof of Proposition 23.2.1(d),
u 6 (b(z ) * [ z 2^ b ] ) :
So one obviously has
^
u 6
(b(z ) * [ z 2^ b ] ) :
z2dom(b)
On the other hand one similarly has for each z 2 dom(b ) the inequality
u b (z ) 6 u [ z 2^ b ] 6 b(z ) 6 [ z 2^ b] ;
and hence also u 6 (b (z ) * [ z 2^ b] ). So one also has
^
u 6
(b (z ) * [ z 2^ b] ) :
z2dom(b )
According to (23.13) one thus has proved (23.14).
Having in mind that VB j= (y = x ^ H (x) ) H [x=y]) holds true
according to Corollary 23.2.1, one gets from (23.13) and (23.14) for each
< :
u 6 [ b =^ b ] [ H (b )]] 6 [ H (b)]] :
Again because of (23.13) one nally gets from these inequalities
[ 9 xH (x)]] = f[ H (b )]] j < g = fu j g 6 [ H (b)]]
which nishes the proof.
Now we are able to state and prove the main result of this section.
_
551
Theorem 23.2.2. Each B-interpretation VB is a model of all the axioms of

ZFC and hence of all the theorems of ZFC.
Proof: Using bounded quantiers, the axiom of extensionality (ZF1) can
equivalently be written as

8 x 8 y 8 z 2 x(z 2 y) ^ 8 z 2 y(z 2 x) ) x = y :
Then one has immediately VB j= (ZF1) from Proposition 23.2.2 and denition (23.9).
In the following discussion of the other axioms of ZFC we always can
consider some xed VB -assignment f which has to be modied only for
variables which shall be explicitly mentioned in the context. Therefore we
use also in the present proof the shorthand notations [ H ] or also [ H (x)]] for
Val (H; f ), depending on whether it seems to be of interest to mention the
variable x explicitly or not, and [ H (a)]] for Val (H; f [x=a]) as in the previous
proof. And for ZFC-axioms of the form 8 u 9 vH (u; v) the proof of VB j=
8 u 9 vH (u; v) is obviously given if one is able to determine for each b 2 V B
some c 2 V B such that Val (H; f [u=b; v=c]) = 1 holds true.
Consider now the axiom of separation (ZF2) and some b 2 V B . Choose
dom(c) = dom(b) and for each z 2 dom(b) let:
c(z ) =def b(z ) [ H (z )]]
for the w H which occurs in (the actual version of) the schema (ZF2). One
has c 2 V B , obviously, and also
[ 8 x(x 2 v , x 2 U ^ H (x)]] =
= [ 8 x 2 v(x 2 u ^ H (x))]] [ 8 x 2 u(H (x) ) x 2 v)]] :
We choose c as the value for v and b as the value for u. By Proposition 23.2.2
one has
^
[ 8 x 2 c(x 2 b ^ H (x))]] =
(b(z ) [ H (z )]] * [ z 2^ b] [ H (z )]]) = 1) ;
z2dom(c)
using also Proposition 23.2.1(b) together with (23.4) and (23.1). And one has
also
^
[ 8 x 2 b(H (x) ) x 2 c)]] =
(b(z ) * ([[H (z )]] * [ z 2^ c] ))
=
z2dom(b)
^
z2dom(b)
(b(z ) + [ H (z )]] + [ z 2^ c] ) = 1
because of b(z ) + [ H (z )]] = (b(z ) [ H (z )]]) and of b(z ) [ H (z )]] 6 [ z 2^ c] .

So we have immediately [ 8 x(x 2 v , x 2 U ^ H (x)]] = 1 (taking by
abuse of language here b; c additionally as individual constants) and thus
also VB j= (ZF2).
552
For the discussion of (ZF3) suppose again that some b 2 V B shall be

given. Then from previous results we have immediately
^
_
[ 8 x 2 b 9 yH (x; y)]] =
(b(z ) * f[ H (z; w)]] j w 2 V B g) :
z2dom(b)
Because the carrier B of the Boolean algebra B is a set, there exists for
each z 2 dom(b) some ordinal (z ) such that
_
_
f[ H (z; w)]] j w 2 V B g = f[ H (z; w)]] j w 2 VB(z) g :
Let 2 On be an upper bound of all the ordinals (z ) for z 2 dom(b).
Choose c 2 V B such that c(z ) = 1 for each z 2 dom(c) = VB . Then one has
[ 8 x 2 b 9 yH (x; y)]] 6 [ 8 x 2 b 9 y 2 cH (x; y)]]
and thus also VB j= (ZF3).
The axiom of union (ZF4) we rst rewrite equivalently as

8 u 9 v 8 x 2 v 9 y 2 u(x 2 y) ^ 8 y 2 u 8 x 2 y(x 2 v) :
Then consider some b 2 V B and dene a function c 2 V B by choosing
[
dom(c) = fdom(w) j w 2 dom(b)g ;
and for each z 2 dom(c) furthermore
_
c(z ) =
(b(z ) [ z 2^ w] ) = [ 9 y 2 b(z 2 y)]]:
w2dom(b)
From Proposition 23.2.2 one then gets rst

[ 8 x 2 c 9 y 2 b(x 2 y)]] =
^ ;

[ 9 y 2 b(z 2 y)]] * [ 9 y 2 b(z 2 y)]] = 1 ;
z2dom(c)
and furthermore also

[ 8 y 2 u 8 x 2 y(x 2 v)]] =
z2dom(b) w2dom(z)
b(z ) z (w) * [ w 2^ c] = 1
because of
b(z ) z (w) 6 b(z )m [ w 2^ z ] 6 [ 9 y 2 b(w 2 y)]] = c(w) 6 [ w 2^ c] :
Hence one has immediately VB j= (ZF4).
As a simplication of the axiom of power set (ZF5) we use the common
inclusion relation x j y =def 8 z 2 x(z 2 y). For a given b 2 V B form another
object c 2 V B with dom(c) = dom(b) B such that one has
^
c(z ) = [ z j b] =
(z (w) * [ w 2^ b] )
w2dom(z)
553
for each z 2 dom(c). For the proof that VB is a model of the axiom of power
set (ZF5) it now suces to show
[ 8 x(x 2 c , x j b)]] = 1 :
And this reduces immediately to the problem to show
[ 8 x(x j b ) x 2 c)]] = 1 ;
(23.15)
because one has that from denition (23.8) the relationship
_
[ z 2^ c] =
([[w j b] [ z =^ w] ) 6 [ z j b]
w2dom(c)
results for each z 2 V B , and hence has [ 8 x(x 2 c ) x j b)]] = 1. The way
to get (23.15) is again by constructing for each z 2 V B some w 2 V B such
that
[ z j b ) z = w ] = [ z j b ) w 2 c] = 1
holds true, and hence also [ z j b ) z 2 c] = 1. Put
w(s) = [ s 2^ z ] for each s 2 dom(w) = dom(z ) :
This immediately gives [ s 2^ w] 6 [ s 2^ z ] for each s 2 V B , and hence gives
[ w j z ] = 1. Furthermore one has
_ ;

[ s 2 b ^ s 2 z] =
b(t) [ s =^ t] [ s 2^ z ] 6 [ s 2^ w]
t2dom(b)
because of
b(t) [ s =^ t] [ s 2^ z ] 6 [ s =^ t] [ t 2^ z ] = [ s =^ t] w(t) ;
hence [ b \ z j w] = 1. But this yields also [ z j b ) z = w] = 1 because of
[ z j b] 6 [ z j b ^ b \ z j w ^ w j z ]
6 [ z j w ^ w j z] = [ z =^ w] ;
and therefore also
[ z j b] 6 [ w =^ z ^ z j b] 6 [ w j b] = c(w) 6 [ w 2^ c] ;
and thus nally [ z j b ) w 2 c] = 1. Hence we have also VB j= (ZF5).
For the treatment of the axiom (ZF6) of innity rst consider an embedding of the class of all (usual) sets into V B determined inductively on the
rank of the sets a by
a 7! a =def f(s; 1) j s 2 ag :
W
It is easy to check directly that one always has [ z 2^ a] = f[ z =^ s] j s 2 ag as
well as
a 2 c i [ a2^ c] = 1 ; a = c i [ a=^ c] = 1 :
554
Hence it suces to consider ! , and to use it to derive VB j= (ZF6) in a

straightforward way.
It is suitable to refer for the discussion of the axiom of foundation (ZF7)
to a principle of inductive proofs which is valid for the class V B , and which
reads for any subclass C of V B as:
;

8 x 2 V B dom(x) j C ) x 2 C ) C = V B :
(23.16)
(The simplest way to prove this principle goes via a proof
by contradiction: if
;
C 6= V B would be the case together with 8 x 2 V B dom(x) j C ) x 2 C ,
then consider some z 2 V B n C of smallest rank to get a contradiction.) Let
;

t = [ 8 x 8 y 2 xH (y) ) H (x) ] :
Our goal is to show t 6 [ H (z )]] for each z 2 V B , because this immediately
yields VB j= (ZF7). So consider the class Ct = fz 2 V B j t 6 [ H (z )]]g and
suppose that for w 2 V B one has dom(w) Ct . Then it is
t 6 f[ H (z )]] j z 2 dom(w)g
6 ^fw(z) * [ H (z)]] j z 2 dom(w)g = [ 8 y 2 wH (y)]] ;
and by the choice of t even
t 6 [ 8 y 2 wH (y)]] [ 8 y 2 wH (y) ) H (w)]] 6 [ H (w)]] ;
i.e. w 2 Ct . Thus one has Ct = V B by (23.16), and hence VB j= (ZF7).
^
So it remains to consider the axiom of choice (AC). As a shorthand notation let

Gu (x; y) =def y 2 x ^ 9 =1 z 2 u(y 2 z ) :
Consider some b 2 V B . Then one has to show that
;

[ 8 x 2 b 9 y Gb (x; y) ) 9 =1 y 2 x(y 2 c) ] = 1
(23.17)
B
holds true for some suitable c 2 V .
The meaning of Gu (x; y) is that y is an element of the set u which is
\characteristic"
for the set x. Starting with b 2 V B we form the set U =
S
fdom(w) j w 2 dom(b)g and assume that the elements of U are (uniquely)
enumerated by ordinals, such that U = fw j < g for some suitable ordinal
2 On. We choose dom(c) = U , and intend to realize that c \collects"
all those elements of U which are the rst ones of z 2 dom(b) under this
enumeration, and which are in this sense \characterizing" elements of the
sets z 2 dom(b). This idea, however, has to be formulated and realized in its
\many-valued", i.e. Boolean understanding. Therefore we dene for each
z 2 U:
_
_;
^

c(z ) =def
[ z = w ^ Gb (w; w )]] [ :Gb (w; w ] :
w2dom(b) <
<
Looking at the universally quantied implication which is the essential part

of the w considered in (23.17), one knows from (23.5) that its succedent is
555
a conjunction of two conjuncts which formulate an existence condition and a

uniqueness condition for y 2 x. We shall treat both conjuncts separately and
show that both are \entailed" by the antecedent of the \essential implication"
of the w in (23.17).
Let us start with the existence condition. Consider some w 2 dom(b) and
some z 2 dom(w). Then one has dom(w) U = dom(c), and hence according
to Proposition 23.2.1(b)
^
[ Gb (w; w )]] [ :Gb (w; w ] [ z =^ w ] w(z ) 6 [ z 2^ c] w(z )
<
for each < . And from this one gets

^
[ Gb (w; w )]] [ :Gb (w; w ] [ w 2^ w] 6 [ 9 y 2 w(y 2 c)]] ;
<
by taking the supremum over all z 2 dom(w), and using (23.8) as well as
Proposition 23.2.2(a). But because one has [ Gb (w; w )]] 6 [ w 2^ w] by the
choice of Gb , one gets
_
[ 9 y Gb (x; y)]] = f[ Gb (w; w )]] j < g
_
^
= f[ Gb (w; w )]] [ :Gb (w; w ] j < g
6 [ 9 y 2 w(y 2 c)]]
<
in a similar way as (23.13) previously. And this is just the B-validity of the
\essential" implication from (23.17) with the existential condition as succedent.
Now consider the uniqueness condition as succedent for (23.17). By elementary transformations one gets that in this case one has to show
;
[ 8 x 8 y1 8 y2 9 y Gb (x; y) ^ x 2 b ^

y 1 2 c ^ y1 2 x ^ y 2 2 c ^ y 2 2 x ) y 1 = y 2 ] = 1 :
From the denition of c one easily gets c(z ) = [ z 2^ c] for each z 2 dom(c) = U .
Consider for any z1 ; z2 2 dom(c) the truth degree
sw = [ w 2 b ^ z1 2 c ^ z1 2 w ^ z2 2 c ^ z2 2 w]
= [ w 2^ b] c(z1 ) [ z1 2^ w] c(z2 ) [ z2 2^ w] :
Because one immediately has
_
c(z ) 6
[ Gb (w0 ; z )]]
w0 2dom(b)
for each z 2 dom(c) from the denition of c, and because

Gb (x0 ; y) ^ y 2 x ^ x 2 b ) x0 = x
is a PL2 -logically valid w according to the choice of Gb , one also has the
inequalities
556
c(z ) [ z 2^ w] [ w 2^ b]
6 _ [ Gb (w0 ; z) ^ z 2 w ^ w 2 b]
w0 2dom(b)
6 [ Gb (w; z)]] :
Therefore one has

sw 6 [ Gb (w; z1 )]] [ Gb (w; z2 )]] :
For these z1; z2 2 dom(c) there exist ordinals ; < such that z1 = w and
z2 = w , and we can additionally suppose < . Thus one gets
sw 6 sw c(z2 )
_
_;
^

= sw
[ w = w ] [ Gb (w0 ; w )]] [ :Gb (w0 ; w ]
w0 2dom(b) <
<
6 [ w = w ] [ Gb(w; w )]] [ :Gb (w; w ]

<
6 [ Gb (w; w )]] [ :Gb(w; w ] = [ Gb (w; z2 )]] [ :Gb(w; z1 )]] ;
_;
and therefore
sw 6 [ Gb (w; z1 )]] [ Gb (w; z2 )]] [ :Gb (w; z1 )]] = 0 6 [ z1 =^ z2 ] ;
which gives the B-validity of the crucial implication in the uniqueness case.
Hence one also has VB j= (AC), and the proof is nished at all.
2
The length of this proof indicates that rst-order independence considerations are much more dicult than propositional ones. This proof also uses
on the metatheoretical level set theoretical principles which are part of the
theory ZFC. Therefore this proof of the existence of a B-valued model of ZFC,
also called Boolean valued model of ZFC, is not a consistency proof for ZFC
in an \absolute" sense { it is only a \relative" consistency proof: relative to
the (intuitive) set theoretic principles used (metatheoretically) within this
proof. Even the fact that this proof itself used the axiom of choice is not a
true problem because one knows that the addition of the axiom of choice to
the usual principles of set theory { as e.g. formalized in the theory ZFC {
does not create inconsistency, cf. e.g. [175, 200].
This { weak { relationship between the fact that each B-interpretation
for ZFC is a Boolean valued model of ZFC and the consistency problem
for ZFC is not of prime importance. It is much more important that these
B-interpretations provide a means to get, by suitable choice of the Boolean
algebra B of truth degrees, models of ZFC which give some crucial sentences
of the language of set theory a truth degree dierent from 1, i.e. which are not
models of some suitable sentences of the language of set theory. Therefore the
Boolean valued models of ZFC provide a { very useful { tool for unprovability
proofs relative to the theory ZFC.
For details of such unprovability proofs the reader should consult the
numerous original papers or e.g. the books [36, 287, 320, 464].
24. Consistency Considerations for Set Theory
The B-interpretations VB which have been discussed in Section 23.2, i.e. the
Boolean valued models for the set theory ZFC, are not the proper tool to
discuss the consistency problem for set theory, as mentioned at the end of
Section 23.2. It is interesting to see, however, that there are other approaches
toward consistency investigations for set theory which also refer to manyvalued logic. Approaches of this kind shall be discussed in the present section.
Intuitively set theory, as mainly created1 by Cantor and Dedekind, is
based on two basic principles. The principle of extensionality says that sets
are uniquely characterized by their elements. It may e.g. be written down as
;
(Ext)
8 x 8 y 8 z (z 2 x , z 2 y) , x = y
under the additional assumption that one is only dealing with sets. The \unrestricted" comprehension principle says that sets are arbitrary collections of
objects. It may be written down as
(Comp) 9 x 8 y(y 2 x , H (y))
with reference (in each particular case) to some property H (x) of sets. (Which,
for formal reasons, has to be written down in such a way that y is not a free
variable of the w H (x).)
The naive2 set theory is based just on these two principles. Unfortunately
already the unrestricted comprehension principle (Comp) is inconsistent as
became well known from results of B. Russell3 [470] who took for H (x) the
simple w :(x 2 x) to form according to (Comp) a set xR such that
1 An excellent survey of set theory including its historical aspects is [175]. Life and
work of Cantor are discussed in detail e.g. in [116, 362], and e.g. [154, 254, 358,
375] explain dierent aspects of the history of set theory.
2 Naive set theory, in the usual sense of this word, is based on principles which are
essentially determined only informally. It is, thus, not a formalized mathematical

theory and hence capable of dierent formalizations. Each such formalization
may, however, dier more or less from its \naive" origin.
3 The fact that an unrestricted use of the comprehension principle could lead to
inconsistencies was, however, known earlier e.g. to Cantor himself, who spoke
of \inkonsistente Vielheiten" (i.e. inconsistent collections) and of \absolut unendliche" (i.e. absolutely innite) collections, having in mind e.g. the (proper)
classes of all cardinals or of all ordinals, cf. [116, 415, 432]. Also Burali-Forti
[77] was aware of such inconsistencies. The great advantage of Russell's discov-
558
8 y(y 2 xR , :(y 2 y)) :
Choosing here the particular \set" xR for y yields the contradiction

xR 2 xR , :(xR 2 xR ) ;
(24.1)
which is known as Russell's Paradox.
The various axiomatizations of set theory, which have been developed
to avoid this and other contradictions derivable from (Ext) and (Comp),
usually proceed in the framework of classical (rst-order) logic and restrict the
comprehension principle { and sometimes distinguish between sets and proper
classes. Thus one has, e.g., axiom (ZF2) as well as the axioms (ZF4), (ZF5)
of the system ZFC as particular cases of (Comp). In this way one successfully
avoids in the usual axiomatizations of set theory all the known contradictions.
A consistency proof, however, for these axiomatized set theories does not
exist { and cannot be expected according to the well known incompleteness
theorems of Go del [198, 512].
Other possibilities to approach the consistency problem of naive set theory
could perhaps be opened if one leaves the framework of classical logic and
discusses the principles (Ext) and (Comp) within the framework of some other
system of logic, e.g. in the frameworks of many-valued, of paraconsistent, or
of modal logics. We restrict our considerations to systems of many-valued
logic as a framework for these discussions.
Such a turn is rather natural having in mind that e.g. Bocvar [66] designed in 1938 his three-valued system B 3 for the analysis and avoidance of
the logical and semantical antinomies.4 The problem of the consistency or
inconsistency of the principle (Comp) in a many-valued setting was, however, only attacked in 1957 by Skolem [508]. He chose the Lukasiewicz
rst-order systems L to formalize (Comp) { as well as (Ext) { within their
language. However, the nitely many-valued Lukasiewicz systems Lm are
not suitable for the purpose of having another \consistency behavior" as in
classical logic: each of the systems Lm allows one to derive paradoxes which
are essentially similar to the Russell Paradox as was recognized (already)
in [370].
And this situation is not restricted to the nitely many-valued Lukasiewicz systems. Suppose that S is a system of many-valued (rst-order)
logic with nitely many truth degrees which has an implication connective
! satisfying the detachment property
if j=S H and j=S (H ! G); then j=S G
ery was the extreme simplicity of its construction which does not need to refer
to any more specic set theoretical notions.
4 Even more so because it has been shown in [250] that one can consistently add
to the Peano arithmetic, i.e. to the formalized arithmetic of natural numbers,
a truth predicate { as long as the formalization proceeds within the 3-valued
Lukasiewicz logic, and that one knows for long that the addition of such a
truth predicate gives an inconsistent theory as long as the formalization proceeds
within classical logic.
559
together with j=S (H ! H ) for all ws H; G of S. For such a system S the

absorption property of some level n:
i! (H; G) j= i!n (H; G)
n+1
(Absn )
=1
=1
together with (Comp) allows the construction of a contradiction in S. Such

a construction of a contradiction not even needs the full power of (Comp), it
is already sucient to have a set an;H available which has the property that
for some w H of S, which is not logically S-valid, the formulas
i n
x 2 an;H and ! (x 2 x; H )
=1
are \equivalent" (e.g. in the sense of semantical equivalence, of equivalence

by denition, or of provable equivalence), and that for this notion of \equivalence" there holds true a corresponding equivalence theorem.
In such a case one has
i n
j=S an;H 2 an;H ! ! (an;H 2 an;H ; H ) ;

=1
(24.2)
which means by denition of ! nothing but
n+1
j=S ! (an;H 2 an;H ; H ) ;

=1
and one has also
i n
j=S ! (an;H 2 an;H ; H ) ! an;H 2 an;H :

=1
(24.3)
The absorption property of level n then allows one to get from (24.2) the
antecedent of the implication in (24.3), and therefore to get j=S an;H 2 an;H
by the detachment property. But this together with (24.2) gives j=S H by
an iterated application of detachment. This, however, was supposed not to
hold true by choice of H . The construction of this contradiction is essentially
similar to the derivation of Russell's Paradox.
The crucial point for this derivation of a contradiction is the availability
of the absorption property (Absn ) of some suitable level n. One immediately
sees that each one of the nitely many-valued Lukasiewicz systems Lm has
the absorption property (Absm;1 ). Therefore (Comp) is inconsistent also in
each one of the nitely many-valued Lukasiewicz systems Lm .
The innitely many-valued Lukasiewicz system L1 on the other hand
does not have the absorption property (Absn ) for any level n. To show that
(Absn ) fails in L1 for each n, one simply has to consider in the property
n and 0.
(Absn ) such ws H and G which have truth degrees n+1
Therefore L1 is the only one of the Lukasiewicz systems which reasonably can be taken as a framework for discussing the consistency behavior
560
of the comprehension schema (Comp). For doing this, i.e. for discussing set
theory in the framework of L1 , we suppose that the rst-order language of
L1 has besides the identity symbol $ only one (binary) predicate symbol
" to denote the membership relation { and that all other predicate symbols
have been introduced by denition. For each set of ws of this set theoretic
L-language LL we say that (Comp) is L1 -consistent for i the set
f9x8y(y " x $L H ) j H 2 g
of ws of LL has an L1 -model.
We consider the following particular sets i of ws of LL :
1 = set of all ws H (y; x; x1 ; : : : ; xn ) of LL with their free variables among y; x; x1 ; : : : ; xn , which do not contain quantications,
2 = set of all ws H (y; x) of LL with their free variables among
y; x,
3 = set of all ws H (y; x; x1 ; : : : ; xn ) of LL with their free variables among y; x; x1 ; : : : ; xn , in which one has z1 z2 for
each subformula of the form z1 " z2 which occurs in the
scope of a quantication of z1,
4 = set of all ws H (y; x; x1 ; : : : ; xn ) of LL with their free variables among y; x; x1 ; : : : ; xn , in which the variable y occurs
only in the rst argument place of the membership symbol
" , i.e. which do not contain any subformula of the form
z " y.
Skolem, who initiated this whole line of investigations, has proved in [508]
that (Comp) is L1 -consistent for 1 . A simplied proof is given in [153].
This result was generalized by Chang [93] who proved that (Comp) is L1 consistent for 3 . And the inclusion relation 1 3 is immediately recognized. In [93] it was also proved that (Comp) is L1 -consistent for 2 . And
Fenstad [153] proved that (Comp) is also L1 -consistent for 4 .
Having another look at the ZF-axioms (ZF1), . . . , (ZF7) as given in Section 23.2, cf. p. 542, one recognizes that e.g. in the case of the axiom of
separation (ZF2) no instance of it belongs to the set 2 , and that each one
of the sets 1 ; 3 ; 4 contains only some of the instances of (ZF2). In the
case of the axiom of union (ZF4) the crucial L-formula5
9x(y " x ^ x " x1 ) ;
5 It is common usage to translate the quantiers of classical logic by the standard
quantiers 8; 9 of the L-systems. For the translation of the connectives it is

obviously rather unimportant here whether classical conjunction is translated as
weak or as strong conjunction of the L-systems. The same holds true also for the
further cases we are going to consider here.
561
which has to be considered for the discussion of this axiom in the framework
of the L-systems, is only a member of 4 and not of the other three sets. And
the corresponding L-formula
8x(x " y !L x " x1 ) ;
which has to be considered for the discussion of the power set axiom (ZF5)
in the framework of the L-systems, is not a member of any one of the sets
i , i = 1; : : : ; 4.
This means that the results of Skolem, Chang and Fenstad we mentioned for the L1 -consistency of (Comp) give only a rather limited partial
solution to this problem which was the starting point for the investigations
of Skolem. And the situation becomes even more complicated if one adds
the principle of extensionality (Ext) to these considerations, e.g. translated
in the language LL by the w
;

8x8y 8z (z " x $L z " y) $L x $ y :
(24.4)
Having in mind the absolute point of view for identity $ in many-valued
logic, then it was proved in [93] that one loses the L1 -consistency of (Comp)
for 2 as well as for 3 if one adds the principle (Ext) in the form of (24.4)
in these cases. The reason is that any L1 -interpretation A which is a model
of (Comp) either for ws of 2 or for ws of 3 satises
A j=L 9x8y(y " x $L :(y " y)) ;
which means that there exists for each real number r > 0 and each Aassignment f an object c 2 jAj such that
ValLA (x " x $L :(x " x); f [x=c]) > 1 ; r
holds true, i.e. such that the dierence of ValLA (x " x; f [x=c]) and the truth
degree 12 is \suciently small". Hence one has because of
A j=L 9x8y(y " x $L y " y) ;
A j=L 9x8y(y " x $L y " y Y y " y)
objects a; b 2 jAj such that ValLA (a $ b) = 0 because of a 6= b, but such that
the truth degree of 8z (z " a $L z " b) is dierent from 0. And this contradicts
the principle (Ext).
Having in mind, on the other hand, the liberal point of view for identity $
in many-valued logic, then Chang mentions in [93] that in this case (Comp)
together with (Ext) become L1 -consistent for 2 . It seems that there do not
exist further results for this particular version of the present L1 -consistency
problem.
All these investigations of Skolem, Chang and Fenstad which we have
discussed up to now, and which do not solve the L1 -consistency of (Comp),
are model theoretic in nature, i.e. essentially consist in the construction of
suitable L1 -models for subsets of all the ws which fall under the comprehension schema (Comp).
562
An interesting progress was reached with proof theoretic methods, i.e.

with syntactical investigations on suitable logical calculi, by Grishin [229,
231] and White [572]. The rst one of them developed a sequent as well
as a Hilbert type calculus for some (decidable) rst-order system L0 which
proved to be a weaker system as the system L1 . Its logically valid ws are
characterized as the class of all those ws which always assume value 1 by
any interpretation in an integral commutative (completely) lattice-ordered
monoid hA; 6; +; ;i (with universal upper bound 1 and universal lower bound
0) which is equipped with an additional unary operation ; satisfying the
laws:6
; ; a = a and a 6 b , (;a) + b = 1 :
Hence they are (in their propositional part) primarily formulated in negation
and disjunction. The implication connective then is introduced via
H1 ! H2 =def :H1 _ H2 ;
and the conjunction is understood as the lattice meet. This means that,
contrary to the situation in monoidal logic, in L0 conjunction and implication
do not form an adjoint pair. This (identity free) system is proved in [229] to
remain consistent if one enriches it with class terms (and the membership
predicate) in its language and adds also the axiom of comprehension.
And the starting point for the proof theoretic considerations in [572] is
provided by the results of Hay [256] and Belluce/Chang [44] discussed in
Section 9.3. The logical calculus under consideration in [572] is an extension
of { an inessential modication of { the calculus which is determined by the
axiom schemata (L1 1); : : : ; (L1 10) of Section 9.3. This calculus refers to a
set theoretic language which has the membership predicate " as its only
predicate symbol, but which allows for class terms fx k H g, H any w, and
which adds the further axiom schemata
(L1 11) H [x=t] !L t " fx k H g ,
(L1 12) t " fx k H g !L H [x=t]
for each term t and each L-formula H . (The substitution operation in these
schemata is to be taken in the liberalized sense, i.e. including renaming of
bound variables if necessary to make sure that no free occurrence of x in
H occurs in the scope of a quantication of a variable which occurs free in
the term t.) Additionally also in the schemata (L1 7); (L1 8) one allows for
the (liberalized) substitution of any terms instead of only variables y. The
inference rules of the logical calculus under consideration in [572] are the
rules (MP) and (Gen) together with the innitary rule of inference
Qn
H
Y
i=1 H for each n 1 ;
(inf )
L
6 The algebraic properties of these structures are investigated in more detail in
[230].
563
which is just rule (9.119) from Section 9.3. The result of Theorem 9.3.4 remains true: this logical calculus allows the derivation of just the L1 -logically
valid formulas.
Now one can form for each w H the class term t1 = fz k H [y=z ]g and
derive via (L1 1); : : : ; (L1 10) the w y " t1 $L H and hence also the w
9x8y(y " x $L H ) :
(24.5)
Therefore the problem of the L1 -consistency of (Comp) can be reformulated
as the problem to prove that there is no inconsistency, i.e. no w of the
form G & :G, derivable from the axiom schemata (L1 1); : : : ; (L1 12) via the
inference rules (MP), (Gen) and (infL ).
This (reformulated) problem was solved in [572] by transforming the
logical calculus K L (of Hilbert type), which is determined by the axiom
schemata (L1 1); : : : ; (L1 12) and the inference rules (MP), (Gen), and (infL ),
into a calculus of natural deduction7 K which does not allow one to derive
any inconsistency of the form G & :G, but which derives all the L-formulas
which are K L -derivable.
These proof theoretic considerations did not include the principle (Ext) of
extensionality up to now. This completion can e.g. be reached by introducing
a graded identity via the denition
x $ y =def 8z (x " z $L y 2 z ) :
(24.6)
The principle (Ext) then becomes
;

8x8y 8z (z " x $L z 2 y) !L x $ y :
(24.7)

Unfortunately one then is able to derive in the calculus K , which is extended
by (24.7) as additional axiom, the w x $ y _:x $ y and hence also to derive
each one of the ws A _:A for any w A. Thus the extended logical calculus
is no more sound w.r.t. L1 and hence loses its meaning.
Even worse: it is neither the axiom (24.7) of extensionality nor the particular logical calculus K which causes this diculty, but the schema (Comp)
of comprehension in the L1 -setting. This can be seen from a result of Ragaz
who has proved in [434, 436] that one already has for the set
1 = f9x8y(y " x $L H ) j H 2 1 g
of ws the (semantic) results
1 j=L1 8x8y;(x $ y _ :x $ y) ;

1 j=L1 8x9y :x $ y ^ 8z (z " x $L z 2 y)
with $ dened as in (24.6).
7 Calculi of natural deduction are essentially characterized by the fact that they
have quite a lot of inference rules, viz. usually for each connective and each
quantier a rule which governs its introduction and another one which governs
its elimination, and often only very few axioms (or even no axioms at all).
564
Denition (24.6) of the graded identity $ is quite natural, at least from

the point of view of set theory in the framework of classical logic. Nevertheless
these last mentioned results oer the problem whether this denition (24.6)
provides the \right" approach toward graded identity in the L1 -framework,
or whether other approaches toward graded identities in many-valued logic
could be preferable here. The problem is open up to now.
The interesting and important results of [572] hence do not really solve the
L1 -consistency problem for naive set theory. Similarly the problem remains
(partly) open whether another framework as that one which is provided by
the Lukasiewicz systems and their standard connectives and quantiers
oers a better situation for the discussion of the consistency problem for
naive set theory. It was shown in [353], however, that for most of the usual
systems of many-valued logic the situation is similar as the situation is for the
nitely many-valued systems Lm : (Comp) remains inconsistent. Exceptions
among the better known classes of systems seem to be only an innitely
many-valued version of the Post systems, and another version of L1 which
has all truth degrees 12 as designated ones and which has the implication
H !1 G =def :H _ G
as its \main" implication.
It seems that one of the central diculties, particularly for the model
theoretic approach, comes from the fact that one does not have a suciently
clear and natural intuition of what a universe of \many-valued sets" should
look like. In classical set theory such an idea is provided by the intuition of
the cumulative universe of all sets which is subdivided into levels which may
be numbered by the ordinals, which have either the empty set or some class
of urelements as lowest level, and which constructs the levels successively by
transnite iteration of the power set operation. Perhaps the further development of the theory of fuzzy sets, which has been discussed in Chapter 18,
sometimes may change this situation. However, actually also these investigations do not provide suciently clear intuitions for the \higher levels" of
such a cumulative universe of \many-valued" sets.
As a side remark we mention also another possibility to attack the consistency problem of set theory via many-valued logic: one can try to avoid
one of the core axioms (Ext) or (Comp) of naive set theory, but to retain
as much as possible of the other axioms of some standard system of classical
set theory, and can then discuss the problem of consistency of this \reduced"
system of set theoretic axioms in the setting of many-valued logic. Because
one is convinced in classical set theory to have consistent systems with full
extensionality and weakened comprehension,8 it seems less advisable to avoid
or weaken the comprehension axiom here, but it may be interesting to avoid
8 Of course, this is not a provable result of set theory according to the second
incompleteness theorem of Godel. However, all set theorists are convinced that
the systems ZF of Zermelo-Fraenkel and NGB of von Neumann-GodelBernays as well as also the system KM of Kelley-Morse really are consistent.
565
or weaken the axiom of extensionality { having in mind that this may be

combined with some intensional understanding of the equality sign. Such approaches have, e.g., been discussed in [569, 570] but shall not be considered
in detail here.
566
References
References
1. Ackermann, R. (1967): An Introduction to Many-Valued Logics. Routledge &

Kegan Paul, London.
2. Adillon, R.J. { Verdu, V. (1997): On product logic, Soft Computing 1, 141{
146.
3. Akama, S. (1997): A proof method for the six-valued logic for incomplete information, in: 27th Internat. Symp. Multiple-Valued Logic, Antigonish/Nova
Scotia, 1997, Proceedings, IEEE Computer Soc., Piscataway/NJ, 223{226.
4. Alsina, C. { Trillas, E. { Valverde, L. (1983): On some logical connectives
for fuzzy sets theory, J. Math. Anal. Appl., 93, 15{26.
5. Alsina, C. (1996): As you like them: connectives in fuzzy logic, in: 26th Internat. Symp. Multiple-Valued Logic, Santiago de Compostella, 1996, Proceedings,
IEEE Computer Soc., Piscataway/NJ, 2{7.
6. Anderson, A.R. { Belnap, N.D. (1975{92): Entailment: The Logic of Relevance and Necessity. Vols. 1{2, Princeton Univ. Press, Princeton.
7.
Aqvist, L. (1962): Re ections on the logic of nonsense, Theoria 28, 138{157.
8. Arruda, A.I. (1977): On the imaginary logic of N.A. Vasil'ev, in: A.I. Arruda
et al. (eds.), Non-Classical Logics, Model Theory and Computability. NorthHolland Publ. Comp., Amsterdam, 3{24.
9. Asser, G. (1959): Einfuhrung in die mathematische Logik. Part I. Aussagenkalkul. Teubner, Leipzig.
10. Asser, G. (1972): Einfuhrung in die mathematische Logik. Part II: Pradikatenkalkul der ersten Stufe. Teubner, Leipzig.
11. Asser, G. (1981): Einfuhrung in die mathematische Logik. Part III: Pradikatenlogik hoherer Stufe. Teubner, Leipzig.
12. Avron, A. (1991): Natural 3-valued logics { Characterization and proof theory,
J. Symbolic Logic 56, 276{294.
13. Avron, A. (1996): The method of hypersequents in proof theory of propositional non-classical logics, in: W. Hodges et al. (eds.), Logic: From Foundations
to Applications. European Logic Colloq., Keele, UK, 1993, Clarendon Press,
Oxford, 1{32.
14. Avron, A. (1999): On the expressive power of three-valued and four-valued
languages, J. of Logic and Computation 9, 977{994.
15. Baaz, M. (1992): Automatisches Beweisen fur endlichwertige Logiken, Mitteilungen math. Gesellsch. Hamburg 12, 1141{1155.
16. Baaz, M. (1996): Innite-valued Godel logics with 0-1-projections and relativizations, in: P. Hajek (ed.), Godel '96. Logical Foundations of Mathematics,
Computer Science and Physics { Kurt Godel's Legacy, Lecture Notes Logic,
vol. 6, Springer, Berlin, 23{33.
17. Baaz, M. { Ciabattoni, A. { Fermuller, C.G. { Veith, H. (1998): Proof
theory of fuzzy logics: Urquhart's C and related logics, in: L. Prim (ed.), Math.
568
References
Foundations of Computer Science { MFCS '98, 23rd Internat. Symp., Proceedings, Lecture Notes Computer Sci., vol. 1450, Springer, Berlin, 203{212.
18. Baaz, M. { Fermuller, C.G. (1992): Resolution for many-valued logics, in:
A. Voronkov (ed.), Logic Programming and Automated Reasoning. Internat.
Conf. LPAR '92, St. Petersburg, Russia, July 1992. Lecture Notes Artif. Intell.,
vol. 624, Springer, Berlin, 107{118.
19. Baaz, M. { Fermuller, C.G. (1995): Resolution-based theorem proving for
many-valued logics, J. Symbolic Comput. 19, 353{391.
20. Baaz, M. { Fermuller, C.G. { Quirchmayr, G. { Zach, R. (1993): A
many-valued logic for default reasoning, in: L. Gun/R. Onvural/E. Gelenbe
(eds.), Proc. 8th Internat. Symp. Computer and Information Sci., ISCIS VIII,
Istanbul, Nov. 1993, Middle East Techn. Univ., Ankara, 476{479.
21. Baaz, M. { Fermuller, C.G. { Salzer, G. (200x): Automated deduction
for many-valued logics, in: A. Robinson/A. Voronkov (eds.), Handbook of Automated Reasoning. Elsevier (to appear).
22. Baaz, M. { Fermuller, C.G. { Salzer, G. { Zach, R. (1998): Labeled
calculi and nite-valued logics, Studia Logica 61, 7{33.
23. Baaz, M. { Fermuller, C.G. { Zach, R. (1993): Systematic construction
of natural deduction systems for many-valued logics, in: 23rd Internat. Symp.
Multiple-Valued Logic, Sacramento/Ca., 1993, Proceedings, IEEE Computer
Soc., Piscataway/NJ, 208{213.
24. Baaz, M. { Fermuller, C.G. { Zach, R. (1994): Elimination of cuts in rstorder nite-valued logics, J. Inform. Process. Cybernetics EIK 29, 333{355.

25. Baaz, M. { Hajek, P. { Krajicek, J. { Svejda,
D. (1998): Embedding
logics into product logic, Studia Logica 61, 35{47.
26. Baaz, M. { Hajek, P. { Montagna, F. { Veith, H. (200x): Complexity of
t-tautologies. Annals Pure applied Logic (to appear).
27. Baaz, M. { Leitsch, A. { Zach, R. (1996): Incompleteness of a rst-order
Godel logic and some temporal logics of programs, in: E. Borger et al. (eds.)
Computer Science Logic. Selected Papers from CSL '95, Springer, Berlin, 1{15.
28. Baaz, M. { Zach, R. (1994): Approximating propositional calculi by nitevalued logics, in: 24th Internat. Symp. Multiple-Valued Logic, Boston/Mass.,
1994, Proceedings, IEEE Computer Soc., Piscataway/NJ, 257{263.
29. Baaz, M. { Zach, R. (1998): Compact propositional Godel logics, in: 28th Internat. Symp. Multiple-Valued Logic, Fukuoka, 1998, Proceedings, IEEE Computer Soc., Piscataway/NJ, 108{113.
30. Bar, G. { Rohleder, H. (1967): U ber einen arithmetisch-aussagenlogischen
Kalkul und seine Anwendung auf ganzzahlige Optimierungsprobleme, Elektronische Informationsverarbeitung Kybernetik EIK 3, 171{195.
31. Bandemer, H. { Gottwald, S. (1995): Fuzzy Sets, Fuzzy Logic, Fuzzy Methods with Applications. Wiley, Chichester.
32. Barba, J. (1993): A modal reduction for partial logic, J. Philosophical Logic
22, 429{435.
33. Barwise, J. (ed.) (1977): Handbook of Mathematical Logic. North-Holland
Publ. Comp., Amsterdam.
34. Baudry, L. (1989): The Quarrel over Future Contingents (Louvain 1465{
1475). Kluwer, Dordrecht.
35. Beaver, D.I. (1997): Presupposition, in: J. van Benthem/A. ter Meulen (eds.),
Handbook of Logic and Language. Elsevier, Amsterdam and MIT Press, Cambridge/Ma., 939{1008.
36. Bell, J.L. (1977): Boolean-Valued Models and Independence Proofs in Set Theory. Clarendon Press, Oxford.
References
569
37. Bell, J.L. { Machover, M. (1977): A Course in Mathematical Logic. NorthHolland Publ. Comp., Amsterdam.
38. Bell, J.L. { Slomson, A.B. (1969): Models and Ultraproducts. North-Holland
Publ. Comp., Amsterdam.
39. Bellmann, R. { Giertz, M. (1973): On the analytic formalism of the theory
of fuzzy sets, Information Sciences 5, 149{156
40. Bellmann, R.E. { Zadeh, L.A. (1970): Decision-making in a fuzzy environment. Management Science 17, B141{B164. [cf. also[590]]
41. Bellmann, R.E. { Zadeh, L.A. (1977): Local and fuzzy logics, in: J.M.
Dunn/G. Epstein (eds.), Modern Uses of Multiple-Valued Logic. Reidel, Dordrecht, 105{165.
42. Belluce, L.P. (1964): Further results on innite valued predicate logic, J.
Symbolic Logic 29, 69{78.
43. Belluce, L.P. (1997): Generalized fuzzy connectives on MV-algebras, J. Math.
Anal. Appl. 205, 485{499.
44. Belluce, L.P. { Chang, C.C. (1963): A weak completeness theorem for innite valued rst-order logic, J. Symbolic Logic 28, 43{50.
45. Belnap, N.D. (1977): How a computer should think, in: G. Ryle (ed.) Contemporary Aspects of Philosophy. Oriel Press, Stockeld, 30{56.
46. Belnap, N.D. (1977): A useful four-valued logic, in: J.M. Dunn/G. Epstein
(eds.), Modern Uses of Multiple-Valued Logic. Reidel, Dordrecht, 8{37.
47. Bencivenga, M. (1986): Free logic, in: D. Gabbay/F. Guenthner (eds.), Handbook of Philosophical Logic , vol. 3, Reidel, Dordrecht, 373{426.
48. Bendova, K. (1999) A note on Godel fuzzy logic, Soft Computing 2, 167.
49. Bergmann, M. (1981): Presupposition and two-dimensional logic, J. Philosophical Logic 10, 27{53.
50. Bergmann, M. (1981) Only, even, and clefts in two-dimensional logic, in:
11th Internat. Symp. Multiple-Valued Logic, Oklahoma 1981, Proceedings, IEEE
Computer Soc., Long Beach, 117{123.
51. Berka, K. { Kreiser, L. (1971): Logik-Texte. Kommentierte Auswahl zur
Geschichte der modernen Logik. Akademie-Verlag, Berlin (3rd, rev. ed. 1983).
52. Bernays, P. (1926): Axiomatische Untersuchungen des Aussagenkalkuls der
\Principia Mathematica", Mathematische Zeitschrift 25, 305{320.
53. Bertolini, F. (1971): Kripke models and manivalued logics, in: Symposia
Mathematica , vol. 5, Academic Press, London, 113{131.
54. Beth, E.W. (1955): Semantic entailment and formal derivability, Mededelingen
Koninkl. Nederlandske Akad. Wetensch., Aft. Letterkunde, vol. 18, no. 13, 309{
342.
55. Beth, E.W. (1965): The Foundations of Mathematics. North-Holland Publ.
Comp., Amsterdam.
56. Biacino, L. { Gerla, G. (1998): An extension principle for closure operators,
J. Math. Anal. Appl. 198, 1{24.
57. Biacino, L. { Gerla, G. (1998): Logics with approximate premises, Internat.
J. Intelligent Systems 13, 1{10.
58. Biacino, L. { Gerla, G. { Ying, M.S. (2000): Approximate reasoning based
on similarity, Math. Logic Quarterly 46, 77{86.
59. Birkhoff, G. (1948): Lattice Theory. American Math. Soc., New York.
60. Black, M. (1937): Vagueness: an exercise in logical analysis, Philosophy of
Science 4, 427{455.
61. Black, M. (1963): Reasoning with loose concepts, Dialogue 2, 1{12.
62. Blamey, S. (1986): Partial logic, in: D. Gabbay/F. Guenthner (eds.), Handbook
of Philosophical Logic, vol. 3, Reidel, Dordrecht, 1{70.
570
References
63. Blau, U. (1978): Die dreiwertige Logik der Sprache: ihre Syntax, Semantik und
Anwendung in der Sprachanalyse. de Gruyter, Berlin.
64. Bochmann, A. (1998): Biconsequence relations: a four-valued formalism of
reasoning with inconsistency and incompleteness, Notre Dame J. Formal Logic
39, 47{73.
65. Bocklisch, S.F. { Orlovski, S. { Peschel, M. { Nishiwaki, Y. (eds.)
(1986): Fuzzy Sets Applications, Methodological Approaches, and Results.
Akademie-Verlag, Berlin.
66. Bocvar, D.A. (1938): Ob odnom trechznacnom iscislenii i ego primenenii
k analizu paradoksov klassiceskogo rassirennogo funkcional'nogo iscislenija,
Matematiceskij Sbornik 4 (46), 287{308.
[English translation: Bochvar, D.A., On a three-valued logical calculus and its
application to the analysis of the paradoxes of the classical extended functional
calculus, History and Philosophy of Logic 2, 87{112.]
67. Bocvar, D.A. (1943): K voprosu o neprotivorecivosti odnogo trechznacnogo
iscislenija, Matematiceskij Sbornik 12 (54), 353{369.
68. Bocvar, D.A. { Finn, V.K. (1972): O mnogoznacnych logikach, dopuskajuscich formalizaciju analiza antinomij. 1, in: Issledovanija po Matematiceskoj Lingvistike, Matematiceskoj Logike i Informacionnym Jazykam. Nauka,
Moscow, 238{295.
69. Bocvar, D.A. { Finn, V.K. (1974): O kvazilogiceskich funkcijach, in: Issledovanija po Formalizovannym Jazykam i Neklassiceskim Logikam. Nauka,
Moscow, 200{213.
70. Bocvar, D.A. { Finn, V.K. (1976): Nekotorye dopolnenija k stat'jam o
mnogoznacnych logikach, in: Issledovanija po Teorii Mnozestv i Neklassiceskim
Logikam. Nauka, Moscow, 265{325.
71. Bodenhofer, U. (1999): A Similarity-Based Generalization of Fuzzy Orderings. Schriften der J.-Kepler-Univ. Linz, Reihe C { Technik und Naturwiss.,
vol. 26, Univ.-Verlag R. Trauner, Linz.
72. Bodenhofer, U. (2000): A similarity-based generalization of fuzzy orderings preserving the classical axioms, Internat. J. Uncertainty, Fuzziness and
Knowledge-Based Systems 8, 593{610.
73. Bolc, L. { Borowik, P. (1992): Many-Valued Logics, 1. Theoretical Foundations. Springer, Berlin.
74. Boicescu, V. { Filipoiu, A. { Georgescu, G. { Rudeanu, S. (1991):
Lukasiewicz-Moisil-Algebras. Annals Discr. Math., vol. 49, North-Holland Publ.
Comp., Amsterdam.
75. Borkowski, L. (1976): Formale Logik. Akademie-Verlag, Berlin.
76. Brouwer, L.E.J. (1975): Collected Works. Vol. 1: Philosophy and Foundations
of Mathematics (ed.: A. Heyting). North-Holland Publ. Comp., Amsterdam.
77. Burali-Forti, C. (1897): Una questione sui numeri transniti, Rendiconti
Circ. Mat. Palermo 11, 154{164.
78. Buridan, J. (1966): Sophisms on Meaning and Truth. Translated and with an
introduction by T.K. Scott, Appleton-Century-Crofts, New York.
79. Burton-Roberts, N. (1989): On Horn's dilemma, presupposition and negation, J. of Linguistics 25, 95{125.
80. Burton-Roberts, N. (1990): Trivalence, gapped bivalence, and ambiguity of
negation: a reply to Seuren, J. of Linguistics 26, 455{470.
81. Butnariu, D. { Klement, E.P. (1993): Triangular Norm-Based Measures
and Games with Fuzzy Coalitions. Kluwer, Dordrecht.
82. Byrd, M. (1979): A formal interpretation of Lukasiewicz' logics, Notre Dame
J. Formal Logic 20, 366{368.
References
571
83. Carnielli, W.A. (1987): Systematization of nite many-valued logics through

the method of tableaux, J. Symbolic Logic 52, 473{493.
84. Carnielli, W.A. (1987): The problem of quanticational completeness and
the characterization of all perfect quantiers in 3-valued logic, Zeitschr. math.
Logik Grundl. Math. 33, 19{29.
85. Carnielli, W.A. (1991): On sequents and tableaux for many-valued logics,
J. Non-Classical Logic 8, 59{76.
86. Carvallo, M. (1968): Logique a trois valeurs. Logique a seuil. Gauthier-Villars,
Paris.
87. Cat-Ho, Nguyen (1973): Generalized Post Algebras and Their Applications
to Some Innitary Many-Valued Logics. Dissertationes Math., vol. 57, PWN,
Warsaw.
88. Chakraborty, M.K. { Das, M. (1983): Studies in fuzzy relations over fuzzy
subsets, Fuzzy Sets Systems 9, 79{89.
89. Chakraborty, M.K. { Das, M. (1983): On fuzzy equivalence. I, II, Fuzzy
Sets Systems 11, 185{193 and 299{307.
90. Chang, C.C. (1958): Proof of an axiom of Lukasiewicz, Transactions American
Mathematical Society 87, 55{56.
91. Chang, C.C. (1958): Algebraic analysis of many valued logics, Transactions
American Mathematical Society 88, 476{490.
92. Chang, C.C. (1959): A new proof of the completeness of the Lukasiewicz
axioms, Transactions American Mathematical Society 93, 74{80.
93. Chang, C.C. (1963): The axiom of comprehension in innite valued logic,
Mathematica Scandinavica 13, 9{30.
94. Chang, C.C. (1965): Innite valued logic as a basis for set theory, in: Y. BarHillel et al. (eds.), Logic, Methodology and Philosophy of Science. Proceedings
Internat. Congress Jerusalem 1964, North-Holland Publ. Comp., Amsterdam,
93{100.
95. Chang, C.C. { Keisler, H.J. (1966): Continuous Model Theory. Princeton
University Press, Princeton.
96. Chang, C.C. { Keisler, H.J. (1973): Model Theory. North-Holland Publ.
Comp., Amsterdam.
97. Chapin, E.W. (1974{75): Set-valued set theory. I{II, Notre Dame J. Formal
Logic 15, 614{634; 16, 255{267.
98. Chellas, B.F. (1980): Modal Logic. An Introduction. Cambridge Univ. Press,
Cambridge.
99. Cignoli, R. (1970): Moisil Algebras. Notas de Matematica 27, Universidade
Nacional del Sur, Bahia Blanca.
100. Cignoli, R. (1980): Some algebraic aspects of many-valued logics, in: A.I. Arruda/N.C.A. da Costa/A.M. Sette (eds.), Proceedings 3rd Brazilian Conference
on Mathematical Logic. Sociedade Brasileira de Logica, Sao Paulo, 49{69.
101. Cignoli, R. (1982): Proper n-valued Lukasiewicz algebras as S-algebras of
Lukasiewicz n-valued propositional calculi, Studia Logica 41, 3{16.
102. Cignoli, R. (1984): An algebraic approach to elementary theories based on
n-valued Lukasiewicz logics, Zeitschr. math. Logik Grundl. Math. 30, 87{96.
103. Cignoli, R. { Esteva, F. { Godo, L. { Torrens, A. (2000): Basic Fuzzy
Logic is the logic of continuous t-norms and their residua, Soft Computing 4,
106{112.
104. Cignoli, R. { d'Ottaviano, I.M.L. { Mundici, D. (2000): Algebraic Foundations of Many-Valued Reasoning. Trends in Logic, Studia Logica Library,
vol. 7, Kluwer, Dordrecht.
572
References
105. Cignoli, R. { Mundici, D. (1997): An invitation to MV algebras, in: M.

Droste/R. Goebel (eds.), Advances in Algebra and Model Theory. Gordon and
Breach, Readings, UK, 171{197.
106. Cignoli, R. { Torrens, A. (2000): An algebraic analysis of product logic,
Multiple{Valued Logic 5, 45{65.
107. Cleave, J.P. (1974): The notion of logical consequence in the logic of inexact
predicates, Zeitschr. math. Logik Grundl. Math. 20, 307{324.
108. Cohen, P.J. (1963{64): The independence of the continuum hypothesis. I{II,
Proc. Nat. Acad. Sci. USA 50, 1143{1148; 51, 105{110.
109. Cohen, P.J. (1969): Decision procedures for real and p-adic elds, Communications Pure Applied Mathematics 22, 131{151.
110. Cohn, P.M. (1981): Universal Algebra. Reidel, Dordrecht.
111. D'Agostino, M. { Gabbay. D. { Hahnle, R. { Posegga, J.(eds.) (1999):
Handbook of Tableau Methods. Kluwer, Dordrecht.
112. da Costa, N.C.A. { Alves, E.H. (1981): Relations between paraconsistent
and many-valued logic, Bulletin Section of Logic 10, 185{191.
113. Dalla Chiara, M.L. (1976): A general approach to non-distributive logics,
Studia Logica 35, 139{162.
114. Dalla Chiara, M.L. (1981): Logical foundations of quantum mechanics,
in: E. Agazzi (ed.), Modern Logic { A Survey. Synthese Library 149, Reidel,
Dordrecht, 331{351.
115. Dalla Chiara, M.L. (1985): Quantum logic, in: D. Gabbay/F. Guenthner
(eds.), Handbook of Philosophical Logic, vol. 3, Reidel, Dordrecht, 427{469.
116. Dauben, J. (1979): Georg Cantor: His Mathematics and Philosophy of the
Innite. Harvard Univ. Press, Cambridge/Mass.
117. De Baets, B. { Mesiar, R. (1996): Residual implicators of continuous tnorms, in: H.-J. Zimmermann (ed.), Fourth European Congress on Intelligent
Techniques and Soft Computing, EUFIT '96, Aachen, Sept. 1996. Proceedings,
Verlag Mainz, Aachen, 27{31.
118. Dienes, P. (eds.) (1949): On ternary logic, J. Symbolic Logic 14, 85{94.
119. Dilworth, R.P. (1938): Abstract residuum over lattices, Bulletin American
Mathematical Society 44, 262{268.
120. di Nola, A. { Ventre, A.G.S. (eds.) (1986): The Mathematics of Fuzzy
Systems. Interdisciplinary Systems Research, vol. 88, Verlag TU V Rheinland,
Cologne.
121. Dishkant, H. (1978): An extension of the Lukasiewicz logic to the modal
logic of quantum mechanics, Studia Logica 37, 149{155.
122. Doherty, P. { Lukasewicz, W. (1992): Defaults as rst-class citizens, in:
22nd Internat. Symp. Multiple-Valued Logic, Sendai, 1993, Proceedings, IEEE
Computer Soc., Piscataway/NJ, 146{154.
123. Dombi, J. (1982): A general class of fuzzy operators, the DeMorgan class of
fuzzy operators and fuzziness measures induced by fuzzy operators, Fuzzy Sets
Systems 8, 149{163.
124. Domingo, X. { Trillas, E. { Valverde, L. (1981): Pushing LukasiewiczTarski implication a little farther, in: 11. Internat. Symp. Multiple-Valued Logic,
Norman/Oklahoma, 1981, Proceedings, IEEE Computer Soc., New York, 232{
243.
125. Drake, F.R. (1974): Set Theory. An Introduction to Large Cardinals. NorthHolland Publ. Comp., Amsterdam.
126. Dubois, D. { Prade, H. (1980): Fuzzy Sets and Systems. Theory and Applications. Academic Press, New York.
127. Dubois, D. { Prade, H. (1992): Gradual inference rules in approximate
reasoning, Information Sciences 61, 103{122.
References
573
128. Duffy, M.J. (1979): Modal interpretations of three-valued logic. I{II, Notre
Dame J. Formal Logic 20, 647{657; 658{673.
129. Dugundji, J. (1940): Note on a property of matrices for Lewis' and Langford's
calculi of propositions, J. Symbolic Logic 5, 150{151.
130. Dumitriu, A. (1971): Logica polivalenta. Viata Literara, Bucharest.
131. Dummett, M. (1959): A propositional calculus with denumerable matrix, J.
132. Dummett, M. (1995): Bivalence and vagueness, Theoria 61, 201{216.
133. Dunn, J.M. (1976): Intuitive semantics for rst-degree entailments and `coupled trees', Philosophical Studies 29, 149{168.
134. Dunn, J.M. (1986): Relevance logic and entailment, in: D. Gabbay/F. Guenthner (eds.), Handbook of Philosophical Logic, vol. 3, Reidel, Dordrecht, 117{224.
135. Dunn, J.M. (1998): A comparative study of various semantical treatments of
negation: a history of formal negation, in: D. Gabbay/H. Wansing (eds.), What
is Negation. Kluwer, Dordrecht, 23{51.
136. Dunn, J.M. { Epstein, G. (eds.) (1977): Modern Uses of Multiple-Valued
Logic. Reidel, Dordrecht.
137. Dunn, J.M. { Meyer, R.K. (1971): Algebraic completeness results for Dummett's LC and its extensions, Zeitschr. math. Logik Grundl. Math. 17, 225{230.
138. Dwinger, Ph. (1968): Generalized Post algebras, Bull. Acad. Polon. Sci., Ser.
Sci. Math., Astronom., Phys., 16, 559{563.
139. Dwinger, Ph. (1977): A survey of the theory of Post algebras and their
generalizations, in: J.M. Dunn/G. Epstein (eds.), Modern Uses of MultipleValued Logic. Reidel, Dordrecht, 53{75.
140. Ebbinghaus, H.-D. (1969): U ber eine Pradikatenlogik mit partiell denierten
Pradikaten und Funktionen, Archiv math. Logik Grundlagenforsch. 12, 39{53.
141. Epstein, G. (1960): The lattice theory of Post algebras, Transactions American Mathematical Society 95, 300{317.
[Reprinted in: D.C. Rine (ed.) Computer Science and Multiple Valued Logic.
North-Holland Publ. Comp., Amsterdam 1977, 17{34.]
142. Epstein G. (1990): The Semantic Foundations of Logic. Vol.1: Propositional
Logics. Kluwer, Dordrecht.
143. Epstein G. (1993): Multiple-Valued Logic Design. Institute of Physics Publishing, Bristol.
144. Epstein, G. { Horn, A. (1974): Chain based lattices, Pacic J. Mathematics
55, 65{84.
145. Epstein, G. { Horn, A. (1976): Logics which are characterized by subresiduated lattices, Zeitschr. math. Logik Grundl. Math. 22, 199{210.
146. Epstein, G. { Rasiowa,
H. (1986): On P-algebraic extensions of Post algebras of order !+, in: 16th Internat. Symp. Multiple-Valued Logic, Blacksburg/Va., 1986, Proceedings, IEEE Computer Soc., New York, 4{7.
147. Epstein,
G. { Rasiowa, H. (1990): Theory and uses of Post algebras of order
! + !. I, in: 20th Internat. Symp. Multiple-Valued Logic, Charlotte/NC, 1990,
Proceedings, IEEE Computer Soc., New York, 42{47.
148. Epstein,
G. { Rasiowa, H. (1991): Theory and uses of Post algebras of order
! + !. Part II, in: 21st Internat. Symp. Multiple-Valued Logic, Victoria/B.C.,
1991, Proceedings, IEEE Computer Soc., New York, 248{254.
149. Esteva, F. { Trillas, E. { Domingo, X. (1981): Weak and strong negation
functions for fuzzy set theory, in: 11. Internat. Symp. Multiple-Valued Logic,
Norman/Oklahoma, 1981, Proceedings, IEEE Computer Soc., New York, 23{
26.
574
References
150. Esteva, F. { Godo, L. (1999): QBL: towards a logic for left-continuous

t-norms, in: European Soc. Fuzzy Logic and Technol.: Eus at-Estylf Joint Conference, Palma (Mallorca), 1999, Proceedings, Univ. de les Illes Balears, 35{37.
151. Evans, T. { Schwartz, P.B. (1958): On Slupecki T-functions, J. Symbolic
Logic 23, 267{270.
152. Faucett, W.M. (1955): Compact semigroups irreducibly connected between
two idempotents, Proceedings American Mathematical Society 6, 741{747.
153. Fenstad, J.E. (1964): On the consistency of the axiom of comprehension in
the Lukasiewicz innite valued logic, Mathematica Scandinavica 14, 65{74.
154. Ferreiros, J. (1999): Labyrinth of Thought. A History of Set Theory and its
Role in Modern Mathematics. Birkhauser, Basel.
155. Feys, R. (1965): Modal Logics. Gauthier-Villars, Paris, and Nauwelaerts Eds.,
Louvain.
156. Finch, P.D. (1969): On the structure of quantum logic, J. Symbolic Logic 34,
275{282.
157. Finn, V.K. (1971): Ob aksiomatizacii nekotorych trechznacnych logik, Naucno-Techniceskaja Informacija, ser. 2, no. 11, 16{20.
158. Finn, V.K. (1974): A criterion of functional completeness for B3, Studia Logica 33, 121{125.
159. Finn, V.K. (1974): Aksiomatizacii nekotorych trechznacnych iscislenij vyskazyvanij i ich algebr, in: Filosoja v Sovremennom Mire. Filosoja i Logika.
Nauka, Moscow, 398{438.
160. Fisch, M. { Turquette, A.R. (1966): Peirce's triadic logic, Transactions
Charles Sanders Peirce Society 2, 71{85.
161. Fitting, M.C. (1969): Intuitionistic Logic, Model Theory and Forcing. NorthHolland Publ. Comp., Amsterdam.
162. Fitting, M.C. (1983): Proof Methods for Modal and Intuitionistic Logics.
Reidel, Dordrecht.
163. Fitting, M.C. (1988): Logic programming on a topological bilattice, Fundamenta Informaticae 11, 209{218.
164. Fitting, M.C. (1989): Bilattices and the theory of truth, J. Philosophical
Logic 18, 225{256.
165. Fitting, M.C. (1990): First-Order Logic and Automated Theorem Proving.
Springer, New York.
166. Fitting, M.C. (1991/92): Many-valued modal logics. I{II, Fundamenta Informaticae 15, 235{254; 17, 55{73.
167. Fitting, M.C. (1995): Tableaus for many-valued modal logic, Studia Logica
55, 63{87.
168. Fodor, J. (1993): A new look at fuzzy connectives, Fuzzy Sets Systems 57,
141{148.
169. Fodor, J. (1993): On contrapositive symmetry of implications in fuzzy logic,
in: H.-J. Zimmermann (ed.), First European Congress on Fuzzy and Intelligent
Technologies, EUFIT '93, Aachen, Sept. 1993. Proceedings, Verlag Augustinus
Buchhandlung, Aachen, 1342{1348.
170. Fodor, J. { Roubens, M. (1994): Fuzzy Preference Modeling and Multicriteria Decision Support. Kluwer, Dordrecht.
171. Font, J.M. { Rodriquez, A.J. { Torres, A. (1984): Wajsberg algebras,
Stochastica 8, 5{31.
172. Fraassen, B. van (1966): Singular terms, truth-value gaps, and free logic, J.
of Philosophy 63, 481{495.
173. Fraassen, B. van (1968): Presupposition, implication, and self-reference, J.
of Philosophy 65, 136{152.
References
575
174. Fraassen, B. van (1969): Presuppositions, supervaluations, and free logic,

in: K. Lambert (ed.), The Logical Way of Doing Things. Yale University Press,
New Haven, 67{91.
175. Fraenkel, A.A. { Bar-Hillel, Y. { Levy, A. (1973): Foundations of Set
Theory. North-Holland Publ. Comp., Amsterdam.
176. Frank, H.J. (1979): On the simultaneous associativity of F (x; y) and x + y ;
F (x; y). Aequationes Math. 19, 194{226.
177. Fuchs, L. (1963): Partially Ordered Algebraical Systems. Pergamon Press,
New York.
178. Gabbay, D. { Guenthner, F. (eds.) (1983{87): Handbook of Philosophical
Logic. Vols. 1{4, Reidel, Dordrecht.
179. Gaines, B.R. (1978): Fuzzy and probability uncertainty logics, Information
and Control 38, 154{169.
180. Gaines, B.R. { Kohout, L.J. (1977): The fuzzy decade: a bibliography of
fuzzy systems and closely related topics, Internat. J. Man-Machine Studies 9,
1{68.
181. Garcia, O. { Moussavi, M. (1990): A six-valued logic for representing incomplete information, in: 20. Internat. Symp. Multiple-Valued Logic, Charlotte/NC,
1990, Proceedings, IEEE Computer Soc., New York/NY, 110{114.
182. Garmendia, L. { Trillas, E. { Salvador, A. (1999): On measuring -Tconditionality of fuzzy relations (Part II), in: European Soc. Fuzzy Logic and
Technol.: Eus at-Estylf Joint Conference, Palma (Mallorca), 1999, Proceedings, Univ. de les Illes Balears, 235{238.
183. Gentilhomme, Y. (1968): Les ensembles ous en linguistique, Cahiers de
Linguistique Theorique et Applique, V, 47, Bucharest.
184. Gentzen, G. (1934{35): Untersuchungen uber das logische Schlieen, Mathematische Zeitschrift 39, 176{210 and 405{431.
185. Gerla, G. (1994): An extension principle for fuzzy logics, Zeitschr. math.
186. Giles, R. (1976): Lukasiewicz logic and fuzzy set theory, Internat. J. ManMachine Studies 8, 313{327.
187. Giles, R. (1979): A formal system for fuzzy reasoning, Fuzzy Sets Systems 2,
233{257.
188. Giles, R. (1982): Semantics for fuzzy reasoning, Internat. J. Man-Machine
Studies 17, 401{415.
189. Giles, R. (1982): Foundations for a theory of possibility, in: M.M. Gupta/E.
Sanchez (eds.), Fuzzy Information and Decision Processes. North-Holland Publ.
Comp., Amsterdam, 183{195.
190. Giles, R. (1985): A resolution logic for fuzzy reasoning, in: 15. Internat. Symp. Multiple-Valued Logic, Kingston/Ontario, 1985, Proceedings, IEEE
Computer Soc., New York/NY, 60{67.
191. Ginsberg, M.L. (1987): Multi-valued logics, in: M.L. Ginsberg (ed.), Readings
in Nonmonotonic Reasoning. Morgan Kaufman, Los Altos/Ca., 251{255.
192. Ginsberg, M.L. (1988): Multivalued logic: a uniform approach to reasoning
in articial intelligence, Computational Intelligence 4, 265{315.
193. Ginzer, J.A. { Butler, J.T. (1979): Multiple-Valued Logic: 1974{1978 {
survey and analysis, in: 9th Internat. Symp. Multiple-Valued Logic, Bath, 1979,
194. Girard, J.Y. (1987): Linear logic, Theoret. Computer Sci. 50, 1{102.
195. de Glas, M. (1984): Representation of Lukasiewicz' many-valued algebras.
The atomic case, Fuzzy Sets Systems 14, 175{185.
576
References
196. Gluschankof, D. (1992). Prime deductive systems and injective objects in

the algebras of Lukasiewicz innite-valued calculi, Algebra Universalis 29, 354{
377.
197. Goddard, L. { Routley, R. (1973): The Logic of Signicance and Context.
Scottish Academic Press, Edinburgh.
198. Godel, K. (1931): U ber formal unentscheidbare Satze der Principia Mathematica und verwandter Systeme. I, Monatshefte f. Math. und Physik 38, 173{
198.
199. Godel, K. (1932): Zum intuitionistischen Aussagenkalkul, Anzeiger Akademie der Wissenschaften Wien, Math.-naturwiss. Klasse 69, 65{66; also: Ergebnisse eines mathematischen Kolloquiums 4 (1933), 40.
200. Godel, K. (1940): The Consistency of the Axiom of Choice and of the Generalized Continuum Hypothesis with the Axioms of Set Theory. University Press,
Princeton.
201. Goguen, J.A. (1967): L-fuzzy sets, J. Math. Anal. Appl. 18, 145{174.
202. Goguen, J.A. (1968{69): The logic of inexact concepts, Synthese 19, 325{
373.
203. Goldberg, H. { Leblanc, H. { Weaver, G. (1974): A strong completeness
theorem for 3-valued logic, Notre Dame J. Formal Logic 15, 325{330.
204. Goldblatt, R.I. (1974): Semantic analysis of orthologic, J. Philosophical
Logic 3, 19{35.
205. Goodman, I.R. { Nguyen, H.T. (1985): Uncertainty Models for KnowledgeBased Systems. North-Holland Publ. Comp., Amsterdam.
206. Gottwald, S. (1974): Mehrwertige Anordnungsrelationen in klassischen
Mengen, Mathematische Nachrichten 63, 205{212.
207. Gottwald, S. (1976): A cumulative system of fuzzy sets, in: W. Marek/M.
Srebrny/A. Zarach (eds.), Set Theory and Hierarchy Theory. A Memorial Tribute to A. Mostowski, Bierutowice 1975. Lecture Notes Math. 537, Springer,
Berlin, 109{119.
208. Gottwald, S. (1976{77): Untersuchungen zur Mehrwertigen Mengenlehre.
I{III. Mathematische Nachrichten 72, 297{303; 74, 329{336; 79, 207{217.
209. Gottwald, S. (1979): Set theory for fuzzy sets of higher level, Fuzzy Sets
Systems 2, 125{151.
210. Gottwald, S. (1980): Fuzzy uniqueness of fuzzy mappings. Fuzzy Sets Systems 3, 49{74.
211. Gottwald, S. (1981): Fuzzy Mengen und ihre Anwendungen. Ein U berblick,
Elektronische Informationsverarbeitung Kybernetik EIK 17, 207{235.
212. Gottwald, S. (1984): On the existence of solutions of systems of fuzzy equations, Fuzzy Sets Systems 12, 301{302.
213. Gottwald, S. (1984): Fuzzy set theory: some aspects of the early development, in: H.J. Skala/S. Termini/E. Trillas (eds.), Aspects of Vagueness. Theory
and Decision Library, vol. 39, Reidel, Dordrecht, 13{29.
214. Gottwald, S. (1985): A generalized Lukasiewicz-style identity logic, in: L.P.
de Alcantara (ed.), Mathematical Logic and Formal Systems. Lecture Notes
Pure and Applied Mathematics, vol. 94, Marcel Dekker, New York, 183{195.
215. Gottwald, S. (1986): Characterizations of the solvability of fuzzy equations,
Elektronische Informationsverarbeitung Kybernetik EIK 22, 67{91.
216. Gottwald, S. (1986): On some theoretical problems concerning the construction of fuzzy controllers, in: S.F. Bocklisch et al. (eds.), Fuzzy Sets Applications,
Methodological Approaches, and Results. Akademie-Verlag, Berlin, 45{55.
217. Gottwald, S. (1986): Fuzzy set theory with t-norms and '-operators, in: A.
di Nola/A.G.S. Ventre (eds.), The Mathematics of Fuzzy Systems. Interdisciplinary Systems Research, vol. 88, Verlag TU V Rheinland, Cologne, 143{195.
References
577
218. Gottwald, S. (1989): Mehrwertige Logik. Eine Einfuhrung in Theorie und

Anwendungen. Akademie-Verlag, Berlin.
219. Gottwald, S. (1993): Fuzzy Sets and Fuzzy Logic. The Foundations of Application { from a Mathematical Point of View. Vieweg, Braunschweig/Wiesbaden
and Teknea, Toulouse.
220. Gottwald, S. (1995): An approach to handle partially sound rules of inference, in: B. Bouchon-Meunier/R.R. Yager/L.A. Zadeh (eds.), Advances in
Intelligent Computing { IPMU '94, Selected Papers, Lecture Notes Computer
Sci., vol. 945, Springer, Berlin, 380{388.
221. Gottwald, S. (1999): Many-valued logic and fuzzy set theory, in: U. Hohle/
S.E. Rodabaugh (eds.) Mathematics of Fuzzy Sets. Logic, Topology, and Measure Theory. The Handbooks of Fuzzy Sets Series, Kluwer, Boston, 5{89.
222. Gottwald, S. { Pedrycz, W. (1985): Problems of the design of fuzzy controllers, in: M.M. Gupta et al. (eds.), Approximate Reasoning in Expert Systems.
North-Holland Publ. Comp., Amsterdam, 393{405.
223. Gottwald, S. { Pedrycz, W. (1986): Solvability of fuzzy relational equations and manipulation of fuzzy data, Fuzzy Sets Systems 18, 1{21.
224. Grale, W. (1976): Das U bergehen von Typen in der Lukasiewicz-Logik.
Ph.D.-Thesis, Albert-Ludwigs-Universitat, Freiburg/Br.
225. Gratzer, W. (1968): Universal Algebra. van Nostrand, Princeton.
226. Greechie, R.J. { Gudder, S.P. (1973): Quantum logics, in: C.A. Hooker
(ed.), Contemporary Research in the Foundations and Philosophy of Quantum
Theory. University of Western Ontario Series Philosophy of Science 2, Reidel,
Dordrecht, 143{173.
(1977): Algebraic analysis of Lukasiewicz { Tarski's n-valued
227. Grigolia, R.S.
logical systems, in: R. Wojcicki/G. Malinowski (eds.) Selected Papers on Lukasiewicz Sentential Calculi. Ossolineum, Wroclaw, 81{92.
{ Finn, V.K. (1979): Algebry Bocvara i sootvetstvujuscie
228. Grigolija, R.S.
im propozicional'nye iscislenija, in: Issledovanija po Neklassiceskim Logikam i
Teorii Mnozestv. Nauka, Moscow, 345{372.
229. Grishin, V.N. (1974): Ob odnoj nestandartnoj logike i ee primenenii k teorii
mnozestv, in: Issledovanija po Formalizovannym Jazykam i Neklassiceskim
Logikam. Nauka, Moscow, 135{171.
230. Grishin, V.N. (1976): Ob algebraiceskoj semantike logiki bez sokrascenij, in:
Issledovanija po Teorii Mnozestv i Neklassiceskim Logikam. Nauka, Moscow,
247{264.
231. Grishin, V.N. (1979): Teorema Erbrana dlja logiki bez sokrascenij, in: Issledovanija po Neklassiceskim Logikam i Teorii Mnozestv. Nauka, Moscow, 316{
329.
232. Guccione, S. { Tortora, R. { Vaccaro, V. (1981): Deduction theorems
in Lukasiewicz propositional calculi, Rendiconti Sem. Mat. Univers. Politecn.
Torino , 39, 53{65.
233. Guedj, D. (1985): Nicholas Bourbaki, collective mathematician. An interview
with Claude Chevalley, The Mathematical Intelligencer 7, no. 2, 18{22.
234. Gupta, A. { Belnap, N. (1993): The Revision Theory of Truth. MIT Press,
Cambridge/Mass.
235. Gupta, M.M. { Kandel, A. { Bandler, W. { Kiszka, J.B. (eds.) (1985):
Approximate Reasoning in Expert Systems. North-Holland Publ. Comp., Amsterdam.
236. Gurevic, J.S. { Kokorin, A.I. (1963): Universal'naja ekvivalentnost' uporjadocennych abelevych grupp, Algebra i Logika 2, 37{39.
237. Hahnle, R. (1993): Automated Deduction in Multiple-Valued Logics. Clarendon Press, Oxford.
578
References
238. Hahnle, R. (1994): Many-valued logic and mixed integer programming, Annals of Math. and Artif. Intelligence 12, 231{263.
239. Hahnle, R. (1997): Proof theory of many-valued logic { linear optimization
{ logic design: connections and interactions, Soft Computing 1, 107{119.
240. Hahnle, R. (1998): Commodious axiomatization of quantiers in multiplevalued logic, Studia Logica 61, 101{121.
241. Hahnle, R. (1999): Tableaux for many-valued logics, in: M. d'Agostino et al.
(eds.) Handbook of Tableau Methods. Kluwer, Dordrecht, 529{580.
242. Hahnle, R. (200x): Advanced many-valued logics, in: D.M. Gabbay (ed.)
Handbook of Philosophical Logic. 2nd ed., vol. 6: Alternatives to Classical Logic,
Kluwer, Dordrecht (to appear).
243. Hajek, P. (1995): Fuzzy logic and arithmetical hierarchy. Fuzzy Sets Systems
73, 359{363.
244. Hajek, P. (1997): Fuzzy logic and arithmetical hierarchy II. Studia Logica
58, 129{141.
245. Hajek, P. (1998): Basic fuzzy logic and BL-algebras, Soft Computing 2, 124{
128.
246. Hajek, P. (1998): Metamathematics of Fuzzy Logic . Trends in Logic, vol. 4,
Kluwer, Dordrecht.
247. Hajek, P. (200x): Fuzzy logic and arithmetical hierarchy III. Studia Logica
(to appear).
248. Hajek, P. { Godo, L. { Esteva, F. (1996): A complete many-valued logic
with product-conjunction. Arch. Math. Logic 35, 191{208.
249. Hajek, P. { Havranek, T. { Jirousek, R. (1992): Uncertain Information
Processing in Expert Systems. CRC Press, Boca Raton/FL.
250. Hajek, P. { Paris, J. { Shepherdson, J. (2000): The liar paradox and fuzzy
logic, J. Symbolic Logic 65, 339{346.
251. Hajek, P. { Paris, J. { Shepherdson, J. (2000): Rational Pavelka predicate
logic is a conservative extension of Lukasiewicz predicate logic, J. Symbolic Logic
65, 669{682.
252. Hajek, P. { Valdes, J.J. (1990): Algebraic foundations of uncertainty processing in rule-based expert systems. I, Comp. Articial Intelligence 9, 325{334.
253. Hallden, S. (1949): The Logic of Nonsense. Uppsala Univ., Uppsala.
254. Hallett, M. (1984): Cantorian Set Theory and Limitation of Size. Clarendon
Press, Oxford.
logische Aggregationen nicht-binar explizierter
255. Hamacher, H. (1978): Uber
Entscheidungskriterien. Rita G. Fischer Verlag, Frankfurt/Main.
256. Hay, L.S. (1963): Axiomatization of the innite-valued predicate calculus, J.
257. Heijenoort, J. van (ed.) (1967): From Frege to Godel. A Source Book in
Mathematical Logic, 1879{1931. Harvard Univ. Press, Cambridge/Mass.
258. Hempel, C.G. (1939): Vagueness and logic, Philosophy of Science 6, 163{180.
259. Henkin, L. (1949): The completeness of the rst-order functional calculus, J.
260. Hermes, H. (1967): Einfuhrung in die Verbandstheorie. Springer, Berlin.
261. Herzberger, H. (1973): Dimensions of truth, J. Philosophical Logic 2, 535{
556.
262. Heyting, A. (1930): Die formalen Regeln der intuitionistischen Logik,
Sitzungsberichte Preuische Akademie der Wissenschaften Berlin, Physikal.mathemat. Klasse II, 42{56.
263. Hintikka, K.J.J. (1955): Form and content in quantication theory, Acta
Phil. Fennica 8, 7{55.
References
579
264. Hisdal, E. (1986): Innite-valued logic based on two-valued logic and probability. Part 1.1: Diculties with present-day fuzzy set theory and their resolution
in the TEE model, Internat. J. Man-Machine Studies 25, 89{111.
265. Hisdal, E. (1986): Innite-valued logic based on two-valued logic and probability. Part 1.2: Dierent sources of fuzziness and uncertainty, Internat. J.
Man-Machine Studies 25, 113{138.
266. Hisdal, E. (1988): Are grades of membership probabilities?, Fuzzy Sets Systems 25, 325{348.
267. Hisdal, E. (1998): Logical Structures for Representation of Knowledge and
Uncertainty. Studies in Fuzziness and Soft Computing, Physica-Verlag, Heidelberg.
268. Hodes, H. (1986/87): Individual-actualism and three-valued modal logic. Part
I: Model theoretic semantics; Part II: Natural deduction formalization, J. Philosophical Logic 15, 369{401; 16, 17{63.
269. Hohle, U. (1992): M -valued sets and sheaves over integral commutative CLmonoids, in: S.E. Rodabaugh/E.P. Klement/U. Hohle (eds.), Applications of
Category Theory to Fuzzy Subsets. Kluwer, Dordrecht, 33{72.
270. Hohle, U. (1994): Monoidal logic, in: R. Kruse/J. Gebhardt/R. Palm (eds.),
Fuzzy Systems in Computer Science. Vieweg, Wiesbaden, 233{243.
271. Hohle, U. (1995): Commutative, residuated l-monoids. In: U. Hohle/E.P.
Klement (eds.), Non-Classical Logics and Their Applications to Fuzzy Subsets.
Kluwer, Dordrecht, 53{106.
272. Hohle, U. (1995): Presheaves over GL-monoids. In: U. Hohle/E.P. Klement
(eds.), Non-Classical Logics and Their Applications to Fuzzy Subsets. Kluwer,
Dordrecht, 127{157.
273. Hohle, U. (1996): On the fundamentals of fuzzy set theory; J. Math. Anal.
Appl. 201, 786{826.
274. Hohle, U. (1998): Many-valued equalities, singletons and fuzzy partitions,
Soft Computing 2, 134{140.
275. Hohle, U. (1999): Allgemeine Bemerkungen zu nichtklassischen Logiken.
In: R. Seising (ed.), Fuzzy Theorie und Stochastik. Vieweg: Braunschweig/Wiesbaden, 226{236.
276. Hohle, U. (2000): Classication of subsheaves over GL-Algebras. In: R.
Buss/P. Hajek/P. Pudlak (eds.) Logic Colloquium '98. Lecture Notes in Logic,
vol. 13, A.K.Peters, Natick/Ma., 238{261.
277. Holdsworth, D.G. { Hooker, C.A. (1981{82): A critical survey of quantum
logic, Scientia 116{117 (Supplementum), 127{246.
278. Holmblad, L.P. { Ostergaard, J.-J. (1982): Control of a cement kiln by
fuzzy logic, in: M.M. Gupta/E. Sanchez (eds.), Fuzzy Information and Decision
Processes. North-Holland Publ. Comp., Amsterdam, 389{399.
279. Horn, A. (1969): Logic with truth values in a linearly ordered Heyting algebra, J. Symbolic Logic 34, 395{408.
280. Horn, L.R. (1985): Metalinguistic negation and pragmatic ambiguity, Language 61, 121{174.
281. Horn, L.R. (1989): A Natural History of Negation. University of Chicago
Press, Chicago.
282. Hughes, G.E. { Cresswell, M.J. (1968): An Introduction to Modal Logic.
Methuen, London.
283. Jablonskij, S.V. (1952): O superpozicijach funkcij algebry logiki, Matematiceskij Sbornik 30 (72), 329{348.
284. Jablonskij, S.V. (1958): Funkcional'nye postroenija v k-znacnoj logike,
Trudy Matematiceskogo Instituta Akademii Nauk SSSR 124, 5{142.
580
References
285. Jaskowski, S. (1936): Recherches sur le systeme de la logique intuitioniste,

in: Actes du Congres Internationale de Philosophie Scientique 1936. vol. 6,
Paris, 58{61. [English translation: Studia Logica 34 (1975), 117{120.]
286. Jech, T.L. (1971): Lectures in Set Theory with Particular Emphasis on the
Method of Forcing. Lecture Notes in Mathematics, vol. 217, Springer, Berlin.
287. Jech, T.L. (1978): Set Theory. Academic Press, New York.
[2nd, corr. ed. Springer, Berlin, 1997.]
288. Kalmbach, G. (1974): Orthomodular logic, Zeitschr. math. Logik Grundl.
Math. 20, 395{406.
289. Kandel, A. { Lee, S.C. (1978): Fuzzy Switching and Automata: Theory and
Applications. Crane Russak, New York and E.Arnold, London.
290. Kandel, A. { Yager, R.R. (1979): A 1979 bibliography on fuzzy sets, their
applications and related topics, in: M.M. Gupta/R.K. Ragade/R R. Yager
(eds.), Advances in Fuzzy Set Theory and Applications. North-Holland Publ.
291. Kang Tai, { Chen Gouchen (1998): Modications of Bellman{Giertz's theorem. Fuzzy Sets Systems 94, 349{353.
292. Karpenko, A.S. (1997): Mnogoznacnye Logiki. Logika i Komp'juter, vol. 4,
Nauka, Moscow.
293. Kearns, J.T. (1974): Vagueness and failing sentences, Logique et Analyse, N.
S. 17, 301{315.
294. Kearns, J.T. (1979): The strong completeness of a system for Kleene's threevalued logic, Zeitschr. math. Logik Grundl. Math. 25, 61{68.
295. King, R.E. (1982): Fuzzy logic control of a cement kiln precalciner ash
furnace, in: Proceedings IEEE Conference Applications of Adaptive and Multivariable Control, Hull. Long Beach, 56{59.
296. Kirin, V. (1966): Gentzen's method for the many-valued propositional calculi,
Zeitschr. math. Logik Grundl. Math. 12, 317{332.
297. Kirin, V. (1968): Post algebras as semantic bases of some many-valued logics,
Fundamenta Mathematicae 63, 279{294.
298. Klaua, D. (1964): Allgemeine Mengenlehre. Ein Fundament der Mathematik.
Akademie-Verlag, Berlin.
299. Klaua, D. (1966): U ber einen zweiten Ansatz zur mehrwertigen Mengenlehre,
Monatsberichte Deutsche Akademie der Wissenschaften Berlin 8, 161{177.
300. Klaua, D. (1966): Grundbegrie einer mehrwertigen Mengenlehre, Monatsberichte Deutsche Akademie der Wissenschaften Berlin 8, 781{802.
301. Klaua, D. (1968): Partiell denierte Mengen, Monatsberichte Deutsche
Akademie der Wissenschaften Berlin 10, 571{578.
302. Klaua, D. (1969): Partielle Mengen mit mehrwertigen Grundbeziehungen,
Monatsberichte Deutsche Akademie der Wissenschaften Berlin 11, 573{584.
303. Klaua, D. (1970): Stetige Gleichmachtigkeiten kontinuierlich-wertiger Mengen, Monatsberichte Deutsche Akademie der Wissenschaften Berlin 12, 749{
758.
304. Kleene, S.C. (1938): On notation for ordinal numbers, J. Symbolic Logic 3,
150{155.
305. Kleene, S.C. (1952): Introduction to Metamathematics. North-Holland Publ.
Comp., Amsterdam and van Nostrand, New York.
306. Klement, E.P. (1997). Some mathematical aspects of fuzzy sets: triangular
norms, fuzzy logics, and generalized measures. Fuzzy Sets Systems 90, 133{140.
307. Klement, E.P. { Mesiar, R. { Pap, E. (1997): A characterization of the
ordering of continuous t-norms, Fuzzy Sets Systems 86, 189{195.
References
581
308. Klement, E.P. { Mesiar, R. { Pap, E. (1999): Bausteine der Fuzzy Logik:
t-Normen { Eigenschaften und Darstellungssatze, in: R. Seising (ed.), Fuzzy
Theorie und Stochastik. Vieweg, Braunschweig/Wiesbaden, 205{225.
309. Klement, E.P. { Mesiar, R. { Pap, E. (2000): Triangular Norms. Kluwer,
Dordrecht.
310. Kneale, W. { Kneale, M. (1962): The Development of Logic. Clarendon
Press, Oxford.
311. Korner, S. (1966): Experience and Theory. Routledge & Kegan Paul, London.
312. Kolesarova, A. (2000): A note on Archimedean triangular norms, BUSEFAL
80, 57{60.
313. Kolmogoroff, A. (1932): Zur Deutung der intuitionistischen Logik, Mathematische Zeitschrift 35, 58{65.
314. Konikowska, B. { Morgan, C.G. { Orlowska, E. (1998): A relational
formalisation of arbitrary nite valued logics, Logic J. of the IGPL 6, 755{774.
315. Kreiser, L. { Gottwald, S. { Stelzner, W. (eds.) (1988): Nichtklassische
Logik. Eine Einfuhrung. Akademie-Verlag, Berlin.
316. Kreschnak, H. (1985): Computergestutzte Analysen von Schulerleistungen.
Ein Beitrag zur Logik und Methodologie der Diagnostik. Volk und Wissen Verlag, Berlin.
317. Kripke, S.A. (1975): Outline of a theory of truth, J. of Philosophy 72, 690{
716.
318. Krzystek, P.S. { Zachorowski, S. (1977): Lukasiewicz logics have not the
interpolation property, Reports Mathematical Logic 9, 39{40.
319. Kubin, W. (1979): Eine Axiomatisierung der mehrwertigen Logiken von
Godel, Zeitschr. math. Logik Grundl. Math. 25, 549{558.
320. Kunen, K. (1980): Set Theory. An Introduction to Independence Proofs.
North-Holland Publ. Comp., Amsterdam.
321. Kuratowski, K. { Mostowski, A. (1976): Set Theory. With an Introduction
to Descriptive Set Theory. North-Holland Publ. Comp., Amsterdam and PWN,
Warsaw.
322. Kuros, A.G. (1956): Theory of Groups. Chelsea, New York.
323. Kutschera, F. von (1967): Elementare Logik. Springer, Wien.
324. Kutschera, F. von (1975): Partial interpretations, in: E.L. Keenan (ed.),
Formal semantics of Natural Language. Cambridge University Press, Cambridge, 156{174.
325. Lakoff, G. (1973): Hedges: a study in meaning criteria and the logic of fuzzy
concepts, J. Philosophical Logic 2, 458{508.
[Reprinted: D. Hockney/W. Harper/B. Freed (eds.), Contemporary Research in
Philosophical Logic and Linguistic Semantics. Reidel, Dordrecht 1975, 221{271.]
326. Langholm, T. (1988): Partiality, Truth and Persistence. CSLI Lecture Notes,
vol. 15, CSLI, Stanford/Ca.
327. Lano, K. (1992): Fuzzy sets and residuated logic, Fuzzy Sets Systems 47,
203{220.
328. Larsen, R.M. (1980): Industrial applications of fuzzy logic control, Internat.
J. Man-Machine Studies 12, 3{10.
329. Leblanc, H. (1977): A strong completeness theorem for 3-valued logic: part
II, Notre Dame J. Formal Logic 18, 107{116.
330. Lewis, C.J. { Langford, C.H. (1932): Symbolic Logic. Century Comp., New
York. [2nd ed., Dover, New York, 1959.]
331. Ling, C.H. (1965): Representation of associative functions, Publ. Math. Debrecen 12, 189{212.
582
References
332. Lovett, E.O. (1900{01): Mathematics at the International Congress of Philosophy, Paris 1900, Bulletin American Mathematical Society 7, 157{183.
333. Lukasiewicz, J. (1913): Die logischen Grundlagen der Wahrscheinlichkeitsrechnung. Krakow. [English translation in [338].]
334. Lukasiewicz, J. (1920): O logice trojwartosciowej, Ruch Filozocny 5, 170{
171. [English translation in [338].]
335. Lukasiewicz, J. (1930): Philosophische Bemerkungen zu mehrwertigen Systemen des Aussagenkalkuls, Comptes Rendus Seances Societe des Sciences et
Lettres Varsovie, cl. III, 23, 51{77.
336. Lukasiewicz, J. (1935): Zur Geschichte der Aussagenlogik, Erkenntnis 5,
111{131.
337. Lukasiewicz, J. (1953): A system of modal logic, J. Computing Systems 1,
111{149.
338. Lukasiewicz, J. (1970): Selected Works. (ed.: L. Borkowski) North-Holland
Publ. Comp., Amsterdam and PWN, Warsaw.
339. Lukasiewicz, J. { Tarski A. (1930): Untersuchungen uber den Aussagenkalkul, Comptes Rendus Seances Societe des Sciences et Lettres Varsovie,
cl. III, 23, 30{50.
340. MacNeille, H. (1937): Partially ordered sets, Transactions American Mathematical Society 42, 416{460.
341. MacVicar-Whelan, P.J. (1978): Fuzzy sets, concept of height, and hedge
very, IEEE Transactions Systems, Man and Cybernetics SMC 8, 507{512.
342. Maksimova, L. { Vakarelov, +D. (1974): Representation theorems for generalized Post algebras of order ! , Bull. Acad. Polon. Sci., Ser. Sci. Math.,
Astronom., Phys., 22, 757{764.
343. Maksimova, L. { Vakarelov, D. (1974): Semantics for !+-valued predicate
calculi, Bull. Acad. Polon. Sci., Ser. Sci. Math., Astronom., Phys., 22, 765{771.
344. Mal'cev, A. I. (1973): Algebraic Systems. Springer, Berlin, and AkademieVerlag, Berlin.
345. Mal'cev, A. I. (1976): Iterativnye algebry Posta. Novosibirsk.
346. Malinowski, G. (1977): Classical characterization of n-valued Lukasiewicz
calculi, Reports Mathematical Logic 9, 41{45.
347. Malinowski, G. (1993): Many-Valued Logics. Clarendon Press, Oxford.
348. Mamdani, E.H. { Gaines, B.R. (eds.) (1981): Fuzzy Reasoning and Its Applications. Academic Press, New York.
349. Mangani, P. (1973): Su certe algebre connesse con logiche a piu valori,
Boll. Unione Math. Italiana, ser. 8, 4, 68{78.
350. Martin, J.N. (1975): A many-valued semantics for category mistakes, Synthese 31, 63{83.
351. Martin, J.N. (1977): An axiomatization of Herzberger's 2-dimensional presuppositional semantics, Notre Dame J. Formal Logic 18, 378{382.
352. Martin, R.L. { Woodruff, P.W. (1975): On representing `True-in-L' in L,
Philosophia 5, 213{217.
353. Maydole, R.E. (1975): Paradoxes and many-valued set theory, J. Philosophical Logic 4, 269{291.
354. McCawley, J.D. (1981): Everything that Linguists have Always Wanted to
Know about Logic. University of Chicago Press, Chicago [2nd ed. 1993].
355. McColl, H. (1897): Symbolic reasoning. II, Mind, N.S., 6, 493{510.
356. McGee, V. (1991): Truth, Vagueness, and Paradox. Hackett Publ. Comp.,
Indianapolis/Ind.
357. McNaughton, R. (1951): A theorem about innite-valued sentential logic,
J. Symbolic Logic 16, 1{13.
References
583
358. Medvedev, F.A. (1965): Razvitie Teorii Mnozestv v XIX Veke. Nauka, Moscow.
359. Menger, K. (1942): Statistical metrics, Proc. Nat. Acad. Sci. USA 8, 535{
537.
360. Menger, K. (1979): Selected Papers in Logic and Foundations, Didactics,
Economics. Reidel, Dordrecht.
361. Meredith, C.A. (1958): The dependence of an axiom of Lukasiewicz, Transactions American Mathematical Society 87, 54.
362. Meschkowski, H. (1967): Probleme des Unendlichen. Werk und Leben Georg
Cantors. Vieweg, Braunschweig.
363. Mesiar, R. { Thiele, H. (2000): On T-quantiers and S-quantiers, in: V.
Novak/I. Perlieva (eds.) Discovering the World with Fuzzy Logic. Studies in
Fuzziness and Soft Computation, vol. 57, Physica-Verlag, Heidelberg (in print).
364. Michalski, K. (1937): Le probleme de la volonte a Oxford et a Paris au XIVe
siecle, Studia Philosophica 2, 233{365.
365. Michalski, R.S. (1977): Variable-valued logic and its applications to pattern
recognition and machine learning, in: J.M. Dunn/G. Epstein (eds.), Modern
Uses of Multiple-Valued Logic. Reidel, Dordrecht, 506{534.
366. Miyakoshi, M. { Shimbo, M. (1985): Solutions of composite fuzzy relational
equations with triangular norms, Fuzzy Sets Systems 16, 53{63.
367. Mizumoto, M. (1982): Fuzzy inference using max-^: -composition in the compositional rule of inference, in: M.M. Gupta/E. Sanchez (eds.), Approximate
Reasoning in Decision Analysis. North-Holland Publ. Comp., Amsterdam, 67{
76.
368. Mizumoto, M. { Tanaka, K. (1979): Some properties of fuzzy numbers, in:
M.M. Gupta/R.K. Ragade/R.R. Yager (eds.), Advances in Fuzzy Set Theory
and Applications. North-Holland Publ. Comp., Amsterdam, 153{164.
369. Mizumoto, M. { Zimmermann, H.J. (1982): Comparison of fuzzy reasoning
methods, Fuzzy Sets Systems 8, 253{283.
370. Moh Shaw-Kwei (1954): Logical paradoxes for many-valued systems, J. Symbolic Logic 19, 37{40.
371. Moisil, G.C. (1940): Recherches sur les logiques non chrysipiennes, Annales
Scientiques de l'Universite de Jassy 26, 431{466.
372. Moisil, G.C. (1941): Notes sur les logiques non-chrysipiennes, Annales Scientiques de l'Universite de Jassy 27, 86{98.
373. Moisil, G.C. (1972): Essais sur les Logiques Non Chrysipiennes. Acad. Rep.
Soc. Roumanie, Bucharest.
374. Montagna, F. (200x): Three complexity problems in quantied fuzzy logic,
Studia Logica (to appear).
375. Moore, G.H. (1982): Zermelo's Axiom of Choice. Its Origins, Development,
and In uence. Springer, New York.
376. Moore, R.E. (1966): Interval Analysis. Prentice Hall, Englewood Clis.
377. Moore, R.E. (1979): Methods and Applications of Interval Analysis. SIAM
Studies Applied Mathematics 2, Society Industrial and Applied Mathematics,
Philadelphia.
378. Morgan, C.G. (1974): A theory of equality for a class of many-valued predicate calculi, Zeitschr. math. Logik Grundl. Math. 20, 427{432.
379. Morgan, C.G. (1975): Similarity as a theory of graded equality for a class of
many-valued predicate calculi, in: 5th Internat. Symp. Multiple-Valued Logic,
Bloomington/Ind., 1975, Proceedings, IEEE Computer Soc., Long Beach, 436{
449.
380. Morgan, C.G. (1979): Local and global operators and many-valued modal
logics, Notre Dame J. Formal Logic 20, 401{411.
584
References
381. Morikawa, O. (1989): Some modal logics based on a three-valued logic, Notre
Dame J. Formal Logic 30, 130{137.
382. Morikawa, O. (1994): A sequential formulation of a logic based on fuzzy
modalities. Fuzzy Sets Systems 63, 181{185.
383. Mostert, P.S. { Shields, A.L. (1957): On the structure of semigroups on
a compact manifold with boundary, Annals of Mathematics 65, 117{143.
384. Mostowski, A. (1961): An example of a non-axiomatizable many-valued
logic, Zeitschr. math. Logik Grundl. Math. 7, 72{76.
385. Mostowski, A. (1961{62): Axiomatizability of some many valued predicate
calculi, Fundamenta Mathematicae 50, 165{190.
386. Muller, G.H. { Lenski, W. { Rautenberg, W. (eds.) (1987): OmegaBibliography of Mathematical Logic. vol. II: Non-Classical Logics. Springer,
Berlin.
387. Mundici, D. (1986): Interpretation of AF C*-algebras in Lukasiewicz sentential calculus, J. Functional Analysis 65, 15{63.
388. Mundici, D. (1988): The derivative of truth in Lukasiewicz sentential calculus, in: W.A. Carnielli/L.P. de Alcantara (eds.) Methods and Applications of
Mathematical Logic. Proc. VII Latin American Symp. Mathematical Logic 1985,
Contemporary Mathematics, vol. 69, American Math. Society, Providence/RI,
209{227.
389. Mundici, D. (1994): A constructive proof of McNaughton's theorem in
innite-valued logic, J. Symbolic Logic 59, 596{602.
390. Muskens, R. (1989): Meaning and Partiality. PhD Thesis, University of Amsterdam.
391. Muzio, J.C. { Wesselkamper, T.C. (1986): Multiple-Valued Switching Theory. Adam Hilger, Boston/Ma.
392. Nakamura, A. (1992): On a logic based on fuzzy modalities, in: 22nd Internat. Symp. Multiple-Valued Logic, Sendai, 1992, Proceedings, IEEE Computer
Soc., Piscataway/NJ, 460{466.
393. Nakamura, A. (1993): On a multi-modal logic based on the graded classications, Found. Comput. Decision Sci. 18, 275{291.
394. Nakamura, A. { Gao, Jian-Ming (1992): On a KTB-modal fuzzy logic,
Fuzzy Sets Systems 45, 327{334.
395. Nguyen, H. { Walker, E. (1997): A First Course in Fuzzy Logic. CRC
Press, Boca Raton.
396. Novak, V. (1989): Fuzzy Sets and Their Applications. Adam Hilger, Bristol.
397. Novak, V. (1990): On the syntactico-semantical completeness of rst-order
fuzzy logic. I{II. Kybernetica 26, 47{66; 134{154.
398. Novak, V. (1992): The Alternative Mathematical Model of Linguistic Semantics and Pragmatics. Plenum Press, New York.
399. Novak, V. { Perfilieva, I. { Mockor, J. (1999): Mathematical Principles
of Fuzzy Logic. Kluwer, Boston.
400. Novikov, P. S. (1977): Konstruktivnaja Matematiceskaja Logika s Tocki Zrenija Klassiceskoj. Nauka, Moscow.
401. Ockham, William of (1969): Predestination, God's Foreknowledge, and Future Contingents. Translated with an Introduction, Notes, and Appendices by
M. McCord Adams, N. Kretzmann, Appleton-Century-Crofts, New York [2nd
ed., Hackett Publ. Comp., Indianapolis, 1983].
402. Oh, C.-K. { Dinneen, D.A.(eds.) (1979): Syntax and Semantics. Vol. 11:
Presupposition. Academic Press, New York.
403. Ohya, T. (1968): Many valued logics extended to simple type theory, Science
Reports Tokyo Kyoiku Daigaku , 9, 260{270.
References
585
404. Orlowska, E. (1991): Post relation algebras and their proof systems, in:
21st Internat. Symp. Multiple-Valued Logic, Victoria/B.C., 1991, Proceedings,
IEEE Computer Soc., New York, 298{305.
405. Orlowska, E. (1997): Many-valuedness and uncertainty, in: 27th Internat. Symp. Multiple-Valued Logic, Antogonish/Nova Scotia, 1997, Proceedings,
IEEE Computer Soc., Piscataway/NJ, 153{160.
406. Ostermann, P. (1988): Many-valued modal propositional calculi, Zeitschr.
math. Logik Grundl. Math. 34, 343{354.
407. Ostermann, P. (1988): Many-valued modal logics: uses and predicate calculus, Zeitschr. math. Logik Grundl. Math. 36, 367{376.
408. Ovchinnikov, S.V. (1981): Structure of fuzzy binary relations, Fuzzy Sets
Systems 6, 169{195.
409. Ovchinnikov, S.V. { Roubens, M. (1991): On strict preference relations.
410. Paalman-deMiranda, A.B. (1964): Topological Semigroups. Mathematisch
Centrum, Amsterdam.
411. Panti, G. (1998): Multi-valued logics, in: D. Gabbay/P. Smets (eds.) Handbook of Defeasible Reasoning and Uncertainty Management Systems. vol. 1: P.
Smets (ed.) Quantied Representation of Uncertainty and Imprecision. Kluwer,
Dordrecht, 25{74.
412. Paris, J.B. (1997): A semantics for fuzzy logic, Soft Computing 1, 143{147.
413. Patzig, G. (1973): Aristotle, Lukasiewicz and the origins of many-valued
logic, in: P. Suppes et al. (eds.), Logic, Methodology and Philosophy of Science
IV. PWN, Warsaw and North-Holland Publ. Comp., Amsterdam, 921{929.
414. Pavelka, J. (1979): On fuzzy logic. I{III, Zeitschr. math. Logik Grundl. Math.
25, 45{52; 119{134; 447{464.
415. Peckhaus, V. (1990): Hilbertprogramm und Kritische Philosophie. Vandenhoeck & Ruprecht, Gottingen.
416. Pedrycz, W. (1982): Fuzzy control and fuzzy systems. Dept. Math., Delft
Univ. of Technology, Report 82 14.
417. Pedrycz, W. (1983): Sterowanie i systemy rozmyte. Zeszyty Naukowe Politechniki Slaskiej, Ser. Automatyka, no. 70, Gliwice.
418. Peirce, Ch.S. (1931{58): Collected Papers (eds.: C. Hartshorne et al.), Harvard Univ. Press, Cambridge/Mass.
419. Peters, S. (1977): A truth-conditional formulation of Karttunen's account
of presuppositions, Texas Linguistic Forum 6, 137{149.
420. Pirog-Rzepecka, K. (1966): Rachunek zdan w ktorym wyrazenia trace sens,
421. Pirog-Rzepecka, K. (1973): A predicate calculus with formulas which loose
sense and the corresponding propositional calculus, Bulletin Section of Logic 2,
22{29.
422. Pirog-Rzepecka, K. (1977): Systemy nonsense-logics. PWN, Warsaw-Wroclaw.
423. Pogorzelski, W.A. (1964): The deduction theorem for Lukasiewicz manyvalued propositional calculi, Studia Logica 15, 7{23.
424. Poschel, R. { Kaluznin, L.A. (1979): Funktionen- und Relationenalgebren.
Deutscher Verlag der Wissenschaften, Berlin and Birkhauser, Basel.
425. Post, E. L. (1920): Determination of all closed systems of truth tables, Bulletin American Mathematical Society 26, 437.
426. Post, E. L. (1921): Introduction to a general theory of elementary propositions, American J. Mathematics 43, 163{185.
427. Pottinger, G. (1983): Uniform, cut-free formulations of T, S4 and S5, J.
586
References
428. Prade, H. (1982): Modeles mathematiques de l'imprecis et de l'incertain

en vue d'application au raisonnement naturel, These d'Etat, Universite Paul
Sabatier, Toulouse.
429. Prior, A.N. (1953): Three-valued logic and future contingents. Philosophical
Quarterly 3, 317{326.
430. Prior, A.N. (1957): Time and Modality. Clarendon Press and Univ. Press,
Oxford.
431. Priest, G. { Sylvan, R. (1992): Simplied semantics for basic relevant logics,
J. Philosophical Logic 21, 217{232.
432. Purkert, W. { Ilgauds, H.J. (1987): Georg Cantor. Vita Mathematica,
vol. 1, Birkhauser, Basel.
433. Pynko, A.P. (1999): Functional completeness and axiomatizability within
Belnap's four-valued logic and its expansions, J. Applied Non-Classical Logics
9, 61{106.
434. Ragaz, M. (1981): Arithmetische Klassikation von Formelmengen der unendlichwertigen Logik, PhD-Thesis, ETH Zurich.
435. Ragaz, M. (1983): Die Unentscheidbarkeit der einstelligen unendlichwertigen
Pradikatenlogik, Archiv math. Logik Grundlagenforschung 23, 129{139.
436. Ragaz, M. (1983a): Die Nichtaxiomatisierbarkeit der unendlichwertigen Mengenlehre, Archiv math. Logik Grundlagenforschung 23, 141{146.
437. Raju, P.T. (1954): The principle of four-cornered negation in Indian philosophy, Review of Metaphysics 7, 694{713.
438. Rasiowa, H. (1973): On generalised Post algebras of order !+ and !+-valued
predicate calculi, Bull. Acad. Polon. Sci., Ser. Sci. Math., Astronom., Phys., 21,
209{219.
439. Rasiowa, H. (1974): An Algebraic Approach to Non-Classical Logics. PWN,
Warsaw and North-Holland Publ. Comp., Amsterdam.
440. Rasiowa, H. (1986): Rough concepts and !+-valued logic, in: 16th Internat.
Symp. Multiple-Valued Logic, Blacksburg/Va., 1986, Proceedings, IEEE Computer Soc., New York, 282{288.
441. Rasiowa, H. (1992): Toward fuzzy logic, in: L.A. Zadeh/J. Kacprzyk (eds.),
Fuzzy Logic for the Management of Uncertainty. Wiley, New York, 5{25.
442. Rasiowa, H. (1994): Axiomatization and completeness of uncountably valued
approximation logic, Studia Logica 53, 137{160.
443. Rasiowa, H. { Cat Ho, N. (1992): LT -fuzzy logic, in: L.A. Zadeh/J. Kacprzyk (eds.), Fuzzy Logic for the Management of Uncertainty. Wiley, New York,
121{139.
444. Rasiowa, H. { Sikorski, R. (1963): The Mathematics of Metamathematics.
PWN, Warsaw.
445. Reichenbach, H. (1932): Wahrscheinlichkeitslogik, Sitzungsberichte Preussische Akademie der Wissenschaften Berlin, Phys.-math. Klasse, 476{488.
446. Reichenbach, H. (1949): Philosophische Grundlagen der Quantenmechanik.
Birkhauser, Basel.
447. Reischer, C. { Simovici, D. (1987): A functional characterization of post
algebras, in: 17th Internat. Symp. Multiple-Valued Logic, Boston/Mass., 1987,
448. Rescher, N. (1969): Many-Valued Logic. McGraw Hill, New York.
449. Rescher, N. { Urquhart, A. (1971): Temporal Logic. Springer, Wien.
450. Restall, G. (1995): Four-valued semantics for relevant logics (and some of
their rivals), J. Philosophical Logic 24, 139{160.
451. Rine, D.C. (ed.) (1977): Computer Science and Multiple Valued Logic. NorthHolland Publ. Comp., Amsterdam [2nd rev. ed. 1984].
References
587
452. Rodabaugh, S.E. (1982): Fuzzy addition in the L-fuzzy real line, Fuzzy Sets
Systems 8, 39{51.
453. Rodabaugh, S.E. (1985): Complete fuzzy topological hyperelds and fuzzy
multiplication in the fuzzy real lines, Fuzzy Sets Systems 15, 285{310.
454. Rolf, B. (1981): Topics on Vagueness, PhD-Thesis, University of Lund.
455. Rose, A. (1950): Completeness of Lukasiewicz-Tarski propositional calculi,
Mathematische Annalen 122, 296{298.
456. Rose, A. (1952): Eight-valued geometry, Proc. London Math. Soc., Ser. 3, 2,
30{44.
457. Rose, A. { Rosser, J.B. (1958): Fragments of many-valued statement calculi,
Transactions American Mathematical Society 87, 1{53.
458. Rose, G.F. (1953): Propositional calculus and realizability, Transactions
American Mathematical Society 75, 1{19.
459. Rosenberg, I. (1970): U ber die funktionale Vollstandigkeit in den mehrwer
tigen Logiken, Rozpravy Ceskoslovensk
e Akademie Ved, R ada Matematickych
a Prirodnich Ved 80, no. 4, 3{93.
460. Rosenberg, I. (1973): The number of maximal closed classes in the set of
functions over a nite domain, J. Combinatorial Theory, Ser. A, 14, 1{7.
461. Rosenberg, I. (1977): Completeness properties of multiple-valued logic algebras, in: D.C. Rine (ed.), Computer Science and Multiple Valued Logic. NorthHolland Publ. Comp., Amsterdam, 144{186.
462. Rosenbloom, P.C. (1942): Post algebras. I. Postulates and general theory,
American J. Mathematics 64, 167{188.
463. Rosser, J.B. (1960): Axiomatization of innite valued logics, Logique et Analyse, N.S., 3, 137{153.
464. Rosser, J.B. (1969): Simplied Independence Proofs. Boolean Valued Models
of Set Theory. Academic Press, New York.
465. Rosser, J.B. { Turquette, A.R. (1952): Many-Valued Logics. NorthHolland Publ. Comp., Amsterdam.
466. Rousseau, G. (1967{70): Sequents in many-valued logic. I{II, Fundamenta
Mathematicae 60, 23{33; 67, 125{131.
467. Rousseau, G. (1970): Post algebras and pseudo-Post algebras, Fundamenta
Mathematicae 67, 133{145.
468. Ruitenberg, G. (1993): Basic logic and Fregean set theory, in: H. Barendregt/M. Bezem/J.W. Klop (eds.), Dirk van Dalen Festschrift , Questiones Innitae, vol. 5, Utrecht Univ., Utrecht, 121{142.
469. Ruitenberg, G. (1998): Basic predicate calculus, Notre Dame J. Formal
Logic 39, 18{46.
470. Russell, B. (1903): The Principles of Mathematics. I, Univ. Press, Cambridge.
471. Russell, B. (1923): Vagueness, Australasian J. Philosophy 1, 84{92.
472. Rutledge, J.D. (1959): A preliminary investigation of the innitely manyvalued predicate calculus, PhD-Thesis, Cornell University, Ithaca.
473. Rutledge, J.D. (1960): On the denition of an innitely-many-valued predicate calculus, J. Symbolic Logic 25, 212{216.
474. Salomaa, A. (1959): On many-valued systems of logic, Ajatus 22, 115{159.
475. Saloni, Z. (1972): Gentzen rules for the m-valued logic, Bull. Acad. Polon.
Sci., Ser. Sci. mathematiques, astronomiques et physiques, 20, 819{826.
476. Salzer, G. (1996): MUltlog: an expert system for multiple-valued logics, Collegium Logicum 2, 50{55.
477. Sanchez, E. (1976): Resolution of composite fuzzy relation equations. Information and Control 30, 38{48.
588
References
478. Sanchez, E. (1977): Solutions in composite fuzzy relation equations: application to medical diagnosis in Brouwerian logic, in: M.M. Gupta/G.N.
Saridis/B.R. Gaines (eds.), Fuzzy Automata and Decision Processes, NorthHolland Publ. Comp., Amsterdam, 221{234.
479. Sanchez, E. (1984): Solution of fuzzy equations with extended operations,
480. Scarpellini, B. (1962): Die Nichtaxiomatisierbarkeit des unendlichwertigen
Pradikatenkalkuls von Lukasiewicz, J. Symbolic Logic 27, 159{170.
481. Schmidt, H.A. (1960): Mathematische Gesetze der Logik. I, Vorlesungen uber
Aussagenlogik. Springer, Berlin.
482. Schmucker, K.J. (1984): Fuzzy Sets, Natural Language Computations, and
Risk Analysis. Computer Science Press, Rockville/Md.
483. Schotch, P.K. { Jensen, J.B. { Larsen, P.F. (1977): A strong completeness theorem for Lukasiewicz nitely many-valued logics, in: 7th Internat. Symp.
Multiple-Valued Logic, Charlotte/NC, 1977, Proceedings, IEEE Computer Soc.,
Long Beach/Ca., 89{91.
484. Schotch, P.K. { Jensen, J.B. { Larsen, P.F. { MacLellan, E.J. (1978):
A note on three-valued modal logic, Notre Dame J. Formal Logic 19, 63{68.
485. Schroter, K. (1955): Methoden zur Axiomatisierung beliebiger Aussagenund Pradikatenkalkule, Zeitschr. math. Logik Grundl. Math. 1, 241{251.
486. Schroter, K. (1955{58): Theorie des logischen Schlieens. I{II, Zeitschr.
math. Logik Grundl. Math. 1, 37{86; 4, 10{65.
487. Schweizer, B. { Sklar, A. (1958): Espaces metriques aleatoires, Comptes
Rendus Acad. Sci. Paris, Ser. A, 247, 2092{2094.
488. Schweizer, B. { Sklar, A. (1960): Statistical metric spaces, Pacic J. Mathematics 10, 313{334.
489. Schweizer, B. { Sklar, A. (1961): Associative functions and statistical
triangle inequalities, Publicationes Mathematicae Debrecen 8, 169{186.
490. Schweizer, B. { Sklar, A. (1983). Probabilistic Metric Spaces. NorthHolland Publ. Comp., Amsterdam.
491. Scott, D.S. (1967): Lectures on Boolean-valued models for set theory. (Unpublished mimeographed typescript of lectures given at the Symposium on Axiomatic Set Theory of the American Math. Society in Berkeley.)
492. Scott, D.S. (1973): Background to formalization. I, in: H. Leblanc (ed.),
Truth, Syntax and Modality. North-Holland Publ. Comp., Amsterdam. 244{
273.
493. Scott, D.S. (1974): Completeness and axiomatizability in many-valued logic,
in: L. Henkin et al. (eds.), Proceedings Tarski Symposium. Proceedings Symposia
Pure Mathematics 25, American Math. Society, Providence, 411{435.
494. Scott, D.S. (1976): Does many-valued logic have any use?, in: S. Korner
(ed.), Philosophy of Logic. University of California Press, Berkeley, 64{74.
495. Scott, D.S. (1979): Identity and existence in intuitionistic logic, in: M.P.
Fourman/C.J. Mulvey/D.S. Scott (eds.), Applications of Sheaves , Lecture Notes
Math., vol. 753, Springer, New York, 660{696.
496. Segerberg, K. (1965): A contribution to nonsense-logics, Theoria 31, 199{
217.
497. Segerberg, K. (1965): Some modal logics based on a three-valued logic,
Theoria 33, 53{71.

498. Sestakov,
V.I. (1964): O vzaimootnosenii nekotorych trechznacnych logikach
iscislenij, Uspechi Matematiceskich Nauk 19, 177{181.1
499. Sette, A.M. (1973): On the propositional calculus P , Mathematica Japonica
18, 173{180.
References
589
500. Sette, A.M. { Carnielli, W.A. (1995): Maximal weakly-intuitionistic logics, Studia Logica 55, 181{203.
501. Seuren, P.A.M. (1984): Logic and truth-values in language, in: F. Landman/F. Veltman (eds.), Varieties of Formal Semantics., Foris Publ., Dordrecht,
343{363.
502. Seuren, P.A.M. (1985): Discourse Semantics. Blackwell, New York.
503. Seuren, P.A.M. (1990): Burton-Roberts on presupposition and negation, J.
of Linguistics 26, 425{453.
504. Shimoda, M. (1981): Categorical aspects of Heyting-valued models for intuitionistic set theory, Comment. Math. Univ. Sancti Pauli 30, 17{35.
505. Sierpinski, W. (1945): Sur les fonctions de plusieurs variables, Fundamenta
Mathematicae 33, 169{173.
506. Simons, P. (1989): Lukasiewicz, Meinong, and many-valued logic, in: K. Szaniawski (ed.), The Vienne Circle and the Lvov-Warsaw School. Kluwer, Dordrecht, 249{292.
507. Simons, P. (1999): MacColl and many-valued logic: an exclusive conjunction,
Nordic J. Philosophical Logic 3, 85{90.
508. Skolem, Th. (1957): Bemerkungen zum Komprehensionsaxiom, Zeitschr.
math. Logik Grundl. Math. 3, 1{17.
509. Slupecki, J. (1936): Der volle dreiwertige Aussagenkalkul, Comptes Rendus
Seances Societe des Sciences et Lettres Varsovie, cl. III, 29, 9{11.
510. Smets, P. { Magrez, P. (1987): Implication in fuzzy logic, Internat. J.
Approx. Reasoning 1, 327{347.
511. Smullyan, R.M. (1968): First Order Logic. Springer, Berlin.
512. Smullyan, R.M. (1992): Godel's Incompleteness Theorems. University Press,
Oxford.
513. Soames, S. (1989): Presupposition, in: D. Gabbay/F. Guenthner (eds.), Handbook of Philosophical Logic. vol. 4, Reidel, Dordrecht, 553{616.
514. Strawson, P.F. (1950): On referring, Mind, N.S., 59, 320{344.
515. Strehle, P. (1983): Zur Begrundung und Darstellung mehrdimensionaler
mehrwertiger Logiken, Dissertation B, Universitat Leipzig.
516. Suchon, W. (1974): La methode de Smullyan de construire le calcul n-valent
de Lukasiewicz avec implication et negation, Reports Mathematical Logic 2,
37{42.
517. Sugeno, M. (1977): Fuzzy measures and fuzzy integrals: a survey, in: M.M.
Gupta/G.N. Saridis/B.R. Gaines (eds.) Fuzzy Automata and Decision Processes, North-Holland Publ. Comp., Amsterdam, 89{102.
518. Sugeno, M. (ed.) (1985): Industrial Applications of Fuzzy Control. NorthHolland Publ. Comp., Amsterdam.
519. Sugeno, M. { Nishida, M. (1985): Fuzzy control of a model car, Fuzzy Sets
Systems 16, 103{113.
520. Sundholm, G. (1983): Systems of deduction, in: D. Gabbay/F. Guenthner
(eds.), Handbook of Philosophical Logic, vol. 1, Reidel, Dordrecht, 133{188.
521. Surma, S.J. (1973): Jaskowski's matrix for the intuitionistic propositional
calculus, in: S.J. Surma (ed.), Studies in the History of Mathematical Logic.
Ossolineum, Wroclaw, 87{121.
522. Surma, S.J. (1984): An algorithm for axiomatizing every nite logic, in: D.C.
Rine (ed.) Computer Science and Multiple Valued Logic. North-Holland Publ.
523. Surma, S.J. { Wronski, A. { Zachorowski, S. (1975): On Jaskowski-type
semantics for the intuitionistic propositional logic, Studia Logica 34, 145{148.
524. Suzuki, Nobu-Yuki (1997): Kripke frame with graded accessibility and fuzzy
possible world semantics. Studia Logica 59, 249{269.
590
References

525. Svejdar,
V. { Bendova , K. (2000) On inter-expressibility of logical connectives in Godel fuzzy logic, Soft Computing 4, 103{105.
526. Takahashi, Moto-o (1968): Many-valued logic of extended Gentzen style. I,
Science Reports Tokyo Kyoiku Daigaku , 9, 271{292.
527. Takeuti, G. { Zaring, W.M. (1971): Introduction to Axiomatic Set Theory.
Springer, New York.
528. Takeuti, G. { Zaring, W.M. (1973): Axiomatic Set Theory. Springer, New
York.
529. Tarski, A. (1951): A Decision Method for Elementary Algebra and Geometry.
Univ. of California Press, Berkeley (2nd printing).
530. Thiele, H. (1958): Theorie der endlichwertigen Lukasiewiczschen Pradikatenkalkule der ersten Stufe, Zeitschr. math. Logik Grundl. Math. 4, 108{142.
531. Thiele, H. (1993): On the denition of modal operators in fuzzy logic, in:
23rd Internat. Symp. Multiple-Valued Logic, Sacramento/Ca., 1993, Proceedings, IEEE Computer Soc., Piscataway/NJ, 62{67.
532. Thiele, H. (1994): On t-quantiers and s-quantiers, in: 24th Internat. Symp.
Multiple-Valued Logic, Boston/Mass., 1994, Proceedings, IEEE Computer Soc.,
Piscataway/NJ, 264{269.
533. Thiele, H. (1995): On fuzzy quantiers, in: Z. Bien/K.C. Min (eds.), Fuzzy
Logic and Its Applications to Engineering, Information Sciences, and Intelligent
Systems. Selected Papers 5th IFSA World Congress, Seoul, Korea, July 1993,
Kluwer, Dordrecht, 343{352.
534. Thiele, H. (1998): Axiomatic considerations of the concepts of R-implication
and t-norm, in: Proc. Seventh Internat. Conf. Information Processing and Management of Uncertainty in Knowledge-Based Systems { IPMU '98, Sorbonne,
Paris 1998, 1798{1805.
535. Thole, U. { Zimmermann, H.J. { Zysno, P. (1979): On the suitability of
minimum and product operators for the intersection of fuzzy sets, Fuzzy Sets
Systems 2, 167{180.
536. Thomason, S.K. (1978): Possible worlds and many truth values, Studia Logica
37, 195{204.
537. Tokarz, M. (1977): A method of axiomatization of Lukasiewicz logics, in:
R. Wojcicki/G. Malinowski (eds.) Selected Papers on Lukasiewicz Sentential
Calculi. Ossolineum, Wroclaw, 113{117.
538. Tong, R.M. (1977): A control engineering review of fuzzy systems, Automatica 13, 559{570.
539. Tong, R.M. { Beck, M.B. { Latten, A. (1980): Fuzzy control of the activated sludge wastewater treatment process, Automatica 16, 659{701.
540. Traczyk, T. (1963): Axioms and some properties of Post algebras, Colloquium Mathematicum 10, 193{209.
541. Traczyk, T. (1967): On Post algebras with uncountable chain of constants.
Algebras and homomorphisms, Bull. Acad. Polon. Sci., Ser. Sci. Math., Astronom., Phys., 15, 673{680.
542. Traczyk, T. (1977): Post algebras through P0 and P1 lattices, in: D.C. Rine
(ed.) Computer Science and Multiple Valued Logic. North-Holland Publ. Comp.,
Amsterdam, 115{136.
543. Trzesicki, K. (1990): Many-valued tense logic and the problem of determinism, in: 20th Internat. Symp. Multiple-Valued Logic, Charlotte/NC, 1990,
544. Trillas, E. (1979): Sobre funciones de negacion en la teoria de conjuntos
difusos, Stochastica III, 47{60.
References
591
545. Trillas, E. { Valverde, L. (1985): On implication and indistinguishability in the setting of fuzzy logic, in: J. Kacprzyk/R.R. Yager (eds.) Management
Decision Support Systems using Fuzzy Sets and Possibility Theory. TU V Rheinland, Cologne, 198{212.
546. Turquette, A.R. (1967): Peirce's Phi and Psi operators for triadic logic,
Transactions Charles Sanders Peirce Society 3, 66{73.
547. Turunen, E. (1997): Algebraic structures in fuzzy logic, Fuzzy Sets Systems
52, 181{188.
548. Turunen, E. (1995): Well-dened fuzzy sentential logic, Math. Logic Quarterly 41, 236{248.
549. Turunen, E. (1997): Rules of inference in fuzzy sentential logic, Fuzzy Sets
Systems 85, 63{72.
550. Turunen, E. (1999): Mathematics Behind Fuzzy Logic, Advances in Soft
Computing, Physica-Verlag, Heidelberg.
551. Urquhart, A. (1973): An interpretation of many-valued logic, Zeitschr. math.
552. Urquhart, A. (1986): Many-valued logic, in: D. Gabbay/F. Guenthner (eds.),
Handbook of Philosophical Logic, vol. 3, Reidel, Dordrecht, 71{116.
553. Valverde, L. { Trillas, E. (1985): On modus ponens in fuzzy logic, in: 15th
Internat. Symp. Multiple-Valued Logic, Kingston/Ontario, 1985, Proceedings,
IEEE Computer Soc., New York/NY, 294{301.
554. van Dalen, D. (1986): Intuitionistic logic, in: D. Gabbay/F. Guenthner
(eds.), Handbook of Philosophical Logic , vol. 3, Reidel, Dordrecht, 225{339.
555. van der Sandt, R.A. (1988): Context and Presupposition. Croom Helm,
London.
556. Vasilev, N.A. (1910): O castnych suzdeniach, o treugol'nike protivopoloznostej, o zakone isklucennogo cetvertogo. Ucenie Zapiski Kazanskogo Universiteta, Kasan.
557. Vasilev, N.A. (1912): Voobrazaemaja (nearistoteleva) logika, Zurnal Ministerstva Narodnogo Prosvescenija 40, 207{246.
558. Visser, A. (1912): Four-valued semantics and the Liar, J. Philosophical Logic
13, 181{217.
559. Wade, L.J. (1945): Post algebras and rings, Duke Mathematical J. 12, 389{
395.
560. Wajsberg, M. (1931): Aksjomatyzacja trojwartosciowego rachunku zdan,
Comptes Rendus Seances Societe des Sciences et Lettres Varsovie, cl. III, 24,
126{148. [English translation in [562].]
561. Wajsberg, M. (1935): Beitrage zum Metaaussagenkalkul. I, Monatshefte
Mathematik Physik 42, 221{242.
562. Wajsberg, M. (1977): Logical Works. (ed.: S.J. Surma), Ossolineum, Wroclaw.
563. Ward, N. (1937): Residuation in structures over which a multiplication is
dened, Duke Mathematical J. 3, 627{636.
564. Ward, N. { Dilworth, R.P. (1939): Residuated lattices, Transactions American Mathematical Society 45, 335{354.
565. Webb, D.L. (1936): Denition of Post's generalized negative and maximum
in terms of one binary operation, American J. Mathematics 58, 193{194.
566. Weber, S. (1983): A general concept of fuzzy connectives, negatives and
implications based on t-norms and t-conorms, Fuzzy Sets Systems 11, 115{134.
567. Wechler, W. (1978): The Concept of Fuzziness in Automata and Language Theory. Studien zur Algebra und ihre Anwendungen 5, Akademie-Verlag,
Berlin.
592
References
568. Weingartner, P. (1968): Modal logics with two kinds of necessity and possibility, Notre Dame J. Formal Logic 9, 97{159.
569. Weingartner, P. (1973): A predicate calculus for intensional logic, J. Philosophical Logic 2, 220{303.
570. Weingartner, P. (1975): A nite approximation to models of set theory,
571. Weidner, A.J. (1981): Fuzzy sets and Boolean-valued universes, Fuzzy Sets
Systems 6, 61{72.
572. White, R.B. (1979): The consistency of the axiom of comprehension in the
innite-valued predicate logic of Lukasiewicz, J. Philosophical Logic 8, 509{534.
573. Wojcicki, R. { Malinowski, G. (eds.) (1977): Selected Papers on Lukasiewicz Sentential Calculi. Ossolineum, Wroclaw.
574. Wolf, R.G. (1977): A survey of many-valued logic (1966{1974), in: J.M.
Dunn/G. Epstein (eds.), Modern Uses of Multiple-Valued Logic. Reidel, Dordrecht 167{323.
575. Woodruff, P.W. (1970): Logic and truth value gaps, in: K. Lambert (ed.),
Philosophical Problems in Logic. Reidel, Dordrecht, 121{142.
576. Woodruff, P.W. (1974): A modal interpretation of three-valued logic, J.
Philosophical Logic 3, 433{439.
577. Wronski, A. (1987): Remarks on a survey article on many valued logic by
A. Urquhart, Studiea Logica 46, 275{278.
578. Yager, R.R. (1980): On the lack of inverses in fuzzy arithmetic, Fuzzy Sets
Systems 4, 73{82.
579. Yager, R.R. (1980): On a general class of fuzzy connectives, Fuzzy Sets Systems 4, 235{242.
580. Yager, R.R. (1983): Some relationships between possibility, truth and certainty, Fuzzy Sets Systems 11, 151{156.
581. Ying, Mingsheng (1988): On standard models of fuzzy modal logics. Fuzzy
Sets Systems 26, 357{363.
582. Zadeh, L.A. (1965): Fuzzy sets, Information and Control 8, 338{353.
583. Zadeh, L.A. (1965): Fuzzy sets and systems, in: J. Fox (ed.) Systems Theory.
Microwave Research Institute Symposia Series 15, Polytechnic Press, Brooklyn,
29{37.
584. Zadeh, L.A. (1971): Similarity relations and fuzzy orderings, Information
Sciences 3, 159{176.
585. Zadeh, L.A. (1975): The concept of a linguistic variable and its application
to approximate reasoning. I, Information Sciences 8, 199{250.
586. Zadeh, L.A. (1975): Fuzzy logic and approximate reasoning, Synthese 30,
407{428.
587. Zadeh, L.A. (1978): Fuzzy sets as a basis for a theory of possibility, Fuzzy
Sets Systems 1, 3{28.
588. Zadeh, L.A. (1979): A theory of approximate reasoning, in: J.E. Hayes/D.
Michie/L.I. Mikulich (eds.), Machine Intelligence 9. Halstead Press, New York,
149{194.
589. Zadeh, L.A. (1982): Possibility theory as a basis for representation of meaning, in: Sprache und Ontologie. Akten 6. Internat. Wittgenstein Symposium
1981, Kirchberg/Wechsel. Holder-Pichler-Tempski, Wien, 253{262.
590. Zadeh, L.A. (1987): Fuzzy Sets and Applications. Selected Papers (R.R. Yager
et al. eds.), Wiley, New York.
591. Zadeh, L.A. (1996): Fuzzy Sets, Fuzzy Logic, and Fuzzy Systems. Selected
Papers (G.J. Klir/B. Yuan eds.), World Scientic, Singapore.
592. Zawirski, Z. (1935): U ber das Verhaltnis mehrwertiger Logik zur Wahrscheinlichkeitsrechnung, Studia Philosophica 1, 407{442.
References
593
593. Zhang, Jinwen (1979): The normal fuzzy set structures and Boolean-valued
models, J. Huazhong Institute Technology 7, no. 2, 1{9.
594. Zhang, Jinwen (1979): Some basic properties of normal fuzzy set structures,
J. Huazhong Institute Technology 7, no. 3, 1{9.
595. Zhang, Jinwen (1980): A unied treatment of fuzzy set theory and Booleanvalued set theory { fuzzy set structures and normal fuzzy set structures, J.
Math. Anal. Appl. 76, 297{301.
596. Zhang, Jinwen (1982): Between fuzzy set theory and Boolean valued set
theory, in: M.M. Gupta/E. Sanchez (eds.), Fuzzy Information and Decision
Processes. North-Holland Publ. Comp., Amsterdam, 143{147.
597. Zimmermann, H.J. (1979): Theory and applications of fuzzy sets, in: K.B.
Haley (ed.), Operational Research '78. Proceedings 8th IFORS Internat. Conference Toronto 1978, North-Holland Publ. Comp., Amsterdam, 1017{1033.
598. Zimmermann, H.J. (1991): Fuzzy Set Theory and its Applications. Kluwer,
Boston.
599. Zimmermann, H.J. { Zadeh, L.A. { Gaines, B.R. (eds.) (1984): Fuzzy Sets
and Decision Analysis. North-Holland Publ. Comp., Amsterdam.
600. Zimmermann, H.J. { Zysno, P. (1980): Latent connectives in human decision
making, Fuzzy Sets Systems 4, 37{51.
601. Zinov'ev, A.A. (1963): Philosophical Problems of Many-Valued Logic. Reidel,
Dordrecht.
594
References
SUBJECT INDEX
595
Subject Index
absorption property, 558
adjoint pair, 93
adjunction property, 92
algebra
BCK-, 234
BL-, 351
Boolean, 9, 539
complete, 539
G-, 277
Heyting, 275
Lindenbaum, 215, 236
Lukasiewicz, 248
MTL-, 351
MVm -, 240
MV-, 216
P-, 316
Post, 318
product, 295
pseudo-Boolean, 275
Wajsberg, 244
alphabet, 16
approximation, sequential, 532
assignment, 27
automorphism
of the unit interval, 82
axiom, 107
axiomatizable, 108
axiomatization, adequate, 108
basis, interpretational, 6
BCK-algebra, 234
biconsequence, 389
bijection, 10
bilattice, 400
bivalence, principle of, 3, 4
BL-algebra, 351
bound
lower, 8
universal lower, 9
universal upper, 9
upper, 8
bounded occurrence, 23
branch, of a tableau
closed, 139
complete, 143
open, 139
calculus
logical, 6, 107
logical, of Gentzen-type, 137
logical, of Hilbert-type, 108
sequent, 149
tableau, 137
cardinality, 7
carrier, 9
cartesian product, fuzzy, 435
chain, 8
Heyting, 275
MV-, 224
product, 295
character, nite, 147
class term, 7
closure operator
fuzzy, 487
closure operators, 486
closure system
fuzzy, 487
closure systems, 487
comparable, 8
complement, 318, 426
complete
Post-, 196
complete, functionally, 161
completeness
of a logical calculus, 108
compositionality, principle of, 3, 4,
24
comprehension principle, 557
conjunction
innitary, 434
Lukasiewicz, 66, 92
strong, 181
weak, 181
connective, 15
596
References
main, 17
principal, 17
consequence, 33, 40
consistent, -, 457
continuous, 10
left, 10
contradiction, 32
countermodel, 33
cover, 528
optimal, 530
crisp set, 424
cut, 372
de Morgan law, 251, 335
strong algebraic, 351
decidable, 171
deduction theorem, 130
degree
of containment, 429
of derivability, 477
of equality, 429
of existence, 413
of soundness, 481
derivability
degree of, 477
notion of, 107
derivable, 109
derivation, 110, 153
derivation rule, 107
diagonal, 445
disjunction
innitary, 434
Lukasiewicz, 90, 92
strong, 181
weak, 181
domain, 439
domination (for t-norms), 458
double-negation property, 351
element, complemented, 318
embeddable, locally, 11
empty fuzzy set, 431
entailment, 33, 40
equality, L-valued, 414
equation, MV-, 225
equivalence class, 8
fuzzy, 446
equivalence relation, fuzzy, 446
equivalent
denitionally, 10
semantically, 19, 39
exchange principle, 97
extension
conservative, 486
consistent schematic, 281
elementary, 50
proper schematic, 281
schematic, 281, 369
extension principle, 437
extensionality, principle of, 3, 4, 24,
557
falsity, logical, 32
lter, 41, 277, 300, 366, 370
prime, 277, 300, 366, 370
principal, 41, 277, 300
proper, 41, 277, 300
lter equivalence, 43
formula
atomic, 16, 23
bounding, 521
compound, 16
signed, 138
well-formed, 16
free occurrence, 23
full image, 442
function
generalized characteristic, 424
algebraic, 320
choice, 9
membership, 424
negation, 85
negation, strict, 85
negation, strong, 85
Sheer-, 165
truth degree, 17
generalized, 26
truth degree, of H , 20
function algebra, 163
future contingents, 55
SUBJECT INDEX
fuzzication, 457

fuzzy equivalence relation, 446
graded, 465
fuzzy logic, 471, 473
rational, 479
fuzzy partial ordering, 454
graded, 461
fuzzy preordering, 454
fuzzy relation, 438
fuzzy set, 424
empty, 431
universal, 431
fuzzy subset, 424
fuzzy theory, 473
Godel systems, 269
gap, 394
generator, additive (of a t-norm),
81
glut, 394
graded properties, 460
graded set, 49
group, 11
-`, 11
-o, 11
abelian, 11
ordered, 11
Heyting algebra, 275
Hintikka set, 144
saturated, 145
homomorphism
for MV-algebras, 224
Horn formula, 427
basic, 427
hull
t-transitive, 452
consequence, 40
hypersequent, 158
ideal, 221
prime, 221
proper, 221
idempotent, 67
597
idempotent (w.r.t. some t-norm),

67
identity relation
crisp, 402
many-valued, 402, 429
i, 7
implication
Godel, 92
Lukasiewicz, 91
R-, 93
S-, 93
implication function, 98
inclusion relation, 428
many-valued, 429
independent, 538
inference rule
admissible, 129
sound, 129
injection, 10
interpretation, 26, 305
$-absolute, 404
$-normal, 410
faithful, 305
intersection
of a family, 433
innitary, 433
t-norm based, 426
inverse, pseudo-, 77
involution, 85
irreducible, 372
isomorphism, 10
partial, 11
j-connective, 104
K -based, 528
-complement, 85
lattice, 8
distributive, 9
residuated, 348
left continuous, 68
Lemma of Zorn, 8
L-interpretation, 375
safe, 375
598
References
logic
basic t-norm, 352, 366, 376
fuzzy, 471, 473
rational, 479
intuitionistic, 525, 526
monoidal, 352, 376
monoidal t-norm, 352, 363, 376
partial, 495
logical matrix, 18, 538
Lukasiewicz systems, 179
mapping
polynomial, 320
projection, 10
matrix
logical, 18
matrix, logical, 18, 538
maximal, 8
maximum, nilpotent, 90
membership function, 424
metavariable, 16, 22
minimal, 8
minimum, nilpotent, 73
model, 32, 40
()-, 33
t-, 33
Boolean valued, 556
of a graded set, 49
of a signed formula, 143
sequent, 149
model class, 32, 40
modus ponens, 332
generalized, 479
monoid
divisible, 349
integral, 347
monoidal logic, 376
MTL-algebra, 351
MVm -algebra, 240
MV-algebra, 216
injective, 235
semisimple, 233
simple, 233
negation, t-norm based, 104, 335
normal condition, 6, 31
normal form, prenex, 253
notation
inx, 17
prex, 17
n-tuple, 7
occurrence
bounded, 23
free, 23
omitted, locally, 255
operation, consequence, 34
operator, -, 97
ordered pair, 7
ordering
knowledge, 399
partial, 8
truth, 399
partial ordering
fuzzy, 454, 458
graded fuzzy, 461
irre exive fuzzy, 454
partition
L-valued, 448
polymorphism, 168
Post algebra, 318
Post systems, 313
Post-complete, 196
powerset, 7
pre-linearity, 351
predicate
W -valued, 25
L-valued, 374
preordering, fuzzy, 454
presupposition, 494
principle
of bivalence, 3, 4
of compositionality, 3, 4, 24
of comprehension, 557
of extensionality, 3, 4, 24, 557
problem of satisability, 172
product
bold(face), 66
bounded, 66
SUBJECT INDEX
cartesian, 7
direct, 9
drastic, 73
fuzzy cartesian, 435
relational, 439
subdirect, 224
product system, 174
projection, 163
proof, in a logical calculus, 110
property
degree ranking, 97
double-negation, 351
graded, 460
signed consistency, 147
pseudo-Boolean algebra, 275
quantication, 23
restricted, 8
quantier, 22
existential, 26, 250
t-conorm based, 343
universal, 26, 250
t-norm based, 343
radical, 221
range, 439
realized, locally, 255
reducible, 372
relation, 8
(t; )-antisymmetric , 458
-re exive, 458
t-antisymmetric, 445
t-asymmetric, 445
equivalence, 8
fuzzy, 438
inverse, 439
irre exive, 445
linear, 459
re exive, 445
satisfaction, 27
similarity, 447
symmetric, 445
t-transitive, 445
weakly linear, 459
rule
599
constant introduction, 479

introduction, 153
of detachment, 109, 332
graded, 479
of generalization, 480
of inference, 107
admissible, 129
sound, 129
of universal quantication, 126
tableau extension, 140
thinning, 153
Russell's paradox, 558
satisfaction relation, 27
satisability problem, 172
satisable, 33, 38, 137, 143
saturated, 372
scope, 23
sentence, 23
universal, 12
sequent, 137, 149
valid, 149, 158
sequent model, 149
set
L-valued, 415
separated, 415
fuzzy, 424
crisp, 424
graded, 49
sign, 138
similarity relation, 447
similarity type, 9
singleton
fuzzy, 431
M-valued, 448
u-, 431
soundness, 108
degree of, 481
partial, 483
standard condition, 6, 30
of a universal quantier, 125
standard extension, 129
structure, algebraic, 9
subformula, 17
subsequent, nite, 154
600
References
subset, 7
fuzzy, 424
substructure, elementary, 50
sum
algebraic, 90
bounded, 90
drastic, 90
generalized ordinal, 84
ordinal, 74, 371
superposition, 162
supervaluation, 495
symbol, individual, 22
t-conorm, 89
t-norm, 67
Archimedean, 69
Lukasiewicz, 72
strict, 72
t-norm logic
basic, 352, 366, 376
monoidal, 352, 363, 376
tableau, 138
closed, 143
complete, 143
open, 143
tableau extension rule, 140
tautology, 31
term, 23
theorem
algebraic completeness, 231, 303
axiomatizability, for Lm , 242
compactness, 34, 49, 154
compactness, for L1 , 261
completeness, 111, 127, 134,
148, 155, 356, 365, 368,
379, 406, 412, 414, 485
restricted, 485
weak, 309
completeness, for , 304
completeness, for G1 , 281, 289
completeness, for L1 , 239
deduction, 130
extended completeness, 381
niteness, for entailment, 37,
130, 154
Lowenheim-Skolem, 380
model existence, 147
of a logical calculus, 107, 109
omitting types, 256
replacement, 19, 39
soundness, 110, 144, 153, 377,
406, 411
strong completeness, 133
transfer, 428
theory, fuzzy, 473
truth degree, 4
antidesignated, 29
designated, 29
multi-dimensional, 173
undesignated, 30
truth degree table, 64
truth value, 5
ultralter, 41
ultraproduct, 46
union
of a family, 433
innitary, 433
t-norm based, 426
unit element, 9
universal set, 431
u-, 431
universe of discourse, 424
vague notion, 423
valid, 137
logically, 5, 31, 38
valuation, 18
variable
individual, 21
propositional, 15
Wajsberg algebra, 244
well-ordered, 8
w, 16
zero divisor, 72
zero element, 9
Zorn's Lemma, 8
INDEX OF NAMES
601
Index of Names
Ackermann, R., 59
Aristotle, 55
Arruda, A.I., 55
Gluschankof, D., 235

Godo, L., 352, 363
Grishin, V.N., 561
Belluce, L.P., 262, 562

Belnap, N.D., 399
Bernays, P., 56, 535
Beth, E.W., 137
Blau, U., 390
Bocvar, D.A., 56, 385, 386, 558
Brouwer, L.E.J., 419, 525
Burali-Forti, C., 557
Buridan, J., 173
Byrd, M., 4
Hahnle, R., 485

Hohle, U., 352, 413
Hajek, P., 59, 352, 366, 485
Hamacher, H., 73
Hay, L.S., 262, 562
Heyting, A., 525
Hintikka, J., 137
Cantor, G., 557

Chang, C.C., 196, 216, 220, 262,
560{562
Cignoli, R., 250
Cohen, P.J., 538
Craig, W., 198
Kleene, S.C., 56, 385, 388

Kolmogorov, A., 525
Dedekind, R., 557

Duy, M.J., 505
Dumitriu, A., 59
Dummett, M., 281
Dunn, J.M., 398
Dwinger, Ph., 318
Malinowski, G., 4
Mangani, P., 216
McColl, H., 55
McNaughton, R., 201
Menger, K., 66
Moisil, G.C., 248
Morgan, C.G., 409
Mostowski, A., 127, 267
Mundici, D., 201, 234
Epicureans, 55
Epstein, G., 317
Esteva, F., 352, 363
Fenstad, J.E., 560, 561
Fitting, M.C., 513
Fraenkel, A.A., 542
Frank, H.J., 67, 73, 91
Godel, K., 56, 65, 85, 92, 327, 336,
525, 558
Gaines, B.R., 58
Gentzen, G., 137
Girard, J.Y., 361
Jaskowski, S., 56, 174, 526, 527

Jablonskij, S.V., 168
Lukasiewicz, J., 55, 56, 58, 65, 84,

91, 174, 179, 196, 313, 327,
336, 393, 421, 503{505, 508
Novak, V., 59
Pavelka, J., 59
Pedrycz, W., 97
Peirce, Ch.S., 55
Peters, S., 498
Post, E.L., 55, 84, 165, 168, 196,
313
Ragaz, M., 485, 563
Rasiowa, H., 322, 323
602
References
Reichenbach, H., 58
Rescher, N., 59
Rose, A., 196
Rosenberg, I., 168
Rosenbloom, P.C., 316
Rosser, J.B., 56, 59, 108, 134, 196
Rousseau, G., 322
Russell, B., 557
Sanchez, E., 97
Sanjaya, 394
Schroter, K, 58
Schroter, K., 120, 149, 193
Schweizer, B., 66
Scott, D., 4
Sestakov, 388
Seuren, P.A.M., 496, 497
Sklar, A., 66
Skolem, Th., 542, 558, 560, 561
Slupecki, J., 56, 511
Stoics, 55
Strawson, P.F., 496
Sugeno, M., 85
Tarski, A., 200, 504
Thiele, H., 402
Turquette, A.R., 56, 59, 108, 134
Urquhart, A., 4
van Fraassen, B.C., 495
Vasilev, N.A., 55
Wajsberg, M., 56, 195, 196
Webb, D.L., 165
Weber, S., 74
White, R.B., 561
Woodru, P.W., 496, 505, 509, 511
Yager, R., 74
Zadeh, L.A., 424, 447, 460
Zawirski, Z., 58
Zermelo, E., 542
Zinov'ev, A.A., 59
INDEX OF SYMBOLS
Index of Symbols
G (A), 227
seq , 292
(a; b), 7
;t , 335
=def , 7
A B, 7
Di , 319
G+a , 227
Pm(n) , 163
[a]S , 8
&, 180
CnS , 34
,, 7S
Mod , 32
), 7
IP (A) , 7
ValS (H; ), 18
ValSA (H; ), 27
, 137
G ,S271
VerQ , 26
: ai , 27
[y1 =b1; : : : ; yn =bn], 27
T, 391
,7
S
V, 7
126
V, W
;
V W, 539
; , 187
?; >, 5
\t ; [t , 426
card A, 7
[:x " A ] , 425
=, 426
$L , 180
$t , 328
;, 7
t , 429
et , 386
et1 ; et2 ; et3 , 65
8; 9, 26
$, 402
, 7
^, 180, 217
^t , 328
hFi, 163
6R, 8
_, 180, 217
_t , 328
K , 293
K BTL , 368
K L , 235
K ML , 353
K MTL , 363
K mG , 152
Di, 322
DS+ ; DS; , 29
J S , 15
KS , 15
K6=S t , 150
L0 , 471
LS , 15
R; R+ , 11
V0 , 15
W, 5
W S , 18
Wm ; W0 ; W1 , 63
X =R, 446
A j= , 474
; (G ; a), 227
, 291
wH , 20
0, 216
1, 217
AxBTL , 367
AxML , 353
AxMTL , 363
AxRT , 109
G(W ), 270
L , 179
PL2 , 6
Pm , 313
603
604
References
Sval, 149
prnk , 163
s : H , 139
tautL , 189
var(H ), 19
F -Prod, 43
j=S , 27, 33
:, 179, 216
non0 , 85
non1 , 84
non2 , 84
nont , 104
, 292
, 216
, 217
! , 109
Qi , 10
,9
Q P
; , 186
^, 7
seq 1 ; seq 2 , 91
9, 7
8, 7
, 269, 313
:, 7
_, 7
u, 295
t, 295
tautG , 272
tautS , 31
, 7
jt, 429
w~H , 20
!, 217
! , 292
!Bl , 391
!t , 328
!G , 269
!L , 179
>; ?, 3
't , 329
't , 97
`K , 107
`?K , 129
`L, 235
`G , 153
`RT , 109
Y, 180
vel1 ; vel2 ; vel3 , 90

verS' , 18
fx jj H (x)g, 425
fx j H (x)g, 7
f [D], 10
f (;1), 77
p(n) , 15
IF (X ), 426
B3 , 386
K 3, 388
ei , 318
st, 89
ts, 90
G , 269
Js ; js , 105
AB, 7
(DEDj= ), 129
(DED` ), 129
(FINj= ), 129
(FIN` ), 129
(MP), 109

A Treatise On Many-Valued Logics

Hochgeladen von

Dokumentinformationen

Originalbeschreibung:

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

A Treatise On Many-Valued Logics

Hochgeladen von

Copyright:

Verfügbare Formate

See

A Treatise on Many-Valued Logics

Available from: Siegfried Gottwald

class of nitely-valued systems, including axiomatizations of the entailment

Part I. Basic Notions

2. The Formalized Language and its Interpretations : : : : : : : : : 15

3. Logical Validity and Entailment : : : : : : : : : : : : : : : : : : : : : : : : : : 29

Designated Truth Degrees . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

4. Outline of the History of Many-Valued Logic : : : : : : : : : : : : : 55

6.1 The Axiomatizability Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . 107

6.4 Axiomatizing the Entailment Relation . . . . . . . . . . . . . . . . . . . . 128

7. Sequent and Tableau Calculi : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 137

7.1 Tableau Calculi for Many-Valued Logic . . . . . . . . . . . . . . . . . . . . 138

8. Some Further Topics : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 161

8.1 Functional Completeness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161

Part III. Particular Systems of Many-Valued Logic

9.1 The Propositional Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179

10. The Godel Systems : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 267

10.1 The Propositional Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267

11. Product Logic : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 291

11.1 The Propositional System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291

12. The Post Systems : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 313

12.1 The Original Presentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313

13. t-Norm Based Systems : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 327

13.1 The Propositional Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327

14. Axiomatizing t-Norm Based Logics : : : : : : : : : : : : : : : : : : : : : : : 345

14.1 The Propositional Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345

15. Some Three- and Four-Valued Systems : : : : : : : : : : : : : : : : : : : 385

15.1 Three-Valued Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 385

16. Systems with Graded Identity : : : : : : : : : : : : : : : : : : : : : : : : : : : : 401

Graded Identity Relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 401

Part IV. Applications of Many-Valued Logic

18.1 Vagueness of Notions and Fuzzy Sets. . . . . . . . . . . . . . . . . . . . . . 423

19. Fuzzy Logic : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 471

Many-Valued Logic with Graded Consequences . . . . . . . . . . . . . 472

20. Treating Presuppositions with Many-Valued Logic : : : : : : : : 493

20.1 The Phenomenon of Presuppositions . . . . . . . . . . . . . . . . . . . . . . 493

21. Truth Degrees and Alethic Modalities : : : : : : : : : : : : : : : : : : : : 503

21.1 Interpreting Modal Logic as Many-Valued Logic . . . . . . . . . . . . 503

22. Approximating Intuitionistic and Other Logics : : : : : : : : : : : 525

22.1 Many-Valued Approaches toward Intuitionistic Logic . . . . . . . . 525

23. Independence Proofs : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 535

23.1 The Propositional Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 535

24. Consistency Considerations for Set Theory : : : : : : : : : : : : : : : 557

1.1 Classical and Many-Valued Logic

logics may consult [178] or also [315].

1.1 Classical and Many-Valued Logic

Deleting the principle of bivalence does for many-valued logic, however,

1.2 Preliminary Notions

1.2 Preliminary Notions

Sometimes we also write fAi j i 2 I g for i2I Ai in case of this generalized

1.2 Preliminary Notions

By a slight abuse of language, these projection mappings i with 1  i  n

1.2 Preliminary Notions

For functions f : A ! B one has to distinguish their values f (a) for a 2 A

1.2 Preliminary Notions

Corollary 1.2.1. If a universal sentence of the language of `-groups is valid

2. The Formalized Language and its

2.1 Propositional Syntax

2. The Formalized Language and its Interpretations

to represent propositional variables.1

2.2 Propositional Semantics

10. The Godel Systems : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 267

By a slight abuse of language, these projection mappings i with 1 i n

Given a compound formula H '(H1 ; H2 ; : : : ; Hn ) the { in this case

or there exists some characteristic class KS of algebraic structures (of coin-

and such that each one of the ws Hi00 , 1 i n, can be built up

its (nonempty) family J S of (basic) propositional connectives,

its family P S of predicate symbols together with an arity function assigning

It then is a routine matter to dene the class of well-formed formulas and

ValSA ((Qx1 ; : : : ; xk )(H1 ; : : : ; Hn ); ) =def VerSQ(AH ):

DS+ [ DS; W S and DS+ \ DS; = ; ;

ValS (H ; ) 2= DS for some H 2 , is often also called a countermodel of H

As a side remark it should be mentioned that the notion of ( t)-model

Therefore we now suppose that each nite subset of has a model. To

It is easy to recognise that each 2 ModS^ ( ) has to satisfy (p0 ) = 0.

in A. By ModS ( ) the class of all S-models of shall be denoted. And a

I shall be not restricted to specic cases; it shall be allowed to be innite, even

For each lter F on I one denes a binary relation F in Xi2I Ai via

f F g =def fi 2 I j f (i) = g(i)g 2 F

This relation F is an equivalence relation: it is re exive because I 2 F