Beruflich Dokumente
Kultur Dokumente
discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/259645593
CITATIONS
READS
398
91
1 author:
Siegfried Gottwald
University of Leipzig
180 PUBLICATIONS 2,218 CITATIONS
SEE PROFILE
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
Contents
3
6
Propositional Syntax . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Propositional Semantics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
First-Order Syntax . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Many-Valued Predicates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
First-Order Semantics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
15
17
21
24
26
29
31
38
40
Conjunction Connectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
Negation Connectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
Disjunction Connectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
Implication Connectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
The J{Connectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
6. Axiomatizability : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 107
VIII
Contents
Contents
IX
Contents
Part I
Basic Notions
1. General Background
1. General Background
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
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.
1. General Background
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.
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
1. General Background
T
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
1. General Background
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
11
12
1. General Background
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
14
1. General Background
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.
16
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".
17
18
either there exists some particular truth degree structure MS for S, the
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
20
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.
22
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
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.
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
26
27
28
30
31
32
33
34
35
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-
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
38
the proof of Theorem 2.2.1. The details are left for the reader as an exercise.
39
40
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
42
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
43
(3.21)
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
45
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
j =1
(3.29)
47
j =1
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
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
49
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
52
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
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
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
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
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
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
64
;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)
65
66
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
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
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)
70
t-norm t such that the unary function f (u) = t(u; u) is right continuous (in the
standard sense).
71
(5.17)
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
72
Proposition 5.1.4. The only t-norm which has all u 2 [0; 1] as idempotents
73
74
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
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.
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
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
d2 ; d1
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
[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 ); if z 2 [0; f (0)]
0;
if z 2 (f (0); 1].
(5.38)
(5.39)
79
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
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
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.
83
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.
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
(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
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
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
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".
89
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.
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].
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
Corollary 5.4.1. For left continuous t-norms t; t one has for all u; v 2
[0; 1]
logic.
95
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
97
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
98
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
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
(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
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
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
105
106
6. Axiomatizability
108
6. Axiomatizability
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)
1
1
(MP)
A; A ! B
B
110
6. Axiomatizability
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
tional system S of many-valued logic satises the conditions (RT1) and (RT2).
112
6. Axiomatizability
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) :
=1
=1
i k
i k
k
k
! (Hi ; A)) ! (Hk !
! (Hi ; B )) ;
!
i
i
113
=1
! ((Hk+1
+1
=1
i k
=1
i k
k
k
! (Hi ; A) !
! (Hi ; B )) :
! (
i
i
(6.10)
=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)
=1
=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
(6.20)
j =1;j 6=u
i r
;
`RT ! (Ai ; G) ! Hu ! Hv !
=1
! (Hj ; G)
j ;j 6 u;v
k+1
=1 =
(6.21)
115
!
k+1
j =1;j 6=u;v
! (Hv ! (Hu !
(Hj ; G)))
!
k+1
j =1;j 6=u;v
(Hj ; G))) :
(6.22)
`RT Hu !
k
! (Hj ; G) !
! (Hj ; G) ;
j ;j 6 u;v
j ;j 6 v
k+1
+1
=1 =
=1 =
(6.23)
`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
i n
i n
;
! ! (Jti (Hi ); A) ! ! (Jti (Hi ); B ) :
=1
=1
(6.25)
+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
(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
=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
(6.35)
which via an application of (6.3) gives (6.34) in general for this rst case.
117
i n
=1
But because of
=1
(6.36)
(6.37)
j r
=1
(6.39)
i k
i n
(6.40)
(6.41)
(6.42)
i n
i n
=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
(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
=1
=1
(6.45)
i n
(6.46)
i n
(6.47)
We now iterate the transition from (6.44) to (6.46) another r ; 2 times. This
nally gives
j n
j n
=1
=1
(6.49)
j n
(6.50)
j k
(6.51)
(6.52)
im
(6.53)
1
1
=1
119
m;1
=1
(6.54)
m;2
(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
=1
1
1
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.
121
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
11
21
up to the order of the terms in the disjunctions and the conjunctions. But this
does not matter for the present purposes.
^
^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
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
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)
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
(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
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
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.
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
130
6. Axiomatizability
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
i n
j=S H i `K ! (Hi ; H ) :
=1
n;1
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
134
6. Axiomatizability
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
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
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.
139
140
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
f0g : H1
f0; 21 g : H2
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
143
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
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
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
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 )
:
+ f^sj : Hij j (j; ^sj ) 2 M g M 2 T ('; s)
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 .
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.
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
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.
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
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
m
[
j =1
g(i; j )
for all i = 1; : : : ; m.
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
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
159
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
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
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) :
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
166
167
168
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
171
functionally complete.
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
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
H is mn provided
173
174
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
177
180
&
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
182
183
184
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.
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
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
+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
i=1
1
_
i=1
Hi =def H1 ;
Hi =def H1 ;
n^
+1
i=1
n_
+1
i=1
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
ValL(
i=1
188
;2
;kY
i=1
p; (k ; 2) 2 ; (k ; 2) + 1
= 1 ; (k ; 2) m;2 1 ; 2 = m;2
ValL :p ^
kY
;2
i=1
p; m;2 :
;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
Gn
i=1
(:p) ^ p Y
;
;1
;nX
i=1
;1
;nX
i=1
i=1
p :
(:p); = ValL p Y
nY
;2
i=1
p; = 1 :
and therefore
nX
;1
;
ValL
i=1
190
ValL p Y
;
nY
;2
i=1
p;
nY
;2
i=1
p; = 1 , (p) nn;;12 :
G2
3
X
i=1
(p & p) _
3
X
i=1
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
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 :
;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
+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
192
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
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
j=L H $L
;1
m
Y
i=0
(Jm;i (A) !L B ) !L B ;
;1
Y
;m
i=0
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
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
(9.28)
(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
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 :
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
m;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
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
200
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
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
n
X
i=1
ai xi g
202
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 +
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
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 )
+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
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
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
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 .
209
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
210
Dk , and that Dk is bounded. If, on the other hand, Q1; Q2 belong to dierent
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
211
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
s = bd +
n
X
i=1
ai ki
g(x1 ; : : : ; xn ) = b +
n
X
i=1
ai xI ;
213
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 ) .
10 And the two resulting Boolean algebras are not only isomorphic, but even
identical.
215
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 r^ole 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 :
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
A1 ^ : : : ^ Ak;1 ) Ak
with term equations A1 ; : : : ; Ak . This means, in algebraic terms, that the class
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
218
219
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)
13 The details of these denitions do not matter here. In general we assume that
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
222
a = a _ (a ^ b) = (a
:(a ^ b)) (a ^ b) 2 I ;
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
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
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
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)
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
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
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
231
The MV-algebras, thus, have for the Lukasiewicz system of innitevalued logic essentially the same r^ole 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 )) :
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
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
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.
indicates a compactness theorem for fuzzy logic mentioned later on in Section 19.
235
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)
237
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
240
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
242
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
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 :
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.
247
v w 5 (u v ) (u w ) :
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
249
;
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].
250
251
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
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 .
m;1
H1 ! H2 =def ! (H1 ; H2 )
=1
254
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
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 ; : : :
x1 ; x2 ; x3 ; : : :
256
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
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
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
b2jAj
0)
259
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
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,
262
HY
n
Y
i=1
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 , `
i=1
H weakly L1 -valid , `H
n
X
i=1
H for some n = 1; 2; : : : :
263
, `H H Y
n
Y
i=1
n
Y
i=1
H for each n = 1; 2; : : :
H for each n = 1; 2; : : :
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 ; ) :
265
ni
^
+ ValLB
k
^
i=1
H (G; ; Q) 8x8y
k
; ^
i=1
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.
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-
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
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
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) :
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
277
278
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.
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
283
284
285
8x H (x) !G H [x=t]
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
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
288
289
290
292
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
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
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
298
` ((B ! C ) ! (B A ! (B ^ C ) A))
299
(11.5)
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
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
; ! 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
302
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
304
305
306
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
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).
309
310
311
312
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.
314
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)
315
m
_
i=2
:H =def Dm (p) _
or even simpler
:H =def Dm (p) _
m
_
i=2
m
_
i=2
(12.4)
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
Contrary to the situation for the Lukasiewicz systems, for the Post
systems in their original form there exist only very few syntactically oriented
316
t(a) =def
nX
;1
k=0
ak ;
d1 (a) =def
;1
n
X
i=1
ai
n;1
;a =def
nX
;1
k=1
(dk (a))k ;
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);
318
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 :
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)
1 As a complement of
319
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) :
321
This result does not yet establish the relationship between P-algebras and
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
$ dened as
323
Dm;1 H
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-
8 Also further generalizations of the standard Post algebras have been considered
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
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,
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
fw t s j u t s vg fz j u t z v t wg :
H1 ; : : : ; Hn :
H
333
Proposition 13.1.7. For any left continuous t-norm t the following ws are
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
334
Proposition 13.1.8. For any left continuous t-norm t the following ws are
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 .
Proposition 13.1.9. For any t-norm t the following expressions are logically valid:
(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
337
338
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
Proposition 13.2.2. For any left continuous t-norm t the following ws are
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
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
341
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
t1 (x) =def x ;
343
344
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.
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
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
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
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.
353
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
355
`ML H1 ^ H2 ! H2 :
356
357
358
(a)
359
360
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 :
362
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
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
365
366
Proposition 14.1.17. Starting from the pre-linear monoidal logic one gets
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
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 ;
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 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
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
following assertions are equivalent:
(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
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
i2I
11 This means the (in case of existence uniquely determined) j 2 I with i @ j such
372
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
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).
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
376
i2I
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)
(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
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.
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
380
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
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
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
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
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
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
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-
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 ;
1
2
1
2
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 .)
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)
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
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
functionally complete.
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 :
interpretations for dierent systems of relevance logic. We will not discuss details
here. The interested reader may e.g. consult [431, 450].
399
= (1; 1)
s
;@
= (1; 0)
;
;
@
@
;
;
@
@
@
@
;
;
@s F = (0; 1)
;
@;
s
Fig. 15.3.
= (0; 0)
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].
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.
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) :
403
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) :
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
406
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
i=1
Hi =def H1 ;
kG
+1
i=1
Hi =def
k
G
i=1
Hi t Hk+1 ;
!G
(s;t)
r=(s;t)
Jr (P (x1 : : : yi : : : xn )) ;
(16.14)
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
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
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
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 r^oles
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).
413
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
415
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
417
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
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
424
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
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.
426
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 :
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)
^ ;
A;B;C
A \t (B \t C ) = (A \t B ) \t C
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 :
431
Proposition 18.2.3. Let the t-norm t be left continuous. Then for all fuzzy
432
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
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
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 ) :
equality of truth degrees of the left and right sides of the t -equations. They
can be suitably weakened if necessary.
2
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)
(v) j=
(vi) j=
i2I
\
i2I
(Ai t B )
Ai t B j
i2I
i2I
437
Ai t B ;
(Ai t B ) ;
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.
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
R; S; T
439
440
that
and
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
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
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.
443
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 :
445
446
447
448
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 ]
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
j 2I
Aj ] [ x " Ak ]
and therefore
;
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 ]
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 ] :
^ [y "
j 2I
Aj ] iinf
([[y " Ai ] [ x " Ai ] ) :
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
451
E0 (x;
y) (E0 (y; y) E0 (y; z ))
6 [ z " [ Aj ] iinf2I ([[z " Ai] [ x " Ai ] ) :
j 2I
j 2I
j 2I
Aj ] inf
([[x " Ai ] [ y " Ai ] ) [ x "
i2I
j 2I
[ x " Ai ] iinf
([[x " Ai ]
2I
j 2I
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
j 2I
Aj ] [ x " Ak ]
j 2I
j 2I
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
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.
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 :
453
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 ]
9 At present it is an open problem whether the results of the next proposition can
454
(ii) Again deleting all indices tG we have from Proposition 18.5.3(i) that
455
Proposition 18.5.7. For each fuzzy relation R and any left continuous t-
R =
i=1
Ri
i<
]
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
457
458
459
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
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
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
464
465
466
467
468
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)
2
[
i=0
R i jt R :
(18.32)
3
[
i=0
Ri jt R :
(18.34)
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)] ]
i=0
1
Y
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-
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
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.
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.
474
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.
475
476
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)
477
478
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
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 :
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
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.
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
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
(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.
(a) A C(A) ;
(b) A B ) C(A) C(B ) ;
(c) C(C(A)) = C(A) ;
487
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
489
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
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
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
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
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].
495
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
497
^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
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
499
500
501
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
504
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-
505
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
B A; H
(Exch) A B;H
0
507
n =def
W
n_
;1 _
n
i=1 k=i+1
(pi $ pk )
508
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
510
511
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
513
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
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 ; ).
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
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
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 r^ole denoted by T and F .
So the tableau calculus of [167] works only with signed bounding formulas
of the forms s : (^u ! 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 : (^u ! v^) for truth degrees u 66 v;
2. F : (^u ! v^) for truth degrees 0 6= u 6 v 6= 1 ;
3. F : (0 ! H ) ;
4. F : (H ! 1) ;
5. T : (^u ! H ) and F : (^v ! H ) for truth degrees v 6 u .
522
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 : (^u ! (H1 ! H2 ))
F : (w^ ! H1 ) + T : (w^ ! H2 )
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 :
(^u ! 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
;
;
; 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
524
u^ ! v^
u^ ! v^
A
2A
526
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
(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.
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)
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.
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)
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
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
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
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
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
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
537
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.
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
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
541
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.
(ZF5)
(ZF6)
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
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)
[ a =^ b] [ b =^ c] 6
547
(c(w) * [ w 2^ a] ) ;
w2dom(c)
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
548
u2dom(a)
^
u2dom(a)
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)
_
v 2V B
u2dom(a)
u2dom(a)
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
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
552
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)
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
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).
^
<
555
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)
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)]] :
<
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].
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
558
559
n+1
(Absn )
=1
=1
i n
x 2 an;H and ! (x 2 x; H )
=1
i n
(24.2)
n+1
i n
(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
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
[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
565
566
References
References
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
572
References
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.
Symbolic Logic 24, 97{106.
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
References
575
576
References
References
577
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.
Symbolic Logic 28, 77{86.
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.
Symbolic Logic 14, 159{166.
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
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.
Fuzzy Sets Systems 43, 319{326.
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,
Studia Logica 18, 139{164.
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.
Symbolic Logic 48, 900{901.
586
References
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,
Fuzzy Sets Systems 12, 237{248.
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.
Comp., Amsterdam, 143{149.
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,
Proceedings, IEEE Computer Soc., New York, 228{236.
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.
Logik Grundl. Math. 19, 111{114.
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,
Studia Logica 34, 45{58.
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
597
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
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
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
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