LOGIC, SEMANTICS,

METAMATHEMATICS
PAPE RS FROM 1923 TO 1938

BY

ALFRED TARSKI

T B A N SL A T K B BY
J. H. W O G D G E R

OXFORD
AT TH E CLAREN D O N PRESS
1956
PRINTED IN GREAT BRITAIN
 T R A N SL A T O R ’S PREFACE

T he setting free o f Poland after the First W orld W ar wa s fol­

lowed b y intensive activity in her Universities. In the depart­
ments o f philosophy and mathematics this took the form, in a
number o f places, o f new and powerful investigations in the
fields o f mathematical logic, the foundations o f mathematics,
and the m ethodology o f the sciences. Prominent in this m ove­
ment was the W arsaw school led b y Lukasiewicz, Kotarbinski,
and Lesniewski. Under their skilled guidance a younger genera­
tion grew up and among these Alfred Tarski quickly dis­
tinguished himself.
Ever since I first enjoyed the hospitality o f Professor Tarski
in Warsaw in 1936, it has seemed to me that the importance
and scope o f the Polish school o f logicians were insufficiently
known and appreciated in the English-speaking world. Then
came the Second W orld W ar, bringing ruin once more to Poland,
killing men, destroying laboratories, and burning manuscripts
and libraries. After this war it occurred to me that I should be
performing a public service, as well as acknowledging in some
small measure m y debt to m y Polish friends, i f I prepared a
collected edition o f some o f Professor Tarski’s publications in
English translation. W hen he visited England in 1950 to deliver
the Sherman Lectures at University College, London, I men­
tioned m y plan to him and received his approval.
This volume contains Tarski’s m ajor contributions to logic,
semantics, and metamathematics published before the Second
W orld W a r ; their arrangement here corresponds to the chrono­
logical order in which they first appeared in print. W ith the
exception o f articles I I and X I (which are too closely connected
with, and too often referred to in, the other articles to be
omitted), the volume does not include Tarski’s studies in the
foundations o f special mathematical disciplines— set theory,
group theory, etc. Neither does it contain his papers o f a
pronouncedly mathematical character, dealing with special
topics from the domain o f set theory, measure theory, abstract
viii T R A N S L A T O R ’ S PREFACE

algebra, elementary geometry, etc. Also excluded are short

notes, abstracts, and preliminary reports whioh are closely
related to some o f the articles included in the volume and the
contents o f which are more fully presented in these articles.
A longer paper from the domain o f logic and m ethodology
which has been omitted is cSur la method© deductive 5in Travaiuc
du IX e Congrls International de Philosophic (Paris 1937); this
paper is a purely expository one, and its ideas have been fully
developed in Chapter V I o f Tarski’s book Introduction, to logic
(New Y ork, 1941).
In a sense the present work is more than a volume o f transla­
tions. Naturally an attempt has been made to remove the
misprints and errors whioh occur in the originals. Moreover, the
articles have been provided b y the author with cross-references
to other articles in the volume and with notes referring to later
developments and recent literature. Occasionally some new
remarks have been added for the purpose o f clarifying certain
passages in the original text. Articles I I and V I contain more
serious changes, Tarski having inserted in them several passages
which, he hopes, will help to clarify and amplify their contents.
W hile the work o f translation was in progress, the passages
whioh seemed to me doubtful and difficult wer6 noted down and
sent to Professor Tarski in Berkeley, California. In this way it
was possible to adjust the text o f the translations in many
points, so as to meet the author’s wishes. However, in view o f
the time limit and the geographical distance between the resi­
dences o f the author and the translator, it was impracticable to
send the whole manuscript to Professor Tarski before it was
set in print. Instead he received galley proofs, on which for
obvious reasons he could not suggest too extensive changes. It
also proved impossible to discuss the suggested changes in
detail, and it was left to m y decision which changes were
actually to be carried through. Thus Professor Tarski is not
responsible for the final text or the technical aspect o f the book.
Three articles in this volume are joint publications o f Tarski
and other authors: Professor C. Kuratowski (article V II), Dr.
A. Lindenbaum (article X H I), and Professor J. Lukasiewicz
 T R A N S L A T O R ’ S PREFACE ix

(article IV ). W e are greatly obliged to Professors Kuratowski

and Lukasiewicz for their permission to include the translations
o f the jointly published articles in the volume, Dr. Lindenbaum
fell a victim to the Gestapo during the war.
The papers included in the volume originally appeared in the
following periodicals and collective works: Actes du Congres
International de Philosophic Scientifique (articles X V and X V I),
Comptes Rendus de la SociiU des Sciences et des Lettres de Varsovie
(articles H I and IV ), Ergebnisse tines mathematischen Kol-
loquiums (articles X I I I and X IV ), Erkenntnis (article X ), Funda-
menta Mathematical (articles I, V I, V II, X I, X U , and X V II),
Ksiyga Pamiqtkowa Pienvszego Polskiego Zjazdu Hatemaiycznego
(article II), Monatshefte fUr Mathematik und Physik (articles V
and I X ), Przeglqd Filozoficzny (articles I, X , X V , and X V I), and
Travaux de la Sociiii des Sciences et des Lettres de Varsovie
(article V III). Acknowledgements should be made to the pu b­
lishers and editors o f these periodicals for their generosity.
I am obliged to Mr. S. W . P. Steen for kindly reading the
proofs o f article VILl and throughout the work I have received
much help from m y son, Michael W oodger. W e are also indebted
to several colleagues and students o f Professor Tarski— Dr.
C. C. Chang, Professor A. C. Davis, Professor J. Kalicki, Mr.
R. Montague, Professor J. Myhill, Professor D. Rynin. and Mr.
D . Scott— for their assistance in revising the original text o f the
articles and in reading galley proofs.
Finally it is a pleasure to acknowledge the courtesy and help
which we have received from the staff o f the Clarendon Press at
all stages in the production o f the book.
J. H. W .
 A U T H O R ’S ACKN O W LED G EM EN TS

I t is a rare privilege for an author to see a volume o f his

collected papers published during his lifetime, and especially so
if the papers be translated into a single language from originals
in a number o f other languages. I cannot therefore but be
deeply m oved b y the appearance o f this volume and b y the
thought o f the m any and great sacrifices which its publication
has laid upon m y friend, Professor Joseph H. W oodger. For
five long years he has devoted to this work an immense amount
o f effort and time, which otherwise could have been used for
fruitful research in his chosen field, theoretical biology and its
foundations.
The task o f a translator is rarely a gratifying one. Circum­
stances have made it especially thankless in the present ca se;
let me indicate some o f them. The papers whose translations
constitute the volume were originally published over a period
o f fifteen years and in several different languages. They vary
considerably in subject-matter, style, and notation. Under
these conditions, the task o f combining the papers in one book
provided with a reasonable degree o f terminological consistency
and conceptual uniform ity presents extreme difficulties. In a
few cases (in particular, in the case o f the monograph on the
concept o f truth, which occupies nearly one-third o f the volume)
the translation had to be based not upon the original, which
was published in Polish, but upon the French or German version.
This made it even harder for the translator to give a fully
adequate rendering o f the original intentions and ideas o f the
author. In addition, due to the factors o f space and time, the
translator was deprived o f the benefit o f extensively discussing
with the author even the m ajor difficulties encountered in his
work, and so achieving a meeting o f minds before the text was
set up in type. T o illustrate this point I may mention that, for
various reasons, I have been unable so far to read a considerable
part o f the present text, and it seems more than likely that I
shall not have read it before receiving a copy o f the published
Zll AUTHOR’S ACKNOWLEDGEMENTS

book. The realization o f the difficulties involved makes me feel

all the more indebted to him whose initiative, devotion, and
labour have brought this volume into existence.

I t is needless to say that I fully appreciate the assistance and

consideration o f all the persons mentioned in the translator’s
preface. But I should feel unhappy if at this place I did not
make special mention o f the man who helped me more than
anyone else in m y part o f the job — m y younger colleague, the
late Jan Kalieki. He spent the last two years o f his short life in
Berkeley; he generously offered his help on the day o f his
arrival and continued it untiringly and patiently, with the
greatest devotion and conscientiousness, until his last day. He
studied the originals o f the articles, translated for Professor
W oodger various passages from the Polish text, prepared refer­
ences to recent literature, discussed with me the remarks which
I planned to insert in the translations, assumed most o f the
burden o f the extensive correspondence connected with the
publication o f the book, and read the first batch o f galley proofs.
His tragic and untimely death (in November 1953) was a cause
o f considerable delay in the publication o f this volume.
A. T.
University of California
Berkeley>August 1955
 CONTENTS

I. ON T H E P R I M I T I V E T E R M OF L O G IS T IC 1

II. F O U N D A T I O N S OF T H E G E O M E T R Y OF SO L ID S 24

III. ON SO M E FUNDAMENTAL CONCEPTS OF

M ETA M A TH E M ATIC S 30

IV. IN V E ST IG A T IO N S IN T O THE SE N TEN TIA L

CALCULUS 38
(by Jan L U K A S I E W I C Z and Alfred T A R S K I )

V. F U N D A M E N T A L C O N CEP T S OF T H E M E T H O D O ­
L O G Y OF T H E D E D U C T I V E S C IE N C E S 60

V I. ON D E F I N A B L E SE TS OF R E A L N U M B E R S no

V II. L O G I C A L O P E R A T I O N S A N D P R O J E C T I V E SETS
(by Casimir K U R A T O W S K I and Alfred T A R S K I ) 143

V III. T H E C O N C E P T OF T R U T H I N F O R M A L I Z E D
LANGUAGES 152
Introduction 152
§ 1. The Concept of True Sentence in Everyday or Colloquial
Language 154
§ 2. Formalized Languages, especially the Language of the
Calculus of Classes 165
§ 3. The Concept of True Sentence in the Language of the
Calculus of Classes 186
§ 4. The Concept of True Sentence in Languages of Finite
Order 209
$ 6. The Concept of True Sentence in Languages of Infinite
Order 241
§ 6. Summary 265
§ 7. Poateoript 268

IX . SOME O B S E R V A T I O N S ON T H E C O N C E P T S OF
cu-CO N SIST E N CY A N D ^ -C O M P L E T E N E S S 278

X. SOM E M E T H O D O L O G I C A L I N V E S T I G A T I O N S ON
T H E D E F I N A B I L I T Y OF CONCEPTS 296

XL O N T H E F O U N D A T I O N S OF B O O L E A N A L G E B R A 320

X II. FO U N D ATIO N S OF THE CALCULUS OF

SYSTEMS 342

X III. ON T H E L I M I T A T I O N S OF T H E M E A N S OF
E X P R E S S I O N OF D E D U C T I V E T H E O R I E S
(by Adolf L J K D E N B A U M and Alfred T A R S K I ) 384
xiv CONTENTS

X IV . ON E X T E N S I O N S OF I N C O M P L E T E SYSTEMS
OF T H E S E N T E N T I A L C A L C U L U S 393

XV. THE E STAB LISH M E N T OF SCIE NTIFIC

SEM AN TICS 401

X V I. O N T H E C O N C E P T OF L O G I C A L C O N S E Q U E N C E 409

X V II. SE N T E N T IA L CALCULUS AND TOPOLOGY 421

A B B R E V IA TIO N S 455

B IB L IO G R A P H Y 456

SUBJECT I N D E X 463

I N D E X OF N A M E S OF P E R S O N S 468

I N D E X OF S Y M B O L S 470
 I

ON THE PR IM ITIVE TERM OF

LOGISTICf

I n- this article I propose to establish a theorem belonging to

logistic concerning some connexions, not widely known, which
exist between the terms o f this discipline. My reasonings are
based on certain sentences which are generally accepted among
logisticians. But they do not depend on this or that particular
theory o f logical types. Among all the theories o f types which
could be constructed1 there exist those according to which my
arguments in their present form are perfectly legitimate.2
The problem o f which I here offer a solution is the following:
is it possible to construct a system of logistic in which the sign of
equivalence is the only primitive sign (in addition o f course to
the quantifiers8) ?
This problem seems to me to be interesting for the following
reason. We know that it is possible to construct the system o f
logistic by means o f a single primitive term, employing for this
purpose either the sign o f implication, if we wish to follow the
1 The possibility of constructing different theories of logical types is also
recognised by the inventor of the best known of them. Cf. Whitehead, A . N .,
and Bussell, B . (90), vol. 1, p. vii.
1 Such a theory was developed in 1920 by S. Le6niewski in his course on
the principles of arithmetic in the University of Warsaw; an exposition of the
foundations of a system of logistic based upon this theory of types can be
found in Le&niewski (46), (47), and (47 b).
* In the sense of Peiroe (see Peirce, C. S. (58 a), p. 197) who gives this name to
the symbols ‘ I T (universal quantifier) and ‘2* (particular or existential
quantifier) representing abbreviations of the expressions: ‘ for every significa­
tion of the terms . . and ‘ for some signification of the terms . ,

t B ibliographical N ote . This article constitutes the essential part of the

author’s doctoral dissertation submitted to the University of Warsaw in 1923.
The paper appeared in print in Polish under the title *0 wyrazie pierwotnym
logiBtyki’ in Przeglqd Filozoficzny, vol. 26 (1923), pp. 68-89. A somewhat
modified version was published in French in two parts under separate titles:
‘Sur le terme primitif de la Logistique*, Fundamental Mathematicae, vol. 4
(1923), pp. 196-200, and ‘ Sur les truth-functions au sens de MM. Bussell et
Whitehead’, ibid., vol. 6 (1924), pp. 69-74. The present English translation
is based partly on the Polish and partly on the French original.
2 ON THE P R IM IT IV E TERM OF LOGISTIC I

example o f Russell,1 or b y making use o f the idea o f Sheffer,2

who adopts as the primitive term the sign o f incompatibility,
especially introduced for this purpose. Now, in order really to
attain our goal, it is necessary to guard against the entry o f
any constant special term into the wording o f the definitions
involved, if this special term is at the same time distinct from
the primitive term adopted, from terms previously defined, and
from the term to be defined .8 The sign o f equivalence, if we
em ploy it as our primitive term, presents from this standpoint
the advantage that it permits us to observe the above rule quite
strictly and at the same time to give to our definitions a form
as natural as it is convenient, that is to say the form o f
equivalences.
The theorem which will be proved in § 1 o f this article,
[p,q]::p .q . == :.[/]:.*> s :[r].p s / ( r ) . ss .[r].q s / ( r ) «
constitutes a positive answer to the question raised above. In
fact it can serve as a definition o f the symbol o f logical product
in terms o f the equivalence sym bol and the universal quantifier;
and as soon as we are able to use the symbol o f logical product ,
the definitions o f other terms o f logistic do not present any
difficulty, as appears, e.g. from the following sentences:
[?]:• ~ 0P) ss :p s .[q].q,
[ p , q ] : . p o q . = :p ss .p.q,
i s s .
1 See Russell, B, (61 d)f pp. 16-18.
* See Sheffer, H . M. (63 a). See also Russell, B. (61 s).
8 In this article we regard definitions as sentences belonging to the system
of logistic. I f therefore we were to use some special symbol in formulating
definitions we could hardly claim that only one symbol is accepted in our
system as a primitive term. It may be mentioned that, in the work of White-
head and Russell cited above all the definitions bave the form 4a =» b Df. *
and thus actually contain a special symbol which occurs neither in axioms nor
in theorems; it seems, however, that these authors do not treat definitions as
sentences belonging to the system.
4 In this note I adopt the notation of Whitehead and Russell with some
slight modifications; in particular, instead of expressions of the form *<j>x9
I write *<f>(x)9; also the use of dots differs in some details from that in Principia
Mathematical
b Some further developments related to this result are contained in the
doctoral dissertation of Henry H it, ‘ An economic foundation for arithmetic*,
Harvard University, 1948; see also H it, Henry (31 a).
I ON TH E P R IM IT IV E TERM OF LOGISTIC 3

It will be seen from the discussion in § 2, that the results

obtained can be considerably simplified within a system o f
logistic which contains the following sentence among its axioms
or theorems: [ * , 5, / ] : f s , . f y ) . o , ( „ .

However, this sentence cannot be proved or disproved with­

in any o f the systems o f logistic which are known from the
literature. This gave rise to a further study o f the sentence in
question, and in particular to a. search for other sentences
equivalent to it. The results obtained will be presented in §§ 3-6
o f this work.

§ 1. F u n d a m e n t a l T h e o r e m

I start b y introducing a few definitions, Def. 1-3, which will

be used in this and the following sections. Then I give several
lemmas, Th. 1- 10, and finally prove Th. 11 which is the main
result o f this work.
The proofs offered are strictly speaking incomplete; they
must be regarded as commentaries indicating the course o f the
reasoning. The structure o f these commentaries is in part bor­
rowed from Whitehead and Russell; they do not, I think, require
more detailed explanation.

Dee. 1. [p] :^r(p) = .p ~ p

Dee. 2. [jp].&s(2>) s p
Dee. 3. [p] :fl{p) == .p == ~ {p)
For symmetry, one could also introduce the following defini­
tion:

D ee. 2'. [p].ngr(p) s ~ (p ).

However, this definition would be quite useless, since the symbol

‘ ngr would have the same meaning as the negation symbol
already occurring in logic.

T h . 1. 0 1 .to{p) (Def. 1)

T h . 2. 0 ] : [g] = tr(q) . o p
4 ON T H E P R IM IT I V E TER M OF LOGISTIC 1 .5 1

Proof.1 [p~\: . H p. 0 :
(1) p == .[g].fr(g):
Cn (1, Th. 1)
T h . 3. \ p ,q\ :p ^ .p = tr(q)
Proof. [p ]:H p .o.
(1) tr{q) (Th. 1)
Cn (Hp, 1)
T h . 4. M ••[?]•.? = M a) •= (Th. 2, Th. 3)

III

Hi
T h . 5.

is

u
Dp . a ] " [ / ] s :[d-3> =/(*■)•

2
2
fTV
r—l
£2j

Proof.
u

<l) P s :[r}.p == tr(r). = .[r].g = ir(r):.

(2) \r\.p == <r(r). = p : (Th. 4)
T’

(Th. 4)
&

(3)
III

II!

(4) p ~ .p ^ q : (1. 2, 3)
(5) (4)2
III

II!

(6) p=zp
Cn (6, 6)
T h . 6. [»]• ~ (Isi-p = « (a ))
Proof. Dp I-
(1) ~ (p = ~ ( p )}.
(2) ~ [ [ q ] . p s s g ). (1)
~ ([«?] ^ 33 « (e )) (1, Def. 2)
T h . 7. [p . g ] : [r] .p = as(r ) . s . [r ]. q — as(r)
1 In the proofs of theorems which have the form of a conditional sentence,
the symbol *Hp' stands for the hypothesis of the theorem, and the symbol
lCn' for its conclusion.
3 I pass from (4) to >;o) by making use of the following theorem which was
kindly communicated to pae by J. Lukasiewicz:
iP, 2 , r ] :. p s . g sj.sr.
This theorem, the proof of which is quite easy and is here omitted, expresses
an interesting property of equivalence which is analogous to the associative
property of logical addition and logical multiplication. -
X. §1 ON TH E P R IM IT I V E TER M OF LOG ISTIC 5

Proof. 0 ,g ] :

T
s
(Th. 6)

IT
¥
a.
in
( 1)

•
(2) ~ ( [ r ].2 = 05(r)): (Th. 6)
\f\.p — as(r). — ,[r ].8 = as(r) (1 .2 )
T h . 8. CP*ff] : : [ / ] : - s : W -P = / W - s - M - q = f ( r ) : . o p
Proof. [ p , q ] : : H p : . o :.
( 1) = : [r ].p =. as(r). = .\r\.q = a s (r ):.
Cn ( 1, Th. 7)
T h . 9. [p » 8] : ; [ / ] :•? = : M-:P =/(»•)• = -W -J =
z . p . q (Th. 8, Th. 5)
T h . 10. [p »?>/3 3 s ~ f ( r ) . =s .[r ].g s / ( r )
Proof. ! > , 8 , / ] " - f f p 3 :.
( 1) P
(2) 1 (Hp)
(3) jP = ? (1,2)
i—

HI
III

(3)
II!

(4)
(5) [ r ] :p s / ( r ) . = .q = f { r ) : (4)
(6) [r\.p = /(»• ). = .[r ].g = = / » : . (5)
On ( 1, 6)
T h . 11. \jp,q\v.p.q== : . [ f ] : . p = :[r].j> = / ( r ) .
= . [ r ] . q = f ( r ) (Th. 10, Th. 9)

§ 2. T r u t h - f u n c t io n s a n d t h e L a w o f S u b s t it u t io n
Whitehead and Russell, in their work mentioned above, refer
to a function / as a truth-function1 if it takes sentences as argu­
ment values and satisfies the condition
(a) 0 , 8] 3 /(g ).
The sentence
(A) [p,q,f] :p = q,f(p ) . 3 /(8 ),
which expresses the fact that every function / (taking sentences
a$ argument values) is a truth-function, will be called the 'law
of substitution\
1 A. JST. Whitehead and B. Russell (90), vol, 1, p. 659 ff.
6 ON TH E P R IM IT I V E TERM OF LOGISTIC I, § 2

In Def. 4 I introduce the sym bol ‘Bp’ ; by this definition, the

expression <Bp{f}> will mean the same as ‘the function / is a
truth-fiinction\ Def. 5 will enable us to replace the law o f
substitution b y a single sym bol *Sb\ Am ong the theorems
proved in this section, Th. 17 is the most important. I t shows
that, in a system o f logistic which contains the law o f substitu­
tion among its axioms or theorems, the sentence formulated in
Th. 11 can be replaced b y a simpler sentence,
[p ,q ]:.p.q. ss : [ / ] : p s ./(p ) s / ( j ) ,
which also can serve as a definition o f the symbol o f logical
product in terms o f the equivalence symbol and the universal
quantifier. From Th. 20 it follows that another closely related
aari f/iri OP*
[ p , q ] : . p v q . = : [ 3 / ] :p ss .f(p ) s=f(q)
can be used to define the sym bol o f logical sum.
The remaining theorems o f this section are o f auxiliary
character.
D e f . 4. [ / ] :. 8P{ f } = : [p, q] : p ~ q .f(p ) . of(q)

D e f . 5. Sb = .[f ].6 p { f )
T h . 12. [ / ] : . 8p{f} = -[p,q]:p = g.D ./(p ) = /(g ) (Def. 4)
T h . 13. 8b == : [ p , q j ] :p = q-f(p).of{q) (Def. 5, D ef. 4)
T h . 14. 8b = :[p , q,f] :p q. o ,f(p) ==/(g)
(Def. 5, Th. 12)
T h . 15. 8b = : . \ p , q , f ] : p . q . ? -f(p) s s fiq ) : .
[p,q>f ] ■~ (?)• ~ (?)• 3 -f(p) = /( ? )
(Th. 14)
The theorem just stated shows that the law o f substitution
is equivalent to the logical product o f two sentences, the first
o f which could be called the law of substitution far true sentences,
and the second one the law of substitution for false sentent&s.
I am unable to solve the problem whether either o f these two
sentences alone is equivalent to the general law o f substitution.
T h . 16. [ p , q , f ] : p . q . O .f(p) = /( g ) : D :.[p ,g ] :.p .g . e :
[ / ] -P » f(P)
I, §2 ON TH E P R IM IT IV E TER M OF LOGISTIC 7

Proof. H p : o :.:
( 1) 4 :.
(a) p.q:
(6' [ / ] • /( ? ) = /(?)• '• (Hp,a)
(e) [ / ] : p = ./(p ) a / ( g ) : . : (o, 6)

( 2) [P »ff]” [/]-* P s -/(P) * / t o ) =3 :•

r—i

**>

••
••
III

III
to)

•
(e) P = -M p ) = tr{q) : to)

(/) tr(p)s=tr(q). (Th. 1)

to) P: (e ,/)
(A) p = .as(p) == a s(g ): (d)
••

S'
III

III
•«

W
( j) p = p .s= q : (i )

(*) ?• 0 ‘)
(*) P -?:--* (?>*)
<7» (1- c , 2 - Z )
T h . 17. S&O :.[ p ,g ] : .p .g . a : [ / ] : p a ./(p ) a / ( g )
(Th. 15, Th. 16)
T h . 18. ~ (p )- ~ (?)• ^ * /(p ) s / ( 3 f ) : ^ •••[>>?]:•
~ (p) •~ (?) • = : [ / ] : ~ (p) a ./(p ) a / ( g )
(Th. 1, Def. 2)
I am omitting the p roof o f Th. 18 which is entirely analogous
to that o f Th. 16.
T h . 19. [ p . ? , / ] : ~ (p ). ~ to), o -f(P) ^ / ( ? ) ••3 •••[>.?]:•
p v g . a : [ 3 / ] :p a ./(p ) a f(q) (Th. 18)
T h . 20 . Sbo :.[ p , g ] : . p V g . a : [ 3 / ] : p a ./(p ) a / ( g )
(Th. 15, Th. 19)
The converses o f Ths. 16 and 19 can easily be established.
It is a direct conclusion that the sentence
!>><?] =-P -?- = : [ / ] :P s •/( p ) = / ( ? )
8 ON TH E P R IM IT I V E TERM OF LOGISTIC 1, § 2

not only follows from the law o f substitution for true sentences,
but is equivalent to it. Similarly the sentence
[ p , q ] : . p v q . s : [ 3 f ] : p = ,f(p) ==f(q)
is equivalent to the law o f substitution for false sentences, and
the logical product o f these two sentences is equivalent to the
general law o f substitution.

§ 3. Independence of th e Law of S u b s t it u t io n o f t h e
A x i o m s o f L o g is t ic ; t h e L a w o f t h e N u m b e r o f F u n c t io n s
T o the question whether every function / (having a sentence
as argument) is a truth-function, A . N. Whitehead and B. Russell
give a negative answer. Their answer is based exclusively on
intuitive considerations and does not appear to be convincing .1
On the other hand it seems quite clear that the law o f sub­
stitution can be neither demonstrated nor refuted in any o f the
systems o f logistic hitherto known. Moreover , it is even possible
to prove the independence o f this sentence o f the known systems
o f axioms o f logistic, e.g. the system o f W hitehead and Russell,
by using the method usually applied in proofs o f independence,
i.e. by means o f a suitable interpretation. W ithout giving any
details we m ay mention that such an interpretation can be
found in a system o f logistic based upon the three-valued
sentential calculus constructed by Lukasiewicz .2
In any case, anyone who regards the sentence (A) as true
and wishes to incorporate it in the system o f logistic, must
either admit it as an axiom or introduce another axiom which,
when added to the axioms o f the system, implies the sentence
(A). The theorems which will be established in the sequel can
have the same interest in the construction o f such a system o f
logistic as, for example, theorems concerning the equivalent
forms o f the axioms o f Euclid have for researches on the
foundations o f geometry.

1 It should be mentioned in this connexion that LeSniewski constructed a

general method which makes it possible to eliminate functions not satisfying
condition (a) from all known arguments. This method of Le&niewski has not
been published.
1 See article IV , in particular § 3.
 I, §3 ON TH E P R IM IT I V E TER M OF LOGISTIC 9

Some o f the theorems just mentioned are direct corollaries o f

logically stronger theorems in which conditions are formulated
that are necessary and sufficient for a function / to be a truth-
function. For instance, as will be seen from Th. 41, the condition

(ty b ] -ftp) = tr(p) ■v •bJ -ftp) •= «(p) •v . b ] -ftp) =

~ ( p ) - v . b ] - / ( p ) = fi(p)
is o f this kind. An immediate corollary o f Th. 41 is Th. 43
which states that the law o f substitution is equivalent to the
sentence

(B) [/] : b] -f(P) = tr(P) ■V •b] -f(P) 35 ai!(P) ■v •bl •

f(p) = ~ (p)- v -b]-/b) ^ b ) -
This sentence, owing to its intuitive content, could be called
the law of the number of functions; it expresses the fact that
every function (which has sentences as arguments) is equivalent
for every value o f the argument, to one o f the tour functions
tr, as, and ft.
D e f . 6.1 Tr = .[3 r ].r
D e f . 7 .1 jF 7 = = .[r].r
DBS’. 8. [ / , 0] : = { / , g) ss . b J ftp) -= g(p)
T h . 21 . Tr (Def. 6)
T h . 22. [ p ] : p = . p ~ Tr (Th. 21)
T h . 23. ~ (FI) (Def. 7)
T h . 24. b ] : b ) = -P == FI (Th. 23)
T h . 25. [p ] :P = Tr. v .j> == FI
Proof, b ]:
( 1) 2> V ~ ( i> ) :
( 2) p o . p = Tr (Th. 22)
(3) ~ (p) o .p =s FI; (Th. 24)
p == Tr. v .p = FI (1, 2 , 3)
Th . 26. [p]. ~ ( p = T r . p z s FI) (Th. 21, Th. 23)
Th. 27. b * /]:W } -ftTr) .f(Fl). of(p)
1 I writ© ‘ Tr’ and *FV instead of ‘ 1* and ‘ 0* which occur, for example,
in Sohrdder, E . (62), vol. 1, p. 188.
10 ON TH E P R IM IT IV E TERM OF LOGISTIC I. §3

Proof. 1
(1) p == Tr. v .p = FI: (Th. 25)
(2) p = T r .D /(p ): (Def. 4, Hp)
(3) p == FI. o f ( p ) : (Def. 4, Hp)
Gn (1, 2, 3)
T h . 28. [ f ] : e P{f}.f { T r ) .f ( F l ) .^ = {f,tr}
Proof. [f]:.H p .o:
(1) [p]-f(p)- (Th. 27)
(2) [p]-f(p) = H p ) ■ (1, Th. 1)
Gn (Def. 8, 7)
T h . 29. [ / ] : W ) Ji.Tr). ~ (f(Fl)). o — {/, a*}
Proof. [fV.'.Hp.D:.
(1) W :
(a) P D.
(<*) p = Tr. (Th. 22)
03) f(p )' (Def. 4, a, Hp)
(b)
(y) p = P i. (Th. 22)
(S) (/(*>)): (Def. 4, y, Hp)
(«) P = /(p )- (a—j8, b— 8)
(d) <w(p) —fip) (Def. 2, c)
Gn (Def. 8, 1—d)
T h . 30. [ f ] : e P{ f } . ~ ( f ( T r ) ) . f ( F l ) . D = { f
(Th. 22, Def. 4, Th. 24, Def. 8)
I omit the proof which is analogous to that o f Th. 29.
T h . 31. 0 ] . ~ (flip)) (Def. 5)
T h . 32. [ p j ] : W }• - (/(T r)). ~ {f{FI)), d - (/(p ))
(Th. 25, Def. 4)
The proof is analogous to that o f Th. 27.
T h . 33. [ /] : W } - - (f(Tr)). ~ (/(J 7 )). d = {/,./*}
(Th. 32, Th. 31, Def. 8)
The proof is analogous to that o f Th. 28.
I, §3 ON THE P R IM IT IV E TERM OF LOGISTIC 11

Th . 34. [ / ] : 6P[f i 3 • = {/. *»■} V ={/> as} V

*=» {f, V =* i f f i i
Proof.
( 1)
f{Tr).f{Fl). v , f { T r ) . ~ (f(Fl)}. V
- (f(Tr)).f(Fl). v . - (/(3 V )). - {/{FI)):
On (£fp, 1, Th. 28, Th. 29, Th. 30 Th. 33}
T h , 35. 6p{tr}
Proof.
( 1) \jp,q] :p =5 q.trip ) • tr(q) ■. (Th. 1)
6p{tr} (Def. 4, 1)
T h . 36. eP{as}
Proof.
( 1) *
(«> p — q.p.Dq:
(b) as(p) = p . (Def. 2)
(e) as(q) s= q : (Def. 2)
(d) p ~ q.as(p). Dcs(?):. {a, 6, c)
0/>{<m } (Def. 4, 1— d)
Th . 37. (Def. 4)
T h . 38. (Th. 31, Def. 4)
The proofs o f the last two theorems are analogous to that o f
Th. 35.
T h . 39. [/>?] : ^ {?>/}• W } - ^
Proof.
( 1) [P>2] ;
(a) p == g ./( p ) , d /(v >• (Def. 4)
(b) £(P) = ,% )• (Def. 8, J?p)
(e) 2 (2 )s /(2 ): (Def. 8, ffp )
(<*) p q.gip).og{q} :. (a,b,c)
Cn (Def. 4, 1— d)
 12 ON TH E P R IM IT I V E TERM OF LOGISTIC 1. §3

T h . 40. [ / ] : == { / , tr} v = { / , ew} V = { /, V

= {f,M- o ¥ J )
Proof. [f]:H p .o .
( 1) = { f , t r } 3 6P{ f } (Th. 39, Th. 35)
( 2) = { / . ^ } D 0p { / } (Th. 39, Th. 36)
(3) = { / , ~ } d BP{ f } (Th. 39, Th. 37)
W - U J} 3 W f) (Th. 39, Th. 38)
Cn (Hp, 1, 2, 3, 4)
T h . 41. [ / ] •H I ) s •- { /, tr) V = { /, a ,} V - { /, - } V
= (Th. 34, Th. 40)
T h . 42. 8h ~ . [ / ] . = { / , tr} V = { / , as} V = { / , V

=-- i f ft) (Th. 41)

T h . 43. Sb = : [ / ] :[p].f(p) ss lr(p) . v . [p].f(p) s as(p).
v •[p] ■f(P) S3 ~ (p) •V .[p ] J(p) = fl(p)

§ 4, T h e L a w o f D e v e l o p m e n t

The most important theorem o f this seotion is Th. 60, where

I give a new necessary and sufficient condition which a function
must satisfy if it is to be a truth-function. According to this
condition the function f should possess the following property:
(c) j> ] : ,f(p) ~ :/(7V ) . p . V .f(Fl). ~ (p).
Th. 51, which follows from Th. 50, establishes the equivalence
o f the sentence (A) and the sentence:
(C)[p ,n : .f(p) s= :f(Tr) . p . V .f(Fl). ~ (p):
in this last sentence it is easy to recognize the law of development
well known in the algebra o f logic .1
T h . 44, [ p j J : •°P\f} f(p) ■^ -f(Tr) . p . V .f(Fl) . ~ p
Proof. [pJ\:.H p.o :
( 1) p ~= Tr. V .p=s FI: (Th. 25)
(2) p = T r. d -f(Tr).p : (Def. 4, Hp, Th. 22)
(3) p s F I . d .f(Fl) . ~ ( p ) : (Def. 4, Hp, Th. 24)
Cn ( 1, 2, 3)
1 Cf. L, Couturat (14 a), § 29; Schrader, E. (62), vol. I, p. 409, 44+.
1,5 4 ON TH E P R IM IT I V E TERM OF LOGISTIC 13

T h . 45. [p>f\ •W ) -f(Tr) -P •Df(P) (Th. 22, Def. 4)

*n
(Th. 24, Def. 4)
A

i—i

/"•s
T
3
Th . 46.

*
Th . 47. ! > , / ] :. M / } -f(Tr) -P ■V .f(Fl) . ~ (p) : z>f(p)
(Th. 45, Th. 46)
Th . 48. [ / ] : : Op{f}o : . [ * ] : -f(p) s if(T r).p . V .f(Fl) . ~ (p )
(Th. 44, Th. 47)
T h. 49. [f}::[p]:.f(p)
m ;f ( T r ) . p . V .f(Fl) . ~ (* ):. a 6p{f}
Proof. U ]:.:H p :.o ::
(1) [p,q\-..p s q.f(p ) . 0 :
<*) P = ?•/(?) =
(b) /(T r ).p . V .f(Fl). ~ i p ) : (Hp, a)

(«) f{T r).q .V .f(F l).~ (q ): (a, b)

(d) /( ? ) :: (Hp, c)
Cn (Def. 4, l - d )
T h. 50. [ / ] : : W } S :.[p ]:./(p )
= :/<ZV).j>. V ./(F J). ~ ( p )
(Th. 48, Th. 49)
T h. 51.
= :f(Tr).p. V ./(F I). ~ (p) (Th. 50, Def. 5)

§ 5. T he F irst T heorem on the B ounds of a F unction

As well as satisfying the law o f development, in virtue of Th. 50,
truth-functions also possess other properties which the algebra
o f logic attributes to its functions. Nevertheless the correspond­
ing theorems are sometimes easier to prove by the use o f the
specific laws o f logistic.
In this section I prove, in particular, Ths. 52-54 according
to which all truth-functions satisfy the following conditions:
(d) [ p i m - ® .f(Tr).f(Fl)
(e) [lp].f(p). = ./(Tr)vf(Fl)
U ON TH E P R IM IT I V E TERM OF LOGISTIC I, § 6

or, in equivalent terms:

(d') [p ] :f(Tr).f(Fl).o.f(p).
(e') [p] .f(Tr)vf(FI)
and also the condition:
(f) [p] :f(Tr) , f(Fl ). of(p) o ./(ZV) v / ( i 7 ) .
The theorems mentioned only express the conditions which
are necessary in order that a function / should be a truth-
function. It is impossible to prove that these conditions are
sufficient .1 It is, however, possible to prove that each o f the
sentences which attributes to each fu n ctio n / (having a sentence
as argument) the properties (d), (e), or (f ),2 implies the law o f
substitution. This has enabled me to establish, in Ths. 68, 62,
and 64, the equivalence o f this law and the following sentences:
fl>) [ f ] - [ p ) J ( P ) - ^ ■KTr).f(Fl),
(E) lf]:\3p].f{p). = .f(Tr)vf(Fl),
(F) [P,f] :f(Tr) .f(F l). z>f(p) o .f(Tr) vf(Fl).
The sentence (F) is known in the algebra o f logic .3 I call it
the first theorem on the bounds of a function . Sentences (D)
and (E) result from it almost immediately.
T h . 52. [ f ] : . e P{ f } o :[p].f(p). = ,f(Tr).f(Fl)
Proof. [f]:.H p o:
(1) &].f(p).z>.f{Tr).f(Fl):
( 2) f ( T r ) .f ( F l ). D . [ p]-f(p ) : (Th. 27)
Cn ( 1, 2)
T h . 53. [ / ] : . dp{f] o : [3\p] .f(p ) . = .f(Tr) yf(Fl)
1 On© con even prove that the sentences expressing the sufficiency of these
conditions are independent of the axioms of logistic; as in the case of the law of
substitution, the proof is based upon an interpretation in three-valued logic
(cf. the remarks at the end of § 2).
3 Lukasiewicz has already called attention to the fact that in attributing
to every function / the properties indicated here as conditions (d) and (e) (and
the properties corresponding to these conditions for functions of several argu­
ments) we can bring about notable simplifications in the constructions of
logistic. Cf. Lukasiewicz, J. (50 a).
3 Cf. L. Couturat (14 a), $ 28; E. Schroder (62), vol. I, p. 427, 48+.
I, $ 5 ON TH E P R IM IT I V E TERM OF LOGISTIC 15

Proof.
a) ~ (/(2 V )). ~ (/( * * )). 0 - M - ~ ( / ( ? ) ) : (Th. 32)
(2) P r i./(P )-3 -/(2 V )v /(^ ): (1)
(3) /(T V J v /fF Q .D .p p j./C p ):
Cn (2, 3)
T h . 54. [ f ] : . 0P{f}z> : b ] :f{Tr).fKFi). of(p) o .f ( T r ) v f ( F l )
(Th. 52, Th. 53)
T h . 55. ::[ /] :!> ] • /( ? ) • 25 -f(Tr).f{Fl) :.g(Tr,Tr).
. flr(-2V, F i ) . ?( Fi, T r ). g( FJ, FJ) : . o g(q, r)
Proof. [q,r,g]::E p:.D :
(1) [p].g(Tr,p). ~ .g(Tr, Tr).g(Tr, FI):
( 2) [p].g(Tr,p): (1, Bp)
(3) g(Tr,r): ( 2)
(4) [ p ] . g ( F l , p ) ~ .g(Fl,Tr).g(Fl,Fl)-. {Bp)
(5) [p].g{Flrp): (4, Hp)
( 6) g(Fl,r): ( 6)
(?) {p\.g($,r). == ,g{Tr,r).g{Fl,r): {Bp)
(8) b>]-g{p,r)- (7, 3, 6)
Cn ( 8)
T h . 56. [ / ] :{p].f(p). = .f(Tr).f(Fl) :o
Proof. H p : o ::
( 1) [/]:•
(a) Tr ~ Tr.f(Tr). i f ( T r ) :
(6) ~ (T r == F i ) : (Th. 21, Th. 23)
(c) Tr ~= FL A Tr ) . ::,f(Fl): (6)
(<*) FI s Tr J (FI). j f { T r ) : (b)
(«) F l ~ FI,f (FI), o f (FI):.
(/) [p ,q}:p z~ q.f(p).of(q):: (Th. 55, Hp, a, c, d, e)
Cn (Def. 4, 1—/ )
T h . 57. Sb = : [ f ] : [ p ] .f ( p ) . = .f(Tr).f(Fl)
(Th. 52, Th. 66)
16 ON THE P R IM IT I V E TERM OF LOGISTIC 1,5 6

Th . 58. Ip,q,f ] :p == q.f(p). o f ( q ) : ~ : [ / ] :[ p ] ./( p ) .

~ . / ( Tr).f(Fl ) (Th. 57, Def. 4)
T h . 59. [ /] : [2p] .f(p) . - ./(TV) vf(Fl) : d: [ /] : [p] .f(p).
f(Tr).f(Fl)
Proof. Hp : .

(1 ) I f ] : P i>] •- (f(P)) ■ = •~ (M 'r)) V - ( f ( F l)) : .

On (1)
T h . 60. [ / ] : [ > ] f(p) ■= -f(Tr) .f(Fl ) : o : [ / ] : [3pJ ./( p ) .
- J{Tr)vf{Fl)
Th© p roof is analogous to that o f the preceding theorem.
T h . 61. [ /] :[ 3 j> ] ./( p ) . s ./(TV) v f ( F l ) : = : [ / ] : [p] ./( p ) .
S3 ./(TV) ./(FT) (Th. 59, Th. 60)
T h. 62. Sb - : [ / ] : pp] ./(p). s .f(Tr) vf(Fl)
(Th. 57, Th. 61)
Th. 63. Sb s : [ / ] : [p ] .f(p ) . - ./(TV) ./(F T ): [Bp] .f(p ) .
-/(TV) vf(Fl) (Th. 57, Th. 62)
T h. 64. Sb = : [ p , / ] :/(TV) .f(Fl ) . of(p) o ./(TV) v / ( Fi)
(Th. 63)
In addition to the properties o f a function / which have been
examined in the course o f this section, it is possible to study
other properties which are stronger from the logical point o f
view:
(g) [?] : M -f(P) • = •/(?)
(h) [ g ] : [3p] .f(p ) . = ./(g) v /( ~ ( g ) ) ,
or, in equivalent formulations:
(g') [p, ? ] : / ( ? ) - / ( ~ (?)) •=>/(p),
(hO [P» ?] -/(F) •^ •/(?) v /( ~ ( g ) ) ,
and the property:
(i) |>>2] :/ ( 3 ) •/(■•'' iq))-3f(p}~-' /fc> v / f — (?)).
B y applying analogous methods to those which were used
above, the following theorems are easily proved:
T h . 65. [ / ] : . 0 p {/} d : [ g ] : [p ] ./ ( p ) . ss . / ( g ) . / ( ~ (g))
L 1.5 ON TH E P R I M I T I V E TER M OF LOGISTIC 17

T h . 66. E : [ ? ] : [ 3 p ] . / ( p ) . - • / ( ? ) v / ( ~ (q))
T h . 67. [ / ] : J p {f) 3 =[P> ?] •'/<?) • / ( - (?)) •3 m ^ •/(?)
v / ( ~ (2))
T h . 68. Sb = :[q,f]:[p].f(j>). s ,/(<?) • / ( - (q)\
T h . 69. Sb = :[q,f]:[3p].f(p). = .f(q)V / ( ~ (q))
Th. 70. 8b == :[p,q,f] ♦/(— <2))- ^f(p)^ -/(2) v / ( ~ <2))

§ 6 . T he Second T heorem on th e B ounds of a F unction

I shall now show that every truth-function possesses the
following properties (see Ths. 71-73):
(i) [p]-f(p)- = / ( / ( ^ ) ) »
(k) [3p].f(p). = f(f(Tr)),
or, in equivalent terms:
(i'J b>l-f(fm )of(p),
(k') b?].f(p)of(f(Tr)),
as welhas the properly:
(l ) b ? lf(fm )^ f(P )
These conditions, like those o f § 5, are only necessary in order
that a function / should be a truth-function. Moreover, in
contrast to the situation in the preceding section, I am unable
to show that one o f the following sentences:
w [f]-[p ]-f(p )-^ M m
(K ) [f]:[lp ).f(p).^ .f(f(T r)),
according to which every function / possesses the properties
(j) and (k), implies the law o f substitution.
N ow we shall see that this law results from the sentence:
<io [ p , f ] - f ( f m ^ f i p ) ^ f ( m ’r)h
which is equivalent to the logical product of the sentences (J)
and (K). Thanks to this result I obtain, in Ths. 79 and 80, new
formulations, equivalent to the sentence which interests me in
this work.
18 ON TH E P R IM IT I V E TER M OF LOGISTIC I, § 6

We may call the sentence (L), which is also known in the

algebra o f logic,1 the second theorem on the bounds of a function.
T h . 71. [/ ] : . W } 3 :[p ]-/(p ). = /(/(* ? ))
Proof. [ / ] ::Hp d ::
(1) W : -fiP) s :f(Tr).p V .f(Fl) . ~ ( p ) : : (Th. 48)
(2) f(f(Fl)) = :/(2V ) .f(Fl ) . v . / ( J i ) . ~ (f(FQ ) :. ( 1>
(3) f(f(Fl)) = .f(Tr).f(Fl): (2)
On (Th. 52, J3p, 3)
T h . 72. [ / ] : . 6P{ f ] d : [3p] .f(p) . == /(/(T V ))
(Th. 48, Th. 53)
The proof is analogous to that of the preceding theorem.
T h . 73. [ / ] : . dP{ f } o . [p] .f(f(Fl)) o /(p ) of(f(Tr))
(Th. 71, Th. 72)
T h . 74. [ /] :[ > ] .f(p) . = f ( f ( F l } ) : o : [q,f] : ~ (?) o .f(q) ■

Proof. U p : d :::
(1) [ 2 ./] :.:
(«) P 2] : :
(«)2 |> ] : -2(3>) = : ~ (p) o .f(p) = f ( F l )
(J3) g(Fl) ~ : ~ (FI) o .f(Fl) = f ( F l ) :. («)
(y) f(Fl) = f ( F l ) :
(8) ~(Fl)o.f(Fl)==f(F l) (y)
(«) g(Fl):. 08,8)
a) g(g(Fl)) == : - (g(Fl)) o .f(g(Fl)) = f ( F l ) : . («)
( V) ~(g(Fl))o.f(g(Fl))=sf(Fl): («)
m
w I> ]- 2 ( P ): (S p , 2)
(*) 2 (2 ):• : (0
1 Cf. L. Couturat, (14a), § 28 (Remarque).
* The auxiliary definition which I introduce at this place and of which
I make use in the proof, may seem superfluous. But I have adopted this as
a device for making the proof clearer.
1.1 6 ON T H E P R IM IT I V E TERM OF LOGISTIC 19

(b) ~ (g) D ./(g)= f ( F l ) ::: { * - k)

On (1-6)
T h. 76. [ / ] : [p],f(p) . == f(f(Fl)) : o : [ p , q , f ] :
~ (jp ).~ (q).f(p).of(q)
Proof. Hp : d :.
(1) [P, 2 , / ] : ~ (p) ■~ (g) -/(jp) • •
(а) f(p) = f ( F l ) ./(g) ^ f ( F l ) . f ( p ) (Th. 74)

(б) / ( p ) = /(2 )-/(P ) («)

(c) /( g ):. (6)
<?» (1—c)
T h . 76. [ / ] : [3p] ./(p ). = f { f { T r ) ) : o : [ g , /] : g o ./(g)
= f(Tr)
Proof. Hp : d : . :
( 1) M ~ (2 (P » • = •~ {(fteiTr))) ■•:
(2) fc /]::
(а) P 2] :.
(a )1 [p ] : 2<P) = - P - ( / ( p ) = / ( T r ) ) :.
03) g(Tr) = .T r.~ (/(2 V )= = /(2 V )): («)
(y) /(2 V )= /(2 V ).
(*) ~ (2(2V)): (ft y)
(«) g(2(^)) = -2(Tr). ~ (/(g(Tr)) s /(2 V )) : (a)
(0 ~ (2(2( ^ ) ) ) : (e, ft
(ij) [>]• ~ (g(p)): (1 ,0
(ft /'* '( 2(2) ) :: (ft
(б) ~ (g. ~ (/(g) = /(2V))):: (a—a, ft
<7» (2— 6)
T h . 77. [ /] :[ 3 p ] ./( p ) .= /( /( 2 V ) ) :
3 : [P>2»/] :P -2 -f(P) •'->/(2) (Th. 76)
The p roof is analogous to that o f Th. 76.
T h . 78. [ / ] : O ] ./(p ). « f ( f ( F l ) ) : [3p] ./(p ).
^f(f{T r)):o.[f].6p{f)
1 See the preceding note.
20 OI\T TH E P R IM IT I V E TERM OF LOGISTIC I* § 6

Proof. Up : d ::
(1) [}y>q . f ] : . p S £ g ./(p ).3 :
(a) P = q-j'(P) ■
(b) p.q. V . (p). ~ (q): (“ )
(c) p.q .ofiq ): (Th. 77, U p , a)
(d) - ^ ( p ) . ~ (q). (Th. 75, U p , a)
(e) /( ? ) :: ip, c, d)
On (Def. 4, 1—e)
Tn. 79. ,% hh : [ / ] :{p].f(p). rzf{f(Fl))-.[3p].f(p).
~ W T r ) ) (Ths. 71, 72, 78}
T h . 80. *b m ,[p,f].f{f(Fl)) o f ( p ) - f ( f ( T r ) ) {Tli. 79)

N o t e I. On F u n c t io n s of S everal A rgum ents

When we examine functions o f several arguments (all these
arguments being sentences) we reach results which are analogous
to the preceding. Their proofs present no essentially novel
features.
For example, if we limit ourselves to functions o f two argu­
ments we can establish the following:
In order that such a function Vf mayV be a truth-fun etioiu i.e.
in order that it should satisfy the condition:
(%) [?-• ?» '-*>*] :p-szq.r== s.f{p, r). z, ,f(q, s),
it is necessary that it should possess the following properties:
•i) b > ? ]-/b > ? )- v :
b > ? l :/ b > ? ) = ■ p vq:\ r-[p ,q ]:f(p ,q ) s . ~ ( p ) V f f :V :
b > ? ] :/ ( * > ,? ) = .p v ~ {q ):V :[p ,q ]:f(p ,q )
=h . ~ (j)) \/ (g ): v .

b > ? 3 -f (p>q ) = p . V . [ p , q ] . f ( p , q ) S3 q . V : b > q] : f ( p , q )

==.p~q: V :

[p a ] :f { p , q ) s ~ (? ): V .[p ,q].f(p ,q) =

~ (?)• V . f j p , q ] . f ( p , q ) = ~ (i>). V :
b > ? ] ;/ b > ? ) = .;p .g :V :b » ? ]:/b > ? ) = -i> .~ (? ):V :
lp ,q ]:f(p ,q ) = •~ ( p ) . g : V : b . ? ] :/(2 > »?)
= . ( p) . ~ (? ): V .
b »?]« ~ {f(p>q))i
I ON TH E P R IM IT I V E TERM OF LOGISTIC

(Ci) 31 '■•f(P>?) ^ -f(Tr, Tr) .p.q. V .f{Tr, F I ) . p . ~ ( q ) .

vf(Fl, Tr). ~ ( p ) . q . V .f(Fl, FI). ~ (r ). ~ Iq),
(di) U>, ?] •/(?, 2) - s ./(T r, Tr) ./(Tr, 27) .f{Fl. T r ) F I ) ,
(ex) [3P,9}-f(p,9) • = -f(Tr. Tr) Vf [ T * 27) v/{27, Tr) V
/(27, 27),
(fi) [p, 2l Tr) ./(Tr, 27) .f(Fl, Tr) .f(Fl, FI). of(p, q) d .
./(Tr, Tr) v /(T r, 27) V/(27, Tr) v/(27, 27).
Conditions (bx) and ( c j are necessary and sufficient in order
that a function/should be a truth-function, rhe others are only
necessary.1
We can similarly prove the equivalence of the sentences (AO
the law of substitution. (BO the law of the number of f unctions,
(Ci) the law of development, (D;)> (E^, and (FO—the theorem on
the bounds of a function, which attribute to each function / the
properties (a1)-(f1) mentioned above.2
It is easy to show that the sentences (A1)-(F1) are equivalent,
not only among one another, but also to the sentence (A)— the
law" of substitution for functions with a single argument. More
over, without leaving the domain of functions with two argu­
ments, we can again construct a series of sentences equivalent,
to the sentence (xA.) and, from the point of view of content,
intermediate between the sentences of the preceding sections
and those we have just discussed. The following are some
examples:
(AD [p ,q,r,f ] :p = q.f(p,r). z>f(qt r),
(DD [ » , / ] : [q] .f(p, q). = ./(p, Tr) .f(p, FI).
1 I do not cite here the conditions corresponding to conditions (j), (k)> and
(1) which were studied in the course of § 6. because they will be rather compli­
cated and do not seem to me to be very interesting.
2 In an unpublished manuscript of Lukasiewicz I have found an argument
which may be summarized in these terms: The laws of the number of functions
result from the following hypothesis:
**J{Tr). V dip) ^ f ( F l )
and from analogous hypotheses for functions of several arguments. It is easy
to prove that each of these hypotheses is equivalent to the law of substitution
(for functions with the same number of arguments).
22 ON THE P R IM IT IV E TERM OF LOGISTIC I

N ote II. O n F u n c t io n s o f w h ic h t h e A r g u m e n t s
are not Se n ten c e s
Analogous problem s present themselves in the study o f logis­
tic functions o f which the arguments are not sentences but
functions.
I shall restrict m yself to the consideration o f only a single
case, to be specific that o f a function o f one argument which is
itself one o f the functions studied in §§ 3-6.
B y analogy with the term inology o f W hitehead and Russell,1
we call such a function <f> an extensional function if it satisfies
the condition:
(**) [/> s ] ■ = {/. 9} • 3 •
The sentence
(b2) =
which attributes to every function <f> the property (a2), is
probably independent o f the axiom s o f logistic, even if we add
to them the sentence (A ). Nevertheless I do not know o f a
p roof o f this.
It is possible to form ulate a series o f theorems (analogous to
those o f the preceding sections) which express the necessary and
sufficient conditions which a function (f> must satisfy if it is to
be extensional or which provide equivalent form s for the sen­
tence (A 2). However, the proofs o f these theorems require that
the sentence (A) be adm itted as hypothesis. Otherwise they are
quite like the proofs o f §§ 3 -6 ; but the role analogous to that o f
Th. 25 is here played by the sentence:
[/]• * = {/» « « } V = - {/ ,$ ,
which we already know to be equivalent to our hypothesis (see
Th. 42 o f §3 ).
I here give the m ost characteristic sentences which are
equivalent to the sentence (A 2):
(C2) [/:<£]:.<£{/} = : == {/, tr).<f>{tr}. V . = {/, as).J>{as). v .
= {/> -}.# -}• v . =
P 2) M :[/]•#/}• s
1 Whitehead and Ruasell (90), vol'. I, p. 22.
I ON TH E P R IM IT I V E TER M OF LOGISTIC 23

(Ea) M ' - m ■*{/}■ ® .<f>{tr}V 4>{a$ ) V V

(Ea) [ /, <f>] ■ . <f>{as} . •3 4>{J) 3 •<Htr} V
<i{cs} V $ {~ } V
(M) [<f>] :.[3g ] : [ / ] . # / } ^ g(f(Tr),f(Fl)).
The sentences (C2)-(F 2) correspond, strictly speaking, to
the sentences (C )-(F ) o f §§ 2 and 3, or o f N ote I.
B ut the sentence (M) has no correlate in the preceding series o f
sentences. It m ay perhaps be added that we can easily derive
from (M) the sentence (B 2)— the law of the number of functions.
For technical reasons I refrain from giving this law here; it
would contain sixteen logical summands and its form ulation
would be very laborious w ithout the help o f some auxiliary
definitions.
 II

FO U N D A TIO N S OF TH E GEOMETRY
OF SO LID S!

Some years ago LeSniewski suggested the problem o f establish­

ing the foundations o f a geometry of solids, understanding by
this term a system o f geom etry destitute o f such geom etrical
figures as points, lines, and surfaces, and adm itting as figures only
solids— the intuitive correlates o f open (or closed) regular sets
o f three-dim ensional Euolidean geom etry.1 The specific charac­
ter o f such a geom etry o f solids— in contrast to all point g eo­
m etries— is shown in particular in the law according to which
each figure contains another figure as a proper part. This problem
is closely related to the questions discussed in the works o f
W hitehead and N icod.2
In this resume I propose to sketch a solution o f this problem
om itting the question o f its philosophical im portance.
The deductive theory founded by S. Lesniewski and called by
him mereology* w ill be essentially involved in our exposition.
In this theory the relation of part to whole (which is treated

1 By a regular open set, or an open domain, we understand a point set

which coincides with the interior of its closure; by a regular closed set, or a
closed domain, we understand a point set which coincides with the closure
of its interior. These notions have been defined by Kuratowski, C. (42) and
(41), pp. 37 f.
* See Whitehead, A. N. (88) and (89), and Nicod, J. (56).
s The first outline of this theory appeared in the monograph Lesniewski, S.
(45a). See also Lesniewski, S. (47 a). From a formal point of view mereology
is closely related to Boolean algebra. In this connexion, cf. X I, p. 333 footnote.

t B ibliographical N ote . An address on the foundations of the geometry

of solids was given by the author to the First Polish Mathematical Congress,
held in Lwow in 1927. A summary of this address appeared in French under
the title ‘ Les fondements de la g6om6trie des corps’ in Ksi$ga Pamiqtkowa
Pierwazego Polskiego Zjazdu Matematycznego, supplement to Annalea de la
SocietS Polonaiae de Math&natique, Krakdw 1929, pp. 29-33. The present
article oontains a translation of this summary, with additions designed to
clarify oertain passages of the French text.
II F O U N D A T IO N S OF TH E G E O M E T R Y OF SOLIDS 25
as a relation between individuals1) can be taken as the only
primitive notion. In terms of this relation various other notions
are defined as follows:
D efinition I. An individual X is called a proper part of an
individual Y if X is a part of Y and X is not identical with Y.

D efin itio n II. An individual X is said to be disjoint from an

individual Y if no individual Z is apart of both X and Y*
D efin itio n I II . An individual X is called a sum of all elements
of a class a of individuals1if every element of a is a part of X and if
no part of X is disjoint from all elements of a.
The following two statements can be accepted as the only
postulates o f mereology:2
P o stu la te I. I f X is a pari of Y and Y is a part of Z , then
X is a part of Z.
P o stu la te n .For every non-empty class a of individuals
there exists exactly one individual X which is a sum of all elements
of a*
The geometry o f solids is based upon mereology, in the sense
that the relation between part and whole is included in the
gyvtem of primitive notions o f the geometry o f solids, and
similarly Posts. I and I I are included in the postulate system
of that theory.
The notion o f a sphere is accepted as the only specific primitive
1 The terms ‘ individual* and ‘ class’ are used here in the same way as in
the well-known work of A. N . Whitehead and B. Russell (90). The termi­
nology adopted in this article differs in several details from that of Le&niewski’s
mereology. In fact Le6niewski uses the terms ‘ ingredient’, ‘ part*, ‘ exterior*,
and ‘set* in the sense in which I use ‘ part’, ‘ proper part*, ‘ disjoint’, and ‘ sum’
respectively, and he employs the term ‘ class* in a sense completely different
from ours.
* The postulate system for mereology given in this article (Posts. I and II)
was obtained by simplifying a postulate system originally formulated by
S. Le6niewski in (47 a), Ptzeglqd Filozoflczny, vol. 33 (1930), pp. 82 ff. The
simplification consisted in eliminating one ofLesniewski’s postulates by deriving
it from the remaining ones. A postulate system obviously equivalent to the
system formed by Posts. I and II can be found in the book of J. H. Woodger,
The Axiomatic Method in Biology, Cambridge 1937, Appendix E (by A . Tarski),
p. 160.
20 F O U N D A T IO N S OF TH E G E O M E T R Y OF SOLIDS II

notion o f the geom etry o f solids.1 In terms o f this notion (and

o f the notions o f m ereology previously introduced) I shall now
successively define a series o f further notions which will prove
helpful in eventually form ulating the specific postulates for the
geom etry o f solids.2
D efinition 1 . The sphere A is externally tangent to the sphere
B if (i ) the sphere A is disjoint from the sphere B; (ii) given two
spheres X and Y containing as a pari the sphere A and disjoint from
the sphere B , at least one of them is part of the other.
D efinition 2. The sphere A is internally tangent to the sphere
B if (i ) the sphere. A is a proper part of the sphere. B ; {ii) given two
spheres X and Y containing the sphere A as a part and forming
pari of the sphere J5, at least one of them is a pari of the other.
Definition 3. The spheres A and B are externally diametrical
to the sphere C if (i) each of the sphei es A and B is externally tang&n t
to the sphere 0 ; (ii) given two spheres X and Y disjoint from ike
sphere G ami such that A is a pari of X and B a part- of Y, the
sphere X is disjoint from the sphere Y.
D efinition 4. The spheres A. and B are internally diametrical
to the sphere 0 if (i) each of the spheres A and B is internally
tangent io th„ sphere C; (ii) given two spheres X and Y disjoint
from Hit xpht'-c C and such that the sphere A is externally tangent
to X and B to Y, the sphere X is diArfp from the sphere Y.
D efint.cion 5. The sphere A is concentric vntk the sphere B if
one of the following conditions is satisfied: (i) the spheres A and B
are identical; (if) the sphere A is a •proper part of the, sphere, B and
besides, given tv:o spheres X and Y externally diametrical to A afnd
internally tangent to B, these spheres are internally diametrical to
B ; (Hi) the sphere B is a proper part of the sphere A and besides,
1 As regards three-dimensional point- geometry, a method of bating it on
the notion of n2>kere the only primitive notion has been developed in
Huntington. Ih V. (33;; but the construction outlined in the present article
has very little in '.ommon with Huntingdon s development.
2 The system of definitions given below embodies some simplifications (in
particular in the formulation of D el 3) tor which the author is indebted to
B. Knaster. For further simplifications of these definitions gee a recent note
o f Jaskowoki, S. :36 a).
n FO U N D A T IO N S OF TH E G E O M E T R Y OF SOLIDS 27

given two spheres X and Y externally diametrical to B and internally

tangent to A, these spheres are internally diametrical to A .
Definition 6. A point is the class of all spheres which are con­
centric with a given sphere}
D efinition 7, The points a and b are equidistant from the point
c if there exists a sphere X which belongs as ehmeni to the point c and
which moreover satisfies the following condition: no sphere Y
belonging as element to the point a or to the point b is a part of X
disjoint from X .
D efinition 8. A solid is an arbitrary mm of spheres.
D efinition 9. The point a is interior to the solid B if there exists
a sphere A which is at the same time an element of the point a and a
part of the solid B.
It is known that all the concepts o f Euclidean geom etry can
be defined b y means o f those o f point and o f equidistance of two
points from a third} Consequently, by regarding the concepts
introduced by Defs. 6 and 7 as correlates o f their homonyms in
ordinary geom etry, we are able to define in the geom etry o f
solids the correlates o f all the other notions o f point geom etry.
In particular, we are able to establish the meaning o f the ex­
pression 'the class a of points is a regular open set9,
I turn now’ to the form ulation o f the specific postulates which
— together with Posts. I and II o f m ereology— form a postulate
system sufficient for the developm ent o f the geom etry o f solids.
In the first place I adopt the follow ing:
P ostulate I. The notions of paint and of equidistance of two
points from a third satisfy all the postulates of ordinary Euclidean
geometry of three dimensions.3
In addition to this postulate, which is o f fundamental im por
tance, we m ay adm it certain auxiliary postulates which render
our system categorical. The postulates which I adopt for this
1 Thus, in the geometry of solids, spheres are treated as individuals, i.e.
objects of the first order, while points are classes of spheres and hence objects
of the second order. In or dinar/ point geometry the opposite is true,
3 Of. Pieri, M. (59).
* A postulate system for ordinary geometry involving only these two
notions as primitive can be found in Pieri, M. (59).
28 F O U N D A T IO N S OF TH E G E O M E T R Y OF SOLIDS II

purpose establish exact connexions between notions o f the

geom etry o f solids— those o f a solid and o f the relation of apart to a
whole— and the corresponding pair o f notions o f ordinary point
geom etry, those o f a regular open set and o f the relation of in­
clusion,.
P ostulate 2. I f A is a solid, the class a of all interior points
of A is a non-empty regular open set.
P ostulate 3. I f the class a of points is a non-empty regular
open set, there exists a solid A such that a is the class of all its interior
points.
P ostulate 4. I f A and B are solids, and all the interior points
of A are at the same time interior to B , then A is a part of B.
The postulate system given above is far from being simple
and elegant; it seems very likely that this postulate system can
be essentially sim plified b y using intrinsic properties o f the
geom etry o f solids. F or instance, it is not difficult to show that
Post. 4 can be replaced by either o f the follow ing tw o postulates:
P ostulate 4'. I f A is a solid and B a part of A } then B is
also a solid.
P ostulate 4". I f A is a sphere and B a part of A, there exists
a sphere C which is a part of B.
It should be noticed that the class o f all solids (and n ot that
o f all spheres) constitutes what is called the ‘universe o f dis­
courseJ for the geom etry o f solids. F or this reason it m ay be
convenient to adopt the notion o f a solid as an additional
prim itive notion. In this case D ef. 8 should, o f course, be om itted;
in its place the fa ct that the class o f solids coincides with that
o f arbitrary sums o f spheres should be stated in a new postulate.
As regards m ethodological properties o f the geom etry o f
solids, I should like to state the follow ing two results (the proofs
o f which present no serious difficulties):
T heorem A. The postulate system of the geometry of solids
is categorical.
More specifically, given any two models o f this postulate
system , we can establish a one-to-one correspondence between
H F O U N D A T IO N S. OF TH E G E O M E T R Y OF SOLIDS 29

solids o f the first m odel and those o f the second in such a way
that if X ,Y , Z are any solids o f the first m odel, and X\ Y\ Z'
the corresponding solids o f the second m odel, we have:
(i) X is a sphere i f and only i f X ' is a sphere;
(ii) Y is a part o f Z i f and only i f Y' is a part o f Z '.
T heorem B . The postulate system of the geometry of solids,
with the postulates of mereology included, has a model in ordinary
three-dimensional Euclidean geometry. To obtain such a model
we interpret spheres as interiors of Euclidean spheres, and the
relation of a part to a whole as the inclusion relation restricted to
non-empty regular open setsJ1
Conversely, the postulate system of three-dimensional Euclidean
geometry has a model in the geometry of solids. Such a model is
obtained by interpreting points and the relation of equidistance in
the way indicated in Defs. 6 and 7.
Consequently, the consistency problems for the two postulate
systems.are equivalent.
In conclusion, when we com pare the results which have been
summarized with those o f W hitehead and N icod, the follow ing
m ay b o stated: The procedure which has enabled us here to
form ulate definitions and postulates (especially D efs. 6 and 7
and Post. 1) can be regarded as a special case o f the method of
extensive abstraction developed b y W hitehead. N icod has already
drawn attention to the equivalence o f the problem s o f con ­
sistency for the tw o system s o f geom etry : that o f the geom etry
o f solids and that o f ordinary point geom etry. In m y opinion
what is to be regarded as a new result is the precise, m ethod o f
establishing the m athem atical foundations o f the geom etry o f
solids w ith the help o f a categorical system o f postulates con­
taining only tw o prim itive notions: the notion o f sphere and
that o f being a part.
1 This fact has an interesting consequence formulated in terms o f Boolean
algebra; cf. X I, p. 341, note 2.
 Ill
ON SOME F U N D A M E N T A L CONCEPTS
OF M E T A M A T H E M A T IC Sf

O ur object in this com m unication is to define the meaning,

and to establish the elem entary properties, o f some im portant
concepts belonging to the methodology of the deductive sciences,
which, follow ing H ilbert, it is custom ary to call metamaihe-
matics.1
Form alized deductive disciplines form the field o f research of
metaniathemafcics roughly in the same sense in which spatial
entities form the field o f research in geom etry. These disciplines
are regarded, from the standpoint o f metamathematics, as sets
o f sentences, Those sentences which (following a suggestion o f
3. Lesniewski) are also called meaningful sentences, are them ­
selves regarded as certain inscriptions o f a well-defined form .
The set o f all sentences is here denoted bv the sym bol ‘S ’ . From
the sentences o f anv/set X certain other sentences can be obtained
%
by means o f certain operations called rules of inference. These
sentences are called the ccmsequences of the set X. The set o f all
consequences is denoted by the sym bol (Cn(X)\
A n exact definition o f the two concepts, of sentence and o f
consequence, can be given only in those branches o f meta-
mathematics in which the field o f investigation is a concrete
form alized discipline, On account o f the generality o f the present
considerations, however, these concepts will here be regarded as
prim itive and will be characterized by means o f a series o f
1 Many ideas and resume outlined m this report have been presented in
a more detailed way in two later articles of the author, V and X II.

+ BiBLioGRAPHiCAii The main ideas of this article were outlined by

the author in a lecture to the Polish Mathematical Society, Warsa ~ Section, in
192S. For & summary f this lecture goo Tarski, A. (72). The comm unicat ion
was presented {by 3. ; •-'ke.siewies) to the Warsaw Scientific Society on 27
March 1920; it was published under the title 4Vber eimge fundamental
Begriffe der Metamathemarik5 in Cotnpte* Bendus des seance* c*. la &ocidU de*
Sciences dcs LeMree de Varsovie, vol 22, 1920. cl. iii, pp, 22-29.
n i M E TA M A T H E M A TIC S 31

axiom s. In the custom ary notation o f general set theory these

axioms can be form ulated in the follow ing way:

A xiom 1. $ < X0.

A xiom 2. I f X s S, then X £ C'n(JT) c 8,
A xiom 3. I f X £ $, tftew Cn(Cn(X)) = Orf(X).
A xiom 4. / / X s then Cn(X) == 2 -

A xiom 5. There exists a sentence x e S such that Cn({x}) =

W ith a view to reaching m ore profound results, other axioms
o f a m ore special nature are added to these. In contrast to the
first group o f axiom s those o f the second group apply, not to all
deductive disciplines, but only to those which presuppose the
sentential calculus, in the sense that in considerations relating
to these disciplines we m ay use as premisses all true sentences
o f the sentential calculus.1 In the axiom s o f this second group
there occur as new prim itive concepts two operations b y means o f
which from simple sentiences more com plicated one3 can
be form ed, nam ely the operations o f form ing im plications and
forming negations. The im plication with, the antecedent x and
the consequent y is here denoted by the sym bol lc{xyy)\ and
the negation o f x b y the sym bol *n(x)\ The axiom s are as
follows:1
A xiom 6*. I f x e 8 and y e S , then c(x, y) e 8 and n{x) e $,3
1 i.e. all sentences which belong to the ordinary (two-valued) system of the
sentential calculus; cf. IV, § 2, Def. 5.
* The numbering of the axioms of the second gro-jp and of the theorems
which follow from them is distinguished from that of the remaining axioms
and theorems by the presence of an asterisk ** \
• As already explained, sentences are here regarded as material objects
(inscriptions). From this standpoint the content of As. 6* does not correspond
exactly to the intuitive properties of the concept* recurring in it. It is not
always possible to form the implication of two sentences (they may occur in
widely separated places). In order to simplify matters we have, in formulating
t.hia axiom, committed an error; this consists m uleniifyi.xg equiform sentences
(as S. Le&bewski calls them}. This error ^-au be removed by interpreting <5?
as the set c f all types of sentences (and not oi sentences) and by modifying in
an analogous maimer the intuitive sense of other primitive concepts. In this
connexion by the type of a sentence x we understand the set of all sentences
which are equiforuo with x.
32 ON SOME F U N D A M E N T A L CONCEPTS IH

A xio m 7*. I f X s 8, y e 8, z e 8 and e{y,z) e Gn(X), then

z e (7 n (X + {y }).

A xio m 8*. I f X £ 8, y e S, z e 8 and z e Gn(X-\-{yj), then

c(y,z) 6 Cn(X).-\

A x io m 9*. I f x e S, then Gn({x,n(x)}) — 8.

A xio m 10*. I f x e S, then Cn({x} ) . Gn{{n(x)}) — Cn(0).

A xiom s 8* and 10* are satisfied only in connexion with those

form alized disciplines in the sentences o f which no free variables
occu r.1 Instead o f A x. 8* in its full extent, it suffices to adopt
as an axiom the follow ing special case o f this sentence:

I f y e 8, z e 8 and z e Cn({y\), then c(y, z) e Gn(0).

On the basis o f these axiom s a series o f theorems concerning

the concepts involved can be proved, for exam ple:

T heorem 1. I f X £ 7 £ 8, then Cn(X) £ Gn(Y).

T heorem 2. I f X + Y £ S, then

C n {X + Y ) = Cn{X+C n{Y)) = O n (0 « (Z )+ C » (r )).

This theorem can be generalized to cover an arbitrary (even

an.infinite) num ber o f summands.

1 This means that expressions (sentential functions) with free variables are
not regarded as sentences.

f Ax. 8* is one of the formulations of what is known in the literature as

the deduction theorem. This theorem, in its application to the formalism of
Principia Mathematical was first established by the author as far back as 1921
(in connexion with a discussion in the monograph of K . Ajdukiewicz (2)); the
result was discussed in the author's lecture to the Warsaw Philosophical
Institute, Section of Logic, listed by title in B u cli F ilo z o fic z n y , vol. 6 (1921-2),
p, 72 a. Subsequently the deduction theorem was often applied in metamathe-
matic&l discussion. Thus e.g. some theorems stated in the note (85) of A.
Lindenbaum and A . Tarski, as well as in the note of A. Tarski (72), were
obtained with the essential help of this; result. In particular, Th. II in the
first of these notes is simply a specialized form of the deduction theorem.
A proof of the deduction theorem for a particular formalized theory is outlined
in IX , p. 286 (proof of Th. 2 a). For reference to other appearances of the
deduction theorem in the literature see a review by A. Church (of a book by
W . V. Quine) in the Joumal of Symbolic Logic, vol. 12 (1947). pp. 60-61.
m OF M E T A M A T H E M A T IC S 33

T heorem 3*. I f x e S, y e S and z e &, then

c(c(x, y), c(c(y, z), c(x, z))) e Cn(0), c(x, c(n(x), y)) e Cn(0)
and c(c(n(x),x),x) e Cn(0).
This theorem asserts that every sentence obtained b y substitu­
tion from one o f the three axiom s o f the ordinary system o f
sentential calculus, w hich are due to J. Lukasiewicz,1 is a conse­
quence o f the null set 0 (and hence is also a consequence o f
every set X o f sentences). B y the use o f A x. 7* this theorem
can be extended to all substitution instances o f true sentences
o f the sentential calculus.
B y means o f the concepts 8 and Cn(X) other im portant con­
cepts o f m etam athem atics can be defined. F or exam ple:
D efinition 1. A set X of sentences is called a deductive system
(or simply a system ), in symbols X e S , if
Cn(X) = I s 5 .
The follow ing properties o f system s are easily proved:
T heorem 4. For every set X £ 8 there exists the smallest
system 7 which includes X , and in fact 7 = Cn(X).
In consequence o f this theorem the set Cn(0) is the sm allest
system o f all; this set can be called the system of all logically
true sentences.
T heorem 6. I f ft £ S and ft # 0, l e S (the inter­
section of any number of systems is itself a system).
T heorem 6. I f for any two systems X e R and Y e ft there is
a system Z e R with I s Z and 7 s Z, and R # 0, then 2 X e <5.
X e ft

T heorem 7* (of A . Lindenbaum ). I f R s g , ft < X0 and

^ I e S , t h e n X e f t ; in other words, no system•
, can be repre­
sented as a sum of a finite number of systems distinct from itself.
T heorem 8*. I f ft s 6 , ft < X0, Y e S, 7 s 2 X and
XeA
7 9* 0, then there is a system X e ft such that Y £ X .

1 cf. rv, § 2.
34 ON SOME F U N D A M E N T A L CONCEPTS m

W e n est introduce the notion o f (logical) equivalence, together

with the im portant notions o f consistency and completeness.
D efin itio n 2. The sets X and 7 of sentences are called (logi­
cally) equivalent, ir symbols X ~ 7 , if X + 7 s= S and
Cn{X) = Cn(Y).
D efin itio n 3. The set X of sentences is .called consistent, in
symbols X e SB, */ X s S and if the formula X ~ 8 does not hold
(i.e. if On(X) # S).
D e fin itio n 4. The set X is said to be com plete, in symbols
X e 18, if X c S and if every setT e SB which includes X satisfies
the formula X rv F .
W ith the help o f the axiom s o f the second group it can be
shown that Defs. 3 and 4 agree with the usual definitions o f
consistency and com pleteness:
Theorem 9*. X e SBif and only if X £ S and if for no sentence
y e S do we have both y e Gn(X) and n(y ) e On(X).
T heorem 10*. X e 18 if and only if I s /S and if for every
sentence y e 8 at least one of the formulae y e Cn(X) and
n(y) e Cn{X) holds..
The follow ing theorem s are also derivable:
T h e o r e m 11. I f it £ <SJ and if for every finite class 2 £ Si
a set 7 e St exists which satisfies the formula V X £ Y, then
y x e aas. cs
T heorem 12 (o f A . landenbaum ). I f X e SB, then there is a set
Y e S . SB. S3 such that X £ 7 : in other words, every consistent
set of sentences can be enlarged to iotrn a consistent and complete
system.
T heorem 13*. Let X s S, In order that y e Cn(X) it is neces­
sary and sufficient that y e S o d that the formula X - f { « (y)} e SB
should not hold.
The concept o f com pleteness is often oonfused with tw o other
concepts which are related to it in content: that o f categoricity
in OF M E TAM AT H EM ATIC S 35

and o f non-ramifiability.1 The next notion to be defined is closely

related to that o f completeness:
D efin itio n 5. The degree o f completeness o f a set X c S o f
sentences, in symbols y(X ), is the smallest ordinal number a # 0
which satisfies the following condition: there exists no increasing
sequence o f consistent non-equivalent sets Xg of sentences o f tqpe a
which begins with X (i.e. no sequence of sets X ( satisfying the
formulas: X 0 = X , X$ c X n s /S' and Cn(Xg) ^ Cn(Xv) for
£ < i) < a).
From this definition it follows that:
T heorem 14. y(X ) 1 if and only if X ^ 8 (i.e. if X £ S
—

and if the formula X e 903 does not hold); y(X) •— 2 if and- only
if X e ffi.93; y(X ) > 2 if and only if X e SB— 93.
Finally we introduce the follow ing concepts:
D efin itio n 6. A set X of sentences is called independent, in
symbols X e U, if X £ 8 and if Y — X always follows from the
formulas: T X and Y £ X .
D efin itio n 7. A set Y of sentences is called a basis o f the set
X of sentences, in symbols Y e SB(X), if X and Y e H.
D efinition 8. A set Y of sentences is called a finite axiom
system, or for short an axiom system^of the set X o f sentences,
in sy m b o lsY eV x (X ),ifX Y and Y < X 0.
DEFINITION 9 . A set X of sentences is said to be finitely axioma-
tisable, or for short axiomatizable, in symbols X eS l, if
*x(X ) # 0.
T heorem 15*. X e U if and only if X S and for every y e X
the formula X —{y}-f{n (y)} 6 SB holds.
T heorem 16. I f X s S and % < X0, then there exists a set
7 G, X such that Y e © (X ); i.e. every finite set of sentences contains
a basis as a subset.
T heorem 17*, I f X £ S, then © (X ) *£ 0; i.e . every set of sen­
tences possesses a basis.
1 Cf. the remarks concerning these notions in article V* at the end of $ 7-
36 ON SOME F U N D A M E N T A L CONCEPTS m

T heorem 18. The following conditions are equivalent: (1)

X e 51; (2) there is a set T £ X , such that Y e 9lr(X)j_(3) 9 lr(X ).
23(X) =£ 0; (4) JAere exists a set Y £ X , such that Y < Xo
Fe93(X).

T heorem 19*. /n order $Aa$ X e 51 it is necessary and sufficient

that X c S and that X possess no infinite basis.

T heorem 20*. Let X e in order that X 6 51 it is necessary

&;

and sufficient that no class ft c 0 exist which satisfies the follow­

ing conditions: X e f t and X = 2 Y (i.e. that X cannot be repre-
sented as a sum of systems distinct from itself).

T heorem 2L We have § 7 1 < N0 and ® < 2*;

i / an infinite set X e U exists, £Aen

1 7 1 = X0 and = 1 = 2**.

I t should be noted that in almost all known deductive dis­

ciplines an infinite and at the same tim e independent set o f
sentences can be constructed, which thus realizes the hypothesis
o f the second part o f the last theorem. In these disciplines there
are therefore m ore non-axiom atizable than axiom atizable
system s; system s are, so to speak, only exceptionally axiom a­
tizable.1
B y some authors the concept o f the independence o f a set
o f sentences is sharpened in various directions (complete inde­
pendence b y E . H . M oore,2 maximal independence by H . M.
Sheffer3). These notions will not be discussed here.
W ithin the conceptual fram ework o f this paper we can carry
out m etam athem atical investigations relating to concrete deduc­
tive disciplines. For this purpose in each single case the concepts
o f sentence and o f consequence must first be defined. W e then
take as a starting-point some set X o f sentences in which we are

1 Lindenbauin was the first to draw attention to this fact (in connexion
with the sentential calculus), thus indirectly suggesting the notion of axio-
matizability; cf. IV , § 3.
Q Introduction to a form of general analysis, New Haven 1910, p. 82.
1 Sheffer, H. M. (63), p. 32.
m OF M E T A M A T H E M A T IC S 37

interested. W e investigate it from the point o f view o f con­

sistency and axiom atizability; we try to determine its degree
o f com pleteness, and possibly to specify all the systems, in
particular all the consistent and com plete systems, which include
X as a subset.f
t For an example of investigations in these directions which deal with
the simplest deductive discipline, namely the sentential calculus, the reader
is referred to IV . Some results concerning other deductive disciplines are
briefly discussed in X II, § 5.
 IV
IN V E S T IG A T IO N S IN TO T H E S E N T E N T IA L
CALCULUS!

I n the course o f the years 1920-30 investigations were carried

out in W arsaw belonging to that part o f metamathematics— or
better m etalogic— which has as its field o f study the simplest
deductive discipline, nam ely the sentential calculus. These
investigations were initiated b y Lukasiewioz; the first results
originated both with him and with Tarski. In the seminar
for m athem atical logic which was conducted b y Lukasiewicz in
the U niversity o f W arsaw beginning in 1926, m ost o f the results
stated below o f Lindenbaum , Sobooifiski, and W ajsberg were
found and discussed. The- system atization o f all the results
and the clarification o f the concepts concerned was the work o f
Tarski.
In the present com m rm ication the m ost im portant results o f
these investigations -for the m ost part not previously pub­
lished— are collected together.^

§ 1. G e n e b a e Concepts
I t is our intention to refer our considerations to the conceptual
apparatus which was developed hi the preceding article (see I II ).
t BrouoaBAPHicAL N oth:. This joint <’emmnnioation of J\ Lukasiewicz and
A. Tarski was presented (by Lukasiewicz) to the Warsaw Scientific Society on
27 March 1930; it was published under the title 4Untersuohungen iiber den
Ausaagenkallriil ’ in Oomptes Rendus des stances de la SocteU des Sciences ei des
Lettres de Vcrzovie, vol. 23, 1930, cl. iii, pp. 80-50.
% To avoid misunderstandings it should be stated that the present article
does not contain results discovered by both the authors jointly, but is a com­
pilation of theorems and concepts belonging to five different persons. Each
theorem and conoept is scribed to its respective originator. Theorem 3, for
instance, is not a theorem of Lukasiewicz and Tarski, but a theorem of
Lindenbaum, Nevertheless, some scholars mistakenly referred to both authors,
Lukasiewicz and Tarski, the many-valued systems of logio ascribed in the
article to Lukasiewicz alone. In spite of a correction which appeared in 1933
in the Journal of Philosophy, vol. 30, p< 364, this mistake persists till today. 2t
clearly follo ws from § 3 and notes of this article that the idea of a logio different
from the ordinary system called by Lukasiewicz the two-valued logic, and
the construction of many-valued systems of logic described here, are entirely
due to Lukasiewicz alone and should not be referred to Lukasiewicz and Tarski.
IV, J I S E N T E N T IA L CALCULUS 36

F or this purpose we wish first to define the notion o f a (meaning­

ful) sentence and that o f a consequence of a set of sentences w ith
respect to the sentential calculus.

D efinition 1. The set S o f all sentences is the intersection of

all those sets which contain all sentential variables (elementary
sentences) and are closed under the operations of forming implica­
tions and negations.1
The concepts o f sentential variable, o f implication and o f nega­
tion cannot be explained further; th ey m ust rather be regarded as
prim itive concepts o f the m etatheory o f sentential calculus (i.e.
o f that part o f metamathematiGS in which sentential calculus
is investigated). The fundam ental properties o f these concepts,
which suffice for the construction o f the part o f m etam athe­
m atics w ith w hich we are here concerned, can be expressed in
a series o f sim ple sentences (axiom s) which need not now be
stated. U sually the letters ‘p ’, ’g ', V\ etc,, are used as sentential
variables. In order to express in sym bols the sentences ‘p im ­
plies q' (or ‘i f p, then q’ ) and ‘it. is n ot the case that p ’ ,
Lukasiewicz em ploys the form ulas ‘ Cpq’ and ‘N p ’ respectively.2
W ith this notation the use o f such technical signs as parentheses,
dote, eto., is rendered unnecessary. W e shall encounter several
examples o f sentences w ritten in this sym bolism in subsequent
sections. In addition to the form ation o f im plications and nega­
tions other similar operations are com m only used in the senten­
tial calculus. B u t as these are all definable b y means o f the tw o
m entioned above they w ill n ot be considered here.
The sym bol ‘ Cpq’ , w hich in the sentential calculus expresses
the im plication betw een ‘p ’ and ‘q’ , is to be clearly distinguished
from the m etam athem atical sym bol :c(x, y )’, which denotes the
im plication w ith the antecedent x and the consequent y. The
expression ‘ Cpq’ is a sentence (in the sentential calculus) whilst
the expression ‘c(x, y )’ is the name of a sentence (in the m eta-
1 A set— according to the usual terminology of abstract set theory— is
said to be closed under given operations if as the result of carrying out these
operations on elements of the set in question one always obtains elements of this
same set.
1 Cf. Lukasiewicz, «T. (51 a), p. 610 note, and (51), p. 40.
40 I N V E S T I G A T I O N 'S IN T O TH E IV, § 1

sentential calculus). An analogous remark applies to the sym­

bolic expressions (N p ’ and
The consequences o f a set o f sentences are formed with the
help of two operations, that of substitution and that of detach­
ment (the modus ponens scheme of inference). The intuitive
meaning of the first operation is clear; we shall not, therefore,
discuss its character more closely. The second operation consists
in obtaining the sentence y from the sentences x and z = c(xyy ) .
We are now in a position to explain the concept of conse­
quence:
D efinition 2, The set of consequences Cn{X) of the set X of
sentences is ike intersection of all those sets which include the set
X c S and are closed under the operations of svbsfituticm and
detachment.
From this we obtain:
T heorem 1. The concepts id and Cn{X) satisfy the axioms
1-5 given in article I I I . 1
We are especially interested in those parts X of the set 5
which form deductive systems, i.e, which satisfy the formula
Cn(X) = X, Two methods of constructing such systems are
available to us. In the first, the so-called axiomatic method, an
arbitrary, usually finite, set X of sentences—-an axiom system
— is given, and the set Cn(X)} i.e. the smallest deductive system
over X , is formed. The second method, which can best be called
the maHx method, depends upon the follow ing definitions of
Tarski:2
1 See article III, p. 31.
3 The origin o f thi.-; method is to be sought in the well-known verification
procedure fo r the usual two-valued sentential calculus {see below Def. 5),
which was used by Peirce (‘ On the algebra o f Logic \ Am. J oum. of Math.
vol. 7 (I8S5), p. 101} and Schroder. This was thoroughly treated in Lukasiewicz.
J. (50 o). Lukasiewicz was also the first to define by means of a matrix a
system o f the sentential calculus different from the usual one, namely his
three-valued system (see below, p. 47, note 2). This he did in the year 1920.
Many-valued systems, defined by matrices, were also known to Post (see Post,
E. L. (60), pp. 180 ffi). The method used by P. Bernays (‘ Axiomatische
Untersiichung des Ausaagenkalkiiis der “ Principia Mathematica” \ Math. Z .%
vol. 25, 1926, pp. 305-20) for the proof of his theorems on independence also
rests on matrix formation. The view o f matrix formation as a general method
o f constructing systems is due to Tarski.
IV. §1 S E N T E N T IA L CALCULUS 41

Definition 3. A (logical) matrix is an ordered quadruple

3Wl = [A, B ,f, g\ which consists of two disjoint sets (with dements
of any bind whatever) A and B , a function f of two variables end a
function g of one variable, where the two functions are defined for
all elements of the set A -f-i? and take as values dements of A d B
exclusively.
The matrix 351 — [A y B. /, g] is called normal if the formulas
x e B and y e A always imply f ( x } y) e A,

Definition 4. The function h is called a value function of the

matrix 50t — [A, B, /, <?] if it satisfies the following conditions:
( 1 ) the function h is defined for every x e S; (2 ) if c ?.->a sentential
variable, then h(x) e A-\-B\ (3) if x e S and y e St then

h(c(x,y)) =f(h(x),h(y)y,

(4) if x e S then h(n(x)) = g(h(x)).

The sentence x is satisfied (or ■verified) by the matrix
501 = [A, B,f, <7], in symbols x e C(501), if the formula h(x) e B
holds for every value function h of this matrix.

The elements of the set B are, following Beraays,1 called

designated elements.
In order to construct a system of the sentential calculus with
the help of the matrix method a matrix 3JI (usually normal) is
set up and the set ffi(SR) of all those sentences which are satisfied
by this matrix is considered. This procedure rests upon the
following easily provable theorem;

Theorem 2. I f 501 is a normal matrix, then @(50t) e S.

I f the set ©(50?) forms a system (as it always will, according to

Th. 2, if the matrix 501 is normal), it is called the system generated
by the matrix 501.
The following converse of Th. 2, which was proved by Linden -
baum, makes evident the generality of the matrix method
described here:

1 See Bernaya, P. (5), p. 316.

« IN V E S T IG A T IO N S IN T O TH E IV, *1

T heorem 3. For every system l e S there exists a normal

matrix 2U = [A, B, f, g], with an at most denumerable set A + B ,
which satisfies the formula X —

E ach o f the tw o m ethods has its advantages and disadvan­

tages. System s constructed b y means o f the axiom atio m ethod
are easier to investigate regarding their axiom atizability, but
system s generated b y m atrices are easier to test for com plete­
ness and consistency. In particular the follow ing evident theo­
rem holds:

T heorem 4. I f 2R = [d , B ,f, g] is a normal matrix and

A ^ 0, then ®(S») € SB.

§ 2. T he O r d in a r y (T w o -v alu ed ) System oe the

Sen ten tial Calculus

In the first place we consider the m ost im portant o f the

system s o f the sentential calculus, nam ely the well-known
ordinary system (also called b y Lukasiewicz1 the tw o-valued
system ), which is here denoted b y {L'.
Using the m atrix m ethod, the system L m ay be defined in
the follow ing w ay:

D efin itio n 5. The ordinary system L o f the sentential cal­

culus is the set of all sentences which are satisfied by the matrix
SOI [A, B ,f, gr] where A = {0}, B — {1}® and the functions f and
g are defined by the formulas: /(0 , 0) = /(0 , 1) = / ( l , 1) — 1,
/ ( l , 0) - 0, fif(0) = 1, <7(1) = 0.

1 See (60 a).

1 The set having a as its only element is denoted by ‘ {o}

t A proof of this theorem has recently been published in J. Z na«

tryeach logicznyoh’ (*0x1 logical matrices % in Polish), Travaux de la SocitU
des Sciences et dta Lettres de Wroclaw, Ser. B, Nr. 19, Wroclaw (1949), 42 pp.
See also H. Hermes, *Zur Theorie der aussagenlogischen Matrizen’, Math. Z.t
vol. 53 (1951), pp. 414-18.
IV, §2 S E N T E N T IA L CALCULUS 48

From this definition it follow s easily that the system L is

consistent and com p lete:
T heorem s . ie S .2 B .3 3 .
The system L can also be defined by means o f the axiom atic
method. The first axiom system o f the sentential calculus was
given b y C. Frege.1 Other axiom system s have been given by
W hitehead and Bussell2 and b y H ilbert.3 O f the system s at
present know n the simplest is that o f Lukasiewicz; ho also has
proved in an elem entary manner the equivalence o f the tw o
definitions o f L.*1 His result m ay be stated thus;

T heorem 0 . Let X be the set consisting of the three sentences:

*CCpqCCqrCpr\ 'CCNppp\ cCpCNpq
then X 6 2tx(L). Consequently L is amomatizable, L e 21.
A ccording to a m ethod for investigating the independence o f
a set X o f sentences developed by B em ays and Lukasiewicz,5
1 BegriffMchrift, Halle a/S (1879), pp. 26-30. Frege’s system is based upon
the following six axioms: iCpCgp\ ‘ COpCqrCCvqCpr', 1COpGqrCqCpr',
*OOpqCNqNp’, *CNNpp\ lCpNJSp\ Lukasiewicz has shown that in this
system the third axiom is superfluous smee it can be derived from the preced­
ing two axioms, and that the last three axioms can be replaced by the single
sentence *CONpNqCqp\
a Principia McUhematica, vol. 1 (1925). p. 91.
• See (29 a), p. 153.
4 Cf. Lukasiewicz, J. (61), pp. 46 and 121 ff. The proof of the equivalence
of the two definitions of L aroountr to the same thing as proving the complete­
ness of the system L when defined by meana of the axiomatic method. The
first proof of completeness of this kind is found in Post (60).
6 Bemays has published, in his article (6), which dates from the year 1926
(but according to the author’s statement contains results from his unpublished
Habilitationeschrift presented in the year 1918), a method based upon matrix
formations, which enables us to investigate the independence of given Bets of
sentences. The method given by Bemays was known before its publication
to Lukasiewicz who, independently of Bemays, and following a suggestion
of Tarski (cf. I, pp. 8-14), first applied his many* valued systems, defined
by means of matrices, to the proofs of independence, and subsequently
discovered the general method. On the basis of this method Lukasiewicz
had already in 1924 investigated the independence of the axiom systems
given by Whitehead and Russell and by Hilbert, and had shown that
neither of them is independent. These results {without proof) are contained
in the following note by Lukasiewicz: ‘ Demonstration de la compatibility
dee axioms de la th6orie de la deduction’, Ann . Soc. Pol. M a t h vol. 3 (1925),
p. 149.
u I N V E S T I G A T I O N S IN T O T H E IV, § 2

a norm al m atrix 9Jly is constructed for every sentence y e X

which verifies all sentences o f the set X with the exception o f y.
W ith the help o f this method Lukasiewicz proved that in
contrast to the previously mentioned axiom system the follow ing
theorem holds:
T heorem 7 . The set X of sentences given in Th. 6 is indepen­
dent; consequently X is a basis of L. X c 33(L).
Tarski developed another structural m ethod for the study o f
independence. Although less general than the m ethod o f m atrix
form ation, this can be used successfully in some cases.
The follow ing general theorem is due to T arsk i:f
T heorem 8. The system L, as well os every axiomaiizubie
system of the sentential calculus which contains the sentences:
'CpCqp* and ‘ CpCqCGpCqrr (or cCpCqCGpCarGsr'), possesses
a basis consisting of a single sentence.1
The p roof o f this theorem enables us in particular it* give
effectively a basis o f the system L which contains a single
elem ent.2J Lukasiewicz has sim plified Tarski's p roof and, with
the help o f previous work o f B. Sobocinski, has established the
follow ing:
T heorem 9. The set which consists of the single sentence z:
‘ CCCp Cqp C 0 CN rCsNt 0 CrOsu C OtsCiu vGwv*
is a basis of the system L, i.e. {z} e 33(i> ).

3 .lii analogous, but/ quite trivial, theorem applies to all axiornalivable

systems of those deductive disciplines which already presuppose the sentential
calculus and satisfy not only Axs. 1-5, but also Axs. 6*- 10* of III.
2 This result was obtained by Tarski in the year 1925; of. Lesniewski. S. (46),
p. 58. An axiom system o f the ordinary sentential calculus consisting of a single
axiom was set up by Nicorl in the year 1917 (see Nicod, J. (55)). The axiom
o f Nicod is constructed with the She Cor disjunction 'p j qTas the only primitive
term, and the rule of detachment formulated by Nicod m connexion with this
term is stronger than the rule of detachment for implication. This facilitated
the solution of the problem.

f Compare in this connexion a recent paper of & Pel*ofttor {oZ a ), in

particular pp. 294 15.
X The axiom originally found by Tarski is explicitly fonnuhih-d n; the
article b y B. Sobocihski, ‘ Z badah nad teorja dedukcji’, P r a e g ^ d Filc;;oJtcznyt
vol. 35 (1932), pp. 172-93, in particular p. 189. It oonsists of 53 letters.
IV, §2 S E N T E N T IA L CALCULUS 45

This sentence z which has 33 letters, is the shortest sentence

known at the present tim e which suffices as the only axiom
for developing the system L. The sentence z is not organic
with respect to the system L. F or a sentence y e X is said to
be organic with respect to a system X if no (meaningful) part o f
y is an elem ent o f X (the term ‘organic’ comes from S. Les-
niewski, and we owe the definition o f organic sentence to
M. W ajsberg). The sentence z is not organic with respect to L
because it contains parts, c.g. ‘ C p C q p which are elements o f L.
Sobocihski has given an organic axiom for the system L which
contains 47 letters.*}*
The foliowin.g theorem is & generalization o f Th. 8:
Theorem 10. The system L. as well as every aziomatizable
system of the sentential cahulus which contains the sentences
'CpCqp' and 'CpGqCCpCqrrpossesses for every natural number
m a basis containing exactly m elements-
For the system L has effectively proved this
theorem ; the generalization to other systems is due to Tarski4
In contrast- to this property o f che system L. Tarski has
effectively shown that:
Theorem For every %wUurai number m, systems of the
11 .
sentential calculus exist every basis of vjh^ch contains exactly rn
elements.
The follow ing considerations o f Tarski, col-conn the special
case o f this theorem when m 1 (D e i 0 and Ths. 12-14).
D efinition 6, The sentence x is called ^decomposable if
x e S and if every oasis of the system Cn({.r)) consists of o?Uy one

t The results discussed in the last paragraph of irv> text were improved
after the original publication of this article. I d fact Lukasiewicz found in 1032
a single noil-organic axiom consisting of 29 letters: see Sobocuis&i *60 c),
especially pp. 181 f. In 1936 on the ground of a result of Sobociiiski ho pub­
lished without proof a single organic axiom of 23 letters in J. Lukasiewicz p?2),
p. 121, note 10. The shortest hitherto known single organic axiom consisting
of 21 letters was found in 1952 bj7C. A. Meredith; see his article, *Single axioms
for the systems (C, N), {O, O), and {A, N) of the wo-valued propositional
calculus'. The Journal of Computing Systems, voi, 1 (1953), pp, 155-64.
X See B. Sobocihski, ‘Z bad an nad teorja dedukcji’ (‘Some investigations
upon the theory of deduction1), Frzeglqd Filozoficzny, voi. 36 (1932), pp, 178 ff.
46 I N V E S T I G A T I O N S IN T O T H E IV, § 2

sentence (i.e. if no independent set of sentences containing more

than one element is equivalent to the set {x}).
I f this condition is not satisfied then the sentence x is said to be
decom posable.
It is found that alm ost all known sentences o f the system L
are indecom posable, in particular:
T heorem 12. The sentences
cCpp‘, ‘ GpGCpqgf, ‘ C C O p q p p ‘ CCCpgqCCqpp
‘ CGpqOOqrGpr\ ,CGqrGGpqCpr>
are indecomposable.
T heorem 13. I f x s Af, y g S, and s e S then the sentences
» (* ), c(n(x), y), c{cln(x), y), e), c(x, c(n{y), z))a?e indecomposable; in
particular, this holds for the sentences :
' GNNpp*, ‘ CpNNp\ ‘ CNpCpq’, !CpCNpq
‘ GCJs/ppp’, C CpJS'pNp ".
From Ths. 12 and 13 it results that the set o f sentences given
in Th, 6 consists exclusively o f indecom posable sentences.
On the other hand the follow ing theorem has been proved:
T heorem 14. The sentences
'GpCqp', ’GQCpqrCqr\ and ,GCpCqrOqGpr>
are decomposable*
A notew orthy theorem on axiom systems o f L has been
proved b y W ajsberg:
f The following remarks may help the reader to reconstruct th© proofs of
Ths* 12-I4i
(i) Let x *=* *Cpp\ Ifc can easily shown that the system Cn({x}) consists
of all those and only tho^j sontoneea which can b© obtained from w by sub­
stitution; more generally^ if Y i&any subset of <7»({«}), then Cn(Y) consists of
those and only tho.se sentences which are obtainable from sentences of Y by
substitution. Hence we conclude without difficulty that every independent
set of sentences which is equivalent to {r} coxisists of just one sentence, in fact,
of a sentence c{vt v), where v is an arbitrary variable.
By means of a similar argument man> of the sentences mentioned in
Ths. 12 and IS can be proved to be ^decomposable,
(ii) Let x = iCpOgp\ y - - cCOpUqpOpCcp‘ end ~ — c(t/, x) , Clearly the set
{x} is equivalent to the set, {y, *}. Also it can easily be shown that the set
{y, z) is independent. Hence the sentence x ia dactmaposable.
IV, §2 S E N T E N T IA L CALCULUS 47

T heorem 16. In every basis (and in general in every axiom

system) of the system L, as well as of every subsystem of L which
contains the sentence ‘ CpCqCrp\ at least three distinct sentential
variables occur 1 In other words, if X is the set of all those sentences
of the system L in which at most two distinct variables occur, then
L —Cn(X) # 0; in particular, the sentence 'GpCqCrp* belongs to
L but not to Cn(X ) .f

§ 3. M a n y -valu ed System s of the Sen ten tial Calculus

In. addition to the ordinary system o f the sentential calculus
there are m any other systems o f this calculus which are w orthy
o f investigation. This was first pointed out b y Lukasiewicz who
has also singled out a specially im portant class o f such system s.2
The system s founded b y Lukasiewicz are here called n-valued
systems o f the sentential calculus and denoted b y the sym bol
‘Lpf (where either n is a natural number or n ~ K0). These
systems can be defined b y means o f the m atrix m ethod in the
follow ing w ay:
D e fin itio n 7. The n-valued system Ln o f the sentential cal­
culus (where n is a natural number or n = H0) is the set of all
1 It is not necessary to explain any further the meaning of the expression
‘ in the sentence x two or three distinct variables occur* since it is intuitively
clear. ‘ Distinct’ here means the same as 'no* equif'orra ‘ (of H I, p. 31,
note 3}>
* What is called the three-valued system of the sentential caiouiua was
constructed by Lukasiewicz in the year 1920 and described m a lecture given
to the Polish Philosophical Society in Lw6w, A report by tftp. author, giving
the content of that lecture fairly thoroughly was published in the journal
Buck Filozoficzny, vol. 6 (1920), p* 170 (in Polish). A short account of the
n-valued systems, the discovery of which belongs to the ygar i922; in given in
Lukasiewicz, J. (61), pp. 115 ff. The philosophical implications of ^ valued
systems of sentential calculus are discussed in the oi Lukasiewicz,
4Philosophische Bemerkungen zu mehrwertigea Systemc^ des Aussagen
kalkuis \ Cmipies Bendus des stances de la SccidU ass Sciences & des Lettres de
Vwrsovie> vol. 28 (1980), ol. iii, pp. 61-77.

f Wajsbarg’s proof of Th. 15 is given at the end oi his paper *S7 cf). Another
proof of the result discussed can be obtained by th\r use of the method de­
veloped in the note of A, H . Diamond and J. G <k McKinley, "Algebras and
their subalgebras5, Bulletin of the American Mathematical Society, vol. 53
(1947), pp. 968-62. For another proof of that part of Th, 16 winch concerns
the whole system L, soe also S. Jankowski, ‘ Trois contribution.® au calcul des
propositions bivalent*, Shjdia Societatis Sdentiarum T w nensis, section A,
vol. 1 (1948), pp. 3-15, in particular pp. 9 f.]
13 I N V E S T I G A T I O N S IN T O T H E IV, §3

sentences which are satisfied by the matrix 501 =: [A . B,f, g] where, in

the case n = 1 the set A is null, in the. case. ] < n < K0 A consists
of all fractions of the form £/(/?-—1 ) for 0 k < n—l, and in the
casen = X 0 it consists of all fractions Jellfor 0 ^ k < l;further the
set B is equal to { 1 } and the f unctions f and g are defined by the
formulas'. f(x,y) — m in (l, 1 —x-\-y), g(x) = l —x.
As Lindenbaum has shown, the system Z Ko is not changed
if, in the definition of this system, the set A o f all proper frac­
tions is replaced by another infinite sub-set of the interval
(0, I):

T heorem 16. Let 50} — [A, B , f , #] be a matrix where B = { 1 },

the functions f and g satisfy the formulas
f(x,y) = min(l, 1 —x+y)> 9{x) = 1 —
and A be an arbitrary infinite set of numbers which satisfies the
condition: 0 < x < 1 for every x e A, and is closed und<er the two
operations f and g; then (£(SR)
Prom D ef. 7 the follow ing facts established by Lukasiewicz
are easily obtained:

T heorem 17. (cl) Lx = 8, L2 = L;

(&) if 2 m ^ X q, 2 ^ X0 and/ 1 is a divisor of m — 1,
then Lm ^ Ln;
(6) -A. = IT
l<n<K«
A,-
T heorem 18. All systems L n for 3 < n Ko are consistent
but not complete: Ln e S . 308— 35.
The oonverse of Th. 17 6 was proved by Lindenbaum:
T heorem 19. For 2 < m < N0 and 2 < w < X0 we have
Lm c: Ln and only if n— i a divisor of m— 1 .1

1 Lindenbaum gave a lecture at the first congress of the Polish mathemati­

cians (Lw6w, 1927) on mathematical methods of investigating the sentential
calculus in which, among other things, he formulated the above-mentioned
theorem. Cf. his note ‘ M6thodes mathdmatiques dans les recherches snr le
system© de la theorie de deduction’, Ksiqga Pamiqtkowa Pierwszego PoteJciego
Zjazdu Matematycznego, Krakow (1929), p. 36.
IV, §3 S E N T E N T IA L CALCULUS 49

Th. 17c was im proved b y Tarski by means o f Th. 16:

T heorem 20. Ln for every increasing sequence
1 < i < K»
n* of natural numbers.
Concerning the problem o f the degree o f completeness for
systems Ln the follow ing partial result has been obtained.
T heorem 21. I f n— I is a prime number (in particular if
n = 3), then there are only two systems, namely S and L> which
contain Ln as a proper part; in other words, every sentence
x e S - - L n satisfies one of the formulas: Cn(Lu-\-{x^ = S or
Cn(Ln+ {x }) - L ; y(Ln) - 3.
This theorem was proved for n = 3 by Linaenbaum ; the
generalization to all prim e numbers given in the theorem is due
to T arski.f
Regarding the axicm atizability o f the system Ln we have the
following theorem which was first proved by W ajsberg for n = 3
and for all n for which n — 1 is a prim e number, and was later
extended to all natural numbers b y Lindenbaum :
T heorem 22. For every n, 1 < n < X0, we have Ln e 91.
The effective p roof o f Th. 22 enables us to give a basis for
every system Ln where 1 < n < Xo- In particular W ajsberg
has established:
T heorem 23. The set X consisting of the sentences (CpCqp\
*CCpqCCqrCpr\ 'C C N p N q C q p ‘ CCCpNppp’ forms a basis of
£g, i.e. X e 5&(Lf).
The follow ing theorem o f W ajsberg is one o f the generaliza­
tions o f Th. 22 at present know n:

| In May 1930 while the original printing of this article was in progress,
Th. 21 was improved and the problem of the degree of completeness was
•olved for systems L n with an arbitrary natural n ; this was a joint result of
members of a proseminar conducted by Lukasiewicz and Tarski in the Univer-
ttty of Warsaw. A proof of Th. 21 and its generalizations appeared in print
reoently; see A. Rose, 'The degree of completeness of m-valued Lukasiewicz
propositional calculus’, The Journal of the London Mathematical Society,
▼ol. 27 (1952), pp. 92-102. The solution of the same problem for has been
given in A. Rose, ‘ The degree of completeness of the N0-valued Lukasiewicz
propositional calculus’, The Journal of the London Mathematical Society,
▼oL 28 (1953), pp. 176-84.
60 IN V E S T IG A T IO N S IN TO T H E TV, f 3

T heorem 24. Let 9K = [A, B ,f,g] be a normal matrix in

which the set A-\-B is finite. I f the sentences
:CCpqCCqrCpr’, ‘CCqrCCpqCpr', ‘ QQqrCpp',
' GCpqCNqNp\ ‘ CNqCCpqNp’

are satisfied by this matrix, then e iJ.|

The Ths. 8,10, and 16 o f § 2 can be applied to the systems Ln.

A ccordingly we have:

T heorem 26. Every system Ln, where 2 ^ n- < K0, possesses,

for every natural number m (and in particular for m =-• 1), a basis
which has exactly m elements.
T heorem 26. In every basis (and in general in every axiom
system) of ike aystern Ln at least three distinct sentential variables
occur.
As regards the problem o f extending Th. 22 to the case n = Ka
Lukasiewicz has form ulated the hypothesis that the system L*t

*j There L3 a comprehensi ve .iterature related to The, 22-2-1 and, more

generally, concerning the axiomatizability of various systems of sentential
calculus, We list a few papers on this subject in which further bibliographical
references can also be found: M. W&jsberg,' ‘ Aksjomatyzaeja trOjwartoscio-
wego rachunku zdah’ Axiomatization o f the three-valued sentential calculus
in Polish), CompUs Bendus des stances de la SociStd des Sciences et- des Leltres
de Varsovie, vol. 24 (1S31/, ol. iii, pp. 126-48. A further relevant paper by
the same author; ’ Beiirage zum Mefcaaussagenkalklil I *, Monatsheftc fCr Mathc-
matik und Physik, vol. 42 (1935), pp. 221—42. B. Sobocihski, *Aksjomatyzaeia
pewnych wieiowartosciowych systemdw teorji dedukcji ‘ (‘ Axiomatization
of certain many- valued systems of the theory of deduction * in Polish). Bocznik:
prao naukr.i&ych Zrzeszenia AsyeterUdw Uniucrsyteiu Jdzsfa PilsudsHego w
V/arsiawic. vol, 1 (1936), Wydfcial Matematycss^o Przyrodniczy JSTr. pp.
399-419, J. Slapecki, *Dow6a aksjom&tyzowalno^ci peinych sysfcemdw wielo*
wartoseiowyoh rachunku -dah ’ (*A proof of the axiomatizability o f functionally
complete systems o f many-valued sentential calculus \ in Polish), Comptes
Bendus des stances de la Society des Sciences et des Lettres de Vareovie, cl. iii,
vol, 32 (1939), pp. 110-28; and by the same author: *PMny ur6jw£rto?e'‘6*wy
raohunsk zdah’ ('The full three-valued sentential calculus’, in Polish), Annate*
Universitaii* M iriae Cime-Sldodowsha, vol. 1 (1946 ?, pp. 193-209. J. B. Koasor
and A. R , Turquette, *Axiom schemes for w-valued propositional calculi
Journal of Symbolic Logie, vol. 10 (1946), pp. 6I-£2, and ‘ A note on the
deductive completeness of -valued propositional calculi’, ibid., vci. 14
(1949), pp. 219-26.
in the first paper of Wajsberg listed above we find a proof of Th, 23, in the
second a proof of Th. 24.
IV, S3 S E N T E N T IA L CALCULUS 61

is axiomatizable and that the set consisting of the following five

sentences
‘CpCqp’, ‘GGpqGGqrCpr’, ‘CCCpqqCCqpp,>
‘GCCpqCqpCqp’, ‘ CCNpNqCqp’
forms an axiom system for A *, t
It must be emphasized that, as defined here, the systems L n
for n > 2 have a fragmentary character, since they are incom­
plete and are only sub-systems o f the ordinary system L . The
problem o f supplementing these systems to form complete and
consistent systems which are at the same time distinct from L can
be positively solved, but only in one way, namely by widening
the concept o f meaningful sentence o f the sentential oalculus,
and by introducing, beside the operations o f forming implications
and negations, other analogous operations which cannot be
reduced to these two (of. also § 5).
Finally we may add that the number o f all possible systems
of the sentential oalculus was determined by Lindenbaum.
T heorem 27. 1 = 2*>, but I T ! = X04
This result was improved by Tarski as follow s:
T heorem 28. S . 2 B . § = 2* but f T t O T l = K0.§

f This hypothesis has pioved to be oorreot; see M. Wajsberg, ‘ Beitrage

mm Metaaussagenkallrtil 1% Monatshefte fur Mathemalik und Phymk, vol. 42
(1935), pp. 221-42, in particular p. 240. As far as we know, however, W ajs-
berg’s proof has not appeared in print.
The axiom-system above is not independent: C. A. Meredith has shown that
*OCCpqOqpCqp*is deduoible from the remaining axioms.
% A proof of the first part of Th, 27 (and in fact of a somewhat stronger
result) can be found in the paper of K . Sohroter (62 a), pp. 301 ft. The proof
of the seoond part of Th. 27 is almost obvious.
{ The proof of Th. 28 can be outlined as follows: For any given natural
number n *= 1 ,2 , 3,... let xn be the sentence which is formed by n symbols 4C 9
followed by n-f-1 variables *p \ Given any set N of natural numbers, let X N be
the set consisting of all sentences xzn where n belongs to N and of ail sentences
9m+l where n does not belong to N . It can easily be shown that the set Cn{XN)
coincides with the Bet of all those sentences which can be obtained from sen­
tencesof X y by substitution. Hence the set X y is consistent and can therefore
be extended to form a complete and consistent system Xy> On the other hand,
ifM and N are two different sets of natural numbers, then the sum of X jf and
X jf is clearly inconsistent, and hence the systems X jf and X y cannot be
identical. The remaining part of the proof is obvious.
52 IN V E S T IG A T IO N S IN TO TH E IV, §4

§ 4 . T h e R e s t r ic t e d S e n t e n t ia l C a l c u l u s

In investigations into the sentential calculus attention is some­

tim es restricted to those sentences in which no negation sign
occurs. This part o f the sentential calculus can be treated as
an independent deductive discipline, one which is still simpler
than the ordinary sentential calculus and will be called here the
restricted sentential calculus.
F or this purpose we must first o f all m odify the concept o f
m eaningful sentence b y om itting the operation o f form ing
negations from D ef. 1. In a corresponding way the concept o f
substitution is also sim plified, and this brings with it a change
in the concept o f consequence. After these modifications Th. 1
remains valid.
F or the construction o f closed systems o f the restricted sen­
tential calculus both o f the m ethods described in § 1 are used:
the axiom atic and the m atrix m ethod. B ut a logical m atrix is
now defined as an ordered triple [A, B ,f] and not as an ordered
quadruple (D ef. 3); consequently condition (4) in D ef. 4 o f a
value function disappears. Ths. 2 -4 remain valid.
The definition o f the ordinary system L+ o f the restricted
sentential calculus is com pletely analogous to D ef. 5, with one
obvious difference which is called for by the m odification in the
concept o f m atrix. This system has been investigated by Tarski.
From the definition o f the system its consistency and com plete­
ness are easily derivable; hence Th. 5 holds also in the restricted
sentential calculus. The axiom atizability o f the system is estab­
lished in the follow ing theorem :

T heorem 29.The set X consisting of the three sentences

{CpCqp\ ‘ C C pqC C qrC pr‘ CCCpqpp9forms a basis of the system
L +; consequently L + e 21.

This theorem originates w ith Tarski; it contains, however,

a sim plification com m unicated to the authors by P. Bem ays.
In fact the original axiom system o f Tarski included, instead
o f the sentence ‘ CCCpqpp9, a m ore com plicated sentence,
iv ,H S E N T E N T IA L CALCULUS 53

lCCCpqrGGprr\xj* The independence o f both axiom systems

was established by Lukasiewicz.
Ths. 8, 10, 11, 12, 14, and 15 from § 2 have been extended to the
restricted sentential calculus b y their originators. Tarski, in
particular, has succeeded in setting up a basis o f the system L+
consisting o f only a single sentence.2 Tw o simple examples o f
such sentences, each containing 25 letters, are given in the next
theorem. The first is an organic sentence and was found by
W ajsberg, the second is not organic and is due to Lukasiewicz:
T heorem 30. The set of sentences consisting either of the single
sentence ‘ CCCpqCCrstCCuCCrstCCpuCst'
or of the single sentence
‘ GCGpCqpCGCCCrstuCCsuCruvv’
forms a basis of the system L+.%
D ef. 7 o f the w-valued system Ln can be applied at once to
the restricted sentential calculus provided only that the con­
cept o f m atrix is suitably m odified. Ths. 16-22 as well as 24-26,
which describe the m utual relations among the various systems
Ln, determine the degree o f completeness o f the systems and
establish their axiom atizability, have been extended to the re­
stricted sentential calculus b y their originators. (In the case o f
Th. 21 this was done b y Tarski; for Th. 22 b y W ajsberg. In
Th. 24 the sentences with negation signs are to be om itted.)
The problem o f the axiom atizability o f the system L£9 is left
open.
Finally, the number of all possible systems of the sentential

1 Cf. Le&niewski, S. (46), p. 47, note. 2 Cf. Le6niewski, S. (46), p. 68.

t Th© original proof of Th. 29 has not been published. But a proof of
this result can easily be obtained by means of a method developed in M.
Wajsberg, ‘ Metalogisehe Beitr&ge ’, Wiadomotci Matematyczne, vol. 18 (1936),
pp. 131-68, in particular pp. 154-7; the derivations which are needed for
applying Wajsberg’8 method can be found, for example, in W . V. Quine,
System of Logistic, Cambridge, Mass. (1934), pp. 60 ff.
J More recently Lukasiewioz has shown that the sentence ‘ CCCpqrOOrpCsp *
can also serve as a single axiom for system L+ and that there is no shorter
sentence with this property. See J. Lukasiewicz, ‘ The shortest axiom of the
implicational calculus of propositions ’, Proceedings of the Royal Irish Academy,
stfotion A , vol. 52 (1948), pp. 25-33.
54 IN V E S T IG A T IO N S IN TO TH E IV, J 4

calculus, whioh was determ ined b y Lindenbaum and Tarski in

Ths. 27 and 28, also remains unchanged in the restricted sentential
calculus.^

§ 6. T h e E x t e n d e d S e n t e n t ia l C alcu lu s
B y the extended sentential calculus we understand a deduc­
tive discipline in the sentences o f which there occur what are
called universal quantifiers in addition to sentential variables
and the im plication sign.x F or the universal quantifier Lukasie­
w icz uses the sign ‘J J ’ whioh was introduced b y Peirce.2 W ith
this notation the form ula ‘JJ pq’ is the sym bolic expression o f
the sentence, ‘for slip , q (holds)’ . The operation which consists
in putting ths universal quantifier ‘ J J ’ with a sentential variable
x in front o f a given sentence y is called universal quantification of
the sentence y with respect to the sentential variable x, and is
3 In (46) Leibjevf.ki has described the outlines of a deductive system,
called by him Protothetic, which, compared with tho extended sentential
calculus, goes still further beyor.d the ordinary sentential calculus in the
respect that, in addition to quantifiers, variable functors are introduced. (In
the sentence ‘ Gpq ’ the expression *0* is called a Junctor, and *p* and ‘ g* are
called the arguments. The word functor we owe to T. Kotarbidski. In both the
ordinary and the extended sentential calculus only constant functors are used.)
In addition to this principal distinction, there are yet other differences between
ihe extended sentential calculus and the protothetic as it is described by
Le&niewski. In contrast to the extended sentential calculus, in the protothetic
only those expressions are regarded as meaningful sentences in which no free,
but only bound (apparent) variables occur. Some new operations (rules of
inference or directives) are also introduced by means of which consequences are
derived from given sentences, such, for example, as the operation of distri­
buting quantifiers, which is superfluous in the extended sentential calculus.
Finally it must be emphasised that LeSniewski has formulated with the utmost
precision the conditions whioh a sentence must satisfy if it is to be admitted
as a definition in the system of the protothetic, whereas in the present work
the problem of definitions has been left untouched. Article I belongs to proto-
thetio. A sketch of the extended sentential calculus is given in Lukasiewicz, J.
(51), pp. 154-69; this sketch rests in great part on results of Tarski (cf. loc.
oit. Preface, p. vii). The two-valued logio of Lukasiewicz (52 a) has many points
of contact with the extended sentential calculus. Finally, there are many
analogies between the extended sentential calculus and the functional calcu­
lus of Hilberi and Aokermarm (see Hilbert, D ., and Aokermann, W . (30),
especially pp. 84-85).
• The expression ‘ quantifier* occurs in the work of Peirce (58 a), p. 197,
although with a somewhat different meaning.

t The footnote concerning Th. 27 on p. 61 applies to the restricted

sentential calculus as well.
IV, J 5 S E N T E N T IA L CALCULUS 00
W
denoted b y ‘nx{y)' in m etam athem atioal discussions. This con­
cept is to be regarded as a prim itive concept o f the m eta-
sentential calculus.1
D ebtnition 8. The set 8 * o f all m eaningful sentences (o f the
extended sentential caloulus) is the intersection of aU those sets
which contain all sentential variables and are closed under the two
operations of forming implications and of universal quantifica­
tion {with respect to an arbitrary sentential variable).*
The operations o f form ing negations and o f existential quan­
tification (which consists in prefixing to a given sentence y the
existential quantifier ‘2 ’ with a sentential variable «), are n ot
considered here because, in the system o f the extended sentential
calculus in which we are interested, they can be defined w ith the
help o f the tw o operations previously m entioned. F or exam ple,
we can use the form ula ‘ Op J J qq’ as definiens for *Np\
In deriving consequences from an arbitrary set o f sentences
Lukasiewicz and Tarski make use o f the operations o f insertion
and deletion of quantifiers, in addition to those o f substitution*
and detachm ent. The first o f these operations consists in obtain­
ing a sentence y = c(z, nt{u)) from a sentence of the form
x — c(z, u), where z e S* and u e 8 X, under the assum ption that
t is a sentential variable which is not free in z.* The second
operation is the inverse o f the first and consists in deriving the
sentence x — c(z, u) from the sentence y — c(z, irt(u)) (in this
case w ithout any restriction concerning the variable £}.5
1 Cf. the remarks following Def. 1 in § 1.
3 In contrast to Hilbert, D ., and Aekermann, W . (30), p. 52, as well as
to the standpoint taken in Lukasiewicz, J. (51), p. 155, the expression wa(y)
is also regarded as meaningful when x either occurs as a bound variable in y
or does not occur in y at all.
8 The operation of substitution undergoes certain restrictions in the ex­
tended sentential calculus (cf. Lukasiewicz, J, (51), p. 100, and Hilbert, D .,
and Aekermann, W . (30), p. 54.)
4 We do not need to discuss the meaning of the expression occurs in the
sentence y as a free (or bound) variable ’ since it is sufficiently clear (cf. Lukasie­
wicz, J. (61), p. 166, and Hilbert and Aekermann (30), p. 64).
5 In the restricted functional calculus only the £rst operation is used.
Instead of the second an axiom is set up (of. Hilbert, IX, and Aokerzfiann, W .
(30), pp. 53-54). An analogous procedure would not be possible in ouroaloulus;
for if we drop the second operation the system L* to be discussed below would
not have a finite basis.
56 IN V E S T IG A T IO N S IN TO TH E rM 5
9. The set Cnx(X) o f consequences o f the set X o f
D e f in it io n
sentences (in the sense o f the extended sentential calculus) is the
intersection of all those sets which include the given set X ^ S*
and are closed under the operations of substitution and detach-
ment, as well as insertion and deletion of quantifiers.
W ith this interpretation o f the concepts S x and Cnx(X) Th.
1 from § 1 remains valid.
A s before, tw o m ethods are available for the construction o f
deductive system s: the axiom atic and the m atrix m ethods.
The second m ethod has not yet received a sufficiently clear
general form ulation, and in fact the problem o f a simple and
useful definition o f the concept o f m atrix still presents m any
difficulties. Nevertheless this m ethod has been successfully
applied b y Lukasiewicz in special cases, namely for the con­
struction o f the n -valued system s Lx (for n < K0) and in par­
ticular for the construction o f the ordinary system L x o f the
extended sentential calculus. The construction o f the systems
L x is precisely described in the following.
First let us introduce the following auxiliary
D e f in it io n 10.
notation: b = 'p\ g = 7r6(6) (falsehood), n(x) = c(x, g) for every
x g S x (the negation o f the sentence x), a(x, y) = c(c(x, y), y) and
h(xf y) = n(a(n(x), n(y))) for every x e S x and every y e S x (the
disjunction or rather alternation, and the conjunction o f the
sentences x and y);1 furthermore &<li(a^) = x1 for m ~ 1 and
kiLifai) = A ^ iL l1^ ) , xm) for every arbitrary natural number
m > 1, where xi e S x for 1 < i < m (the conjunction o f the
sentences xv x2,..., xm). Further we put bm = b for m = 1,
bm = c(n(b), 6m_i) for every natural number m > 1, and finally
am = iTb(c(bmyb)) for every natural number m.2
1 The logical expressions *Apq' {*p or g*) and 'Kpq' ('p and g’) correspond,
in the symbolism introduced by Lukasiewicz, to the metalogical expression*
'a{x, y)' and y)* respectively. Of the two possible definitions of the
alternation, which in the two-valued, but not in the n-valued, system are
equivalent: o(a?, y) = c(c(a;, y), y) and a(x, y) = c(n(x), y), the first was chosen by
Lukasiewicz for various, partly intuitive, reasons (cf. Lukasiewicz (50 a), p. 201).
* For example,
—‘p’y&»—'oop nppp’f &
»“ ‘oop nppocj) n ppp\
and ®i- *np° pp'< °t— ‘Tipoocp npppp'y
<»,=•npoccp nppocp npppp'-
IV, J5 S E N T E N T IA L CALCULUS 67

N ow let n be a definite natural number > 1. W e choose »

sentences called basic sentences, and denote them b y the sym bols
V > V ..... V ; in fa ct we put = g, fif2 = a „_lf and
j7m = c(«(<72), ? m_x) for every to, 2 < to < vi.1 L et 0 be the
smallest set o f sentences which contains all sentential variables
and basic sentences and is closed with respect to the operation
o f form ing im plications.
A function h is called a value function (of the n-th degree) if it
satisfies the follow ing conditions: (1) the function h is defined
for every sentence x e O; (2) i f z is a sentential variable, then
h(x) is a fraction o f the form (m — l)l(n— 1), where m is a natural
number and 1 < m < n; (3) for every natural number m,
1 < m < n9 we have h(gm) — (m—l)/(n—l); (4) i f x e 0 and
y e <?, then h(c(x9y)) = m in (l, 1—h(x)+h(y)).2
W ith every sentence x e S* a sentence f(x) e O is correlated
b y recursion in the follow ing w ay: (1) if a; is a sentential variable
or a basic sentence, then f(x) = x; (2) if x e S * , y e S x , and
c(x9 y) is not a basic sentence, we put f(c(x9 y)) = c(f(x)9f(y));
(3) if a; is a sentential variable whioh is not free in the sentence
y e S * 9then f(rrx(y)) = f(y)\ (4) but i f the sentential variable x
is free in the sentence y e 8 X and rrjy) is not a basic sentence,
then we put f(nx(y)) = {fiVi)) where the sentence yi for
every i 9 1 < i < n9 arises from y b y the substitution o f the
basic sentence gi for the free variable x.
The n-valued system Lx of the extended sentential calculus,
where 2 < n < X0, is now defined as the set o f all those sen­
tences x e 8 X which satisfy the form ula h(f(x)) = 1 for every
value function h (o f the nth degree); in addition L x is set equal
to j8 x . The system L$ = L* is also called the ordinary system
of the extended sentential calculus.8
1 For example, for n = 3:
0* = 'JJpOOOpJJpppp',
0t - ‘ C C n p C O O p U p p p p U p p U p o c o p U p p p p '-
1 Of. Defs. 4 and 7 above.
s In the definition adopted by Lukasiewicz, instead of the basic sentences
0i» 0s>“ ‘» 0»» there occur what are called sentential constants, clt clv.., cn, i.e.
special signs distinct from sentential variables. The concept of meaningful
sentence is thereby temporarily extended. The rest of the definition rims
quite analogously to the definition in the text. In the final definition of the
58 IN V E S T IG A T IO N S IN TO TH E IV, §5

From this definition o f the system s L* the follow ing facts

easily result (they are partly in opposition to Tha, 18 and 19
o f § 3);
T heorem 31 . L% e <5 . 308. SBfor every natural n, 2 < n < Ko.1
T heorem 32 . For 2 < m < K0 and 2 < n < X0 we have
Z& £ L* if and only if m =» n (no system of the sequence L * ,
where 2 < n < X0, is included in another system of this sequence).
T heorem 33 . The set of all sentences of the system L% (where
1 <; n < Rg) in which no bound variables occur is identical with
the corresponding system L% of the restricted sentential calculus.
Regarding the axiom atizability o f these systems Tarski has
shown that The. 8, 10, 22, 29, and 30 also hold in the extended
sentential calculus. In this connexion Tarski has also proved
the follow ing:
T heorem 34 . Every axiom system of the system L% = L+ in
the restricted sentential calculus is at the same time an axiom
system of the system = L 'A in the extended sentential calculus.1
On the other hand} not every basis of the system L+ in the restricted
sentential calculus is at the same time a basis in the extended
calculus (and not every set of sentences which is independent in the
restricted sentential calculus remains independent in the extended
calculus).
T heorem 35 , For 3 < n < X0 universal quantifiers and, bound
variables occur in at least one sentence of every basis (and in general
of every axiom system) of the system L * .
systems all expressions which contain sentential constants are eliminated,
and the concept of meaningful sentence is reduced to the original expressions.
By moans of the modification introduced in the text, which is due to Tarski,
the definition of the systems L„ certainly takes on a simpler form from the
metalogical standpoint, but at the same time it becomes less perspicuous. In
order to establish the equivalence of the two definitions it suffices to point out
that the expressions chosen as basic sentences satisfy the following condition;
for every value function h (in the sense of the original definition of Lukasiewicz),
m—1
W/tom)) ** H °m ) ------ 7 * uikenre 1 < m < n.
n—x
1 The completeness and axiomatizability of the system L£ was proved by
Tarski in the year 1927. Hie proof was subsequently simplified by S. Jankowski.
IV, §6 S E N T E N T IA L CALCULUS 59

I t is w orthy o f note that the p roof given b y Tarski o f Th.

22 in the extended sentential calculus makes it possible to
construct effectively an axiom system for every system
1% (3 < K0). R elatively sim ple axiom system s o f this kind
were constructed b y W ajsberg; in the case n = 3 his result is
-as follow s:
T h eo rem 36. Let X be the set consisting of the follovnng sen­
tences:
‘ CCCpqCrqCCqpCrp\ *OpCqp’ , ‘ CCCpGpqpp
‘c n pcccp xx ppppG n pcccp h pppp n pp\
‘ CO n pCCCp X I PPPP I I P P I I pOCCp X I PPPP';
then X 6 $tx(Zg).
A n exact definition o f the denum erably-valued system L$e
o f the extended sentential calculus presents m uch greater diffi­
culties than that o f the finite-valued systems. This system has
not yet been investigated.
Ths. 27 and, 28, which determine the number o f all possible
system s, remain correct in the extended sentential calculus.
In conclusion we should like to add that, as the simplest
deductive discipline, the sentential calculus is particularly suit­
able for m etam athem aticai investigations. It is to be regarded
as a laboratory in whioh m etam athem aticai m ethods can be
discovered and m etam athem aticai concepts constructed which
can then be carried over to m ore com plicated m athem atical
systems.
 V
FUNDAMENTAL CONCEPTS OF TH E
M E T H O D O L O G Y OF TH E D E D U C T IV E
SCIENCES*)-
I n t r o d u c t io n
T he deductive disciplines constitute the subject-m atter o f the
methodology of the deductive sciences, which today, follow ing
H ilbert, is usually called metamathematics, in m uch the same
sense in which spatial entities constitute the subject-m atter o f
geom etry and animals that o f zoology. N aturally not all
deductive disciplines are presented in a form suitable for objects
o f scientific investigation. Those, for example, are not suitable
which do not rest on a definite logical basis, have no precise rules
o f inference, and the theorems o f which are form ulated in the
usually ambiguous and inexact terms o f colloquial language—
in a word those which are not form alized. M etamathematical
investigations are confined in consequence to the discussion o f
form alized deductive disciplines.
Strictly speaking m etam athem atics is not to be regarded as a
single theory. F or the purpose o f investigating each deductive
discipline a special m etadiscipline should be constructed. The
present studies, however, are o f a more general character: their
aim is to make precise the meaning of a series of important meta­
mathematical concepts which are com m on to the special m eta­
disciplines, and to establish the fundamental properties of these
concepts. One result o f this approach is that some concepts which
can be defined on the basis o f special m etadisciplines will here
be regarded as prim itive concepts and characterized b y a series
o f axiom s.
A n exact p roof o f the follow ing results naturally requires,
besides the above-m entioned axiom s, a general logical basis.
t B ibliographical N ote . This article originally appeared under the title
*Fundamentals BegriSe der Methodologie der deduktiven Wissenschaften, I
in MonatshefU far Mathematik und Phyaik, vol. 37 (1930), pp. 361-404. For
earlier publications of the author on the same topics see the bibliographical
note to m .
V M ETH ODOLOGY O F D E D U C T IV E S C IE N C E S 61
This basis should n ot be thought o f as too com prehensive: for
exam ple, those chapters o f the celebrated work o f W hitehead
and Bussell1which com prise the sentential calculus, the theory o f
apparent variables, the calculus o f classes, and the elements o f
the arithm etic o f cardinal and ordinal numbers are quite
sufficient for our purposes. The axiom o f choice is not used in
this discussion, and the axiom o f infinity can also be easily
elim inated.
As regards notation, we shall for practical reasons form ulate
these considerations in the terms o f everyday language; but we
shall also make use o f a series o f sym bols, almost all o f which are
custom ary in textbooks and articles from the dom ain o f set
theory.
The variables V , cy\ V ... here denote individuals (objects
o f low est type) and in particular sentences; the variables
*B\ CX \ T \ .., sets o f individuals; ‘B ’, classes o f sets (or
systems o f sets) and finally ‘A\ V , ordinal numbers. The
sym bols o f the calculus o f classes and arithm etic, *c , ’
V, T T> ‘ < \ e^c *> are used with their usual
m eanings; ‘a / denotes the smallest ordinal number o f the
second and ‘£2’ o f the third number class. The formulas *x e A ’ or
‘x e A ’ express as usual that the set A respectively contains or
does n ot contain the element x. ‘A ’ denotes the cardinal o f the
set A . The sym bol MSf...]* denotes the set o f all values o f the
/(*)
function / corresponding to those values o f the argument x
which satisfy the condition form ulated in the brackets c[...]\ In
particular we put {#} = E[y = x] (the set which contains x as
v
its only elem ent), ?P(A) = E [X c A ] (the power set o f A ),
.3T
= E[A Q X ], (E = E [S < K0] (the class o f all finite,
x x ________
‘inductive’ sets), and £ = E[-q < £] (the cardinal number o f
i)
the ordinal num ber £).
The results o f the present article are applicable to arbitrary
deductive disciplines, and in particular to the simplest deduc­
tive discipline, the sentential calculus. This article was originally
1 Whitehead, A. N., and Russell, IL (90).
68 F U N D A M E N T A L CO N CEPTS OF V

intended as the first part o f a m ore com prehensive paper. The

discussion in the second part dealt w ith those deductive disci*
plines which presuppose a certain logioal basis, including at
least the whole o f the sentential calculus (approxim ately in the
sense that all the valid statem ents o f the sentential oalculus
can be used as premisses in all reasoning within these deductive
disciplines). This second part was developed independently o f
the first part and was based upon a special axiom system . In
its original form the second part has never appeared in print.1
In conclusion it should be noted that no particular ph ilo­
sophical standpoint regarding the foundations o f m athem atics
is presupposed in the present w ork. Only incidentally, there­
fore I m ay m ention that m y personal attitude towards this
question agrees in principle with that which has found em phatic
expression in the writings o f S. Lefeiew ski3 and whioh I w ould
call intuitionistic formalism,.^

§ 1. Meaningful Sentences; Consequences of Sets of

Sentences
From the standpoint o f m etamathematics every deductive
discipline is a system o f sentences, which we shall also oail
meaningful sentences.* The sentences are m ost conveniently
regarded as inscriptions, and thus as concrete physical bodies.
N aturally, not every inscription is a m eaningful sentence o f
a given discipline: only inscriptions o f a well-determ ined
structure are regarded as meaningful. The concept o f meaning­
ful sentence has no fixed content, and must be relativized to a
1 A modified version of the second part is to be found in XII. The discus­
sion in Hertz, P. (27) has some points of contact with the present exposition.
As examples of Generate investigations (within special mel&disoiplines) based
upon the conceptual framework of this paper we may mention IV and
Presburger, M, (61).
3 Cf. especially Le&oiewski, S. (46), especially p. 78.
* Instead of ‘meaningful sentence* we could say ‘well-formed sentence*.
I use the word ‘meaningful’ to express my agreement with the doctrine of
intuitionistic formalism mentioned above.

f This last sentence expresses the views of the author at the time when
this article was originally published and does not adequately reflect his present
attitude.
V ,§ 1 M ETHODOLOGY O F D E D U C T IV E S C IE N C E S 03

concrete formalized discipline. On the basis o f the special

metadisdplines mentioned in the introduction this oonoept can
be reduced to intuitively simpler oonoepts;1 in the present
work, however, it must be taken as a primitive concept. The
set o f all meaningful sentences will be denoted by the symbol ‘S ’ .
Let A be an arbitrary set o f sentences o f a particular dis­
cipline. With the help o f certain operations, the so-called rules
of inference, new sentences are derived from the set A , called
the consequences o f the set A . To establish these rules o f inference,
and with their help to define exaotly the oonoept o f consequence,
is again a task o f special metadisciplines; in the usual terminology
o f set theory the schema of such a definition can be formulated
as follows:
The set o f all consequences of the set A is the intersection o f all
sets which oontain the set A and are closed under the given rules
o f inference.®
On account o f the intended generality o f these investigations
we have no alternative but to regard the concept o f consequence
as primitive. The two concepts—of sentence and of consequence—
are the only primitive concepts which appear in this discussion.
The set o f all consequences o f the set A o f sentences is here
denoted by the symbol ‘ Gn(A)’ .
W e will now state four axiom s (A x. 1-4) which express
certain elem entary properties o f the prim itive concepts and are
satisfied in all known form alized disciplines.
We assume in the first plaoe that the set S is at most
denumerable.
A x io m 1. S < K0.
This axiom scarcely requires comment since the contrary
hypothesis would be unnatural and would lead to undesired
complications in proofs. Moreover, the question arises whether
the assumption that the set S is infinite (even denumerably in­
finite) is consistent with the intuitive view o f sentences as oon-
1 See rv, p. 39 (Def. 1) and p. 65 (Def. 8).
1 Cf. IV , p. 40 (Def. 2) and p. 66 (Def. 9). A set is said to be closed under
given operation* if each of these operations, when applied to elements of the set,
always yields again an element of the set (of. FrSchet, M. (17)).
34 F U N D A M E N T A L CONCEPTS OF V,§1
Crete inscriptions. W ithout going deeper into this disputable
question, which is irrelevant for our discussion, it m ay be noted
here that I personally regard such an assum ption as quite
sensible, and that it appears to m e even to be useful from a m eta-
m athem atical standpoint to replace the inequality sign b y the
equality sign in A x. l . f
From the definition schema o f the set o f consequences form u­
lated above it follow s that every sentence which belongs to a
given set is to be regarded as a consequence o f that set, that the
set o f consequences o f a set o f sentences consists solely o f
sentences, and that the consequences o f consequences are them ­
selves consequenoes. These facts are expressed in the next tw o
axiom s.
A xiom 2. I f A £ S, then A £ Cn(A) £ S.
Axiom 3. I f A £ S, then Cn(Cn(A)) = Cn(A).
F inally it should be noted that, in concrete disciplines, the
rules o f inference with the help o f which the consequences o f a
set o f sentences are form ed are in practice always operations
which can be carried out only on a finite number o f sentences
(usually even only on one or tw o sentences). H ence every
consequence o f a given set o f sentences is a consequence o f a
finite subset o f this set and vice versa. This can be shortly
expressed b y the follow ing
A xiom 4. I f A £ 8, then Cn(A) = T CntX).
x>mk)-9
B y w ay o f exam ple we give here som e elem entary conse­
quences o f the above axiom s.
T heorem 1. (a) I f A £ B £ 8, then Cn(A) £ Gn{B).
[A x. 4]1
(b) I f A-\-B £ S, then the formulas
A £ Cn(B) and Cn(A) £ Cn(B)
are equivalent. [A x. 2, A x. 3, T h. la ]

1 In cases where a proof is clear I content myself with a reference to the

theorems to be used.

t Compare the discussion of related problems in Y in, pp. 174 and 184.
V, §1 M E T H O D O L O G Y OF D E D U C T IV E SCIENCES 65

In accordance w ith Th. 1 a the operation Cn in the dom ain

o f sets o f sentences is m onotonic; this property can be expressed
in several equivalent form s, e .g .:

gC M X ) s < * .( £ 4 or s flO H Z ).

for every non-null class St £ $ ( 5 ) .

T heorem 2. (a) I f A + B £ S, then
C n (A + B ) = Cn{A+C n(B )) = Cn(Cn(A)+Cn{B));
(b) if, more generally, St is an arbitrary class £ $p(S), then

/ ) =
Proof, (a) A ccording to A x. 2,
A + B £ A + C n {B ) £ Cn{A)+G n(B) £ 8,
whence b y Th. 1 a
(1) C n (A + B ) £ Cn{A+C n(B)) £ Cn(Cn(A)+Cn(B)).
A gain according to Th. 1 a and A x. 2 we have
Cn(A) £ C n (A + B ) £ 8 and Cn(B) £ C n (A + B ) £ 8,
thus also On{A)-\~Cn(B) £ Cn(A-{-B) £ 8 and therefore
(2) Cn{Cn{A)+Cn(B)) £ Cn(Cn(A+B)).
F inally, from A x. 3 it follow s th at
(3) Cn(G n(A+B)) = G n(A +B ).
The form ulas (1) to (3) give im m ediately:
C n (A + B ) = Cn(A+G n{B)) = Cn(Cn(A)+Cn(B)), q.e.d.
(b) is proved in an analogous manner.
I f A £ 8, 0 e (£, and G £ Cn(A), then there
T h e o r e m 3.
exists a set B which satisfies the formulas B e d , B £ A , and
0 £ Cn(B).
Proof. B y the hypothesis and A x. 4, a set G(x) oan be corre­
lated uniquely w ith every elem ent x e C, in such a w ay that the
form ulas
(1) x e Cn(G(x)), G{x) e <£, and G(x) £ A for x e C
are satisfied. (Since the set G is finite, the existence o f such a
66 F U N D A M E N T A L CONCEPTS OF V, § 1

correlation can be established w ithout using the axiom o f

ohoioe).
I t follow s from (1) that
(2) C s TCn{0(x)).
0€C
W e put

(3) £ -2 <?(*).
XGC

Since C e (£, we infer from (1) and (3) that

(4) B e (E and B £ A.
From (3) and (4) we also obtain 0{x) £ B £ A £ S for
x e G. A ccording to Th. l o we thus have Cn(G(x)) £ Cn(B)
for every elem ent x o f C, whence 2 Cn(G(x)) £ Gn(B); and
®c<7
com bining this inclusion w ith (2) w e obtain
(5) C£
B y (4) and (5) the set B satisfies the conclusion.
T heorem 4. Lei Si be a class which satisfies the following
condition: (a) for every finite subclass SI of Si there exists a set
Y e Si, such that £ I . I f also Si £ $ (£ ), then
XeS
C n ( g X ) = J ',Cn(X).

Proof. Let x be a sentence such that

(1) z e C n Q g X ).

Since b y the hypothesis ^ X £ S, the existence o f a set Z

which satisfies the form ulas

(2) Ze<£, f s T I ,
xSt
and
(3) a: 6 Cn{Z)
follow s from (1) and from A x . 4. From (2) can further be in­
ferred the existence o f a finite subclass £ o f Si, such that
Z £ y X . A ccording to the prem iss (a) there corresponds to
■»8
V, § 1 M E T H O D O L O G Y OF D E D U C T I V E S C I E N C E S 67

the class £ a set Y e ft such that T I £ 7, Consequently we

X?£
have
(4) Y e it and .£ £ F.
By Th. la , from (4) it follows that Cn{Z) £ On(Y), and
therefore by (3) * e Cn(Y); hence with reference to (4) we obtain
(5) x s T Cn(X).

Thus we have shown that the formula (1) always implies (5},
accordingly the following inclusion holds:
(«> Cn{ 2 X) s T Cnpq.
\Xeft ' Xsft
On the other hand we have 7 £ T I £ 5, whence accord-
mg to Th. la Cn{Y) £ Gni T X^i for every set F e conse-
'w?vs /
quentiy
,7) £ < * * > s < * (£ * ).
The formulas (6) and (7) give

An immediate consequence of the last theorem is

Cobgllaby 5, Let R be a class which satisfies the following
condition: (at) R 0 and for any two sets X and Y belonging
to R cither X G F Of F G X. If R G %[8), then

0n( i / ) ~ L 0 n m - P * - 4!
Tins corollary is often applied to the class of all terms of an
increasing sequence of sets of sentences.
T heorem 6. Lei- B + C £ 8, and pvt
F{X) = C.Cn(X-t-B)
for every set X £ 8 (and in particular F(X) ~ Cn{X+B) in
case C ~ S). We then have:
(a) 6 < K0;
(b) if A £ Gs then A £ F(A) £ 0:
(c) if A £ 0, then F(F(A)) - F(J),
(d) if A £ G, then F{A) = J F(X\.
xetifto.e
68 F U N D A M E N T A L CONCEPTS OF V, § 1

{In other words, Axs. 1-4 remain valid if 'S' and 'Cn are every­
where replaced in them by *C’ and (F' respectively.)
Proof. First let it be noted that in case 0 = S we have, b y
A x. 2, Cn(X-\-B) £ C for every X £ Sy so that the function
F(X) = C . Cn{X-\-B) reduces indeed to
F(X) = Cn[X+B).
(а) follow s im m ediately from A x. 1.
(б) is easily obtained from A x. 2.
(c) B y A x. 2 and Th. 1 a we have
C . C n { A + B ) + B £ Cn(A-uJB)+B £ £
and accordingly
G »(C . C n ( A + B ) + B ) £ Cw(Cn(.4+ £ ) - } - B);
in view o f Th. 2 a
Cn{Cn(A + B) + B) = On((A -j-.Z?) + i3) - (7n(A+B).
From this it follow s that
G\Cn(C.Cn{A + B ) + B ) £ 6bC%(A + iJ );
thus b y the hypothesis
F(F(A)) £ F (A ).
Since the inverse inclusion F(A) £ JF(F(A)) results im m e­
diately from (b) (if we replace ‘A* in it by ‘ F (A )’ ), we finally
obtain the desired form ula F(F(A)) = F (A ).
(d) B y A x. 4:
(1) C n (A + iJ ) = y Cn(X).

For every set X e we obviously have X £ X j - f Bt

where X x ~ A . X e ^P(A); from this by Th. 1 a we have
Cn{X) £ Cn{Xx+ B ) .
If, m oreover, I g S, then X x e G and accordingly X x e G. $(A )«
Thus:
(2)

On the other hand, if X e then X -f-B £ A + B , whence

C n {X + B ) £ Cn(A+B)i therefore
(3)
V, §1 M E T H O D O L O G Y OF D E D U C T I V E S C I E N C E S 09

The form ulas (1)—(3) im m ediately give

C n(A +B ) = y C n { X + B },
and thus
O .C n (A + B )= y C .C n (X + B ).
JTeCg.iP (A)

B y the hypothesis it follow s from this that

F {A )= y F(X), q.e.d.
X ed M *U )

In consequence o f the above theorem, the concepts in ­

vestigated in this discussion can be relativized in tw o distinct
directions: 1. Instead o f considering all meaningful sentences we
can restrict our attention to the elements o f a given set C o f
sentences— a sentential domain. 2. A fixed set o f sentences B> a
basic set, is chosen, and in the form ation o f the consequences o f an
arbitrary set X o f sentences all the sentences o f the basic set
are added to the set X , so that the consequences o f the set X
in the new sense are identical with the consequences o f the set
X - f 2? in the old sense. A ccording to Th. 6 (above) such a
m odification in the meaning o f the prim itive concepts does not
affect the validity o f the underlying axioms. Therefore all
consequences o f these axiom s also remain valid, and in particular
those theorems which will be presented in the follow ing sections;
one must only see to it that the non-prim itive concepts in these
theorems undergo an analogous relativization. On these grounds
the theorem in question deserves the name relativization theorem.

§ 2. D eductive (Closed) Systems

W ith the help o f the tw o concepts introduced in the preceding
section, those o f sentence and consequence, alm ost all basic
concepts o f m etam athem atics can be defined; on the basis o f
the given axiom system various fundam ental properties o f these
concepts can be established.
In the first place an especially im portant category o f sets o f
sentences w ill be singled out, nam ely the deductive systems.
Every set o f sentences which contains all its consequences is
70 F U N D A M E N T A L CONCEPTS OF V, 5 2

called a deductive system, or possibly a closed system, or sim ply a

system.1
D eductive systems are, so to speak, organic units which form
the subject m atter o f m etam athem atical investigations. Various
im portant notions, like consistency, completeness, and axio*
m atizability, which we shall encounter in the sequel, are theo­
retically applicable to any sets o f sentences, but in practice are
applied chiefly to systems.
The class o f all system s is denoted by the sym bol
D efinition 1. S = E[Cn(X) g I c: $].
x
A n easy transform ation o f the above definition yields
T heorem 7. In order that A e S , it is necessary and sufficient
that Cn(A ) ~ A ^ S. [D ef. 1, A x. 2]
Further properties o f system s are expressed in the following
theorem s:
Theorem 8. // B s <5, then the conditions (a) A £ B and
(;8) A £ S and Cn(A) c; B are equivalent. [Th. 16, 7]
T heorem 9. (a) I f A £ 8, then Cn(A) e S and also
Cn(A) — n X-
X e Q .J t).< 5

(b) In particular, Cn(0) e S and Cn{0) == X.

Proof. (a) B y Axs. 2 and 3,
A £ (7 r i(A ) and Cn(Cn(A)) C n (A ) g 5',

whence b y Th. 7 Cn(A) e Q (A ). ®. A ccordingly

n C?n(4);

on the other hand from Th. 8 it follow s that Cn(A) £ X for

every set X e £ i(A ). S , thus that Gn{A) £ ]T X , Conse-
Xfc.Gui).<3
quently we have C'n(A) e Q (A ). <5 and XT X,
X eX i(A ).5
q.e.d.
(6) results im m ediately from (a) if we pat ^4 =-= 0.
1 The term ‘ deductive system ’ was used in earlier publications of the author;
see III and the bibliographical note to that article. To Hertz, P. (27) ‘ closed
system’ is used; and in Zermelo, E. (92) wo find the phrase logically closed
system* used in the same sense.
V, § 2 M E T H O D O L O G Y OF D E D U C T ! Y E S C I E N C E S 71

In accordance with the above theorem On{A) is the smallest

closed system o v e r A (i.e, the smallest o f all systems which
include the set A), and Cn{0) is the smallest system in general.
The system Cn(0) can be called the system of all logically provable
or logically valid se n te n c e s).

T heorem 10,
8 e <S and 8 — T X. [D e l 1, A x. 2]
.X5--0
8 is thus the largest o f all system s.
T heorem 11. (a) I f A e 5 and. £ e <3 , th en A. B e S.
(6} In generala if it sl: 3 and S\ 0, ik e n ^ ^ X e <5.

Proof, (a) From D e l 1 we obtain Cn(A) s A £ 8 and

Cn{B) c 2? cr B y Th. i a we have Gn{A*B) s Gn{A) and
O n (A .j3) c thus Vn(.t.B) ^ Cn{A).Cn(B). A ccord ­
ingly Cn{A.B) s <4. i>* c: s whence, by D e l l , i . £ e S , q.e.d.
(b) is proved analogously.
In contrast to the preceding theorem, a sum o f systems is n ot
always a new system . h\ this connexion only the follow ing can
be proved,
T heorem 12. I f the clam Si satisfies th e condition (a) of Th. 4
and if & cr 6 , th e n T AT g S,

Proof. B y D e l 1 it follow s from the form ula it s S that

it £ ^P(aS). Th© premisses c *f'Th. -$ (or o f Cor. 5) are thus satisfied,
whence Cn( T x ) =*- T Ch?>(A'j.

F rom this last form ula and. the inclusion it £ S it follow s

b y Th. 1 that <7n{ V X ) ^ 5‘ s and consequently that
T i e s , q .e l

The theorem above establishes a sufficient condition for a class

o f system s to be such that The sum o f all systems o f this class is
itself a system . It can he shown that this condition is not
necessary. H ow ever, \t wo restrict ourselves to those deductive
disciplines which presuppose sentential calculus, and i f we
assume the class Si to be finite, t hen the converse o f Th. 12 holds
(a *1 'LU: \
72 F U N D A M E N T A L C O N C E P T S OF V, § 3

§ 3. L ogical (I n feren tial ) E quivalence of T wo S ets of

Sentences
Two sets of sentences are called logically or inferentially equi­
valent, or simply equivalent, if they have all their consequences
in common (:Lt\ their sets of consequences coincide). The sig­
nificance of the concept consists in the fact that almost every
property to be considered here applies to all sets equivalent to a
given set A whenever it applies to A,
The class o f all sets which are equivalent to a given set A of
sentences is denoted b y ‘ 3tq(.4}*; the formula ‘jBe$la(Ay
accordingly states that the sets A and B are equivalent. But we
shall not introduce a special relation symbol for the equivalence
o f sets.
D efinition 2. 2tq(.4) = .Ej\.l~'r X $= £ and On{A) ~*= Cn(X )].

T heorem 13. (a)The formulas A £ 8 y A e ^ q { A ) and

* (\{A) * 0 are equivalent. [Def. 2]
(b) The formulas A e 'itq(B ), B e 2lq(.4), ) . 5lq(J5) ^ 0
and 2Iq(.4) — 3lq(B) ^ 0 are equivalent. [D ef. 2]
T h eorem 14. I f A e B s C and A e 3tq(C), then B e $ q (C )
and S q (^ ) = $tq(B) = Jtq(C).
Proof. B y virtue o f D ef. 2 it follow s from the hypothesis
that A S B s G £ 8 and Cn(A ) = C'w(C); from this b y Th.
l a we obtain Cn(A) c Cn{B) s Cn(0) and Cn(B ) = Cn{C).
B y another application o f D ef. 2 we then obtain B e 5tq(0),
which finally on the basis o f the hypothesis and o f Th. 136
yields the second o f the required form ulas,
« q ( 4 ) = Utq(B) = «q(<7).
T heorem 15. (a) I f A e 3la(.B) and G e S, then
A + CeKqlB+C),
(6) I f A , e 3Iq(B1) and A 3 e 3tq(JS2), then
-41+^42 e iSlq(.8-s4'-®2)-
(c) In general, i f to every set X e 51 a set Y e £ satisfying the
formula X e 3tq(T) corresponds, and vice versa, then

£ x * M £ r )-
V, §3 M E T H O D O L O G Y OF D E D U C T IV E SCIENCES 73

Proof, (a) B y means o f D ef. 2 the hypothesis yields

(A +C )+(B +C ) £ S and Cn(A) = Cn(B).
Prom this b y applying Th. 2 o we obtain
Cn (A+C ) = Cn{Cn(A)+C) = Cn(Cn(B)+C) = < 7n (£+ C ).
A ccordin gly b y D ef. 2 we have .4 - f O e $ q (i? + C '), q.e.d.
(6) and (c ) follow in an analogous w ay b y means o f Th. 2 o, b.
T heorem 16. For every set A £ 8 there is a corresponding
sequence of sentences xv o f type n cA which satisfies the formulas

(a) x v s Cn^ E\ji O J ) for every v < tt

and (fi) E[v < tt] e 2tq(J.).

Xy

Proof. In consequence o f A xs. 1 and 2 as well as o f the

hypothesis o f the theorem , the set Gn(A) £ 8 is at m ost de­
num erable; the elements o f this set can therefore be ordered in
an infinite sequence (with possibly repeating terms) o f the type co,
so that
(1) Cn(A) = E [ \ < a > ] c: S.
v\
L et
(2) A0 be the smallest of the numbers A < a> which satisfy theformula
yx e Cn{0);
Further, let
(3) for 0 < v < co, Av be the smallest of the numbers A < co which
satisfy the formula

F inally, let
(4) 7t be the smallest ordinal number with which the conditions (2)
and (3) do not correlate any number
W e put
(5) xv = y^ fo r v < n .

1 Where possibly n = 0 (the empty sequence, which has no terms at all,

is regarded as a sequence of type 0).
74 F U N D A M E N T A L CONCEPTS OF V,§3

I t follow s from (2 )-(4 ) that

(6) 7T ^ Oi.
From (l)-(5 } we easily obtain

(7) xv e Cn^E\jL < v]j for v < n

and
(8) E[v < tt] £ On(A).
Xv
L et us suppose that
(9) there exist numbers A < to such that y\ s Cn(E\y < tt]^.
A ccordin gly (in view o f (5)) let
(10) A' be the smallest of the numbers A which satisfy the formulas
A < to and y\ e Cnl E [v < rc]).

I f 7r were less than to, then, b y com paring the statements (2),
(3), and (10), we should obtain A' — A^, which contradicts the
condition (4). Consequently, on the basis o f (6),
(11) 7t = to.
In accordance w ith Th. l a and w ith reference to (l)-(3 ) we
have Cn^ E{ji < c Cn^ E[ji < 7r]^ for every v < tt. W e

thus have b y (10) e Cn(E[fi < v]^ for v < w and in particular

y# e 6%(0) for v == 0, whence, in view o f (2) and (3),

(12) A' ^ A„ for every v < w.
F inally it is easy to show that
(13) K < K for v, < v3 < n.
In fact, let < v2 < w. B y A x. 2 and Th. 1 a, the conditions
(1)~(4) im ply then the follow ing form ulas:

< r2] £ CnlE{ji < v2]\

and 0 » ( E[y. < v,]) c O « ( % < v j).

Vhfx VAft
A consequence o f the first inclusion is

1 'VA '
V, § 3 M E T H O D O L O G Y OP D E D U C T IV E SCIENCES 76

since b y (3) (for v = v2) y\Vt e Cn^ E [/x < v2]j, the numbers

AVi and AVl cannot be identical. B y com bining the last form ula
with the second inclusion we obtain

to„£Cb( % < 4

In other words A„f is one o f the numbers A which satisfy the

condition (3) for v = vx (or the condition (2) in the case vx = 0);
since AVi is b y definition the smallest o f these numbers, we have
AVi ^ AVl. W e thus finally reach the desired form ula XVi < Ayj.
From the form ulas (11)-(13) we conclude at once that A' ^ co,
which contradicts (10) and hence refutes the assum ption (9). It
must thus be accepted that

(14) yx e Cn{E\y < 7r]j for every A < co.

From (1) and (14) it follow s that

Cn(A) £ Cn{E\v < ?r]j;
'»C|>
the application o f Th. 16 to the inclusion (8) gives

Cn^E[v < 7r]^ s Cn(A).

Thus we have the identity

Cn(A) = Cn{E[v < tt]),
Xp
which in accordance with D ef. 2 leads to the form ula
(15) E[v < n] g $ cj(A ) .
Xp

The form ulas (6), (7), and (15) state that the sequence o f
sentences xv satisfies all conditions o f the conclusion.
In the p roof above we have used a m ethod o f reasoning which
is repeatedly em ployed in set-theoretical considerations. F or
the idea o f applying it to m etamathematieal problem s we are
indebted to Lindenbaum , who em ployed it in the p roof o f
Th. 56 given below.
A sequence o f sentences which satisfies the condition (a ) o f
Th. 16 could be called cardinally independent; if at the same time
the form ula (/?) is satisfied, then this sequence is to be called an
76 F U N D A M E N T A L CONCEPTS OF V, § 3

ordered basis o f the set A o f sentences. In this term inology the

theorem states that every set o f sentences possesses an ordered
basis.
T heobem 17. I f A c S, then Cn(A) e 9tq(A) and
Cn{A) = 2 [D ef. 2, A xs. 2, 3]
Xem(A)
T h e o b e m 18. (a) In order that A e 9lq(I?) and B e S , it is
necessary and sufficient that A £ S and B = Gn(A).
[D ef. 2, Ths. 7, 9 a]
(6) I f A s 5 , {Cn(A)} = S .9 Iq (A ); [Ths. 13a, 18a]
(c) I f A e <5, then {A } = S .5 tq (A ). [Ths. 7, 18a, 6]
B y Ths. 17 and 18a, Cn(A) is both the greatest o f all sets
equivalent to the set A and the only system equivalent to A;
consequently the set Cn(A) can be regarded and used as a
representative o f the whole class 3tq(A).

§ 4. A n A xiom System of a S et of Sentences ,

A xiom atizable Sets of Sentences
A n axiom system o f a set o f sentences is a finite set which is
equivalent to that set; a set o f sentences which possesses at least
one axiom system is called axiomatizable. Lindenbaum has
directed our attention to the concept o f axiom atizability. In
his investigations on the m eta-sentential-calculus he has
established the interesting fact that, in addition to axiom atizable
sets o f sentences, non-axiom atizable sets also exist.1
Using to denote the class o f all axiom systems o f a
set A o f sentences, a n d 1219to denote the class o f all axiom atizable
sets o f sentences, we reach
D e f in it io n 3. (a) 2ls(A ) = G .2tq(A).
(b) 91 = E[SHx(X) # 0].
21

The follow ing consequences o f this definition axe easily

provable:
(a) In order that A e %%(A) it is necessary and
T h e o b e m 19.
sufficient that A e (£. '$ (5 ). [D ef. 3 a, Th. 13 a]
(6) $ .$ (& ) s 91. [Th. 19a, D ef. 36]
1 Cf. IV, p. El (Th. 27).
V,§4 M E T H O D O L O G Y OF D E D U C T IV E SCIENCES 77

T heorem 20. (a) I f 2l£(A ).2l£(f?) # 0, then A e $q(.B).

[D ef. 3 a, Th. 136]
(6) I f A e ftq(B), then ftx(A) = 2ls(.B). [Th. 136, D ef. 3 o]
(c) I f A e f t then ftq{A) £ ft.
T heorem 21. (a) I f A x e 2lx(.B1) and A 2 e 21x(B2), then
A i + A z e 21*(Bj+ .B 2). [D ef. 3 a, Th. 156]
(6) I f A e 21 and B e ft, then A-\-B e 2t.
[D ef. 3 6, Th. 21 a]
T heorem 22. I f A e f t , then $(.4).2t5E(.4) # 0.
Proof. B y Defe. 2 and 3 it follow s from the hypothesis that
there is a set X which satisfies the form ulas
( 1) Xe<£,
(2) A-+-X £ S and Cn{A) = Cn(X).
From (2) w ith the help o f A x. 2 we obtain A £ 8 and
X £ Cn(A); from this, b y applying Th. 3 and b y reference to
(1), we infer that there exists a set Y such that
(3) 7 6®,
(4) Y g: A <= S and X £ Cn(Y).
B y Th. 1 a, 6 the form ulas (4) im ply
Cn(X) £ Cn(Y) £ Cn(A),
whence, in view o f (2), Cn{A) = Gn(Y); and from this b y means
o f D ef. 2 we obtain:
(5) Y e 2Tq(A).
B y D ef. 3 a the form ula Y e ftx(A) follow s from (3) and (5).
B y com bining this form ula w ith (4) we obtain at once
^(A).ftx(A) ^ 0, q.e.d.
T heorem 23. I f the class it satisfies the condition (a) of
Theorem 4 (or of Cor. 5) and if ^ X e f t , then

*M & *)*°-
Proof. B y Th. 22 it follow s from the hypothesis that
78 F U N D A M E N T A L CONCEPTS OF V,§4

H ence b y D ef. 3 a we have, for some Y,

N“
Ul
(1)

(2) Fee,
aud
(3)
Prom (1) and (2) it is easily inferred that there is a class
f i e ® . *$(&) which satisfies the form ula f g V I , Since
Xe2
b y hypothesis the condition (a) o f Th. 4 (or o f Cor. 5) is satisfied,
it follow s that there is a set Z e SL which includes the set T X

and a fortiori the set Y. W e thus have

(4) Z e ft,
(5) Y c Z c y X.
Xek
B y Th. 14 the form ulas (3) and (6) give at once
(6) Zet

Finally from (4) and (6) we obtain

^ 0, q.e.d.

T heorem 24. Let a sequence of sentences xv of the type re < w

be given which satisfies the formula (a) xv e S—Cn{E\ji < r]j
for every v < rr. In order that E\y < n ] e % it is necessary and
Xv
sufficient that rr < cd.
Proof. A . L et us suppose that
(1) E\y < rr] E %
Xv
and nevertheless assume
(2) TT= CO,
B y Th. 22 and D ef. 3 a it follow s from (1) that there is a set
Y which satisfies the formulas
(3) Y ^ E [v < n l Fe<£,
Xv
V. $4 M E T H O D O LO G Y OP D E D U C T IV E SCIENCES 79

and

(4) Y e $tq(E [v < w]).

M oreover, from (3) we infer the existence o f a number v such

that

(6) v < €o and Y £ E\ji ^ v].

The form ulas (2) and (5) give

(6) v +1 < IT,

whence b y the hypothesis o f the theorem we have

(7) avu e C n{E [n < v ]) .

Since E[ji < v] £ 8, we obtain from (5) with the help o f Th.

l o Cn(Y) c Cn[E[ft ^ v ]j; b y com paring this inclusion with (7)

we reach the form ula

(8 ) x y+1 e C n(Y).

B y A x . 2 and D ef. 3 it results from (4) that

E[v < n] £ Cn(E[v < tt] ) = Cn(Y).

Thus b y virtue o f (6) we have xv+1 e Cn(Y), which obviously

contradicts the form ula (8) and refutes the assumption (2).
H ence we conclude that
(9) 7T < co.
B . I f on the other hand the inequality (9) is satisfied, then
according to the hypothesis we have E\y < n] e C£. $ (£ ), from
Xy

which b y Th. 196 the form ula (1) follow s.

W e have thus established the equivalence o f the form ulas (1)
and (9), and this concludes the proof.
T heorem 26. The following conditions are equivalent: (a)
A . e 31; (j8) there is no sequence of sentences xv of type co which
satisfies both of the formulas (a) and (ft) of Th. 16, and in addition
i c S ; (y) there is a sequence of sentences x v of the type n < a>,
which satisfies both of these formulas.
80 F U N D A M E N T A L CONCEPTS OF V ,§ 4

Proof. I f the condition (a) holds, then b y Th. 20 c we have

2tq(A) £ 21 and accordingly E\y < «•] e 21 fo r every sequence
o f sentences w hich satisfies the conclusion o f Th. 16; from Th.
24 it thus results that the inequality it > o» does n ot hold.
B ut since A £ S (by D efe. 2 and 3 a, b), w e can assert that
(j8) follow s from (a). On the other hand b y Th. 16 there exists
for every set A £ S a sequence o f sentences either o f type io
or o f type n < a>, which satisfies the conclusion o f this theorem ;
(/?) thus im plies (y). F inally, i f (y) is satisfied, then we have
E [v < 7r] 6 © . $lq(A), w hich on the basis o f D ef. 3 a, 6 gives the
condition (a).
A ccordin gly the conditions (a )— (y) are equivalent, q.e.d.
Further theorem s deal w ith axiom atizable and unaxiom atiz-
able system s.
T heorem 26. <5.2t = E [X e <£. *P(&)].
(M X )
[D ef. 3 a, 6, Th. 18 a]
T h eorem 27. In order that A e S —21, it is necessary and
sufficient that there exists a sequence of sets X „ of the type to which
satisfies the following formulas: (a) I , e S for every v < to;
(P) Xp £ X v and X^ # X v for p. < v < a>; (y) A = 2 The
formula (a) can be replaced by (a ) X v e S . 21.
Proof. A . F irst let us assume that

(1) A 6 S — 21.
B y applying Ths. 16 and 25 (and taking in to account D ef. 1)
w e infer from (1) that a sequence o f sentences xv o f type to exists
w hich satisfies the form ulas

(2 )
and
(3) E[v < a»] e 2tq(A).
3b
B y T h. 18 a from (1) and (3) we obtain

(4)
V, §4 M E T H O D O L O G Y OF D E D U C T IV E SCIENCES SI

W e pu t
(5) V < <Jt).

B y T h . 26 it followB from (4) and (5) that

(6) X v e <5 for v < a*
and even that
(6 ') X ve S .? l for v < a>.
In accordance w ith Th, 1 a and b y reference to (4) we obtain
Cn(i?[A < £ Cn[E\X < and from this b y (5)

(7) X p £ X v for n < v < co.

From (4), (6), and A x. 2 we obtain x■ e X v, but in view o f (2)
and (6) e X^ for < v < w; consequently
(8) X^ # X v for n < v < to.
F inally it is to be noted that the class it' o f sets o f sentences
Tv = i£[A < v], where v < o>, satisfies the hypothesis o f Cor. 5,
whence (7 n (^ r r ) — ^ Cn(Y ); on the basis o f (4) and (5) we
thus have
(9) -4 = 2
v<a>

From (6 )-(9 ) (as well as (6 ')) it results that

(10) the sequences of sets X v of type a> satisfies the formulas
(a )-(y ) (as well as the formula (ex')) of the theorem.
B. Consider now a sequence o f sets X v o f the kind described
in (10) (with the reference to (a ') om itted). Put
(11) R = E[v < a>].

From (10) and (11) it follow s at once that

(12)
Xe&
and that
(13) the doss R satisfies condition (a) of Cor. 5.
Since further b y (10) and (11) 51 £ <3, we infer from (13) and
(12) b y means o f Th. 12 that
(14) A — T I e S .
x3t
82 F U N D A M E N T A L CONCEPTS OF V, $4

I f it were the case that ft. XJ # 0, e.g. if

Feft.aq(j\r),
then b y (10) and (14) w e should have 7 e @ . $ q (A ) and A e S ,
whioh b y Th. 18c gives Y = A e ft; but the last form ula is in
contradiction w ith (10) and (11) (since the sum o f all terms o f
an increasing sequence o f sets o f type u> contains every term
as a proper part). Consequently
(16)

B y the use o f Th. 23 we infer from (13)-(15) that

b y com bining this form ula w ith (14) we at once obtain (1).
W e have thus shown that the existence of a sequence of sett
described in (10) forms a necessary and sufficient condition for
formula (1), q.e.d.
From tho last tw o theorems the follow ing corollary is easily
obtained:
COROLLARY 28. 6 . % < X0.
Proof. A ccording to a well-known theorem o f set theory
A x. 1 has the consequence that the class (£. ty(S) is at most
denum erable. B y Th. 26 the function On establishes a m any-to-
one correlation between the m embers o f this class and those
o f the class S . 31; accordingly this latter class is also at most
denum erable, < X®, q.e.d.
C orollary 29. (a) I f A e <5—% then ?P(A). S . 31 = X0.
(b) I f © - « =£ 0 (or I p ( /S ) - « # 0), then W .1 = X0.
Proof, (a) B y Th. 27 the hypothesis gives iP (A ).<5.3l > X0;
since the inverse inequality follow s im m ediately from Cor. 28,
we finally have fp (A ). ®. 3t = X0;
(b) results directly from (a) and Cor. 28 (with the help of
Ths. 18 a and 20 c there is n o difficulty in showing that the
second premiss, Sp(3) — 51 # 0, im plies the first, S — 91 ^ o).
T heorem 30. Hither 1 < < X0 or X-0 < <s < 2*'.
V, §4 M E T H O D O LO G Y OF D E D U C T IV E SCIENCES 8S

Proof. A ccording to the well-known theorem on the cardinal

number o f the power set we first infer from Ax* 1 and D ef^ l that
< ^P(S) < 2*#; and b y virtue o f Th. 10 we also have § > 1.
I f now S c (or, what am ounts to the same thing, S = 6 .21),
then b y Cor. 28 we have
1 < 1 < Ko;
if, how ever, <5 —21 7^ 0 , then it follow s from Cor. 296 that
K0 < § < q.e.d.
It is to be noted that the theorem above can be established
without the use o f the axiom o f choice (otherwise the result
would be quite trivial); the same applies also to the generaliza­
tion o f this theorem given below in Cor. 65. I f we restrict our­
selves to the consideration o f deductive disciplines which pre-
suppose sentential calculus, we can im prove Th. 30 b y showing
that the class <5 is either o f the power 2* for some v < a> or else o f
the power 2K®.f

§ 5. I n dependen t Sets of S entences ; B asis of a Set

of S entences
A set o f sentences is called independent if it is not equivalent
to any o f its proper subsets. The class o f all independent sets
o f sentences is denoted by ‘IT:
D efin itio n 4. U = 2 ? [f p (X ).8 q (Z ) = {X}].
x
Some equivalent transform ations o f the above definition are
given in the next tw o theorems.
T heorem 31. The following conditions are equivahrd: (a)
i e U ; (£) A.Cn (X) £ X £ S for every set X s A ; (y) the
formulas X-\-Y £ A and Cn(X) =■= Cn(Y) always imply X = Y+
and in addition A c S.
Proof A . First we suppose that
(1) A e U.
B y Defe. 4 and 2 it follow s that:
(2) A £ S.
+ See X II, Ths. 37 and 38 (p. 367}.
84 F U N D A M E N T A L CONCEPTS OF V ,{ 5

L et X be any subset o f A .
B y the use o f Th. 2 a and w ith the help o f (2) we obtain:
C n { ( A - C n { X ) ) + X ) = Cn {(A -C n (X ))+ C n (X ))
= Cn(A+Cn(X)) = C n ( A + X ) = Cn(A),
and accordingly b y D ef. 2, (A — (7 »(X ))H -X e 93(A ).5tq(A ).
From this, b y virtue o f D ef. 4 and w ith the help o f (1), we infer
that ( A — Cn(X ) ) + X — A and consequently.4 . ( Cn(X)—X ) = 0.
Thus b y (2) we have
(3) A .Cn(X) £ X £ S for every set X £ A.
B . W e assume next the condition (3). W e consider any two
sets X and T which satisfy the form ulas X + Y £ A and
Cn(X) — Cn(Y). It follow s from (3) that
A.Cn(Y) = A .Cn (X) £ X ;
since further (3) gives the inclusion (2), we have b y A x . 2
Y s A.Cn{Y).
Consequently Y £ X . In an exactly analogous manner we
reach the inverse inclusion X £ Y, so that finally X = Y. W e
have thus shown that
<'4) the formulas X -j-F £ A and Gn{X) = Cn(Y) always imply
X — Y, and in addition (2) holds.
C. Finally, let (4) be given. A ccording to D ef. 2 every set
X e 93(A ).2tq(A ) satisfies the form ula Cn(X) == Cn(A)\ hence,
i f in (4) w e put Y = A , w e obtain X = A . Thus we have
93(A). $ q (A ) £ {A }; bu t since b y Th. 13 a the inclusion
{A } £ 93(A ).5tq(A )
also holds, we reach the identity 93(A). SSq(A) = {A }, w hich by
D ef. 4 gives the form ula (1).
A ccording to the above argum ent, (3) follow s from (1), (4)
from (3), and (1) from (4). The conditions (1), (3), and (4) are
thus equivalent, q.e.d.
T heorem 32. In order that A e JI, it is necessary and sufficient
that z e 8 — Cn(A—{« }), for every x e A .
Proof. A . L et us assume that
(1) A e It,
V, §5 M E T H O D O L O G Y OF D E D U C T IV E SCIENCES 85

Mid apply Th. 31. According to the condition (y) o f this theorem
it follows directly from (1) that
(2) A £ 8.
By putting X — A —{x} in condition (/J) o f the same theorem
we further obtain A . C n ( A —{x}) c A —{x}, whence
{x }.A . Cn(A—{x}) — 0.
Under the assumption that x e A the last formula gives
{x}. Cn(A—{x}) = 0 and finally, by virtue of (2),
(3) x e 8 — Cn(A — {«}) for every x e A.
B. We now assume formula (3) and note initially that from
this the inclusion (2) immediately follows. Let us assume that
there exists a set X £ A which satisfies the formula
A .C n (X )-X # 0 .
Let for example x e A . Cn(X)—X and accordingly X £ A —{x},
which by Th. l a gives the inclusion Cn(X) £ Cn(A —{2 });
since then x e A . Cn(X), we infer at once that x e Gn{A—{x}),
which contradicts formula (3). Our assumption is thus dis­
proved; consequently, in view o f (2), we must assume that
(4) A . Cn(X) £ X £ S for every set X £ A .
B y Th. 31 formula (1) follows from (4).
We have thus established the equivalence o f the formulas
(1) and (3) and thus proved the theorem.
From the last theorem we obtain immediately
Corollary 33. (a) OeU; [Th. 32]
(6) in order that {* } eVL, it is necessary and sufficient that
x e S—Cn(0). [Th. 32]
T heorem 34. I f A e U, then ?P(A) £ U. [Th. 31]
T heorem 35. I f A e U , then <P(A).2l = ip(A).<£.
Proof. According to Th. 32 (or else 31) the hypothesis gives
(1) A £ S.
86 F U N D A M E N T A L CONCEPTS OF V, § 5

Consequently we have s p ($ ).g , from which b y

Th. 196 the inclusion
(2)
follow s.
On the other hand let us assume that
(3) $(A).9!-« # < > ,
and accordingly let
(4) Xe
W ith the help o f A x. 1 we easily infer from (l)-(4 ) that the
set X is denum erable; accordingly all elements o f this set can
be ordered in an infinite sequence o f the type co, with all terms
distinct, such that
(5) X = E\y < w] tohere x^ ^ xvfor p < v < o>.
Xy
A ccording to Th. 34 it follow s from (4) and the hypothesis that
X eVL; from this b y Th. 32 and with the help o f (5) we obtain
(6) xv e S— Cn(X—{xv}) for every v < to.
From (5) and ( 6) we further obtain
E\fi < v\ £ X —{#„} c S for v < a>.
Xyt '
B y Th. 1 a this implies the inclusion
Cn(E[(x < v]) £ C n {X -{x v)y,
X fj,

hence b y ( 6 ) it follow s that

(7) xv e S-~Cn(E{ji < v]) for V < Ct>.

B y com bining (5) and (7) w ith Th. 24 we easily obtain

X = E[v < ca] e 91,
Xv
contrary to form ula (4).
W ith this, assumption (3) is disproved and therefore we have
(8 ) $ { A) M c

Inclusions ( 2) and ( 8} at once give the required identity

^ (,4 ). 9 1 = $ ( 4 ).® , q.e.d.
V, §5 M E T H O D O L O G Y OF D E D U C T IV E SCIENCES 87

Theorem 36. I f A e U—(£, then ty(Cn(A)). 6.31 = Sq, but

y(Cn(A)).<5 = $ ( < 7 » ( A ) ) . 6 - « = 2*.
Proof. B y Ths. 32 and 36 as well as A x. 1 we infer from the
hypothesis that
(1) A c s , A e St,
and
(2) I - K
A ccording to T h. 18 a it follow s from ( 1) that Cn(A) e S and
A e $tq(Cn(A))-, from this b y Th. 20 c we obtain Cn(A) e S —
H ence, as a consequence o f Cor. 29 a, we obtain at once

(3) y {C n (A )).G .K = K0.

Since b y the hypothesis the set A is independent, we can apply
Th. 31. Condition (y) o f this theorem asserts that the function
Gn m aps the class $ (A ) on I [ I g i l in one-to-one fashion;
On(X)
consequently the tw o classes have the same cardinal number.
B y a well-known theorem it follow s from ( 2) that ty{A) = 2*«;
hence
(4) I [ I c A] = 2*°.
Cn(X)

B y means o f Ths, l a and 9a, using ( 1) we obtain w ithout

difficulty that E [X £ A ] c ty(Cn(A)).(5; b y com bining this
Cn(X)
form ula w ith (4) we get ip(C w (A )). 6 > 2s*. B ut since b y Th. 30
we also have . 6 ^ 2*», we finally obtain

(5) f(O i(;l))7 s = 2*».

Form ulas (3) and (5) at once give

(6) $ (0 7 » (A )).6 -3 t = 28»~ X 0 = 2*.

A ccording to (3), ( 6), and ( 6) the theorem is com pletely
proved.
Corollary 37. I f U -(£ ^ 0, then 6.31 = K0, but

S 6 -3 1 = 2*.
[Th. 36, Cor. 28, Th. 30]
88 F U N D A M E N T A L CONCEPTS OF V ,§ 6

I t is to be noted that within alm ost all deductive disciplines,

and in particular w ithin the sim plest o f them — the sentential
calculus— it has been found possible to construct a set o f sen­
tences w hich is both infinite and independent, and thus to realize
the hypothesis o f the last corollary. H ence, it turns out that in
all these disciplines there are m ore unaxiom atizable than axiom a-
tizable system s; the deductive system s are, so to speak, as a rule
unaxiom atizable, although in practice we deal alm ost exclusively
w ith axiom atizable system s. This paradoxical circum stance was
first n oticed b y Lindenbaum in application to the sentential
calculus .1
E very independent set o f sentences which is equivalent to a
given set A is called a basis o f the set A ; the class o f all such sets
o f sentences is here denoted b y ‘ 93(A)’ .
D e f in it io n 5. 98(A) = <5q(.4).lX.
The above definition can be expressed otherwise as follow s:
T heorem 38. In order that B e 93(A), it is necessary and
sufficient that $P(j9) . 9Iq(A) = {5 } . [D efs. 6, 4, Th. 136]
In the usual term inology o f set theory a set X is said to be
irreducible with respect to a class R of sets when © (X ). R = {-X}-2
Thus the theorem above asserts that ‘ a basis o f the set A o f sen­
tences’ means the same as ‘ a set irreducible w ith respect to the
class 9Iq(^4)’ .
Other properties o f this concept are expressed in the follow ­
in g theorem s:
T heorem 39. In order that A e ©(^4), it is necessary and
sufficient that A e It. [D efe. 5, 4]
T heorem 40 . (a) I f 93(A). 93(B) # 0, then A e 5tq(.B);
[D ef. 5, T h. 136]
(6) if A s %q(B), then © (^ ) = 93(B). [T h. 136, D ef. 5]
T heorem 41. (S.93(A) = 11.91a-(.4). [D efs. 5, 4, 3a]
T heorem 42. I f A e % then $P(-4). © . 93(A) ^ 0 .
1 Cf. IV , Th. 27 (p. 51).
* Cf. Tarski, A . (71), p. 48 (Def.,1).
V, §5 M E T H O D O L O G Y OF D E D U C T IV E SCIENCES 89

Proof. B y Th. 22 and D ef. 3 a the hypothesis im plies that a set

X exists w hich satisfies the form ulas
(1 ) X<=A,
(2 ) X e © and X e 9Iq(A).
B y T h. 13 a, 6 it results from ( 2 ) that X e 9tq(X ); consequently
the class 51 = ?P (X ). 2tq(X), w hich consists o f subsets o f the
finite set X , is distinct from 0 . H ence, b y a fam iliar definition o f
finite sets ,1 we conclude that this class contains am ong its
elements at least one irreducible set Y (i.e. a set with the property
© (F ). 51 = {Y}). B y Th. 38 this set T form s a basis o f X , and
thus we have
(3) Y ^X and Y e © (X ).
F rom ( 1)—(3) we obtain at once
(4) 7 s i and Y e ©.
B y Th. 406 it follow s from ( 2) that © (A ) = © (X ), whence
by (3)
(5) 7 e 8 (4 ).
Form ulas (4) and (5) give at once
© (A ). © .© (A ) # 0, q.e.d.
T h e o r e m 43. I f A e SI, then 58(A) s ffi.
Proof. I f the conclusion were false, we should have, b y D ef. 5,
iitq (A ).lI— © ^ 0 . A ccordin gly let X e 9tq(A) and X e U —©.
B y Th. 35 the second o f these form ulas yields X e 51; and hence,
with the help o f the first form ula b y using Th. 20 c we infer that
4 e 91, contrary to the hypothesis.
W e thus have 58(A) s ©, and so the theorem is proved.
C o r o l la r y 44. The following conditions are equivalent:
(a) A e 91; (j8) 35(A) ^ 0 and 58(A) s ©; (y) © .© (4 ) 0.
Proof. B y Ths. 42 and 43, (/J) is at once obtainable from (a);
(y) results im m ediately from (/J); finally, on the basis o f Th. 41
and D ef. 36, (a) follow s from (y). A ccordingly the form ulas
(a)-(y) are equivalent, q.e.d.

1 Of. Tarski, A . (71), p. 49 (Def. 3).

90 F U N D A M E N T A L CONCEPTS OF V,$5

B y the corollary just established (or b y Th. 42) every axio-

m atizable set o f sentences possesses at least one basis. This result
cannot be extended to unaxiom atizable sets o f sentences on the
basis o f the axiom s underlying this discussion1 (it can only be
shown that every set o f sentences possesses an ordered basis;
see remarks follow ing Th, 16). On the other hand, for those de­
ductive disciplines which presuppose the sentential calculus it
can be proved that every set o f sentences has a ba sis.f

§ 6. Consistent S ets of Sentences

The concepts o f consistency and com pleteness, with which
we are concerned in this and the next sections, are am ong the
m ost im portant concepts o f m etam athem atics; around these
concepts is centred the research which is carried on tod ay within
special m etadisciplines.
A set o f sentences is called consistent if it is n ot equivalent to
the set o f all m eaningful sentences (or, in other words, if the set
o f its consequences does n ot contain as elements all meaningful
sentences).
A ccording to the usual definition, a set o f sentences is called
consistent if there is no sentence which together with its negation
belongs to the consequences o f this set. Our definition thus
diverges from the usual one, and indeed it has a m uch more
general character since knowledge o f the concept o f negation is
not presupposed; in consequence this definition can be applied
even to those deductive disciplines in w hich the negation
concept is either entirely lacking or at least does not exhibit the
properties usually ascribed to it.2 H ow ever, the tw o definitions
o f consistency prove to be equivalent for all those disciplines
which are based upon the ordinary system o f sentential calculus.

1 For example, in the sentential calculus it is easy to construct sets of

sentences without a basis. (A proof of this resuit appears in SchrSter, K.
(62 a), pp. 299-301.)
* Such a discipline, the so-called restricted sentential calculus is treated in
IV, §4.

f Cf. the remarks following Th. 25 in XII (p. 362).

V, $6 M E TH O D O LO G Y OP D E D U C T IV E SCIENCES 91

W e ow e the idea o f the definition adopted here to E . P ost:

in his investigations on the sentential calculus he has made
use o f a olosely related definition.1
D enoting the class o f all consistent sets o f sentences b y ‘ SB’
we reach
D e fin itio n 6. SB = yS(S)—$L<\(S).
The content o f this definition is m ore clearly form ulated in the
follow ing:
T heorem 45. In order {hat A e SB, it is necessary and sufficient
that A £ 8 and Cn{A) yt 8. [D ef. 6, Ths. 10, 18o ]
The follow ing elem entary properties o f consistent sets o f
sentences deserve consideration:
T heorem 46. (a) I f A e SB, then $ (A ) £ 2B;
[D ef. 6, Th. 14]
(6) if A 6 SB, then $lq(A) £ SB. [Th. 45, D ef. 2]
T heorem 47. S — SB = {£ }. [D ef. 6, Ths. 7, 18c]
T heorem 48. Let S e SI; in order that (a) A e SB, it is necessary
and sufficient that (ft) $P(A).ffi £ SB.
Proof. A ccording to Th. 46 a the condition (fi) is necessary for
(a) to h old; it remains only to show that this condition is also
sufficient.
L et us assume (/?) and suppose that nevertheless
(1) A € SB.
B y D ef. 6 it follow s from (j8) that $P(A). ® £ $ ($ ); from this
we easily obtain:
(2) A £ 8.
B y another application o f D ef. 6 we obtain from (1) and (2)
A e ®cj(5). Since b y hypothesis 8 e 21 holds, we infer b y Th. 20 c
that A e 91, whence b y Th. 22 the form ula ^P(A).Sfce(A) # 0
follow s. A ccordingly b y D ef. 3 there exists a set X which satisfies
the form ulas

(3)
1 Poet, E. L. (60), p. 177.
92 F U N D A M E N T A L CONCEPTS OF V, J 6

and
(4) X e $tq(.4).
Prom (j8) and (3) it follow s at once that X e 933; according
to Th. 136, (4) is equivalent to the form ula A e 9tq(X). Conse­
quently b y virtue o f Th. 46 b we obtain the form ula A e 933,
w hich coincides w ith (a) and contradicts the assum ption (1).
W e have thus shown that (/?) is a sufficient condition for (a),
and with this the p roof is com plete.

T heorem 49. Let 8 e 91; if the class - f t satisfies the condition

(a) o f Th. 4 (or of Cor. 5) and if at the same time £ 933 holds,
then y X 6 933.
XeSt

Proof. Consider an arbitrary set Y e<£. X ^. B y virtue

o f condition (a) o f Th. 4 we infer w ithout difficulty that there is a

set Z w hich satisfies the form ulas Z e St and Y £ Z (of. the p roof
o f Th. 23). On the basis o f the hypothesis the first form ula yields
Z e 933; hence b y using Th. 46 a and with the help o f the second
form ula w e obtain T € 933.
I t is thus proved that the form ula Y e ® . 93 X ) always

im plies Y e 933. Consequently we have the inclusion

C. £ 9B;

in accordance w ith the hypothesis S e 91 and w ith the help o f

Th. 48 we get from this at once

q.e.d.

The form ula 8 6 91, which oocurs as a premiss in the two

preceding theorem s, as w ell as in some later ones (Ths. 51, 56,
and 67), cannot be derived from the axiom s listed in § 1. It is,
how ever, satisfied for all known form alized disciplines; even the
follow ing logically stronger assertion holds: there is a sentence
x e 8 such that Cn({x}) — S. Nevertheless it does n ot seem
desirable to include this form ula am ong the axiom s, on account
o f its special and, in a certain sense, accidental character.
V, 57 M E T H O D O L O G Y OF D E D U C T IV E SCIENCES 93

§ 7. D ecision D omain of a Set of Sentences, Complete

Sets of Sentences
B y the decision domain o f the set A o f sentences we under­
stand the set o f all sentences w hich are either consequences o f
A or which, when added to A , yield an inconsistent set o f sen­
tences. A set o f sentences is said to be complete or absolutely com­
plete i f its decision dom ain contains all m eaningful sentences.1
W ith regard to this definition o f com pleteness we can repeat
all remarks w hich were m ade above in § 6 in connexion with the
definition o f consistency. This definition is also due to P ost, who
has used it in his investigations into the sentential calculus.2
As is easily seen the usual definition o f com pleteness, resting upon
the ooncept o f negation, is quite unsuitable for the sentential
calculus (as w ell as for all disciplines w hich contain sentences
w ith so-called free variables). In foot, b y the usual definition, a
set o f sentences is called com plete if, for every sentence, either
the sentence itself or its negation belongs to the consequences o f
this set; and, as is well known, the ordinary system o f sentential
calaulus is com plete in the sense th at its decision dom ain contains
all m eaningful sentences, although it is not com plete according to
the usual definition. The tw o definitions prove to be equivalent
for all those disciplines w hich presuppose sentential calculus
(and in w hich no expression containing free variables is regarded
as a sentence).
The concept o f absolute com pleteness is o f great im portance
for those m etadisciplines in whioh the objects o f investigation
are ‘p oor’ , elem entary disciplines o f an uncom plicated logical
structure (e.g. the sentential calculus, or disciplines w ithout
predicate variables).3 On the other hand this concept has not yet
played an im portant part in investigations on ‘rich ’ , logically
more com plicated disciplines (e.g. the system o f Principia
Mathematical). The cause o f this is perhaps to be sought in the
1 The ooncept of completeness also occurs with another meaning, discussed,
for instance, in Fraenkel, A. (16), § 18.4, pp. 347-64; the meaning used here
corresponds to FraenkeFs Entscheidungedefinitheit.
• See Post, E. L. (60), p. 177.
* Cf. Langford, C. H. (44), p. 115, as well as IV and Presburger, M. (61).
For later developments in this,domain consult Tarski, A. (84).
94 F U N D A M E N T A L CONCEPTS OF V,§7

widespread, perhaps intuitively plausible, bu t not always strictly

founded, belief in the incom pleteness o f all system s developed
w ithin these disciplines and known at the present day .1 I t is
nevertheless to be expected that the concept o f com pleteness
w ill som e day attain a greater im portance even for the latter
disciplines. F or, although all known consistent system s o f this
kind m ay be incom plete, yet Th. 56 to be established below offers
at least a theoretical possibility o f extending every suoh system
to one w hich is both consistent and com plete. The question now
arises how this extension is to be carried out so as to be ‘ effective’ ,
as natural as possible, and at the same tim e agreeing w ith som e
philosophical view point .2
M oreover, it is to be noted th a t som e m ore special problem s
w hich are closely connected w ith the concept o f com pleteness
have already been successfully investigated even w ith respect
to those ‘rich ’ disciplines. The investigations referred to have
aim ed at and succeeded in proving that all m eaningful sentences
o f a particular logical form (e.g. all sentences w ithout function
variables) belong to the decision dom ain o f some given system .
I t m ight be useful for such investigations to introduce the con­
cept o f relative com pleteness: a set A o f sentences is said to be
relatively complete with respect to the set B of sentences if the decision
dom ain o f A includes the set B ; this concept w ill n ot be discussed
further here .3
The decision dom ain o f the set A o f sentences w ill be denoted
b y £Crnt(.4)’ and the class o f all com plete sets o f sentences b y ‘ 33’ .

D efin itio n 7. (a) Cent(4 ) = Cn{A)+S. E [A + {x} g SB].

X
(6) 35 = $ (8 ). E[<£ni(X) = &].
x
1 See Fraenkel, A. (16), pp. 347-54.
a Since this article first appeared in print, several important contributions
concerned with the concept of completeness have been published, throwing
much light on the questions discussed in the last paragraph. See in the first
place Godel, K. (22).
s The investigations on absolute completeness in a ‘poor* discipline are as
a role equivalent to those on relative completeness in a more extensive dis­
cipline. For this reason the works cited on p. 93, note 3 can be regarded as
examples of investigations on relative completeness, if they are related to an
extended discipline (e.g. to Principia McUhemcUica).
V, S 7 M E T H O D O L O G Y OF D E D U C T IV E SCIENCES 95

The follow ing elem entary properties o f decision dom ains

should be noted:
T heorem 50. (a) I f A £ 8 , then A £ (£rtt(-4) £ 8.
[A x. 2, D ef. 7 a]
(b) I f A s B s S, then CErtt {A) £ (Eitt(.B).
[The. 1 a, 46 a, D ef. 7 a ]
(c) I f A e then C nt(A ) = <£nt(B).
[D ef. 2 , Ths. 136, 15 a, 466, D ef. 7 a]
T heorem 51. Let 8 e 51; if A £ 8, then
ffint(A) = Y <Ent(X).
XecTi).®
Proof. W e consider a sentence y e 8 such that A~j-{y} e 2B.
Hence, b y Th. 48 and with the help o f the hypothesis, we con­
2
clude that S3(A+{ /}). (£—333 ^ 0. Thus there is a set Y which
satisfies the form ulas Y e SP (A + { y } ) . ® and Y s SB. The first
form ula gives us Y —{y} e SP(A). (E, and with the help o f Th. 46 a
obtain from the second form ula T + {y ) = (Y — e SB.
Consequently there exists a set X such that X e ip (A).<£ and
X-j-{yj 6 2B (and in fa ct X = Y —{y}); thus we have

y% I (-E p '+ {x }e S 3 3 ]).

From this argument it follow s im m ediately that

(1) 8 . E [ A + { x } e 2B] £ Y l s . M X + { x } e 2B]).

x X }

A ccording to A x. 4 we have further

(2) On(A) = Y Cn(X).
xssJCo.®
Form ulas ( 1) and (2 ) give at once
C «(A )+ )S .X [A + {a :}6 2B] j C n { X ) + S . E [X + {x] e 2B]),

whence b y D ef. 7 a
(8) ®nt(A ) £ Y ® nt(X ).
XeSR3t).«
On the other hand it follow s from Th. 506 that
® nt(X ) £ ®nt(A)
90 FUNDAMENTAL CONCEPTS OF V,§
fo r every set X e . <£, whence

(4) T « n t(X ) £ <£nt(A).

x & S td ),®

Finally, from (3) and (4) we obtain

(£rtt(*4) = y ffint(X), q.e.d.

D ef. 7 6 can be m odified in various w ays:

T heorem 52. The following conditions are equivalent:

(a) A e © ; (j3) A £ S and £S(A).2B £ # q (A ); (y) A £ S anc

& ( A ) . S . W £ (C w (A )}; (S) £ ( A ) .® = {S, Cn{A)}.
Proof. A . L et us assume form ula (a) and consider any se1
X e £ }(A ). 233. Thus according to X)ef. 6 A £ X £ S; hence bj
D ef. 7 a, b and w ith the help o f (a) we infer that X £ (Srtt(A) and
that consequently for every x e X either x e Gn(A) or A -f-{z} 6 2B;
but since A-[-{x} £ X, it follow s from Th. 46 a that A-\-{%} € SB
and therefore x e Cn(A) for every x e X . From this it follows
that X £ Cn(A) and further that
A £ X £ On{A)\

on the basis o f Th. 1 a, b this last form ula gives

Cn(A) £ Cn(X) £ Cn(A), i.e. Cn(A) = Cn{X),

whence b y D ef. 2 the form ula X e 3lq (A ) follow s.
From these considerations we conclude at once that

& (A ). 2B £ S q (A );

b y D ef. 7 b, (a.) gives also the inclusion A £ S. I t is thus proved

that
( 1) (jS) follows from (a).

B . Further b y Th. 186 we have

( 2) (j8) implies (y).

C. It is likewise easy b y Ths. 47 and 9 a to establish that
(3) ( 8) follows from (y).
V, § 7 M E T H O D O L O G Y OF D E D U C T IV E SCIENCES 97

D. N ow we assume that (8) holds; from this it results im ­

m ediately that A £ S. L et further an arbitrary sentence x e 8
be given. B y virtue o f Th. 9 a we have
Cn{A-\-{x}) e £ ( 4 + { # } ) . S .
Thus a fortiori Cn(A-\-{%)i) e £ l ( 4 ) .S . Hence b y ( 8) we obtain
Cn{A+{x})e{S ,Cn{A)}.
HCn(A-\-{x}) = 8, then it follow s from Th. 45 that A -f{a ;} e 205;
^ Cn(A+{x}) = Cn(A),
then, b }' A x. 2, we have x e Cn(A + {* }) and consequently
x e Cn(A). Thus b y D ef. 7 a it follow s in either case that
* e <Ent(A).
B y means o f this argument we reach the inclusion S £ (fnt(A ),
which b y Th. 50 a leads to the identity Cent(A ) = S\ and, in
accordance with D ef. 7 b, (a) is a consequence o f this identity.
Thus
(4) (a) follows from (8).
B y com bining ( 1)—(4) we infer at once that the conditions (a )-
(8) are equivalent, q.e.d.
T heorem 53. The following conditions are equivalent: (a)
.A e S .2 B .9 3 ; (/?) A e 903 and for every x e 8 either x e A or
A + { x } e 2B; (y) 52(A). 2B = {A }.
Proof. A . B y com paring D ef. 7 a, b and Th. 7 we easily see that
(1 ) (/S) follows from (a).
B . L et us assume (j8). W e consider an arbitrary set
X e 5 2 (A ).903,
which thus, b y D ef. 6, satisfies the form ula A £ X £ S. F or
every x e X we have £ X and consequently, b y Th. 46 a
A + { x } e 2B, w hich b y virtue o f (j8) gives the form ula x e A ;
consequently we have X £ A and in fact X = A. Hence we
conclude that 52(A ). 203 £ {A }; since the inverse inclusion follow s
directly from (/?) we finally reach the form ula (y), i.e.
52(A).2B = {A }.
08 FUNDAMENTAL CONCEPTS OF V ,§7

I t is thus proved that

(2) (y) follows from (/3).
C. N ext we assume that (y) holds. From this it results im ­
m ediately that A e i B and therefore A £ S. B y Ths. 9 a and
18 6 we thus have Cn(A) e & (A ). € . & q(A); hence, w ith the help
o f Th. 46 b, we obtain Cn(A) e Q (A ). 9GB. The com parison o f the
last tw o form ulas w ith (y) gives first
On(A) = 4 e ®
and secondly & (.4).2B £ $tq(4).
N ow , since the condition (J3) o f T h . 62 is satisfied, we have also
A e 93, so that finally the whole form ula (a), A e G .2B.33, is
derived. A ccordingly w e have
(3) (y) implies (a).
From (l)-(3 ) it follow s at once that the conditions (ot)-(y) are
equivalent. This com pletes the proof.
T heorem 54. (a) I f A e 95, then & (A ). $ ($ ) £ 91;
[D ef. 76, Th. 50 a, 6]
(6) if A e 93, then 9lq(A) £ S . [D efe. 2 , 7 6, Th. 50 c]
T heorem 55. (a) ty(S) — 9QB-f ®> [T h. 45, Defe. 6, 7 a, 6]
(6) S e 93. [Ths. 47, 55 a]
The follow ing interesting theorem , due to Lindenbaum , asserts
that (provided the set 8 is axiom atizable) every consistent set
o f sentences can be extended to a consistent and com plete system .
T heorem 56. Let S e %; i f A e SB, then
JQ (A ).S .9B .9J = £ 0 .
Proof. A ccording to D ef. 6 and Th. 47 the hypothesis gives
A £ S and A # 8, whence 8 ^ 0 . Thus in view o f A x. 1 all
sentences can be ordered in an infinite sequence o f type <o (not
necessarily with all term s distinct), such that
(1) S — JJ[A < « ] .
*A
L et
(2) A0 be the smallest o f the numbers A < at which satisfy the
formula A +{a:^} 6 9B.
V, | 7 M E T H O D O L O G Y OP D E D U C T IV E SCIENCES 90
Farther, let
(3) A* (where 0 < v < <o) be the smallest o f the numbers A < u>
which satisfy the conditions
A + E [ p < v]+{*a} e 3B and X > A„ for every p < v.
* am

Finally, let
(4) it be
the smallest ordinal number with which the conditions (2)
and (3) correlate no number A^.
N ow we pu t
(6) X 0= A, X v+1 = A + E \ji < v] for v < rr
and
(6) X = X 0-f- 2 Xy+V
v<ir
The conditions (2 )-(4 ) give at once
(7) w < w.
From (4 )-(6 ) we obtain
(8) A s X,
and in general
(9) Xp —X for (i < v < ir.
F rom (2 )-(5 ) w e infer w ithout difficulty that
(10) X 0 6 28 and X y+16 28 for v<n.
L et ft — (X 0}-f- E \v < «•]. B y (9) and (7) the class St
Xy+\
satisfies the condition (a) o f Cor. 5; by (10) the inclusion ft S 8B
holds. Taking in to account the form ula S e 2t assumed in the
hypothesis we see that all premisses o f Th. 49 are satisfied.
A ccordingly we have Y e 2B, or, b y (6),

(11) Xe2B.
From (4 )-(6 ) we obtain the form ula
(12) X~A+E[ii<ir].
*a^
B y an indirect argum ent (analogous to the one used in refuting
the assum ption (9) in the p roof o f Th. 16) w® easily infer from
100 F U N D A M E N T A L CONCEPTS OF V, $ 7

(2 )-(5 ) that for every number A < ' co which is distinct from all
numbers A^ w ith /x < it either X 0+ { x y} e 2B holds or there is a
num ber v < n w hich satisfies the form ula

B y virtue o f (9) and Th. 46 a we have a fortiori X-{-{xv} e 2B.

H ence b y ( 1) and ( 12) we obtain
(13) for every x e S either x e X or e 2B.
A ccording to ( 11) and (13) the set X satisfies the condition
(jS) o f Th. 53 (for A = X ); consequently,
(14) Z e S .2 B .S .
The form ulas ( 8) and (14) at once give
£ (A ).S .9 B .3 3 # 0, q.e.d.
Besides the concept o f completeness, and perhaps m ore often
than this concept (especially with regard to logically more com ­
plicated and com prehensive disciplines), tw o other concepts
are treated in m etam athem atics which are related in content to
com pleteness, although they are logically weaker and thus more
general, viz. non-ramifiability and categoricity (m onom orphy ).1
These concepts are not reducible to those o f sentence and conse­
quence, f The definition and the establishm ent o f the funda­
m ental properties and o f the m utual connexions o f these concepts,
as well as the clarification o f their relation to the concept of
com pleteness, m ust be left to special m etadisciplines.

§ 8 . C a r d in a l and O r d in a l D e g r e e op Co m p l e t e n e s s
In order to obtain a classification and characterization o f in­
com plete sets o f sentences, we introduce here the concept of
the degree o f completeness o f a set o f sentences, and we do this
in tw o ways, nam ely b y correlating a cardinal number and an
ordinal num ber w ith every set o f sentences. The cardinal degree

1 Information about these concepts is given in Fraenkel, A. (16), pp. 347-64,

where the literatute on the subject is also listed. For the definitions of these
notions see also X, pp. 310-314 £., and X III, p. 390.

t Nor can they be defined in terms of those notions which are discussed
in xn.
V, §8 M E T H O D O L O G Y OF D E D U C T IV E SCIENCES 101

of completeness o f the set A o f sentences, sym bolically g(4 ), is

the number o f all system s which include the set A ; the ordinal
degree of completeness, in sym bols y(A), is identical with the
smallest ordinal number tt for which there is no strictly increasing
sequence o f type tt o f consistent systems which include the set A .
In form ulas we thus have
D efin ition 8. (a) g(J.) = Q(^4) . S ;
(6) y(A) is the smallest ordinal number rr for which there is no
sequence of type tt of sets X v which satisfy the formulas:
(a) X v e £*(-4 ) . S . 2B for v < tt, where A £ S ,
and
(/?) Xp c X v and* X [ l ^ X v fo r p < v < rr.
D ef. 86 m ay be transform ed as follow s:
Theorem 57. y { A ) is identical with the smallest ordinal num ber
rr f o r which there is no sequence o f typ e tt of sentences x v which
sa tisfy the form u la (a) x v e S — C n ^ A -|-25[/x < v]^ f o r v < n,

where A £ S .
P r o o f. L et a num ber £ be given such that

(1) I < y{A).

B y D ef. 86 there is a sequence o f sets X v o f type £ which
satisfy the form ulas:
(2) X ve . S . 2B f o r v <£, where A £ 8 ;
(3) X p ^ X v and X ^ ^ X v f o r fi < v < £.
W e put further
(4) X g = S (in case £ is not a lim it num ber).
B y Ths. 45 and 47 the form ulas ( 2) and (4) give
(5) X v £ S = X g and X v =£ S = X g f o r v <i £.
From ( 3) and (5) we infer at once that the sets X v+1—X y9
where v < £, are distinct from 0 ; hence with every such set
one o f its elements x v can be correlated in such a w ay that
(6) x v e X v+1—X v fo r v <£.
(This correlation does not require the axiom o f choice since by
102 F U N D A M E N T A L CONCEPTS OF V, 5 8
(5) and A x. 1 all sets X K+1—•X v w ith v < £ are subsets o f the
at m ost denumerable set 8.)
I t follow s from ( 2), (3), and ( 6) that A +E \ji < v] s X „ for

every v < £; hence w ith the help o f ( 2) and Th. 8 w e obtain

Cn(A+E{jjL < v]j c X v. B ut since b y (5) and ( 6) xv e 8 —X v,
w e finally have
(7) xv e S -G n (A 4 -E [fi < v]) far v < £ (where A £ 8).

W ith this we have proved that ( 1) always implies (7): for

every number £ < y(A) a sequence o f sentences o f type £ can
be constructed w hich satisfies form ula (a) o f the theorem here
discussed. B ut since b y hypothesis there is no such sequence for
the num ber it, the inequality w < y(A ) cannot hold. Conse­
quently,
( 8) tt y(A ).
B y an analogous argument the inverse inequality
(9) w < y(A)
can be proved. F or let us consider an arbitrary num ber £ < it .
A ccording to the hypothesis there is a sequence o f sentences
x v o f type £ which satisfies the form ula (7). B y putting
for v < £,

we easily conclude from Ths, 9 a, 47, and 1 a that the sequence o f

sets X v satisfies the form ulas ( 2) and (3). B u t b y D ef. 8 b no such
sequence o f sets can be constructed for the number y (A ). Con­
sequently the inequality y(A ) < tt cannot hold and we obtain
the form ula (9).
Form ulas ( 8) and (9) at once give the required identity
y(A) — it , q.e.d.
T heorem 58. (a) I f A s B, thm
0(A ) > q(B) and y { A ) ^ y [ B ) \ [D ef. 8 a, 6]
(b) if A e 5lq(J3), then g(A ) = g(iB) and y(A) = y(B).
Proof o f (b). B y D ef. 2 the hypothesis gives Gn(A) = Gn(B).
From this b y Th. 8 we infer that for every set X e S the in­
clusions A £ X and J S s I are equivalent, whence
£ ( A ) . <3 = = & (£ ). S .
V, |8 METHODOLOGY OF DEDUCTIVE SCXEK'. ES 10S

B y virtue o f B ef. 8 a, b wa obtain at once from the last form ula

the tw o required identities:
s(-d) ~ 3 ^ ) a&d y(A) ~ y(B).
I t is left to the reader to establish the connexions between
the numbers c(A ) and g(i?), or y(A) and y(B), under more special
assumptions (e.g. A e ty(3)—
In the follow ing theorem a characterization is given o f the
m ost im portant categories o f sets o f sentences, in fact o f con­
sistent and com plete sets, in terms o f the notions just defined.
T heorem 58, (a) The formulas A £ 8. g(A ) ^ 1 avdy'A) ~> 1
are equivalent; [ B e f 8 a, h, T h . 1 0 ] 1
(b) the formulas A e 5B, 1 g (A) ^ 2, and 1 y(A ) ^ 2 are
equivalent', [D ef. s o , b, Ths, 52, 10, 47, 9a , 46a]
(c) the formulae A s SB, q(A) ^ 2, and y(A) > 2 are equivalent.
[B ef. 8 a, 6, Ths, 9a, 10. 45, 47, 46 a]
The p roof is quite elem entary.
Corollary 60. (a) The formulas
A e 3tq(£) (or A 6 $P (S )-3B ), g(A ) - 1, a,id y(A) « 1
are equivalent; [Th. 69 a, c, Befs. 2, 6]
(6) the formulas A e 38.58, g (A.) 2 , and ylA) — 2 are
equivalent; [Th. 59 6, c]
(c) the formulas J .e 2 B ~ 5 3 , g(.4) 3, arid •y(J) > 3 are
equivalent. [Th. 69 b, c]
T heorem 61. For every set A tee have
g(A) 2 s* and y(A ) < fl.
Proof. By B ef. 1 we have © £2 $ (& ); hence by D ef. 3 s it
follow s that g(A } < IpftT). On the other hand (h j a well-known
theorem on the cardinal number o f the power set) A x. 1 implies
the form ula 3)(B) < 2R«. Thus we finally resell the required
inequality g(A ) 2s*.
I f y(A ) > Q were the case, then b y D efs. Hb and 6 there
w ould exist an increasing sequence (without repe^ in g terms)
o f the type Q o f subsets o f the sat 8 ; hut this contradicts A x. I
1 It is assumed that in the caso of the hypothesis A o at l*ast the empty
sequence o f type 0 satisfies (vacuously) the conditions (*} and \8'> of Bef, S 5*
104 FUNDAMENTAL CONCEPTS OF V, § 8

according to which the set S is at m ost denumerable. Conse­

quently y(A) < 12, and with that the p roof is com plete.
T heorem 62. I f A £ S and if there is a set X such that
I g (5 and x e S— Gn{A-\-{X—{x}))for every z e X , then
q{A) = 2*° and y{A) = Q.
Proof. Applying Ths. 7 and 32, the follow ing is obtained from
Cor. 37:
(1) if there is a set X such that and x e S—Cn(X—{x})
for every x e X , then
E[Cn(Y) = Y~^~S] = 2*o.
F
W e put
(2) F{Y) = Cn(A-\-Y) for every set Y c 8.
B y the reiativization theorem , Th. 6, all the axiom s postulated
in § 1 remain valid if everywhere in them *Cn9is replaced b y the
sym bol iF i just defined (although the variables in the axioms
m ust also be renam ed). Therefore all consequences o f these
axiom s also remain valid (cf. the remarks after Th. 6 ). This
applies in particular to ( 1) which, b y the transform ation de-
cribed above, and in view o f ( 2), becom es
(3) if there is a set X such that X e Q and x e S—■Gn[A + ( X —{#}))
for every x e X , then
E[Gn(A+Y) - f c l ] = 2*.
F
From (3) with the help o f the hypothesis we obtain
(4) E[C n(A+Y) = T c 1 ] = 2*

B y means o f Ths. 7 and 9 a it is now easy to show that the

form ulas Cn(A + F ) = Y c S and Y e Q (A ). <3 are equivalent.
A ccordingly it follow s from (4) that JQ(Aj.S> = 2*«; and hence
b y means o f D ef. 8 a we obtain im m ediately the required formula
(5) q(A) = 2*-.

W e now consider an arbitrary number £ such that

(6) €< a
V, §8 METHODOLOGY OF DEDUCTIVE SCIENCES 105

thus | ^ X0. B y A x. 1 the set X £ S involved in the hypothesis

is countable. H ence a sequence o f type f can be constructed
consisting entirely o f distinct sentences xv o f the set X and thus
satisfying the form ulas:
(7) E[v < tj] c X ^ S and x^ ^ xv for /x < v < £.
Xv

L et v be an arbitrary number < £. From (7) we infer that

< v] s { x v}) £ 8, whence b y Th. l a the

inclusion
Gn(A + E [fi < v]) S G n { A + ( X - { x y}))

follow s. B ut since b y virtue o f the hypothesis

xv e 8 — C n { A + { X —{*v})),
we have further
(3) xv e 8 —Cn[A-\-E\jj, < vj) for every v < £.

Thus ( 6) im plies ( 8): for every number £ < Q there is a

sequence o f sentences o f type £ which satisfies the form ula ( 8).
B ut according to Th. 67 there exists no such sequence for the
number y{A). Consequently y(A) ^ Q, and com paring this
form ula with Th. 61 we obtain at once
(9) y(A) = Q.
The form ulas (5) and (9) form the conclusion o f the theorem .
The theorem and the next corollary im m ediately follow ing
were obtained join tly b y Lindenbaum and the author.
T heorem 63. (a) I f y(A) < at, then g(J.) < N0;
( 6) if y(A) > to, then g(.4 ) > K0.
Proof, (a) I f A is n ot a subset o f S, then b y Th. 69 a
g (^ ) = 0, and so g(4i) < X0. H ence, we restrict ourselves to the
case where
(1) A £ 8.
W e put
(2) jF(Jl) = Cn(A + X ) for every set X £ 8.
B y an easy argument (which is com pletely analogous with the
derivation o f (3) in the previous proof) we infer from Th. 6 with
100 F U N D A M E N T A L CONCEPTS OF v,s$
the help o f (1) and (2) that A xs. 1-4 as well as all their con­
sequences remain valid if ‘ Cri is everywhere replaced hy (F\
In particular, b y Def. 2, Th. 16 can be transform ed in the follow ­
ing w ay:
(3) far every set £ £ 8 ihe.rc is a corresponding sequence of sen­
tences xy of type 73* ^ a> which satisfies the formulas
(*) X, e S - F i E f a < v]) for v < rr

and (B) FtE[y < » ]) = F(X).

B y Th. 5? the type tt o f every sequence o f sentences which

satisfies the form ula (a) o f the condition (3) is less than y{£),
thus according to the hypothesis < Further b y Th. 7 it
follow s from. (2) that for every set X eSH(A).<S the form ula
F(X) = X holds. In view o f this we obtain from (3)
(4) for every set X e £ i(A ), S there is a corresponding sequence of
sentences x v e S of type rr < «*>which satisfies the formula
f (M[v < *]) = X.

A ccording to A x. 1 the set 8 is at m ost denumerable; hence, as

is well known, every set oi finite sequences consisting only of
elements o f 3 is likewise at m ost denumerable.
N ow (4) shows that the fim ctkm F maps such a set o f sequences
onto the class Q (A ).S ; consequently this class is also at most
denumerable. From this b y D ef, 8 a it follows im m ediately that

8(-4) < Kw
(6) B y Th. 57 the hypstlissje im plies that
(5) A c? S,
and that, w ith every number «r < ta (thus < y{A)), a sequence
o f sentences o f typ e v can be correlated so that

(6) x f 's 'Sf—CvJ,A4- E[n < y]} for v < w < -a,

(The axiom o f chovr-e is not used in this step, for, as we have

already m entioned, the sot o f all finite sequences o f sen te aces is
at m ost denum erable.)
V, | 8 M E T H O D O L O G Y OP D E D U C T IV E SCIENCES 107

W e now p u t
(7) Xj,'* = Cn^A+JSJjx, < v]j fo r v < ir < <a,

and let
(8) ft = E Jy < it < » ].

B y the use o f Th. 9 a it easily results from (4 )-(8 ) that

(9) ft s £ ( 4 ) .« .
From (8) it follow s at once that ft ^ N0. B ut on the other
hand the class ft is infinite. F or b y A x. 2 we infer w ithout
difficulty from (5)—(7) that e and therefore
Xf?> # X ’f '1 for < v < it < to, hence b y (8) we obtain ft ^ if
for every it < at. A ccordingly

(10) ft = X 0.
In view o f D ef. 8 a the form ulas (9) and (10) give the required
inequality ^ > K

Cobollary 64. (a) I f y( A ) — a>, then g(A ) =s K0;

[Th. 63:7,6]
(6) if g(A ) < X0, then y(A) < u.-; [Th. 636]
(c) if g(A ) > K0, then y(A) > « . [Th. 83 a]
Cobollary 65. For every set A £ o we h-ave
either 1 < g(A ) < X0 or -< %{A) < 2**.
[The. 39 a, Oi, 63 a, b]
From this corollary Th- 30 is at once derivable as a- special
case (for A. = 0).

T heorem 66, I f y{A) --F- then y(A) <; g(J.}.

Proof. In the case y{A) — 0 the eon elusion is obvious. If
0 < y(A) < a), then the class JQ(»4). £> by virenc of Det. Su.
contains at least y(A) — 1 consistent systems and moreover, by
Th. 47, one contradictory system, namely 3, Tims we have
h U p S > y ( A )— 1 + 1 =a y { A ), which bv Def. 8 5 gives the
formula y{A) ^ g(x4). Finally, if we have to < v(J. j <; fi. then
y ( 2 ) - K 0 and, b y Th. 636, $(A) ^ K0; from this again the
108 F U N D A M E N T A L CONCEPTS OF V, §8

conclusion follow s: y(A) < g(^4). Since b y Th. 61 and the hypo­
thesis y{A) < Q, all possible cases have been dealt with and so
the theorem is proved.
It remains undecided whether this theorem can be extended
to the case y(A) = 12.
The follow ing theorem is o f a more special nature.

T heorem 67. Let 8 e 2t; then, if tt is a limit number, we have

y(A) 7T+ 1 for every set A.
Proof. Let us suppose, in contradiction with the conclusion,
that for a certain set A
( 1) y(A )^7r+l.
H ence, b y D ef. 86 , for the num ber 77 (as well as for every
number £ < tt+ 1) there exists a sequence o f sets X v o f type 77
which satisfies the form ulas:

( 2) I , e J 2 (4 ).S .S B for v< tt,

( 3) Xp c: X v and X^ ^ X p for /jl < v < tt.

L et us put

(4) = 2 x v.
V < TT

Let St = E\y < 7rl. Since 77 ^ 0, we also have St ^ 0; from

this b y (3) we conclude that the class St satisfies condition (a) o f
Cor. 5. Thus, by (2), Jft £ ® .2B holds. Consequently we can
apply Tbs. 12 and 49 to this class; thus b y the form ula 8 e 31
assumed in the hypothesis we obtain Y X e S .2B . B ut since,

b y (2) and (4), A £ X „ - I , we have X „ e £ (.4 ). <5.2B, and

accordingly (2) can be generalized in the follow ing w ay:

(6) I , e Q ( 4 ) . S .2 B for v < 77+ I .
Since b y hypothesis 77 is a lim it number, (4) provides an
analogous generalization o f (3):
(6) Xll<
^ X v and X p =£ X v for f*<v<7r+l.

In consequence o f D ef. 86 the existence o f a sequence of

V, f 8 M E T H O D O L O G Y OF D E D U C T IV E SCIENCES 109

sets (o f type 7 r + l) satisfying the form ulas (5) and (6) contra­
dicts the assum ption (1). W e must have therefore
y(A) tt+1, q.e.d.
The results obtained here which concern the notions g(J.)
and y{A) are rather fragm entary. Only by res dieting ourselves
to those deductive disciplines which presuppose the sentential
calculus are we able to obtain m ore com plete results. In parti­
cular we can then show that, for every set A o f sentences, g (4 )
is either finite or equals 2**, and sim ilarly yUl) is either finite
or equals £2.
In general, when developing the metamatb.ematies o f deduc­
tive disciplines which presuppose the sentential calculus, we
chiefly concentrate upon the same notions which have been
discussed in the present paper. Since, however, the develop­
m ent is based upon logically stronger (though more special)
assumptions, various results obtained here can he supplemented
and im proved.
 VI

ON D E F I N A B L E SETS OF R E A L
NUMBERS!

M athesia ^xoiaws, in general., do not like to operate with, the

notion o f defin ability; their attitude towards this notion is one
o f distrust and reserve. The reasons for this aversion are quite
clear and understandable. First, the meaning o f this notion is
not at all preside: a given ob ject m ay or m ay not be definable
w ith respect, to the deductive system in which it is studied,
the r.vies o f definition which are adopted, and the terms that
are taken a* prim itive. It is thus permissible to use the notion
o f definabiiitv c-nlv in a relative sense. This fact has often
been neglect 3d m m athem atical considerations and has been
the -source ot nmnerc*us contradictions, o f which the classical
exam ple is fum lsiicd b y the well-known antinom y o f Richard 1
The d i s t r u s t o f m a t h e m a t i c i a n s towards the notion in question
is reinforced bv the current opinion that this notion is outside
the proper ku-its o f m athem atics altogether. The problem s o f
making it* ruuuring more precise, o f rem oving the confusions
and misunOTs landings connected with it, and o f establishing
its fimdamental properties, belong to another branch o f science
—-metama'ihematics.
In this article I shall try to convince the reader that the
opinion just m entioned is net altogether correct, W ithout doubt
1 Cf. i*>eenkeh />. (1C, where precise UbUogr&pkieai references w*U be
found.

t . r.Trr al N o t f , The m ein '’ dees of thia article originated in

1929 adc ^ 'ii-v*e*opef! by the author iti his talk oo the T'olish Mathematical
Society, -be n-xon of Lvrb/, in 1930; so3 a >uixmx&ry c f this talk mid^r- the title
Eliber def dficibaie Mengen reeller £&h!en', AanaZsn dc Ui Soci4t6 Poloruii&a ie
/oh 9 (1030), pp. 20 C- 7. The article itself first appereed mider
the title 1 ^nB^rc^ifS do norrifcres r£e(3.1 .’ in JlPvndnifrwitote
Mathetnc&ii.xe. vol. 17 (1931), pp. SI‘>-39, The present English edition has
been suppler^cntod by some new passages which contain elaborations o f the
original te^t. Ths most, important addition has been made in § 2, after Th. 6 ;
s?-e in this oormesicix footnote 7 on p. 120 .
 VI ON D E F IN A B L E SETS OF R E A L NUMBERS n :

the notion o f definability as usually conceived ia o f a m eta-

m athem atical nature. I believe, how ever, that I have found a
general m ethod which, with a reservation to be discussed at the
end o f § 1, allows us to reconstruct this notion in the dom ain
o f m athem atics. This m ethod is also applicable to certain other
notions o f a m etam athem atical nature. The reconstructed con ­
cepts do not differ at all from other m athem atical notions and
need arouse neither fears nor doubts; their study remains en­
tirely within the dom ain o f norm al m athem atical reasoning.
Finally, it seems to me that this m ethod allows us to derive
certain results w hich could n ot be obtained b y operating only
with the m etam athem atical conception o f the notions studied.
A description o f the m ethod in question which is quite general
and abstract %vould involve certain technical difficulties, and, i f
given at the outset, would lack that clarity which I should like
it to have. F or this reason I prefer in this article to restrict
consideration to a special ease, one which is particularly im ­
portant from the point o f view o f the questions which interest
m athem aticians at the present tim e. I shall in particular a nalyse
the notion o f definability o f a single category o f objects, nam ely,
sets o f real numbers. M oreover, my considerations will be in the
nature o f a sketch; I shall content m yself sim ply with co n ­
structing some precise definitions, either om itting the conse­
quences which follow from them , or presenting them w ithout
dem onstration.1

§ 1. T he Concept of D e f in a b l e Sets of R eal N um bers

FROM TitE MffiTAldATREMATlOAL STANDPOINT

The problem set in this article belongs in principle to the

type o f problem s which, frequently occur in the course o f m athe­
m atical investigations. Our interest is directed towards a term
o f which we can give an account that is more or less precise in
its intuitive content, but the significance o f which has not at
present been rigorously established, at least in m athem atics.
1 In this connexion it should be noted that an analogous method can be
profitably applied for the definition o f many concepts in the field o f meta-
mathems,tics> e.g. that o f true sentence or o f a tuniversally valid sentential
function* Of. TO J.
112 ON D E F IN A B L E SETS OF B E A L NUM BERS VI, § 1

W e then seek to construct a definition o f this term which, while

satisfying the requirements o f m ethodological rigour, w ill also
render adequately and precisely the actual meaning o f the term .
I t was just such problem s that the geometers solved when they
established the meaning o f the terms 'm ovem ent’ , ‘line’ , ‘sur­
face’ , or ‘ dim ension’ for the first tim e. Here I present an
analogous problem concerning the term ‘ definable set o f real
num bers’ .
Strictly speaking this analogy should not be carried too far.
In geom etry it was a question o f making precise the spatial
intuitions acquired em pirically in everyday life, intuitions which
are vague and confused b y their very nature. Here we have to
deal w ith intuitions more clear and conscious, those o f a logical
nature relating to another dom ain o f science, metamathematics.
T o the geom eters the necessity presented itself o f choosing one
o f several incom patible meanings, but here arbitrariness in
establishing the content o f the term in question is reduced
alm ost to zero.
I shall begin then b y presenting to the reader the content o f
this term , especially as it is now understood in metamathematics.
The remarks I am about to make are n ot at all necessary for
the considerations that will follow — any more than empirical
knowledge o f lines and surfaces is necessary for a m athem atical
theory o f geom etry. These remarks will allow us to grasp more
easily the constructions explained in the follow ing section and,
above all, to judge whether or not they convey the actual
meaning o f the term . I shall confine m yself to sketching this
m atter briefly w ithout attem pting too much precision and
rigour in its form ulation.
As already m entioned, the notion o f definability should always
be relativized to the deductive system in which the investiga­
tion is carried out. N ow in our case it is quite im m aterial which
o f the possible systems o f the arithm etic o f real numbers is
chosen for discussion. I t w ould be possible, for example, to
regard arithm etic as a certain chapter o f m athem atical logic,
w ithout separate axiom s and prim itive terms. B ut it will be
m ore advantageous here to treat arithm etic as an independent
VI, §1 ON D E F IN A B L E SETS OF R E A L NUMBERS 113

deductive science, form ing as it were a superstructure o f

logic.
The construction o f this science m ay be thought o f more or
less in the follow ing w ay: as a basis we adm it some system o f
m athem atical logic, w ithout altering its rules o f inference and
o f defin ition ; we then enlarge the system o f prim itive terms
and axiom s b y the addition o f those which are specific to arith­
m etic. As a suitable basis we can use, for example, the system
developed in Principia Mathemati-ca b y W hitehead and Russell.1
In order to avoid needless com plications it is, however, con­
venient to subject this system first to certain sim plifications,
especially to sim plify the language, taking care o f course not to
diminish its capacity for expressing all the ideas which could
be form ulated in the original system. These m odifications con­
cern the follow ing points: the ram ified theory o f types is replaced
b y the sim plified theory, so that the axiom o f reducibility is
rejected;2 also the axiom o f extensionalitv is adopted. A ll con­
stant terms, except the prim itive terms, are eliminated. Finally,
we shall suppress sentential variables and variables representing
tw o- and m any-term ed relations.8
In view o f this last point we shall only have variables in the
system which represent (1) individuals, that is to say objects o f
the first order (and in particular real numbers, which are to be
regarded as individuals); (2) sets (classes) o f individuals, i.e. ob ­
jects o f the second order; (3) sets (families) o f these sets, i.e.
objects o f third order, etc. I t is desirable, further, to fix exactly
the form o f the signs which are to serve as variables, and
especially to distinguish them according to the order o f the
objects which they represent. Thus we can agree to em ploy
the signs (x',’3 (x*\...3 (x^k) ’ ,... as variables o f order 1;
(xik)’,'.. as variables o f order 2 ; and, in general, the signs
1 Whitehead, A. N., and Russell, B. A. W. (90).
3 Cf. Chwistek, L. (14) and Carnap, R. (8).
3 An arbitrary two-termed relation R can be replaced in all discussions by
the set o f ordered couples [x, y] satisfying the formula xR y; ordered couples
can also be interpreted as certain sets (cf. Kuratowski, C. (38), Chwistek, L.
(13)). In an analogous fashion many-termed relations can be eliminated. As
regards sentential variables they are completely avoided in Neumann, J. v,
(54).
114 ON D E F I N A B L E SETS OF B E A L NUMBERS VI, §1

‘x'(lf 'xfff’ ,.,. as variables o f order l, taking as values

objects o f the same order,1
As prim itive terms o f the system o f logic it is convenient to
adopt the sign o f negation ' the signs o f logical sum,
and o f logical product ‘ ; the quantifiers: universal ‘XX’ and
existential' and finally the sign ‘ e ’ o f set membership
The meaning o f these signs does not require further remark.8
I t now remains to add to this system the specific prim itive
terras o f the arithm etic o f real numbers. The three signs V , ‘p,’ ,
and V are known to suffice for the definition o f all the concepts
o f this science. The sentential functions ‘^(x^ , xP) ’
and <<s{z)k),x <
,l'!, riOT>)’ assert respectively that ~ 1; and
•x?' are real numbers such that X}k) < afP; and r i%), xp, and
are real numbers such that — xP-j-x{m'1.
la addition to variables and constants it is necessary to have
soma technical signs in the system , nam ely left T and right
‘ Y parentheses. From these three kinds o f sign a great variety
o f expressions can be form ed. Sentential functions constitute
an especially im portant category.8 In order to make this notion
precise we distinguish first the primitive sentential functions,
nam ely expressions o f the type e * ^ +j))!. ‘v(x'k})’ ,
‘pix)*\z{P y and ‘<j{xftk\x'P,xYn)y. Then we consider certain
operations on expressions called the fundam ent,?!- operations.
These are the operation a f negation which when applied to a
given expression p ' yields its negation ‘f \ the opera;lions o f
logical addition and m ultiplication, which form from tw o
expressions Kr/ and ‘q’ , their logical sum \-p { g)‘ and logical

1 I 036 the symbols ! x'F‘ ' x'f',..., only -■>£ «oh«jnata; in s v e v y

particular case, instead o f the indices '(&)’ and 'U}’ there is. aa appropriate
number of small stroke*.
* The signs ‘ I I ' {or '£•') and (or ' + ’ ) sire not iodiioenssble, for they
can be defined by means o f other signs. In principle tbs sign ■£’ is also super­
fluous, for in the pisos o f such expresdona as ' * s Z ' T ® is an element o f the
•set X*) we can uae equivalent expressions o f the ftra> S.’!:) ‘ (*« has the
property J f \ or '« se-tishes the condition X '); I prefer here, however, to
conform to the current language o f mathematical works.
s I especially emphasize that I constantly use the torso 'sentential function'
as a metamathetaaticai term, denoting expressions ol one category (for this
term is sometimes interpreted in a logical sense, a meaning fceiag given to it
like that of the terms ‘ property’ , ‘ condition’, or ‘ clses’ ).
VI, { 1 ON D E F IN A B L E SETS OF R E A L NUM BERS 115

product ‘ (p .q )’ . Finally, the operations o f universal and o f

existential quantification w hich from the expression (p' and
the variable yield the universal quantification ‘ U p ’

and the existential quantification ‘ S p ’ of <

p >with respect to the

variable ‘x $ \
The sentential functions are those expressions which can be
obtained from the prim itive functions b y applying to them , in
any order and any (finite) num ber o f tim es, the five fundam ental
operations. In other words the set o f all sentential functions
is the smallest set o f expressions which contains a? elements all
the prim itive functions and is closed w ith respect to the above
operations. The follow ing are examples o f sentential functions:
2 c(x,.x't Xf )t XT ((•*'' £ r,.)-[-(*, e *»)), ^ Tj (*•■» s xm).
W ithin this style o f notation (which clearly departs from the
sym bolism ox Principia Mathematica1), we introduce in practice
m any sim plifications. Thus we shall use as variables o f the
three lowest orders the signs ‘X ® ’ , and respectively.
In the place o f V ', V ’, V ” , V * '’ we write V , ‘y\ V , V ,
and sim ilarly for the tw o follow ing orders. As negations o f the
prim itive sentential functions w e write
* i W . W * " . a**)*. and '*(**>,30>,**>)>-
Finally, sotting aside the m echanical w ay o f using parentheses
exem plified above in the construction o f sentential functions,®
we omit, parentheses everywhere so long as no nfisrmderstandmg
w ill arise.
W e can classify sentential functions according to the order
o f the variables w hich occur in ' hem. The function which con­
tains at least one variable o f order » , but no variable o f higher
order, is called a sentential function o f order n; thus In the
exam ples given above we have functions o f orders 3, 2, and 3.
1 On the other banc? our natation a g with that o f Sehrdder, K, (02).
* This method Is doe to Lewis {of1, L-swis, O, L (4fi), p 3£7). By using
the Ida*. o f Lukasiewicz it la possible so cfd ihe nse of parentheses completely
in this system: for this purpose it su£k:.e*. use tha dcspxfission* s scX \
*V&y\ {crxyz \ *4 p q %an d4.pq\ i m it t i oi 1 g X ) \ 1v(* i . L;*■{*>y) \ i a{xty Tz) h
>+<?)% and •{*.*)’ IV).
1X6 ON D E F IN A B L E SETS OF R E A L NUMBERS VI, § 1

Am ong the variables which enter into the com position o f a

given sentential function we can distinguish free (or real) and
bound (or apparent) variables. The precise distinction between
these tw o kinds presents no difficulty.1 Thus in the function
TT((*e Y) -f- (x e Z )) 5the variables (Y 9and (Z* are free and V is
X

bound. Sentential functions which are devoid o f free variables,

like the expression ‘T XI (x ^ X )\ are called sentences.
X x
In order to carry out the construction o f the form al system
o f arithm etic which I shall sketch here, it w ould be necessary to
form ulate explicitly those sentences which are to be regarded as
axioms (both the general logical ones and those which are specifi­
cally arithm etical), and then to form ulate the rules of inference
(rules of proof) with the help o f which it is possible to derive
from the axioms other sentences called theorems o f the system.
The solution o f these problem s occasions no great difficulty.
I f I om it their analysis here, it is because they are not o f m uch
im portance for what is to follow .2
L et us now consider the situation m etam athem atically. For
each deductive system it is possible to construct a particular
science, nam ely the ‘m etasystem 5, in which the given system
is subjected to investigation. Hence, to the dom ain o f the
m etasystem belong all such terms as ‘ variable o f the nth
order5, fsenten1ial function5, ‘free variable o f a sentential
fu nction5, ‘ sentence’ , etc., that is to say, terms which denote
the individual expressions o f the system under consideration,
sets o f these expressions, and relations between them . On
the other hand nothing forbids our introducing into the m eta­
system arithm etical notions, in particular real numbers, sets
o f real numbers, etc. B y operating with these tw o categories
o f term s (and with general logical term s) we can try to define
the sense o f the follow ing phrase; ‘A finite sequence of objects
satisfies a given sentential function.’ The successful accom plish­
m ent o f this task raises difficulties which are greater than would
1 Cf. for example, Hilbert, D., and Ackerinauxi, W. (30), pp. 62-54; cf. also
article V III, § 2, Def. 11, p. 178.
* As regards the logical axioms and the rules of deduction see article
V III, § 2.
VI, §1 ON D E F IN A B L E SETS OF R E A L NUM BERS 117

appear at first sight.| However, in whatever form and to what­

ever degree we do succeed in solving this problem, the intuitive
meaning of the above phrase seems clear and unambiguous.
It will therefore suffice to illustrate it here by some particular
examples. Thus the function 6X e is satisfied by those,
and only those, sequences of two terms of which the first term
is a set X of individuals and the second term is a family of
those sets which contains X as an element. The functions
sa(x9y,zy and ‘v(x).v(y).fji(y,zy are satisfied by all sequences
composed of three real numbers x y y t z, where x = y + z and
x — 1 = y ^ z respectively. The function

‘ 2 2 (®(*» y>z) ■/*(«» *)•«(**))’

z u

is satisfied by sequences of two real numbers x and y where

x ^ y - f- 1 . The function i<r{x9y 9y)-\-5(x9y 9y y is satisfied by
any sequence formed by two individuals, and the function
‘(x e X ) . (x e X)' is satisfied by no sequence. Finally, a function
without free variables, i.e. a sentence, is satisfied either by the
empty sequence or by no sequence, according as the sen­
tence is true or false. As we see from these examples, in all
situations where we say that a given finite sequence satisfies
a sentential function, it is possible to establish a one-one
correspondence between the terms of the sequence and the
free variables which occur in the function. Also the order
of every object which constitutes a term of the sequence is
equal to that of the corresponding variable, The special case
of the notion in question where the sentential function contains
only a single free variable is of particular importance for us.
The sequences which satisfy such a function are also composed
of a single term. Instead, therefore, of sequences we may speak
quite simply of objects (namely the unique terms of the corre­
sponding sequences) and say that they satisfy the given function.
Thus the function 6^ ( x e X Y is satisfied by all individuals,
the function ‘T (x g X ) ’ by any set of individuals except the

f For an exhaustive discussion of the subject, see VIII.

118 ON D E F I N A B L E SETS OF B E A L NUSJBEKS VI, § 1

null set. The function x >x )’ is satisfied by only one indi­

vidual, the real num ber 0, and the function ‘2 (/*(y, *) •<*(y>y, y))'
v
b y every non-negative real num ber and b y no other indi­
vidual.
W ith every sentential function with m variables we can now
correlate uniquely the set o f all w -term ed sequences which satisfy
this fu n ction ; in case m = 1, we again replace the set o f one-term ed
sequences b y the set o f objects which are the unique terms o f
these sequences. Consequently, a function which contains a
variable o f order 1 as its only free variable determines a certain
set o f individuals, which, in particular, m ay be a certain set o f
real numbers. The sets thus determined b v sentential functions
are precisely the definable.sets in the arithmetical system considered.
Am ong them v e can distinguish sets o f the first, second, etc.,
orders, corresponding to the orders o f the functions which deter­
mine them .
The m etam athem atical definition o f definable sets, based on
the notion o f satisfaction (o f sentential functions by objects),
then takes the follow ing form :
A set X is a definable set (or a definable set of order n) if there
ie a sentential function (or a sentential function of order n at
most1) which contains some variable of order 1 as its only free
variable, and which satisfies the condition that, for every real
number x e X if and only if x satisfies this function.
The following will serve as examples of definable sets in the
sense of the foregoing definition—and in particular o f those of
order 1: th.u set composed of the single number 0, the set of ail
positive numbers, the set of ail numbers x such that 0 < & < 1,
and many others, The functions which determine these sets
are, respectively *, x) ’, ‘5(x, x, x ) . T (^(y, z ) . a{y, y, y))\
V
*2 2 *)•«(*)}’ > etc* A relatively simple
V *
example of a definable set- of order 2, of which it can be
shown that it is not of order 1. is furnished by the set of ail
natural numbers (including 0), determined by the sentential
1 It is easy to show that the words ‘ at most/ can be omitted without
modifying the meaning of the notion defined.
VI, {1 ON DEFIN ABLE SETS OF REAL NUMBERS 119

funotion

* y>y) •(y e x ))+

+ 2 2 2 (fffo y>0 •” (*). (y e X ) . (s e X ) ) +( z e X )) *.

Such examples can be m ultiplied indefinitely. Eaoh particular
set o f numbers with which we are concerned in mathematics is
a definable set, inasmuoh as we have no other means o f intro­
ducing any set individually into m athem atics than b y con­
structing the sentential function which determines it, and this
construction is itself the p roof o f the definability o f the set.
M oreover it is not difficult to show that the fam ily o f all definable
sets (as well as that o f the functions which determine them) is
only denumerable; while the fam ily o f all sets o f numbers is not
denumerable, The existence o f undefinable sets follow s imme­
diately. Also, the definable sets can be arranged in an ordinary
infinite sequence; b y applying the diagonal procedure it is
possible to define in the 'metasystem a concrete set which would not
be definable ir< the system itself. In this there i3 clearly no trace
o f any antinom y, and this fact will not appear at all paradoxical
if we take proper note o f the relative character p f the notion o f
definability.
In the next section I shall attem pt a partial reconstruction o f
the notion just defined within m athem atics itself. It is evident
a priori that there can be no question o f a total m athem atical
reconstruction o f the notion o f definability. In fact, if this
were possible, we could, b y applying the diagonal procedure,
define within the system o f arithm etic a set o f numbers
provablv not definable in this system, and we should this tim e
find ourselves involved in the original antinom y o f Richard.
For the same reason, it is im possible to define in mathematics
the general notion o f a definable set o f order n. In other words,
it is im possible to construct a funotion which correlates with
every natural number n the fam ily o f all definable sets o f order n.
I shall show, however, that it is possible to reconstruct by purely
m athem atical m ethods the notions o f definable sets o f orders
120 ON D E F IN A B L E SETS OF R E A L NUMBERS VT, § 1

1, 2,..., i.e. the reconstruction can be carried out for each natural
number separately.
The idea o f the reconstruction is quite simple in principle.
W e notice that every sentential function determines the set o f
all finite sequences which satisfy it. Consequently, in the place
o f the m etam athem atical notion o f a sentential function, we can
make use o f its m athem atical analogue, the concept o f a set o f
sequences. I shall therefore introduce first those sets o f sequences
which are determined b y the prim itive sentential functions. Then
I shall define certain operations on sets o f sequences which
correspond to the five fundam ental operations on expressions.
Finally, in im itation o f the definition o f sentential function,
I shall define the concept o f definable set o f sequences o f order n.
This notion w ill lead us easily to that o f definable sets o f indivi­
duals o f order n.
The whole construction w ill be carried out in detail for the case
n = 1 and in its main features for n — 2. I believe that after
this explanation the m ethod o f construction for higher values
o f n will be clear, f

§ 2. D efin able Sets of Or d e r 1 from the Standpoint of

M athem atics
As the field o f our considerations in this section, we can make
use o f any o f the known systems o f m athem atical logic including,
in particular, the calculus o f classes, the logic o f relations, and
the arithm etic o f real numbers, that is to say a system o f logic
with a system o f arithm etic axiom atically constructed upon it.
F or exam ple, we can use for this purpose the system o f logic o f
Principia Mathematica. I t w ould, however, be necessary to re­
place the ram ified theory o f types b y the sim plified theory.1
1 I omit completely the other Simplifications of the Bystem mentioned in § 1,
since here they would be very inconvenient.
t This artiele was conceived as the first part of a more comprehensive
paper. The second half was intended to include a discussion of definable sets
of order n > 2. Since the second half has never appeared in print, we have
included in the present translation some remarks concerning definable sets
of higher order, and in particular a precise mathematical definition of definable
sets of order 2 (see remarks in § 2, following Th. 6). In this connexion compare
also a recent article by the author 'A problem concerning the notion of
definability’, Journal of Symbolic Logic, vol. 13 (1948), pp. 107-11.
VI, §2 ON D E F IN A B L E SETS OF R E A L NUM BERS 121

M ost t)f the sym bols to be used are com m only found in
works on set theory. In particular, the sym bols ‘ O’ and
({x9y>—9&Y denote respectively the null set and the finite set
com posed o f the elements x , z. The sym bol ‘E<f>(xy denotes
X

the set o f all objects x which satisfy the condition <f>. The
sym bol ‘N t9 denotes the set o f all natural numbers (zero in­
cluded) and the sym bol lRV that o f all real numbers. Further,
when I have need to refer to the considerations o f the preceding
section, I shall em ploy the logical and arithm etical signs which
have there been explained.1
The principal instrum ent o f these investigations will be form ed
b y finite sequences o f real numbers and sets o f such sequences.
In order to avoid all possible m isunderstanding the notion o f
finite sequence w ill first be considered more closely.
Let r be any binary relation. B y the domain of the relation r9
in sym bols D(r), I understand the set E {there exists a y such
X

that xry). The counter domain (or range) d{r) is. defined b y the
form ula <T(r) = E {there exists an x such that xry). The relation r
v
is called a one-many relation (or function) i f for all x} y, z the
form ulas xrz and yrz always im ply the equality x = y .f
E very function s, whose counter dom ain is a finite subset
o f Nt I call a finite sequence. The set o f all finite sequences
w ill be denoted b y cSf\ The unique x which satisfies the
form ula xsk for the given sequence s and natural number jfc,
will be called the k-th term of the sequence s9 or the term with
the index k of the sequence s9 and denoted b y ‘sk\ It is not at
all necessary (contrary to the custom ary interpretation o f the
notion o f a finite sequence) that the set d(s) should be a
1 Thus the symbols *+* and V are used in several different senses: namely
as logical signs, as signs of the theory of sets, and as signs of arithmetic. I do
not believe that this will occasion any error within the limits of this article.

t Contrary to the notation adopted in this paper, fonetions are usually

identified not with one-many but with many-one relations, i.e. with relations
r such that for arbitrary xt y, z, the formulas xry and xrz imply y — z. Con­
sequently, what is called in this paper the domain of a function (or, in
particular, of a sequence) is usually referred to as the range or counter domain
of this function, and conversely.
122 O N D E F I N A B L E S E T S O F B E A L N TJM B E K S V I, §2

segment o f the set N t. A sequence a m ay, for example* have

a second term w ithout a first. W hen we have a e S f and
D(a) £ El, the sequence a is called a finite sequence, of real
number8. In accordance w ith these conventions, the null
relation, i.e. the relation r such that D(r) — 0 = d(r), also
belongs to the set Sf.
Tw o relations r and e are identical, in sym bols r — a, when
the form ulas xry and xay are equivalent for every x and y.
In jpartioular, the identity o f the sequences r and a is charac­
terized b y the conditions d{r) — d(s) and rk ~ ab for every
k e <7(r). I denote b y the sym bol V /X ’ (where r is any relation
and X any set) the relation t which satisfies the condition: for
any x and y we have xty if and only if we have xry and y £ X .
W e assume it can easily be proved that a e S f always im plies
ajX e Sf. ajX is called a subsequence of the sequence a, i.e. it is
com posed o f all the term s o f the sequence a whose indices belong
to the set X . Conversely, the sequence a m ay be called a pro­
longation of the sequence ajX.
I shall be concerned for the present with sets o f finite se­
quences. I introduce in the first place the notions o f the domain
and counter domain of a set of sequences (or of arbitrary relations)
S, in sym bols D(S) and d(S). I define them b y the form ulas
D efin itio n 1. (a) 1){S) — Y D{s),

(p) a{S)

The counter dom ain o f a set o f finite sequences is evidently

the union o f the finite sets o f natural numbers.
I shall distinguish here a category o f sets o f sequences which
is especially im portant for us, nam ely, the homogeneous sets.
D efin itio n 2. £ is a homogeneous set of sequences (or o f
arbitrary relations), if (7(a) — d(S) for all s e S.
A ll the sets o f sequences w ith which we shall deal in the sequel
will be hom ogeneous sets. This is explained b y the fa ct that I
operate here with sets o f sequences as the m athem atical analogues
o f sentential functions. N ow all sets o f sequences determined by
sentential functions are com posed o f sequences having the same
VI, § 2 ON D E F I N A B L E S E T S OF B E A L N U M B E R S

counter domain; they are thus homogeneous in the seme of Def. 2

(ef. § 1 in this connexion).
The largest homogeneous set o f sequences with a given
counter domain N is the set i? v defined by the formula
D efin itio n 3. JlK — 8f.J8((7(«) ~ N).
$
By putting N = Ci[8) we can associate a finite set 2v c.
such that 8 c: with every homogeneous set $ of sequences.
Except in the case S = 0, tins correspondence is unique. Con­
versely, when S £ IiN, 8 is a homogeneous set and we have
(again except when 5 — 0) <T{&) s= Ar.
I next define certain special sets of sequences Vk> and
8 ^ m which I call primitive .sets:
D efin itio n 4. (a) 77 —
C/j. ... 1^'fi.xu
7rwyb;.
. ^ — -
a
<*) ^hri ~ £ ^ (°\ ^
&
(c)

This the primitive sets are the sets of sequences determined

by the primitive sentential functions of order I:
and
I next introduce five fundamental operations on sets o f se­
quences: complementation-, addition? and multiplication, as well
as summation and mxdtiplication with respect to the k-ih terms.
The results of these operations wail be denoted respectively by
the symbols *8% lS ?T \ « % S\ and The
& A*

results o f these operations arc always homogeneous sets of

sequences.
D efin itio n 5. & ~ Ra ^)~- 3.
D efin itio n 6.
(a) # 4 - ® t±l-2(*') e -S' or «;<T(T) e n

(*) £ ? 2’ = B o t t o m .* ( « /W ) e and u/(J(y) e 27).

According to these definitions the set S consists of all se­

quences whose counter domains equal C(S.} and which do not
124 ON D E F I N A B L E S E T S OF R E A L N U M B E R S VI, § 2

belong to the set 8 . The set 8-%-T is the set o f all sequences
o f real numbers with the counter dom ain Q (S)+Q (T) which
contain am ong their subsequences either a sequence o f the
set S or a sequence o f the set T (or both ); sim ilarly the set
8 ? T is the set o f all sequences w ith the counter dom ain
Q{S)Jr G{T) which contain am ong their subsequences both
a sequence o f the set 8 and a sequence o f the set T. Thus
3 is the set o f all sequences s com posed o f three real
numbers sv s2, and sz such that we have 8± # 1 or s2 < s3,
and M1>2? M23°S3AA is the set o f all sequences 8 w ith four term s
8V s2, and s4 satisfying the form ula s^ < s2 < s3 = 2.s4.
The operations o f com plem entation, addition, and m ultiplica­
tion just introduced approxim ate to the corresponding opera­
tions o f the algebra o f sets and offer m any analogous form al
properties, especially in the dom ain o f hom ogeneous sets. For
exam ple, they satisfy the commutative and associative laws, the
— o— .a .2.
tw o distributive laws, the law of De Morgan, S~\~T = S ? T , the
form ulas
54-0 = 5, -S?0 = 0, 59$ = 0,
and m any others.
This correspondence is im paired only in a single detail,
nam ely, there does not exist any set having exactly the same
form al properties as the universal set 1 o f the algebra o f sets,
J3.
and there exist m any different sets S such that $ — 0 (am ong
hom ogeneous sets all sets o f the type RN have this property),
o
while 0 is a special set, nam ely B0, com posed o f a single element,
the null relation. Consequently the form ula
o JL 0.2.
5 + 5 = T+T
is not in general satisfied; the law of double complementation
dr -a.
5 = 5 fails in the case where 5 = 0; the second law of De
Morgan, „ 0 * _o
S °.T = B + T ,
requires the hypothesis S°- T + 0 ; etc. From the practical
point o f view this fact does not involve m uch inconvenience;
VI, §2 ON D E FIN AB LE SETS OF REAL NUMBERS 125
.O
in the place o f the set 1 we can usually em ploy 0 , which pos­
sesses a series o f analogous properties, e.g.

fif-J-0 = 8 ± 8 , *S?0 = S , etc.

I f we restrict ourselves to the consideration o f subsets o f
any fixed set ltN, by letting BN = 1 and changing Def. 5
~ ©
in only one point, nam ely b y putting 0 = 1, the operations 8,
S j - T , and 8 ° T coincide com pletely, as is easily seen, with
the corresponding operations in the algebra o f sets.
The operations o f addition and o f m ultiplication can easily
be extended to an arbitrary finite number (and even to an in ­
finite num ber) o f sets o f sequences; the results o f these generalized
operations are denoted respectively b y the sym bols

and
D efin itio n 7. (a) 'f, 8 — Rocg^a,}. E (there exists a sequence
K 8
t e Ra(g) satisfying the form ulas tjd(s) — s and t e 8),
o
(b) J ^ S = •E (for all sequences t e Ba(g) the form ula

t/d(s) = e im plies the form ula t e 8).

L et 8 be any hom ogeneous set o f sequences. I f k e d(S), the
sets ^ S and J J 8 coincide, according to the above definition,

with the set 8. If, how ever, k e d(8), then ^ 8 is the set o f

all the sequences whioh are obtainable from those o f the set 8
o
b y the om ission o f the it h term s, and the set J j 8 is com posed

o f all sequences not containing the it h terms and whose every

prolongation b y the addition o f some K h term belongs to
the set 8. F or exam ple, we verify im m ediately that the set
2 (-3*8,1 *8 formed o f all sequences composed o f a single
2
term sx whioh is a non-negative real num ber; the set
^ ($ 1,2,3 i® com posed o f all sequences with tw o identical

terms s1 = s2; JX Mlfi is the em pty set, and J J -?-& ■>«) is the
2 3
set o f all sequences s with tw o terms where 31 =
126 ON D E F I N A B L E SE T S OF R E A L N U M B E R S VI, § 2

The laws which express the form al properties o f the opera-

O Q
tions V S and y S and the relations between these operations
and those exam ined above are com pletely analogous, especially
where hom ogeneous sets are concerned, with the laws o f logic
for the quantifiers ‘ ]£! and ‘ J J ’ , for the signs o f negation,
logical sum. and logical product. As examples, I m ention the
form ulas

t p - t p , t P ^ P ’

also the distributive laws

! < « + * ) - ! and £ (S!T) = £ S ? Q T,

and finally the laws o f D e M organ

(this last form ula fails for the case where S — O).*1
The analogy between the laws ju st discussed and those o f
logic is n ot at all a m atter o f chance. I t is easy to show that
it is possible to set up a very strict correspondence between the
fundamental operations on sentential functions studied in § i,
and those on sets o f sequences introduced in Defs. 6-7. This
can be done in the follow ing w ay: let f lt /*>••• be any senten­
tial functions (o f order 1) and Sv 8Z,,.. the sets o f sequences
determ ined b y them : every function obtained b y one o f the
fundam ental operations when applied to the given functions
determines the set o f sequences whioh can be obtained with the
help o f the appropriate operation when applied to the sets
corresponding to these functions. A certain departure from this
general phenom enon only appears in the case where the func­
tion considered is obtained b y the negation o f a function whioh
determines the null set. Thus, for exam ple, to the function
z*t xf" ) . 5{%', x ”, z!*)' corresponds, to be sure, the set

1 For the reason* dismissed above-, p. 124.

VI, § 2 ON D E F I N A B L E S E T S O F B E A L N U M B E B S 127
... ■
fy lA * } — an d not the set -^0 = ^1,2^?^.2,3 (°f - the
remarks on the subject o f Defe. 5-6).
Only one m ore operation on sets o f sequences remains to be
m entioned, nam ely the replacem ent o f the index k b y l in the
sequences o f a given set. The result o f this operation m il be
‘k ’
denoted by the symbol j-S .

D e f in it io n 8. -S Sf. E (there exists a sequence s e S such

L
that we have either k e d(s), d(t) = <Z(s)-~ {& }+ {!}, sk = tu and
«l(d{s)~{k}) — t/(d(8)—{k}) or k e d(a) and s = t).
Thus, when S is a hom ogeneous set o f sequences, l e Nt and
k
k e d(S), the set y S is form ed o f all the sequences which can
t
be obtained from certain sequences o f the set 8 b y replacing
in the &th term s the index k b y l, w ithout altering any other
term s. F or exam ple,

tU lA — -^ L ,3 - % 8l&» — ^1,2,2,

If, how ever, k dots n ot belong to the counter dom ain o f the
j
set S o f sequences, we have j S — 8', e.g.
if

f = XJV | 2 ^2,1,3 “ jt^2,l,2> e^°*

2 2
k
The operation j S , when applied to a hom ogeneous set o f
l
sequences, always yields a hom ogeneous set o f sequences; if,
in particular, S £ 2li;, where N s Nt and l e Nt, we have
k k
y S s -Kv-tei+a; or v»$ £ Em according to whether k e N or
l if
k e N * For homogeneous sets, the definition o f this operation
pan be simplified and formulated in the follow ing w ay:

if ft /.
128 ON D E F IN A B L E SETS OF B E A L NUM BERS VI, § 2

In the same sense in which the fundam ental operations on

sentential functions correspond to those on sets o f sequences,
a certain operation, which I have not had occasion to m ention
k
in § 1, corresponds to the operation jS . This is the operation
V
o f substituting (he free variable for the free variable ‘of®’ in
a given sentential function. The intuitive significance o f this
operation is, incidentally, quite clear F or exam ple, b y substi­
tuting in the sentential function

X
the free variable ‘y ’ for the free variable V we obtain
‘A y , y ) - 1 o { x , y , z ) \
X

These functions determ ine the sets o f sequences

^ 1,2 ° 2 ^1,2,3 -^2,2 ° *^1,2(8

respectively.
N ow we are ready to define the notion o f definability of order
1, first for finite sets o f sequences o f real numbers, and then
for sets o f real numbers them selves. The fam ily o f definable sets
o f sequences will be denoted b y 'B f and that o f definable sets
o f individuals b y ‘B ’ . Instead o f saying definable of order 1 we
shall say elementarily definable or arithmetically definable.
D e f in it io n 9. is the product (intersection) o f all the
B f

fam ilies o f sets X which satisfy the follow ing conditions:

(a) Uke X , Mk le X , a n d SkJm e X for any natural numbers k, l,
and m ; (j8) if S 6 X , then 8 e X ; (y) i£ 8 e X and T e X , then
S-i-T e X and S t T e X ; (8) if k e Nt and 8 e X, then f S e X
V
aad^SeX .

D efinition l.o. B = E (there exists a set S e B f such that

D(S) = X and d(S) is a set com posed o f only one element).
N ow the question arises whether the definitions just constructed
(the form al rigour o f w hich raises no objection) are also adequate
VI, § 2 ON D E F IN A B L E SETS OF B E A L N U M BER S 129

materially; in other words do they in fact grasp the current meaning

of the notion as it is known intuitively ? Properly understood, this
question contains no problem o f a purely m athem atical nature,
bu t it is nevertheless o f capital im portance for our considerations.
In order to pu t this question into a precise form , let us sup­
pose that the m aterial adequaoy o f the m etam athem atical
definition o f definable set o f order n, which we reached in the
preceding section, is beyond doubt. The question proposed then
reverts to a problem which is quite concrete, nam ely, whether
D ef. 10 is equivalent to a certain particular case o f this m eta-
m athem atical definition, nam ely the case where n — 1. This
last problem obviously belongs to the dom ain o f m etamathe­
m atics and is easily solved in the affirmative.
W e have in fa ct already established that there is a strict
correspondence between prim itive sets o f sequences and the
fundam ental operations on these sets on the one hand, and the
prim itive sentential functions o f order 1 and the fundamental
operations on these expressions on the other. W e have even
rigorously defined this correspondence (pp. 122 and 127). From
these facts, D ef. 9, and the definitions o f § 1 o f a sentential
function and its order, we can show w ithout difficulty b y in ­
duction that the fam ily Stif is exactly the fam ily o f sets o f
sequences whioh are determ ined b y sentential functions o f order
l .1 I t follow s alm ost im m ediately that the fam ily 3) coincides
w ith that c f the definable sets o f order 1 in the sense o f § 1.
I f we wish to convince ourselves o f the material adequacy
o f D ef. 10 and o f its conform ity w ith intuition w ithout going
beyond the dom ain o f strictly m athem atical considerations, we
must have recourse to the em pirical m ethod. In fact, b y exam in­
ing various special sets w hich have been arithm etically defined
(in the intuitive sense o f this term ), we can show that all o f
them belong to the fam ily conversely, for every particular
set belonging to this fam ily we are able to construct an ele­
m entary definition. M oreover, we easily notice that the same

1 The breakdown mentioned on p. 126 in the correspondence between the

negation of sentential functions and complementation of sets of sequences
only complicates the proof a little.
ISO ON D E F I N A B L E S E T S OF R E A L N U M B E R S VI, § 2

com pletely m echanical m ethod o f reasoning can be applied in

all the cases concerned.
The follow ing shows what this m ethod in volves: Let A be any
set o f numbers elem entarily defined with the help o f the prim itive
notions ‘ 1’ , ' and The definition o f the set A m ay be
pu t in the form :

(1) A = E jm .

The sym bol here represents a certain sentential fu nc­

tion containing the variable ‘of®9 o f order 1 as its sole free
variable. It m ay also contain a series o f bound variables
€a W , etc., provided that all these variables are also o f order 1
(since if this were not the case the definition could not be called
elem entary). As we know from logic, the function can
be constructed in such a w ay that it contains no logical constants
other than the signs o f negation, o f logical sum and product as
well as the universal and existential quantifiers. In the same
w ay we can eliminate all the arithm etical constants except the
signs V , and V , corresponding to the three prim itive
notions.
I t is not difficult to show in each particular case that the
form ula (1) can be transform ed in the follow ing w ay: the sen­
tential functions ‘/i(xfh\aPy, and <o(ufk\xfiiQfln)y (or
‘at® ^ and are replaced by
sym bols o f the form ‘ Uk , ‘Mkj , and cSkjtm\ respectively; the
signs o f logical operations b y those o f the corresponding opera­
tions on sets o f sequences, introduced in Defs. 5 -7 ; and finally
the sym bol ‘ E 9by ‘D 9.1 B y this transform ation the form ula (1)

takes the form

(2) A = D(S),
where in the place o f ‘S 9there is a com posite sym bol, the struc­
ture o f which shows at once that it denotes a set o f sequences
o f the fam ily £&f with counter domain consisting o f a single

1 In the course of these transformations use may be made of certain formulas

which are established in VII.
VI, §2 ON D E F IN A B L E SETS OF R E A L N U M BER S 131

element. B y applying D ef. 10 we conclude at once that the

set A belongs to the fam ily
I shall now give som e concrete exam ples:
1. L et A be a set consisting o f the single number 0. W e see
at once that A — E(x — z-f-.r), i.e. that A — Eo{x,x,x). B y
X X

transform ing this form ula in the way just described we obtain
A = D(Sltl'j), so that A e 2 .
2. L et A be the set o f all positive numbers. As is easily
established, we have A = E {x =£ and there exists a y such
X

that y < x and y = y + y ) , i.e.

A = E(o { x , x, x ). J,{ti(y,x).a{y,y,y))\,
x ' y 1
jS. o x

(<S|,1|t ° y r (My, °8w )\, and again we have

A s 2$.
3. Finally let A — E (0 < x < 1). W e transform this form ula
X

successively as follow s:
A — E (ithere exist y and z such that y < x, y = y + y 9
X
x < z, and z = 1)
= ^ 2 2 (My.*) •°(y>y. y) •M*»*) •u(2))
x y z

W e thus have in this case also A e 2$.

The same m ethod can be applied in the opposite direction.
A single exam ple w ill suffice to illustrate this:
4. L et A be a particular element o f the fam ily for example
let A = d ( ^ 2 ($ 2,1,3 -$ 222° ^ ) ) • W e easily obtain the follow -
'2 3 '' '' '
ing transform ation o f this form ula

^ = ^ 2 2 ( a(y,x,z) . o(y.y,y).v(z)) — E (there exist y

X v Z X

and z such that y = x-\-z, y = y + y , and z — 1).

W e see that the set A can be arithm etically defined. It is, in
fact, com posed o f a single number, — 1.
132 OK D E F IN A B L E SETS OF R E A L NU M BERS VI, § 2

As a consequence of these considerations the intuitive adequacy

of Def. 10 seems to be indisputable.
The sets o f finite sequences and especially the arithmetically
definable sets o f numbers in the sense established above are
(from the point o f view o f analytic geometry) objects o f a very
elementary structure. In order to characterize them in terms
as familiar as possible, I shall use the well-known concept
o f a linear polynomial with integer coefficients. Polynomials are
here understood as functions whioh correlate real numbers with
finite sequences o f real numbers (in other words as one-many
relations whose domains contain only real numbers and counter
domains only finite sequences o f such numbers). More precisely,
the linear polynomial p with integer coefficients is determined
by a finite set N o f natural numbers, by a sequence a o f whole
numbers with counter domain N, and by a number 6, and it
correlates the number

p (*) ==Jk° ‘ " ,‘ + 6

■with every sequence s e R#. With the help o f this concept I shall
define a special category of sets o f finite sequences, the elementary
linear sets.
D efin itio n 11. S is called an elementary linear set if there
exists a linear polynom ial p w ith integer coefficients such that
either s = = 0 and £>(«) s Rl),
$

or 8 — R(p(s) > 0 and D(s) c Rl).

An elementary linear set is simply the set o f all sequences e

o f real numbers which are also solutions either o f a linear
equation o f the type ‘p{s) — O’, or o f an inequality o f the type
*jp(«) > O’ .
Erom Def. 11 and the interpretation o f a linear polynomial
adopted here, it follows that every elementary linear set is
homogeneous.
We are now in a position to formulate the following theorem
characterizing the family o f sets 3 f.
VI, §2 ON D E F IN A B L E SETS OF R E A L NU M BER S 133

T heorem 1. In order that a set S of sequences of real numbers

should belong to the family 3)f, it is necessary and sufficient that
S should be a finite sum of finite products of elementary linear sets
(with a common counter domain).1

Proof. That the condition is sufficient is easily proved from

Def. 9 and an easy lemma stating that every elementary linear
set belongs to the family 3 f. To show that this condition is
necessary we use an argument by induction: we show that the
family o f all sets which satisfy the given condition contains the
sets Uk, Mkj and Skii>m(which is clear), and then we 3how that it
is closed with respect to the five operations introduced in Defs.
5-7. Hence, we oonclude at once, with the help of Def. 9, that
every set 8 e 2fif satisfies this condition. A certain difficulty
arises with the operation ^ (to which by the De Morgan laws
o
the operation TT is reducible). The question reduces to proving
a lemma o f algebra showing that a necessary and sufficient con­
dition for a system o f linear equations and inequalities in several
unknowns to have a solution with respect to one o f the unknowns,
is that it be representable as a logical sum o f systems in which
only the other unknowns appear.
We see in this demonstration the intimate connexion between
the operations and the elimination o f an unknown
from a system o f equations and inequalities fam iliar in
algebra.
Sets o f numbers o f the follow ing types: E (x = a), E ( x > a),
X X

E( x < a), and E( a < x < 6),2 where a and b are tw o arbitrary
X X

rational numbers, I shall call intervals with rational end points.

With this convention we can easily deduce from Th. 1 the follow­
ing corollary concerning the family

1 Sum and produot may here be understood rather in the sense of Def. 6
(and the condition of the common counter domain is then superfluous) than
in the ordinary sense of the algebra of sets.
* And, if necessary, Buoh other analogous types as E(x < a), E(a < * < & ) ,
* X
eto., but this is not indispensable here.
IS4 ON D E F IN A B L E SETS OF R E A L NUM BERS VI, § 2

T heorem 2. In order that a set X of real numbers should

belong to the family it is necessary and sufficient that X is the
sum of a finite number of intervals with rational end points.
We can easily formulate (and prove) a metamathematical
theorem which is an exact analogue o f Th. 1. This metamathe­
matical result leads us to a conclusion that, in the system of
arithmetic described in § 1, every sentence of order 1 can be
proved or disproved. Moreover, by analysing the proof of this
result, we see that there is a mechanical method which enables
us to decide in each particular case whether a given sentence
(of order 1) is provable or disprovable.
It should be emphasized that the two preceding theorems
have only an accidental character and have only a slight bear­
ing upon the leading idea of this work. These theorems are
due to the fact that we have chosen the sets Uk, Mkj, and Skj m
as the primitive sets of sequences, i.e. restricted ourselves to a
system of arithmetic which contains as the only primitive
notions the number 1, the relation ^ and the operation -f.
Now it is well known that arithmetic can also be based on many
other systems of primitive concepts, and these systems may
even contain superfluous concepts which can be defined by
means of other notions. Nothing compels us, then, to make the
definitions of the families 2 f and 2 depend just on the parti­
cular sets Uky Mkp and Sk^m. On the contrary, these sets may
be replaced by others, and the extension and the structure of
the two families may thus undergo quite essential modifications.
It is easy to see, for example, that the arithmetic of real numbers
can be based on the two primitive notions of sum and of product.1
We can then replace in Def. 9 the sets Uk and Mkl by the sets
Pc,l,m ” (s k G sl - S7n 6 an^ s k ~ Sl ‘ Sm)>
s
while retaining the sets Skttm; Def. 10 remains unchanged. To
obtain for the new notion of definability the results analogous
to Ths. 1 and 2, we make the following changes:
(1) In Def. 11 and Th. 1 we delete the term *linear’ every­
where (i.e. instead of linear polynomials, we consider poly-
1 Cf. Veblen, O. (87).
VI, I 2 ON D E F IN A B L E SETS OF B E A L N U M B EB S 135

normals o f arbitrary degrees); (2) in Th. 2 we replace the term

‘rationed’ by ‘algebraic\ Ths. 1 and 2 in the modified formula­
tions then still hold; their proof, however, becomes more in­
volved and difficult.! Thus in this case the family 2 has
undergone an extension. We have here, in fact, a system o f
primitive concepts which is stronger in the logical sense. In
many cases this extension is still greater, especially when the
notion o f natural number is taken as a primitive notion, and
the sets Nk = . E(sk e Nt) are primitive sets. The family 2
8

may then contain sets which are quite complicated from the
point o f view of analytic geometry, e.g. the so-called projective
sets of arbitrarily high classes.1
In order to deprive the notion o f elementary definability (of
order 1) o f its accidental character, it is necessary to relativize
it to an arbitrary system of primitive concepts or—more pre­
cisely—to an arbitrary family o f primitive sets of sequences.
In this relativization we no longer have in mind the primitive
concepts o f a certain special science, e.g. o f the arithmetic of
real numbers. The set Rl is now replaced by an arbitrary set
V (the so-called universe of discourse or universal set), and
the symbol Sfis assumed to denote the set of all finite sequences
8 such that the domain o f a is included in F ; the primitive sets
o f sequences are certain subsets o f Sf. We can even abstract
from the type of objects which constitute the set F, and we
can treat the terms occurring in the definition which we are to
construct as ‘systematically ambiguous’2 terms, intended to
1 Cf. V H . I would point out that it seems profitable from the practical point
of view to base arithmetic on ‘strong* systems of primitive notions, for it is
convenient to have a sufficiently large category of arithmetical concepts which
do not require definitions of order greater than 1.
# Cf. (90), vol. 1, pp. 39-41.

t By analysing this proof we obtain (just as in the case of the original

Ths. X and 2) some conclusions of a metamathematical nature; we see that,
in the modified system of arithmetic, every sentence of order i is still provable
or disprovable, and that there is a mechanical method which permits us in
each particular case to decide whether or not such a sentence is provable.
For a detailed proof of all these results see the monograph A. Tarski, A Decision
Method for Elementary Algebra and Geometry, 2nd ed., Berkeley, California,
X951.
136 ON D E F IN A B L E SETS OF B E A L N U M BER S VI, § 2

denote simultaneously objects o f different types. In this way,

while considerably exceeding the limits o f the task we set our­
selves at first, we reach the general notion o f an arithmetically
definable set of finite sequences (with all the terms of a given type)
with respect to the family & of primitive finite sequences (with all
the terms o f the same type). The family o f all these definable
sets will be denoted by the symbol 'ShflfP)'.
12. 3>f(!F) is the product o f all the families
D e f in it io n
which satisfy the condition (a) S' c2£, as well as the conditions
09)-(8) o f Def. 9.
In imitation o f Def. 10, we shall make use o f the family
S>f(^) in order to define the family @{<F) o f all arithmetically
definable sets (composed o f objects o f a given type) with respect
to the family & of primitive sets of sequences (with terms o f the
same type):
D efin itio n 13, = E (there exists a set S e thf{!F) such
x
that jD(8) == X and d (S ) is a set composed o f a single element).
In a completely analogous way we introduce the notion o f an
arithmetically definable two-, three-, and n-termed relation (between
objects o f a given type) with respect to the family& .
From Defs. 12 and 13 it is easy to deduce various elementary
properties o f the families 2 f { ^ ) and & {& ), such, for example,
as the following:
T heorem 3. (a) @)f{!F) is a family which satisfies the condi­
tions (a)-(S) of Def. 12; it is in fact the smallest of these families',
(b) 3>f{F) = 2 9 f(9 ) and & (& ) = T where (5 is the set
»E(S #6(5
of all finite subfamilies of the family
(c) i f & £ S?, we have S>f(^) £ &ft@) and
(d) ® f{9 f(& )) = ® f{3r) and 2>{3>f{&)) = £>(&);
(e) i f the family & is at most denumerable, then the families
t&fi^) and are also at most denumerable.
It is to be noted that in the proof o f the above theorem we
need not make use of the specific properties of the operations
which figure in the conditions (/3)—(S) o f Def. 12. It can be based
exclusively on the fact that the family 3)f{& ) is defined as the
VI, §2 ON D E F IN A B L E SETS OF B E A L N U M B ER S 137

product o f all the families SC containing J*" and closed with

respect to certain operations applicable to a finite number o f
sets o f families SC (this last detail occurs in the proof o f (6)
and (e)).
We are able—and this is an important point— to prove Th.
3 (e) in an effective fashion; we are able to define a (well-deter­
mined) function which correlates an infinite sequence S* satis­
fying the formula
m ^ ) = D(JS*) (i.e. m * ) = {£?> Si,...,S*,...})
with every infinite sequence 8 o f sets (of type to) such that
3? = D (8) (i.e. such that^" = {8v 8z,...,Sn,
In all the applications o f Defs. 12 and 13 to concrete families
we never have to deal with families J*-o f a completely arbitrary
nature; they always satisfy certain conditions, defining what we
shall call regular families.
14. S? is a regular family, if (a) 3F is a non-null
D e f in it io n
family which is at most denumerable; (/}) every set 8 e 3? is a
homogeneous set o f sequences; (y) if &e Nt, l e Nt, and 8 e & ,
we have j 8

An example o f a regular family is the family composed o f all

the sets Uk, Mk}, and where k, l, and m are any natural
numbers.
The postulate o f regularity, imposed on the families 3? to
which we apply Defs. 12 and 13, appears quite natural when
we recall that we are operating with primitive sets as the mathe­
matical representatives o f the primitive terms o f some deductive
science. Now, as is well known, it is possible to choose for each
deductive science a system of primitive terms consisting exclu­
sively o f so-called predicates, i.e. o f signs such as V , and V
which, when accompanied by a certain number o f variables,
tiie arguments, form sentential functions. It is precisely these
sentential functions which we call primitive functions when
we introduce the sets of sequences determined by them into
the family All the expressions, in particular therefore all
the primitive functions, in a given deduotive system are at
138 OK D E F IN A B L E SETS OF R E A L NU M BER S VI, § 2

most denumerable in number—it is for this reason that we

have condition (a) of Def. 14; the sets of sequences determined
by sentential functions are always homogeneous—hence we
have condition (J3). Finally, every time we consider a sen­
tential function as one o f the primitive functions, we con­
sider also every function which can be obtained from it by
substituting one variable for another—therefore the family 3?
should be closed with respect to the operation j S which is

the mathematical analogue o f tho operation o f substitution.1

It is precisely this which explains condition (y) o f Def. 14.
It is to be noted that the condition (y) may be omitted in
Def. 14 provided we introduce into Def. 12 an analogous assump­
tion on the families SC. Such a procedure would conform to
that metamathematical attitude towards the method of con­
struction of deductive systems which admits, as primitive
expressions of the system, not predicates, e g. V , ‘/x\ and V ,
but concrete sentential functions, e.g. ‘u(x)s, y)\ and
'c(x,y, z)\ which are identified with the primitive functions;
then the sentential functions are obtained from primitive
expressions net, only by using the five fundamental operations,
but also with the help o f the operation of substitution. Such
an attitude, which incidentally is not common in practice, is in
principle possible and correct, although it would involve some
inessential complications.
However, according to the point of view we have adopted,
the inclusion of this assumption in Def. 12 (and in Def. 9) is
superfluous, as the following shows:
T heorem 4. I f 3? is a regular family, then is also
regular.
This is easily obtained from Defe. 5-8 and 12 and Th. 3e.
The other properties of the family &f{*F) under the hypothesis
that 3F is regular can be obtained by analogy from theorems
o f metamathematics, esj>ecially from those which express rela­
tions between arbitrary sentential functions and the primitive
1 Cf. the remarks on p. 127 following Def. 8.
VI, §2 ON D E F IN A B L E SETS OF B E A L N U M BER S 139

functions. As an example I give here the following theorem,

which corresponds to the well-known theorem on the reduction
o f every sentential function to the so-called normal form,1 and
which greatly facilitates reasoning in this domain:
T h eo rem 5. For any regular fam ily a necessary and suffi-
_o
dent condition that 8 e is that 8 = 0 or that 8 is of the
form

where T is a finite sum of finite products of sets X e 8F and

o
their complements X ,a where n is an even natural number and
where lt (for 1 < * < n) and ktJ (for 1 < i < n and 1 < j < lt)
are arbitrary natural numbers.
Proof. I f follows immediately from Th. 3 a that the condition
in question is sufficient. To prove that it is necessary we proceed
by induction: from Def. 14 and the formal properties of the
fundamental operations on sets o f sequences, we show that the
family SCo f all the sets which satisfy the condition of the theorem
has all the properties (a)-(S) of Def. 12. Consequently every set
8 = belongs to this family.3
As an example I also give an easy consequence of Defe. 12-14
concerning the family
I f & is a regular fam ily, and X e
T h e o r e m 6. £ # {& ) and
7 e & {& ), then we have
X e $ {& )* X + Y e @ (& ), and
The relativized notions CSf(CP) and SSifC?) are of fundamental
importance for the whole mathematical theory o f definability.
With their help we can, for example, easily define the notions
o f definable sets o f orders higher than 1. In fact, for any given
natural number » , let us agree to denote by ‘^ / n’ the family of all
1 Cf. Hilbert, D ., and Ackermann, W . (30), pp. 83-64.
* Cf. p. 133, note 1.
8 This is analogous to the corresponding proof in metamathematics;
of. note 1, jvbove.
4 Here X is the complement of the set X (with respect to the universe of
discourse F).
HO ON D E F IN A B L E SETS OF R E A L N U M B ER S VI, $ 2

definable sets o f order n o f finite sequences of real numbers,

and by ‘2 nf the corresponding family o f all definable sets o f
order n o f real numbers themselves. Thus the families 2 f x
and 2 t coincide with 2 f and 2 respectively. We now restrict
ourselves to the case » = 2 and construct mathematical defini­
tions o f 2 f 2 and 2 2 in the following way:
For the universe o f discourse we take the family o f all sets
o f real numbers (i.e. o f all sub-sets o f Rl); thus Sf is now the
set o f all finite sequences whose terms are sets o f real numbers,
and the meaning o f the symbol ‘R # ’ (see Def. 3) changes corre­
spondingly. X being any set o f finite sequences o f real numbers,
we replace, in every sequence « e l , each term ek by the set
{»*}, and we denote the set o f all finite sequences thus obtained
by X *. In particular we have (cf. Def. 4);
(a) U*k = Rm .E (sk = {l}y,
8

(b) = R{kfi.E(for some x and y, sk = {x}, st = {y}, and

8
* < y)\
(c) = -Bfcxm-^for some x, y and 2 , sk = {a;}, e, = {y},
8
sm * {*}, and x = y+a).
We also put
(d) Ek( = R{kfi.E {ior some x , 8k = {x} and x e 8t).
8

The sets Uk, M%tl, S%i>m, and Ekl are regarded as primitive
sets of sequences, and the family of all these primitive sets is
denoted by
We now define 2 f % as the family o f all sets X such that
X * e 2 m ) . The definition o f 2 %is obtained from Def. 10
by providing the symbols '2 ' and ‘2 f ’ with subscripts ‘2’ ; we
could also define 2 t as the family o f all sets X such that
{X } 6 2(& r0). To convince ourselves that the definition just
outlined is materially correct we prooeed as in the oase o f 2
(cf. the remarks following Def. 10).
In a similar way we define the notions 2 f n and 2 n as well
as the relativized notions 2 f n(& ) and 2 n{SF) for every natural
number n. We can show that each o f the families 2 n (or 2 f n),
is contained as a part in *^»+l (or 2 f n+1y, moreover, by means
VI, 5 2 ON D E F IN A B L E SETS OF B E A L NTTMBEKS 141

o f a diagonal procedure, we can always construct a set which

belongs to &n+i without belonging to 3>n.
Professor Kuratowski has drawn my attention to a geo­
metrical interpretation o f the concepts introduced here. In
particular, a finite sequence with counter domain {k,
can be interpreted as a point in a space o f a finite number o f
dimensions and o f which X k, X „..., X p are the coordinate
axes. The set Brv, ri is then the entire space; the homo­
geneous sets are the sets contained in one such space. The set
Uk is composed o f a single point with coordinate 1 situated on
the axis X k. The sets Mkt and Sk} mbecome known elementary
figures o f analytic geometry, namely half-planes and planes
in certain special positions. The geometrical interpretation o f
the operations on sets introduced in Defe. 5-8 is not at all
c
difficult. In particular, the operation £ is easily recognized
as that o f projection parallel to the X k axis. In this interpreta­
tion it is not necessary to distinguish between a term and the
sequence composed o f this term; Defs. 10 and 13 become super­
fluous. Ths. 1 and 2 take the form of certain theorems o f analytic
geometry.
From the abstract character o f these considerations we see that
the hypothesis that the axes X k, X ,.... X v are Euclidean straight
lines is no longer necessary. They may, on the contrary, be
abstract spaces o f absolutely any kind; the space R^ t is
then their combinatorial or Cartesian product.
The joint note by Kuratowski and myself, which immediately
follows this article, as well as further investigations by Kura­
towski,1 seem to testify to the great heuristic importance o f
the geometrical interpretation o f the constructions sketched
in this article. In particular, as we shall show in the next
article, the following theorem is a consequence of Def. 12:
I f & is a family of projective sets, so also is
In conclusion we may point out that, instead of finite
sequences, we can operate in all the preceding constructions,
1 See Kuratowski, C. (3d), and also (41), pp. and 243 ff.
142 ON D E F IN A B L E SETS OF R E A L NUM BERS VI, §2

with ordinary infinite sequences (with counter domain identical

with the set o f natural numbers) or, in other words, with pointB
o f space with an infinite number of dimensions (Hilbert space).
We could even use exclusively operations on sets o f sequences
which never lead out o f this space. Moreover the operations o f
complementation, addition, and multiplication would then
coincide with ordinary operations o f the algebra o f sets. Thus,
the modification suggested would certainly simplify the develop­
ment and enhance its intrinsic elegance. On the other hand,
the modified construction would have some defects from the
standpoint of applications, and its heuristic value might pos­
sibly be diminished.
 VII
LO G IC A L O P E R A T IO N S AND
PR O JE C TIV E SETSf

A s e n t e n t ia l function cj>(xli..^xn) o f n real variables will be

called projective if the set E x1?..., xn) } i.e. the set o f points
**...*•
o f n-dimensional space satisfying the function xn), is
projective.2
We propose to show that the five logical operations (see § 1)
when applied to projective functions (a finite number o f times)
always yield projective functions. That is to say every set of points
definable by means of sentential functions is projective*
In the definitions which we usually encounter, especially in
classical mathematics, the functions <f>(xX)...}xn) are projective
functions o f a particularly simple type: the sets E (f>(x1,...,xn)
which correspond to them are curves, surfaces, etc., which are
elementary in the sense o f analysis (such as the plane z = x-\-y,

1 This notation is due to Lebesgue. If, for example, y, z) stands for the
equation z = x -ty , the set E y f z) is the plane given by this equation; if
mjt

4>{x, y) s (x < y),

the set E <j>{xt y) is the upper part of the X V plane determined by the
mi
diagonal x — y.
3 A set of points is called protective, according to Lusin, when it is obtained
from a closed set by applying, any finite number of times, two operations:
(1) orthogonal projection, (2) passage to the complement.
* It is a question of explicit (cf. § 4) and arithmetical (or elementary) defini­
tions in the sense established in V I, Dfs. 9 and 12, pp. 128 and 136. Cf. here
the theorem on p. 141 of the latter article, which is formulated without using
the concept of sentential function.

t BrBX*ioOKAPSiCAii N ote. The main results c f this article were presented

by the authors to the Polish Mathematical Society, Section of Lw6w, in two
talks on 12 June 1930 and 16 December 1930. The summaries of these talks
are given in Kuratowski, C. (40) and Tarski, A. (74), The article itself appeared
under the title *Les operations logiques et les ensembles projectifs ’, Fundamenta
Mathematicae, vol. 17 (1931), pp. 240-8.
 144 LO G IC A L O P E R A T IO N S AN D P R O JE C T IV E SETS VH

the hyperbolic paraboloid z — xy, the surface z = xv, the set

o f positive integers, etc.). All these definitions define projective
sets. It is, on the other hand, remarkable that it is possible to
specify by means o f these elementary functions sets non-
measurable B, functions which cannot be represented analy­
tically, and projective sets o f an arbitrary class.
In order to specify non-projective sets (in an explicit fashion)
it is indispensable to use, apart from real variables, variables o f
a higher type which admit sets o f points as values.1
We thus see that the concept o f the projective set of Lusin
suggests itself in a natural way as soon as we wish to study sets
(or functions) which can be specified effectively.

§ 1. L o g ic a l N o t a t io n . T r a n s f o r m a t io n F o r m u l a s
I f a represents a sentence, then a' represents the negation o f
this sentence; if a and /? represent two sentences, then a+/J
(read ‘ a or /3’ ) represents the logical sum, and a.j8 (read ‘a and
j8’ ) represents the logical product o f these sentences.
I f <f>(x) represents a sentential function, then 2 4>(x) 18 read
X

‘there exists an x such that <f>(x)’, J I <j>(x) is read ‘for every x we

X
have
I f A and B represent two sets (of real numbers), and A x
represent;* a set dependent on a parameter x, then the following
equivalences hold:
(1) ( « 6 i ) 's ( * e A ') ,
(2) (x e i ) + ( * e fi) = ® e (4 + B ),
(3) (x eA ).(x eB ) ==xeA.B,
(4) 2 ( t e A x) = = t e ' Z A x,
X X

(5 ) =
X x

1 The definable non-projective Bets are then of the order n > 2 in the sense
of VI.
* For example, H 2 *tates that for every x there exists a y such
« v
that x < y. The condition for a function f(x) to be bounded is expressed

thU8: 211 !/(*)! < v -

VII, §1 LO G IC A L O P E R A T IO N S AND P R O J E C T IV E SETS 146

The symbols *•’ » *2’ >and ‘J p on the left-hand sides of

these equivalences denote the five logical operations and on the
right-hand sides the corresponding set-theoretical operations
(complementation, set addition, and set multiplication).
Remembering the formula
(6) t e E<f>(x) = m
X
which defines the operation E, we can easily deduce, from the
preceding formulas, the following identities:
(7) E(<f,(x))' =

(8) E(4>(x)+>/i(x)) = E<j>{x)4-Eijj{x),

X X X
(9) E(<f>(x) = E<f>(x) . Eiftipc),
X X X
(10) E 2 <f>(x, y) = 2 y)>
x y y x
(11) E J l ^ x .y ) = JlE<l>(x,y).
X V v X
By way o f example, we prove the formula (10).
By (6), t e E 'Z <f>(x,y) == 2 4>(t>y)- Denoting by A y the set
* t v
Eif>(x,y) we have, according to (6), y) = teE<f>(x,y) = t e A y,
X X
whence, by (4),
2,<f>(t,y) ss 2 < e d y = t e ^ A y = t e j , E<j>(x,y).
V V V V X

§2. L oqico-matsbmatioal D uality

The formulas (7 )-(ll) show that the operation E may be
permuted with each o f the five logical operations, each o f which
changes the logical sense into the mathematical sense (or vice
versa). We thus see that if the sentential function <x(x) is obtain­
able from the functions <f>k (with different variables or
not) by applying the five logical operations, then the set E<x(x)
X
is obtained in the same way from the sets E<f>v ..., E<f>k.x
X X
1 This justifies the use of the same symbols in two senses, logical and
mathematical, without danger of misunderstanding. We thus see that the
passage from the sentential function <x(x) to the set Eot(x) is only another way
x
of reading the expression which defines ol(x ).
146 LO G IC A L O P E R A T IO N ’S AN D P R O JE C T IV E SETS VH, § 2

This also remains true if a is a function o f several variables,

for the foregoing formulas can be generalized immediately by
replacing the variable x by a system o f n variables.1
In the sequel this logioo-mathematical duality will be ex­
pressed in another form by taking into account the following
geometrical interpretation o f the logical operation
(12) The set E^^>(x,y) is the projection of the set E </>{x,y)
x y xv
parallel to the Y axis.
From (6) we see

*o e e
x y
2 <£(*, y) = 1y y ) = 2y Lf (*o> y) s xy y)lJ
and this last sentenoe asserts that there exists a. point having
the abscissa xQwhich belongs to the set E<f>(x, y); in other words
xy
that x0 belongs to the projection of this set onto the X axis.
Examples
1. Let A be the set of all integers. Let d>{x) be the largest
integer which does not exceed x; in symbols
& = ^(*)] S ( y e A ) . ( y < x).{x < y + 1).
Hence, by formula (9),
E[y = <?(*)] — Ely e A) . E{ y < *). E(x < y + 1 ).
xy xy xy xy
We thus see that the graph of the equation y = &(x) is the
common part o f the three sets: (1) the set of horizontal straight
lines with integer ordinates, (2) the half plane y < x, (3) the
half plane x < 1.
2. Evidently we have (x > 0) == £ (x = y2). Consequently
v
the set of non-negative numbers = E(x ^ 0) = E 2 = y2)
X x y
= the projection parallel to the Y axis o f the parabola E(x = t/2).
xy
3. Let us express in symbols the fact that, if a; is a rational
1 The case n = 0 ia included; in this case $(x) becomes a sentence and,
denoting by 0 and 1 its logical value (i.e. ‘ false’ or ‘ true’ ) and employing the
same symbols to denote the empty set and the space of real numbers respec­
tively, we have
(a ie O js O , (® s 1) s 1, E 0 = 0, £ 1 = 1.
(C sc
V II,§ 2 LO G IC A L O P E R A T IO N S AND P R O J E C T IV E SETS 147

number (x e R), it is of the form y/z where y and z are integers

o f which the second is not zero:
R = ^ I 2 ( y e 4 ) . ( « e 4 ) . ( a a = y).(z ^ 0).
x y z

The set E (y e A), ( z e A), (xz = y).(z 0) is, according to

xyz
(9), the common part o f four sets: o f the set formed by the
planes E (y e 41), of the set formed by the planes E ( ze A), o f
xyz xyz
the hyperbolic paraboloid y = xz, o f the whole space minus the
plane z = 0. By projecting this set onto the X Y plane, and then
onto the X axis, we obtain, in conformity with (12), the set R.

REM ARKS

(1) In all the foregoing the hypothesis that the variables x ,

y> z,... range over the set o f real numbers is in no way essential.
We may suppose that x ranges over an arbitrary set X , y over
a set Y , distinct or not from X , z over a set Z, etc. The
Euclidean plane X Y should then, be replaced by the combina­
torial or Cartesian product of X and Y, i e. by the set o f pairs
(x, y)9where x e X and y e Y (m an analogous way, the combina­
torial product of X , Y , Z is the set of points (x, y,z), where
x e X, y e Y . z e Z). The extension of the concept of projection
to the general case is obvious.
(2) According to (12) a continuous geometrical operation corre­
sponds to the logical operation ]£. Numerous topological applica­
tions of the logical calculus are connected with this fact (see
Kuratowski, C. (39)).
(3) Regarding the operation JJ, it is easily seen that the set
E Y [ </>{%,y) is composed of all the x0 such that the straight line
x y
x x0 is completely contained in the set E<f>(x,y). This geo-
xy
metrical operation, not being in general continuous, is often
more advantageously defined by means of the operation 2 with
the help o f the generalized De Morgan’s law:

as) n#*) = f2w*))f.

* L X J
148 LOGICAL OPERATIONS AND PROJECTIVE SETS VII, §2

which is a generalization o f the well-known formula

a.jS = ( e t + P ) ’.

We thus see that the five logical operations considered can

be reduced to three: negation, summation, and the operation 2*
(4) The relation o f implication, symbolically represented by
a->/3 also reduces to the same operations for we have

§ 3. P r o je c tive S e ts and Se n t e n t ia l F un ctio n s

Projective sets possess the following fundamental properties:1
(а) the complement o f a projective set is projective;
(б) the sum (and also the product) o f two projeotive sets is
a projective set;
(c) if E<f>(x) is projective, so also is E<f>{x)-,z
X XV

(d) the projection o f a projective set is a projective set.

From this we deduoe, by means o f the formulas (7)—(9), (12),
the principal theorem o f this artiole:
The five logical operations, when applied to projective sentential
functions, always lead to projective sentential functions.
First, by remark (3) we can reduce the five operations to three:
negation, summation, and the operation 2*
Now, (1) I f the function <f>(x) is projective, the set E<f>(x) is
X

likewise projective; for by (a) [E<f>(x)^' is projective, from which,

by (7), E(<f>(x)Y is likewise projective, that is to say the function
X

(<f>(x))' is projective.
(2) Let^(z1,...,*„) = ^(xlh,...,x kl) + w here^,...,k},
h>—> lm ^ » ;3 if the functions <f> and x 9X6 assumed to be
1 See, for example, N. Lusin, L egons sur les ensem bles analypiques (Paris,
1980), pp. 270-7.
* The set E<j>(x) is obtained by passing a vortical straight line through each
wv
point of the set E
»
* e.g. 4>(x, y, z) e s xf,(x, y)+x(v, *)•
V n ,§ 3 LOGICAL OPERATIONS AND PROJECTIVE SETS 149

projective, we conclude from (c) that the sets E *p(xki, . . . , x k.)

and E x(xh> 6X6 projective and since, by (8),

*x -~ X n

E <f>(xv .. .,x n) ~ E >P(xkt,. ..,x kj) + E x(xu -> X t J ,

X f.X n • X t .. .X n X i„ . X n

we conclude from (b) that is a projective function.

(3) Assuming that (f>(xv ...t x n) is a projective function, we have
to prove that 2 is also projective.1 This is evident for
xt
the case where h > n , for in this case 2
Xk
Suppose then that k ^ n. Now according to (12) the set
E 2 %n)18 an orthogonal projection of the set
E <f>(xv .. .,x n)-, and, by (d), this is a projective set. q.e.d.

E X A M P L E S AN D NOTES
Let <f>(x, y) be a projective sentential function o f two variables.
The set M = E <j>(x, y ) is then a plane projective set. Let Q be
xy
the set formed by the union o f all the straight lines contained
in Jf. In symbols
( x ,y ) e Q == {{a x + b y = c).(a 2+ 6 a # 0).

•XI [(««+&> = c) <f>(u,v)]\.a

uv '
The set E (ax-\-by = c) is a closed set (in the space o f
abcxy
5 dimensions), and the set E (oa+ 6 2 = 0) is composed o f a
db
single point (0,0), thus it results directly from our theorem that
Q is a projective set.3
We thus see that the operation o f deriving Q from M does
not take us out o f the domain o f projective sets. This is the
1 I f k = n = I, this is a sentence. Now, each sentence is a projective
sentential function, since the whole space as well as the empty space are
projective sets; cf. p. 146, footnote.
% Q may also be defined as follows:
(*, y) 6 Q sb 2) I I {[(«— t = (v-y)coB (] ->■<j>(u, v)}.
( M>
* Which may, moreover, be non-measurable B when M is open. See
Nikodym et Sierpiiiski, Fundamenta Mathematu&e, vol. 7,.p. 259.
150 LOGICAL O PERATIO N S AND PR O JECTIVE SETS V II,§3

case for the greater part o f the operations which are considered
in mathematics. In order to convince oneself of this it suffices
to write their definitions in logical symbols.
Regarding the formulation o f the theorem o f § 3, it is to be
noted that if, instead o f supposing that the given sentential
functions are projective, we make more restricted hypotheses,
the function which is obtained may also be characterized in a
more specific manner. To give an example let us call every
sentential function linear if it is obtained by addition and multi­
plication from sentential functions of the form f(x v ...,x n) = 0
or > 0 , where / is a polynomial o f degree 1 with integral co­
efficients . Then the property of being a linear sentential function
proves to be invariant under the five logical operations coil-
sidered. In particular, the sets o f real numbers defined by means
o f linear functions prove to coincide with finite sums of intervals
(closed or open) with rational end points.1
On the other hand important examples o f classes of sentential
functions are known which are invariant, not under all, but
under some o f the logical operations discussed. Such is, for
instance, the class o f all functions corresponding to closed sets
or to Borelian sets, etc.; see Kuratowski, C. (41), pp. 168-9
and 246.

§ 4 . I m p l ic it D e f in it io n s
Hitherto we have supposed that the 'unknown’ function a is
obtained from a system o f given functions by perform­
ing upon them logical operations; the type of this definition is

These are definitions properly so called, or explicit definitions.

I f we carry out the five logical operations as well as a change
o f variables2 on the system <f>n augmented by the unknown
1 W e obtain another example of this kind if we replace, in the example
just given, linear sentential functions (corresponding to polynomials of the
first degree) by algebraic sentential functions (corresponding to polynomials
of an arbitrary degree) and also rational numbers by algebraic numbers. Both
examples have been discussed (although in different terms) in V I, see in
particular the first footnote on p. 133.
3 For example, if we pass from 4>(xt y) to <j>(xt z).
V II, § 4 L O G IC A L O P E R A T IO N S A N D P R O J E C T IV E SE TS 15)

function a, and if the equation

admits one and only one solution (for a), we can say that this
equation defines a implicitly.
A very common type of implicit definition is represented by
definitions by induction (finite or transfinite).
By means o f implicit definitions we are able to pass out of
the domain o f projective sets. They lead to a new class o f sets,
implicitly protective sets which it would be interesting to study.
 VIII
T H E C O N C E P T OF T R U T H IN
F O R M A L IZE D L A N G U A G E S!
iNTBODTTOTIOir
T he present' article is almost wholly devoted to a single prob­
lem— the definition of truth. Its task is to construct—with
reference to a given language—a materially adequate and
formallv correct definition' of the term Hrue sentence*. This prob­
lem, which belongs to the classical questions of philosophy,
raises considerable difficulties. For although the meaning o f
the term 'true sentence’ in colloquial language seems to be
quite clear and intelligible, all attempts to define this mean­
ing more precisely have hitherto been fruitless, and many in­
vestigations in which this term has been used and which started
with apparently evident premisses have often led to paradoxes
and antinomies (for which, however, a more or less satisfactory
solution has been found). The concept o f truth shares in this
respect the fate of other analogous concepts in the domain o f
the semantics o f language.
The question how a certain concept is to be defined is cor­
rectly formulated only if a list is given o f the terms by means
o f which the required definition is to be constructed. I f the
definition is to fulfil its proper task, the sense o f the termB in
this list must admit o f no doubt. The question thus naturally
arises: What terms are we to use in constructing the definition
o f truth ? In the course o f these investigations I shall not neglect
to clarify this question. In this construction I shall not make use
t B ib lio graphical ISTote. This article was presented (by J. Lukasiewicz
to the Warsaw Scientific Society on 21 March 1931. For reasons beyond
the author’s control, publication was delayed by two years. The arada
appeared in Polish in Tarski, A. (73). A German translation was published
under the title ‘ Der Wahrheitsbegrif? in den formalisierfcen Sprachen*, in
Studia Pkilosophica, vol. 1 (1930) (reprint dated 1935), pp. 201-405; it is
provided with a Postscript in which some views which had been stated in the
Polish original underwent a rather essential revision and modification. The
present English version is based upon the German translation. For earlier
publications and historical information concerning the results of this work see
p. 154, footnote, p. 247, footnote, and the concluding remarks of the Postscript.
vm CONCEPT OF TRUTH IN FORMALIZED LANGUAGES 153

o f any semantical concept if I am not able previously to reduce

it to other concepts.
A thorough analysis o f the meaning current in everyday life
o f the term 'true’ is not intended here. Every reader possesses
in greater or less degree an intuitive knowledge of the concept
o f truth and he can find detailed discussions on it in works on
the theory of knowledge. I would only mention that throughout
this work I shall be concerned exclusively with grasping the
intentions which are contained in the so-called classical con­
ception o f truth (‘true— corresponding with reality') in contrast,
for example, with the utilitarian conception ( ‘true—in a certain
respect useful').1
The extension of the concept to be defined depends in an
essential way on the particular language under consideration.
The same expression can, in one language, be a true statement,
in another a false one or a meaningless expression. There will
be no question at all here o f giving a single general definition o f
the term. The problem which interests us will be split into a
series o f separate problems each relating to a single language.
In § 1 colloquial language is the object o f our investigations.
The results are entirely negative. With respect to this language
not only does the definition of truth seem to be impossible, but
even the consistent use o f this concept in conformity with the
laws o f logic.
In the further course o f this disoussion I shall consider ex­
clusively the scientifically constructed languages known at the
present day, i.e. the formalized languages o f the deductive
sciences. Their characteristics will be described at the beginning
o f § 2 . It will be found that, from the standpoint o f the present
problem, these languages fall into two groups, the division being
based on the greater or less stock o f grammatical forms in a
particular language. In connexion with the ‘poorer’ languages
the problem o f the definition o f truth has a positive solution:
there is a uniform method for the construction o f the required

1 Cf. KotarbiAski, T. (37), p. 126 (in writing the present article I have
Hpeatedly consulted this book and in many points adhered to the terminology
there suggested).
154 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES VIII

definition in the case o f each o f these languages. In §§ 2 and 3

I shall cany out this construction for a concrete language in full
and in this way facilitate the general description o f the above
method which is sketched in § 4. In connexion with the ‘richer’
languages, however, the solution of our problem will be negative,
as will follow from the considerations o f § 5. For the languages
o f this group we shall never be able to construct a correct
definition o f the notion o f truth.f Nevertheless, everything
points to the possibility even in these cases—in contrast to the
language o f everyday life—o f introducing a consistent and
correct use o f this concept by considering it as a primitive notion
o f a special science, namely o f the theory of truth, and its funda­
mental properties are made precise through axiomatization.
The investigation o f formalized languages naturally demands
a knowledge o f the principles o f modem formal logic. For the
construction o f the definition o f truth certain purely mathemati­
cal concepts and methods are necessary, although in a modest
degree. I should be happy if this work were to convince the
reader that these methods already are necessary tools even for
the investigation o f purely philosophical problems.1

§ 1. T he Concept of T rue Sentence in E v e r y d a y or

C olloquial L anguage
For the purpose of introducing the reader to our subject, a
consideration—if only a fleeting one-—o f the problem o f defining
truth in colloquial language seems desirable. I wish especially
1 This was communicated to the Society of Sciences in Warsaw by J. Luka­
siewicz on 21 March 1931. The results it contains date for the most part from
1929. I have reported on this, among other things, in two lectures which
I gave under the title ‘ On the Concept of Truth in relation to formalized
deductive systems * at the logical section of the Philosophical Society in
Warsaw (8 October 1930) and at the Polish Philosophical Society in Lw6w
(15 December 1930), aresum6 of which appeared in Tarski, A. (73), For reasons
unconnected with me the printing of this work was much delayed. This enabled
me to supplement the text with some rather important results (cf. p. 247, foot­
note). In the meantime a r£sum6 of the chief results has appeared in Tarski,
A . (76).

t Regarding this statement compare the Postscript.

vm , §I OONGErT OF TRUTH IN FORMALIZED LANGUAGES 155

to emphasize the various difficulties which the attempts to solve

this problem have encountered,1
Amongst the manifold efforts which the construction o f a
correct definition of truth for the sentences of colloquial language
has called forth, perhaps the most natural is the search for a
semantical definition. By this I mean a definition which we can
express in the following words:
( 1 ) a true sentence is one which says that the state of affairs is so
and so, and the state of affairs indeed is so and so.2
From the point of view of formal correctness, clarity, and
freedom from ambiguity o f the expressions occurring in it, the
above formulation obviously leaves much to be desired. Never­
theless its intuitive meaning and general intention seem to be
quite clear and intelligible. To make this intention more definite,
and to give it a correct form, is precisely the task of a semantical
definition.
As a starting'point certain sentences of a special kind present
themselves which could serve as partial definitions o f the truth
of a sentence or more correctly as explanations of various con­
crete turns o f speech o f the type (x is a trae sentence’. The
general scheme of this kind of sentence can be depicted in the
following way:
(2 ) x is a true sentence if and only if p.
In order to obtain concrete definitions we substitute in the
x The considerations which I shall put forward in this connexion are, for
the most part, not the result of my own studies. Views are expressed in them
which have been developed by S. Le^uiewski in his lectures at the University
of Warsaw (from the year 1919/20 onwards), in scientific discussions and in
private conversations; this applies, in particular, to almost everything which
I shall say about expressions in quotation marks and the semantical antinomies.
It remains perhaps to add that this fact does not in the least involve Le&niewski
in the responsibility for the sketchy and perhaps not quite precise form in
which the following remarks are presented.
* Very similar formulations are found in Kotarbihski, T. (37), pp. 127 and
136, where they are treated as commentaries which explain approximately
the classical view of truth.
Of course these formulations are not essentially new; compare, for example^
the well-known words of Aristotle.* 'To soy of what is that it is not, or of what
is not that it Lb, is false, while :*> say of what is that it is, or of what is not
that it is not, is true.’ (Aristotle, MeUsphysiea, I\ 7, 27; Works, vol. 8,
English translation by W . D. Rose, Oxford, 1908.)
im CONCEPT OF TRUTH IN FORMALIZED LANGUAGES V III,} 1

place o f the symbol *p* in this scheme any sentence, and in the
place of V any individual name o f this sentence.
I f we are given a name for a sentence, we can construct an
explanation o f type (2 ) for it, provided only that we are able
to write down the sentence denoted by this name. The most
important and common names for which the above condition is
satisfied are the so-called quotation-mark names. We denote by
this term every name o f a sentence (or of any other, even mean­
ingless, expression) which consists o f quotation marks, left- and
right-hand, and the expression which Jies between them, and
which (expression) is the object denoted by the name in question.
As an example o f such a name o f a sentence the name 8‘it is
snowing” will serve. In this case the corresponding explanation
o f type (2) is as follows:
( 3 ) ‘it is snowing* is a true sentence if and only if it is snowing.1
Another category o f names o f sentences for which we can
construct analogous explanations is provided by the so-called
structural-descriptive names. We shall apply this term to names
which describe the words which compose the expiession denoted
1 Statements (sentences) are always treated here as a particular kind of
expression, and thus as linguistic entities. Nevertheless, when the terms
‘expression*, ‘statement*, efco., are interpreted as names of concrete series of
printed signs, various formulations which occur in this work do not appear
to bo quite correct, and give the appearance o f a widespread error which
consists in identifying expressions of like shape. This applies especially to the
sentence (3), since with the above interpretation quotation-mark names must
be regarded as general (and not individual) names, which denote not only
the series of signs in the quotation marks but also every series of signs of like
shape. In order to avoid both objections of this kind and also the introduction
of superfluous complications into the discussion, which would be connected
among other things with the necessity of using the concept of likeness of
shape, it is convenient to stipulate that terms like ‘ word’, ‘ expression\ ‘sen­
tence*, etc., do not denote concrete series of signs but whole classes of such
series which are of like shape with the series given; only in this sense shall
we regard quotation-mark names as individual names of expressions. Cf.
Whitehead, A. N ., and Russell, B. A. W. (90), vol. 1. pp. 661- 6 and— for other
interpretations of the term ‘ sentence’— KotarbiAski, T. (37), pp. 123-5.
I take this opportunity of m e n t i o n i n g that I use the words ‘ name’ and
‘ denote* (like the words ‘ object’, ‘ dates’ , ‘ sr-l&tic?''} not in on?, but in many
distinct senses, because I apply there both to d dect;-. in the narrower tense
(i.e. to individuals) and also to all kinds of dosses and relations, etc. From
the standpoint of the theory of types expounded in Whitehead.. A . N ., and
Russell, B. A . W . (90) (vol. 1, pp. 139-68) these expressions are to be regarded
as systematically ambiguous.
vm . § 1 C O N C E P T O F T R U T H IN - F O R M A L IZ E D L A N G U A G E S 157

b y the name, as well as the signs of whioh each single word is

composed and the order in which these signs and words follow
one another. Such names oan be formulated without the help
of quotation marks. For this purpose we must have, in the
language we are using (in this oase colloquial language),
individual names of some sort, but not quotation-mark
names, for all letters and all other signs of which the words and
expressions of the language are composed. For example we
could use ‘A ’ , ‘E ’, ‘E f’, ‘ Jay ’, ‘P e’ as names of the letters V ,
V, ‘p ’. I t is dear that we can corrdate a structural-
descriptive name with every quotation-mark name, one which
is free from quotation marks and possesses the same extension
(i.e. denotes the same expression) and vice versa. For example,
corresponding to the name “ ‘snow’ ” we have the name ‘a
word which consists of the four letters: E s , E n , O , Double-U
following one another’. It is thus clear that we can construct
partial definitions of the type (2) for structual-descriptive names
of sentences. This is illustrated by the following example:

(4) an expression consisting of three words of which the first is

composed of the two letters I and Te (in that order) the second
of the two letters I and Es (in that order) and the third of the
seven letters Es, En, 0 , Double~U, I, En, and Oe (in that
order), is a true sentence if and only if it is snowing.
Sentences which are analogous to (3) and (4) seem to be clear
and completely in accordance with the meaning of the word
‘true’ whioh was expressed in the formulation (1). In regard
to the clarity o f their content and the correctness o f their form
they arouse, in general, no doubt (assuming o f course that no
suoh doubts are involved in the sentences which we substitute
for the symbol (p ' in (2)).
But a certain reservation is nonetheless necessary here. Situa­
tions are known in which assertions of just this type, in com­
bination with certain other not less intuitively clear premisses,
lead to obvious contradictions, for example the antinomy of
the liar. W e shall give an extremely simple formulation of this
antinomy whioh is due to J. Lukasiewicz.
158 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES V III, § 1

For the sake of greater perspicuity we shall use the symbol

V as a typographical abbreviation o f the expression ‘the sen­
tence printed on this page, line 5 from the top’ . Consider now
the following sentence:
c is not a true sentence.
Having regard to the meaning of the symbol V , we can
establish empirically:
(a) (c is not a true sentence ’ is identical with c.
For the quotation-mark name o f the sentence c (or for any
other o f its names) we set up an explanation of type (2):
(f3) lc is not a true sentence’ is a true sentence if and only if
c is not a true sentence.
The premisses (a) and (ft) together at once give a contra­
diction:
c is a true sentence if a?id only if c is not a true sentence.
The source o f this contradiction is easily revealed: in order
to construct the assertion (/?) we have substituted for the
symbol in the scheme (2) an expression which itself contains
the term 'true sentence’ (whence the assertion so obtained—in
contrast to (3) or (4)—-can no longer serve as a partial definition
o f truth). Nevertheless no rational ground can be given why
such substitutions should be forbidden in principle.
I shall restrict myself here to the formulation o f the above
antinomy and will postpone drawing the necessary conse­
quences of this fact till later. Leaving this difficulty on one
side I shall next try to construct a definition of true sentence
by generalizing explanations of type (3). At first sight this task
may seem quite easy— especially for anyone who has to some
extent mastered the technique of modern mathematical logic.
It might be thought that- all we need do is to substitute in (3)
any sentential variable (i.e. a symbol for which any sentence
can be substituted) in place of the expression ‘it is snowing*
which occurs there twice, and then to establish that the resulting
formula holds for every value o f the variable, and thus without
v m , § 1 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES 159

further difficulty reach a sentence which includes all assertions

o f type (3) as special cases:
(5) for all p, ‘p" is a true sentence if and only if p.
But the above sentence could not serve as a general definition
o f the expression cx is a true sentence’ because the totality o f
possible substitutions for the symbol V is here restricted to
quotation-mark names. In order to remove this restriction we
must have recourse to the well-known fact that to every true
sentence (and generally speaking to every sentence) there corre­
sponds a quotation-mark name which denotes just that sen­
tence.1 With this fact in mind we could try to generalize the
formulation (5), for example, in the following way:
(6) for all x, x is a true sentence if and only if, for a certain p,
x is identical with *p' and p.
At first sight we should perhaps be inclined to regard (6) as
a correct semantical definition o f ‘true sentence’ , which realizes
in a precise way the intention o f the formulation (1) and there­
fore to accept it as a satisfactory solution of our problem.
Nevertheless the matter is not quite so simple. As soon as we
begin to analyse the significance o f the quotation-mark names
which occur in (5) and (6) we encounter a series o f difficulties
and dangers.
Quotation-mark names may be treated like single words o f
a language, and thus like syntactically simple expressions. The
single constituents o f these names—the quotation marks and
the expressions standing between them—fulfil the same func­
tion as the letters and complexes o f successive letters in single
words. Hence they can possess no independent meaning.
Every quotation-mark name is then a constant individual
name o f a definite expression (the expression enclosed by the
quotation marks) and in fact a name of the same nature as the
proper name o f a man. For example, the name “p ” denotes one
1 For example, this fact could be formulated in the following way:
(S') fo r dU x , i f x is a true sentence, then— for a certain p — x is identical with
V 9
from the premisses (5) and (5') the sentenoe (6) given below can be derived
as a conclusion.
160 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES YW L, § 1

o f the letters of the alphabet. With this interpretation, which

seems to be the most natural one and completely in accordance
with the customary way o f using quotation marks, partial
definitions o f the type (3) c anno t be used for any significant
generalizations. In no case can the sentences (5) or (6) be
accepted as suoh a generalization. In applying the rule called
the rule o f substitution to (5) we are not justified in substituting
anything at all for the letter ‘p ’ which occurs as a component
part o f a quotation-mark name (just as we are not permitted
to substitute anything for the letter T in the word ‘true’).
Consequently we obtain as conclusion not (5) but the following
sentence: ‘p ’ is a true sentence if and only if it is snowing. We
see at once from this that the sentences (6) and (6) are not
formulations o f the thought we wish to express and that they
are in fact obviously senseless. Moreover, the sentence (5) leads
at once to a contradiction, for we can obtain from it just as
easily in addition to the above given consequence, the contra­
dictory consequence: ‘p ’ is a true sentence if and only if it is
not snowing. Sentence (6) alone leads to no contradiction, but
the obviously senseless conclusion follows from it that the letter
‘p ’ is the only true sentence.
To give greater clarity to the above considerations it may
be pointed out that with our conception o f quotation-mark
names they oan be eliminated and replaced everywhere by, for
example, the corresponding structural-descriptive names. If,
nevertheless, we consider explanations of type (2) constructed
by the use o f such names (as was done, for example, in (4) above),
then we see no way o f generalizing these explanations. And if in
(5) or (6) we replace the quotation-mark name by the structural-
descriptive name ‘p c’ (or 'the ward which consists of the single
letter P e') we see at once the absurdity o f the resulting formula­
tion.
In order to resoue the sense o f sentences (5) and (6) we
must seek quite a different interpretation o f the quotation-
mark names. We must treat these names as syntactically com­
posite expressions, o f which both the quotation marks and the
expressions within them are parts. Not all quotation-mark
v m , 5 1 CONCEPT OF TROTH IN FORMALIZED LANGUAGES 161

expressions will be constant names in that case. The expression

“ *p* ” occurring in (5) and (6), for example, must be regarded
as a function, the argument o f which is a sentential variable
and the values o f which are constant quotation-mark names o f
sentences. We shall call such functions quotation-functions.
The quotation marks then become independent words belonging
to the domain o f semantics, approximating in their meaning to
the word 'name', and from the syntactical point o f view, they
play the part o f functors.1 But then new complications arise.
The sense o f the quotation-function and of the quotation marks
themselves is not sufficiently clear. In any case such functors
are not extensional; there is no doubt that the sentence “for
all p and q, if p if and only if q, then ‘p ’ is identical with (q” y
is in palpable contradiction to the customary way of using
quotation marks. For this reason alone definition (6) would
be unacceptable to anyone who wishes consistently to avoid
intensional functors and is even of the opinion that a deeper
analysis shows it to be impossible to give any precise meaning
to such functors.2 Moreover, the use o f the quotation functor
exposes us to the danger o f becoming involved in various
semantical antinomies, such as the antinomy o f the liar. This
will be so even if—taking every care—we make use only o f
those properties o f quotation-functions which seem almost
evident. In contrast to that conception o f the antimony of the
liar which has been given above, we can formulate it without
using the expression 'true sentence’ at all, by introducing the
1 W© call such words as ‘ reads* in the expression lx reads’ functors (this is
a sentence-forming functor with one individual name as argument); also ‘ sees’
in the expression sees y* (a sentence-forming functor with two name argu­
ments), and ‘ father’ in the expression ‘ the father of a?’ (a name-forming functor
with one name argument), as well as ‘ or’ in the expression lp or g’ (a sentence-
forming functor with two sentence arguments) ; quotation marks provide an
example of a name-forming functor with one sentence argument. The term
‘ functor’ we owe to T. Kotarbidski, the terms ‘ sentence-forming functor* and
‘ name-forming functor’ to K . Ajdukiewicz; cf. Ajdukiewicz, K . (3).
1 I shall not discuss the difficult problem of extensionality in more detail
here; cf. Carnap, R . (8) where the literature of the problem is given, and
especially Whitehead, A . N ., and Russell, B. A. W . (90), vol. 1, pp. 669-66.
It should be noted that usually the terms ‘ extensional’ and ‘ intensional’ are
applied to sentence-forming functors, whilst in the text they are applied to
quotation marks and thus to name-forming functors.
162 CONCEIT OF TRUTH IN FORMALIZED LANGUAGES VHI, § 1

quotation-functions with variable arguments. We shall give a

sketch o f this formulation.
Let the symbol ‘c ’ be a typographical abbreviation o f the
expression ‘the sentence printed on this page, line 6 from the top \
W e consider the following statement:
for all p, if a is identical with the sentence ‘p\ then not p
(if we accept (6) as a definition o f truth, then the above state­
ment asserts that c is not a true sentence).
We establish empirically:
(a) the sentence ‘for aid p, i f c is identical with the sentence
‘p\ then not p ’ is identical with c.
In addition we make only a single supplementary assumption
which concerns the quotation-function and seems to raise no
doubts:
(/J) for all p and g, if the sentence ‘p 3 is identical with the
sentence ‘q\ then p i f and only if q.
By means o f elementary logical laws we easily derive a contra­
diction from the premisses (a) and (fi).
I should like to draw attention, in passing, to other dangers
to whioh the consistent use o f the above interpretation o f quota­
tion marks exposes us, namely to the ambiguity o f certain
expressions (for example, the quotation-expression which occurs
in (6) and (6) must be regarded in certain situations as a func­
tion with variable argument, whereas in others it is a constant
name which denotes a letter o f the alphabet). Further, I would
point out the necessity o f admitting certain linguistic construc­
tions whose agreement with the fundamental laws of syntax is
at least doubtful, e.g. meaningful expressions whioh contain
meaningless expressions as syntactical parts (every quotation-
name o f a meaningless expression will serve as an example).
For all these reasons the correctness o f definition (6), even with
the new interpretation o f quotation marks, seems to be ex­
tremely doubtful.
Our discussions so fax entitle us in any oase to say that the
attempt to construct a correct semantical definition of the expression
*true sentence meets with very reed difficulties. We know of no
vxn, § 1 CONCEPT OP TROTH IN FORMALIZED LANGUAGES 163

general method which would permit us to define the meaning

o f an arbitrary concrete expression o f the type ‘x is a true
sentence’, where in the place o f V we have a name o f some
sentence. The method illustrated by the examples (3) and (4)
fails us in those situations in which we cannot indicate for a
given name o f a sentence, the sentence denoted by this name
(as an example o f such a name ‘the first sentence which will be
printed in the year 2000’ will serve). But if in suoh a case we
seek refuge in the construction used in the formulation o f
definition (6), then we should lay ourselves open to all the
complications which have been described above.
In the face o f these facts we are driven to seek other methods
o f solving our problem. I will draw attention here to only one
such attempt, namely the attempt to construct a structural
definition. The general scheme o f this definition would be some­
what as follows: a true sentence is a sentence which possesses such
and such structural properties (i.e. properties concerning the form
and arrangement in sequence o f the single parts o f the expres­
sion) or which can be obtained from such and such structurally
described expressions by means of such and such structural trans­
formations. As a starting-point we can press into service many
laws from formal logic which enable us to infer the truth or
falsehood of sentences from certain of their structural properties;
or from the truth or falsehood of certain sentences to infer
analogous properties o f other sentences which can be obtained
from the former by means o f various structural transforma­
tions. Here are some trivial examples o f suoh laws: every ex­
pression consisting of four parts of which the first is the word Hf ’,
the third is the word ‘then’ , and the second and fourth are the same
sentence, is a true sentence; if a true sentence consists of four parts,
of which the first is the word *i f ’, the second a true sentence, the
third the word ‘then’ , then the fourth part is a true sentence. Such
laws (especially those o f the second type) are very important.
With their help every fragmentary definition o f truth, the ex­
tension o f which embraces an arbitrary class o f sentences, can
be extended to all composite sentences which can be built up
from sentences o f the given class by combining them by means
164 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES v m , § 1

o f suoh expressions as ‘i f . . . then’, ‘if and only if’ , ‘or’, ‘and’,

‘not’ , in short, by means o f expressions belonging to the sen­
tential calculus (or theory o f deduction). This leads to the idea
o f setting up sufficiently numerous, powerful, and general laws
for every sentence to fall under one o f them. In this way
we should reach a general structural definition o f a true
sentence. Yet this way also seems to be almost hopeless, at
least as far as natural language is concerned. For this lan­
guage is not something finished, closed, or bounded by clear
limits. It is not laid down what words can be added to this
language and thus in a certain sense already belong to it
potentially. We axe not able to specify structurally those
expressions o f the language which we call sentences, still less
can we distinguish among them the true ones. The attempt to set
up a structural definition of the term ‘true sentence’—applicable
to colloquial language is confronted vnth insuperable difficulties.
The breakdown o f all previous attempts leads us to suppose
that there is no satisfactory way o f solving our problem. Im­
portant arguments o f a general nature can in fact be invoked
in support o f this supposition as I shall now briefly indicate.
A characteristic feature o f colloquial language (in contrast to
various scientific languages) is its universality. It would not be
in harmony with the spirit o f this language if in some other
language a word occurred which could not be translated into it;
it could be olaimed that ‘if we can speak meaningfully about
anything at all, we can also speak about it in colloquial language’ .
I f we are to maintain this universality of everyday language in
connexion with semantical investigations, we must, to be con­
sistent, admit into the language, in addition to its sentences and
other expressions, also the names o f these sentences and ex­
pressions, and sentences containing these names, as well as such
semantic expressions as ‘true sentence’ , ‘name’, denote’ , etc.
But it is presumably just this universality o f everyday language
which is the primary source of all semantical antinomies, like the
antinomies o f the liar or of heterologioal words. These anti­
nomies seem to provide a proof that every language which is
universal in the above sense, and for which the normal laws of
VIII, § 1 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES 165

logic hold, must be inconsistent. This applies especially to the

formulation o f the antinomy o f the liar which I have given on
pages 157 and 158, and which contains no quotation-function
with variable argument. I f we analyse this antinomy in the
above formulation we reach the conviction that no consistent
language can exist for which the usual laws o f logic hold and
which at the same time satisfies the following conditions: (I) for
any sentence which occurs in the language a definite name o f
this sentence also belongs to the language; (II) every expression
formed from (2) by replacing the symbol ‘p* by any sentence
o f the language and the symbol V by a name of this sentence
is to be regarded as a true sentence of this language; (III) in
the language in question an empirically established premiss
having the same meaning as (a ) can be formulated and accepted
as a true sentence.1
I f these observations are correct, then the very possibility of a
consistent use of the expi'ession 'true sentence’ which is in harmony
with the laws of logic and the spirit of everyday language seems to
be very questionable, and consequently the same doubt attaches to
the possibility of construcivng a correct definition of this expression.
§ 2. F o r m a l iz e d L a n g u a g e s , e s p e c ia l l y t h e L a n g u a g e o f
the Ca l c u l u s o f Cl a s s e s
For the reasons given in the preceding section I now abandon
the attempt to solve our problem for the language of everyday
life and restrict myself henceforth entirely to formalized lan­
guages? These can be roughly characterized as artificially con-
1 The antinomy of heterologioal words (which I shall not describe here—
cf. Grelling, K ., and Nelson, L. (24), p. 307) is simpler than the antinomy of
the liar in so far as no empirical premiss analogous to (a) appears in its formula­
tion; thus it leads to the correspondingly stronger consequence: there can
be no consistent language which contains the ordinary laws of logic and
satisfies two conditions which are analogous to (I) and (II), but differ from
them in that they treat not of sentences but of names, and not of the truth
of sentences but of the relation of denoting. In this connexion compare the
discussion in § 6 of the present article— the beginning of the proof of Th. 1, and
in particular p. 248, footnote 2.
1 The results obtained for formalized language also have a certain validity
for colloquial language, and this is owing to its universality: if we translate
into colloquial language any definition of a true sentence which has been con­
structed for some formalized language, we obtain a fragmentary definition of
truth which embraces a wider or narrower category of sentences.
166 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES VUL, $ 2

struoted languages in which the sense o f every expression is

unambiguously determined by its form. Without attempting
a completely exhaustive and precise description, which is a
matter o f considerable difficulty, I shall draw attention here
to some essential properties which all the formalized languages
possess: (a) for each o f these languages a list or description
is given in structural terms o f all the signs w ith which the ex­
pressions o f tiie language are form ed ; (,8) among all possible
expressions which can be formed with these signs those called
sentences are distinguished by means o f purely structural pro­
perties. Now formalized languages have hitherto been con­
structed exclusively for the purpose o f studying deductive
sciences formalized on the basis of such languages. The language
and the science grow together to a single whole, so that we speak
o f the language o f a particular formalized deductive science,
instead o f this or that formalized language. For this reason
further characteristic properties o f formalized languages appear
in connexion with the way in which deductive sciences are built
up: (y) a list, or structural description, is given o f the sentences
called axiom s or p rim itiv e statements; (8) in special rules, called
rules of infercuts, certain operations o f a structural kind are em­
bodied which permit the transformation o f sentences into other
sentences; the sentences which can be obtained from given sen­
tences by one or more applications o f these operations are called
consequences o f the given sentences. In particular the conse­
quences o f the axioms are called provable or asserted sentences.1
It remains iierkaps to add that we are not interested here
in ’formal* languages and sciences in one special sense of the
word ‘formal*, namely sciences to the signs and expressions of
which no material sense is attached. For such sciences the prob­
lem here discussed has no relevance, it is not even meaningful.
1 The formalization of a science usually admits of the possibility of intro­
ducing new signs into that science which were not explicitly given at the out­
set. These signs— called defined signs (in contrast to the primitive signs)—
appear in the science in the first instance in expressions of a special structure
called definitions* which are constructed in accordance with special rules—
the rules of definition. Definitions are sometimes regarded as asserted sen­
tences of the science. This feature of the formalization of languages will not be
considered in the sequel.
V ffl, § 2 CONCEPT OF TROTH IN FORMALIZED LANGUAGES 167

We shall always ascribe quite concrete and, for us, intelligible

meanings to the signs which occur in the languages we shall
consider.1 The expressions whioh we call sentences still re­
main sentences after the signs which occur in them have been
translated into colloquial language. The sentences which are
distinguished as axioms seem to us to be materially true, and in
choosing rules o f inference we are always guided by the prin­
ciple that when such rules are applied to true sentences the
sentences obtained by their use should also be true,2
In contrast to natural languages, the formalized languages
do not have the universality which was discussed at the end o f
the preceding section. In particular, most o f these languages
possess no terms belonging to the theory o f language, i.e. no
expressions which denote signs and expressions o f the same or
another language or which describe the structural connexions
between them (such expressions I call—for lack o f a better
term— structural-descriptive). For this reason, when we investi­
gate the language o f a formalized deductive science, we must
always distinguish clearly between the language about which we
speak and the language in which we speak, as well as between
the science which is the object of our investigation and the
science in which the investigation is carried out. The names
o f the expressions o f the first language, and of the relations
between them, belong to the second language, called the meta­
language (which may contain the first as a part). The descrip­
tion o f these expressions, the definition o f fchs complicated
concepts, especially o f those connected with the construction
o f a deductive theory (like the concept o f consequence, o f
provable sentence, possibly o f true sentence), the determination
o f the properties o f these concepts, is the task of the second
theory which we shall call the nietatheory.
For an extensive group o f formalized languages it is possible
1 Strictly speaking this applies only to the signs called constants* Variables
and technical signs (such as brackets, dots, etc.) possess ro independent mean­
ing ; but they exert an essential influence on the meaning the expressions
of whioh they form parts.
* Finally, the definitions are so constructed that they elucidate or determine
the meaning of the signs which are introduced into the language by means of
primitive signs or signs previously defined (cf. p. 166, note 1)
168 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES VH I, § 2

to give a method by which a correct definition of truth can be

constructed for each of them. The general abstract description
of this method and of the languages to which it is applicable
would be troublesome and not at all perspicuous. I prefer
therefore to introduce the reader to this method in another
way. I shall construct a definition of this kind in connexion
with a particular concrete language and show some of its most
important consequences. The indications which I shall then
give in § 4 of this article will, I hope, be sufficient to show how
the method illustrated by this example can be applied to other
languages of similar logical construction.
I choose, as the object of my considerations, the language of
a deductive science of the utmost simplicity which will surely
be well known to the reader—that of the calculus of classes.
The calculus of classes is a fragment of mathematical logic and
can be regarded as one of the interpretations of a formal science
which is commonly called the algebra of logic,1
Among the signs comprising the expressions of this language
I distinguish two kinds, constants and variables I introduce .2

only four constants: the negation sign the sign of logical sum
{disjunction) the universal quantifier TI\ and finally the
inclusion sign ‘I .3 I regard these signs as being equivalent in
1 Cf. Schroder, E. (62), vol. 1 (especially pp. 160-3) and Whitehead, A. N.,
and Russell, B. A. W. (90), vol. i, pp. 205-12.
2 By making use o f an idea of Lukasiewicz I avoid introducing any technical
signs (like brackets, dots, etc.) into the language, and this is due chiefly to
the fact that I always write the functor before the arguments in every
meaningful expression; cf. Lukasiewicz, J. (51), especially pp. v and 40.
3 Usually many other constants occur in the calculus of classes, e.g. the
existence sign, the sign of implication, of logical product (conjunction), of
equivalence, of identity, as well as of the complement, the sum, and the
product o f classes (see p. 168, note 1); for that reason only a fragment of the
calculus o f classes can— formally speaking—be constructed in the language
under consideration. It is, however, to be noted that all constants of the
calculus o f classes could be introduced into this language as defined terms,
if we complete its formalization by making the introduction of new signs
possible by means of definitions (see p. 166, note 1). Owing to this fact our
fragmentary language already suffices for the expression o f every idea which
can be formulated in the complete language o f this science. I would also point
out that even the sign of inclusion *V can be eliminated from our language by
interpreting expressions of the type 4xy ’ (where any variables occur in the place
o f *«’ and 4y ’) in the same way in which in the sequel we shall interpret the
expression
VIH , § 2 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES 169

meaning respectively with the expressions ‘not’, ‘or’, ‘for alT

(in the sense in which this expression was used in statement (6)
of § 1, for example) and ‘is included in’. In principle any
arbitrary symbols could be used as variables, provided only
that their number is not limited and that they are distinct
in form from the constants. But for the further course of our
work it is technically important to specify the form of these
signs exactly, and in such a way that they can easily be ordered
in a sequence. I shall therefore use as variables only such sym­
bols as ‘x,’, cx ,9, Kxm\ and analogous signs which consist of the
symbol V and a number of small strokes added below. The
sign whichhas k such small strokes (k being any natural number
distinct from 0) will be called the k-th variable. In the in­
tuitive interpretation of the language, which I always have in
mind here, the variables represent names of classes of indi­
viduals. As expressions of the language we have either single
constants and variables or complexes of such signs following one
another, for example: ‘x,Nx*\ (NIxfx„\ cAIx,x» Ix*x,\
Tkr,\ (Hx,Ix»xw\ (I xmXtn and so on. Expressions of the type
zNp\ *Apq\ ‘ILrp\ and 'lay9, where in the place of 1 p 9 and fq9
any sentences or sentential functions (this termwill be explained
below), and in the place of ‘x9and ‘y 9any variables, appear, are
read: ‘not p 9(or ‘it is not true that p 9),1 ‘p or qy9 ‘for all classes x
wehavep’, and ‘the classx isincludedinthe classy 9,respectively.
Regarding composite expressions, i.e. those which are not signs,
we can say that they consist of two or more other simple expres­
sions. Thus the expression ‘N Ix ,x *9 is composed of the two
successive expressions ‘A75 and (Ix,Xu9 or of the expressions
*NI9and 'x,xa9or finally of the expressions ‘Nix 9 and ‘av.
But the proper domain of the following considerations is not
the language ofthe calculus ofclasses itself but the corresponding
metalanguage. Our investigations belong to the metacalculus
of classes developed in this metalanguage. From this springs
the need to give the reader some account—if only a very brief
1 For stylistic reasons we sometimes use the expression ‘it is not true that *
instead of the word ‘ not *, the whole expression being regarded as a single word,
no independent meaning being given to the separate parts, and in particular
to the word ‘ true’, which occur in it.
170 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES VIII, § 2

one—ofthestructureofthemetalanguageandofthemetatheory.
I shall restrict myself to the two most important points: (1) the
enumeration of all the signs and expressions which will be used
in the metalanguage, without explaining in more detail their
importance in the course of the investigation, and (2) the setting
up of a system of axioms which suffices for the establishment of
the metatheory or at least will form a foundation for the results
obtained in this article. These two points are closely connected
with our fundamental problem; were we to negleot them, we
Bhould not be able to assert either that we had succeeded in
correctly defining any concept on the basis of the metalanguage,
or that the definition constructed possesses any particular con­
sequences. But I shall not attempt at all to give the metatheory
the character of a strictly formalized deduotive science. I shall
content myself with saying that—apart from the two points
mentioned—the process of formalizing the metatheory shows no
specific peculiarity. In particular, the rules of inference and of
definition do not differ at all from the rules usedin constructing
other formalized deductive sciences.
Among the expressions of the metalanguage we can distin­
guishtwokinds. To thefirstbelongexpressions of a general logical
character, obtainable from any sufficiently developed system
of mathematical logic.1 They can be divided into primitive
expressions and defined expressions, but this would be point­
less in the present case. First we have a series of expressions
which have the same meaning as the constants of the science
we are considering; thus ‘not’ or ‘ii is not true that’,2 ‘or’, fo r dll’,
and ‘is included in ’—in symbols ‘ s ’. Thanks to this circum­
stance we are able to translate every expression of the language
into the metalanguage. For example, the statement ‘for all a
(or for all classes a) a £ o’ is the translation of the expression
sr r x, Ixt x,\ To the same category belongs a series of analogous

1 For example, from the work Whitehead, A. N., and Bussell, B* A* W,

(90). (But I do not intend to use here any special logical symbolism. Apart
from the exceptions which I shall explicitly mention I shall use expressions
of colloquial language.) For the meaning of the general logical expressions
given below see Carnap, B. (8 ).
a See p. 169, note 1.
v m , 5 2 CONCEPT OF TROTH IN FORMALIZED LANGUAGES 171

expressions from the domain of the sentential calculus, of the

first order functional calculus and of the calculus of classes, for
example, ‘i f . .., then', ‘and', ‘if and only if'. ‘for somex' (or ‘there
is an x such that . . . ' ) , ‘is not included in'—in symbols ‘is
identical with'—in symbols ‘= ’, ‘is distinct from'—in symbols
‘is an dement o f '—in symbols V , ‘is not an element o f '—in
symbols ‘e’, ‘individual', ‘class', ‘null class', {dass of all x such
that', and so on. We also find here some expressions from the
domain of the theory of the equivalence of classes, and of the
arithmetic of cardinal numbers, e.g. ‘finite class', ‘infinite class',
4power of a class', ‘cardinal number', ‘natural number' (or ‘finite
cardinal number'), ‘infinite cardinal nu m ber ‘O’, 42\ * < ',
4> ’, 4— ', . . . ♦ Finally I shall need some terms
from the logic of relations. The class of all objects x, to which
there corresponds at least one object y such that xRy (i.e. x
stands in the relation R to y) will be called the domain of the
binary or two-termed relation R. Analogously, the counter domain
of the relation R is the set of all objects y for which there is at
least me object x such that xRy. In the case of many-termed
relations we do not speak of domain and counter domain, but
of the 1st, 2nd, 3rd,..., n4h domain of the relation. The relation
having only one element x in its domain and only one element y
in its counter domain (a relation which thus holds only between
x and y and between no other two objects) is called an ordered
pair, where x is the first and y the second member. Analogously
using many-termed relationswedefineorderedtriples, quadruples,
and in general ordered n-tuples. If, for every object y belonging
to the counter domain of a two-termed relation R, there is only
me object x suchthat xRy, thentherelation R is calledone-many.
The concept of sequence will play a great part in the sequel. An
infinite sequence ia a one-many relation whose counter domain is
the class ofail natural numbers excludingzero. In the sameway,
the term ‘finite sequence of n terms' denotes every one-many
relation whose counter domain consists of all natural numbers
k such that 1 < k < n (where n is any natural number distinct
from 0). The unique x which satisfies the formula xRJc (for a
given sequence R and a given natural number k) is called the
172 CONCEPT OF TROTH IN FORMALIZED LANGUAGES V U I, § 2

k-th term of the sequence R, or the term of the sequence R with

index k, and is denoted b y W e say that the sequences R
and S differ in at most the k-th place, if any tw o corresponding
term s o f these sequences Rtand$< are identical with the exception
o f the &th term s Rk and Skwhich m ay be distinct. In the follow ­
ing pages we shall deal with sequences o f classes and o f natural
numbers, i.e. with sequences all o f whose terms are either classes
o f individuals or natural numbers. In particular, a sequence all
o f whose terms are classes w hich are included in a given class a,
will be called a sequence of subclasses of the class a.
In contrast to the first kind of expression, those of the second
kind are specific terms of the metalanguage of a structural-descrip­
tive character, and thus names of concrete signs or expressions of
the language of the calculus of classes. Among these are, in the
first place, the tern^ ‘the negation sign9, ‘the sign of logical sum9,
Hhe sign of the universal quantifier9, ethe inclusion sign9, ‘the
k-th variable9, Hhe expression which consists of the expressions
x and y following one another9 and ‘expression9. As abbrevia­
tions of the first six terms I shall use the symbols ‘ng9, {sm9,
(un9, ‘in9, ‘vk , and ‘x ' y 9 (the sign V thus denotes a sequence,
the terms of whichare the successivevariablesvv v2ivz,..,). These
terms have already been used in introducing the reader to the
language of the calculus of classes. I hope that, thanks to the
explanations already given, no doubt will remain concerning the
meaning of these terms. With the help of these terms (and pos­
sibly general logical terms) all other concepts of the meta­
language of a structural-descriptive kind can be defined. It is
easy to see that every simple or composite expression of the
language under investigation has an individual name in the
metalanguage similar to the structural-descriptive names of
colloquial language (cf. pp. 156 and 157). For example, the
symbolic expression i{{ng''inY'v-ff'v2 can serve as aname of the
expression Nix, x„\ The fact that the metalanguage contains
both an individual name and a translation of every expression
(and in particular of every sentence) of the language studied
will play a decisive part in the construction of the definition of
truth, as the reader will see in the next section.
V m , § 2 CONCEPT OF TRTJTH IN FORMALIZED LANGUAGES 173

As variables in the metalanguage I shall use the symbols

(1) ‘a’, *b’ ; (2) </\ y , ‘A’ ; (3) ‘Jb\ V , W , V , y ; (4) V, V , V ,
V , y , V ; and (6) ‘X ’, In this order they represent the
names of (1) classes of individuals of an arbitrary character,1
(2) sequences o f such classes, ( 3 ) natural numbers and sequences
o f natural numbers, (4 ) expressions, and (5 ) classes o f expressions.
We turn nowto the axiomsystemof the metalanguage. First,
it is to be noticed that—corresponding to the two kinds of ex­
pressions in the metalanguage—this system contains two quite
distinct binds of sentences: the general logical axioms which
suffice for a sufficiently comprehensive system of mathematical
logic, and the specific axioms of the metalanguage which describe
certainelementaryproperties of the above structural-descriptive
concepts consistent with our intuitions. It is unnecessary to
introduce explicitly the well-known axioms of the first kind.2
As axioms ofthe secondkindweadopt thefollowingstatements:3
Axio m 1. ng, sm, un, and in are expressions, no two of which
are identical.
A x io m 2. vkis an expression if and only if his a natural number
distinct from 0;v kis distinct from ng, sm, un, in, and also from vz
if Jc =£1.
A xio m 3 . xTy is an expression if and only if x and y are ex­
pressions; xTy is distinct from ng, sm, un, in, and from each of the
expressions vk.
A xiom 4 . I f x, y, z, and t are expressions, then we have
x^y = z^t if and only if one of the following conditions is satis­
fied: (a) x =■ z and y — t; (j3) there is an expression u such that
x — z^u and t = vTy\ (y) there is an expression u such that
z = x^u and y = u^L
A xio m 5 . (The principle o f induction.) Let X be a class
which satisfies the following conditions: (a) ng e X , sm e X, un e X
1 Although in the cases (1) and (4) I use distinct variables I here treat
expressions as special classes of individuals, namely as classes of concrete series
of printed signs (cf. p. 156, note 1).
* They may again be taken from Whitehead, A. N., and Russell, B. A. W.
(90), cf. p. 156, note 1.
* As far as I know the metatheory has never before been given in the form
of an axiomatized system.
174 CONCEPT OP TROTH IN FORMALIZED LANGUAGES V III, f 2

and in e X ) (ft) if Jc is a natural number distinct from 0, then

vk e X ; (y) if x e X and y s i ; then vTy e X . Then every ex­
pression belongs to the class X .
The intuitive senseof Axs. 1-4 requiresno further elucidation.
Ax. 5gives apreciseformulationofthefact that everyexpression
consists of a finite number of signs.
It is possibleto prove that theabove axiom system is categorical.
This fact guarantees to a certain degree that it will provide a
sufficient basis for the construction of the metalanguage.1
Some of the above axioms have a pronounced existential
character and involve further oonsequences of the same kind.
Noteworthy among these oonsequences is the assertion that the
class of all expressions is infinite (to be more exact, denumer­
able). From the intuitive standpoint this may seem doubtful
and hardly evident, end on that account the whole axiom-
system may be subject to serious criticism. A closer analysis
wouldrestriot this critioism entirelyto Axs. 2 and 8 as the essen­
tial souroes of this infinite character of the metatheory. I shall
not pursue this difficult problem any further here.2 The con-
1 I use the term ‘ categorical’ in the sense given in Veblan, O. (8 6 ). I do
not propose to explain in more detail why I see in the categoricity of en
axiom system an objective guarantee that the system suffices for the establish­
ment of the corresponding deductive science 5 a series of remarks on this question
will be found in Fraenkei, A. (16).
Regarding the interpretation of the term ‘ categorical* there are certain*
although not especially important, differences of opinion. Without going into
details I may mention that in the case of one of the possible interpretations
the proof that the system is categorical would require the addition of *wo
further axioms to the system given in the text, in these axioms (which oilier*
wise are not of great importance) the specific conception of expressions as
classes would occur (of. p. 156, note 1). The first axiom would state that two
arbitrary expressions are disjoint classes (i.e. have no element in common),
in the second the number of elements of every expression would be stipulated
in some way.
* For example, the following truly subtle points are here raised. Normally
expressions are regarded as the products of human activity (or as classes of
such products). From this standpoint the supposition that there are infinitely
many expressions appears to be obviously nonsensical. But another possible
interpretation of the term ‘ expression* presents itself: we could consider all
physioal bodies of a particular form and size as expressions. The kernel of the
problem is then transferred to the domain of physics. The assertion of the
infinity of the number of expressions is then no longer senseless and even
forms a special consequence of the hypotheses which are normally adopted in
physics or in geometry.
V m , $ 2 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES 176

sequences mentioned could of course be avoided if the axioms

were freed to a sufficient degree from existential assumptions.
But the fact must be taken into consideration that the elimina­
tionorweakeningofthese axioms, whichguaranteetheexistence
of all possible expressions, would considerably increase the
difficulties of constructing the metatheory, would render im­
possible a series of the most useful consequences and so intro­
duce much complication into the formulation of definitions
and theorems. As we shall see later this will become dear
even in the present investigations. For these reasons it seems
desirable, at least provisionally, tc»base our work on the axiom
system given above in its initial unweakened lorm.
Making use of the expressions and symbols of the meta­
language which have now been enumerated, I shall define
those concepts which establish the calculus of classes as a
formalized deductive science. These are the concepts of sen­
tence, axiom (primitive sentence), consequence and provable sen­
tence. But first I introduce a series of auxiliary symbols which
will denote various simple types of expression and greatly
facilitate the later constructions.
DsFENmON 1. x is an inclusion with vk as first and t>. as second
term—in symbols x — —if and only if x = (iri~'vk)'~'vl.
D e fin itio n 2. x is a negation of the expression y—in symbols
x =e y—i f and only if x — ngTy.
D efin itio n 3 . a; logical sum (disjunction) of the expressions
y and z—in symbols x — y-{-z—if and only if x — (emTyY'z.
D e fin itio n 4 . x is a logical sum of the. expressions tv ia,..Mtn
(or a logical sum of a finite n-termed sequence t of expressions)-—
n
in symbols x s= ^ tk—if and only if t is a finite n termed
sequence of expressions which satisfies one of the following con-
n—1
ditions: (a) n == 1 and x «= tv (fi) n > 1 and x « ^ 1

1 As will be seen, Def. 4 is a recursive definition which, as such, raises certain

methodological misgivings. It is, however, well known that with the help of a
general method, the idea of which we owe to G. Frege and R. Dedekind, every
recursive definition caD be transformed into an equivalent normal definition
176 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES V m , § 2

D efin itio n 5 . x is a logical product (conjunction) of the

expressions y and z—in symbols x — y.z—if and only if
x = y+z.
D efin itio n 6. x is a universal quantification of the expression
y under the variable vk—in symbols x — y—if and only if
x = (unTvk)''y.
D e fin itio n 7 . x is a universal quantification of the expression
y under the variables vPi, vPt,..., vPn—in symbols x = f ] $ f ny—i f
and only if p is a finite n-termed sequence of natural numbers
which satisfies one of the following conditions: (a) n = 1 and
* = f k 3A (P)n > 1 andx = D p? ” - 1 flp .2 /-

D efin itio n 8. x is a universal quantification of the expression

y if and only if either x = y or there is a finite n-termed sequence p
of natural numbers such that x — C\pfnV-
D efin itio n 9 . x is an existential quantification of the expres-
siony under the variable vk—in symbols x =• \ J ky—if and only if
« = n J .
We have thus introduced three fundamental operations by
means of which compound expressions are formed from simpler
ones: negation, logical addition, and universal quantification.
(Logical addition is, of course, the operation which consists in
forming logical sums of given expressions. The terms ‘negation*
and ‘universal quantification* are thus used to refer both to
certain operations on expressions and to expressions resulting
from these operations.) If, beginning with the inclusions ikj,
we apply to them the above operations any number of times
we obtain an extensive class of expressions which are called
sentential functions. We obtain the concept of sentence as a
special case of this notion.
(cf. Dedekind, R. (15), pp. 33-40, and Whitehead, A. N., and Russell, B. A. W.
(90), vol. 1, pp. 550-7, and vcl. 3, p. 244). This, however, is unpractical in so
far as the formulations so obtained have a more complicated logical structure,
are less clear as regards their content, and are less suitable for further deriva­
tions. For these reasons I do not propose to avoid recursive definitions in the
sequel.
v m , $ 2 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES 177

D efinition 10. x is a sentential function if and only if x is an

expression which satisfies one of the four following conditions:
(a) there exist natural numbers k and l such that x = ikl', (/3) there
exists a sentential function y such that x = y ; (y) there exist sen­
tential functions y and z such that x — y + z ; (8) there exists a
natural number k and a sentential function y such that x —
The following expressions will serve as examples of sen­
tential functions according to Def. 10: 'lx,x*\ 'Nix,x„,\
cAIx,xmIxmx,\ ‘XT x,.NIx,x/, and so on. On the other hand
the expressions ‘I\ sIx,\ 'AIx,x,„\ ‘XT I^x„\ etc., are not
sentential functions. It is easily seen that for every sentential
function in the language we can automatically construct a
structural-descriptive nameofthis functioninthe metalanguage,
by making use exclusively of symbols which were introduced
in Defs. 1, 2, 3, and 6. For example, the following symbolic
1 Def. 10 is a recursive definition of a somewhat different type from that
of Def. 4 since the usual ‘transition from n—1 to n* is lacking in it. In order
to reduce this to an ordinary inductive definition we must first inductively
define the expressions *x is a sentential function of the nth degree* (inclusions
tfc I would then be functions of the 0th degree, the negations and logical sums
of these inclusions, as well as their generalizations for any variable, functions
of the 1st degree, and so on), and then simply stipulate that *x is a sentential
function ’ means the same as *there is a natural number n such that x is a sentential
function of the nth degree \ Def. 10 could also be transformed into an equivalent
normal definition in the following way:
xis a sentential function if and only if the formula x e X holds for every doss
X which satisfies the following four conditions: (a) if k and l are natural numbers
distinct from 0, then e X ; ifi) if y e X , then y $ X ; (y) if y 6 X and z e X
then y+ z e X ; (8) if k is a natural number distinct from 0 and y e X , then
n kV ^ x.
It should be emphasized that recursive definitions of the type of Def. 10
are open to much more serious methodological objections than the usual
inductive definitions, einoe in contrast to the latter, statements of this type
do not always admit of a transformation into equivalent normal definitions
(see p. 175, note 1). The fact that such a transformation is possible in the
present case is owing to the special nature of the concepts occurring in the
definition (to the fact, namely, that every expression consists of a finite number
of signs and that the operations given in conditions (j3)-(5) always lead from
shorter to longer expressions). If, nevertheless, I sometimes give definitions
of this kind in the present article in the place of equivalent normal definitions
(Defs. 10, 11, 14, 22, and 24), I do so because these definitions have important
advantages of quite another kind: they bring out the content of the concept
defined more clearly than the normal definition does, and—in contrast to the
usual recursive definition—they require no previous introduction of accessory
ooncepts which are not used elsewhere (e.g. the accessory concept of a sentential
funotion of the nth degree).
178 CONCEPT OP TRUTH IN FORMALIZED LANGUAGES VH I, § 2

expressions function as names of the above examples of sen­

tential functions: and ‘Hih,*’-
D efinition 11. vkis a free variable of the sentential function x
if and only if k is a natural number distinct from 0, and x is a sen­
tential function which satisfies one of the following four conditions:
[a) there is a natural number l such that x = i^ orx — tZ(fc; (/?) there
is a sentential function y such that vk is a free variable of y and
x~ (y) there are sentential functions y and z such that vk is a
free variable ofy and x — y-\-z orx — z-\-y; (8) there is a number l
distinct from k and a sentential function y such that vk is a free
variable o fy and x = f V -
Variables which occur in a sentential function but are not free
variables of this function, are usually called bound (apparent)
variables.1
D efinition 12. x is a sentence (or a meaningful sentence)—
in symbols x e S—if and only if x is a sentential function and no
variable vk is afree variable of thefunction x.
Thus the expressions: H i1!,!’ flxU* Ll,%>
n A i + r i i i W ) are sentences, but the functions: tw, n« ■%2.
l3,I+flx U« H.Xare nofc sentences because they contain the free
variable vv By virtue of the above definition the symbol ‘S*
denotes the class of all meaningful sentences.
The system of prim itive sentences of the calculus of classes
will contain two kinds of sentences.2 The sentences of the first
kind are obtained by taking any axiom system which suffices
as a basis for the sentential calculus and contains the signs of
negationandlogical additionasthe onlyconstants- Forexample,
the axiom system consisting of the following four axioms:
‘ANAppp’, *ANpApq\ 'ANApqAqp',
and 'ANANpqANArpArq9.3
1 Cf. Hilbert, D., aod Ackermann, W, (30), pp. 62-54.
1 Concepts which I shall discuss in the further course of § 2 do not occur in
the definition of true sentence itself. I shall, however, make use of them in the
preparatory discussions at the beginning of $ 3 which establish the definitive
form of the definition. I shall also use them in the formulation of certain
consequences of this definition (Tbs. 3-6 of $ 3) which express characteristic
and materially important properties of true sentences.
* This axiom system is the result of a modification and simplification of
the axiom system which is found in Whitehead, A. N., and Bussell, B. A. W.
(90), vol. 1, pp. 96-97; cf. Hilbert, D., and Ackermann, W. (30), p. 22.
V U I, § 2 CONCEPT OF TROTH IN FORMALIZED LANGUAGES 179

In these axioms we replace the sentential variables ‘p ’, ‘q\ and

V by any sentential functions, and then to the expressions thus
obtained, if they are not already sentences, we apply the opera­
tion of universal quantification a sufficient uumber of times
until ail the free variables have disappeared. The following will
serve aa examples:
‘ANA J I x, lx,x, I J x, lx , x, JJ x, lx, x,’,
X I*'II x* ANIx,Xn AIxtX* lx*x,\ etc.
In order to obtain the sentences of the second kind we shall
take as our starting-point some axiom system of the as yet
unformalized calculus of classes which contains the inclusion
sign as the only undefined sign,1 and we than translate the
axioms of this system into the language of the present article.
Naturally we must first eliminate all constants which are defined
by means of the inclusion sign* as well as ell term** belonging to
the sentential calculus and the functional calculus which are
distinct in meaning from the universal quantifier, the negation
sign and the sign of logical addition. As examples of sentences
of this second kind we have
TI x*9 an<^ T I n Xr XTXwa n a n ix* IXiXur’ .
DEmariON 13. x is an axiom (primitive sentence) if and only
if x satisfies one of the two following conditions, (a) x e S and
there exist sentential f unctions y} z, and u such that x is a universal
quantification* of one of the four fuvxtions y-ry-ty, y+ sy- f-z),
y+z-'riz ^y), and y+z^{u+p+i;u+z})'} (fi) & is identical with
one of thefive sentences
Hi ‘l,!* Hi fls
f i i n » u - 3 ( ^ - *s,0c ^ 4+

f l i P:a cs,2* “**,&/)•

and
fli Uaffls *3,2+-3»*)• d' ht$)) • _
PUds. 2~>~Uev^if *%s)))'
1 I have chosen here the system o f postulates which is given in Huntington,
E. V. (32), p. 297 (this system has, however, been simplihed by the elimination
of, among others, certain assumptions of an existential nature).
180 CONCEPT OF TROTH IN FORMALIZED LANGUAGES VILE, § 2

In the formulation of the definition of the concept of con­

sequence I shall use, among others, the following expression:
*u is an expression obtainedfrom the sentential function w by sub­
stituting the variable vkfor the variable vf. Theintuitive meaning of
this expression is clear and simple, but in spite of this the defini­
tion has a somewhat complicated form:
D efin itio n 14. x is an expression obtained from the sentential
function y by substituting the {free) variable vkfor the (free) variable
vt if and only if k and l are natural numbers distinct from 0, and x
and y are sentential f unctions which satisfy one of thefollowing six
conditions: (a) x = ikk and y = ilf,; (fi) there exists a natural num­
ber m distinct from l>such that x = ikm and y = ilm or x — i^kand
y = (y) vt is not a free variable of the function y and x = y\
(S) there exist sentential functions z and t such that x = zsy = i,
and z is an expression obtained from t by subsiihvting the variable
vkfor the variable (e) there exist sententialfunctions z, t, u, and w}
such that x = z+u, y = t-\-w, where z and u are obtained from t
and w respectively by substituting the variable vkfor the variable vt\
(£) there exist sentential functions z, t and a natural number m
distinctfrom k and l such that x — “ Om an^ z i's obtained
from t by substituting the variable vkfor the variable
For example, it follows from this definition that the expres­
sions pj3 (^i+*i,s) *1,3+ 02*2,3are obtained from the
functions: t2>2, Hs (4.2+4,s) and 12,3+ fl 24.3 respectively by
substituting v1 for v2. But the expression 0 1*1,3cannot be ob­
tained in this way from the function 02*2,3nor the expression
Oi*i,i from the function 02*2>i«
1 The following is a normal definition which is equivalent to the above
recursive one (cf. p. 177, note 1):
x is an expression obtained from the sentential function y by substituting the
variable vk for the variable if and only if k and l are natural numbers distinct
from 0 and if the formula xRy holds for every relation R which satisfies the
following six conditions: (a) tktk Ril l ; (ft) i f m is a natural number distinct from
0 and l, then Rii>mand imtkRim,t > (y) if z is a sentential function and vj is not
a free variable of z, then zR z; (8) if zRtt then zR l; (c) if zRt and uRw, then
z + u R t+ w ; (£) if m is a natural number distinct from 0, kt and l and zRtt then
rim**
The definitions of substitution in Lesmeweki, S. (46), p. 73 (T.E. xlvii),
and (47), p. 20 (T.E. xxvn°) depend on a totally different idea.
VIII, § 2 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES 181

Among til© consequences of a given class of sentences we

include first all the sentences belonging to this class, and all
the sentences which can be obtained from these by applying,
an arbitrary number of times, the four operations of substi­
tution, detachment, and introduction and removal of the universal
quantifier.1 If we had wished to apply these operations not
only to sentences, but to arbitrary sentential functions, ob­
taining thereby sentential functions as results, thenthe meaning
of the operation of substitution would be completely deter­
mined by Def. 14, the operation of detachment would correlate
the function z with the functions y and y+ z, the operation
of introduction of the universal quantifier would consist in
forming the function y+fl* z fromthe function y + z (provided
that vk is not a free variable of the function y), the operation of
removal oftheuniversal quantifierwouldproceedinthe opposite
direction—from the function y+flfc2 function y + z .1
In order to simplify the constructionI first definethe auxiliary
concept of consequence of the n-th degree.
D efinition 15. x is a consequence of the wth degree of the
class X of sentences if and only if x e S, X £ S, n is a natural
number and either (a) n = 0 and x e X , or n > 0 and one of the
following five conditions is satisfied: (J3) x is a consequence of the
n— Ith degree of the class X ; (y) there exist sententialfunctions u and
w, a sentence y and natural numbers k and l such that x is the univer­
sal quantification of thefunction u, y is the universal quantification
of thefunction w, u is obtainablefrom the function w by substituting
the variable vkfor the variable vl9and y is a consequence of the class
X of the n—lth degree; (S) there exist sententialfunctions u and w
as well as sentences y and z such that x, y, and z are universal
quantifications of the functions u, w-\-u, and w respectively, and y
and z are consequences of the doss X of the n— Ith degree; (*) there
exist sententialfunctions uand w, a sentence y and a natural number
k such that x is a universal quantification of thefunction u-\-f\kw,
y is a universal quantification of the function u-\-w, vk is not a free
variable of u, and y is a consequence of the doss X of the n—lth
1 Cf. Lukasiewicz, J. (61), pp. 169-63; IV> p. 56.
182 C O N C E PT O F T R U T H IN F O R M A L IZ E D L A N G U A G E S Vm . f 2

degree; (0 there exist sentential functions u and w, a sentence y and

a natural number k, such that x is a universal quantification of the
function u+w, y is a universal quantification of the function
«-}- wand y**a> consequence of the doss X of the n—1th degree.
D efinition 10. xisa consequenceof the class X of sentences—
symbolically x e On(X)—if and only if there is a natural number n
such that x isa consequence of the nth degree of the doss X .1
D efinition 17. x is a provable (accepted) sentence or a
theorem—in symbols x e Pr—if and only if x is a consequence of
the set of all axioms.
From this definition, it is easy to see that we shall have,
among the provable sentences, not only all the sentences which
can be obtained from the theorems of the sentential calculus
in the same way in whioh the axioms of the first kind (i.e. those
satisfying the oondition (a) of Def. 13) were obtained from the
axioms of the sentential calculus, but also all known theorems
of the unformalized calculus of classes, provided they are first
translated into the language under investigation. In order to
become convinced of this weimitate in the metatheory, in every
particular case, the corresponding proof from the domain
of the sentential calculus or of the calculus of classes. For
example, it is possible in this way to obtain the sentence
n x K x + l o ) fr°m fch® well-known theorem ‘A N pp’ of the
1 The concept of consequence could also be introduced directly (i.e. without
the help of consequence of the nth degree) in the following way:
x € Cn{X) if and only if XQ S and if the formula x e Y holdsfor every doss
Y which satisfies thefollowing conditions: (at) X Q Y ; (ft) if y e S and is a universal
quantification of thefunction u,z is a universal quantification of the function w, u
is obtainablefrom the function w by substituting the variable v*. for the variable V;
and z e Y, then y e Y ; (y) if y GS, y> z, and t are universal quantifications of the
functions u, w+u, and w respectively and ze Yandte Y,thenye Y ; {b)ifysS,
u and w are sentential functions, y is a universal quantification of the function
u fflfc) z a universal quantification of the function u-^w, vk is not a free
variable of thefunction u and z e Y9then y e Y ; («?) if y e u and w are sentential
functions, y is a universal quantification of the function u+w, z a universal
quantification of the function u+ w and z e Y, then y e Y.
It is, however, to be noted that by transformation of the definition just given
into a recursive sentence of the type of Def. 10 we obtain a sentence which
is equivalent neither with the above definition nor with any other normal
definition (cf. p. 177, note 1).
V m , $ 8 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES 183

sentential calculus. Translating the proof o f this theorem,1 we

show successively from D ef. 13 that

ru<*rHu+i4>. nx (‘•
ia iu) )>

n i ( , l A + ti A + ‘ i A + ( tW + ( ‘ x , i + tx . i ) + (ti A + tl ,x ) ) )

are axiom s; consequently by D ef. 15

n i ( li A + ( ti A + tL l ) "fr* ( h . i + h . i ) )

is a consequence of the 1st degree and is a conse­

quence of the second degree of the class of all axioms. Hence by
Defs. 16 and 17 ni(ti,i+ti,i)18 a provable sentence.
From examples of such inferences the difficulties can be
imagined which would at once arise if we wished to eliminate
from the axioms of the metatheory the assumptions which are
of an existential nature. The fact th&t the axioms would no
longer guarantee the existence of some particular sentenoe,
which we wish to demonstrate, is not of much consequence.
Serious importance attached only to the fact that, even as­
suming the existence of some concrete sentence, we could
not establish its provability; since in the proof it would be
necessary to refer to the existence of other, as a rule more com­
plicated, sentences (as is seen in the proof of the theorem
Tlxfry+ha) which was sketohed above). So long as we
are dealing with speoial theorems of the type ‘x e Pr\ we oan
take measures to provide these sentences with premisses which
guarantee the existence of the sentences necessary for the proof.
The difficulties would increase significantly if wepassed to sen­
tences of a general character which assert that all sentences of
a certain kind are provable—or, still more generally, are con­
sequences of the given class ofsentences. It would then often be
necessary to include among the premisses general existential
assumptions which would not be weaker than those which, for
intuitive reasons, we bad eliminated fromthe axioms.8
* Of. Whitehead, A, N.. and Russell, B. A. W (90), vol. 1, p. 101, *2.1.
1 This is t^sily seen from tho examples' of Ths. 11, I2t 24, and 28 in § 3.
184 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES V m , § 2

For these reasons the standpoint might be taken that Def.

17, in case the existential assumptions are rejected, would no
longer embrace all the properties which we ascribe to the con­
cept of theorem. The problem of a suitable ‘correction’ of the
above definition would then arise. More precisely expressed,
it would be a question of constructing a definition of theorem
which would be equivalent to Def. 17 under the existential
assumptions and yet—quite independently of these assump­
tions—would have as consequences all theorems of the type
*if the sentence x exists, then x e Pr\ provided the corresponding
theorem ‘x e Pr ’ could be proved with the help of the existential
assumptions. I shall give here a brief sketch of an attempt to
solve this problem.
It can easily be shown that the axiom system adopted in the
metatheory possesses an interpretation in the arithmetic of the
natural numbers. A one-one correspondence can be set up be­
tween expressions and natural numbers where operations on
numbers having the same formal properties are correlated with
the operations on expressions. If we consider this correspon­
dence, we canpickout, fromthe class of all numbers, those which
are correlatedwithsentences; amongthesewill be the ‘primitive’
numbers. We can introduce the concept of a ‘consequence’ of
a given class of numbers, and finally define the ‘accepted’
numbers as ‘consequences’ ofthe class ofall ‘primitive’ numbers.
If wenoweliminatethe existential assumptions fromthe axioms,
the one-one correlationdisappears: to every expressionanatural
numberstill corresponds, but notto everynumber, anexpression.
But we can still preserve the concept of ‘accepted’ number
previously established and define the theorems as those which
are correlated with ‘accepted’ numbers. If we try, on the basis
of this new definition, to prove that a concrete sentence is a
theorem, we shall no longer be compelled—as is easily seen—to
refer to the existence of any other sentences. Nevertheless the
proof will still require—and this must be emphasized—an
existential assumption, the assumption, namely, that there
exist sufficiently many natural numbers or—what amounts
to the same thing—sufficiently many distinct individuals. Thus
VJU, $ 2 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES 185

in orderto derive all desired conclusions fromthe newdefinition,

it would be necessary to include in the metatheory the axiom of
infinity, i.e. the assumption that the class of all individuals is
infinite.1 I know of no method, be it even less natural and more
complicated than the one just discussed, which would lead to a
satisfactory solution of our problem which is independent of
the above axiom.
In connexionwiththe concepts of consequenceandoftheorem
I have mentioned rules of inference. When we have in mind the
construction of a deductive science itself, and not the investiga­
tion of such a science carried out on the basis of the metatheory,
we give, instead of Def. 17, a rule by which we may add to the
science as a theorem every consequence of the axioms. In our
case this rule can be divided into four rules—corresponding to
the four operations which we use in the construction of con­
sequences.
By means of the concepts o f sentence and o f consequence all
the most important methodological concepts can be introduced
into the metatheory, in particular the concepts o f deductive
system, o f consistency and o f completeness.*
D efinition 18. X is a deductive system if and only if
Cn(X) c X s 8.
D efinition 19. X isa consistent class of sentences if and only if
X £ S and if, for every sentence x, either x e Gn(X) or x e Cn(X).
D efinition 20. X isa complete class of sentences if and only if
I g S and if, for every sentence x, either x e Cn(X) o rx e Cn(X).
In the sequel yet another concept will prove useful:
D efinition 21. The sentences x and y are equivalent with
respect to the class X of sentences if and only if x e S ,y e S, X £ S
and both x + y e Cn(X) and y + x e Cn{X).
A more detailed analysis o f the concepts introduced in this
section would exceed the limits o f the present work.
1 Cf. Whitehead, A. N., and Russell, B. A. W. (90), vol. 2, p. 203.
a Cf. pp. 70, 90, and 93 of the present volume.
186 CONCEPT OP TRUTH IN FORMALIZED LANGUAGES VXII, § 3

§ 3. T he Concept op T rue Sentence in the L anguage op

the Calculus op Classes
I passonnowtothechiefproblemofthis article—theoonstruo-
tion ofthe definition of true sentence, the language of the oaloulus
of classes still being the object of investigation.
It might appear at first sight that at the present stage of our
discussion this problem can bo solved without further difficulty,
that ‘true sentence’ with respect to the language of a formalized
deductive science means nothing other than ‘provable theorem’,
and that consequently Def. 17 is already a definition of truth
and moreover a purely structural one. Closer reflection shows,
however, that thisviewmustberejectedfor thefollowingreason:
no definition of true sentence which is in agreement with the
ordinaryusage oflanguage shouldhave any oonsequences which
contradict the principle of the excluded middle. This principle,
however, is not valid in the domain of provable sentences. A
simple example of two mutually contradictory sentences (i.e.
such that one is the negation of the other) neither of which is
provable is provided by Lemma E below. The extension of the
two concepts is thus not identical. Prom the intuitive stand­
point all provable sentences are without doubt true sentences
(the Defs. 13-17 of § 2 were formulated with that in mind).
Thus the definition of true sentence whichwe are seeking must
also cover sentences whioh are not provable.1
1 The fact must also be taken into consideration that—in contrast to the
concept of true sentence—the ooncept of provable sentence has a purely
accidental character when applied to some deductive scienoes, whioh is ohiefiy
oonneoted with the historical development of the science. It is sometimes
difficult to give objective grounds for narrowing or widening the extension
of this concept in a particular direction* For example, when we are dealing with
the calculus of classes the sentence which stipulates the existence of
at least two distinct classes, is not accepted on the basis of the definitions of
§ 3—whioh will be expressed in Lemma E. Moreover, this sentence cannot
be derived from the formal hypotheses upon which the work of Schroder is
based, although in this case the matter is not quite clear (cf. Schrdder, E. (62),
vol* 1, pp. 245 and 246; voh 2, Tart 1, p. 278; vol. 3, Part 1, pp. 17 and 18);
but in many works this sentence occurs as an axiom, of the algebra of logic
or forms an obvious consequence of these axioms (cf. Huntington, E. V. (32),
p, 297, Post, 10). For quite different reasons, whioh will be discussed below
in connexion with Th. 24 (cf. especially p. 207, footnote), it would be desirable
to include th© sentence n^rit^i-i- Ua^i* among the
VU I, { 3 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES 187

L et ns try to approach the problem from quite a different angle,

b y returning to the idea o f a semantical definition as in § 1, A s
we know from § 2, to every sentenoe o f the language o f the cal­
culus o f olasses there corresponds in the metalanguage not only a
name o f this sentence o f the structural-descriptive kind, but
also a sentence having the same meaning. For exam ple, corre­
sponding to the sentenoe 'JJ a;,JJ x.Alx,x» Ix,x,’ is the name
and the sentence ‘for any olasses a and b we
have o c i or 6 £ o’.In order to make dear the content o f the
concept o f truth in connexion with some one concrete sentenoe
o f the language with whioh we are dealing we can apply the same
method as was used in § 1 in formulating the sentences (3) and (4)
(of. p. 156). W e take the scheme (2) and replace the sym bol V
in it by the name o f the given sentenoe, and ‘p ’ by its translation
into the metalanguage. A ll sentences obtained in this w ay, e.g,
‘P li HaCb,8+ 1*4 ) w a true sentence if and only if for any dosses a
andbwehavea £ 6 or 6 £ a ’, naturally belong to the metalanguage
and explain in a precise w ay, in accordance with linguistic usage,
the meaning o f phrases o f the form ‘a; is a true sentence’ which
ooour in them . N ot much more in principle is to be demanded o f
a general definition o f true sentence than that it should satisfy
the usual conditions o f methodological correctness and indude
all partial definitions o f this type as spedal oases; that it should
be, so to speak, their logical product. A t m ost we can also
require that only sentences are to belong to the extension o f the
defined concept, so that, on the basis o f the definition con­
structed, all sentences o f the type ‘x is not a true sentence*, in
which in the place o f ‘a?’ we have the name o f an arbitrary ex­
pression (or o f any other objeot) which is not a sentence, oan
be proved.
Using the symbol ‘ 2 V to denote the class ofall truesentences,
the above postulate can be expressed in the following conven­
tion:
Convention T . A formally correct definition of the symbol

theorems, although this is not usually done. In the course of this work I shall
have several occasions to return to the problem of the mutual relations of
these two concepts: of theorem and of true sentenoe.
188 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES VUE, § 3

*Tr\ formulated in the metalanguage, will he called an adequate

definition of truth if it has the following consequences:
(a) all sentences which are obtained from the expression *x e Tr
if and only if p 9 by substituting for the symbol V a structural-
descriptive name of any sentence of the language in question a/ndfor
the symbol *p’ the expression which forms the translation of this
sentence into the metalanguage;
(/?) the sentence for any x ,i f x e Tr then x e S' (in other words
*Tr c S').1

It should be noted that the second part of the above con­

vention is not essential; so long as the metalanguage already
has the symbol ‘ Tr9which satisfies the condition (a), it is easy
to defineanewsymbol 'Tr'9whichalso satisfies the condition (jS).
It suffices for this purpose to agree that Tr' is the common part
of the classes Tr and S.
If the language investigated only contained a finite number of
sentences fixedfromthe beginning, andif we couldenumerate all
these sentences, then the problemofthe constructionof a correct
definition oftruthwouldpresent no difficulties. For this purpose
it would suffice to complete the following scheme: x e Tr if and
only if either x = xxandpv or x — x2andp2,>.. or x — xn and pn,
the symbols 'xx , ‘x29,..., (xn9 being replaced by structural-
descriptive names of all the sentences of the language investi­
gated and 'px > 'p2i..., (pn9by the corresponding translation of
these sentences into the metalanguage. But the situation is not
like this. Whenever a language contains infinitely many sen­
tences, the definition constructed automatically accordingto the
aboveschemewouldhaveto consist ofinfinitelymanywords, and
such sentences cannot be formulated either in the metalanguage
1 If we wished to subject the metalanguage and the metatheory expressed
in it to the process of formalization, then the exact specification of the mean­
ing of various expressions which occur in the convention T would present
no great difficulties, e,g. the expressions ‘formally correct definition of the
given symbol \ *structural-descriptive name of a given expression of the language
studied\ 'the translation of a given sentence (of the language studied) into the
metalanguage1. After unimportant modifications of its formulation the con­
vention itself would then become a normal definition belonging to the meta­
theory.
V m , § 3 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES 189

or in any other language. Our task is thus greatly com­

plicated.
The idea of using the recursive method suggests itself. Among
the sentences of a language we find expressions of rather varied
kinds from the point of view of logical structure, some quite
elementary, others more or less complicated. It would thus
be a question of first giving all the operations by which simple
sentences are combined into composite ones and then deter­
mining the way in which the truth or falsity of composite
sentences depends on the truth or falsity of the simpler ones
contained in them. Moreover, certain elementary sentences
could be selected, from which, with the help of the operations
mentioned, all the sentences of the language could be con­
structed; these selected sentences could be explicitly divided
into true and false, by means, for example, of partial definitions
of the type described above. In attempting to realize this idea
wearehowever confrontedwithaserious obstacle. Even asuper­
ficial analysis of Defe. 10-12 of § 2 shows that in general com­
posite sentences are in no way compounds of simple sentences.
Sentential functions do in fact arise in this way fromelementary
functions, i.e. frominclusions; sentences on the contrary are cer­
tain special cases of sentential functions. In viewof this fact, no
method canbegivenwhichwouldenableusto definethe required
concept directly by recursive means. The possibility suggests
itself, however, of introducing a more general concept which
is applicable to any sentential function, can be recursively
defined, and, when applied to sentences, leads us directly to the
concept of truth. These requirements are met by the notion of
the satisfaction of a given sentential function by given objects, and
in the present case by given classes of individuals.
Let us try first to make clear by means of some examples the
usual meaning of this notion in its customary linguistic usage.
Thewayinwhichweshall dothis represents anatural generaliza­
tion of the method which we have previously used for the con­
cept of truth.
The simplest and clearest case is that in which the given sen­
tential function contains only one free variable. We can then
190 CONCEPT OP TRUTH IN FORMALIZED LANGUAGES V m , $ 8

significantly say of every single object that it does or does not

satisfy the given function.1 In order to explain the sense of this
phrasewe oonsiderthe followingscheme:
for all a, a satisfies the sentential function x if and only if p
andsubstituteinthis sohemefor ‘p’ thegivensentential funotion
(after first replacing the free variable occurring in it by V )
and for V some individual name of this function. Within
colloquial language we can in this way obtain, for example, the
following formulation:
for every a, toe have a satisfies the sentential function ‘x is white’
if and only if a is white
(and from this conclude, in particular, that snow satisfies the
function 'x is white’). A similar construction will be familiar to
the reader from school algebra, where sentential functions of a
special type, called equations, are considered together with the
numbers which satisfy these functions, the so-called roots of the
equations (e.g. 1 is the only root of the equation 'x-\-2 => 3’).
When, in particular, the funotion belongs to the language of
the calculus of classes, andthe correspondingexplanation of the
expression *a satisfies the given sentential funotion’ is to be
formulated wholly in the terms of the metalanguage, thenin the
above soheme we insert for ‘p ’ not the sentential function itself,
but the expression of the metalanguage having the same mean­
ing, andfor V we substitute anindividual name of this funotion
whioh likewise belongs to the metalanguage. For example, this
method gives the following formulation in connexion with the
function ‘JJ x, Ix,x.’ :
for all a, a satisfies the sentential function. p)2tl a if and only
if for all dosses b we have a s b
(whenoe it follows at once that the only class which satisfies the
function ‘JJ x» Ix,x/ is the null class).
In oases where the sentential function has two distinct free
variables we proceed in an exactly analogous manner. The only
1 Provisionally I ignore problems connected with semantical categories (or
logioal types); these problems will be discerned in f 4.
V m . 13 C O N C E P T O F T R U T H IN F O R M A L IZ E D LA N G U A G E S 10 1

difference is that the concept of satisfaction now refers not to

single objeots but to pairs (more accurately to ordered pairs) of
objects. In this way we reach the following formulations:
for aU a and b, a and b satisfy the sentential function 'x sees y’
if and only if a sees b\for all a and b, a and b satisfy the sentential
function, t2i8 (i.e. ‘Ix ,x m’) if and only if a £ 6.
Finally we pass to the general case, where the given sentential
function contains an arbitrary number of free variables. For the
sake of a uniform mode of expression we shall from now on not
say that given objeots but that a given infinite sequence of objects
satisfies a given sentential function. If we restrict ourselves to
functions from the calculus of classes, then the establishment of
an unambiguous explanation of this expression is facilitated by
the fact that all the variables whioh occur in the language of this
science are ordered (enumerated) in a sequence. In considering
the question of whioh sequences satisfy a given sentential func­
tion, we shall always have in mind a one-many correspondence
of certain terms of a sequence / with the free variables of the
sentential function, where with every variable corresponds the
term of the sequence with the same index (i.e. the term f k will
be correlated with the variable vk). No account will be taken of
the terms which are not correlated with any variable.1 We can
explain the procedure best by means of concrete examples.
Consider the function fish,a already mentioned. This function
contains only one free variable vv so that we oonsider only the
first terms of sequences. We say that the infinite sequence f of
classes satisfies the sentential function */ an^ ^
1 This is a simplification of a purely technical nature. Even it we could
not order ell the variables of a given language in a sequence (o.g. because we
used symbols of arbitrary shapes as variables), we could still number ail the
signs, and thus all the variables, of every given expression, e.g. on the basis of
the natural order in whioh they follow one another in the expression: the
sign standing on the extreme left could be called the first, the next the second,
and so on. In this way we could again set up a certain correlation between the
free variables of a given function and the terms of the sequence. This correla­
tion (in contrast to the one described in the text) would obviously vary with
the form of the function in question j this would carry with it rather serious
complications in the formulation of Def. 22 given below and especially of
conditions (y) and ($).
192 CONCEPT OP TROTH IN FORMALIZED LANGUAGES VIH , J a

class f i satisfies this function in the former sense, i.e. if for alt
classes b, we haveft £ 6. In an analogous w ay the infinite sequence
f of classes satisfies the sententialfunction t2g if and only if the classes
/ s and / 8 satisfy the function in the previous sense, i.e. if f %£ / 8.
Ih is process m ay be described in general term s as follow s:
W e consider the follow ing sohem e:
The sequence f satisfies the sentential function x if and only if f
is an infinite sequence of classes and p. I f we have a sentential
function from the calculus o f classes, then in the above we replace
the sym bol V b y an individual (struotural-descriptive) name o f
this function form ulated in the m etalanguage, bu t ‘p ’ b y a
translation o f the function in to the m etalanguage, where all free
variables vk, vlt eto. are replaced b y corresponding sym bols
‘fk> ‘f i , etc.
W e shall us© a recursive m ethod in order to form ulate a
general definition o f satisfaction o f a sentential function b y a
sequence o f classes, w hich will include all partial definitions o f
this notion as special cases w hich are obtained from the given
scheme in the w ay described above. F or this purpose it will
suffice, bearing in m ind the definition o f sentential function, to
indicate which sequences satisfy the inclusions ik l and then to
specify how the notion we are defining behaves when the three
fundam ental operations o f negation, disjunction, and universal
quantification are perform ed on sentential functions.
The operation o f universal quantification calls for special
consideration. L et x be any sentential function, and assume
that we already know which sequences satisfy the function x.
Considering the m eaning o f the operation o f universal quan­
tification, we shall say that the sequence / satisfies the func­
tion f\kx (where h is a particular natural num ber) only if this
sequence itself satisfies the function x and does not cease to
satisfy it even when the Jdih term o f this sequence varies in any
w ay; in other words, i f every sequence which differs from
the given sequence in at m ost the &th place also satisfies the
function. For exam ple, the function PI2li ,2 satisfied b y those,
and on ly those, sequences / for which the form ula f x £ f 2 holds
w ithout regard to the w ay in which the second term o f this
V m , § 3 CONCEPT OP TRUTH IN FORMALIZED LANGUAGES 193

sequence is allow ed to vary (as is easily seen, this is only possible

when the first term is the null class).
A fter these explanations the understanding o f the follow ing
definition should not be difficult.

D efin itio n 22. The sequence f satisfies the sentential function

x if and only i f f is an infinite sequence of classes and x is a sentential
function and these satisfy one of the following four conditions:
(a) there exist natural numbers k and l such that x = ikj and
fk — fi> (P) there is a sentential function y such that x ~ y and f
does not satisfy the function y\ (y) there are sentential functions
y and z such that x = y + z and f either satisfies y or satisfies z\
(S) there is a natural number k and a sentential function y such that
sc = p j k y and every infinite sequence of classes which differs from
f in at most the k-th place satisfies the function y.1

The follow ing are exam ples o f the application o f the above
definition to concrete sentential functions: the infinite sequ en ce/
satisfies the inclusion ih2 if and only if f x £ / 2, and the function
*2,3+13,a if and only if /a the functions f t 2 *1,2 a:ad fl2*2,s
are satisfied b y those, and on ly those, sequences / in which f t
is the null class and / 3 the universal class (i.e. the class o f all
individuals) respectively; finally, every infinite sequence o f
classes satisfies the function tlfl and no such sequence satisfies
the function t1>2. t1>2.
The concept just defined is o f the greatest im portance for
investigations into the sem antics o f language. W ith its help the
m eaning o f a whole series o f concepts in this field can easily be

1 The normal definition, which is equivalent to the above recursive one, is

as follows (cf. pp. 70, 90, and 93):
The sequence f satisfies the sentential junction x if and only if we have fR x for
every relation R which satisfies the following condition:
For any g and y, in order that gRy it is necessary and sufficient that g is an
infinite sequence of classes, y is a sentential function and either (a) there are
natural numbers k and l such that y = an^ 9k £ 9l or (?) there is a sentential
function z such that y — z and the formula gRz does not hold; or (y) there are
sentential functions z and t such that y = z-j-t and gRz or gRt; or finally (8) there
is a natural number k and a sentential function z such that y — f\k z and hRz
for every infinite sequence h of classes which is distinct from g at the k-th place
at most.
194 CONCEPT OP TRUTH IN FORMALIZED LANGUAGES VIII, § 3

defined, e.g. the concepts o f denotation, definability,1 and truth,

with the last o f which we are especially concerned here.
The concept o f truth is reached in the follow ing way. On the
basis o f D ef. 22 and the intuitive considerations which preceded
it, it is easy to realize that whether or not a given sequence satisfies
a given sentential function depends on ly on those terms o f the
sequence which correspond (in their indices) with the free
variables o f the function. Thus in the extrem e case, when the
function is a sentence, and so contains no free variable (which
is in no way excluded by D ef. 22), the satisfaction o f a function
b y a sequence does n ot depend on the properties o f the terms o f
the sequence at all. Only tw o possibilities then remain: either
every infinite sequence o f classes satisfies a given sentence, or
no sequence satisfies it (cf. the Lemmas A and B given below ).
The sentences o f the first kind, e.g. (J i Li,v are th© ^rue sentences',
those o f the second kind, e.g. can correspondingly be
called the false sentences
1 To say that tha name x denotes a given, object a is the same as to stipulate
that the object a (or every sequence of which a is the corresponding term)
satisfies a sentential function of a particular type. In colloquial language
it would be a function which consists of three parts in the following order:
a variable. Ui-f w^rd 'is ’ and the given name x . As regards the concept
o f definability, I eh&ii try to explain its content- on ly in a particular case.
I f we consider which properties of classes wo regard as definable (in reference
to the system of rhe calculus of classes discussed here), we reach the following
formulations.
We say that the aer>Unt*<*l Junction x defines the property P of classes if and
only if for a natural number k (at) x contains as its only free variable, and
(ft) in orde/; that an infinite sequence f of classes should satisfy x, it is necessary
and- sufficient that should have the property P ; vje say that the property P of
classes is defi+*abi-e . / 2 n d only if there is a sentential function x which defines P .
On the basis of these stipulations it can be shown, for example, that such
properties of classes as emptiness, of containing only one, two, three, etc.,
elements aro definable. On the other hand the property of containing infinitely
many elements >, uu, definable (of. the remarks given bolow in connexion
with Ths. 14-ilV.. h will ahe be seen that with tins interpretation the concept
of definability do. - not depend at all on whether the formalization of the
science investigated oormts of the possibility of constructing definitions. More
exact discussion? of definability will 1)6 found in articles VI and X of the
present volume.

f A method of defining truth which is essentially equivalent to the method

developed in this work, but is based upon a. different idea, has recently been
suggested by J. C. C. McKimey in his paper ‘A na*v definition of truth*,
Syntlidse, vol. 7 (i94S'-v»)? pp. 428-33.
VIII, § 3 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES 195

D efin itio n 23. x is a true sentence— in symbols x e Tr — if and

only if x e S and every infinite sequence of classes satisfies x.1
The question now arises whether this definition, about the
form al correctness o f which there is no doubt, is also m aterially
correct— at least in the sense previously laid down in the con ­
vention T . It can be shown that the answer to this question is
affirm ative: Def. 23 is an adequate definition of truth in the sense
of convention T , since its consequences include all those required
by this convention. Nevertheless it can be seen without diffi­
culty (from the fa ct that the number o f these consequences is
infinite) that the exact and general establishment o f this fact
has no place within the limits o f the considerations so far brought
forward. The p roof w ould require the setting up o f an entirely
new apparatus: in fact it involves the transition to a level one
step higher— to the m eta-m etatheory, which would have to be
preceded b y the form alization o f the m etatheory which form s
the foundation o f our investigations.2 I f we do not wish to
depart from the level o f our previous discussions, only one

1 In the whole of the above construction we could operate with finite

sequences with a variable number of terms instead of with infinite sequences.
It would then be convenient to generalize the concept of finite sequence. In
the usual interpretation of this term a sequence which has an nth term
must also have all terms with indices less than n— we must now relinquish
this postulate and regard any many-one relation as a finite sequence if its
counter domain consists of a finite number of natural numbers distinct from 0.
The modification of the construction would consist in eliminating from the
sequences which satisfy the given sentential function all ‘ superfluous’ terms,
which have no influence on the satisfaction of the function. Thus if vit etc.,
occur as free variables in the function (of course in finite number), only those
terms with the indices k, l, etc., would remain in the sequence which satisfies
this function. For example, those, and only those, sequences / of classes would
satisfy the function ilA which consist of only two terms j %and /< verifying the
formula / , c / 4, The value o f such a modification from the standpoint of
naturalness and conformity with the usual procedure is clear, but when we
come to carry it out exactly certain defects of a logical nature show themselves:
Def. 22 then takes on a more complicated form. Regarding the concept o f
truth, it is to be noted that— according to the above treatment— only one
sequence, namely the ‘ empty* sequence which has no member at all, can
satisfy a sentence, i.e. a function without free variables; we should then have
to call those sentences true which are actually satisfied by the ‘ empty*
sequence. A certain artificiality attaching to this definition will doubtless
displease all those who cure not sufficiently familiar with the specific procedures
which are commonly used in mathematical constructions.
2 See p. 188, footnote.
198 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES VIH, § 3

m ethod, the em pirical m ethod, remains— the verification o f the

properties o f D ef. 23 in a series o f concrete examples.
Consider, for exam ple, the sentence f } i U 2li 2> i.©*
• n x,NYlxmNIxtX*\ A ccording to D ef. 22 the sentential
function i12 is satisfied b y those, and only those, sequences / o f
classes for which f x £ f 2 holds, bu t its negation, i.e. the function
t12, only b y those sequences for whioh f x / 2holds. Consequently
a sequence / satisfies the function if every sequence g
which differs from / in at m ost the 2nd place satisfies the
function t12 and thus verifies the form ula gx $£ g2. Since g1 = f x
and the class gr2 may be quite arbitrary, only those sequences /
satisfy the function P j2i1>2, which are such th a t/x §£ 6 for any class
6. I f we proceed in an analogous w ay, we reach the result that the
sequence / satisfies the function (J 2ti,2>he. *be negation o f the
function f ) 2112, only i f there is a class b for which f x ^ 6 holds.
M oreover, the sentence is only satisfied (by an arbi­
trary sequ en ce/) if there is for an arbitrary class a, a class b for
w hich a £ 6. Finally by applying D ef. 23 we at once obtain one
o f the theorems which were described in the condition (ot) o f the
convention T :

f l i U . 4,2 6 Tr if and only if for all classes a there is a class b

such that a c: 6.

From this we infer w ithout difficulty, by using the known

theorems o f the calculus o f classes, that n.u. is a true
sentence.
W e can proceed in an exactly analogous w ay with every other
sentence o f the language we are considering. I f for such a sen­
tence we construct a corresponding assertion described in the
condition (a) and then apply the m ode o f inference used above,
we can prove w ithout the least difficulty that this assertion is a
consequence o f the definition o f truth which we have adopted.
In m any cases, w ith the help o f only the simplest laws o f logic
(from the dom ain o f the sentential calculus and the calculus of
classes), we can draw definitive conclusions from theorems ob­
tained in this w ay about the truth or falsity o f the sentences in
V III, $ 3 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES 197

question. Thus, for exam ple, H i U a(4 ,2+ 4 ,2)18shown to be a true

and f\x P |2^2 a false sentence. W ith respect to other sentences,
e.g. the sentence run.n. (4 ,2+ * 2,3+ 4 , 1) or negation, the
analogous question cannot be decided (at least so long as we do
not have recourse to the special existential assum ptions o f the
m etatheory, cf. p. 174): D ef. 23 alone gives no general criterion
for the truth o f a sentence.1 Nevertheless, through the theorem s
obtained, the meaning o f the corresponding expressions o f the
type *xe Tr' becom es intelligible and unambiguous. I t should
also be noted that the theorem expressed in the condition (/3)
o f the convention T is also an obvious consequence o f our
definition.
W ith these discussions the reader will doubtless have reached
the subjective conviction that D ef. 23 actually possesses the pro­
perty which it is intended to have: it satisfies all the conditions
o f convention T . In order to fix the conviction o f the material
correctness o f the definition which has been reached in this way,
it is w orth while studying some characteristic general theorems
that can be derived from it. W ith a view to avoiding encum ber­
ing this work with purely deductive m atter, I shall give these
theorems without exact proofs.2

T heorem 1 (The principle o f contradiction). For all sentences

x, either x e Tr or x e Tr.

This is an alm ost im m ediate consequence o f Defs. 22 and 23.

T heorem 2 (The principle o f excluded m iddle). For all

sentences x , either x e T r o r x e Tr.

1 At least when it is regarded from the methodological viewpoint this is

not a defect of the definition in question; in this respect it does not differ
at all from the greater part of the definitions which occur in the deductive
sciences.
a The proofs are based on the general laws of logic, the specific axioms of
the metascience and the definitions of the concepts occurring in the theorems.
In some cases the application of the general properties of the concepts of
consequence, of deductive system, etc., which are given in article V of the
present volume is indicated. We are able to use the results obtained there
because it can easily be shown that the concepts of sentence and consequence
introduced here satisfy all the axioms upon which the above-mentioned work
was based.
198 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES VIII, § 3

In the p roof the follow ing lemma, which follow s from Defs.
11 and 22, plays an essential part:
L em m a A . I f the sequence f satisfies the sentential function x,
and the infinite sequence g of classes satisfies thefollowing condition:
for every k ,fk = gk if vk is a free variable of the function x; then the
sequence g also satisfies the function x.
As an im m ediate consequence o f this lemma and D ef. 12 we
obtain Lemm a B which, in com bination with Defs. 22 and 23
easily leads to Th. 1 :
L emma B. I f x e S and at least one infinite sequence of classes
satisfies the sentence x, then every infinite sequence of classes
satisfies x.
T heorem 3. I f X ^ Tr then Cn(X) c= Tr; thus in particular
Cn(Tr) £ Tr.
This theorem is proved b y strong induction based chiefly
on Defs. 15, 16, 22, and 23; the follow ing simple lemma is also
useful in this connexion:
L em m a C. I f y is a universal quantification of the sentential
function x, then in order that every infinite sequence of classes
should satisfy x , it is necessary and sufficient that every infinite
sequence of classes satisfies y.
The results contained in Ths. 1-3 m ay be summarized in the
follow ing (obtained with the help o f Defs. 18-20):
T heorem 4. The class Tr is a consistent and complete deductive
system.
Th eo rem 5. Every provable sentence is a true sentence, in other
words, Pr s Tr.
This theorem follows im m ediately from D ef. 17. from Th. 3,
and from Lemma D, the p roof o f which (on the basis o f Def. 13
and Lemm a C am ong others) presents no difficulty.
L emma D . Every axiom is a true sentence.
Th. o cannot be inverted:
Theorem 6. There exist true sentences which are not provable,
in other words, Tr Pr.
VIH, § 3 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES 199

This is an im m ediate oonsequence o f Th. 2 and the follow ing

lem m a, the exact p ro o f o f which is n ot quite easy:

L em m a E . Both ("I, n 2 li,s s Pr and P|x H a 4,2 e Pr.1

As a corollary from Ths. 1 ,5 , and 6 ,1 give finally the follow ing
theorem :
T h eo rem 7. The doss Pr is a consistent, but not a complete
deductive system.
In the investigations w hich are in progress at the present day
in the m ethodology o f the deductive sciences (in particular in the
w ork o f the G ottingen school grouped around H ilbert) another
concept o f a relative character plays a m uch greater part than the
absolute concept o f truth and includes it as a special case. This
is the concept o f correct or true sentence in an individual domain
«.*•» B y this is m eant (quite generally and roughly speaking) every
sentence which is true in the usual sense if we restrict the extension
o f the individuals considered to a given class a, or— somewhat
m ore precisely— when we agree to interpret the terms ‘individual ’,
‘class o f individuals’ , etc., as ‘elem ent o f the class a ’ , ‘subclass o f
the class a ’ , e tc., respectively. W here we are dealing with the con­
crete case o f sentences from the calculus o f classes we m ust inter­
pret expressions o f the type ‘J J xp' as ‘/ o r every subclass x of the
1 I f we wish to include the sentence f|i 0**1,* among the acceptable
sentences (as is often the case, cf. p. 180, footnote) we could use here, instead of
Lemma E , the following Lemma E ':
Both n i n ,K .+ t M) e Pr and f l i fli(‘y + ‘i,i) S Pr.
The idea of the proof of both of these lemmas is the same as that of the
proofs of the consistency and incompleteness of the lower functional calculus
which is found in Hilbert, D ., and Ackermann, W . (30), pp. 66-68,
3 The discussion of these relativized notions is not essential for the under­
standing of the main theme of this work and can be omitted by those readers
who are not interested in special studies in the domain of the methodology
of the deductive sciences (only the discussions on pp. 208-9 are in closer
connexion with our main thesis).
8 In this connexion see Hilbert, D ., and Ackermann, W . (30), especially
pp. 72-81, and Bemays, P ., and Schonfinkel, M. (6 a). But it should be
emphasized that the authors mentioned relate this concept not to sentences
but to sentential functions with free variables (because in the language of the
lower functional oaloulus which they use there are no sentences in the strict
sense of the word) and, connected with this, they use the term ‘generally valid *
instead of the term ‘ correct* or ‘ true’ ; cf. the second of the works cited above,
pp. 347-8.
200 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES VIII, § 3

doss a we have p\ and expressions o f the type cIx y ’ as ‘the sub­

class x of the class a is contained in the subclass y of the class a\
W e obtain a precise definition o f this concept b y means o f a
m odification o f Defe. 22 and 23. As derived concepts we in tro­
duce the notion o f a correct sentence in an individual domain with
k elements and the notion o f a correct sentence in every individual
domain. I t is w orthy o f note that— in spite o f the great im por­
tance o f these terms for m etam athem atioal investigations— they
have hitherto been used in a purely intuitive sense w ithout any
attem pt to define their m eaning m ore closely.1
D e fin itio n 24. The sequence f satisfies the sentential function x
in the individual dom ain a if and only if a is a class of individuals,
/ an infinite sequence of subclasses of the class a and x a sentential
function satisfying one of the following four conditions: (a) there
exist natural numbers k and l such that x = ikfl andfk £ fy (/}) there
is a sentential function y such that x = y and the sequencef does not
satisfy y in the individual domain a; (y) there are sentential func­
tions y and z such that x = y-\-z and f satisfies either y or z in the
individual domain a ; (8) there is a natural number k and a sen­
tential function y such that x = HfcJ/ an^ everH infinite sequence
g of subclasses of the class a which differs from f in at most the
k-th place satisfies y in the individual domain a.
D efin itio n 25. x is a correct (true) sentence in the individual
dom ain a if and only if x e S and every infinite sequence of sub­
classes of the class a satisfies the sentence x in the individual
domain a.
26. x is a correct (true) sentence in an individual
D e f i n it io n
dom ain w ith k elem ents '—in symbols x e Ctk— if and only if there
exists a class a such that k is the cardinal number of the class a and x
is a correct sentence in the individual domain a.
1 An exception is furnished by Herbrand, J. (26) in which the author
defines the concept of true sentence in a finite domain (pp. 108-12). A com­
parison of Rerbrand’s definition with Refs. 25 and 26 given in the text will
lead the reader at once to the conclusion that we have to do here with like-
sounding terms rather than with a relationship of content. Nevertheless, it
is possible that with respect to certain concrete deductive sciences, and under
special assumptions for the corresponding metatheory, Herbrand’s oonoept
has the same extension (and also the same importance for metamathematioal
investigations) as a certain special case of the concept introduced in Def. 25.
V m , § 3 CONCEPT OF TROTH IN FORMALIZED LANGUAGES 201

D e fin itio n 21.x is a correct (true) sentence in every individual

dom ain— in symbols x e Ct—if and only if for every class a,
x is a correct sentence in the individual domain a.
I f we drop the form ula *x e S' from D ef. 25, and thereby
m odify the content o f Defs. 26 and 27, we reach concepts o f a
m ore general nature which apply not only to sentences but also to
arbitrary sentential functions.
Exam ples o f the application to concrete sentences o f the
concepts defined w ill be given below . In the interest o f m ore
convenient form ulation o f various properties o f these concepts,
I introduce some further sym bolical abbreviations.
D efin itio n 28. x — ek if and only if

x ~ flfc+i ^ + 1 • r u + i( r w 2 cA+i,ft+2+cfc+i,&+t*,fc+i)*
D efin itio n 29. x = a if and only if

x = r i l ( n 2 tl,2“l“ U 2(t2>l ' €2))-

As is easily seen, the sentential function ek states that the class
denoted by the variable vk consists o f on ly one element ; the
sentence a, which plays a great part in subsequent investiga­
tions, states that every non-null class includes a one-elem ent
class as a part.
D efin itio n 30. x = j8n if and only if either n = 0 and
_ /ti-fl_ n l v
* = ¥= 0 andx = n £ <n+1( 2 « * + 2 2 (*w+i•«*«*))•
* h I k 1

D efin itio n 31. x ~ ynif and only if either n = 0 and x — A »

or n ^ 0 and x = pn^x.pn.
It follow s from these definitions that the sentences fin and yn
(where n is any natural num ber) establish that there are at m ost
n , and exactly n , distinct one-elem ent classes respectively, or,
what am ounts to the same thing, that there are n distinct
individuals.
D efin itio n 32.x is a quantitative sentence (or a sentence about
the num ber o f individuals) if and only if there exists afinite sequence

p of n natural numbers such that either x = Tn Ypk or x = | y „ .

202 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES VIII, § 3

I shall now give a series o f characteristic properties o f the

defined concepts and the m ore im portant connexions which
relate them w ith notions already introduced. This is the place
for some results o f a m ore special nature which are connected
with the particular properties o f the calculus o f classes and
cannot be extended to other disciplines o f related logical struc­
ture (e.g, Ths. 11-13, 24, and 28).
T heorem 8. I f a is a class of individuals and k the cardinal
number of this doss , then in order that x should be a correct sentence
in the individual domain a it is necessary and sufficient that
x e Otk.
The p roof is based on the follow ing lemma (among other
things) which follow s from D ef. 24:
L emma F. Let a and b be two classes o f individuals and It a
relation which satisfies the following conditions: (a) for any / '
and g\ if f ,Rgf then f* is an infinite sequence of subclasses of a,
and g' of subclasses of b ; (jS) if f ' is any infinite sequence of sub­
classes of a, then there is a sequence g' such thatf'Rg'; (y) ifg ' is any
infinite sequence o f subclasses of b, then there is a sequence / ' such
t h a t f ' R g (8) for all g \ f\ g\ k and l, if f'R g ', f'B g ”, and
k and l are natural numbers distinct from 0, thenf’k £ f\ if and only
if gk G g”i. I ffR g and the sequence f satisfies the sentential function
x in the individual domain a, then the sequence g also satisfi.es this
function in the individual domain b.
From this lemma, with the help o f D ef. 25, we easily obtain
Lem m a G which, together w ith D ef. 26, at once gives Tii. 8:
L emma G. I f the classes a and b o f individuals have the same
cardinal number, and x is a correct sentence in the individual
domain a , then x is also a correct sentence in the individual domain 6.
A ccording to Th. 8 (or Lem m a G) the extension o f the con­
cept ‘a sentence which is correct in the individual domain a9
depends entirely on one property o f the class a, nam ely on its
cardinal number. This enables us to neglect in the sequel all
results concerning this concept, because they can be derived
im m ediately from the corresponding theorems relating to the
classes Ctk.
V m , § 3 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES 203

W ith the help o f Defs. 24 and 25 the Ths. 1-6 and Lemmas
A -D can be generalized b y replacing the expressions cinfinite
sequence of classes', ‘the sequence . . . satisfies the sentential func­
tion . . 'true sentence\ and so on, b y cinfinite sequence of sub­
classes of the class a\ cthe sequence . . . salisfies the sentential
function . . . in the individual domain a\ *correct sentence in the
individual domain a\ and so on, respectively. As a consequence o f
Th. 8 the results so obtained can be extended to sentences which
belong to the classes Gtk. in this w ay we reach, am ong other
things, the follow ing generalizations o f Ths. 4 -6 :
T heorem 9, For every cardinal number k the class Ctk is a
consistent and complete deductive system.
T heorem 10. For every cardinal number k we have Pr c Gtk,
but Ctk £ Pr.
In reference to Th. 10 the follow ing problem presents itself:
how is the list o f axiom s in D ef. 13 t-o be com pleted, so that the
class o f all consequences o f this extended class o f axiom s m ay
coincide with the given class CtjJ Ths. 11 and 12 which follow
im m ediately below contain the solution o f this problem and also
prove that— with respect to the language o f the calculus o f classes
— the definition o f a correct sentence in a domain with k elements
(D ef, 28) can be replaced b y another equivalent one which is
analogous to the definition o f provable sentence (Def. 17) and
therefore has a structural character.
T heorem 11. I f k is a natural number, and X the class con­
sisting of all the axioms together with the sentences a and yk, then
Ctk = Cn(X).
T heorem 12. I f k is an infinite cardinal number, and X the
doss consisting of all the, axioms together with the sentence a and all
sentences yl (where l is any natural number), then Ctk - - Cn(X).
The p roof o f these theorems is based chiefly on Ths. 9 and 10
and the three follow ing lemmas:
L emma H . For every cardinal number k we have a e Ctk.
L emma I. I f k is a natural number and l a cardinal number
distinct from k, then yk e Ctk and yk e Gt}, but yTk e Ctk and yk £ Ctv
204 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES V m , § 8

L e m m a .K . I f x e 8 and X is the class consisting of ail the

axioms together with the sentence a, then there is a sentence y which
is equivalent to the sentence x with respect to the doss X and such
that either y is a quantitative sentence, or y e Pr or $ e Pr.
Lemm as H and I are alm ost im m ediately evident, but the
p ro o f o f the very im portant and interesting Lemm a K is rather
difficult.1
B y means o f Th. 9 and Lem m a I it is possible from Th. 12 to
derive the follow ing consequence which com bined with Th. 11
brings out the essential differences existing in the logical struc­
ture o f the classes Ctk according to whether the cardinal num ber k
is finite or infinite:
T heorem 13. I f k is an infinite cardinal number, then there is no
class X which contains only a finite number of sentences which are
not axiotns, and also satisfies the formula
Ctk = Cn(X)*
From Lemma I and Ths. 11 and 12 we easily obtain the follow­
ing consequences:
T heorem 14. I f k is a natural number and l a cardinal number
distinct from k, then Gtk $£ Ctf and Ct{ ^ Ctk.
T heorem 15. I f k and l are infinite cardinal numbers, then
Ctk — Ctf,
Theorem 16. I f k is an infinite cardinal number and x e Ctk,
then there is a natural number l such that x e Ctt (in other words, the
class Ctk is included in the sum, o f all the classes Ctt).
A ccording to Ths. 14-16 (or Lem m a I) there exists for every
natural num ber k a sentence w hich is correct in every domain
1 In its essentials this lemma is contained in the results to be found in
Skolem, Th. (64), pp. 29-37.
a The idea of the proof of this theorem is the same as that of the proofs
of Ths. 24 and 25 in article V of the present volume, pp. 78-9. I f we take
over from the latter Def. 3, p .76, and at the same time extend our present
concept of consequence by adding the words 'or x is an aaciom' to the condition
(a) of Def. 15, then we could derive the following consequence from Ths. 11
and 13:
In order that the class Cik should be an axiomatixable deductive system, it is
necessary and sufficient that h be a natural number.
V m , | 3 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES 205

w ith k elem ents and in no dom ain w ith any other cardinal num ­
ber. On the other hand, every sentence which is correct in one
infinite dom ain is also correct in every other infinite dom ain
(w ithout reference to its cardinal num ber) as well as in certain
finite dom ains. From this we infer that the language in question
allows us to express such a property o f classes o f individuals as
their being oom posed o f exactly k elements, where k is any
natural num ber; bu t we find in this language no means by which
we can distinguish a special kind o f infinity (e.g. denum erability),
and we are unable, either with the help o f a single or o f a finite
num ber o f sentences, to distinguish tw o such properties o f
classes as finiteness and infinity.1
B y means o f Ths. 9, 11, and 12 we can prove

T heorem 17. I f X is a consistent class of sentences which

contains all the axioms together with the sentence a, then there is a
cardinal number k such that X £ Ctk; if X is a complete deductive
system, then X = Gtu.

I f we com bine this theorem with Ths. 11 and 12, we obtain a

structural description o f all com plete deductive system s which
contain all the axiom s and the sentence a. I t should be noted that
the presence o f the sentence a is essential here; the m ultiplicity
o f the system s which do n ot contain this sentence is significantly
greater and their exhaustive description w ould not at all be a
sim ple m atter.2
The rem aining considerations concern sentences which are
correct in every individual dom ain, i.e. belong to the class Gt.

1 These results, as well as Th. 19 given below, we owe to Ldwenheim; of.

Ldwenheim, L . (49) (especially Th. 4, p. 459) and Skolem, Th. (64).
a I have occupied myself in the years 1926-8 with problems of this type,
i.e. with the structural description of all complete systems of a given science,
in application to various elementary deductive sciences (algebra of logic,
arithmetic of real numbers, geometry of straight lines, theory of order, theory
of groups); on the results of these investigations, reports were made in the
seminar exercises in the methodology of the deductive sciences which I con­
ducted in Warsaw University in the years 1927/8 and 1928/9. Cf. Pres-
burger, M. (61) (especially note 4 on p. 95), and X I I , § 5. For a detailed
discussion of certain closely related problems (as well as for further biblio­
graphical references) see also the recent publications of the author, Tarski, A.
(84) and (84 a).
206 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES V ffl, § 3

T heorem 18. In order that x e Ct it is necessary and sufficient

that, for every cardinal number k, x e Ctk (in other words, the
class Ct is the product of all the classes Ctjf).
This theorem , which is an im m ediate consequence o f D ef. 27
andT h. 8, can be essentially sharpened b y means o f Ths. 9 and 16:
T h e o r e m 19. In order that x e Ct it is necessary and sufficient
thatf for every natural number Jcy x e Ctk.
The correctness o f a sentence in all finite domains thus entails
its correctness in every individual dom ain.
The follow ing tw o corollaries are derivable from Ths. 9, 14,
and 18:
T h e o r e m 20. For every cardinal number kwehave Ct £ Ctk, but
Ctk £ Ct.
T heorem 21. The class Ct is a consistent but not a complete
deductive -system.
T h eo r em 22. Pr c Ct, but Ct & P r.
This theorem follow s from Ths. 10 and 18 and Lemm a L :
L em m a L .ote Ct but a e Pr.
That <x e Ct follow s at once from Lemma H and Th. 18. The
exact p roof o f the second part o f the lemma is considerably more
difficult.
T heorem 23. I f x is a quantitative sentence then x e Ct,
The proof, which Is based on Lemma I, Th. 18, and D ef. 32,
offers no difficulties.
T heorem 24. i f X is the class consisting of all the axioms
together with the sentence a , then Ct Cn(X).
This theorem is m ost easily proved with the help o f Ths. 11,
12, and 18. B y using Lemma K we obtain from it at once:
T heorem 25. I f x e S , x e Ct and x e Ct, then there is a quanti­
tative sentence y y which is equivalent to the sentence x with respect
to the class Ct.
B v reference to Lem m a L and Th. 24 we notice that we have
the follow ing situation: the concept o f a sentence which is cor­
rect in every individual dom ain has a larger extension than the
V m , § 3 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES 207

concept o f provable sentence, since the sentence a belongs to the

extension o f the first concept but not to that o f the second. B ut
i f we increase the system o f axiom s b y just this single sentence «,
the tw o concepts becom e identical in extension. Because it
seems to me desirable that, w ith respect to the calculus o f
classes, the concepts o f theorem and o f correct sentence in
every individual dom ain should not be distinct in extension,1
I w ould advocate the inclusion o f the sentence a am ong the
axiom s o f this science.
The problem still remains o f clarifying the relation o f the
absolute concept o f truth defined in D ef. 23 to the concepts we
have just investigated.
I f we com pare D efs. 22 and 23 with Defs. 24 and 25 and apply
Th. 8, we easily obtain the follow ing result:
T h e o r e m 26. I f a is the class of aU individuals then x e T r if
and only if x is a correct sentence in the domain a\ thus if k is the
cardinal number of the cla^s a, then Tr = Ctk.
As an im m ediate consequence o f Ths. 20 and 26 we have:
T heorem 27. Ct c Tr, but Tr £ Ct.
I f we bring together Ths. 26 and 14 or Ths. 11 and 12, we
reach the conclusion that those assumptions o f the m etatheory
which determine the cardinal num ber o f the class o f all indivi­
duals (and which do not intervene in the p roof o f Th. 26 itself)
exert an essential influence on the extension o f the term 'true
sentence’ . The extension o f this term is different according to
whether that class is finite or infinite. In the first case the
extension even depends on how big the cardinal number o f this
class is.

1 Tills tendency will be discussed in the next paragraph. It should be

mentioned that Schroder, although beginning with other ideas, has made the
suggestion of completing the sys tem of hypotheses of the calculus of classes with
the sentence a (and even with still other sentences which, however, as can easily
be shown, follow in a simple way from the sentence a); cf. Schrader, E. (62),
vol. 2, Part 1, pp. 318-49. In this connexion I may remark that it seems to
me that the inclusion of the sentence a in the ‘ formal* system of the algebra
of logic (of which the calculus of classes is an interpretation) would not be
useful, for many interpretations of this system are known in which the sentence
in question is not satisfied.
208 CONCEPT OP TRUTH IN FORMALIZED LANGUAGES V m , § 3

Because we can show, on the basis o f the system o f assum p­

tions here adopted, that the class o f all individuals is infinite,
T h. 26 in com bination w ith T h. 12 makes a structural character­
ization o f true sentences possible:

T heorem 28, In order that x e Tr, it is necessary and sufficient

that x is a consequence of the class which consists o f all the axioms
together with the sentence a. and aU sentences y„ where l is any
natural number.

This sentence could, in its form , obviously be regarded as

a definition o f true sentence. I t w ould then be a purely struc­
tural definition, com pletely analogous to D ef. 17 o f provable
theorem . B u t it m ust be strongly em phasized that the pos­
sibility o f constructing a definition o f such a kind is purely
aooidental. W e 6we it to the speoifio peculiarities o f the science
in question (to those peculiarities which, am ong others, have
been expressed in Lem m a K , w hich is the m ost essential prem iss
in the p ro o f o f Ths. 12 and 28) as well as— in some degree— to the
strong existential assum ptions adopted in the m etatheory.
On the other hand— in oontrast to the original definition— we
have here no general m ethod o f construction which could be
applied to other deductive sciences.
I t is w orth noticing that b y analysing the p ro o f o f Th. 28 and
o f the lemmas from w hich this theorem follow s, we can obtain a
general structural criterion o f truth for all sentences o f the
language investigated. From Th. 28 such a criterion for quantita­
tive sentences is easily derivable, and the p ro o f o f Lem m a K
allows us effectively to correlate w ith every sentence o f the
language a sentence which is equivalent to it and which, i f it is
n ot quantitative, is m anifestly true or m anifestly false. An
analogous rem ark holds for the concept o f correctness in a given,
or in every, individual dom ain.
Summarizing the most important results obtained in this
section we can say:

We have succeeded in doing for the language of the calculus of

classes what we tried in vain to do for colloquial language: namely
V m , | 3 CONCEPT OP TRUTH IN FORMALIZED LANGUAGES 200

to construct a form ally correct and materially adequate semantical

definition of the expression *true sentence\
Moreover, by making use of the special peculiarities of the
oaloulus of classes, we have been able to transform this definition
into an equivalent structural definition which even yields a
general criterion of truth for the sentences of the language
of this oaloulus.

§4. T he Concept op T rue Sentence in L anguages op

F inite Order
The methods of constructionwhichI have used in the previous
section for the investigation of the language of the oaloulus of
classes can be applied, without very important ohanges, to
many other formalized languages, even to those with a consider­
ably more complicated logical structure. In the following pages
the generality of these methods will be emphasized, the limits
of their applicability will be determined, and the modifications
which they undergo in their various concrete applications will be
briefly described.
It is by no means my intention, in these investigations, to con­
sider all languages that can conceivably be imagined, or which
any one at any time could or might wish to construct; such an
attempt would be condemned to failure from the start. In what
I shall say here I shall consider exclusively languages ofthe same
structure as those which are known to us at the present day (in
the perhaps unfounded conviction that they will form in the
future, as they have done hitherto, a sufficient basis for the
foundation of the whole of deduotive knowledge). And even
these languages show suoh great differences in their construction
that their investigation in a perfectly general, but at the same
time precise, way must encounter serious difficulties. These
differences are, of course, rather of a ‘calligraphioaT nature. In
some languages, for example, only constants and variables
ocour, in others it is not possible to avoid the use of so-oalled
technical signs (brackets, points, and so on). In some languages
symbols of an exactly specified form are used as variables, so
that the form of the variables depends on the part they play
210 CONCEPT OP TRUTH IN FORMALIZED LANGUAGES VTU, § 4

and their significance. In others quite arbitrary symbols may

be used as variables, so long as they are distinguished by their
formfrom the constants. In some languages every expressionis a
system of linearly ordered signs, i.e. signs following one another
in a line, but in others the signs may lie at different levels, not
only alongside but also below one another. This calligraphy of
the language nevertheless exerts a fairly strong influence on the
form of the constructions in the domain of the metalanguage, as
will doubtless be seen from a brief survey of the preceding para­
graphs.1 For those reasons alone the following exposition will
have the nature of a sketch; wherever it takes a more precise
form, it is dealing with concretely described languages which are
constructed in the same way as the language of the calculus of
classes (i.e. languages without technical signs, with variables
of an exactly specified form, with linear arrangement of the signs
in every expression and so on).2
Before we approach our principal task—the construction of
the definition of true sentence—we must undertake, in every
concrete case, the construction of a corresponding metalanguage
and the establishment of the metatheory which forms the proper
field of investigation. A metalanguage which meets our require­
ments must contain three groups of primitive expressions:
(1) expressions of a general logical kind; (2) expressions having
the same meaning as all the constants of the language to be dis­
cussed or which suffice for the definition of such expressions
(taking as a basis the rules of definition adopted in the meta-
1 Cf., for example, p. 191, footnote.
9 In order to give the following exposition a completely precise, concrete,
and also sufficiently general form, it would suffice if we chose, as the object
of investigation,, the language of some one complete system of mathematical
logic. Such a language can be regarded as a universal language in the sense
that all other formalized languages— apart from ‘ caUigraphioal’ differences—
are either fragments of it, or can be obtained from it or from its fragments
by adding certain constants, provided that the semantical categories of these
constants (cf. below, pp. 215 ff.) are already represented by certain expres­
sions o f the given language. The presence or absence of such constants exerts,
as we shall show, only a minimal influence on the solution of the problem in
which we are interested. As such a language we could choose the language
of the general theory of sets which will be discussed in § 5, and which might
be enriched by means of variables representing the names of two- and of
many-termed relations (of arbitrary semantical categories).
V III, § 4 CONCEPT OP TRUTH IN FORMALIZED LANGUAGES 211

th eory); (3) expressions o f the structural-descriptive type which

denote single signs and expressions o f the language considered,
whole classes and sequences o f such expressions or, finally, the
relations existing between them . That the expressions o f the
first group are indispensable is evident. The expressions o f
the second group enable us to translate every concrete sentence
or, m ore generally, every m eaningful expression o f the language
in to the metalanguage, and those o f the third group provide for
the assignment o f an individual name to every such expression.
These last tw o circumstances taken together play an essential part
in the final form ulation o f the desired definition. Corresponding
to the three groups o f prim itive expressions, the full axiom system
o f the m etatheory includes throe groups o f sentences: (1) axiom s
o f a general logical kind; (2) axiom s which have the same meaning
as the axiom s o f the science under investigation or are logically
stronger than them , but which in any case suffice (on the basis
o f the rules o f inference adopted) for the establishment o f ail
sentences having the same meaning as the theorems o f the
science investigated;1 finally, (3) axiom s which determine the
fundamental properties o f the prim itive concepts o f a structural-
descriptive type. The prim itive expressions and axiom s o f the
first group (as well as the rules o f definition and inference) m ay
be taken from any sufficiently developed system o f m athem atical
logic; the expressions and axiom s o f the second group naturally
depend on the special peculiarities o f the language investigated;
for the third group suitable exam ples are provided in the presen­
tation o f § 2. It is to be noted that the two first groups o f prim itive
1 It ha3 already been mentioned (p. 167) that we are here interested ex­
clusively in those deductive sciences which are not 'formal7 in a quite special
meaning of this word. 1 have, moreover, brought forward various conditions—
of an intuitive not, a formal nature —which are satisfied by the sciences here
investigated: a strictly determinate and understandable meaning of the
constants, the certainty of the axioms, the reliability of the rules of inference.
An external characteristic of this standpoint is jusxi the fact that, among the
primitive expressions and the axioms of the metatheory the expressions and
axioms of the second group occur. For as soon as we regard certain expressions
as intelligible, or believe in the truth of certain sentences, no obstacle exists
to using them as the need arises. This applies also to the rules of inference
which wo may, if need be, transfer from the theory to the metatheory. In
the sequel we shad convince ourselves that this need actually exists in the
cases ’dv'M
212 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES VIEt, § i

expressions and axioms partly overlap one another, and in those

oasesin-whichmathematioallogio, ora fragmentofit, is the object
of investigation (as is the case with the caloulus of olasBes), they
even combine to form one group.
The establishment of the metatheory having been completed,
our next task is to distinguish from the totality of expressions
of the language the especially important oategory of sentential
functions and in particular of sentences. The expressions of the
language investigated consist of constants and variables. Among
the constants, which are usually finite in number, we find, as a
rule, certain signs belonging to the sentential caloulus and the
predicate caloulus: for example the signs of negation, logical
sum, logical product, implication, equivalence, as well as the
universal and existential quantifiers, which we have already
met in § 2. In addition to these we sometimes find other signs
whioh are connected with the individual peculiarities of the
language and denote concrete individuals, classes, or relations;
suoh, for example, as the inclusion sign of the language of the
caloulus of classes, whioh denotes a particular relation between
classes of individuals. Usually there are infinitely many vari­
ables. According to their form, and the interpretation of the
language, they represent names of individuals, classes, or rela­
tions (sometimes there are also variables which represent sen­
tences, i.e. the so-called sentential variables).1 Among the
expressions whioh are formed from the signs of both kinds, we
distinguish first of all the primitive sentential functions, corre­
sponding to the inclusions ikj of the caloulus of classes. The
exact description of the form of the sentential functions and the
specification of their intuitive sense will depend upon the special
peculiarities of the language in question. In any case they are
certain complexes of constants which are names of individuals,
classes, or relations, and of variables whioh represent these
1 In many languages various other categories of constants and variables
occur, e.g. name-forming functors which, in combination with variables, form
composite expressions by which names of individuals, classes, and relations
are represented (e.g. the word ‘ father* in colloquial language, or the sign of
complementation in the complete language of the calculus of classes— cf. p. 161,
note 1, and p. 168, note 3 The languages considered in the present article contain
no signs and expressions of this kind.
V in , § 4 CONCEPT OF TROTH IN FORMALIZED LANGUAGES 213

names. The first sign o f such a com plex is always the name o f a
class or a relation or a corresponding variable, and is called a
(sentence forming) functor of the given primitive sentential func­
tion ? the remaining signs are called arguments, nam ely 1st,
2nd,..., ifcth argument— according to the place they occu py.
F or every constant and variable o f the language studied— with
the exception o f the constants o f the sentential calculus and the
universal and existential quantifiers— a prim itive function can
be constructed which contains this sign (the sentential variables,
even when they appear in the language, do not occu r in the
prim itive functions as functors or arguments, bu t each is re­
garded as an independent prim itive function). N ext we introduce
the fundamental operations on expressions b y means o f which
com posite expressions are form ed from sim pler ones. In addition
to the operations o f negation, logical addition and universal
quantification, which we have m et with in § 2 (Defs. 2, 3, and 6),
we consider here other analogously defined operations, such
as logical m ultiplication, form ation o f im plications and equiva­
lences, as well as existential quantification. Each o f these
operations consists in putting in front o f the expression con­
sidered, or in front o f tw o successive expressions (according to
thekind o f operation), either one o f the constants o f the sentential
calculus w hich belongs to the language, or one o f the tw o quanti­
fiers together w ith the variables im m ediately follow ing it. The
expressions which we obtain from the prim itive functions b y
applying to them any num ber o f times and in any order any o f

1 Thus sentence-forming functors which have names as arguments are here

identified with the names of classes or relations (in fact the one-argument
functors with names of classes and the rest with names of two- or many-
termed relations). This interpretation seems artificial with the interpretation
of the term ‘ functor’ which was given by some examples on p. 161, note 1;
in any case it certainly does not agree with the spirit and formal structure
of the language of everyday life. Without going into details, it seems to me
for various reasons to bo neither necessary nor useful to distinguish between
these two categories of expressions (i.e. sentence-forming functors and names
of classes or relations). Moreover, the whole question is rather of a termino­
logical nature and is without influence on subsequent developments. W e may
either regard the definition of functor given in the text as purely formal and
disregard the current interpretation of the term, or so extend the interpreta­
tion of terms like ‘ name of a class’, ‘ name of a relation’ that we include
expressions which are not names in the usual sense.
214 CONCEPT OF TROTH IN FORMALIZED LANGUAGES VHI, § 4

the fundam ental operations, we call sentential functions. Am ong

the variables which occur in a given sentential function we can
distinguish— by means o f recursive definitions—free and bound
variables. Sentential functions w ithout free variables are called
sentences (cf. Defs. 10-12 in § 2).
N ext we define yet other concepts which are closely con­
nected w ith the deductive character o f the science under in­
vestigation, nam ely the concepts o f axiom, consequence, and
theorem. Am ong the axiom s we include as a rule certain logical
sentences which are constructed in a manner similar to that used
for the first kind o f axiom s o f the calculus o f classes (cf. § 2,
D ef. 13). M oreover the definition o f axiom depends w holly on
the individual peculiarities o f the science investigated, som e­
tim es even on accidental factors which are connected with its
historical developm ent. In the definition o f the concept o f
consequence we follow — mutatis mutandis— the pattern o f § 2.
The operations b y means o f which we form the, consequences o f a
given class o f sentences differ in no essential points from the
operations which were given in D ef. 15. The consequences o f
the axiom s are called 'provable sentences or theorems.
A fter this prelim inary w ork we turn now to our principal task
— the construction o f a correct definition o f true sentence. As we
saw in § 3, the m ethod o f construction available to us presupposes
first a definition o f another concept o f a m ore general kind which
is o f fundam ental im portance for investigations in the semantics
o f language. I mean the notion o f the satisfaction o f a sentential
function by a sequence of objects. In the same section I have
attem pted to clarif y the custom ary meaning o f this expression
in its ordinary usage. I have pointed out that in drawing u p a
correct definition o f the concept o f satisfaction use can be made
o f recursive definition. F or this purpose it suffices— recalling
the recursive definition o f sentential function and bearing in
m ind the intuitive sense o f the prim itive sentential functions
and the fundam ental operations on expressions— to establish
tw o facts: (1) which sequences satisfy the fundam ental func­
tions, and (2) how the concept o f satisfaction behaves under
the application o f any o f the fundam ental operations (or to put
V III, § 4 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES 215

it m ore exactly: which sequences satisfy the sentential functions

which are obtained from given sentential functions b y means
o f one o f the fundam ental operations, assuming that it has
already been established which sequences satisfy the sentential
functions to which the operation is applied). As soon as we have
succeeded in m aking precise the sense o f this concept o f satis­
faction, the definition o f truth presents no further difficulty:
the true sentences m ay be defined as those sentences which are
satisfied b y an arbitrary sequence o f objects.
In carrying out the plan just sketched in connexion with various
concrete languages we nevertheless m eet with obstacles o f a fun­
dam ental kind; in fact, just at the point where we try finally to
form ulate the correct definition o f the concept o f satisfaction. In
order to make clear the nature o f these difficulties a concept
must first be discussed which we have not hitherto had an oppor­
tunity o f introducing, nam ely the concept o f semantical category.
This concept, which we owe to E . Husserl, was introduced
into investigations on the foundations o f the deductive sciences
by Le^niewski. From the form al point o f view this concept plays
a part in the construction o f a science which is analogous to that
played b y the notion o f type in the system Principia Mathematica
o f W hitehead and Russell. B ut. so far as its origin and content
are concerned, it corresponds (approxim ately) rather to the
well-known concept o f part o f speech from the grammar o f
colloquial language. W hilst the theory o f types was thought
o f chiefly as a kind o f prophylactic to guard the deductive
sciences against possible antinom ies, the theory o f seman­
tical categories penetrates so deeply into our fundamental
intuitions regarding the meaningfulness o f expressions, that it is
scarcely possible to im agine a scientific language in which the
sentences have a clear intuitive meaning but the structure o f
which cannot be brought into harm ony w ith the above theory.1
1 Cf. Lesniewski, S. (40), especially pp. 14 and 68; Ajdukiewioz, K . (3),
pp. 9 and 148. From the formal point of view the theory of semantical
categories is rather remote from the original theory of types of Whitehead, A. N .,
and Russell, B. A. W . (90), vol. 1, pp. 37 fit.; it differs less from the simplified
theory of types (cf. Chwistek, L . (12), pp. 12-14; Carnap, R . (8), pp. 19-22)
and is an extension of the latter. Regarding the views expressed in the last
paragraph of the text, compare the Postscript to this article (p. 268).
216 CONCEPT OF TROTH IN FORMALIZED LANGUAGES VIH, | i

For reasons mentioned at the beginning of this section we

cannot offer here a precise structural definition of semantical
category and will content ourselves with the following approxi­
mate formulation: two expressions belong to the same semantical
category if (l) there is a sentential function which contains one of
theseexpressions, andif (2) no sentential function which contains
one of these expressions ceases to be a sentential function if
this expression is replaced in it by the other. It follows from
this that the relation of belonging to the same category is re­
flexive, symmetrical, and transitive. By applying the principle
of abstraction,1 all the expressions of the language which are
parts of sentential functions can be divided into mutually
exclusive classes, for two expressions are put into one and the
same class if and only if they belong to the same semantical
category, and each of these classes iB called a semantical cate­
gory. Among the simplest examples of semantical categories it
suffices to mention the category of the sentential functions,
further the categories which include respectively the names of
individuals, of classes of individuals, of two-termed relations
between individuals, and so on. Variables (or expressions "with
variables) which represent names of the given categories like­
wise belong to the same category.
In connexion with the definition of semantical category the
following question arises: in order to establish the fact that two
given expressions belong to one and the same semantical cate­
gory, is it necessary to consider all possible sentential functions
which contain one of the given expressions and to investigate
their behaviour when one of these expressions is replaced by the
other, or does it suffice to make this observation in some or even
in only one case ? From the standpoint of the ordinary usage oi
language the second possibility seems much more natural; in
order that two expressions shall belong to the same semantical
category, it suffices if there exists one function which contains
one of these expressions and which remains a function whenthis
expression is replaced by the other. This principle, which canbe
called the first principle o f the theory o f semantical categories, is
1 Cf. Carnap, R . (8), pp. 48-60.
VIH , § 4 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES 217

taken strictly as a basis for £be construction o f the form alized

languages here investigated.1 I t is especially taken into account
in the definition o f the concept o f sentential function. I t also
exerts an essential influence on the definition o f the operation o f
substitution, i.e. one o f those operations w ith the help o f which
we form the consequences o f a class o f sentences. F or if we wish
that this operation, when carried out on any sentence, should
always give a new sentence as a result, we m ust restrict ourselves
to substituting for the variables only those expressions which
belong to the same sem antical category as the corresponding
variables.128
* Closely connected w ith this principle is a general
law concerning the sem antical categories o f sentence-form ing
functors: the functors o f tw o prim itive sentential functions
belong to the same category i f and only if the number o f argu­
ments in the tw o functions is the same, and if any tw o arguments
which occu p y corresponding places in the tw o functions also
belong to the same category. From this it follow s that, in par­
ticular, no sign can be sim ultaneously a functor o f tw o functions
w hich possess a different num ber o f arguments, or o f tw o such
functions (even if they possess the same number o f arguments)

1 When applied to concrete languages the formulations given in the text—

both the definition of semantical oategory and the above-mentioned principle—
require various corrections and supplementations. They are in any case too
general, for they also include expressions to which we do not usually ascribe
independent meaning, and which we often include in the same semantical
categories to which meaningful expressions belong (for example, in the language
of the calculus of classes, the expressions <N \ ‘U x ,\ and ‘ A I x , x „ * would
belong to the same semantical category); in the case of these meaningless
expressions, it can easily be shown that even the first principle of semantical
categories loses its validity. This fact is of no essential importance for our
investigations, for we shall apply the concept of semantical category, not to
composite expressions, but exclusively to variables. On the other hand, the
examples which we shall encounter in the sequel show that the above formula­
tions admit of very far-reaching simplifications in concrete cases. Thanks to
a suitable choice of the signs used in the construction of the expressions of
the language, the mere shape of the sign (and even of the composite expression)
decides to which category it belongs. Consequently it is possible that in
methodological and semantical investigations concerning a concrete lan­
guage, the concept of semantical category does not explicitly occur at
All
8 In the language of the calculus of classes, and in the languages which
I shall describe in more detail in the sequel, such expressions can only be
other variables; this explains the formulation of Def. 14 in $ 2.
218 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES VIII, § 4

in which tw o arguments which occu py corresponding places

belong to different categories.
W e require a classification o f the sem antical categories; to
every category a particular natural number is assigned called the
order of the category. This order is also assigned to all expressions
which belong to this category.1 The meaning o f this term can
be determ ined recursively. F or this purpose we adopt the
follow ing convention (in which we have in m ind only those
languages which we shall deal w ith here and we take account
only o f the sem antical categories o f the variables): (1) the
1st order is assigned only to the names o f individuals and to
the variables representing them ; (2) am ong expressions o f the
1th order, where n is any natural number, we include the
functors o f all those prim itive functions all o f whose argu­
ments are o f at m ost the nth order, where at least one o f them
m ust be o f exactly the nth order. Thanks to the above con­
vention all expressions w hich belong to a given semantical
category have the same order assigned to them , which is therefore
called the order o f that category.2 On the other hand the category
1 Cf. Carnap, R , (8), pp. 31-32.
* This classification, by no means includes all semantical categories which
are to be found in formalized languages. For example, it does not include
sentential variables and functors with sentences as arguments— i.e. signs
which occur in the sentential calculus— neither does it include functors which,
together with the corresponding arguments, form expressions which belong
to one of the categories distinct from sentential functions, such as the name­
forming functors mentioned on p. 213, footnote.
In view of this, the definition of order given in the text could be widened
in the following way: (1) to the 1st order belong sentences, names of individuals
and expressions representing them; (2) among expressions of the n -f lth order
we include those functors with an arbitrary number of arguments of order
<: n, which together with these arguments form expressions of order < n,
but are not themselves expressions of the nth order. Even this definition
does not yet cover all meaningful expressions which occur in the deductive
sciences. No signs which ‘ bind* variables fall under this definition (thus such
signs as the universal and existential quantifiers, the signs ‘ E* and ‘ if* of the
theory of sets and analysis or the sign of integration), signs which— in contrast
to the functors— can be called operators. (von Neumann speaks of abstractions
in this connexion, see Neumann, J. v. (54).) On the other hand the latter
classification is completely adapted to the system invented by Le&oiewski
and sketched by him in Le&iiewski (46) and (47). This system contains no
operators except the universal quantifier which belongs to no semantical
category. I may add that, in my view, the lack of operators in Le^niewski’s
system constitutes a deficiency which restricts its .cuniversal* character (in
the sense of p. 210, note 2) to a certain degree*
VIII, § 4 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES 219

is b y no means specified b y the order: every natural number

whioh is greater than 1 can be the order o f m any different
categories. Thus, for exam ple, both the names o f classes o f
individuals and the names o f tw o-, three-, and m any-term ed
relations between individuals are expressions o f the 2nd
order.
It is desirable to classify the sentential functions o f the lan­
guage according to the semantical categories o f the free variables
occurring in them . W e shall say o f tw o functions that they possess
the same semantical type i f the number o f free variables o f every
sem antical oategory in the tw o functions is the same (or, in other
words, if the free variables o f the one function can be put into
one-one correspondence w ith the free variables o f the other in
such a w ay that to every variable a variable o f the same category
corresponds). The class o f all sentential functions w hich possess
the same type as a given function we can call a semantical
type.
W e sometimes use the term 'sem antical category ’ in a deriva­
tive sense, b y applying it, not to the expressions o f the language,
but to the objects w hich they denote. Such ‘hypostatizations’
are not quite correct from a logical standpoint, but they sim plify
the form ulation o f m any ideas. W e say, for exam ple, that all
individuals belong to the same semantical category, but that no
classes or relations belong to this category. From the general law
stated above concerning sentence-form ing functors we conclude
that tw o classes belong to the same category if and only if all
their elements belong to one and the same category. Two tw o-
term ed relations belong to the same category if and only if their
dom ains belong to the same category and their counter domains
belong to the same category. In particular, tw o sequences belong
to the same category if and only if all their terms belong to the
game category. A class and a relation, or tw o relations having
different numbers o f terms never belong to the same category.
It also follow s that there can be no class whose elements belong to
tw o or m ore semantical categories; in an analogous way there can
be no sequence whose terms belong to distinct semantical cate­
gories. Individuals are sometimes called objects o f the 1st order,
220 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES V m , § 4

classes o f individuals and relations between individuals objects

o f the 2nd order, and so on.
The language o f a com plete system o f logic should contain—
actually or potentially— all possible sem antical categories which
occur in the languages o f the deductive sciences. Just this
fa ct gives to the language m entioned a certain ‘universal’
character, and it is one o f the factors to which logic owes its
fundam ental im portance for the whole o f deductive knowledge.
In various fragm entary system s o f logic, as well as in other
deductive sciences, the m ultiplicity o f the semantical categories
m ay undergo a significant restriction in both their number
and their order. As we shall see, the degree o f difficulty which
we have to overcom e in the construction o f a correct definition
o f truth for a given concrete language, depends in the first
place on this m ultiplicity o f the semantical categories appear­
ing in the language, or, m ore exactly, on whether the ex­
pressions and especially the variables o f the language belong
to a finite or an infinite num ber o f categories, and in the latter
case on whether the orders o f all these categories are bounded
above or not. From this point o f view we can distinguish
four kinds o f languages: (1) languages in which all the variables
belong to one and the same sem antical category; (2) languages
in which the number o f categories in which the variables are
included is greater than 1 but finite; (3) languages in which
the variables belong to infinitely m any different categories
but the order o f these variables does not exceed a previously
given natural number n\ and finally (4) languages which
contain variables o f arbitrarily high order. W e shall call the
languages o f the first three kinds languages of finite order, in
contrast to languages o f the fourth kind, the languages of infinite
order. The languages o f finite order could be further divided
into languages o f the 1st, 2nd order, and so on, according to
the highest order o f the variables occurring in the language. By
way o f supplem entation o f the sketch given at the beginning of
this section o f the construction o f a m etatheory, it must be
noted here that the metalanguage, on the basis o f which the
investigation is conducted, is to be furnished with at least all the
V m , § 4 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES 221

sem antical categories w hich are represented in the language

studied. This is necessary if it is to be possible to translate any
expression o f the language into the m etalanguage,1
From the point o f view o f their logical structure the languages
o f the 1st kind are obviously the sim plest. The language o f the
oalculus o f classes is a typical exam ple. W e have seen in § 3
th at for this language the definition o f the satisfaction o f a
sentential function b y a sequence o f objects, and hence the
definition o f true sentence, presents no great difficulties' The
m ethod o f construction sketched there can be applied as a whole
to other languages o f the 1st order. I t is clear that in doing this
certain small, deviations in detail m ay occur. Am ong other things
it m ay be necessary to operate n ot with sequences o f classes bu t
w ith sequences o f other kinds, e.g. w ith sequences o f individuals
or relations, according to the intended interpretation and the
sem antical categories o f the variables occurring in the language.*
A particularly sim ple exam ple o f a language o f the 1st kind
w hich is w orthy o f attention is the language o f the ordinary
sentential calculus enlarged b y the introduction o f the uni­
versal and existential quantifiers. The sim plicity o f this language
lies, am ong other things, in the fa ct that the concept o f variable
coincides w ith that o f prim itive sentential function. In the
m etatheory o f the sentential calculus tw o different definitions
can be given o f provable theorem , the equivalence o f which is in
no w ay evident: the one is based on the concept o f consequence
andis analogous to D efs. 15-17 o f§ 2, the second id connected w ith
the concept o f the tw o-valued m atrix. B y virtue o f this second
definition we can easily determ ine whether any sentence is
provable provided its structure is known.8 I f we now construct
for th is language a definition o f true sentence strictly according

1 Here— mutatia mutandis— the remarks of p. 211, footnote, also apply.

1 Certain complications, which I shall not discuss here, arise if in addition
to variables, composite expressions of the same semantical category also occur
in the language investigated; the complete language of the oalculus of classes
which was mentioned on p. 168, note 3, will serve as an example, or the lan­
guage of a system of arithmetic investigated in Preeburger, M. (61) (cf. also
p. 212, footnote).
* Cf. Hilbert, D ., and Ackermann, W . (30), pp. 84-85; Lukasiewicz, J. (51),
pp. 164 ft.; IV , § 4.
222 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES VH I, § 4

to the pattern given in § 3, we can easily convince ourselves that

it represents a simple transform ation o f the second o f these
definitions o f provable sentence, and thus the tw o terms 'p rov­
able theorem ' and ‘true sentence’ in this case have the same
extension. This fa ct provides us, am ong other things, w ith a
general structural criterion for the truth o f the sentences o f this
language. The m ethod o f construction laid down in the present
w ork could thus be regarded, in a certain sense, as a generalization
o f the m atrix m ethod fam iliar in investigations on the sentential
calculus.
Serious difficulties only arise when we consider languages o f
m ore com plicated structure, e.g. languages o f the 2nd, 3rd,
and 4th kinds. W e m ust now analyse these difficulties and de­
scribe the m ethods which enable us at least partially to overcom e
them . In order to make the exposition as dear and preoise as
possible I shall discuss in som ewhat greater detail som e ooncrete
form alized languages, one o f each kind. I shall try to choose
exam ples which are as simple as possible, are free from all less
essential, subordinate com plications, and are at the same tim e
sufficiently typical to exhibit the difficulties m entioned to the
fullest extent and in the m ost striking form .
The language o f the logic of two-termed relations will serve as an
exam ple o f a language the 2nd order.1 The only constants o f
this language are: the sign o f negation ‘N ’ , the sign o f logical
sum ‘A ’ and the universal quantifier ‘JJ\ As variables we can
use the signs ‘z/, (x„\ ‘x,,,’, . . . and ‘X ,’ , ‘X / , ‘X » \ ___ The sign
com posed o f the sym bol ‘z ’ and o f k small additional strokes is
called the k-th variable of the 1st order, and is denoted b y the
sym bol ‘vk . The sign analogously constructed with the sym bol
‘X ’ is called the k-th variable of the 2nd order, sym bolically
The variables o f the 1st order represent names o f individuals,
those o f the 2nd order names o f tw o-term ed relations between
individuals. From the m aterial and also— in agreement with
the further description o f the language— from the formal

1 This is a fragment of the language of the algebra of relations, the founda­

tions of which are given in Schroder, E. (62), vol. 3— a fragment which never­
theless suffices to express every idea which can be formulated in this language.
V III, $ 4 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES 223

poin t o f view , the signs and % ’ belong to tw o distinct

sem antical categories. Expressions o f the form ‘X yz’ are
regarded as prim itive sentential functions, where in the place
o f ‘X ’ any variable o f the 2nd order, and in the place o f ‘y’
and V any variables o f the 1st order m ay appear. These expres­
sions are read: ‘the individual y stands in the relation X to the
individual z’ and they are denoted— according to the form o f the
variables— b y the sym bols ‘pkj,m’ - 3fy the use o f the sign
from §2 we specify that pkl<m — (Yk^ vd ^ vm. The definitions
o f the fundam ental operations on expressions, as well as those
o f sentential function, sentence, consequence, provable sen­
tence, and so on, are all quite analogous to the definitions o f
§ 2. B ut it m ust always be borne in m ind that in this language
tw o distinct categories o f variables appear and that the expres­
sions Pkjjn play the part o f the inclusions tjy. In connexion with
the first o f these facts we have to consider not one operation o f
quantification (Defe. 6 and 9) but two analogous operations:
w ith respect to a variable o f the 1st order as well as with respect
to a variable o f the 2nd order, the results o f whioh are denoted
b y the sym bols ‘ H * * ’ , an<I ‘ fife * ’ or ‘\J'kx ’ and respec­
tively. Correspondingly there will be tw o operations o f sub­
stitution. A m ong the axiom s o f the logic o f relations we include
the sentences which satisfy the condition (a) o f D ef. 13, i.e.
substitutions o f the axiom s o f the sentential calculus, and uni­
versal quantifications o f these substitutions, and also all sen­
tences w hich are universal quantifications o f expressions o f the

^ >6
where k, l, and m are any natural numbers (l =a to) and y any
sentential function in which the free variable Vk does not occur.
Considering their intuitive m eaning the axiom s o f the last
category m ay be called pseudodefinitions.1
1 This term we owe to Le&mewski, who has drawn attention to the necessity
of including pseudodefinitions among the axioms of the deductive sciences in
those cases in which the formalization of the science does not admit the
possibility of constructing'suitable definitions (of. p. 166, footnote). Pseudo­
definitions can be regarded as a substitute for the axiom of reducibilUy of
Whitehead, A . N ., and Bussell, B. A . W . (90), vol. 1, pp. 55 ff. It would not
be difficult to show the connexion between these sentences and a group of
axioms adopted in Neumann, J. v. (54), p. 18.
224 CONCEPT OF TO U T# IN FORMALIZED LANGUAGES V33I, § 4

To obtain a correct definition o f satisfaction in connexion

with the language we are considering we must first extend our
knowledge o f this concept. In the first stage o f operating with
it we spoke o f the satisfaction o f a sentential function by one,
two, three objects, and so on, according to the number o f free
variables occurring in the given function (of. pp. 189 ff.). From
the semantical standpoint the ooncept o f satisfaction had there a
strongly ambiguous character; it included relations in which the
number o f terms was diverse, relations whose last domain was a
class o f sentential functions, whilst the other domains—in the
case o f the language o f the calculus of classes— consisted of
objects o f one and the same category, namely olasses o f indi­
viduals. Strictly speaking we were dealing not with one concept,
but with an infinite number o f analogous concepts, belonging to
different semantical categories. I f we had formalized the meta­
language it would have been necessary to use infinitely many
distinct terms instead of the one term ‘satisfies’. The semantical
ambiguity o f this concept increases still more when we pass to
languages o f more complicated logical structure. I f we continue
the intuitive considerations o f § 3, analyse the examples given
there and construct new ones after the same pattern, it soon
becomes clear that a strict semantical correlation exists between
the free variables o f the sentential function and the objects which
satisfy these functions: every free variable belongs to the same
semantical category as the name of the object corresponding to
it. If, therefore, at least two different categories occur among
the variables o f the language— as in the case we are investigating
—it does not suffice to restrict consideration to only a single
category o f objects in dealing with the concept of satisfaction.
The domains o f the single relations which are covered by the
term ‘satisfaction*, thus cease to be semantically unambiguous
(only the last domain consists as before exclusively o f sentential
functions). But since the semantical category o f a relation not
only depends on the number o f domains, i.e. the number of
terms standing in the relation to one another, but also on the
categories o f these domains, the category o f the concept of
satisfaction, or rather the category o f each single one of these
V n i, § 4 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES 225

concepts, also depends on two circumstances. It depends on

the number and also on the categories o f the free variables which
appear in the sentential functions to which the concept o f
satisfaction relates. In brief, it depends on what we have called
the semantical type of the sentential function. To functions
which belong to two distinct types two semantically distinct
concepts o f satisfaction always correspond.1 Some examples
will make this clear. We shall say that the objects R, a, and b
satisfy the function p12>3if and only if JRis a relation and a and b
are individuals and we have aRb (i.e. a stands in the relation
R to b). The function pix*'Pz,%,t18 satisfied by the objects R, a,
and 8 if and only if R and 8 are relations, a is an individual and
we have both aRa and aSa, The function f \'z C lzip ^ + p ^ w )
is satisfied by symmetrical relations and only by them, i.e. by
relations such that, for all individuals a and b, if we have
aRb we also always have bRa. The function HKpias+P w )
is satisfied by those and only those individuals a and b which
satisfy the following condition: for every relation R, if aRb,
then bRa, i.e. individuals which are identical. In the above
examples we have sentential functions belonging to four different
semantical types, and we are, therefore, dealing with four differ­
ent relations o f satisfaction, in spite o f the fact that the number
o f free variables and also the number o f terms in the relations is
the same in the first two examples.
The semantical ambiguity attaching to the concept o f satis­
faction in its original conception renders an exact characteriza­
tion o f this concept in a single sentence, or even in a finite number
of sentences, impossible, and so denies us the use o f the only
method so far known to us of constructing a definition of a true
sentence. In order to avoid this ambiguity, in dealing with the
calculus o f classes we had recourse to an artifice which is used by
logicians and mathematicians in similar situations. Instead o f
using infinitely many concepts o f satisfaction of a sentential
1 Moreover, functions o f one semantical type can correspond to several
semantically distinct concepts of satisfaction, provided the free variables of
these functions belong to at least two distinct semantical categories; in addi­
tion to the number and the categories of the variables their arrangement
must also be taken into consideration.
226 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES VIII, $ 4

function by single objects, we tried to operate with the semanti­

cally uniform, if somewhat artificial, concept o f the satisfaction
o f a function by a sequence o f objects. It happened that this
concept is sufficiently more general than the .previous one to
include it—intuitively speaking—as a special case (to define the
logical nature o f this inclusion would, however, be a little diffi­
cult). It will easily be seen that this method cannot be applied to
the present problem without further difficulty. Satisfaction in
its new form is a two-tormed relation, whose domain consists o f
sequences and counter domain o f sentential functions. As
before, there exists between the free variables o f a sentential
function and the corresponding terms o f the sequences which
satisfy it, a strict semantical correlation. Thus if the language
o f the logio o f relations contains variables o f two different
semantical categories, we must likewise use two categories o f
sequences in our investigations. For example, the funotion
fls Htow+Pw* sfttisfisd exclusively by sequences o f two-
termed relations between individuals (namely by those and only
those sequences F whose first term is a symmetrical relation).
But the function DrOhAs+PiAa) is satisfied exclusively by
sequences o f individuals (i.e. by sequences / for whioh / , = / a
holds). The domain o f the relation o f satisfaction and eo ipso
the relation itself thus again becomes semantically ambiguous.
Again we are dealing not with one, but with at least two different
concepts o f satisfaction. But still worse, a closer analysis shows
that the new interpretation o f the concept o f satisfaction can no
longer as a whole be maintained. For one and the same sentential
function often contains free variables o f two different categories.
To deal with such functions we must operate with sequences
whose terms likewise belong to two categories. The first term, for
example, o f the sequence which satisfies the funotion pw must
be a relation, but the two following ones must be individuals.
But it is known that the theory o f semantical categories does
not permit the existence o f such heterogeneous sequences.
Consequently the whole conception collapses. Thus changing
the original interpretation o f the concept o f satisfaction has
removed only one subsidiary cause o f its ambiguity, namely the
V m , $ 4 CONCEPT OP TRUTH IN FORMALIZED LANGUAGES 227

diversity in the number o f terms in the relations which are

the object o f the ocnoept; another far more important factor,
the semantical diversity o f the terms o f the relations, has lost
none o f its force.
Nevertheless the methods used in § S can be applied to the
language now being investigated, although with certain modifica­
tions. In this case also it is possible to find an interpretation o f
the oonoept o f satisfaction in whioh this notion loses its seman­
tical ambiguity and at the same time becomes so general that it
indudes all special oases o f the original concept. In fact, two
different methods are available; I shall call them the method of
many-rowed sequences and the method of semantical unification
of the variables.
The first method requires that we should treat satisfaction
not as a two-termed, but as a three-termed relation whioh holds
between sequences o f individuals, between sequences of two-
termed relations and between sentential functions. We use the
following mode o f expression: ‘the sequence / o f individuals
and the sequence F o f relations together satisfy the sentential
function The content o f this phrase can easily be visualized
by means o f conorete examples. For example, the sequence f
o f individuals and the sequence F of relations together satisfy
the function p, 9y if and only if the individual / s stands in the
relation Ft to the individual / 8. In order to formulate a general
definition we proceed exactly in the manner o f Def. 22 in § 3,
oare being taken to remember that, in the language we are con­
sidering, the expressions th, , „ play the part o f primitive senten­
tial functions and that instead o f one operation o f universal
quantification two related operations occur. The definition o f
true sentence is completely analogous to Def. 23.
This method can now be modified to some extent by treating
satisfaction as a two-termed relation between so-oalled two-
rowed sequences and sentential functions. Every ordered pair
whioh consists o f two sequences f and F is called a two-rowed
sequence (or two-rowed matrix), where the ith term o f the sequence
/ or o f the sequence F is called the fcth term o f the first or second
row respectively o f the two-rowed sequence. In the present
228 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES VID, § 4

case we have to deal with ordered pairs which consist o f a

sequence o f individuals and a sequence o f relations. It is easily
seen that this m odification is a purely form al one and has no
essential effect on the construction as a whole. It is to this
m odification o f the m ethod that the term ‘m ethod o f m any-
rowed sequences * is adapted.
T o understand the m ethod o f semantical unification o f the
variables we begin with certain considerations which are not
im m ediately connected w ith the language we are at present
investigating. It is known that with every individual a a definite
tw o-term ed relation a* can be correlated in such a way that to
distinct individuals distinct relations correspond. F or this pur­
pose it suffices to take as a* an ordered pair whose terms are
identical with a, i.e. the relation R which holds between any two
individuals b and c if and only if b = a and c = a. On the basis
o f this correlation we can now correlate in a one-one fashion with
every class o f individuals a class o f relations, with every m any-
term ed relation between individuals a corresponding relation
between relations, and so on. F or example, to any class A o f
individuals there corresponds a class A* o f all those relations a*
which are correlated with the elements a o f the class A. In this
w ay every sentence about individuals can be transform ed into
an equivalent sentence about relations.
Bearing these facts in m ind we return to the language o f the
logic o f relations and change the intuitive interpretation o f the
expressions o f this language w ithout in any w ay touching their
form al structure. A ll constants will retain their previous mean­
ing, whilst ah variables both o f the 1st and 2nd order are from
now on to represent names o f tw o-term ed relations. To the
prim itive sentential functions o f the type cXyz\ where instead
o f ‘X* some variable Vk and instead o f ‘y* and "zy any two vari­
ables and vm occur, we assign the follow ing meaning: ‘there
exist individuals a and b such that a stands in the relation X to
6, y = a*, and z ~ 6 V In this way the meaning o f the composite
sentential functions w ill likewise be m odified. It is almost
im m ediately evident that every true or false sentence in the
earlier interpretation will remain true or false respectively in
VUE, § 4 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES 229

the new one. B y virtue o f this new interpretation all the vari­
ables o f the language now belong to one and the same semantical
category, not indeed from the form al but from the intuitive
point o f view ; they represent words o f the same 'part o f speech’ .
Consequently the language we are considering can be investi­
gated b y exactly the same m ethods as all languages o f the 1st
kind; in particular, satisfaction can be treated as a two-term ed
relation between sequences o f relations and sentential functions.
A t the same tim e a com plication o f a technical nature— although
an unim portant one— presents itself. Since tw o free variables o f
different orders but the same indices, e.g. vl and m ay occur in
the same sentential function, it is not clear without supplemen­
tary stipulations which terms o f the sequence are to correspond
to the variables o f the 1st, and which to those o f the 2nd order.
To overcom e this difficulty we shall stipulate that to every
variable vk a term o f the sequence with an uneven index 2. 1 k—
Vk
corresponds, and to every variable a term with even index 2 . k
corresponds. F or exam ple, the sequence F o f relations satisfies
the function if and only if there are individuals a and b
such that a stands in the relation F%k to 6, 1 = a*, and
F^m^x = fe*. A part from this detail the definitions o f satisfaction
and o f true sentence differ in no essential point from the defini­
tions given in § 3.
The tw o m ethods described can be applied to all languages o f
the 2nd kind.1 I f the variables o f the language studied belong to
n different semantical categories, we regard satisfaction— under
the m ethod o f m any-rowed sequences— as an ft-fl-term ed
relation holding between n sequences o f the corresponding
categories and the sem antical functions, or as a two-term ed
relation whose dom ain consists o f n-rowed sequences (i.e. ordered
1 This holds even for languages in which variables occur which are not
included in the classification onp. 218 (cf. p. 218, note 2). I shall not deal with
certain (not particularly important) difficulties which may ocoui- here. But
I take this opportunity of mentioning that sentential variables, en if they
occur in the language, do not complicate the construction at »il;, and that,
in particular, it would not be worth while to include them in the process of
semantical unification. Sentences which contain such variables can be ex-
eluded by correlating with each of them, in one-many fashion, an equivalent
sentence which does not contain sentential variables (cf. Hilbert, D., and
Aekermann, W. (30), pp. 84-85).
230 CONCEPT OP TRUTH IN FORMALIZED LANGUAGES VH I, M

n-tuples o f ordinary sequences) and whose oounter domain

consists o f sentential funotions. Constructions based on this
method form the most natural generalization o f the constructions
in §3 and their material correctness appears to leave no
doubts.
In applying the method o f semantical unification o f the
variables, the choice o f the unifying category plays an essential
part, i.e. that semantical category in whioh all the variables o f
the language studied can be interpreted. Only one thing is
required o f the unifying oategory; that with all objects o f every
semantical category which is represented by the variables of
the given language, effective objects o f the chosen category,
can be correlated in a one-one fashion (i.e. so that to distinct
objects, distinct objects correspond). Nevertheless, the choice
o f the unifying category is not always so simple as in the example
discussed above in connexion with the language o f the logic of
relations; this choice cannot always be made from the categories
which occur in the language. If, for example, the variables o f
the language represent names of two-termed relations between
individuals and names of classes whioh consist o f olasses of indi­
viduals, then the simplest unifying category seems to be the
oategory o f two-termed relations between olasses o f individuals.
I do not propose to enter into a further analysis of this problem
(it would presuppose a knowledge o f certain facts belonging to
set theory). I add only the following remarks: (1) the unifying
oategory cannot be o f lower order than any one category among
those occurring in the language; (2) for every language o f the 2nd
kind a unifying category can be found, even infinitely many such
categories and in fact among categories o f the nth order, where n
is the highest order o f the variables occurring in the language.
As soon as the unifying category is specified, and the primitive
sentential functions correspondingly interpreted, the further
course o f the work does not differ at all from the methods of
construction used for languages o f the 1st kind.
In contrast to the method o f many-rowed sequences, there is
no doubt that the second method is somewhat artificial. Never­
theless the definitions constructed by this method prove, on
V III,J 4 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES 231

closer analysis, to be intuitively evident to a soaroely less degree

than the constructions based on the first method. At the same
time they have the advantage o f greater logical simplicity. In
particular, when we are dealing with the definition o f true sen­
tence the proof o f the equivalence o f its two formulations presents
no difficulty in any ooncrete oase. The essential advantages o f
the method o f unification o f the variables only become dear,
however, in the investigation o f languages o f the 3rd kind, since
the method o f many-rowed sequences here proves to be quite
useless.
As a typical example o f a language o f the 3rd kind we choose
the language o f the logic qfmany-termed relations.1 In this soienoe
we deal with the same constants <N\ *A ’, and ‘J J ’ and with the
same variables of the 1st order vk, as in the logio of two-termed
relations. But we also find here variables o f the 2nd order in
greater multiplicity than before. As variables o f this kind we
shall use such signs as X’, ‘X’, X,’,..., X ’, X ’, X ’,...,
‘XT’, XT, X*’,— and so on. The composite symbol constructed
from the sign X ’ with k small strokes below and l such strokes
above will be called the kth variable functor with l arguments,
and denoted by ‘ V\'. Intuitively interpreted, the variables
vk represent, as before, names of individuals, whilst the variables
V\ represent names o f 1-termed relations between indivi­
duals, in particular for l — 1 names o f one-termed relations,
i.e. names o f classes. Both from the intuitive and the formal
points o f view the signs vk, V%, belong to infinitely many
distinct semantical categories o f the 1st and 2nd orders respec­
tively. The fundamental sentential functions are expressions o f
the type 'Xxy...z\ where in plaoe o f X ’ any variable functor
with l arguments and in plaoe o f V , ‘y ’,..., V variables o f the
1st order, l in number, occur. These expressions are read as
X
follows: ‘the ^-termed relation holds between the l individuals
According to the number and form o f the variables we
denote the primitive functions by the symbols <pic,m,n>—>
1 This is a language which resembles the language of the lower predicate
oaleulus of Hilbert, D., and Ackermann, W. (30), pp. 43 ft, but is richer than
the latter because variable funotore can occur in it both as free and as bound
variables.
232 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES V m , § 4

putting pkm = V\~vm, pkim>n = (V l~ v m)~ v n, and so on. In

order to obtain a unified sym bolism , which is independent
o f the number o f variables, we shall use sym bols o f the type
(where ‘p* represents the name o f a finite sequence o f
natural num bers), the meaning o f which is determ ined b y the
form ula p\p = vPi)~ v pi)~ The defi-
nitions o f the m etatheory do not differ at all from the analogous
definitions relating to the logic o f two-term ed relations and even
to the calculus o f classes. As operations o f quantification we
introduce quantification with respect to the variables vk and the
variables Vlk and denote the result o f the operations b y the
sym bols and respectively, The list o f axiom s in­
cludes those which satisfy the condition (<x) o f D ef. 13 o f § 2,
and pseudodefinitions which form a natural generalization o f
the pseudodeftnitions from the logic o f tw o-term ed relations.
Their m ore detailed description seems to be unnecessary.
W e turn now to the problem o f how the concept o f satisfaction
is to be conceived and the definition o f truth to be constructed for
the language we are now considering. A ny attem pt to apply the
m ethod o f m any-rowed sequences in this case fails com pletely.
In this m ethod the term ‘satisfaction5— in whatever form —
expresses the relation o f dependence between n sequences o f
various categories and the sentential functions, where n is exactly
equal to the number o f semantical categories represented by the
variables o f the given language. In the case we are investi­
gating the number n is indefinitely large and the metalanguage
we are using— like all other actually existing form alized lan­
guages— provides no means for dealing with the relation of
m utual dependence between objects which belong to infinitely
m any distinct sem antical categories.2

1 Strictly speaking the meaning of the symbol ‘ plkp ’ should be defined

recursively.
a In those cases in which, in logical and mathematical constructions, we
deal with the mutual dependence between an arbitrary, not previously
determined number of objects of one and the same semantical category, we
mostly use ordinary sequences. For objects which belong to a finite number
of distinct categories many-rowed sequences fulfil the analogous function.
But on the basis of the known languages we find nothing like ‘ sequences with
infinitely many rows* (of distinct semantical categories).
V m , § 4 CONCEPT OF TROTH IN FORMALIZED LANGUAGES 233

The m ethod o f semantical unification o f the variables can, how ­

ever, be applied to this language with com plete success. T o see
this it suffices to note that we can correlate in a one-one fashion,
with every n -term ed relation jR between individuals, a class R*
which consists o f w-termed sequences o f individuals, nam ely the
class o f all sequ en ces/ which satisfy the follow ing condition: the
relation R holds between the individuals f v / 2,..., f n. F or ex­
ample, the class o f all sequences / with tw o terms f x and /* such
that / li? /2 corresponds to the tw o-term ed relation R. Con­
sequently every sentence concerning m any-term ed relations
can be transform ed into an equivalent sentence which asserts
something about classes o f sequences. It will be remembered
that b y sequences o f individuals we mean two-term ed relations
between individuals and natural numbers, A ccordingly all
sequences o f individuals, whatever the number o f their terms,
belong to one and the same sem antical category and therefore
the classes o f these sequences, in contrast to m any-term ed
relations, likewise belong to one and the same category.
On the basis o f these considerations we now partially unify
the semantical categories o f the variables in the follow ing w ay.
T o the variables vk we give— at least provisionally— the same
significance as before. B ut the variables V\ now represent the
names o f any classes which consist o f finite sequences o f indi­
viduals or o f other objects o f the same category (i.e. the names
o f objects o f at least the 3rd order, according to the order which
we assign to the natural num bers).1 The prim itive functions o f
the form ‘Xxy.„z\ which begin with a functor with l arguments
and hence contain l variables o f the 1st order, are interpreted
by phrases o f the ty p e : ‘the sequence o f individuals the first term
o f which is x , the second y,... and the Zth (the last) is s, belongs to

1 In systems of mathematical logic, e.g. in Whitehead, A. N ., and Russell,

B. A. W . (90), vol. 2, pp. 4fL , the cardinal numbers and in particular the
natural numbers are usually treated as classes consisting of classes of individuals
(or other objects), namely as the classes of all those classes which are similar
(in the Principia Mathematica sense) to a given class. For example, the number
1 is defined as the class of all those classes which have exactly one element.
With this conception the natural numbers are thus objects of (at least) the 3rd,
sequences of individuals of the 4th, and classes of these sequences of the 5th
order.
284 C O N C E PT O F T R U T H IN F O R M A L IZ E D L A N G U A G E S V m ,}4

the class X which consists o f i-termed sequences'. From the

intuitive, although not from the formal, standpoint, the variables
from now on still belong to only two distinct semantical cate­
gories; in view o f this circumstance we can use, in the further
course o f our work, the same methods as we employed in investi­
gating languages o f the 2nd kind.
By means o f the phrase: 'the sequence/of individuals and the
sequence F , whose terms form classes o f finite sequences o f
individuals, together satisfy the given sentential function’, we
can bring into service the method o f many-rowed sequences.
To use this concept consistently we must first set up a one-one
correlation between the variables V\ and the terms o f the
sequence F in such a way that terms with different indices
correspond to different variables. This is most easily done by
putting every variable V\ in correspondence with a term having
the index (2.&—1).2,_1. For example, the terms Fv Ft, Fs, Ft,
Ft, correspond to the variables 7J, F|, V}, F|, Ff, Ff,...
With this convention the establishment o f the meaning of the
above phrase in its application to any concrete sentential func­
tion, and even the construction o f a general definition o f the
concept in question, presents no further difficulties. Thus
concerning the primitive functions, those and only those
sequences / and F (of the categories given above) will together
satisfy the function p*^ which satisfy the following condition:
the sequence g o f individuals, whose single term & is identical
w ith /m, belongs to the class In an analogous way, those
functions/and F will together satisfy the function p*^* whioh
satisfy the following condition: the sequence g of individuals with
two terms, where gx = f m and gz = / n, belongs to the olass
•iF&ft-ixs. In general, in order that the sequences / and F should
together satisfy the function it is necessary and sufficient
that the sequence g o f individuals with l terms, where gx =3 fPi,
9t — 9i — fp,t should belong to the olass F(8Jk_^>aj_i (winch
consists o f sequences with the same number o f terms).
1 Instead of the function f(k, l) (2.fc— we ocrnld use any other
function f(k, l) which correlates the natural numbers in one-one fashion with
ordered pairs of natural numbers. Bet theory offers many examples of such
correlations; cf. Fraenkel, A. (10), pp. 30 ff. and 96.ff.
V III, 5 4 C O N C E PT O P T R U T H IN F O R M A L IZ E D L A N G U A G E S 235

I f we wish to apply the method o f unification o f the variables

we again make use o f the fact that a one-one correlation can be
set up between any individuals and certain classes o f finite
sequences, and in such a way that to every individual a there
corresponds the class a* containing as its only element a
sequence whose only member is just the given individual.
'Beginning in this way we next modify the interpretation o f the
variables o f 1st order in exactly the same direction in whioh we
formerly modified the interpretation o f the variables o f the 2nd
order. The primitive functions o f the form 'Xzy...z\ containing
1+ 1 signs, we now regard as having the same meaning as expres­
sions o f the type ‘the Z-termed sequence g o f individuals which
satisfies the conditions: gj — x, gjf = y,..., gf — z, belongs to
the class X , which consists o f sequences with l terms’ . With this
intuitive interpretation all variables now belong to the same
semantical category. The further construction contains no
essentially new features and the reader will encounter no serious
difficulties in carrying it out.
The method o f semantical unification of the variables can be
applied with equal success to the investigation o f any language o f
the 3rd kind. Determining the unifying category may sometimes
be more difficult. As in the case o f languages o f the 2nd kind
it is here impossible to restrict consideration to categories occur­
ring in the language studied. In contrast to those languages it is
never possible to make the choice from among the categories o f
one o f the orders represented in the language. This difficulty
is not, however, essential and exclusively concerns languages o f
the lowest order. For it is possible to prove that for those
languages in which the order o f the variables does not exceed a
given number n, where n > 3, any category of the nth order can
serve as the unifying category.
In this way the various methods at our disposal enable vs to
define the concept of satisfaction and with it to construct a correct
definition o f truth for any language of finite order. We shall see
in the next section that these methods do not extend further;
the totality o f languages o f finite order exhausts the domain o f
applicability o f our methods. This is therefore the place in whioh
236 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES VIII, § 4

to summarize the m ost im portant consequences which follow

from the definitions we have con stru cted.!
First, the definition of true sentence is a correct definition of
truth in the sense of convention T o f § 3. It embraces, as special
cases, all partial definitions which were described in condition
(a) o f this convention and w hich elucidate in a m ore precise
and m aterially correct way the sense o f expressions o f the
type ‘a; is a true sentence’ . A lthough this definition alone
provides no general criterion o f truth, the partial definitions
m entioned do perm it us definitely to decide in m any cases the
question o f the truth or falsity o f the sentences investigated.
In particular, it can be proved— on the basis o f the axiom s o f
the second group adopted in the m etatheory (cf. p. 211)— that
all axioms of therscience under investigation are true sentences. In a
similar manner we can prove, m aking essential use o f the fact
that the rules o f inference em ployed in the m etatheory are not
logically weaker than the corresponding rules o f the science
itself, that all consequences of true sentences are true. These tw o
facts together enable us to assert that the class of true sentences
contains all provable sentences of the science investigated (cf.
Lemm a D and Ths. 3 and 5 o f § 3).
Am ong the mo3fc im portant consequences o f a general nature
which follow from the definition o f truth must be reckoned the
principle of contradiction and the principle of the excluded middle.
These tw o theorem s, together w ith the theorem on the con­
sequences o f true sentences already m entioned, show that the
class of all true sentences forms a consistent and complete deductive
system (Ths. 1, 2, and 4).
As an im m ediate, although a somewhat subsidiary, con­
sequence o f these facts we obtain the theorem that the class of all
provable sentences likewise forms a consistent (although not neces­
sarily com plete) deductive system. In this way we are able to
produce a p roof o f the consistency o f every science for which we
can construct the definition o f truth. The p roof carried out by
t Some further consequences of this type are discussed in the article of
the author ‘ On undeeidable statements in enlarged systems of logic and the
concept of truth’ , Journal oj Symbolic Logic, voL 4 (1939), pp. 105-12; cf. in
particular sect. 9, p. 111.
VIII, § 4 CONCEPT OP TRUTH IN FORMALIZED LANGUAGES 237

means o f this m ethod does not, o f course, add much to our know­
ledge, since it is based upon premisses which are at least as strong
as the assumptions o f the science under investigation.1 N ever­
theless it seems to be w orthy o f note that such a general m ethod o f
p roof exists, which is applicable to an extensive range o f deduc­
tive sciences. It will be seen that from the deductive standpoint
this m ethod is not entirely trivial, and in m any cases no simpler,
and in fact no other, m ethod is kn ow n.f
In those cases in which the class o f provable sentences is not
only consistent but also com plete, it is easy to show that it co­
incides with the class o f true sentences. If, therefore, we identify
the tw o concepts— that o f true sentence and that o f provable
sentence—we reach a new definition o f truth o f a purely struc­
tural nature and essentially different from the original semantical
definition o f this notion.2 E ven when the provable sentences

1 As Ajdukiewicz has rightly pointed out in a somewhat different connexion

(cf. Ajdukiewicz, K . (2), pp. 39-40) it does not at aU follow from this that this
proof is not correct from the methodological standpoint— that it contains in
some form a petitio prvnoipii. The assertion which we prove, i.e. the consistency
of the science, does not occur in any way among the hypotheses of the proof.
1 In the course of this work I have several times contrasted semantical
definitions of true sentence with structural definitions. But this does not mean
that I intend to specify the distinction between the two kinds of definitions
in an exact way. From the intuitive standpoint these differences seem to be
tolerably clear. Def. 23 in § 3— as well as other definitions constructed in the
same way— I regard as a semantical definition because in a certain sense

t In connexion with the problem discussed in the last three paragraphs see
the recent publications: Mostowski, A . (53 e) as weU as Wang, H. (87 c). From
the results of these authors it is seen that in some cases, having succeeded
in constructing an adequate definition of truth for a theory T in its meta­
theory, we may still be unable to show that all the provable sentences of T
are true in the sense of this definition, and hence we may also be unable tc
carry out the consistency proof for T in M , This phenomenon can roughly
be explained as foUows: in the proof that all provable sentences of T are
true a certain form of mathematical induction is essentially involved, and the
formalism of M may be insufficiently powerful to secure the validity of this
inductive argument . Hence a certain clarification of the assumptions (on pp.
174 ff.) concerning foundations of the metatheory may be desirable. In
particular the phrase ‘ from any sufficiently developed system of mathematical
logic* (p. 170) should be understood in a way which does not deprive the
metatheory of any normally applied modes of inference. I f the theory T is
of finite order our purpose will be fully achieved if we decide to provide the
metatheory M with a logical basis as strong as the general theory of classes
discussed in the following section.
23$ C O N C E PT O F T R U T H IN F O R M A L IZ E D LA N G U A G E S VU C, $ 4

do not form a complete system the question of the construction

o f a structural definition is not a priori hopeless. Sometimes it is
possible, by adding certain structurally described sentences, to
extend the axiom system o f the soienoe in a suitable way so that
it becomes a system in which the olass o f all its consequences
coincides with the olass o f all true sentences. But there can be
no question o f a general method o f construction. I suspect that
the attempt to construct a structural definition, even in relatively
simple oases—e.g. in connexion with the logic of two-termed
relations studied in the preceding section—would encounter
serious difficulties. These difficulties would certainly become
muoh greater when it came to the question o f giving a general
structural criterion o f truth, although we have already dealt
with two languages, that o f the oaloulus o f classes and that o f

(which would be difficult to define) it represents a ‘ natural generalization’,

bo to speak an ‘ infinite logical product %of those partial definitions which were
described in convention T and which establish a direct correlation between
the sentences of the language and the names of these sentences. Among the
structural definitions, on the other hand, I include those which are con­
structed according to the following scheme: a class of sentences or other ex­
pressions is described in such a way that from the form of every expression
it is possible to know whether it belongs to the given olass or not. Further
operations on expressions are given of such a kind that if certain expressions
in finite number are given and if the form of an arbitrary other expression is
given, then we can decide whether it can be obtained from the given expressions
by means of the given operations. Finally the true sentences are defined as
those whioh are obtained by applying the given operation to the expressions
of the given class any number of times (it is to be noted that such a structural
definition still in no way provides a general criterion of truth). Certain
differences of a formal nature oan be recognized between these two kinds of
definitions. The semantical definition requires the use of terms of higher order
than all variables of the language investigated, e.g. the use of the term
‘ satisfies’ ; but for the formulation of a structural definition the terms of per*
haps two or three of the lowest orders suffioe. In the construction of a
semantical definition we uce—explicitly or implicitly—those expressions of
the metalanguage which are of like meaning with the expressions of the
language investigated, whilst they play no part in the construction of a struc­
tural definition; it is easy to see that this distinction vanishes when the
language studied is a fragment of logic. Moreover, the distinction as a whole
is not very clear and sharp, as is shown by the fact that with respect to the
sentential calculus the semantical definition can be regarded as a formal
transformation of the structural definition based on the matrix method. At
the same time it must be remembered that the construction of semantical
definitions, based on the methods at present known to us, is essentially
dependent upon the structural definitions of sentence and sentential
function.
VTH , g 4 C O N C E PT O F T R O T H IN F O R M A L IZ E D L A N G U A G E S 239

the sentential calculus, for which this problem oould be rela­

tively easily solved.1
In all cases in which we are able to define satisfaction and the
notion o f true statement, we can—by means o f a modification
o f these definitions— also define two still more general concepts
o f a relative kind, namely the concepts o f satisfaction and cor­
rect sentence—both with respect to a given individual domain a.%
This modification depends on a suitable restriction o f the domain
o f objeots considered. Instead o f operating with arbitrary indivi­
duals, classes o f individuals, relations between individuals, and so
on, we deal exclusively with the elements o f a given class a o f indi­
viduals, subclasses o f this class, relations between elements o f
this class, and so on. It is obvious that in the special case when a
is the class o f all individuals, the new concepts coincide with the
former ones (cf. Defs. 24 and 25, and Th. 26). As I have already
emphasized in § 3 the general concept o f correct sentence in a
given domain plays a great part in present day methodological
researches. But it must be added that this only concerns re­
searches whose objeot is mathematical logic and its parts. In
connexion with the special sciences we are interested in correct
sentences in a quite specifio individual domain for which the
general concept loses its importance. Likewise it is only in con­
nexion with sciences which are parts o f logic that some general
properties o f these concepts, which were proved in § 3 for the
language o f the calculus o f classes, preserve their validity. For
example, it happens that in these sciences the extension o f the
term ‘correct sentence in the individual domain a ’ depends
exclusively on the cardinal number o f the class a. Thus in these
investigations we can replace this term by the more convenient
term ‘correct sentence in a domain with k elements’ (Def. 26,
Th. 8). The theorems previously discussed concerning the con­
cept o f truth, such sb the principles o f contradiction and the
excluded middle can be extended to the ooncept o f correct sen­
tence in a given domain. The concept o f correct sentenoe in every

1 Cf. the remarks on pp, 207 f. and 221; T shall return to this problem in
$ 6 (cf. p. 264, footnote).
* 290; note 2.
240 C O N C E PT O P T R U T H IN F O R M A L IZ E D L A N G U A G E S V IU , § 4

individual domain (Def. 27) deserves special consideration. In

its extension it stands midway between the concept o f provable
sentence and that of true sentence; the class o f correct sentences
in every domain contains all theorems and consists exclusively
o f true sentenoes (Ths. 22 and 27). This class is therefore in
general narrower than the class of all true sentences; it contains,
for example, no sentences whose validity depends on the magni­
tude o f the number o f all individuals (Th. 23). I f it is desired to
transform the system o f the provable sentenoes o f every science
into a complete one, it is neoessary at the outset to add sen­
tences to the system whioh decide the question how many
individuals exist. But for various reasons another point o f view
seems to be better established, namely the view that the decision
regarding such problems should be left to the speoific deductive
sciences, whilst in logic and its parts we should try to ensure
only that the extension o f the concept o f provable sentence
coincides with that o f correct sentence in every individual
domain. For a supporter o f this standpoint the question whether
the extension o f these two concepts is actually identical is of
great importance. In the case o f a negative answer the problem
arises o f completing the axiom system o f the science studied in
such a way that the class o f provable sentences thus extended
now. coincides with the class o f sentences which are correct in
every domain. This problem, which properly is equivalent to
the question o f structurally characterizing the latter concept,
can be positively decided only in a few cases (cf. Th. 24).1 Gene­
rally speaking the difficulties presented by this question are no
less essential than those connected with the analogous problem
o f a structural definition o f true sentence. We meet with similar
difficulties when we attempt to define structurally the concept of
correct sentence in a domain with k elements. Only in the case
where & is a finite number is it easy to give a general method,
modelled on the method o f matrices from investigations on the
extended sentential calculus, which makes a structural definition

1 In the cnee of the lower functional calculus this problem, which is raised
in Hilbert, D., and Ackermann, W. (30), p. 68, has recently been decided by
Gddel, see Gftdel, K. (20).
VIII, § 4 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES 241

o f this concept possible. In this w ay we even obtain a general

criterion which enables us to decide from the form o f any sentence
whether it is correct in a dom ain w ith a previously given finite
number o f elem ents.1
I do not wish to enter here into a more detailed discussion o f
special investigations on the concepts just considered. Some
results which are relevant here, relating to the calculus o f classes,
have already been given as examples in § 3. I will only m ention
that in recent years numerous results have been obtained which
enable us to infer from the correctness o f certain sentences in
special individual dom ains or from their structural properties
their correctness in every dom ain and thus their truth.2 I t is
evident that all these results only receive a clear content and can
only then be exactly proved, if a concrete and precisely form u­
lated definition o f correct sentence is accepted as a basis for the
investigation.

§ 5. T he Concept of T rue S entence in L anguages of

I nfinite Order
W e com e now to languages o f the 4th kind, hence to those
o f infinite order and so lying beyond the scope o f the m ethods o f
construction sketched in the preceding section. The language o f
the general theory of classes will serve as an exam ple. This
language is notew orthy because, in spite o f its elem entary
structure and its poverty in gram m atical form s, it suffices for
1 Cf. Bernaya, P., and SchonfLokel, M. (5 a), p. 352..
3 According to the well-known theorems of Lowenheim and Skolem, certain
categories of sentences are correct in every domain provided they are correct
in all finite and denumerable domains. These sentences include, for example,
all sentences of the logic of two- or many -termed relations, described in this
section, which are generalizations of sentential functions in which variables
of the 2nd order occur exclusively as free variables. In the case of the sentences
of the calculus of classes this result.— as is shown In Ths. 15 and 19 of § 3— can
be essentially sharpened. Certain results of Bernays, Schonfinkel, and Acker -
mann have a narrower domain of application. They allow us to correlate a
particular natural number h with sentences of a special structure in such a
way that from the correctness of a given sentence in the domain with k
elements (thus— as we already know— from purely structural properties of the
sentence) its correctness in every domain follows. Cf. Ackermann, W . (1),
Bernays, P ., and Schtinfinkel, M. (6'a), Herbrand, J. (26), Lowenheim, L. (49),
Skolem, Th. (64), (65), and (66). For a systematic presentation of the results
in this direction including more recent ones, see Church, A. (11 a).
242 C O N C E PT O F T R U T H IN F O R M A L IZ E D L A N G U A G E S VH L $5

the formulation of everyidea which con be expressedin thewhole

language of mathematical logio. It is difficult to imagine a
simpler language which can do this.1
In the general theory o f olasses the same constants oocur as
in the previously investigated sciences, i.e. the signs o f negation
and o f logical sum, as well as the universal quantifier. As
variables we use suoh symbols as ‘X !’, ‘X i\ ‘XV, and so on, i.e.
signs composed o f the symbol ‘X ' and a number o f small strokes
above and below. The sign having n strokes above and k below
is called the k-th variable of the n-th order and is denoted by the
symbol ‘Fg\ The variables V\, V%, V%... represent respectively
names o f individuals, objeots o f the 1st order; olasses o f indivi­
duals, objects o f the 2nd order; classes o f such classes, objects
o f the 3rd order, and so on. These variables obviously belong
to infinitely many semantical categories. As primitive senten­
tial functions we have expressions o f the type X T ’ where in
the place o f X ’ any variable o f the n-j- 1th order, and instead
o f T ’ a variable o f the »th order occurs. This expression is
1 The language of the general theory of clashes is much inferior to the
language of Whitehead, A. N ., and Bussell, B, A* W . (90) in its stock of
semantical categories, and still more inferior in this respect to the language used
by Lefeniewski in his system (of. p. 210, note 2 ; p. 218, note 2). In particular, in
this language no sentential variables and neither names of two- or many-termed
relations, nor variables representing these names, occur. The dispensability
of sentential variables depends on the fact mentioned on p. 229, footnote: to
every sentence which contains sentential variables there is a logically equiva­
lent sentence which does not contain such variables. The results of § 2,
especially Defs. 13-17, suffice to show how such variables are to be avoided in
setting up lists of axioms and in the derivation of theorems; cf. also Neumann,
J. v. (54) (especially note 9, p. 38). The possibility of eliminating two-termed
relations results from the following consideration. With every relation R we
can correlate, in one-one fashion, a class of ordered pairs, namely, the class
of all ordered pairs whose terms x and y satisfy the formula, xRy. If the
relation is homogeneous, i.e. if the domain and counter domain of this relation
belong to the same semantical category, then the ordered pair can be inter­
preted otherwise than we have done on p. 171, namely as olasses having two
classes as elements: the class whose only element is x and the class consisting
of the two elements x and y. In order to apply an analogous method to in­
homogeneous relations we must first correlate homogeneous relations with
them in one-one fashion, and this presents no great difficulty. We proceed
in an analogous way with many-termed relations. In this way every statement
about two- or many-termed relations of arbitrary category can be transformed
into an equivalent statement about individuals, classes of individuals, classes
of such classes, and so on. Cf. Kuratowski, C. (38), p. 171, and Chwistek, L.
(13), especially p. 722.
vm . §8 C O N C E PT O P T R O T H IN F O R M A L IZ E D L A N G U A G E S 243

read; 'the class X (of n + 1th order) has as an element the object
Y (of »th order)’, or ‘the object Y has the property X \ For the
designation o f the primitive functions we employ the symbol
setting ~ Fg+1^ Fj*. The farther development of the
science differs in no essential way from that o f the logic o f two-
or many-termed relations. The quantifications o f the sentential
functions x with respect to the variable V% are denoted by the
symbols x ’ and U* •c - The axioms consist of (1) sentences
which satisfy the condition (a) o f Def. 13 o f § 2, which are thus
derived from the axioms o f the sentential calculus by substitu­
tion, sometimes also followed by generalization; (2) pseudodefini­
tions, i.e. statements which are quantifications o f sentential
functions o f the type

where y is any sentential function which does not contain the

free variable F*+1; (3) the laws of extenMonaHty, i.e. sentences
o f the form

which state that two classes which do not differ in their elements
do not differ in any o f their properties and are thus identical.
In order to obtain in this science a sufficient basis for the estab­
lishment o f various parts o f mathematics and in particular o f
the whole o f theoretical arithmetic, we mu3t add to the above
still one more axiom: (4) the axiom of infinity, i.e. the sentence

U K U I 4 * •n f R x + U I W U - f l i ( ei,i-r4,x) •U M ,1 •4 M
which guarantees the existence o f infinitely many individuals.1
In the derivation o f consequences from the axioms we apply
the operations o f substitution, detachment, and the intro­
duction and removal o f the universal quantifier, analogous to
the operations described in conditions (y}-(£) of Def, 15 in § 3.
When we try to define the concept of satisfaction in connexion
with the present language we encounter difficulties which we
cannot overoome. In the face o f the infinite diversity o f seman-
1 la adopting the axiom of infinity we admittedly give up the postulate
aocording to which only the sentences which are correct in every individual
domain am to be provable sentences of logio (cf. p. 240).
244 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES VIII, § 5

tical categories which are represented in the language, the use

o f the m ethod o f m any-rowed sequences is excluded from the
beginning, just as it was in the case o f the logic o f m any-term ed
relations. B ut the situation here is still worse, because the
m ethod o f sem antical unification o f the variables also fails us.
As we learnt in § 4, the unifying category cannot be o f lower order
than any one o f the variables o f the language studied. Sequences
whose terms belong to this category, and still more the relation
o f satisfaction, which holds between such sequences and the
corresponding sentential functions, must thus be o f higher order
than all those variables. In the language with which we are now
dealing variables o f arbitrarily high (finite) order occu r: con­
sequently in applying the m ethod o f unification it would be
necessary to operate with expressions o f ‘infinite order’ . Y et
neither the metalanguage which form s the basis o f the present
investigations, nor any other o f the existing languages, contains
such expressions. I t is in fa ct not at all clear what intuitive
meaning could be given to such expressions.
These considerations seem to show that it is im possible to
construct a general, sem antically unambiguous concept o f
satisfaction for the language we are studying which will be
applicable to all sentential functions w ithout regard to their
semantical type. On the other hand there appear to be no
difficulties which would render im possible in principle a con­
sistent application o f the concept o f satisfaction in its original
form ulation, or rather— in view o f the semantical am biguity o f
that form ulation— o f an infinite number o f such concepts. Each
o f these concepts is, from the semantical standpoint, already
specified and w ould relate exclusively to functions o f a specific
semantical type (e.g. to functions which contain a variable o f the
1st order as the only free variable). A ctually— independently
o f the logical structure o f the language— the intuitive sense of
none o f these expressions raises any doubt. F or every par­
ticular sentential function we can in fact define this meaning
exactly b y constructing for every phrase o f the type ‘the
objects a, 6, c,... satisfy the given sentential fu nction’ an in­
tuitively equivalent phrase which is expressed -wholly in terms
VIH, § 5 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES 245

o f the metalanguage. Nevertheless the problem o f the construc­

tion o f a correct definition for each o f these concepts again
presents us with difficulties o f an essential nature. On the basis
o f the languages which we have previously studied it was easy
to obtain each special concept o f satisfaction b y a certain
specialization o f the general concept; in the present case this
w ay is clearly not open to us. A brief reflection shows that
the idea o f using the recursive m ethod analogously to the
definition o f sentential function proves, in spite o f its natural­
ness, to be unsuitable. It is easily seen that the com posite
functions o f a particular sem antical type cannot always be
form ed from simpler functions o f the same type. On the con­
trary, if we are to be able to construct arbitrary functions o f a
given type, we must use for that purpose all possible semantical
types.1 It would, therefore, be necessary, in the recursive defi­
nition o f any one o f the special concepts o f satisfaction, to cover,
in one and the same recursive process, infinitely many analogous
concepts, and this is beyond the possibilities o f the language.
The central problem o f our work, the construction o f the
definition o f truth, is closely connected with these considera­
tions. I f we were successful in defining, if not the general, at
least any one o f the special concepts o f satisfaction, then this
problem would not offer the least difficulty.1 2 On the other

1 An external expression of tins state of affairs is that in the definition of

satisfaction not only is it essential to take free variables into account but also
all the bound variables of the function in question, although these variables
have no influence on the semantical type of the function; and whether the
relation of satisfaction holds or not does not depend in any way on the terms
of the sequence which correspond to these variables (cf. Def. 22 of § 3, condition
(8)). It is to be remembered that analogous difficulties to those mentioned in
the text appeared earlier in the attempt to construct a recursive definition
of truth by a direct route (cf. p. 189).
2 For example, let us imagine that we have succeeded in some way in
defining the concept of satisfaction in the case of sentential functions which
contain a variable of 1st order as the only free variable. We could then operate
freely with phrases of the type ‘ the individual a satisfies the sentential func­
tion y\ I f we now consider some one concrete sentential function, e.g. U? d,i*
which is satisfied by every arbitrary individual, we obtain at once the follow­
ing definition of true sentence: x is a true sentence if and only if every individual
a satisfies the function a .U ? *\,i (*•£• tte conjunction of the sentence x and the
function U? Ci.i)* In an exactly analogous way we can pass from every other
specific concept of satisfaction to the concept of truth.
246 CO N C E PT O F T R U T H IN F O R M A L IZ E D L A N G U A G E S V III, $ 6

hand, we know o f no method o f construction which would not—

directly or indirectly—presuppose a previous definition o f the
concept o f satisfaction. Therefore we can say— considering the
failure o f previous attempts—that at present we can construct
no correct and materially adequate definition o f truth for the
language under investigotion.f
In the face o f this state o f affairs the question arises whether
our failure is accidental and in some way connected with defects
in the methods actually used, or whether obstacles of a funda­
mental kind play a part which are connected with the nature
o f the concepts we wish to define, or o f those with the help
o f which we have tried to construct the required definitions. I f
the second supposition is the correot one all efforts intended to
improve the methods of construction would dearly he fruitless.
I f we are to answer this question we must first give it a rather
less indefinite form. It will be remembered that in the con­
vention T of § 8 the conditions whioh decide the material
correctness o f any definition o f true sentence axe exactly stipu­
lated. The construction o f a definition whioh satisfies these
conditions forms in fact the principal object o f our investigation.
3?rom this standpoint the problem we are now considering takes
on a precise form: it is a question o f whether on the basis of the
metatheory of the language we are considering the construction of a
correct definition of truth in the sense of convention T is in principle
possible. As we shall see, the problem in this form can be defi­
nitely solved, but in a negative sense.
It is not difficult to see that this problem exceeds the bounds of
our previous disoussion. It belongs to the field o f the meta­
metatheory. Its definitive solution, even its correct formulation,
would require new equipment for investigation and especially
the formalization o f the metalanguage and the metatheory
which uses it. But without going so far, and still avoiding
t The problem of the possibility of defining satisfaction and truth for the
language under investigation will be considerably clarified by the discussion
in the Postscript. It should be mentioned that the .method of defining truth
recently suggested in McKinsey, J. C. C. (53 b) is not based on a preliminary
definition of satisfaction. Instead, McKinsey has to consider formalized
languages with non-denumerably many constants and has to use a meta­
language which is provided with a very strong set-theoretical apparatus.
V m , §6 C O N C E PT O F T R U T H IN F O R M A L IZ E D LA N G U A G E S 247
various technical complications, I believe I am able to give a
fairly clear account o f everything o f a positive nature that can
at present be established in connexion with the above problem.
In operating with the metalanguage we shall adhere to the
symbolism given in §§ 2 and 8. To simplify the further develop­
ments and avoid possible misunderstandings we shall suppose
the metalanguage to be so constructed that the language we are
studying forms a fragment of i t ; every expression o f the language
is at the same time an expression o f the metalanguage, but not
vice versa. This enables us in certain oases (e.g. in the formula­
tion o f condition (a) o f convention T ) to speak simply o f the
expressions o f the language itself, instead o f expressions of the
metalanguage which have the same meaning.
After these reservations and conventions we turn to the
formulation and proof o f the fundamental result.
T heorem I. (a) In whatever way the symbol ‘ Tr\ denoting a
class of expressions, is defined in the metatheory, it will be possible
to derive from it the negation o f one of the sentences which were
described in the condition (a) o f the convention T ;
(j3) assuming that the does o f dll provable sentences of the meta-
theory is consistent, it is impossible to construct an adequate
definition of truth, in the sense of convention T on the basis of the
metatheory.
The idea o f the proof o f this theorem can be expressed in the
following words:1 (1) a particular interpretation o f the meta-
1 We owe the method used here to Godel, who has employed it for other
purposes in bis recently published work, G5del, K. (22), of. especially pp. 174-5
or 187-90 (proof of Th. VI). This exceedingly important and interesting article
is not directly connected with the theme of our work—it deais with strictly
methodological problems: the consistency and completeness of deductive
systems; nevertheless we shall be able to use the methods and in part also the
results of GodeVs investigations for our purpose*
I take this opportunity of mentioning that Th. I and the sketch o f its
proof was only added to the present work after it had already gone to press.
At the time the work was presented at the Warsaw Society of Sciences (21
March 1931), Gddei’s article—so far as I know—had not yet appeared. In
this place therefore I had originally expressed, instead of positive results, only
certain suppositions in the same direction, which were based partly on my own
investigations and partly on the short report, Gddel, K. (21), which had been
published some months previously.
After I had become acquainted with the above mentioned article I con­
vinced myself, among other things, that the deductive theory which Gddel
248 CONCEPT OF TROTH IN FORMALIZED LANGUAGES V III, f 5

language is established in the language itself and in this way with

every sentence o f the metalanguage there is correlated, in one-
m any fashion, a sentence o f the language which is equivalent to
it (with reference to the axiom system adopted in the m eta­
theory); in this w ay the metalanguage contains as well as every
particular sentence, an individual name, if not for that sentence
at least for the sentence which is correlated with it and equivalent
to it. (2) Should we succeed in constructing in the metalanguage
a correct definition o f truth, then the metalanguage— with
reference to the above interpretation— would acquire that
universal character which was the prim ary source o f the seman­
tical antinomies in colloquial language (cf. p. 164). I t would
then be possible to reconstruct the antinom y o f the liar in the
metalanguage, b y form ing in the language itself a sentence x
such that the sentence o f the metalanguage which is correlated
with x asserts that x is not a true sentence. In doing this it would
be possible, by applying the diagonal procedure1from the theory
o f sets, to avoid all term s which do not belong to the m eta­
language, as well as all premisses o f an em pirical nature which
have played a part in the previous form ulations o f the antinom y
o f the liar.2

had chosen as the object of his studies, which he called the ‘ system P ’, was
strikingly similar to the general theory of classes considered in the present
section. Apart from certain differences of a ' calligraphicaT nature, the only
distinction lies in the fact that in the system P, in addition to three logioal
constants, certain constants belonging to the arithmetic of the natural numbers
also occur (a far-reaching analogy also exists between the system P and the
system of arithmetic sketched in V I (see pp. 113-16)). Consequently the results
obtained for the system P can easily be carried over to the present discussion.
Moreover, the abstract character of the methods used by G<5del renders the
validity of his results independent to a high degree of the specific peculiarities
of the science investigated.
1 Cf. Fraenkel, A . (10), pp. 48 ff.
8 I f we analyse the sketch of the proof given below we easily note that an
analogous reconstruction could be carried out even on the basis of colloquial
language, and that in consequence of this reconstruction the antinomy of
the liar actually approximates to the antinomy of the expression ‘hetero-
logic&r. For a rather simple reconstruction of the antinomy of the liar in this
direction see Tarski, A. (82), note 11, p. 371. It seems interesting that in this
reconstruction all the technical devices are avoided which are used in the proof
of Th. 1 (such as interpretation of the metalanguage in arithmetic or the
diagonal procedure). In connexion with the last paragraph of the text cf.
the concluding remarks of $ 1, pp. 164 f., and in particular p. 165, note 1.
VHI, § 5 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES 249

W e shall sketch the p roof a little more exactly.1

L et us agree for the m om ent to use the sym bol V instead o f
‘-XT*. The existential quantisation o f the sentential function y
with respect to the variable ‘ n' will be denoted b y the sym bol
‘ U f y ’ as before. The variable ‘w’ thus represents names o f
classes the elements o f which are classes o f individuals. Am ong
these classes we find, am ong other things, the natural numbers
and generally speaking the cardinal numbers.2
I have alrea dy m entioned that all facts belonging to the arith­
m etic o f the natural numbers can be expressed in the language o f
the general theory o f classes. In particular, if a natural number
k is given, a sentential function tk is easily constructed in this
language containing the sym bol V as the only free variable and
which asserts that the class whose name is represented b y this
sym bol is identical with the number k (and thus consists o f just
those classes o f individuals which have exactly Jc elem ents).2
F or exam ple:

Li ^ n?(4i.U i H I H 8 ( 4 i - (43+4i+4«))+
+ < 1 •n i U ! U i ( 4 i + . 4.1 . S ) ) .
A general recursive definition o f the sequence o f functions ik
within the metalanguage presents no great difficulty.
As I have already pointed out in § 2 (p. 184) a one-one corre­
spondence can be set up w ithout difficulty between the expres­
sions o f the language and the natural numbers ; we can define in the
metalanguage an infinite sequence <j>o f expressions in which every
expression o f the language occurs once and only once. W ith
the help o f this correlation we can correlate with every operation
on expressions an operation on natural numbers (which possesses
the same form al properties), with every class o f expressions a
class o f natural numbers, and so on. In this way the m eta­
language receives an interpretation in the arithm etic o f the
1 For the sake of simplicity we shall in many places express ourselves as
though the demonstration which follows belonged to the metatheory and not
to the meta-metatheory; in particular, instead of saying that a given sentence
is provable in the metatheory, we shall simply assert the sentence itself. In
any case it must not be forgotten that only a sketch of the proof is given here
and one whioh is far from complete.
1 See p. 233, note 1.
250 C O N C E PT O F T R U T H IN F O R M A L IZ E D L A N G U A G E S V H I, { 5

natural numbers and indirectly in the language o f the general

theory o f classes.
L et us suppose that we hare defined the class Tr o f sentences
in the metalanguage. There w ould then correspond to this olass a
class o f natural numbers whioh is defined exclusively in the terms
o f arithm etic. Consider the expression e Tr\ This is
a sentential function o f the metalanguage which contains V as
the only free variable. From the previous remarks it follow s that
with this function we can correlate another function which is
equivalent to it fo r any value o f ‘n\ but which is expressed
com pletely in terms o f arithm etic. W e shall write this new
function in the schem atic form Thus we have:
(1 ) for any n, U ? ( v 4>n) § Tr if and onto if M;n).
Since the language o f the general theory o f classes suffices
for the foundation o f the arithm etic o f the natural numbers, we
can assume that y*(a)’ is one o f the functions* o f this language.
T he function ty(n)* will thus be a term o f the sequence <£, e.g,
the te n s with the index k, ‘4>{nY « y ?e. I f we substitute ‘k’ for
V in the sentence (1) we obtain:
(2) U f(i;., <}tk) e Tr i f and only if \ji{k).
The sym bol d eletes, o f course., a sentence o f the
language under investigation. B y applying to this sentence
condition {a} o f the convention T we obtain a sentence o f the
form s'£ e Tr if and only if p\ where V is to be replaced by a
structural-descriptive or any other individual name o f the
statem ent U !(l**^s)> but ‘p ’ b y this statem ent itself or b y any
statem ent which is equivalent to it. In particular we can sub­
stitute V t M k V f° r x> and for ‘p }— hi view o f the meaning o f
the sym bol — the statem ent 'there is an a such that n = k
and $ (» )' or, sim ply *0(A)’ . In this w ay we obtain the following
form ulation:

(8) U i(‘* ■4>k) e Tr V an&only »/ <£(*)•

The sentences (2) and (S) stand in palpable contradiction to
one another; the sentence (2) is in fact directly equivalent to the
negation o f (3). In this w ay we have proved the first part of
V m , § 5 CONCEPT OP TRTJTH LSI FORMALIZED LANGUAGES 251

the theorem . W e have proved that am ong the consequences o f

the definition o f the sym bol ‘ 7V’ the negation o f one o f the sen­
tences m entioned in the condition {«) o f the convention T m ust
appear. From this the second part o f the theorem im m ediately
follow s.
The assum ption o f consistency appearing In the part (j3) o f
this theorem is essential. I f the class o f all provable sentences
o f the m etatheory contained a contradiction, then every de­
finition in the m etatheory w ould have among its consequences
all possible sentences (since they all would be provable in the
m etatheory), in particular those described in the convention
T . On the other hand, as we now know ,1 there is no prospect o f
proving the consistency o f the m etatheory which we are working
with, on the basis o f the m eta-m etatheory. It is to be noted
that, in view o f the existence o f an interpretation o f the m eta­
theory in the science itself (a fa ct which baa played such an
essential part in the p roof sketched above), the assumption o f
the second part o f Th. I is equivalent to the assumption c-{ the
consistency o f the scienoe investigated itself and from the
intuitive standpoint is just as evident.
The result reached in Th. I seems perhaps at first sight un­
com m only paradoxical. This impression will, doubtless be
weakened as soon as w e recall the fundam ental distinction
between the content o f the concept to be defined and the nature
o f those concepts which are at our disposal for the construction
o f the definition.
H ie metalanguage in which we carry out the investigation
contains exclusively structural-descriptive terms, such as names
o f expressions o f the language, structural properties o f these
expressions, structural relations between expressions, and so
on, as well as expressions o f a logical kind am ong which (in the
present case) we find all the expressions o f the language studied.
W hat we call m etatheory is, fundam entally, the morphology
of language— a science o f the form o f expressions— a correlate
o f such parts o f traditional grammar as m orphology, etym ology,
and syntax.
1 Cf. Godel, K. (22), ?. 196 (Th. XI).
252 CONCEPT OF TROTH IN FORMALIZED LANGUAGES

The fact that the language studied and the deductive science
carried out in this language are form alized has brought about an
interesting phenom enon; it has been possible to reduce to struc­
tural-descriptive concepts certain other notions o f a totally
different kind, which are distinguished from the form er both in
their origin and in their usual meaning, nam ely the concept o f
consequence together with a series o f related notions.1 It has
been possible to establish what m ay be called the logic of the
given science as a part o f m orphology.
Encouraged b y this success we have attem pted to go further
and to construct in the metalanguage definitions o f certain
concepts belonging to another dom ain, nam ely that called the
semantics of language— i.e. such concepts as satisfaction, de­
noting, truth, definability, and so on. A characteristic feature
o f the sem antical concepts is that they give expression to
certain relations between the expressions o f language and the
objects about which these expressions speak, or that by means
o f such relations they characterize certain classes o f expressions
or other objects. W e could also say (making use o f the suppositio
materiaMs) that these concepts serve to set up the correlation
between the names o f expressions and the expressions them ­
selves.
For a long tim e the semantical concepts have had an evil
reputation among specialists in the study o f language. They have
resisted all attem pts to define their meaning exactly, and the
properties o f these concepts, apparently so clear in their content,
have led to paradoxes and antinomies. F or that reason the
tendency to reduce these concepts to structural-descriptive ones
must seem quite natural and well-founded. The following fact
seemed to favour the possibility o f realizing this tendency:
it has always been possible to replace every phrase which con­
tains these sem antical terms, and which concerns particular
1 The reduction of the concept of consequence to concepts belonging to
the morphology of language is a result of the deductive method in its latest
stages of development. When, in everyday life, we say that a sentence fol­
lows from other sentences we no doubt mean something quite different from
the existence of certain structural relations between these sentences. In the
light of the latest results of Godel it seems doubtful- whether this reduction
has been effected without remainder.
V m , § 5 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES 253

structurally described expressions o f the language, b y a phrase

which is equivalent in content and is free from such terms. In
other words it is possible to form ulate infinitely m any partial
definitions for every sem antical concept, which in their totality
exhaust all cases o f the application o f the concept to concrete
expressions and o f which the sentences adduced in condition (a)
o f convention T are exam ples. It was with just this end in view
that, as a rule, we included in the metalanguage, with regard
to the content o f the sem antical concepts, not only the names
o f expressions but all expressions o f the language itself or ex ­
pressions having the same meaning (even when these expressions
were not o f a logical kind, cf. pp. 210 f.), although such an enrich­
m ent o f the metalanguage has no advantages for the pursuit o f
the ‘pure’ m orphology o f language.
In the abstract the fact m entioned has no decisive im portance;
it offers no path by which an autom atic transition from the
partial definitions to a general definition is possible, which
embraces them all as special cases and would form their infinite
logical product.1 Only thanks to the special m ethods o f con­
struction which we developed in §§ 3 and 4 have we succeeded in
carrying out the required reduction o f the semantical concepts,
and then only for a specified group o f languages which are poor in
gram m atical form s and have a restricted equipm ent o f seman­
tical categories— nam ely the languages o f finite order. Let it
be remem bered that the m ethods there applied required the use
in the metalanguage o f categories o f higher order than all
categories o f the language studied and are for that reason funda­
m entally different from all gram m atical form s o f this language.
The analysis o f the p roof o f Th. I sketched above shows that this
1 In the course of our investigation we have repeatedly encountered similar
phenomena: the impossibility of grasping the simultaneous dependence be­
tween objects which belong to infinitely many semantical categories; the lack
of terms of ‘ infinite order *; the impossibility of including, in on* process of
definition, infinitely many concepts, and so on (pp. 188 f., 232 f., 243, 245).
I do not believe that these phenomena can be viewed as a symptom of the
formal incompleteness of the actually existing languages— their cause is to be
sought rather in the nature of language itself: language, which is a product
of human activity, necessarily possesses a ‘ tmitistic1 character, and cannot
serve as an adequate tool for the investigation of facte, or for the construction
of concepts, of an eminently 4infinitistic * character.
264 CONCEPT OP TRUTH IN FORMALIZED LANGUAGES VTH, § 6

oironmstanee is n ot an accidental one. Under certain general

assumptions, it proves to be im possible to construct a correct;
definition o f truth i f only suoh categories are used which appear m
the language under consideration.1 F or that reason the situation
had fundam entally changed when we passed to the ‘rich ’ lan­
guages o f infinite order. The m ethods used earlier proved to be
inapplicable; all concepts and all gram m atical form s o f the meta­
language found an interpretation in the language and hence we
were able to show conclusively that the semantics o f the language
could not be established as a part o f its m orphology. The signifi­
cance o f the results reaohed reduces just to this.
But. apart’ from this, Th, I has im portant consequences o f a
m ethodological nature. It shows that- it is im possible to define in
the m statheory a class o f sentences o f the language studied
which constats exclusively o f m aterially true sentences and is at
the same tim e com plete (in the sense o f D ef. 20 in § 8). In par­
ticular-; i f we enlarge the class o f provable sentences o f the science
investigated in any way— whether b y supplementing the list o f
axiom s or b y sharpening the rules o f inference— then we either
add false sentences to this olass or w© obtain an incom plete
system. This is all the m ore interesting inasmuch as the enlarge-

1 F t b l ' ; , o»* ^amrdi&tely from certain results contained in Godel, X , (22)

(pp. IS? -91}* -t oao easily be inferred that & structural definition of truth-—
in the cease discussed or vp . 236 especially on p* 237, note 2 —cannot be
constructed even fur languages of finite order which are in aam*i degree richer.
From other investigations of this author (cp. cit., p. 193; Th, IX ) it follows
that in certain elementary oases in which we can construct- such a definition,
it is nevertheless impossible to give a general structural criterion of the truth
o f & sentence. The first of these result* applies, for instance, to the logic of
two-termed and many-tanned relations discussed in § 4. The second result
applies, for example, to the lower predicate calculus {‘engero Fimkticnen-
kalkul*) of HUbert'Aohnrmann (80}> pp. 43 ff.; in this case, however, the
result is applied^ not to the notion of a true sentence, but to the related notion
of a universally valid (‘aligemeingfiltig*) sentential function.
At this point we should like to call attention to the dose connexion between
the notions of ’ structural definition of truth', and oi *general structural
criterion of truth' discussed in this work, and the notions of recursive enu­
merability and general recursiveness known from the recent literature (see,
for example, Mostowskl, A. (6 3 /), chap. 6). In fact, by saying that there is a
£structural definition of truth* for a given formalized theory we essentially
mean that the set of all true sentences of this theory is recursively enu­
merable ; when, we say that there is a ‘ general structural criterion of truth* we
mesa that iho not of all true sentences is general recursive.
v n i, § t> CONCEPT O F TRUTH IN FORMALIZED LANGUAGES 265

ment o f the class o f provable sentences to form a com plete and

consistent system in itself presents no difficulties.1
A n interpretation o f Th. I which went beyond the lim its
given w ould n ot be justified. In particular it w ould be incorrect
to infer the im possibility o f operating consistently and in agree­
m ent with intuition with sem antical concepts and especially with
the concept o f truth. B ut since one o f the possible ways o f con ­
structing the scientific foundations o f semantics is closed we
m ust look for other m ethods. The idea naturally suggests itself
o f setting up semantics as a special deductive science with a
system o f m orphology as its logical substructure. F or this
purpose it w ould be necessary to introduce into m orphology a
given sem antical notion as an undefined concept and to estab­
lish its fundam ental properties by means o f axiom s. The
experience gained from the study o f semantical concepts in
connexion with colloquial language, warns us o f the great
dangers that m ay accom pany the use o f this m ethod. F or that
reason the question o f how we can be certain that the axiom atic
m ethod will not in this case lead to com plications and antinomies
becom es especially im portant.
In discussing this question I shall restrict m yself to the theory
o f truth, and in the first place I shell establish the follow ing
theorem , which is a consequence o f the discussion in the pre­
ceding section:

Theorem IT. For an arbitrary, previously given n itvrol number

le, it is possible to construct a definition of the- symbol ‘ 2 V on the
basis of the meiaiheory, which has among its nav^equences all- those
sentences from the condition (*) of the convention T in which in
the place of the symbol ‘p ‘ sentences with variables c f at most the
k-th order occur {and moreover, the sentence adduced in the condition
(?) of this convention).
B y way o f p roof it suffices to remark that this theorem no
longer concerns the language studied in its whole extent but only
a fragm ent o f it which em braces all those expressions which
contain no variable o f higher order than the kth. This fragment
1 Cf. V, Th. 66, ft result of Lindenbe urn's (see p, 5S of the present volume).
256 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES VIH, § 5

is clearly a language o f finite order and in fact a language o f the

2nd kind. W e can therefore easily construct the required
definition b y applying one o f the tw o m ethods described in § 4.
I t is to be noted that the definition obtained in this way (together
with the consequences given in Th. II) yields a series o f theorems
o f a general nature, like the Ths. 1-5 in § 3, for example, if the
form ulations o f these theorem s are suitably weakened by re­
stricting the dom ain o f their applicability to sentences with
variables o f at m ost the k t h order.
H ence it w ill be seen that, in contrast to the theory o f truth
in its totality, the single fragm ents o f this theory (the objects o f
investigation o f which are sentences which contain only variables
whose order is bounded above) can be established as fragments
o f the m etatheory. If. therefore, the m etatheory is consistent
we shall not find a contradiction in these fragments. This last
result can be extended in a certain sense to the whole theory o f
truth, as the follow ing theorem shows:

T heorem III. I f the class o f a ll provable sentences o f the m eta-

th eory is con sisten t and i f w e add to the m etatheory the sym bol
*Tr* as a new p rim itiv e sig n , and all theorem s w hich are described
in con d ition s (a) and (jS) o f the con ven tion T as new axiom s, then
the class o f provable sentences in the m etatheory enlarged in this w ay
w ill also he con sisten t ,

To prove this theorem we note that the condition (a) contains

infinitely m any sentences which are taken as axiom s o f the
theory o f truth. A finite num ber o f these axiom s— even in union
with the single axiom from condition (,8)— cannot lead to a
contradiction (so long as there is no contradiction already in the
m etatheory). A ctually in the finite number o f axioms obtained
from (a) only a finite number o f sentences o f the language studied
appears and in these sentences we find a finite number o f vari­
ables. There must, therefore, be a natural number k such that the
order o f none o f these variables exceeds &. From this it follows,
b y Th. II, that a definition o f the sym bol ‘ 2 V can be constructed
in the m etatheory such that the axiom s in question becom e con­
sequences o f this definition. In other words: these axioms, with
V in , § 6 C O N C E PT O F T R U T H IN FORMALIZED LANGUAGES 257

a suitable interpretation o f the sym bol ‘ T V , becom e provable

sentences o f the m etatheory (this fact can also be estabHshed
directly, i.e. independently o f Th. I I). I f any class o f sentences
contains a contradiction, it is easy to show that the contra­
diction m ust appear in a finite part o f this class.1 Since, how ever,
no finite part o f the axiom system described in Th. I l l contains
a contradiction, the whole system is consistent, which was to be
proved.
The value o f the result obtained is considerably diminished
b y the fa ct that the axiom s m entioned in Th. I l l have a very
restricted deductive power. A theory o f truth founded on them
w ould be a highly incom plete system , which would lack the m ost
im portant and m ost fruitful general theorem s. L et us show this
in m ore detail b y a concrete exam ple. Consider the sentential
function ' x e T r o r x e Tr\ I f in this function we substitute for
the variable V structural-descriptive names o f sentences, we
obtain an infinite num ber o f theorem s, the p roof o f which on
the basis o f the axiom s obtained from the convention T presents
not the slightest difficulty. B ut the situation changes funda­
m entally as soon as we pass to the generalization o f this sen­
tential function, i.e. to the general principle o f contradiction,
From the intuitive standpoint the truth o f all those theorems is
itself already a p roof o f the general p rin cip le; this principle
represents, so to speak, an ‘infinite logical p rodu ct’ o f those
special theorem s. B ut this does not at all mean that we can
actually derive the principle o f contradiction from the axiom s or
theorems m entioned b y means o f the norm al modes o f inference
usually em ployed. On the contrary, b y a slight m odification in
the p roof o f Th. I l l it can be shown that the principle o f contra­
diction iy not a consequence (at least in the existing sense o f the
word) o f the axiom system described.
W e could, o f course, now enlarge the above axiom system by
adding to it a series o f general sentences which are independent
o f this system . W e could take as new axiom s the principles o f
contradiction and excluded m iddle, as well as those sentences
which assert that the consequences o f true sentences are true,
1 Cf. V , Th. 48, p. 91 of the present volume.
258 C O N C E PT O F T R U T H . IN F O R M A L IZ E D LA N G U A G E S V III, § 5

and also that all prim itive sentences o f the science investigated
belong to the class o f true sentences. Th. I l l could be extended
to the axiom system enlarged in this w ay.3 B ut we attach little
im portance to this procedure. P er it seems that every such
enlargem ent o f the axiom system has an accidental character,
depending on rather inessential factors such, for exam ple, as the
actual state o f knowledge in this field. In any case, various
objective criteria which we should wish to apply in the choice o f
further axiom s prove to be quits inapplicable. Thus it seems
natural to require that the axiom s o f the theory o f truth, together
with the original axiom s o f the m etatheory, should constitute a
categorical system .8 I t can be shown that this postulate co­
incides in the present case with another postulate, according to
which the axiom system o f the theory o f truth should un­
am biguously determ ine the extension o f the sym bol ‘ Tr’ which
occurs in it, and in the follow ing sense: if we introduce in to the
m etatheory, alongside this sym bol, another prim itive sign, e.g.
the sym bol iTr” and set up analogous axiom s for it, then the
statem ent ‘ Tr --- Tr” must be provable. B ut this postulate
cannot be satisfied. F or it is n ot difficult to prove that in the
contrary case the concept o f tr uth could be defined exclusively
b y means o f term s belonging to the m orphology o f language,
which w ould be in palpable contradiction to Th. I . F or other
reasons o f a m ore general nat ure there can be no question o f an
axiom system that w ould be com plete and w ould consequently
suffice for the solution o f every problem from the dom ain o f the
theory under consideration. This is an im m ediate m ethodo­
logical consequence o f Th. I applied n ot to the language o f the
general theory o f classes but to the richer language o f the meta­
theory and the theory o f truth, (cf. the remarks on p. 254).
There is, how ever, quite a different w ay in w hich the founda­
tions o f the theory o f truth m ay be essentially strengthened.
1 For this purpose we must nevertheless to some extent* sharpen the pre­
misses of the theorem by assuming that che class of all provable sentences
of the metatheory is not only consistent, but also consistent in the sense
of Godel, K. (22), p* 187, or in other words, that this class remains consistent
after a single application of the rule of infinite induction, which will be dis­
cussed below.
a See p. 174, note 1.
V E X § 5 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES 259

The fact that we cannot infer from the correctness o f all sub­
stitutions o f such a sentential function as ‘x e T r or x e T r ’ the
correctness o f the sentence which is the generalization o f this
function, can be regarded as a sym ptom o f a certain im perfection
in the rules o f inference hitherto used in the deductive sciences.
In order to make good this defect we could adopt a new rule,
the so-called rule of infinite induction, which in its application
to the m etatheory m ay be form ulated somewhat as follow s: i f a
given sentential function contains the sym bol V , which belongs
to the same sem antical category as the names o f expressions,
as its only free variable, and if every sentence, which arises
from the given function b y substituting the structural-descrip­
tive name o f any expression o f the language investigated for the
variable (x\ is a provable theorem o f the m etatheory, then the
sentence which we obtain from the phrase for every z , if x is an
expression then p y b y substituting the given function for the
sym bol (p\ m ay also be added to the theorems o f the m etatheory.
This rule can also be given another form ulation which differs
from the foregoing only b y the fact that in it, instead o f speaking
about expressions, we speak o f natural numbers; and instead o f
structural-descriptive names o f eg ression s, the so-called specific
sym bols o f natural numbers are dealt with, i.e. such sym bols
as *0J, ‘ l 5, ‘ 1 + 1’ , ‘ i + H - r , and so on. i n this form the rule o f
infinite induction recalls the principle o f com plete induction,
which it exceeds considerably in logical power. Since it is possible
to set up effectively a one-one correspondence between expres­
sions and the natural numbers (of. the p roof o f Th. I) it is easy to
see that the tw o form ulations are equivalent on the basis o f the
m etatheory. B ut in the second form ulation no specific concepts
o f the metalanguage occur at all, and for this reason it is applic­
able to m any other deductive sciences. In the case where we are
dealing with a science in the language o f which there are no
specific sym bols for the natural numbers this form ulation re­
quires certain external m odifications. For example, in order to
form ulate the rale for the general theory o f classes, instead o f
substitutions o f a given sentential function we must operate
with expressions o f the type ^here, in the place o f
260 CONCEPT OP TRUTH IN FORMALIZED LANGUAGES V m , 5 6

‘p ' the function in question occurs and the sym bol ‘tfc’ has the
same meaning as in the p roof o f Th. I .1
On account o f its non-finitist nature the rule o f infinite in­
duction differs fundam entally from the norm al rules o f infer­
ence. On each occasion o f its use infinitely m any sentences must
be taken in to consideration, although at no m om ent in the
developm ent o f a science is such a num ber o f previously proved
theorem s effectively given. I t m ay well be doubted whether
there is any place for the use o f such a rule w ithin the lim its o f
the existing conception o f the deductive m ethod. The question
whether this rule does not lead to contradictions presents no less
serious difficulties than the analogous problem regarding the
existing rules, even if we assume the consistency o f the existing
rules and perm it the use o f the new rule n ot only in the theory
bu t also in the corresponding m etatheory and in particular in
any attem pted p ro o f o f consistency. Nevertheless from the in­
tuitive standpoint the rule o f infinite induction seems to be
as reliable as the rules norm ally applied: it always leads from
true sentences to true sentences. In connexion w ith languages
o f finite order this fa ct can be strictly proved b y means o f the
definition o f truth constructed for these languages. The fact that
this rule enables m any problem s to be solved which are not
solvable on the basis o f the old rules is in favour o f the accept­
ance o f the new rule, n ot only in the theory but also in the m eta­
theory. B y the introduction o f this rule the class o f provable
sentences is enlarged b y a m uch greater extent than b y any
supplem entation o f the list o f axiom s.2 In the case o f certain
elem entary deductive sciences, this enlargem ent is so great that
the class o f theorem s becom es a com plete system and coincides
1 I have previously pointed out the importance of the rule of infinite in­
duction in the year 1926. In a report to the Second Polish Philosophical Con­
gress, in 1927, I have given, among other things, a simple example of a con­
sistent deductive system which after a single application of this rule ceases to
be consistent, and is therefore not ^-consistent (cf. p. 258, note 1; see also IX,
p. 282, note 2). Some remarks on this rule are to be found in Hilbert, D.
(29), pp. 491-2.
* Thus, for example, if we adopt this rule in the metalanguage without
including it in the language, we can prove that the class of provable sentences
of the science is consistent, which previously was not possible. In connexion
with this problem cf. Gfidel, K . (22), pp. 187-91 and 196.
V m , § 5 CONCEPT OP TRUTH IN FORMALIZED LANGUAGES 2*1

w ith the class o f true sentences. Elem entary number theory

provides an exam ple, nam ely, the science in which all variables
represent names o f natural or whole numbers and the constants
are the signs from the sentential and predicate calculi, the signs
o f zero, one, equality, sum, product and possibly other signs
defined w ith their h elp .f
I f it is decided to adopt the rule o f infinite induction in the
m etatheory, then the system o f axiom s to which Th. I l l refers
already form s a sufficient foundation for the developm ent o f
the theory o f truth. The p roof o f any o f the known theorems
in this field will then present no difficulty, in particular the
Ths. 1-6 in § 3 and the theorem according to which the rule o f
infinite induction when applied to true sentences always yields
true sentences. More im portant still, these axiom s, together
w ith the general axiom s o f the m etatheory, form a categorical
(although not a com plete) system , and determine unam biguously
the extension o f the sym bol ‘ I V which occurs in them .
Under these circum stances the question whether the theory
erected on these foundations contains no inner contradiction
acquires a special im portance. U nfortunately this question
cannot be finally decided at present. Th. I retains its full va lid ity:
in spite o f the strengthening o f the foundations o f the m etatheory
the theory o f truth cannot be constructed as a part o f the m orpho­
logy o f language. On the other hand for the present we cannot
prove Th. I l l for the enlarged metalanguage. The premiss which
has played the m ost essential part in the original proof, i.e. the
reduction o f the consistency o f the infinite axiom system to the
consistency o f every finite part o f this system , now com pletely
loses its validity— as is easily seen— on account o f the content o f
the new ly adopted rule. The possibility that the question cannot
be decided in any direction is not excluded (at least on the basis o f
a ‘norm ar system o f the m eta-m etatheory, which is constructed
t This last remark enables us to construct a rather simple definition of
truth for elementary number theory without using our general method. The
definition thus constructed can be further simplified. In fact we can first
structurally describe all true sentences which contain no variables (or quanti­
fiers), and then define an arbitrary sentence to be true if and only if it can
be obtained from those elementary true sentences by applying the rule of
infinite induction arbitrarily many times.
262 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES VIH, § 6

according to the principles given at the beginning o f § 4 and does

not contain the semantics o f the metalanguage). On the other
hand the possibility o f showing Th. I l l to be false in its new
interpretation seems to be unlikely from the intuitive view point.
One thing seems clear: the antinom y o f the liar cannot be directly
reconstructed either in the form ulation m et with in § I or in the
form in which it appeared in the p roof o f Th. I. F or here the
axiom s adopted in the theory o f truth clearly possess,in contrast to
colloquial language, the character o f partial definitions. Through
the introduction o f the sym bol ‘ TV the metalanguage does not
in any w ay becom e sem antically universal, it does not coinoide
with the language itself and cannot be interpreted in that
language (cf. pp. 158 and 248).1
N o serious obstacles stand in the way o f the application o f the
results obtained to other languages o f infinite order. This is
especially true o f the m ost im portant o f these results— Th. I.
The languages o f infinite order, thanks to the variety o f m ean­
ingful expressions contained in them , provide sufficient means
for the form ulation o f every sentence belonging to the arithm etic
1 This last problem is equivalent to a seemingly more general problem of
a methodological nature which can be formulated as follows. Wo presuppose
the consistency of the metatheory supplemented by the rule of infinite in­
duction. We consider an infinite sequence t of sentences of the metatheory;
further we take into the metatheory a new primitive sign *N\ and add
as axioms those and only those sentences which are obtained from the
scheme e N if and. only if p ' by substituting for the sign *n* the £th specific
symbol of the natsiral numbers (i.e. the expression composed of k signs *1 *
separated from one another by the signs ‘ 4- *) and for the sign tp i the ktb
term of the sequence t {/c being here an arbitrary natural number). The question
now arises whether the class of provable sentences of the metatheory, when
enlarged in this way, remains consistent. This problem may be called the
problem of infinite recursive definitions. The axiom system described in it
can— from the intuitive standpoint— be regarded as a definition sui generis
of the symbol ‘ A7’, which is distinguished from normal definitions only by the
fact that it is formulated in infinitely many sentences. In view of this character
of the axioms the possibility of a negative solution of the problem does not
seem very probable. F r o m Th. II and the interpretation of the metatheory in
the theory itself, it is not difficult to infer that this problem can be solved in
a positive sense in those cases in which the order of all variables which occur
in the sentences of the sequence t is bounded above. It is then even possible
to construct a definition of the sym bol4N * in the metatheory such that ell the
axioms mentioned follow from it. This problem obviously does not depend on
the specific properties of the motatheory as such; it can also be presented in the
game or in a somewhat modified form for other deductive sciences, e.g. for the
general theory of classes.
V m , | 5 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES 2.33

o f natural numbers and consequently enable the metalanguage

to be interpreted in the language itself. I t is thanks to just this
circum stance that Th. I retains its validity for all languages o f
this kind.1
Som e remarks m ay be added about those cases in which not
single languages bu t whole classes o f languages are investigated.
A s I have already em phasized in the Introduction, the concept
o f truth essentially depends, as regards both extension and
content, upon the language to w hich it is applied. W e can only
m eaningfully say o f an expression that it is true or not if we treat
this expression as a part o f a ooncrete language. As soon as the
discussion is about a large num ber o f languages the expression
'true sentence’ ceases to be unambiguous. I f wo are to avoid this
am biguity we m ust replace it b y the relative term true sen­
tence w ith respect to the given language’ . In order to make the
sense o f this term precise we apply to it essentially the same pro­
cedure as before: we construct a com m on metalanguage for all
the languages o f the given class; within the metalanguage we
try to define the expression in question with the help o f the
m ethods developed in §§ 3 and 4. I f we are not successful we add
this term to the fundam ental expressions o f the metalanguage
and b y the axiom atic m ethod determ ine its meaning according
to the instructions o f Th. H I o f this section. On account o f the
rdativization, o f this term we should nevertheless expect a -priori
that in carrying out the plan sketched above the earlier diffi­
culties w ould be significantly increased and quite new com plica­
tions m ight arise (connected for exam ple with the necessity o f
1 A reservation is necessary here: if we choose as our starting-point the
classification cf semantical categories sketched on p. 218. nets 2, then we
again encounter languages of infinite order for which Th, T loses its validity.
A typical example is furnished by the language of Le&uiewski’s Protothetic
(ct. Lesniewski, S. (46)). In consequence of tho •finitistio* character of all the
semantical categories of this language, it is easy to construct, in the meta­
language, a correct definition of truth, by choosing as model the matrix method
from the extended sentential calculus. Moreover, such a definition can be
obtained in other ways; as Le&vieweki has shown, the. class of provable
sentences of the protothetic ie complete, and therefore the concept of provable
sentence coincides in its extension with that of true sentence. Th. I on the
other hand applies without restriction to all languages in which the order of
the semantical categories from the domain of Le&oiewskFs Ontology (cf.
Lefaiewski, S. (47)); is not bounded above.
264 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES VIII, § 6

defining the w ord lan gu age’ ). I do not propose to discuss the

problem touched upon in m ore detail in this place. The prospects
for such investigations at the present tim e seem to be rather
lim ited. In particular it w ould be incorrect to suppose that the
relativization o f the concept o f truth— in the direction m entioned
above— w ould open the w ay to som e general theory o f this
ooncept which w ould em brace all possible or at least all form al­
ized languages. The class o f languages which is chosen as th e
ob ject o f sim ultaneous study m ust n ot be too wide. I f, for
exam ple, we include in this class the m etalanguage, which form s
the field o f the investigations and already contains the conoept
o f truth, we autom atically create the conditions which enable
the antinom y o f the liar to be reconstructed. The language o f
the general theory o f truth w ould then contain a contradiction
for exactly the same reason as does colloquial language.
In conclusion it m ay be m entioned that the results obtained
can be extended to other sem antical concepts, e.g. to the concept
o f satisfaction. F or each o f these concepts a system o f postulates
can be set up w hich (1) contains partial definitions analogous to
the statem ents described in condition (a) o f the convention T
which determ ine the m eaning o f the given concept w ith respeot
to all concrete, structurally described expressions o f a given
class (e.g. w ith respect to sentences or sentential functions o f a
specific sem antical typ e), and (2) contains a further postulate
w hich corresponds to the sentence from the condition (jS) o f the
same convention and stipulates that the concept in question
m ay be applied only to expressions o f the given class* W e should
be prepared to regard such a definition o f the concept studied as
a m aterially adequate one if its consequences inoluded all the
postulates o f the above system . M ethods which are similar to
those described in §§ 3 and 4 enable the required definition to be
constructed in all cases where we are dealing w ith languages of
finite order, or, m ore generally, in which the sem antical concept
studied concerns exclusively linguistic expressions in which the
order o f the variables is bounded above (of. Th. I I). In the
remaining cases it can be shown— after the pattern o f the p roof o f
Th. I — that no definition w ith the properties m entioned can be
V in , § 6 CONCEPT OP TROTH IN FORMALIZED LANGUAGES 265

formulated in the metalanguage.1 In order to construct the

theory of the concept studied in these cases also, it must be in­
cluded in the system of primitive concepts, and the postulate
described above must be inoluded in the axiom system of the
metatheory. A procedure analogous to the proof of Th. Ill
proves that the system of the metalanguage supplementedin this
•way remains internally consistent. But the deductive power
of the added postulates is very restricted. They do not suffice
for the proof of the most important general theorems concerning
the concept in question. They do not determine its extension
unambiguously and the system obtained is not complete, nor
is it even categorical. To remove this defect we must strengthen
the foundations of the metatheory itself by adding the rule of
infinite induction to its rules of inference. But then the proof
of consistency would present great difficulties which we are not
able at present to overcome.

§ 6 . Su m m a r y
The principal results of this article may be summarized in the
following theses:
A. For every formalized language of finite order a formally
correct and materially adequate definition of true sentence can be
constructed in the metalanguage, making use only of expressions of
a general logical kind, expressions of the language itself as well as
terms belonging to the morphology of language, i.e. names of lin­
guistic expressions and of the structural relations existing between
them.
B. For formalized languages of infinite order the construction
of such a definition is impossible.

1 This especially concerns the concept of definability (although in this oase

both the formulation of the problem itself, as well as the method of solution,
require certain modifications in comparison with the scheme put forward in
the text). In V I, I have expressed the conjecture that it is impossible to
define this concept in its full extent on the basis of the metalanguage. I can
now prove this conjecture exactly. This fact is all the more noteworthy in
that it is possible— as I have shown in the article mentioned— to construct
the definitions of the particular oases of the concept of definability which
apply, not to the whole language, but to any of its fragments of finite order,
not only in the metalanguage but also in the language itself.
206 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES VIII, § 0

C. On the other hand, even with respect to formalized languages

of infinite order, the consistent and correct use of the concept of truth
is rendered possible by including this concept in the system ofprimi*
tive concepts of the metalanguage and determining its fundamental
properties by means of the axiomatic method (the question whether
the theory o f truth established in this w ay contains no contra­
diction remains for the present undecided).
Since the results obtained can easily be extended to other
sem antical concepts the above theses can be given a m ore general
form : *

A '. The semantics of any formalized language of finite order can

be built up as a part of the morphology of language, based on corre­
spondingly constructed definitions.
B '. It is impossible to establish the semantics of the formalized
languages of infinite order in this way.
O'. But the semantics of any formalized language of infinite
order can be established as an independent science based upon its
own primitive concepts and its own axioms, possessing as its logical
foundation a system of the morphology of language (although a full
guarantee that the semantics constructed b y this m ethod con­
tains no inner contradiction is at present lacking).

From the form al point o f view the foregoing investigations

have been carried out within the boundaries o f the m ethodology
o f the deductive sciences. Some so to speak incidental results
will perhaps be o f interest to specialists in this field. I would
draw attention to the fa ct that with the definition o f true
sentence for deductive sciences o f finite order a general method
has been obtained for proving their consistency (a m ethod which,
how ever, does not add greatly to our knowledge). I w ould point
out also that it has been possible to define, for languages o f finite
order, the concepts o f correct sentence in a given and in an arbi­
trary individual dom ain— concepts which play a great part in
recent m ethodological studies.
B ut in its essential parts the present work deviates from the
main stream o f m ethodological investigations. Its central
VIII, § 6 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES 267

problem — the construction o f the definition o f true sentence

and establishing the scientific foundations o f the theory o f truth
— belongs to the theory o f knowledge and form s one o f the ohief
problem s o f this branch o f philosophy. I therefore hope that this
w ork will interest the student o f the theory o f knowledge above
all and that he w ill be able to analyse the results contained in it
critically and to judge their value for farther researches in this
field, w ithout allowing him self to be discouraged by the appara­
tus o f concepts and m ethods used here, which in places have been
difficult and have n ot hitherto been used in the field in which he
works.
One w ord in conclusion. Philosophers who are not accustom ed
to use deductive m ethods in their daily work are inclined to
regard all form alized languages with a certain disparagem ent,
because they contrast these ‘artificial* constructions with the
one natural language— the colloquial language. F or that reason
the fa ct that the results obtained concern the form alized lan­
guages alm ost exclusively will greatly diminish the value o f
the foregoing investigations in the opinion o f many readers. It
would be difficult foT me to share this view. In xny opinion the
considerations o f § 1 prove em phatically that the concept o f
truth (as well as other sem antical concepts) when applied to
colloquial language in conjunction w ith the norm al laws o f logic
leads inevitably to confusions and contradictions. W hoever
wishes, in spite o f all difficulties, to pursue the semantics o f
colloquial language with the help o f exact m ethods will be
driven first to undertake the thankless task o f a reform o f
this language. He will find it necessary to define its structure,
to overcom e the am biguity o f the terms which occur in it,
and finally to split the language into a series o f languages o f
greater and greater extent, each o f which stands in the same
relation to the next in which a form alized language stands to its
metalanguage. It m ay. however, bo doubted whether the
language o f everyday life, after being 'rationalized’ in this way,
w ould still preserve its naturalness and whether it would not
rather take on the characteristic features o f the form alized
languages.
268 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES VHI, § 7

§ 7. P ostscript
In writing the present article I had in m ind only form alized
languages possessing a structure which is in harm ony with
the theory o f sem antical categories and especially w ith its
basic.principles. This fact has exercised an essential influence on
the construction o f the whole w ork and on the form ulation o f its
final results. It seemed to me then that 'the theory o f the sem an­
tical categories penetrates so deeply into our fundam ental
intuitions regarding the meaningfulness o f expressions, that it is
hardly possible to im agine a scientific language whose sentences
possess a clear intuitive m eaning but whose structure cannot be
brought into harm ony with the theory in question in one o f its
form ulations’ (cf. p. 215). T oday I can no longer defend d e­
cisively the view I then took o f this question. In connexion with
this it now seems to me interesting and im portant to inquire
what the consequences w ould be for the basic problem s o f the
present work if we included in the field under consideration
form alized languages for w hich the fundam ental principles o f the
theory o f sem antical categories no longer hold. In what follow s
I w ill briefly consider this question.
A lthough in this w ay the field to be covered is essentially en­
larged, I do not intend— any m ore than previously— to consider
all possible languages which som eone m ight at some tim e con­
struct. On the contrary I shall restrict m yself exclusively to
languages which— apart from differences connected with the
theory o f sem antical categories— exhibit in their structure the
greatest possible analogy w ith the languages previously studied.
In particular, for the sake o f sim plicity, I shall consider only
those languages in which occur, in addition to the universal
and existential quantifiers and the constants o f the sentential
calculus, only individual names and the variables representing
them , as well as constant and variable sentence-form ing functors
with arbitrary numbers o f arguments. A fter the manner of
the procedure in §§ 2 and 4 we try to specify for each o f these
languages the concepts o f prim itive sentential function, funda­
m ental operations on expressions, sentential function in general,
V in , § 7 C O N C E PT O P T R U T H IN F O R M A L IZ E D L A N G U A G E S 269

axiom , consequence, and provable theorem . Thus, for exam ple,

w e include as a rule am ong the axiom s— ju st as in the language
o f the general theory o f classes in § 6— the substitutions o f the
axiom s o f the sentential caloulus, the pseudo-definitions and the
law o f extensionality (perhaps also other sentences, according to
the specific peculiarities o f the language). In determ ining the
concept o f consequence we take as our m odel D ef. 15 in § 2.
The concept introduced in § 4 o f the order o f an expression
plays a part w hich is no less essential than before in the con­
struction o f the language we are now considering. I t is advisable
to assign to names o f individuals and to the variables represent­
in g them the order 0 (and not as before the order 1). The order
o f a sentence-form ing fu nctor o f an arbitrary (prim itive) sen­
tential function is no longer unam biguously determ ined b y the
orders o f all arguments o f this function. Since the principles o f
the theory o f the sem antical categories no longer hold, it m ay
happen that one and the same sign plays the part o f a functor in
tw o or m ore sentential functions in which arguments occupying
respectively the same places nevertheless belong to different
orders. Thus in order to fix the order o f any sign we must take
in to account the orders o f all arguments in all sentential fu n c­
tions in w hich this sign is a sentence-form ing funotor. I f the
order o f all these arguments is smaller than a particular natural
num ber » , and if there occurs in at least one sentential function
an argument which is e x a ctly b f order n — 1, then we assign to the
sym bol in question th e order n. A ll such sentence-form ing
functors— as w ell as the names o f individuals and the variables
representing them — are inoluded am ong the signs o f finite order.
B ut aocount m ust also be taken o f the possibility that y et other
sentence-form ing functors m ay occur in the language to which
an infinite order m ust be assigned. If, for exam ple, a sign is a
sentence-form ing fu nctor o f only those sentential functions which
have all their argum ents o f finite order, where, how ever, these
orders are n ot bounded above b y any natural num ber, then this
sign w ill be o f infinite order.
In order to classify the signs o f infinite order we m ake use o f
the notion o f ordinal number, taken from the theory o f sets, which
270 C O N C E PT O P T R U T H IN F O R M A L IZ E D L A N G U A G E S V III, $ 7

is a generalization o f the usual concept of natural number.1

As is well known, the natural numbers are the smallest ordinal
numbers. Since, for every infinite sequence o f ordinal numbers,
there are numbers greater than every term o f the sequence,
there are, in particular, numbers which are greater than all
natural numbers. We call them tranefinite ordinal numbers. It is
known that in every non-empty class o f ordinal numbers there
is a smallest number. In particular there is a smallest transfinite
number which is denoted by the symbol ‘<i>\ The next largest
number is a » + l, then follow the numbers <0+2, < o+ 3 ,.„, <o.2,
< o.2 + l, <o,2+2,..., <o.3,..., and so on. To those signs o f infinite
order which are functors o f sentential functions containing
exclusively arguments o f finite order we assign the number <o
as their order. A sign which is a functor in only those sentential
functions in which the arguments are either o f finite order or o f
order <o (and in which at least one argument o f a function is
actually o f order to), is o f the order <o+l. The general recursive
definition o f order is as follows: the order o f a particular sign is
the smallest ordinal number which is greater than the orders o f all
arguments in. all sentential functions in which the given sign
occurs as a sentence-forming functor.2
Just as in§ 4, we candistinguishlanguages offinite andinfinite
order. We can in fact assign to every language a quite specific
ordinal number as its order, namely the smallest ordinal number
which exceeds the orders of all variables occurring in this lan­
guage (the former languages of the nth order—as can easily be
shown—retain their former order under this convention because
the order of the names of individuals has been diminished. The
language of the general theory of classes has the order a>).
It does not at all follow from these stipulations that every
variable in the languages in question is of a definite order. On
the contrary it seems to me (by reason o f trials and other con­
siderations) almost certain that we cannot restrict ourselves to
the use o f variables o f definite order if we are to obtain languages

1 Cf. Praankel, A . (16), pp. 185 ft.

* Cf. the introduction of the system of levels in Carnap, R . (10), pp. 139 ft
(p. 180 in English translation).
VIII, § 7 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES 271

which are actually superior to the previous languages in the

abundance o f the concepts which are expressible b y their means,
and the study o f which could throw new light on the problem s in
which we are here interested. W e m ust introduce into the
languages variables o f indefinite order which, so to speak, ‘run
through' all possible orders, which can occur as functors or
arguments in sentential functions without regard to the order o f
the remaining signs occurring in these functions, and which at
the same tim e m ay be both functors and arguments in the same
sentential functions. W ith such variables we must proceed with
the greatest caution i f we are n ot to becom e entangled in
antinom ies like the fam ous antinom y o f the class o f all classes
which are n ot members o f themselves. Special care must be
taken in form ulating the rule o f substitution for languages which
contain such variables and in describing the axiom s which we
have called pseudodefinitions. B ut we cannot go into details here.1
There is obviously no obstacle to the introduction o f variables
o f transfinite order n ot only into the language which is the object
investigated, but also into the m etalanguage in which the investi­
gation is carried out. In particular it is always possible to con ­
struct the metalanguage in such a w ay that it contains variables

1 From the languages just considered it is but a step to languages of another

kind which constitute a much more convenient and actually much more
frequently applied apparatus for the development of logic and mathematics.
In these new languages all the variables are of indefinite order. From the
formal point of view these are languages of a very simple structure; accord­
ing to the terminology laid down in § 4 they must be counted among the
languages of the first kind, since all their variables belong to one and the same
semantical category. Nevertheless, as is shown by the investigations of
E . Zermeio and his successors (cf. Skolem, Th. (66), pp. 1-12), with a suit­
able choice of axioms it> is possible to construct the theory of sets and
the whole of classical mathematics on the basis provided by this language.
In it we can express so to speak every idea which can be .formulated in the
previously studied languages of finite and infinite order. For the languages
here discussed the concept of order by no means loses its importance; it no
longer applies, however, to the expressions of the language, but either to the
objects denoted by them or to the language as a whole. Individuals, i.e.
objects which are not sets, we call objects of order 0 ; the order of an arbitrary
set is the smallest ordinal number which is greater than the orders of all
elements of this set; the order of the language is the smallest ordinal number
which exceeds the order of all sets whose existence follows from the axioms
adopted in the language. Our further exposition also applies without restric­
tion to the languages which have just been discussed.
272 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES VIH, S 7

o f higher order than ail the variables o f the language studied.

The m etalanguage then becom es a language o f higher order and
thus one w hich is essentially rioher in gram m atical form s than
the language we are investigating. This is a fa ct o f the greatest
im portance from the poin t o f view o f the problem s in w hich we
are interested. F or w ith this the distinction between languages
o f finite and infinite orders disappears— a distinction which was
so prom inent in §§ 4 and 6 and was strongly expressed in the
theses A and B form ulated in the Summary. In fact, the setting
up o f a correct definition o f truth for languages o f infinite order
w ould in principle be possible provided we had at our disposal in
the m etalanguage expressions o f higher order than all variables
o f the language investigated. The absence o f such expressions in
the m etalanguage has rendered the extension o f these m ethods o f
construction to languages o f infinite order im possible. B ut now
w e are in a position to define the concept o f truth for any language
o f finite or infinite order, provided we take as the basis for our
investigations a m etalanguage o f an order w hich is at least
greater b y 1 than that o f the language studied (an essential part
is played here b y the presence o f variables o f indefinite order in
the m etalanguage). It is perhaps interesting to emphasize that
the construction o f the definition is then to a certain degree
sim plified. W e can adhere strictly t o the m ethod outlined in
§ 3 w ithout applying the artifice which we were com pelled
to use in § 4 in the study o f languages o f the 2nd and 3rd
kinds. W e need neither apply m any-row ed sequences nor
carry out the sem antical unification o f the variables, for
having abandoned the principles o f the theory o f semantical
categories we can freely operate w ith sequences whose terms
are o f different orders. On the other hand the considerations
brought forw ard in § 5 in connexion w ith Th. I lose none o f their
im portance and can be extended to languages o f any order. It is
im possible to give an adequate definition o f truth for a language
in w hich the arithm etic o f the natural num bers can b e con­
structed, i f the order o f the m etalanguage in w hich the investiga­
tions are carried out does n ot exceed the order o f the language
investigated (cf. the relevant remarks on .p. 253).
V IU , § 7 C O N C E PT O P T R U T H IN F O R M A L IZ E D LA N G U A G E S 273

Finally, the foregoing considerations show the necessity o f

revising, to a rather important extent, the Theses A and B given
in the conclusions o f this work and containing a summary o f its
chief results:
A. For every formalized language a formally correct and materi­
ally adequate definition of true sentence can be constructed in the
metalanguage with the help only of general logical expressions, of
expressions o f the language itself, and of terms from the morphology
of language—but under the condition that the metalanguagepossesses
a higher order than the language which is the object of investigation.
B. I f the order of the metalanguage is at most equal to that of the
language itself, such a definition cannot be constructed.
From a comparison o f the new formulation o f the two theses
with the earlier one it will be seen that the range of the results,
obtained has been essentially enlarged, and at the same time the
conditions for their application have been made more precise.
In view o f the new formulation o f Thesis A the former Thesis C
loses its importance. It possesses a certain value only when the
investigations are carried out in a metalanguage which has the
same order as the language studied and when, having abandoned
the Construction o f a definition o f truth, the attempt is made to
build up the theory o f truth by the axiomatic method. It is easy
to see that a theory o f truth built up in this way cannot contain
an inner contradiction, provided there is freedom from contra­
diction in the metalanguage o f higher order on the basis o f which
an adequate definition o f truth can be set up and in which those
theorems which are adopted in the theory o f truth as axioms
can be derived.1
Just as in the conclusion of this work, the Theses A and B
can be given a more general formulation by extending them
to other semantical concepts:
A '. The semantics of any formalized language can be established
as apart of the morphology of language based on suitably constructed
1 In particular, the question broached on p. 201 has a positive answer. The
same also holds for the problem of infinite inductive definitions mentioned
pn p. 262, footnote.
274 CONCEPT OP TRUTH IN FORMALIZED LANGUAGES VIII, § 7

definitions, provided, however, language in which the mor­

phology is carried out has a higher order than the language whose
morphology it is .
B\ It is impossible to establish the semantics of a language in
this way if the order of the language of its morphology is at most
equal to that of the language itself.
The Thesis A in its new generalized form is o f no little im por­
tance for the m ethodology o f the deductive sciences. Its con­
sequences run parallel with the im portant results which Godel
has reported in this field in recent years. The definition o f truth
allows the consistency o f a deductive science to be proved on the
basis o f a m etatheory which is o f higher order than the theory
itself (cf. pp. 199 and 236). On the other hand, it follows from
GodePs investigations that it is in general im possible to prove
the consistency o f a theory if the p roof is sought on the basis o f a
m etatheory o f equal or lower order.1 M oreover Godel has given a
m ethod for constructing sentences which— assuming the theory
concerned to be consistent— cannot be decided in any direction
in this theory. A ll sentences constructed according to GodePs
m ethod possess the property that it can be established whether
they are true or false on the basis o f the m etatheory o f higher
order having a correct definition o f truth. Consequently it is
possible to reach a decision regarding these sentences, i.e. they
can be either proved or disproved. M oreover a decision can be
reached within the science itself, w ithout making use o f the
concepts and assumptions o f the m etatheory—provided, o f
course, that we have previously enriched the language and the
logical foundations o f the theory in question b y the introduction
o f variables o f higher order.2
Let us try to explain this somewhat more exactly. Consider
an arbitrary deductive science in which the arithmetic of
natural numbers can be constructed, and provisionally begin
the investigation on the basis o f a metatheory o f the same order
as the theory itself. Godel’s method o f constructing undecidable
sentences has been outlined implicitly in the proof o f Th. I in
1 Of. Gddel, K . (22), p. 196 (Th. X I).
* Cf. G5del, K . (22), pp. 187 fL, and in particular, p. 191, note 48 a.
Vm , §7 C O N C E PT O F T R U T H IN F O R M A L IZ E D L A N G U A G E S 276

§ 6 (p. 249 ff.). Everywhere, both in the form ulation o f the

theorem and in its proof, we replace the sym bol ‘ TV’ b y the
sym bol ‘ P r ’ which denotes the class o f all provable sentences
o f the theory under consideration and can be defined in the
m etatheory (of. e.g. D ef. 17 in § 2). In accordance with the first
part o f Th. I we can obtain the negation o f one o f the sentences
in condition (a) o f convention T o f § 3 as a consequence o f the
definition o f the sym bol ‘P r ’ (provided we replace ‘ T V in this
convention b y ‘P r’ ). In other words we can construct a sen­
tence x o f the science in question which satisfies the follow ing
condition:
it is not true that x e Pr if and only if p
or in equivalent form ulation:
(1) x e Pr i f and only if p
where the sym bol ‘p ’ represents the whole sentence x (in fact
we m ay choose the sentence constructed in the p roof
o f Th. I as x).
W e shall show that the sentence x is actually undecidable
and at the sam e tim e true. F or this purpose we shall pass to a
m etatheory o f higher order; T h. I then obviously remains valid.
A ccording to Thesis A we can construct, on the basis o f the
enriched m etatheory, a correct definition o f truth concerning
all the sentences o f the theory studied. I f we denote the class
o f all true sentences b y the sym bol ‘ TV’ then— in accordance
with convention T — the sentence x which we have constructed
will satisfy the follow ing condition:
{2) x e T r if and only if p ;
from (1) and (2) we obtain im m ediately
(3) x e P r if and only if x e Tr.
M oreover, if we denote the negation o f the sentence x b y the
sym bol ‘x ’ we can derive the follow ing theorems from the defini­
tion o f truth (cf. Ths. 1 and 6 in § 3):
(4) either x e T r o r x e Tr;
(5) if x e Pr, then x e TV;
(6 ) i f x e Pr, then x e Tr;
276 C O N C E PT O F T R O T H IN F O R M A L IZ E D L A N G U A G E S V TH , § 7

From (3) and (5) we infer w ithout difficulty that

(7) xeTr
and th at
(8) x elP r.
In view o f (4) and (7) we have x e Tr, which together w ith (6)
gives the form ula
(9) x e Pr,
The form ulas (8) and (0) together express the fa ct that a: is an
undecidable sentence; m oreover from (7) it follow s that a; is a
true sentence.
B y establishing the truth o f the sentence x we have eo ipso
— b y reason o f (2)— also proved x itself in the m etatheory.
Since, m oreover, the m etatheory can be interpreted in the
theory enriohed b y variables o f higher order (of. p. 184) and
since in this interpretation the sentence x, which contains no
specific term o f the m etatheory, is its own correlate, the p roof o f
the sentence x given in the m etatheory can autom atically be
carried over in to the theory itself: the sentence x which is
undecidable in the original theory becom es a decidable sentence
in the enriched theory.
I should like to draw attention here to an analogous result.
F or every deductive science in which arithm etic is contained
it is possible to specify arithm etical notions which, so to speak,
belong intuitively to this science, bu t w hich cannot be defined
on the basis o f this science. W ith the help o f m ethods which
are com pletely analogous to those used in the construction
o f the definition o f truth, it is nevertheless possible to show
that these concepts oan be so defined provided the science is
enriohed b y the introduction o f variables o f higher order.1
In conclusion it can be affirmed that the definition o f truth and,
m ore generally, the establishm ent o f sem antics enables us to
m atch som e im portant negative results whioh have been obtained
1 Cf. my review, *Uber definierbare Mengea reeller Zahlen,’ Annales de la
SooitU Polonaise de MatMmatique, t. ix, ann£e 1980, Kxakdw, 1921, pp. 206-7
(report on a leoture given on 16 Deoember 1980 at the Lemberg Section of the
Polish Mathematical Society); the ideas there sketched were in part developed
later in V I.
V m , § 7 CONCEPT OF TRUTH IN FORMALIZED LANGUAGES 277

in the m ethodology o f the deductive sciences with parallel

positive results, and thus to fill up in som e measure the gaps
thereby revealed in the deductive m ethod and in the edifice o f
deductive knowledge itself.

H istorical N otes. In the course o f the years 1929 to 1935, in which I

reached the final definition of the concept o f truth and m ost of the remain,
ing results described here, and in the last year o f which the whole work
appeared for the first time in a universal language, the questions here
discussed have boon treated several tim es. In the German language, in
addition to m y summary. Tarski, A . (76), works by Carnap have appeared
in which quite similar ideas were developed (cf. R . Carnap, ‘D ie A n ti-
nomien und die Unvollst&ndigkeit der Matheraatik Monatshefte/ . M ath.
u. Phys. vol. 41 (Leipzig, 1934), pp. 2 6 3-84 and ‘ Ein Gultigkeitskriterium
fur die Satze der klassiechen M athematik, ibid., vol. 42. pp. 1 6 3 -9 0 ; both
articles being supplementations of R . Carnap (10)). The *wo articles have
been incorporated in the English edition of O&roap’s book, entitled Logical
Syntax of Language (London, 1937).
It was to be expected that, in consequence of this lapse of six years,
and o f the nature of the problem and perhaps also o f the language o f
the original text o f m y work, errors regarding the historical connexions
might occur. And in fact Carnap writes in the second ot the above-
mentioned articles regarding m y investigations that they have been
carried out ‘ . . . in connexion with those of God el . . . \ It will there­
fore not be superfluous if I make some remarks in this place about the
dependence or independence of m y studies.
I m ay say quite generally that all m y methods and results, with the
exception o f those at places where I have expressly emphasized thia—
cf. footnotes, pp. 154and 247— were obtained by me quite independently.
The dates given in footnote, p. 154, provide, I believe, sufficient basis
for testing this assertion. I m ay point out further that m y article which
appeared in French (V I), about which I had already reported in Decem ­
ber 1930 (cf. the report in German in A . Tarski (74)) contains precisely
those methods o f construction which were used there for other purposes
but in the present work for the construction o f the definition o f truth.
I should like to emphasize the independence of m y investigations
regarding the following points of detail: (1) the general formulation of
the problem o f defining truth, cf. especially pp. 1 8 7 -8 ; (2) the positive
solution o f the problem, i.e. the definition o f the concept o f truth for the
case where the means available in the metalanguage are sufficiently rich
(for logical languages this definition becomes that o f the term ‘ analytical *
used by Carnap). Cf. pp. 194 and 2 3 6 ; (3) the method o f proving con­
sistency on the basis o f the definition o f truth, cf. pp. 199 and 2 3 6 ;
(4) the axiomatic construction o f the metasystem , cf. pp. 173 fF., and in
connexion with this (5) the discussions on pp. 184 f. on the interpret*-
278 CONCEPT OF TROTH IN FORMALIZED LANGUAGES VIH, $ 7

tion of the metasystem in arithmetic, which already contain the so-

called ‘ method of arithmetizing the metalanguage9which was developed
far more completely and quite independently by Godel. Moreover, I
should like to draw attention to results not relating to the concept of
truth but to another semantical concept, that of definability reported
on p . 276.
In the one place in which m y work is connected with the ideas of
Gddel— in the negative solution of the problem of the definition o f truth
for the case where the metalanguage is not richer than the language
investigated— I have naturally expressly emphasized this fact (cf.
p . 247, footnote); it m ay be mentioned that the result so reached, which
very much completed m y work, was the only one subsequently added
to the otherwise already finished investigation.
 IX
SOME O B S E R V A T IO N S ON TH E
C O N C E P TS OF ^ -C O N S IS T E N C Y A N D
o .C O M P L E T E N E S S t
I n an extrem ely interesting article Gfidel1 introduces the
concept o f m-consistency, and constructs an example o f a de­
ductive system which is consistent in the usual sense, but is not
ai-consistent. In the present article I propose to give another
simple exam ple o f such a system , together with some general
remarks on the concept m entioned as well as on the correspond­
ing concept o f co-com pleteness .2
The sym bolical language in which I shall construct this system
is closely related to the language o f the system P used b y Godel.
It is also the result o f an exact form alization and sim plification,
as far as possible, o f the language in which the system o f Prin -
cipia McUhematica o f W hitehead and Russell3 is constructed.
In spite o f its great sim plicity this language suffices for the
expression o f every idea which can be form ulated in Principia
Mathematical
5 See Godel, K . (22).
2 Already, in the year 1927, at the vSecond Conference of the Polish Philoso­
phical Society in Warsaw in the lecture4Remarks on some notions of the metho­
dology of the deductive sciences1 (listed by title in Buck Fitozoflczny, vol. 10
(1926-7), p. 96), I had pointed out the importance of these two concepts, and the
rule of transfinite induction which is closely related to them and about which
more is said in the text, but I had not suggested special names for these con­
cepts. I also communicated the example of a consistent, and yet not in­
consistent system which I give in the present article in a slightly altered
form. Naturally it is not hereby claimed that I already knew then the results
later obtained by Godel or had even foreseen them. On the oontrary, I
had personally felt that the publication of the work of Godel cited above was
a most exciting scientific event.
8 Whitehead, A. N., and Russell, B. A. W . (90).
4 Cf. articles VI and VIII of the present work, where I have used the same
or a very similar language.
t BiBiiiOGRAFHiOAi N ote . This article first appeared under the title
‘Einige Betrachtungen fiber die Begriffe m-Widerspmchsfreiheit und der
w-Vollst&ndigkeit ’ , Monatshefte ftir Matkematik und Phyeik , vol. 40 (1933),
pp. 97-112. For the historical information about the results of this article
see footnote 2 above.
280 TH E CONCEPTS OF C O N SISTE N C Y IX

In our language two kinds o f signs oocur: constants and

variables. Three constants suffice: the negation sign <N \1 the
sign o f im plication CC' and the universal quantifier ‘(Q \ On
the other hand we shall operate with infinitely m any variables.
As variables wo shall use the sym bols *z' \ lx„\ (x*\ etc., i.e.
sym bols com posed o f the sign V and an arbitrary series o f small
strokes both above and below. The constants and variables
are the sim plest expressions o f the language. B y writing down
these signs one after another in any number and arrangement we
obtain m ore com plicated expressions. E very com posite expres­
sion, i.e. one containing m ore than one sign, can profitably be
regarded as the result o f putting together tw o (or m ore) successive
sim ple expressions. F or exam ple: %',&?* consists o f T T
and ex' xl asf' or o f x, * and (x, xV or also o f x\ x\5 and
'x*,\ A n especially im portant category o f expressions will be
Ibtinguished below, nam ely that o f (meaningful) sentences.
The constants have their usual interpretation: expressions o f
the form *Np\ Opqs and ‘J J xp\ in which, in plaoe o f ‘p* and
qf any sentences and instead o f V any variables, occur, are
read: ‘not p\ Hfp then qy and ‘for every x, p ’ respectively. V ari­
ables with one upper stroke are regarded as names o f individuals
(objects o f 1st order), those with tw o strokes as naming classes
o f individuals (objects o f 2nd order) and so on. The use o f lower
strokes with the variable serves the same purpose as the custom ­
ary use o f letters o f different shapes in the function o f the vari­
ables. A n expression o f the form ‘xy\ where instead o f V any
variable occurs and instead o f V a variable w ith one more
upper stroke, is read: (x is an element of the class y* or *z has the
property y\
The language here briefly described m ay be made the object
o f a special investigation. This task presupposes that we consider
it from the standpoint o f another language which in contrast
to the original one can be called the matalangwage, and on
the basis o f this metalanguage we can establish a special disci­
pline, the so-called metadiscipline♦ In order to avoid unnecessary
1 The negation sign <jan be dispensed with and serves only to simplify the
discussion, Regarding symbolism see IV .
IX AN D ^-C O M P L E T E N E SS 281

com plication we shall not undertake the exact form alization o f

the m etalanguage. It suffices to note that in the metalanguage
the follow ing tw o categories o f concepts occur: ( 1) concepts o f a
general logical nature, belonging to an arbitrary bu t sufficiently
developed system o f ‘m athem atical logic (e.g. to Principia
Mathematics), and in particular those from the dom ain o f the
calculus o f classes, and the arithm etic o f natural num bers ;1
( 2) specific concepts o f a structural-descriptive character which
designate the expressions o f the original language, their struc­
tural properties, and their m utual structural relations. The
follow ing terms are regarded as prim itive concepts o f the
m etalanguage: "expression* or rather "the class o f all expres­
sions’, in sym bols ‘A 9; "negation sign*— V ; "im plication sign*—
V ; "universal quantifier’ — V ’ ; "variable with h strokes below
and l above’ for "the kth. variable o f the ith order’ )—
finally "expression which consists o f the tw o successive
expressions £ and 77’ sym bolically " r " V -2 It can easily be estab­
lished that the above concepts suffice for the exact description
o f every expression o f the original language. W ith their help we
can correlate with every such expression a particular individual
name in the metalanguage. F or example, the expression
"IJ d x>x*’ *s denoted b y the individual name
Corresponding to the tw o categories o f concepts o f the m eta­
language, the system o f assumptions which suffioe for the
deductive establishm ent o f the m etadiscipline consists o f tw o
kinds o f statem ents: ( 1) the axioms of mathematical logic; and
(2) such theses as suffice to establish the basic properties of (he
specific primitive concepts of the metalanguage. It will not perhaps
1 r shall make free use of various symbols whioh are in common use in
works on the theory of sets. For example, formulas of the form tx, y,,.. e Z*
and e Z<y express the fact that x, y,... do and do not belong to the set
Z respectively. <0 ’ denotes the null set and ‘ { 2 , t/,..., 2 } ' the set which con­
tains only xy z as elements. By the symbol ‘Ntf I denote the set of all
patural numbers. Nt—{0} is thus the set of all natural numbers distinct from 0.
* This last primitive concept can be replaced by the concept ‘ the nth sign
from the beginning of the expression f\ The symbol ‘ A* can be defined by
means of the remaining undefined concepts.
* By reference to the associative law which the operation £^*17 obeys (see
below, Ax. 4) the omission of the brackets in expressions of the form ‘ (£^ 1 7 )^ 8 %
((£^ 17)^ 8 ) ^ ’ cannot give rise to> any misunderstandings.
282 TH E CONCEPTS OF ^-C O N S IS T E N C Y IX

be out o f place here to introduce explicitly the full system o f

axiom s o f the second kind (although it will not be possible within
the lim its o f the present sketchy aocount to bring these axioms
into practical use).

A x io m 1. v, t, tt e A , where v # i, v ^ n, and i n.
I f k , l e Z f —{ 0}, then <j>lk e A , where <f>lk =£ v, <f>lk ^ t,
A x io m 2.
and <f% ^ n; if, moreover, kx, lx e N t—{ 0} and fc kx or l ^ lx,
then 4>{ ^
77 e A , then £^17 6 A, where
A x io m 3. / / £, # v, £^77 ^ t,
^ tt, and £^77 ^ $ ./o r any k, l e Nt—{ 0}.
A x io m 4, / / £, 77, S, £ e A , <Aen formula: £^77 = AoWs
{ / and cwfo/ t / eifAer £ — S and 77 ~ g, or if there is a r e A such
that £ =s S^r and | = r^rj, or finally if there is a r e A such that
S = r~> and 77 =
A xiom 5. (JAe principle of induction). Let X be any set which
satisfies the following conditions: ( 1) v, 4, 7r e l ; (2) i/& , Ze ^ —{ 0},
2Aen £ Z ; (3) t / r] e X , then £^77 e Z . Then i c j ,
It is not difficult to prove that the above axiom system is cate­
gorical,1 has an interpretation in the arithmetic of the natural num­
bers awl consists of mutually independent statements.2
In order to give the further developm ents a simpler form a
series o f sym bolic abbreviations is now introduced into the
metalanguage.

1 In the sense of Vebleu, O. (86).

3 The intuitive obviousness of the axioms is not uncontested. Certain
objections may be raised in connexion with the existential assumptions which
are implicit in Axs. 2 and 3 (on the basis of these assumptions it can be proved,
among other things, that the set A has the cardinal number K0, and is thus
infinite). Without going into the analysis of these difficult questions I should
like to say only that the objections mentioned are weakened significantly if
it is assumed: (i) that we regard as expressions of the language not concrete
inscriptions but whole classes of inscriptions of like form, and correspondingly
modify the intuitive sense of the remaining primitive concepts of the meta­
language ; (ii) that we interpret the concept of inscription as widely as possible,
to include not only manuscript inscriptions, but also aU material bodies (or
geometrical figures) of definite form. I discuss this point in more detail in my
article VIII on the concept of truth, where I have given the above axiom
system for the first time. It may be mentioned that other metaecienoee oan,
mutaiis mutandis, be axiomatized after the pattern of this axiom system.
IX AND <o-COMPLETENESS 283

D efin itio n 1. (a) £ e V (£ is a variable), if there are

k , l e Nt-~{0}
such that £ = <f>\\
(b) £ = Vo—Vn (£ i8 an expression consisting o f the successive
expressions tj0, 7jn), i f n e N t . »?0, ijj,..., r)n s A and either
» = 0, £ = i}0 or n > 0, £ — ( ’n o ' ^ - ^ V n - i P V n -

D efin ition 2. (a) £ — 77 (£ ts fAe negation o / 77), if y e A ,

£=
(b) £ = 77 -» 8 (£ is the im plication with rj os antecedent and S
as consequent), if r\, 8 e .4, £ =
(c) £ = 77v 8 (£ is <Ae logical sum or disjunction o / 77 and 8),
if i], 8 e A , £ = i) 8;
(d) £ = t?0 v ... v •!?„ (£ is <Ae logical sum o / ij0, i?n), i /
» e i ? 0, tjx,..., qn e A and either n — 0, £ = q0 or n > 0, and
r = ( n r - r ir r j^ r r ,^
(*) £ = 17A 8 (listh e logical product or conjunction 0/ 77and 8),
if V , b e A , £ = ijv S ;
( / ) £ = Vo A - A ijn (£ is the logical product of r)0. 77J, if
n e Nt, q0, 7}1(..., 7jn e A and £ = 7j0V ... v ijn;
(9) £ = 77 o 8 (£ is <Ae logical equivalence 0 / tj and 8), if
7j, 8 e . 4 , £ = ( tj-> 8 ) a (8 -» -77).

D efin ition 3 . (o) £ = f")*. 77 (£ is fAe universal quantification

of 77 witA respect to the variable <f>lk), if k, l e JVt—{0}, 77 e 4 and
£ = iT fif'-q-,
(b) £ = n i". •••fife .7? (£ *-s tAe universal quantification 0 / 77
toitA respect to the variables $j?0, <$£), i /

77 6 ^ and n, Aq, ig, A^> ^

and £ = ( ( n ^ r . . r ( n ^ n) r r ,
(c) £ = U i rj (£ is the existential quantification o f tjwith respect
to the variable <f>lk), i f l c , l e $ ^ —{ 0}, rj e A and £ =
(d) £ = Ufe —U fe^ (£ is ZAe existential quantification 0/ 77
respect to the variables <5^ ,..., i/

77 £ and A/q, Zq} ^ i

o»d£ = n f c ~ n f c *
284 TH E CONCEPTS OP ^-C O N S IS T E N C Y IX

D efin itio n 4. (a) X *8 ‘Ae atom ic sentence with the

£ =
terms and if k, l, m e N t—{0}, £ = f i T W +l;
(b) £ - r& X i* the identity between (f>%and cffi), if
k, L m e jSft—{0} and £ = P|?+1(€S!i «u)-
Among the most important concepts of the metadiscipline are
those of meaningful sentence (or simply sentence) and o f the
consequences of a set of sentences. The definition of the first
presents no difficulties:
DMBXNiTiON 5. £ e 8 (£ is a meaningful sentence),1 if £ belongs
to every set X which satisfies the foilowing conditions: (1) ejfo e X
for every kyl, in e JV7£—{0}; (2) if ?; e X , then rj c X ; (3) if rj, 5 e X ,
then t) S t I ; (4) if 1c, l e Ni—{0} and r\e X then pjj. rt € X .
Meaningful sentences are thus the expressions we obtain from
elementary sentences : by performing upon then*, any number
of times and in any order, the three operations of negation*
implication, and quantification,
For the definition of the concept of consequence certain pre­
liminary definitions will be useful:
D eeesxtion 6. (a ) £ F v (£ o c c u r s a t th e r a -tk p l a c e i n t> as a
m
free variable), if m e Nt, £ £ V and if v can be represented in the
form : rt — b .f ...^bn, where n e N l , b a, , e {v, i, n}4-V,m ^ n,
Sm — £ and there are no numbers k , l e Nt such that
k < m < k-fl <fn, bk = n, 3fc+1 = £ and 6 8.
(b) £ e Fr(vf) {£ is a free variable of */), if there is an m e N't such
that £ F r;.
m
D efin itio n 7. e L (£ is a logical axiom ), if one of thefollowing
£

conditions is satisfied: (1) there exist statements n, 3, £ e 8 such

that either £ — (ij -> 8) -»■ ((S -> £) -> (17 -> £)) or £ — {fj -*■ rj) -> ij
or £ — -r) (rj -> £)] (2) fAere w a number k e Nt—{ 0} o n i a .sen­
tence 17 e 5 encA that £ — u ? + i n ? ( 4 i o *2}, wdtere 1 e Fr{rj);
(3) there is a number k e N t —{0} such that
£^ <;> *1,2)) ^ j j 1'
1 The expression ‘ sentential function * would be more suitable here; the
expression ‘ sentence* should be reserved for those sentential functions whioh
contain no free variables (see below, Def. 8).
IX A N D a>-COMPLETENESS 285

The sentences w hich satisfy condition (1) in the above defini­

tion are the axioms of the sentential calculus o f Lukasiewicz1 (or
rather substitutions o f these axiom s); the sentences under (2),
clearly related to the axiom o f reducibility o f Principia Mathe-
matica, can be called pseudodefinitions in accordance with the
proposal o f S. Lesniewski; finally we recognize the sentences
under (3) as the so-called laws o f extensionality.

DEiriiSiTTON 8. ^Sb^rj (£ can be obtained from q by the sub­

stitution of the free variable <f>$for the free variable $ * ), if

k , l , m e Nt —{0},
and if £ and q can be represented in the form
t - ~ s n, v-
where (1) n e N t , and S0, |0, 8ls 8„, e {v, c, * } + V ; (2) if
fif* F q, then <f>%°F l.; (3) if i e Nt, i < » and (he formula F q
i i i
does not hold, then Zi == £t.

D efin itio n 9, £ e Cn(X) (£ is a consequence of the set X of

sentences), if £ belongs to every set Y vYhich satisfies the following
conditions'. (1) X + L £ Y ; (2) if SSb^i q and q e T, then B e Y:
(3) if rj, 7] -*■ S e Y , then 5 e 7 ; (4) if lc, l e N t—{0}, q, 8 e 8,
<f>\f Fr{q) and u 8 e Y , then q -*■ f*|J. 8 e Y\ (5) i f k , l e Aft—{0},
q, S e 8 and q --*■ f|i|. 8 e Y, then, q - r b e Y .

A ll sentences are thus regarded as consequences o f the set X

which can be obtained from the sentences o f this set and the
logical axiom s b y means o f the four operations o f substitution,
detachm ent, and the introduction and elim ination o f the
universal quantifier (these rules are com m only called the
rules o f inference).8
From Defs. 5 and 9 various elem entary properties ©f the con-
pept o f consequence are easily derived, e.g .:8

1Cf. rv.
* Cf. article V o f the present work whore the terms * £ ’ and *Cn' occur a s
the only undefined concepts and the parte (a), (ft), (d) of Th. 1 belong to the
axiom system. In connexion with Ths. 2 and 3, cf. H I.
286 T H E CONCEPTS OF «>.C O N SISTE N C Y IX

T heorem 1. (a) I f X s S, then X £ Cn(X) £ 8;

(6) if X £ 8, then Gn(Cn(X)) = On(X);
(c) if X £ Y £ 8 or X £ Gn(Y) and Y £ 8, then

Cn{X) £ On(Y) ;
(d) let X £ toen in order that £ e Cn(X) it is necessary and
sufficient that there exists a finite set Y £ X such that £ e Cn(Y);
(e) i f -£ » £ X n+1 for every n e Nt, then

Gn( f X n) = f Cn(Xn).
ti «»0 7 n «®0

T heorem 2. (a) / / X £ <S, £,ij e 8. Fr{£) = 0 and

r) e <7n(X +{£}), then £ -> v e (7 »(X );

(6) if X Z 8, £ e $ and Fr(£) — 0, toen

0»(X+{£}). £»(*+{$) = C»(X);

(c) if X c. S and £ e S, then Cn(X+{£, £}) = 8;
(d) if £ e S and rj — ... f"|j£ £, where

n & Nt and Aq, jfejf ^

toen Cn({£}) = Gn({tj}); to every £ e 5 toere tons corresponds a

sentence -rj e 8 such that Fr(i j) = 0, Cn({£}) = Cn({rj}) and even
more generally C 'n (X + {^ }) = C n (X + {i 7}) /o r every set X £ 5 .

A certain difficulty m ay arise solely in connexion with the

p roof o f Th. 2 a. W e consider here the set F o f all sentences 8,
such that £ -> 8 e Cn(X) and show that this set satisfies the
conditions (1)~(5) o f the definition, i f ‘X + { £ } ’ is inserted instead
o f ‘X \ From this it follow s im m ediately that the set Y contains
all sentences rj e (7n(X-|-{£}), which was to be proved. From
Th. 2 a we easily obtain 26.
A s is well known, a series o f fundam ental concepts belonging
to the m ethodology o f the deductive sciences can b e defined by
means o f the concepts o f m eaningful sentence and o f corn
sequence. In particular this is the case for the concepts o f a
deductive system and o f consistency and completeness-.
IX AND a>.COMPLETENESS 287

D efin itio n 10. (a) X e S (X is a deductive or closed system ),

i f Gn(X) = X ^ S ;
(6) X e SB (X is a consistent set of sentences), if X £ S and
Cn(X) 96 8;
(c) X e S3 (X is a com plete set of sentences), if X £ 8 and at
the same time we have, for any £e $ , either £ e Cn{X) or X -f-{£ } e SB.
Some transform ations o f D efs. 10 6 and c are w orth empha­
sizing, since they are equivalent to the original form ulations but
in m any cases are easier to m anipulate:
T heorem 3. (a) In order that X e SB, it is necessary and suffi­
cient that X £ 8 and that for every £ e 8 we have either £ e Cn(X)
or %e Gn(X).
(b) Let Fr(£) = 0 ; in order that X - J-{£} e SEB, it is necessary and
sufficient that X c 8, £ £ S, and £ e Gn(X);
(c) in order that X e S3, it is necessary and sufficient that X £ 8
and that for every £ e 8 such that Fr(Q = 0, we have either
t,eCn{X)or £ e C » (X ).
(a) follow s from D ef. 106 and Ths. l a and c and 2c; (6) is
obtained from (a) with the help o f Th. 2 b; finally, to obtain (c)
w e make use o f (6), D ef. 106 and c and Th. 2d.
In contrast to the ordinary consistency and completeness, the
concepts o f <D-consistency and <u-completeness cannot be e x ­
pressed exclusively in terms o f ‘(S’ and ‘Gn\ In defining these
tw o concepts the follow ing supplem entary sym bols will be used:
D airN m oN 1L (a) £ = /Sjf, if k e Nt—{ 0}, n e N t and either
n = 0, £ = f| K fc m
H > 0, ^ = U l •••U» n»+l(e»+l,fc "*• (rl,n+l V ... V )»
(&) £ = yjy. *fk,l 6 { 0} and £ = (J i fH O ii o («U v ’ i.*)) 5
(c) £ = 8k, if k e ^ —{ 0} and
£^ n m im m v A y^ o) £?,i)
The meaning o f the above sym bolical expressions is clear:
the sentence yktl states that the set o f individuals denoted by the
sym bol $ is obtained from the set denoted b y <j>%by the addition
o f a single elem ent; the sentence /?£ means that the set denoted
b y the sym bol <f>%contains at m ost n elements; likewise Sk means
288 T H E CONCEPTS OF oi-CGNSISTENCY IX

that this set is finite (inductive in the sense o f Principia Mathe­

matical).
D efinition 12. (a) X s 2B«, (X is an ^-consistent set of sen­
tences)s if X £ 8 and if for every £ e $ such that Fr(&) = {<£}}*
the condition £) e Cn(X) for every n s Nt always implies
the formula -* i) e Cn(X).
(b) X s 93w (X is an a>-complete set of sentences), if X G S and
i f for every £ e 8 such that Fr(£) = {^ f}, the condition
v m o e < w )
/o r every n s N t always implies the formula f jf ( 8 X-»■ 0 e Cn(X).
I n order to make the content o f these concepts more com ­
prehensible, let us consider any property P which sets o f indi­
viduals can have and which can be form ulated in the present
language. Let £n(P) (where n e Nt ), or ?JK P) be sentences o f the
language which assert respectively that every set which contains
at m ost n individuals, or every finite set o f individuals possesses
the property P. A set X o f sentences will be called ^-consistent
i f it cannot be the case that for some property P all sentences
o f the infinite sequence £n(P ) are consequences o f the set X but
nevertheless the negation o f the sentence £W(P) is also a con­
sequence o f this set. On the other hand, a set X will be called
co-com plete if. whenever all sentences £*(P ) belong to the con­
sequence o f the set X , the sentence $m(P) is also a consequence of
the same set.8
The follow ing relations hold between the classes SB, 2BW,
SB, and SB*:
T heorem 4, (a) if X e SB*,, then X e SB (SB*, s SB);
(6) if X e SB and X e t h e n X s SB* (SB. S * s SBJ;
(c) * / X e © and X e SB*,, then I e S B w (iB. SB^ £ $*,);
(d) if X £ S and X s then X € SB and there is a sentence
i s S such that X + { £ } e SB-SB*,.
1 The formula Fr(f) =* { $ } can here be omitted.
2 Between the definition of aj-oonsistenoy given above and that of GOdel
there exist certain purely formal differences. They are accounted for by the
fact that, in the formal language here in question, specific symbols to denote the
natural numbers do not occur. It is easy to show that with regard to GMel’s
system P (and all more comprehensive systems) the two formulations are equi­
valent.
rx AND a>-COMPLETENESS 289

This theorem is easily obtained from Defe. 106 and 12 a and 6,

and Th. 3.
As was m entioned at the beginning, the converse o f Th. 4 a
does not hold: there exist consistent deductive systems which
are not ^-consistent. Th. 4d shows that such systems can be
obtained b y extending systems which are not ^-com plete. W e
shall adopt this m ethod and construct a simple exam ple o f a
system which is not -com plete— it will be the system T^; b y
extending this system we shall obtain a consistent system T[0
which, however, is not ^-consistent.

D efinition 13. (a) £ = a*, if njE Nt, £ = H P ? ^ U i 4 i ) >

(6) £ « * " , < / £ « n ? (S i^ U l^ i)*
B y the assertion o f an the existence o f at least n-±~ 1 distinct
individuals is obviouslv established, and in the sentence a® we
recognize the axiom of infinity, according to which the sot o f
all individuals is infinite.

D efinition 14. (a) £ £ Tw, if £ e Gn( 2 ;

(6) Z e T ^ i j H e Cn( f {«* }+ {«»}).

'n«»0 '
In order to establish the desired properties o f the systems
Tmand T'w, I n est introduce the follow ing lem m a:

Theorem 5. an+1, a'0 e Cnl 2 {«*}) fa n e Nt.

Sketch of a proof. W e make use o f the well-known m atrix
m ethod from investigations on the sentential calculus.1
L et Sx be the set o f all sentences w ithout free variables, and
Sz the set o f all sentences w ithout universal quantifiers, i.e. those
containing only free variables. W e put ^ (£ ) = £ for £ e»Si and

wo-ru-r&c
for £ e S — where «$£,, ^ . . . . , are all free variables o f th e
sentence £ (the order in which we carry out the operations o f
(juantifioation is indifferent, provided that it is established once
and for all for any sentences ^ e S —Sf). The function obtained
1 See IV , p. 48.
280 T H E CONCEPTS OF ^.-CON SISTENCY IX

establishes a one-many correspondence between elements o f the

set 8 and the set 8V
We shall call designated variables the first» + 1 variables o f the
1st order (i.e. the variables <j>{, ^ +1), the first 2n+1 vari­
ables o f the 2nd order, the first 22*+1 variables o f the 3rd
order, and so on. We denote by ‘S” the set of all sentences
which contain no designated variables, and by 'S '1the set of all
sentences in which, in addition to constants, only designated
variables occur. Further, with every sentence £ e 8 will be
correlated a sentence ^ra(£) e obtained by replacing every
variable <f>l><£*>— by #+*+»••• • In this way, to every
sentence £ e /S* there corresponds a sentence ^s(£) e 8X. 8'.
Further, we define recursively a certain function <fta for all
sentences £ e S'. We put, namely, &,(«&) — «£,, </<8(£) - </«,(£),
’h (Z) -> & (fU 0 = Ci A — A 4+i» wiere
Ci-S6i,*£, £a.S6ii*£, .... £rt+x^ i +x|fc£,
and in general ^8( f j i £) =» £t A ... A £,, where £j, £a,..., Cp are sen­
tences which are obtained from £ by substituting for the free
variables <j>lk all successive designated variables o f the 2th order,
$>• As will easily be seen, the function tfia maps the
set S' onto a part o f the set 8a, in particular the sentences
lM£) e8 g .S “ correspond to the sentences £ 6 8X. /S'. If, therefore,
we put </>(£) a ./f8(^2(^1(£))) we obtain a function which oor-
relates a sentence <fi(Q e 8 a.8 n with every sentence £ e /S.
Finally we consider a function/which assumes only one of the
two values 0 and 1. We define this function for all sentences
£ £/Sa by recursion:

where [«] is by definition the greatest integer not exceeding x,

(in other words /(egj) is equal to the fcth figure in the dyadio
expansion o f the number 2—1),

/( 0 - 1 - / ( 0 .
/(£■>■’y) — i —/(£ )• (!—/ ( ij»
(in other words/(£ 17) — 0, if/(£ ) = 1 and/(i?) — 0, and in the
IX AND « ’ COMPLETENESS 291

remaining cases/( £ -> tj) = 1). In particular, all sentences o f the

form >p(i), where £ e 8, can ocour as arguments o f the fu n ction /.
L et 8* be the set o f sentences £ for which /(</r(£)) = 1 holds.1
n
I f 'X ’ in D ef. 9 is replaced b y *£ {a1} ', it follows w ithout greau
i 0
difficulty th at the set Y — 8* satisfies the conditions (1)~{5)
o f the above definition; whence w e oonolude that

On the other hand, we establish im m ediately that

m '(«n+l)) - /< # * • ) ) = o,
whence an+l, a® 6 8*. Consequently we have

* » « * ,« • I 0 » ( £ {* * }),
w hich was to be proved.
T ebokbm 6. (a) Tw e S — 23w; (6) Tme S .2 B .—-28..,.
Proof. From Th. 1 a, b, and D ef. 14a and b we obtain

(1) | {«»} £ C n ( f {a-}) = T„ = Gn{TJ £ 8

n~o Vi»o •
and

(2) | {«»}+{«»} s cn( f {«>*}+{«•}) - n <?»(n,) = s.

n-0 1
In accordance w ith D ef. 10 a the above form ulas Yield
(3)

B y putting X n =* {a*} in Th. 1 e, we obtain

C n( ! («* )) = f o » (I w );

sinoe, in view o f Th. 5, a” e {a14}) for every natural number

» ( we infer that
(4) a00e C »^ 2o{a"})-

1 3 * may be called the set of all sentences which are generally valid in every
individual domain of n - f 1 elements.
202 T H E CONCEPTS OF &>-C O N SIST E N C Y IX

With the help of ( 1 ) and (4), Def. 126 with X — and

£ — U i ei,i together with 13o, 6 leads to
(5) Tmg

From (4) it follow s that a® e 2 { “ "} j >whence on the basis

o f Th. 36 (w ith X = 2 {“ " } and £ = a®) we obtain the form ula

2 {an} + { ^ } s SB. Hence from (2 ) and Def. 106 we conclude

w-0
that

(6) T'me SB*

Finally, (2), IS a. 6 , and D ef. 1 2 &(for X = and £ = U i €i,i)
give

(7) r me SB,.
By virtue of (3) and (5}~(7) the two formulas desired,
T^e <3—33, and !T ^ e S .33— are proved.
In conclusion, I should like to draw attention, to the system
Tw which is characterized in the following way:
D efinition 15. £e if £ g C7&({«®}).
In its extension this system coincides almost completely with
the system P of Godel mentioned earlier. Thus, using Gttdel’s
methods of proof, we can extend all the results obtained for the
system P to the system 2 ^. In particular we can prove
T heorem 7. If 21
. e 33, tilcn T„ s 3 — 33,.

For this theorem even an effective proof is known; we can

construct, namely, a sentence £0 (of relatively simple logical
structure), which contains as its only free variable, and for
which -> £0) e Gn(2 h) holds for every n e Nt, but at the
same time ~> to) If now vve put

- C » ({« ", n l ( 3 ^ ) } ) ,
we obtain a system with the properties e 3 .3 3 —2B*,,
analogous to those o f T'^, it being assumed that- J7, g 33.
IX AND ^-C O M P LE T E N E SS 293

The system stands in close structural relations to and

; we have in fact

T. - r. - e»(JoM+£?})
and r . = Cn({«-}) -

whence b y means o f Th. 2b the form ula T„ = T'^.T^ follow s.

Nevertheless, in spite o f the form al relationship, there exists a
fundam ental essential distinction between these systems. In a
sense which is not defined m ore closely here, not one o f the three
systems possesses a ‘finitistic interpretation’ , they are not
w holly valid in any finite individual dom ain; thus we have to
do with ‘infinitistic’ system s. Nevertheless, every sentence o f
the system Twor T'wi taken b y itself, holds in some finite dom ain
o f individuals. These systems can be represented as a sum
o f an infinitely increasing sequence o f ‘fm itistic’ system s, e.g.
systems Tn = Cn({ocn}). On the other Hand, the system con­
tains sentences, e.g. the axiom o f infinity a00, which in them ­
selves adm it o f no ‘finitistic interpretation’ . In consequence
this system is m uch more com prehensive than and— in con ­
trast to the latter— is sufficient for the establishm ent o f the
arithm etic o f the natural numbers and also o f analysis in its
whole extent. The system merits the title ‘potentially
infinitistic’ , whilst on the contrary m ight be called ‘actually
infjnitistic’ .
In connexion w ith what has been said, the aj-incompleteness
o f the system appears as a significantly more profound
phenom enon than the analogous property o f the system T^.
The latter is in a certain degree ex definitione a>-incomplete, for
while we include in it all statements an, we have not included the
statem ent a00. F or this reason Th. 6 a is also intuitively clear, if
not trivial, and its p roof is quite simple o f its kind. B y contrast,
the co-incompleteness o f the system discovered b y Godel is
an entirely unexpected fa ct for m odem m athem atics and logic.
The establishm ent o f this fact required subtle and ingenious
m ethods o f inference. B ut that is not all. The p roof o f con-
294 TH E CONCEPTS OF ^-C O N S IS T E N C Y IX

sistenoy for the ‘potentially infinitistio’ system Tu is already

im plicit in the p ro o f o f Th. 6 a (of. Th. 4a) and so presents no
special difficulties. B u t from G odei’s investigations it follows
that the consistency o f the system T„ — and likewise that o f the
other ‘actually infinitastic’ system s— cannot be established at all
on the basis o f the m etadiscipline. From this we can understand
the presence o f the condition ‘T^ e SB’ in the hypothesis o f
Th. 7 .1
The facts here brought forw ard are notew orthy for m any
reasons. Form erly it could be assumed that the form alized
concept o f oonsequence coincides in extension with that con­
cept in everyday language, or at least that all purely structural
operations, w hich unconditionally lead from true statements to
true statem ents, could be reduced w ithout exception to the rules
o f inference em ployed in the deductive disciplines. It m ight also
be thought that the consistency o f a deductive system is in itself
a sufficient guarantee against the appearance o f statements in
the system which— on account o f their m utual structural rela­
tions— cannot both be true. Since, how ever, there are systems
which are on the one hand w-incom plete and on the other con­
sistent, but not ^-consistent, the basis o f both o f these assump­
tions is rem oved. In fact, from the intuitive standpoint the truth
o f the sentence H i^ i £) is w ithout doubt a oonsequence o f the
truth o f all sentences o f the form 0 * an-d hence fol­
low s the falsehood o f the negation o f the form er, namely
fPT ^ (J, B u t it is clear that the sentence ( * | i ^ need
not at all be a consequence o f all sentences -> £), and
that the sentences 0 an<I f l i ^ i £) 0811 occur in
one and the same consistent system .
These disadvantages can be avoided if we add to the rules o f
inference the so-called rule of infinite induction, i.e., if we
enlarge D ef. 9 b y this condition: ( 6) if i\ e S, Fr{ij) =
and n m -»• f]) e 7 for any n e Nt, then f ) f (Sj
E very deductive system will then be w-com plete b y definition,
1 Moreover, we find the same difficulties in the case of the problem of the
^-consistency of the system Ti0: from the positive solution of the problem the
consistency of the system would follow, whilst a negative solution is quite
improbable.
IX AND ou-COMPLETENESS 295

and the usual consistency will coincide with co-consistency, B ut

this w ould have no fundam ental advantage. It m ay be remarked
in passing that such a rule, on account o f its ‘infinitistic’ charac­
ter, departs significantly from all rules o f inference hitherto used,
that it cannot easily be brought into harm ony with the current
view o f the deductive m ethod, and finally that the possibility
o f its practical application in the construction o f deductive
system s seems to be problem atical in the highest degree .1 B ut
the follow ing is w orthy o f em phasis: the profound analysis o f
G odel’s investigations shows that whenever we have undertaken
a sharpening o f the rules o f inference, the facts, for the sake o f
which this sharpening was felt to be necessary, still persist,
although in a m ore com plicated form , and in connexion w ith sen­
tences o f a more com plicated logical structure. The form alized
concept o f consequence will, in extension, never coincide with
the ordinary one, the consistency o f the system will not prevent
the possibility o f ‘structural falsehood'. H owever liberally we
interpret the concept o f the deductive m ethod, its essential
feature has always been (at least hitherto) that in the construc­
tion o f the system and in particular in the form ulation o f its
rules o f inference, use is made exclusively o f general logical and
structurally descriptive concepts. I f now we wish to regard as
the ideal o f deductive science the construction o f a system in
which all the true statem ents (o f the given language) and only
such are contained, then this ideal unfortunately cannot be
com bined with the above view o f the deductive method.*
1 In contrast to all other rules of inference the rule of infinite induction is
only applicable if we have succeeded in showing that all sentences of a par*
ticular infinite sequence belong to the system constructed. But since in every
phase of development of the system only a finite number of sentences is ‘ effec­
tively' given to us, this fact can only be established by means of raetamathe-
matical considerations. The rule of inference mentioned has recently been
discussed by D. Hilbert, see Hilbert, D . (29).
1 The validity of the remarks stated in the last paragraph essentially
depends on the decision not to use in metamathematical discussion any devices
which cannot be formalized within the framework of the theory of types of
JPrincipia Mathematica. But as soon as we abandon this decision and allow
ourselves to use stronger devices in metamathematical discussion, most of the
remarks as originally formulated lose their validity and the whole paragraph
requires a thorough revision. Compare, in this connexion, article VTH, Post­
script, p p .274 ff.
 X

SOME M E T H O D O L O G IC A L
IN V E STIG A TIO N S ON T H E
D E F IN A B IL IT Y OF CONCEPTS!

Introductory R em arks

I n the m ethodology o f the deductive sciences tw o groups o f

concepts occur which, although rather rem ote from one another
in content, nevertheless show considerable analogies, if we con­
sider their role in the construction o f deductive theories, as well
as the inner relations between concepts within each o f the two
groups them selves. T o the first group belong such concepts as
'axiom9, 'derivable sentence9 (or 'theorem9), 'rule of inference9,
'proof9, to the second— ‘primitive {undefined) concept9(or ‘primi­
tive term9), 'definable concept9, 'rule of definition9, 'definition9.
A far-reaching parallelism can be established between the con­
cepts o f the tw o groups: the prim itive concepts correspond to
the axiom s, the defined concepts to the derivable sentences, the
process and rules o f definition to the process and rules o f proof.
H itherto, investigations in the dom ain o f the m ethodology of
the deductive sciences have chiefly dealt with the concepts of
the first group. Nevertheless, in considering the second group
o f concepts m any interesting and im portant problem s foroe
them selves upon us, some o f which are quite analogous to those
which arise in connexion w ith the first group. In the present
article I propose to discuss tw o problem s in this dom ain, namely
the problem of the definability and the mutual independence of

t B ibliograph ical N ote . This article is based upon the text of an address
given by the author in 1934 to the Conference for the Unity of Science in Prague.
It first appeared in print in Polish under the title ‘Z badaii metodologicznych
nad definiowalno6c$ terminbw* in Przeglqd Fllozoficzny, vol. 37 (1934), pp. 438-
60. Later it was published (in a somewhat condensed form) with the title
‘Einige methodologische Untersuchungen ttber die Definierbarkeit der Begriffe’
in Erkenntnis, vol. 5 (1935), pp. 80-100. The present version is a translation
of the German one supplemented by some passages translated from the Polish
original. Further historical references are contained in footnote 1, p. 297.
X TH E D E F IN A B I L IT Y OF CONCEPTS 297

concepts and the problem of the completeness of concepts o f an

arbitrary deductive theory .1
W e shall be concerned here only w ith those deductive theories
which are based upon a sufficiently developed system o f m athe­
m atical logic. Problem s concerning concepts o f logic itself will
n ot be considered. In order to make the discussion more specific
we shall have in mind, as the logical basis o f the deductive
theories discussed, the system o f Principia Mathematical o f A . N .
W hitehead and B . Russell,8 m odified in certain respects. These
m odifications are the follow ing: ( 1) it is assumed that the ram ified
theory o f types is replaced b y the simple theory, and that the
axiom o f reducibility is set aside; ( 2) in the system o f logical
axiom s we include the axiom o f extensionality (for all logical
types), and consequently we identify class-signs w ith one-place
predicates, relation-signs with tw o-place predicates, and so on ;8
(3) we use no defined operators such, for exam ple, as fx(<f>xY or
\'x)(<f>xY\ (4) as sentences, and in particular as provable sen­
tences, o f the system only those sentential functions are con­
sidered which contain no real (free) variables.
D eductive theories based upon a system o f logic can be
roughly characterized as follow s: In addition t o the logical
constants and variables, such a theory contains new term s, the
so-called extra-logical constants or the specific terms o f the theory
in question; to each o f these term s a definite logical type is
assigned. Am ong the sentential functions o f the theory we in­
clude the sentential functions o f logic, as well as all expressions
which are obtainable from them b y replacing real variables (not
necessarily all) b y extra-logical constants o f the corresponding

1 The results concerning the first problem were communicated by me to the

session o f the Warsaw Section o f the Polish Mathematical Society on 17 Decem­
ber 1926; cf. A. Tarski et A. Lindenbaum (85). The discussion and results
relating to the second problem were first communicated by me to the session
o f the Logical Section o f the Warsaw Philosophical Society on 15 June 1932.
* Cf. (90). W e have taken this system as a basis for our discussion chiefly
because it is more developed than other systems of logic and because it is
widely known. It should be mentioned, however, that Principia Mathematica
is not very suitable for methodological investigations because the formalization
o f this system does not satisfy the present-day requirements o f methodology.
# In connexion with these modifications cf. Carnap, R. (8), pp. 21-22, and
(10), pp. 98-101.
298 M E TH O D O LO G IC AL IN V E S T IG A T IO N S ON X

type. Sentential functions in which n o real variables occur are

called sentences. A m ong sentences logically provable sentences
are distinguished; these will be the sentences which are provable
in logic, as well as those which can be obtained from them b y
legitim ate substitution o f extra-logical constants in the place o f
variables (it w ould be superfluous to specify here in what a
legitim ate substitution consists).
L et X be any set o f sentences and y any sentence o f the
given theory. W e shall say that y is a consequence of the set X
o f sentences, or that y is derivable from the set X , i f either y is
logically provable or if there is a logically provable im plication
having a sentence belonging to X , or a conjunction o f such
sentences, as its antecedent, and having y as its consequent.
As is well known, various other concepts belonging to the
m ethodology o f the deductive sciences can be defined in terms
o f the concept o f derivability, such, for exam ple, as the conoepts
o f a deductive (or closed) system, the (logical) equivalence o f two
sets o f sentences, the axiom system o f a set o f sentences, the
axiomatizability, the independence, the consistency and the com­
pleteness o f a set o f sentences. W e shall assume that all these
ooncepts are understood .1* In accordance with custom ary usage
w e shall identify the set which consists o f a single sentence with
the sentence itself; thus we shall speak o f the consequences o f a
sentence or o f the consistency o f a sentence.

§ 1. T he Pboblem op the D evin ability and of the Mutual

I ndependence of Concepts
The problem s to be disoussed in this artdole concern the specific
term s o f any deduotive theory .8

1 Of. for example articles HI and V.

* It has been my aim to keep this article short and to make it generally
intelligible. I have not attempted, therefore, to give my exposition a strictly
deduotive character or to present it in a preoise form. I have, for example,
used expressions in quotation marks, I have often oontented myself with a
schematic representation, rather than an exaot descriptions of expressions, etc.
No essential difficulties would be encountered in modifying this presentation
so as to meet at least those standards of precision at which I have aimed in
some of my other methodological writings (e.g. IX).
x, s i TH E D E F IN A B IL IT Y OF CONCEPTS 299

L et *a ’ be som e extra-logical constant and B any set o f suoh

constants. E very sentence o f the form ;

(I) {x) :x = a. == .<f>(xi b', b”,...),

■where *<f>(x; b', 6',...) ’ stands for any sentential function whjoh
contains V as the only real variable, and in which no extra-
logical constants other than 'bn, ‘b0\.., o f the set B occur, will
be oalled a possible definition or sim ply a definition of the term
V by means of the terms of (he set B. W e shall say that the term
‘o ' is definable by means of the terms of the set B on the basis of the
set X of sentences, if 'o ’ and all terms o f B occur in the sentences
o f the set X and i f at the same tim e at least one possible definition
o f the term ‘o ' b y means o f the terms o f B is derivable from the
sentences o f X .1
W ith the help o f the concept o f definability we can explain
the m eaning o f various other m ethodological concepts which
are exaotly analogous to those defined in term s o f the concept
o f derivability. F or exam ple, the concepts o f the equivalence
of two sets of terms, o f the system of undefined terms for a given
set of terms, etc. It is clear that all these concepts must be
relativized to a set X o f sentences .1 In particular, B will be
called an independent set of terms or a set of mutually independent
terms, with respect to a set X gf sentences, if no term o f the set B
is definable b y means o f the remaining terms o f this set on the
basis o f X .
A ttention w ill be confined here to the case where the set X o f
sentences, and consequently the set B o f term s, is fin ite; the
extension o f the results reaohed to any axiom atizable set o f
sentences presents no difficulties.
M any years ago A . P adoa sketched a m ethod which enables

* It is not difficult to see why the oonoept o f definability, as well as all

derived concepts, roust be related to a set o f sentences: there is no sense in
discussing whether a term can be defined by means o f other terras before the
meaning o f those terms has been established, and on the basis o f a deductive
theory we can establish the meaning o f a term which has not previously been
defined only b y describing the sentences in which the term ocours and which
we accept as true,
800 M E TH O D O LO G IC A L IN V E S T IG A T IO N S ON X, § I

us to establish, in particular cases, the undefinability o f a term

b y means o f other term s .1 In order, by this m ethod, to show
that a term V cannot be defined b y means o f the terms o f a set
B on the basis o f a set X o f sentences, it suffices to give tw o
interpretations o f all extra-logical constants which occur in the
sentences o f X , such that ( 1) in both interpretations all sentences
o f the set X are satisfied and ( 2 ) in both interpretations all
terms o f the set B are given the same sense, but (3) the sense
o f the term V undergoes a change. W e shall here present some
results which provide a theoretical justification for the m ethod
o f Padoa and, apart from this, appear to throw an interesting
light on the problem o f definability .2
W e consider any finite set X o f sentences, an extra-logical
constant V which occurs in the sentences o f X . and a set B
o f such constants which does not, however, include V among
its members. The conjunction o f all sentences o f the set X
(in an arbitrary order) we represent in the schem atic form
b\ c\ c ' , . h e r e ib,\ ‘h*’,... are terras o f the set B ,
and V 5, ‘c "’ ,... are extra-logical constants which occur in the
sentences o f X but are distinct from V , (bn, (if there are
no such constants the form ulation and p roof o f the theorems
given below are somewhat sim plified). It sometimes happens
th a tin a ll sentences o f the set X variables V , (y f\ V '\ V ’ ,
are substituted for (some or all) extra-logical constants, it being
assumed that none o f these variables becom e apparent (bound)
variables as a result o f the substitution .3 In order to represent
schem atically the conjunction o f all sentential functions formed
in this way, the same substitution is made in the expression
‘</r(a; b', ft*,...; c\ c",...)\ On the basis o f these assumptions and
conventions the follow ing theorem can now be proved:

1 Of. Padoa, A. (67), pp, 249-56 and (58). pp. 85-9L

2 An essential advance in this domain has recently been made by E. W.
Beth, who has extended The. 2 and 3 o f this article to a much wider class of
deductive theories, in fact, to all theories based upon the lower functional
calculus (which is a much weaker logical system than Prinoipia Mathematita),
and has thus shown that Padoa’s method can be applied to all theories o f this
olass. Cf. Beth (6a).
8 A similar assumption is tacitly made in all the analogous situations which
occur in this ar tiole.
X,§! TH E D E F IN A B IL IT Y OF CONCEPTS 301

T heorem : I. In order that the term ‘a* should be definable by

means of the terms of the set B on the basis of the set X of sentences,
it is necessary and sufficient that the formula
(II) ( * ) :* - < * . ~
should be derivable from the sentences of X .
Proof. It follow s directly from the definition o f the concept o f
definability that the condition o f the theorem is sufficient; for
(II) is a form ula o f the type (I) and thus & possible definition o f
the term V by means o f the term s o f B.
I t remains to be shown that this condition is at the same tim e
necessary. Let us assume that the term V can be defined by
means o f the terms o f 8 on the basis o f the set X , and thus that
at least one form ula o f type (I) can be derived from the sen­
tences o f X . This form ula is at the same tim e a consequence o f
the conjunction o f all sentences o f X , i.e. o f the sentence
€* (a ; 6',
In accordance with the definition o f the concept o f consequence
we conclude from this that rhe form ula
( 1) 0 (a; b\ 6",..., c\ c *,.,.)• d :( x )
is logically provable. Since the form ula ( 1) is a logically provable
sentence with extra-logical constants, it must be obtainable by
substitution from a logically provable sentence without such
constants, and in fact from the form ula
( 2) (xr, y', y ',..., z', z*,...)
y\ y % .; z', z ',...). d :
:(x) :x = x'. .# e ';y ',y V ..)-
B y means o f easy transform ations we obtain from (2 ) further
logically provable form ulas:
(3) {x, x\ y\ y ",..., z\ z*,...) y\ y*,...; z ',...). 3 :
:x — x'.== 4{x\ y ',y V .) ;
(4) (x>y' *y”, #*; */ ' > ^ :
: x =: x . == .</>(x;y\ y*,-..);
(5) (*»y'> A y'>y',-; A A-)- ^•
.<f>{x;y\ y ”
( 6) (*> y ' . - ) : (3*', y', y*,...; z', z*,...). :> .
•^toy'.yV.)-
802 M E TH O D O LO G IC A L IN V E S T IG A T IO N S ON X, § I

From (3) and (6) we derive

(7) (x, x , y , y z , 2 )•>•) z>>fi(x : y , y )■.•) z , z :
: (3z', z',...)4{x\ y\ y*,...; z', z',...) z o . x — x',
whence
(8) (x, a?', y\ y',..., z', zr, * f > ( x ' ; y\ y " z', z ',...). 3 :
: (3*', z',...).^ (* ; y*, y ',...;2', o . x — x'.
On the other hand the logical definition o f the identity sign
yields
(9) (as, as', y', y ',..., z', :.^(as'; y', y ',...; z', z ',...). 3 z
: x *» 3 .^(*; y', y ',...; s', 2*,...)
whence
(10) (x, x\ y\ y ',..„ z', as',,..) z.tfi{x'i y', y ',...; 2', 2',...). 3 :
d .(3s', 2V.).^(*}y', y',...; z', 2',...),
From (8) and (10) we obtain step by step the formulas:
(11) (*, * f, y', y ',..., 2', 2*,.-) y', y ',...; •a :
**%
'uj

U*
M
^•

*
IS?
N
II

iii
8

00 •
>* •
«

00

00

0
*0
**

V*
•
•

•
•
•

(12) ( » , y , y >•••>2 , 2 ,...) ..^ ( 2 1 y 1 y »•••>2 , 2 ,,..). 3 :

: (x) :x *** s= .(3*', z',.,.).^(as; y', y ',...; z', *',...);
(13) b‘, 6 ',..,; o', c ',...). 3 : (x) zx = 0.

. b . (3z', z",...).tft(x; b', 6",...; 2',...).

Since (18) is logically provable, the consequent o f this im­
plication, i.e. formula (II), is derivable from the antecedent
ty(o; b*, b”,...] c‘, c ',...)’, and hence from the sentences of the
set X . Consequently the condition o f the theorem is not only
sufficient, but also necessary, which was to be proved.
The importance o f the above theorem lies in the following:
if a term can be defined at all by means o f other terms, then
our theorem enables us to construct ‘effectively’ a possible
definition o f this term. Moreover, the theorem shows that the
problem o f the definability o f terms is reducible without re­
mainder to the problem o f the derivability of sentences.
The next theorem is essentially a transformation of Th. 1.
x .! i TH E D E F IN A B IL IT Y OF CONCEPTS 808
T eeo bbm 2 . In order that the term ‘a ' should be definable by
means of the terms of the set B on (he basis of the set X of sentences,
it is necessary and sufficient (hat the formula
(H I) (* ', x ”, y', y*,..., z\ z',..., t', * ',...) :^ (a '; y ', y '....; *’ ....)•
.^>(a>*; y ', y ",...; t\ f*>...). 3 «* as*
should be logically provable.
Proof. I f the term ‘o ’ is definable w ith the help o f the term s
o f the set B, then, b y Th. 1, sentence (II) is a consequence o f the
set X o f sentences and henoe is also a oonsequenoe o f the con­
junction o f all sentences o f this set, i.e. o f the sentence
^ (o; V, 6*,...; o', a",...).
Therefore we conclude that the sentenoe
( 1) tfi(a; b', 6*,...; O', o*,..,). z> :(a?) :* «= a. =s .( 3*', z*,..,).
<P(x; b', z\ s ',...)
is logically provable.
Thus the sentence
( 2) (*\ y ', y \ ..., tr, * ',...) :.# z * ; y ', y*,...;
d : (a;) s .(3 2', z ',...).^ (* ; y f, y * „.,; s ', 2*,..,),
o f which ( 1) is a substitution, m ust also be logically provable.
From ( 2), b y means o f sim ple transform ations, we successively
derive som e further logically provable sentences;
(3) (x , y“, y*..... t', *',...) :.^(a?*; y', y*,...; f",...}.
d :» * ss ,(3 z ', z ',.,.).^ (* ;y ', y ',...;z ', z ',...);
(4) (*, 2/', y*,..., f, . . . ): . ^(x"; y ’, y * ,..,; l', t*,...}.
3 : (3z', z ",...).^ (» ; y ', y *,...; z', z ',...). o
(5) (*, x *, y ', y " ,...,«', < ",...):, (3 z', z ',...) .
y\ y ',...; a', *',...) :^ (* '; y', y * , . . . ; *',...):
: n .* e=
( 6) (*, as*, y ', y ',..., z', z '..... t\ t*,...) :^(a;; y ', y *,...; z ', z*,...).
M»*; y'»y'»-i t\ 13.*=®»*;
(7) (x*, x", y ', y ',..., z', z '.....t*,...) :^ (* '; y ', y*,...; z', z*,...).
# e * ; y ', y',..M *",...)• 3
Sentence (7) coincides with (ITT). H ence we have shown that the
304 M E TH O D O LO G IC A L IN V E S T IG A T IO N S ON X, § 1

condition o f the theorem is necessary for the term V to be

definable b y means o f the term s o f the set B. W e have still to
prove that the condition is sufficient.
F or this purpose we reverse the reasoning sketched above.
W e assume that the sentence (H I), i.e. the sentence (7), is
logically provable. B y successive transform ations we show that
sentences ( 6), (5), and (4) are logically provable. In addition, it
follow s from the definition o f the sym bol ‘ = ’ that the sentence
( 8) (x, y\ t y \ y ' t , t",...).
. D :* = x '. o.<f>{x; y', y",...; t’ ,
is also logically provable, and this im plies
(9) (a, x\ y\ y ’ ,..., t\ t’ ,...):,ift(xr; y '} y" ,...; ? , *',..•)•
. d :x = 3 .(3 z ', z * y ’, y ',...; s', z ',.„ ).
F rom (4) and (9) we im m ediately obtain (3), and then (2)
and ( 1).
Since ( 1) is logically provable, the consequent (II) o f the
im plication ( 1) m ust be a consequence o f b', o', c ",...)’,
the sentence which is the antecedent o f ( 1), and hence (H )
must be a consequenoe o f the set X o f sentences. H ence, by
Th. 1, the term ‘o ’ is definable b y means o f the terms o f the
set B on the basis o f the set X . Thus the theorem has been
proved in both directions.
From the above theorem we obtain as an im m ediate corollary
T heorem 3. In order that the term ‘o ’ should not be definable by
means of the terms of the set B on the basis of the set X of sentences,
it is necessary and sufficient that the formula
(IV ) (3a:', x", i f , y" .....z’ ,..., V, f',...).
y', y ',...; z', z*,...).^ (a ;'; y', y',..., t’, t",...).x' # x"
should be consistent.
In order to show this, it suffices to note that form ula (IV ) is
equivalent to the negation o f form ula (III) o f Th. 2, and then
to apply the general m ethodological principle according to which
a form ula is not logically provable if and only if its negation is
consistent .1
1 Of. for example Th. 13* in article I I I ; we substitute in this theorem ‘O’ in
the place of *X' keeping in mind the remark which follows Th. 4 in that article.
X,§1 TH E D E F IN A B IL IT Y OF CONCEPTS 305

Th. 3 constitutes the proper theoretical foundation for the

m ethod o f P adoa. In fact, in order to establish, on the basis o f
this theorem , that the term V is not definable b y means o f the
term s o f the set i?, we can apply the follow ing procedure. W e
consider a deductive system Y, the consistency o f whioh has
previously been established or is assumed. W e then seek certain
constants (not necessarily extra-logical) €b'\ ‘b*’,...,
*c'\ ‘5*’ ,..., which occur in the sentences o f Y and
satisfy the follow ing conditions: ( 1) if in all sentences o f the set
X we replace the sym bols lb’ \ ‘ft"’,..., V J, V ’,.., b y ‘a’,
4&", V \ ...y ‘c'1, respectively, then the resulting sentences
belong to the system Y ; ( 2) the same holds also if we substitute
the sym bols ‘ 6% ... exactly as before for the terms *V*%
‘i * ’ ,... o f B y but replace the remaining terms V , *cn, ‘c*’ ,,- b y
‘ 8*’ ,..., respectively; (3) the system Y contains the
form ula ld ^ From the foregoing conditions it results that
the follow ing conjunction belongs to the system Y:

b\ c\ 5', & V.; W, d ^ U.

H ence it easily follow s that form ula (IV ) o f Th. 3 which is an
im m ediate consequence o f this conjunction also belongs to the
system Y. A ccording to a general m ethodological principle,
every part o f a consistent set o f sentences is consistent; in parti­
cular, the form ula (IV ) as an elem ent o f the consistent system
Y must be consistent. Thus the condition o f Tfc. 3 is satisfied
and the term ‘a ’ is not definable exclusively b y means o f the
term s o f B . In the above procedure we easily recognize the
m ethod o f Padoa, in an extended form , inasmuch as provision
is made for the possible occurrence o f terms which are identical
neither with ‘a ’ nor with any o f the terms o f B.
The m ethod o f Padoa, or T h ..3, is often successfully applied
in establishing the mutual independence o f the terms o f an
arbitrary set B. The m ethod must obviously be applied as
m any tim es as there are elements in the given set o f terms.
The problem o f independence has a certain practical im portance
in those cases when the set B involved is the system o f prim itive
terms o f a deductive theory. For, should the prim itive term s
306 M ETH O D O LO G IC AL IN V E S T IG A T IO N S ON X, f 1

prove to be n ot m utually independent, then the unnecessary

(i.e. definable) terms can be elim inated and this sometimes makes
possible a sim plification o f the axiom system . I t is perhaps
w orth noting that under the term inology accepted here the
situation can be described in precise terms as follow s: I f the set
B is divided into tw o disjoint sets 0 and B , and every term o f
the set D is definable b y means o f the terms o f the set G on the
basis o f the axiom system X , then the system X can be replaced
b y an equivalent axiom system which again consists o f two
parts 7 and 7- such that the sentences o f the set T, so-called
proper axioms, contain extra-logical constants o f the set 0 ex ­
clusively, whilst every sentence o f the set Z is a possible defini­
tion o f a certain term o f the set D b y means o f the terms o f the
set G.
L et us now give an exam ple. Consider the system o f n-dim en-
sional Euclidean geom etry (where n is any natural num ber).
Let ua assume that the sym bols ‘o ’ and ‘ 65 denote certain
relations between points, ‘a ’ denoting the four-term ed relation
o f congruence o f two pairs o f points, and *b' the three-term ed
relation o f lying between. The expressions 'a lx,y,z,t)’ and
*b(x,y,^y are thus read: ‘the point x is the same, distance from
the point y as the point z is from the point t’ and ‘the. point y
lies between the■points x and z’, respectively. The term ‘o ’ may
be called the metrical primitive term- and *6’ the descriptive
primitive krm. o f geom etry. As is well known, the set consisting
sim ply o f the tw o terms ‘a ’ and ‘b’ can be taken as the system
o f prim itive terms o f geom etry. A ccordingly an axiom system
X o f geom etry can be constructed in which these tw o terms
constitute the only extra-logical constants. The question now
arises whether the term s ‘o ' and ‘ 6 ’ are m utually independent
With respect to the set X o f sentences. The question is to be
answered in the negative. This is so despite the fa ct that, by
P adoa’e m ethod, the term ‘o ’ can be shown n ot to be definable
b y means o f the term ‘ 6 '; in fact, assuming the consistency o f
geom etry it can be proved that form ula (TV) o f Th. 3 is in
this case consistent. (It is interesting that in essence this
p ro o f appears in oertain investigations on the foundations o f
X .| X T H E D E F IN A B IL IT Y OF CONGESTS 307

geom etry which apparently serve a directly opposite goal, in fact

the establishm ent o f m etrical geom etry as a part o f descrip­
tiv e .)1 Nonetheless it is found that, if the system o f geom etry
is at least tw o-dim ensional, then the term is definable b y
means o f the term ‘a ’ on the basis o f the axiom system X ,
The definition can be constructed according to the scheme (II)
o f Th. 1 (although other m uch simpler definitions o f the term
*&’ are also know n). In accordance w ith this, the System X oan
be replaced b y another equivalent axiom system , in the axiom s
o f which the term 'o ’ occurs as the on ly extra-logical constant,
apart from a single ‘improper* axiom , nam ely a possible defini­
tion o f the term ‘b\ As A . Lindenbaum has proved 2 (with the
help o f the axiom o f choice), it is only in the case o f one­
dim ensional geom etry, i.e. the geom etry o f the straight line,
that the tw o term s ‘a ’ and ‘ 6 * are independent o f one another.
The follow ing is also w orth em phasizing. As a starting-point
for this discussion we have used a special form o f definition,
nam ely the scheme (I). Nevertheless the results obtained re­
main valid for other known form s o f definition. In order to
illustrate this b y an exam ple, lot us suppose that *a ’ is a tw o-
place predicate. As a possible definition o f the term 'a* b y
means o f the terms o f the set B we can consider, for exam ple,
every sentence o f the form
(la ) (« ,« ) :a (« ,v ). == .^ (« ,» ;
where v; b',b",...)’ stands for any sentential function which
contains no real variable apart from V and V , and no speoific
terms apart from the terms ‘b' ’ , ‘ 6* *,,.. o f the set B ? It oan
1 Cf. for example Veblen, O. (86).
* Of. A . Tarski and A . Lindenbaum (85).
* The form of definition (la) is much more special than (X). but it has the
advantage that the consistency of the definition is already ensured by its
structure. In order to be sure that within & deductive theory definitions of
type (I) will lead to no oontradictionp a stipulation is usually made not to
introduce such a definition without first having proved the sentence:
(A) < 3 * ).# * ; h\ 6* ...).#(*** b\ b% . ) . ? . * -
I f this sentence is proved, then we can even replace (I) by the simpler
equivalent formula
(B) *(a;
for the two conditions for a correot definition— consistency and re-tranalata-
bility— are guaranteed by sentence (A).
SOS M E TH O D O LO G IC AL IN V E S T IG A T IO N S ON X, § 1

now easily be shown that the definition of the concept of defina­

bility, relativised to sentences of the form (la), is equivalent
to the original definition (although naturally only with respect
to two-termed predicates). From this it follows immediately
that Ths. 1-3 also remain valid on the basis of the new defi­
nition. There is no difficulty in transforming Th. 1 in such a way
that a formula of type (la) occurs in it instead of formula (II).
It suffices, in fact, to replace (II) by
(Eta) (utv):.a(u,v). ss s ' , z',z*
o.x(u,v).

§2. T he Problem of th e Co m pleten ess of Concepts

The problem o f the definability and t he mutual independence

o f concepts is an exact correlate o f the problem o f the derivability
and the mutual independence of sentences. The problem o f the
c o m p l e t e n e s s o f c o n c e p t s , to which we now turn, is also closely

related to certain problems concerning systems of sentences, and

indeed to the problems of completeness and categoricity, al­
though the analogy does not extend so far as in the previous case.
In order to make the problem of completeness precise, we
first introduce an auxiliary concept. Let X and Y be any two
sets? o f sentences. We shall say that the set- Y is essentially richer
iaan the set X with respect to specific terms, if (1) every sentence
o f the set X also belongs to the set Y (and therefore every
specific term of X also occurs in the sect cnees of Y) and if
tV.) in the sentences of Y there occur specific terms which are
ah-enc from the sentences o f X and cannot- bo defined, even on
the basis of the set Y, exclusively by means of those terms
which occur in X .
I f now there existed a set X o f sentences for which it- is im­
possible to construct an essentially richer set Y of sentences
with respect to specific terms, then we should be inclined to
gay that the set X is complete with respect to its specific terms.
L appears, however, that there are in general no such complete
sets o f sentences, apart from some trivial oases. In fact, for
X , §2 TH E D E F IN A B I L IT Y OF CONCEPTS 309

every set X of sentences which is consistent and which does not

contain all extra-logical constants, we can construct a set Y
which is essentially richer in specific terms. For this purpose it
suffices to add to the set X an arbitrary logically provable sen­
tence which contains an extra-logical constant not occurring in
the sentences o f X .1 For this reason the proposed definition
o f completeness requires modification. In this the concept of
ca tegoricity will play an essential part.
As is well known, a set of sentences is called categorical if
any two interpretations (realizations) of this set are isomorphic.2
In order to formulate this more exactly, let us suppose that
symbolic expressions o f the form V x "’ (in words : ‘the, rela ­
tion R maps x f onto x” 1) have been introduced into the system
o f logic. The variable ‘i?’ always denotes a two-termed relation
between individuals, but V 5and *x*' may be o f any logical type,
so long as they are both o f the same type. The precise sense
of the expression V T^x"’ will depend upon the logical type
o f the variables V ’ and ‘x *r. We shall explain it for only a
few cases. For example, if x : and x n are individuals, then
V x'ft has the same meaning as ‘x'Rxni (i.e. V stands in the
relation li to r/‘), i f fx f’ and V '5 denote classes o f individuals,
then the expression has the same sense as the sentential function
(iif) : V e s \ : m (3%*) . u * G x * . u , r Z u * : . { u ” ) :u* e x ’

3.(3u ') . u 'e z ’ .u " % u *

(this involves no circularity since V * and V ” are individual
varia bles, so that the sense o f the expression V T ' has
already been determined). In an exactly analogous manner the
express.k-n V can be denned for class terms of higher
type. Consider again the case when V s and V'* are two-termed
predicates with individual variables as arguments (thu* denoting

1 I.a {'Oiinaxion tribh what. £& paid ao. p. ^99* no to I , it is. obvious that the
new eXuiNviogie&I constant cannot be ieiuied by means o f the terms o f X .
h i foot ‘ he Oiiiy sentence of Y in which fchi& constant occurs fo logically provable,
and i \\m trcu? independently ni the meaning: o f the specitin contained
in it. Jt as® thus be asserted th&t the meaning o f the constant jh question, is
not at all determined by the set Y o f sentences.,
8 Of. for example Yeblen, O. (86).
310 M E TH O D O LO G IC AL IN V E S T IG A T IO N S ON X, | 3

binary relations between individuals); then the form ula V 'a

has the same significance as
{%’, v') :*'(«'. v'). t>.(3u*,v*).x'(ur,*»').«' y u’ .v’ y v” :,
v*). o .(3 u ',v '). x ' ( u ', b' 'a ' v1
The above examples will probably suffice to explain the sense
to be given to the expression discussed with respect to variables
o f arbitrary logical type. M oreover, recalling that, in the nota­
tion o f Prhicipia Mathematicay V is the class o f all individuals
and J -> I the class o f all one-one relations, let us say that the
form ula
y\
R
x ,y ,z ....
is to have the same meaning as the conjunction

Re 1 1 .V y V .* ' y x ".y ' y y\ z' y z*...

(in w ords: ‘R i3 a one-one mapping of the, class of all individuals
onto itself, by which x', z',... are mapped onto x"y y", z’ ....,
respectively").1
Consider now any finite set 7 o f sentences; ‘a ’ , ‘b’y V ,... are
all specific terms which ocour in the sentences o f 7 , and
‘ip(a, b , is the conjunction o f all these sentences. The set
T is called categorical if the form ula

(V ) (x\ x”, y !, y" ,z' y Z ," \ $ (x ‘ ,y ‘yz',...) .${x" ,y" ,z ”,...)

D.(3jg).iz^ ;y -
x »y ,2
is logically provable .5

1 From the standpoint o f the theory o f types the mode of introduction o f the
symbolic expressions ‘ x f ^ x* * and ‘ B
y', z',:. * is not quite free from
x** V*,
objections. It should therefore be emphasized that these symbols are only to
be regarded as schemata o f expressions* In a precise presentation o f this
discussion the use o f suoh symbols becomes unnecessary and all difficulties
connected with them disappear.
3 W s use the word 4categorical *in a different, rcinewhat stronger sense than
te customary : usually it is required o f the relation B which occurs in (V) only
that it maps a1', y 9 z\ .„ onto x*t y*9 z*f.„ respectively, but not that it maps
the class o f all individuals onto itself. The sets o f sentences which are cate.
x , J2 TH E D E F IN A B IL IT Y OF CONCEPTS 311

On various grounds, which will not be entered into further,

great importance is ascribed to categorioity. A non-oategorical
set o f sentences (especially if it is used as an axiom system o f
a deductive theory) does not give the impression of a closed and
organic unity and does not seem to determine precisely the
meaning o f the concepts contained in it. We shall therefore
subject the original definition o f the concept o f completeness to
the following modification: a set X of sentences is said to be
complete with respect to its specific terms if it is impossible to
construct a categorical set ¥ o f sentences which i3 essentially
rioher than X with respect to specific terms. In order to estab­
lish the incompleteness of a set o f sentences it is from now on­
wards requisite to construct a set o f sentences which is not only
essentially richer but also categorical. Trivial constructions in
which the meaning o f the newly introduced specific terms is
quite indeterminate are thus excluded from the beginning.1
With this conception of the notion of completeness the exist­
ence o f both complete and incomplete sets of sentences can
be established. For example, axiom systems o f various geo­
metrical theories are incomplete with respect to their specific
terms, but various axiom systems o f arithmetic are oomplete.2
Let us consider this more olosely with reference to specific
examples. Consider the system of descriptive one-dimensional
geometry, i.e. the totality o f all geometrical sentences concern­
ing points and sub-sets o f a straight line, in which the only
extra-logical concepts are the relation o f lying between, and
the concepts definable by means o f it, For this geometry
gorioal in the usual (V eb W s) sense can be called intrinsically categorical, those
in the new sense absolutely categorical* The axiom systems of various deductive
theories are for the most part intrinsically but not absolutely categorical. It
is, however, easy to make them absolutely categorical. It suffices, for example,
to add a single sentence to the axiom system of geometry which asserts that
every individual is a point (or more generally one which determines the
number of individuals which are not points). It will become clear later (cf.
p. 314 footnote) what it is that leads us to suob a modification of the concept
of categorioity. It may be mentioned that the concept- of categorioity which
we use belongs to the a-concepta in the sense of B. Caraap; but the corre*
sponding f-concept can also be defined.
1 Cf* p. 309, note l and the corresponding place in thfe text,
3 This is only exact if the axiom system of arithmetic contains sentences
to the effect that every individual is a number (cf. p. 310, note 2).
312 M E TH O D O LO G IC AL IN V E S T IG A T IO N S ON X, § 2

it is easy to formulate a categorical axiom system X, which

contains the term of that relation as the only primitive term.
The system X 1 is not complete with respect to its specific
terms. A h w e have already mentioned in the previous section in
connexion with the problem of independence, various concepts
occur in metrical geometry which cannot be defined with the
help of the descriptive ones exclusively, e.g. the relation of the
congruence of two pairs of points. If we now take the symbol
of this relation as a new primitive term and adequately extend
the axiom system X v we obtain an axiom system X 2 for the
full metrical geometry of the straight line. This system X %is
again categorical and at the same time essentially richer than
X 1 with respect to specific terms. But X %is also incomplete
with respect to its specific terms. It ean, for example, be proved
that on the basis of this system not a single symbol can be
defined which denotes an individual point. For that reason it
is easy to set up a new categorical set X 3 o f sentences which is
essentially richer chan X 2 with respect to specific terms, For
this purpose it suffices to introduce a n y t,wo symbols, say
and Mb as new primitive terms, and to supplement X 2 with a
single new axiom to the effect that these symbols denote two
distinct points. The deductive theory founded on the axiom
system Jf3 is identical, from a formal standpoint, with the
arithmetic of real numbers; formally the arithmetic of real
numbers is nothing but the geometry o f xr.e straight iine in
which two points have been ‘effectively’ cb^fcmgnisbed, The
question now arises whether the above process can be continued
ad vnfiniiwta, i.e. whether it is possible to ecuaru'uoi icr X* a new
set X A o f sentences which is categorical an- essentially richer
with respect to specific terms, for this set IY . a, new and
so on. Thie proves to be impossible, the set of sentences is
already complete with respect to its speoirh .erm».
Ho method is apparent a priori which would enable us to estab­
lish the completeness of any particular *q- off rcirnmces, e.g. the
set discussed above. For u set o f can usually be
extended in infinitely many ways, and to proxv* the completeness
of this set it would be necessarv * to show', that- none of these
X,§2 T H E D E F IN A B IL IT Y OF CONCEPTS 313

extensions, if it is categorical, is essentially richer in specific

term s than the set discussed. W e shall give here a result which
makes our task m uch easier in all cases actually known. In fact
we shall show that the completeness o f sets o f sentences, with
respect to the sets o f terms occurring in them, is a consequence o f
another property o f such sets which m ay be called monotrans-
formability. The problem whether a given set o f sentences actually
has this new property presents, in general, no greater difficulties.
The notion o f m onotr&nsform ability is closely related to that
o f categoricity. A set o f sentences is categorical if for any tw o
realizations o f this set at least one relation exists which estab­
lishes the isom orphism o f the tw o realizations; it is m onotrans -
form able if there is at m ost one such relation. In precise terms,
the set X o f sentences is m onotransform able if the sentence
(V I) ( * ',a", y \ y ',...y R" ): y ',...).
x', y\... *i/ , y ,<•>
.R .R " .o . R = R'
y\...
is logically provable, where represents the conjunc­
tion o f all the sentences o f X .3
A set o f sentences which is both categorical and monofcrana-
form able is called strictly (or uniquely) ca tegorica l. The iso­
morphism o f any tw o realizations o f such a set can thus be
established in exactly one way. F or example, the various axiom
system s o f arithm etic are strictly categorical— both the arith­
m etic o f natural numbers (e.g the axiom system o f Peano) ; and
that o f real numbers.2 B ut the axiom system s o f various geo­
metrical theories— o f fopclogy, projective, descriptive, metrical
geometry, etc.— although usuaih categorical, are not mono-
transformable and hence not strictly categorical.
This applies, in particular, to the ^ets X x anu X 2 discussed
above. By way o f example we should like to explain this in
detail for the axiom system X A ? we know the orJv specific
term occurring in this set is the name o f the relation o f lying
1 We can again distinguish two senses of :moziotransfont3able\ intrinsic' end
absolute monoir&nsformabilit/y. IThr^ the term is always used ir the second
sense (cf. p. 310, note 2),
* This is exact only if the axiom system o f arithmetic contains sentences
to the effect that every individual in a number (cf. p. 310, note 2).
914 M E TH O D O LO G IC AL IN V E S T IG A T IO N S O F X, § 2

between. The system X xis categorical; i f we find any tw o inter­

pretations o f this system , w e can always establish a one-one
correspondence between points a, b, c,... in the first interpreta­
tion and points a% b', o’.... in the second in such a w ay that,
whenever b lies between a and c, 5' lies between a ' and o',
and conversely. It is found that a correspondence o f this kind
can be established in infinitely m any different ways. Therefore
the set X x is not m onotransform able and hence not strictly
categorical. On the other hand the axiom system X z (o f the
arithm etic o f real numbers) la m onotransform able and strictly
categorical. In this instance the one-one correspondence between
points (or numbers) for any tw o interpretations can be estab­
lished in exaotiy one way.
W e shall now prove the follow ing:
T heorem 4. Every monotransformable aet of sentences is com­
plete with respect to its specific terms.1
Proof. L et X be any m onotransform able set o f sentences, and
‘<£(a, b,...)’ the conjunction o f all sentences o f X . Form ula (V I)
is thus logically provable. As a substitution into this form ula
we obtain

(1) B) ■4{x,y,...).^x,y,...).Ri^ . l
x, y ,... x, y,~.
the sym bol here denotes, as in Principia Mathematica, the
relation o f identity (and thus has the same meaning as the
sign *=*’ ).
From the definitions o f the expression ‘21 ■■ ~ -■f -~ * a&d o f the
x ,y
identity sign we easily see that the form ula
x, y,...
(2) {x ,y ,...) .!
x, y,...
1 I f we wish to use the words ‘ categorical5and ’ monotransformable5 in the
intrinsic sense (cf. p. 310, note 2 andp. 313, note 1}, then Th. 4 in the above form
would be false, In order that it should remain valid the definition of complete­
ness must be weakened, absolute completeness must be replaced by intrinsic com*
pleteness or by completeness with respect to intrinsic concepts* For this purpose
it would be necessary to give first an explanation of what is meant b y *intrinsic
concept** All this would complicate our exposition considerably. In this con­
nexion compare the concluding remarks in X m ,
X,S3 THE D E F IN A B IL IT Y OF CONCEPTS 316

is logically provable, independently o f the typ e o f the variables

V , ‘y ’ ,.. . . In view o f ( 2), form ula ( 1) can be sim plified as
follow s:

(3) (a;,y,...,.R ) :<£(*,y ,...)..B s> y.~« D .JS ~ I.

a, y,...
(It is w orthy o f note that, not only does the form ula (3) follow
from the form ula (V I), but also, conversely, (V I) follow s from
(3); the tw o form ulas are therefore equivalent. The definition
o f m onotransform ability can thus be sim plified b y replacing
(V I) by (3). The new definition can be expressed as follow s:
a set o f sentences is m onotransform able if there are non-identioal
autom orphism s in none o f its interpretations.)
Consider now any categorical set 7 o f sentences which in­
cludes the set X as a part. W e assume that in the sentences o f
7 new extra-logical constants ‘g’, ‘h ’ ,...— as well as the old ones
V , —occur, and we represent the conjunction o f all these
sentences in the form ‘ift(a, b, k , I t is to be proved that
the set 7 is n ot essentially richer than X with respeofc to specific
terms.
From the categoricity o f 7 it follows that the form ula
(4) (x,,x * ,y ',y ’,,.,.,u ') u*)

2 .( a s ) . jb ~X ' , y , ~ ; V
A.*
i y »»••* ^ 5^
is logically provable. From this b y substitution we obtain
(3) (x,y,..., u', u", v , <?',...): i p ( x , v ' , . . . ) . r//(ar, y ..... v f . v r)
x, y,..„ u\
x, y,,.., u*> v",..
Since 7 contains all sentences o f X , the form ula
(6) (x, y,...,u\ t>',...): y,..., v f, » ',...). d . 4>('Ay ,...)
must also be logically provable. From (5) and ( 6) we obtain
im m ediately
(7) (x , y,.... u\ vf, v', v*,...): tp{x, y,..., u', v',...) .

t(x,y,...,u \ v\ ..,). d . (3R).<f>(x,y,...). I t »

x, y,..., u , v ,...
3ie M E TH O D O LO G IC AL IN V E S T IG A T IO N S ON X, J2

from (7), recalling the meaning o f the expression

s x t y ,..., u ',v 'y...
v\...
we derive
( 8) (x, y ,..., u', u%v', : ifs(x,y u \ v ',...).

d . (3 B ) .<j>{x,y,...) B ^ - - •

The form ulas (3) and ( 8) im ply

(9) {x,y,...,u',u“,v',v’ ,..;) v'
iji(x,y, . o .(3 .8 )..8 “ I .B
^ I^
whence, in accordance with the definition o f the identity sign,

( 10) {x,y,...,u ’ ,u ",v ',v ”,...) :ifi(x,y,...,n',v',...).

ifi(z,y,...,u”,v",...). d .1 it', v',-
u", v",.
Finally, it is easy to see that the sentence

( 11) * •«' =»

is logically provable, and in fact independently of the logical

type of the variables V ’ , V \ V ’ ,... . From formulas ( 1 0 )
and ( 1 1 ) we obtain directly
( 12) (x} y ,..., u9yu”, v\ v'7,...) : t//(x9y,..., u’\ t?',*..).
ifj(x9y,...,u", vn,...). 3 ,uf =s u* .v* ==s v”......
We have thus shown that ( 1 2 ) is logically provable. Since
‘ipia, b,.^9gyh9...)yis the conjunction of all sentences of the set F,
we conclude by applying Th. 2 . that each of the terms cg\ :h\
which are lacking in the sentences of X, can be defined on the
basis of Y by means of the terms V , ‘ 6 ’,... occurring in X .
Consequently the set Y is not essentially richer than X with
respect to specific terms. Hence it is impossible to construct
for the set X a categorical set of sentences-which is essentially
X,§2 TH E D E F IN A B IL IT Y OF CONCEPTS 317

richer than X . The set X o f sentences is thus com plete with

respect to its specific terms, which was to be proved.
The practical im portance o f the above theorem is clear. In
order to establish that a given set o f sentences is com plete with
respect to its specific terms, it suffices to show that this set is
m onotransform able and hence that the form ula (V I) can be
proved in logic. B ut this demands no special m ethod o f investi­
gation and, in the cases so far considered, presents no special
difficulties.
The question naturally arises whether tfie converse o f Th. 4
h old s: in other words, whether the m onotransform ability o f a
set o f sentences is not only a sufficient but also a necessary
condition for its com pleteness with respect to specific term s.
The problem is not yet solved; no one has so far succeeded
either in establishing the converse theorem or in constructing
a counter-exam ple.
In connexion w ith the theorem last given I should like to
make some observations which exceed the strict lim its o f this
article; they deal with the prospects o f deductively construct­
ing theoretical physics either in its entirety or in a particular
part. Suppose we have adopted the system o f four-dim ensional
Euclidean geom etry— the geom etry o f the space-tim e continuum
— as a foundation for the construction o f a physical theory.
Let the system o f axiom s and prim itive term s o f geom etry be
strengthened in such a way that on this basis a particular c o ­
ordinate system can be distinguished. Geom etry then form ally
coincides with the arithm etic o f hypercom plex numbers; its
axiom system is strictly categorical and hence, according to
Th. 4, com plete with respect to undefined terms.
L et us now consider how some part o f theoretical physics,
e.g. m echanics, could be deductively built up on this basis.
Tw o different procedures m ay be considered here. In the first
m ethod we attem pt to define the concepts o f mechanics b y means
o f geom etrical concepts. I f this were successful, then mechanics
w ould, from the m ethodological standpoint, sim ply becom e a
special chapter o f geom etry. I t is easily seen that all attem pts in
this direction m ust be confronted with quite essential difficulties.
818 METHODOLOGICAL INVESTIGATIONS ON X,}8

The specific oonoepts o f mechanics are of a manifold logical

character. They are, for example, properties o f individual
space-time points, or properties of sets o f space-time points
(like the ooncept o f the rigid body), or single-valued functions
which correlate numbers with space-time points (e.g. the
concept o f density). We should be in a position to define
such a concept, e.g. a Amotion, by means o f geometrical con­
cepts, only if—loosely speaking—the behaviour o f the function
over the whole world were known to us, and if we were able
to derive every Amotion value from the mere position o f the
corresponding point in space and time. We have good reasons
for regarding this task as unfeasible from the start. We turn
therefore to the seoond procedure. In this method some speoifio
concepts of mechanics are taken as primitive concepts and some
o f its laws as axioms. Mechanics would then assume the charac­
ter o f an independent deductive theory for which geometry
would be only the foundation. Although the axiom system of
mechanics would be incomplete initially, we could extend and
complete it in accordance with the progress o f empirical science.
If, however, W e wish to apply to mechanics the same criteria
as to every other deductive theory, then we should regard the
problem o f setting up deductive foundations for mechanics as
exhausted only if we obtained a oategorical axiom system for
this theory. But at this stage—and this is the most important
point—Th. 4 would come into action. Since we have adopted
a strictly categorical system o f geometry as a basis, mechanics
cannot be essentially richer than geometry with respect to
specific concepts. The concepts o f mechanics would have to be
definable by means o f geometrical oonoepts, end thus we are
brought back to the first method. The problem of defining the
concepts of mechanics (or any other physical theory) exclusively
by means of geometrical concepts, and the problem of basing
mechanics on a categorical axiom system, are completely equiva­
lent, whoever regards the first problem as hopeless must take
an entirely analogous attitude to the seoond.
The above remarks ju st m ade d o n ot apply to certain new
physical theories in which the basio system o f geom etry is
X, 8 2 TH E D E F IN A B IL IT Y OF CONCEPTS »1«

neither strictly oategorical nor categorical in the usual sense

and in whioh the whole o f physics or som e o f its parts are con ­
ceived ju st as special chapters o f geom etry. N aturally the diffi­
culties do not here vanish, only the centre o f gravity is shifted:
it lies m the problem o f providing the basic geom etry w ith a
categorical axiom system.
 XI

ON T H E F O U N D A T IO N S OF
BOOLEAN ALGEBRAf

also called the algebra of logic, is a form al

B o olean a lg e b r a ,
system w ith a series o f important; interpretations in various
fundam ental departm ents o f logic and m athem atics, The m ost
im portant and best known interpretation is the calculus of classes*
The present article contains a* study o f some alternative
system s o f Boolean algebra. O f the results I have obtained con­
cerning these problem s I shall, give here only those which can
be established with the help o f the usual mathema tical methods
and dem and no special metamathernatieal apparatus.

§ 1. T h e O r d in a r y and the E xtended (o r Co m p l e t e )

System s of B o o lean A lg ebr a

At present several equivalent systems of primitive expres­

sions and postulates are known which suffice for the setting up of
Boolean algebra. In order to make the subsequent developments
more specific I first explicitly introduce one of these systems.
In this system eight primitive expressions occur: ‘ B*— lihe
universe of discourse\ ‘x < y ’— 'the element x is included in
the element y\ (x = y ’— cthe eiernad x stands in the relation of
equality to the element y ’ {(the element x is equal to the element
y li ‘# + 2/ ’— dhe 3um of the elements x and y\ ‘-x.y’— 'the product
of the elements x and y\ *0’— Hhe zero element’ : ‘ V— 'the unit
element9and V * — Hhe complement of the element x\ The symbol
CB ’ denotes a certain set of elements, the symbols 'O' and T
certain elements of this set, the sym bols *< J and ‘ — ’ certain
relations between the elements o f this set; finally, the symbols

t Bibliographical Noth. This article first appeared in German under the

title cZur Grundlegung der Booleschen Algebra. I ’, in Fundamenta Matte-
maticae, voi. 24 (1935), pp, 177-88. (The unpublished Part II is referred to in
the second footnote on p. 341.)
X I, } I BOOLEAN ALGEBRA 321

* + \ *. *, and denote certain operations w hich correlate, w ith

one or tw o elem ents o f this set, another elem ent o f this set .1
The follow ing sentences are taken as postulates:

P o st u la te 3lx. (a) I f x e B , then x < x\

(b) if x , y , z e B, x < y, and y < z, then x < z.
P o st u la te 31*. I f x , y e B, then x = y if and only if both x < y
and y < x.
P o st u la te 31*. I f x, y e B, then
(a ) x + y e B ;
(b) x < x + y and y < x + y ;
(c) if, moreover, g e B, x < z, and y < z, then x + y < g.
P o st u la te 3I4. I f x, y e B, then
(а) x . y e B ;
(б ) x . y < x and x . y < y;
(c) if, moreover, z e B, z < x, and % < y , then z < x.y.

P o st u la te 3I5. I f x , y , z e B, then
(a) x . ( y + z ) = x . y + z . z ;
(b) x + y . z — { x + y ) . ( x + z ) .
B;
P o stu la te 3t«. (a) 0 , 1 e
(b) i f x e B , then 0 < x and x < 1.
P o st u la te I f x e B, then
(a) * ' 6 B ;
(b) x .x ' — 0 ;
(c) x + x ' = 1.

1 In addition to the specific symbols of Boolean algebra the usual symbols

belonging to set theory will also be used. For example, the formula
*x, y,... G X * means that the elements x, y,... belong to the set X . The formula
‘I c T will assert that the set X is included in the set Y ; the symbol
*2Z?[^(a:)]’ denotes the set of all these elements x which satisfy the condition
«
(sentential function) tjt(x); the symbol '{x, y ,...}’ denotes the set which consists
only of the elements » , t/,..., etc. I f it is desired to express that the relation R
holds between x and y, then *zRy * will be written. I f the denial of this is to be
expressed, we shall write 'x non-Ry\
322 ON TH E F O U N D A T IO N S OF X I, § 1
The tota lity o f the Posts. and the theorem s which
follow from them we call the ordinary system of Boolean algebra.
B y introducing tw o operations o f an infinite character, one
o f addition and one o f m ultiplication, this system o f Boolean
algebra qan be extended. F or this purpose we add the follow ­
ing expressions to the list o f prim itive concepts : ‘ T y *— ‘the
yeX
mm of all elements of the set X 9 and * J J y ’— ‘the product of all

elements of the set X \ 2 The system o f postulates is correspond­

ingly supplem ented b y the follow ing sentences:

P o stu la te 2J8. I f X ^ B , then

(a) y y e B\
yeX
(b) x < Y y for every x e X ;
vex
(e) if, moreover, z e B and x < zfor every x e X , then Y y < z.
yeX

P o st u la te iff,. I f X £ B, then
(« ) I l j / e S ;
yeX
&) T i v < x for every x e X ;
VeX
(c) if, moreover, z e B and z < x for every x e X , then z < JT y.
vex
1 W ith slight modifications this is the system of postulates given by L.
Couturat in (14a). Regarding the changes which we have introduced into this
system the following may be said: (1) the expressions 'x =* y \ ‘ x + y 1, 'x.y\
‘ O’, *1’, and *x'* which are defined expressions for Couturat are treated as
primitive, the postulates being correspondingly modified; (2) the symbol lB\
which denotes the universe of discourse, is included in the list of primitive
expressions; (3) the postulate of existence; ‘ 0 ^ 1 * (op, cit., p. 28) is excluded
from the system. In Couturat’s book the reader will learn how all theorems
of Boolean algebra which are used in the further considerations are derived
from the adopted postulate system.
* In addition to the expressions ‘ V j/ * and * fT y ’ symbolic expressions of a
vex ysX
more general nature will be used, * 2 f(x) * and 4 f ] f(x ) * they will denote the
Mat) 0(»)
sum and the product of all elements y of the form y = /(# ), which correspond to
elements satisfying the condition These expressions can easily be defined
by means of the primitive expressions: Let X be the set of all elements y of:
the form: y = f(x)t where x is any element which satisfies the condition
then the formulas hold: 2 /(a ) = 2 V n / ( » ) = u v - Those more general
Mx) yeX Mx) yeX
expressions appear in Post. 2t10.
X I, I 1 BOOLEAN ALGEBRA 323

P o stu la te 3l10. I f x e B and X £ B, then

(a) x . T y — y ( x . y ) ;
yeX yeX

(b) x + n y = IT (* + y )-
yeX yex

The postulate system together with the totality o f

the theorem s which can be derived from it, form s the extended
(or complete) system of Boolean algebra.
I t must be em phasized that the postulates o f this system are
n ot independent o f one another. Thus, for exam ple, any three
o f the four postulates 31®, 3t}0, 9ti0 can be derived from the
remaining postulates o f the system .f M oreover, the prim itive
expressions o f the system are n ot independent. F or exam ple, the
inclusion sign ‘ < ’ can be defined in various ways b y means o f
the other expressions. This is seen from the follow ing theorem
which is easily proved w ith the help o f Posts.
I f x, y e B , then the following formulas are equivalent:
(1) x < y\ (2) x + y = y; (3) x .y = x;
(4) x'+ y = 1 ; (5) x . y ’ = 0.
If, how ever, the sym bol *<* is retained in the list o f prim itive
expressions, then the identity sign can be deleted from it,
because P ost. Slg can be adopted as its definition.
The sym bol ‘ = ’ is frequently treated n ot as a specific sym bol
o f B oolean algebra, bu t as the sign o f logical identity. From now
on we shall adopt this standpoint. I t results in im portant
sim plifications in the system o f prim itive expressions and postu­
lates o f B oolean algebra. I t is especially helpful in reducing the
list o f prim itive expressions. In addition to ‘B\ it suffices to

t This result can be improved. In deriving any three of the postulates

listed above from the remaining postulates of the system we can avoid the use
of $ 8 and $ 9 bo that the derivation of and can essentially be carried out
within the ordinary, and not extended, system of Boolean algebra. To this
end, however, we must supplement the system [3b—$ 7] with appropriate
definitions of ‘ 2 * and ‘ II*. Moreover, Posts. 3If0 and $J0 must be put in a
somewhat weaker form, and in fact provided with certain existential assump­
tions: $ 1 0 with the assumption that 2 y exists, and $ }0 with the assumption
veX
that JQ y exists. Cf. A . Tarski (77), p. 211; also B. J6nsson and A . Tarski,
(36c), part I , p, 905.
m ON T H E F O U N D A T IO N S OF X I, §1

include in this list one o f the following expressions: ‘x < y\

'x + y ’, ‘x.y \ ‘ T y \ or *JT y\ in order to be able to define all
yeX vex.
the remaining expressions. For each of the expressions just
listed a corresponding system of postulates can be constructed
whioh is equivalent to the system [^SCX—9l103 but is much simpler.1
As an example we here introduoe a simple postulate system
based on the two fundamental expressions ‘.B’ and (x < y ’, and
we prove its equivalence with the system The system
consists of four sentences:
P ostulate I f x, y e B, x < y, and y < x, then x = y.
Postulate 9J2. I f x, y, z e B, x < y, and y < z, then x < z.
P ostulate S 3. I f x , y e B and x non- < y, then there is an
dement z e B such that z < x, and z non- < y. and the formulas:
u e B, u <Z y, and u < z always imply u < v, for every v e B.

P ostulate 23*. I f X s B, then there exists an element x e B

which satisfies the following conditions: (1) y < x for every y e X;
(2) if z e B , z < x, and if, for every y e X , the formulas u e B ,
u < y, arid u < z always imply u < v for every v e B, then also
z < v for every v e B.
The content o f the last tw o postulates will be clear if it is
noticed that, on the basis o f the earlier postulate system the
condition ‘the formulas u e B, u < y, and u < z always imply
u < v for every v 6 B ’ is equivalent to the form ula ‘y. z — O’ .
Sim ilarly, the condition ‘z < v for every v e B ’ can be replaced
b y the equation ‘z — O’ .
The rem aining expressions o f B oolean algebra are defined in
the follow ing w ay:

Definition S 6. (a) For all x , y , z e B, z = x-\-y if and only

if z is the only element which satisfies theconditions:
(1) x < z and y < z ,
(2) for every u e B, i f x < u and y < it, then z < u ,

1 E. T. Huntington has constructed such postulate systems for the ordinary

system of Boolean algebra; of. Huntington, B. V. (32) and (34).
X I, { 1 BOOLEAN ALO EBRA 326

(b) for a ilx ,y ,z e B ,z = x .y if and only if z is the only element

which satisfies the conditions:
(1) z < x and z < y ,
(2) for every u e B, i f u < x and u < y, then u < z ;
(c) for every x e B, x — 0 i f and only i f x is the only dement
which satisfies the condition; x < v for every v e B ;
(d) for every x e B, x = 1 */ and only if x is the only dement
which satisfies the condition: v < x for every v e B ;
(e) for every x ,y e B ,y = x' i f and only if y is the only element
which satisfies the conditions;
(1) the formulas u e B , u < x, and u < y always imply u < v
for every v e B ;
(2) the formulas u e B , x < u, and y < u always imply v < u
for every v e B.
Definition S e. I f x e B and X £ B, then
(a) x — jfcV if and wdy i f x is the only element which satisfies
the conditions:
(1) y < x fo r every y e X ;
(2) if z e B and y < z for every y e X , then x < z\
(b) x — JT y if and only if x is the only dement which satisfies
the conditions;
(1) x < y for every y e X ;
(2) i f z e B and z < y for every y e X , then z < x .
T heorem 1. The postulate system w equivalent to the
system o f postulates and definitions [33^—95*].

Proof I. We shall show how the sentences ©x-23^ can be

derived from Posts.
S j trivially follows from Post. %t; 23a is identical with 91}.
We pass on to S 8. Let x, y e B and x non- < y. W e put
z = x .y '. W e see at once that z e B and z < x . I f we had
z < y , i.e. x .y ' < y , we should have (in view o f the formula
x .y < y) x . y + x . y ' < y, whence * .(y + y ') < y, x . l = y, and
x < y , contrary to the hypothesis; hence z n on-< y. Finally if
326 ON TH E F O U N D A T IO N S OF XI, § 1

tie B, u < y> and u < 2, then u < y.z = y.(x.y'); from this
by easy transformations it follows that

u < x.(y.y') = x.O = 0,

and since 0. < v for every v e B, so also u < v for every v e B.
Thus the element z satisfies all conditions o f Post. 53s.
In order to derive 934 we take any s e t l g S and put x = T y.
v£k
According to 9lg*& we have x e B and y < x for every
y e X . Further, we consider any element z e B which satisfies
the premisses o f condition (2) of Post. 934, i.e. (a) z < x 9 and
(jS) for every y e X the formulas u e B , u < y, and u < z always
imply u < v for every v e B. I f in (/?) *u’ is replaced by *z.y\
then it follows that z.y < v for all y e X and v e B whence, by
2 (Z-V) < v> and since, by 9t£0, z.x = z. 2 2/ = 2
we have further z . x < v for every 2? e 5 . On the other hand
(a) gives z = z . x ; thus finally z < v for every v e B. The ele­
ment x consequently satisfies all conditions of the Post. ©4.
S f is immediately derivable from 9I3; we use 9lg order
to prove that x + y is the only element which satisfies conditions
( 1 ) and (2 ) o f Def. 93®. In a similar way 93^ is derived from
?I4, Sg'* from 9I6; 936 from 9l8 and 9V
There remains 23§. Let x , y e B and y = x'. I f u e B> u < 2 ,
and u < y, then u < x . y = x.x' = 0, and since 0 < v9 we also
have u < v for every v e B; hence y satisfies condition ( 1 ). In
a similar way it is established that y satisfies condition (2 ).
I f any other element yx e B likewise satisfies these two con­
ditions, it is easy to prove that we have x .yx = 0 and
x + y x = 1; with the help of the usual methods of Boolean
algebra, we obtain from these formulas yx = x r = y. Conse­
quently y is the only element which satisfies conditions ( 1 ) and
(2 ) . Since, according to 91?, there corresponds an element x' e B
to every element x e B, we easily conclude that, conversely, if
y is the only element which satisfies ( 1 ) and (2 ), then y = x'.
$85 is in this way derived.
We have thus shown that the sentences ©j—Sg actually follow
from the Posts. 9I1 -9I10, q.e.d.
X I, §1 BOOLEAN ALGEBRA 327

II, The p ro o f in the opposite direction likewise presents no

great difficulties.
91* is derived from S 3) b y substituting V for *y' in 33s and so
convincing ourselves that the assum ption: x non-< x leads to
a contradiction. coincides with © 2. is an equivalence,
thus a conjunction o f tw o im plications; one o f these im plica­
tions coincides w ith 23^ the other follow s directly from the
Post. 21* already derived (it m ust not be forgotten that the
sym bol * = ’ is always to be regarded as the sym bol o f logical
identity).
N ow we com e to P ost. 2Ig. L et X be any subset o f 2?. A ccord­
ing to 234 there is an elem ent x which satisfies the follow ing
conditions:
( 1) xeB ;
(2) y < x for every y e X ;
(3 ) if z e By z < x and if, for every y e X , the formulas u e B,
u < y , and u < z always imply u < v for every v e B, then
z < v for every v e B.
W e shall prove that the elem ent x also satisfies the follow ing
condition:
(4) if z e B and if y < z for every y e X , then x < z.
In fa ct let us suppose that this is n ot the case; let z be an
elem ent o f the set B such that

( 5) y < z for every y e X ,

and that nevertheless x n on -< z. B y P ost. 233 ( V and V
being substituted for ‘y* and V respectively) there is an element
i w hich satisfies the follow ing conditions:
( 6) t e B and t < x;
(7) tn o n -< z ;
( 8) the formulas u e B, u < z , and u < t always imply u < v
for every v e B.
L et us now consider any element y e X and suppose that the
form ulas u e B ,u < y, and u < t hold. B y form ula (5) we have
328 ON T H E F O U N D A T IO N S OF X I, $ I

y < z; b y virtue o f 58a the inclusions u < y and y < z give

u < z. W ith the help o f ( 8) w e infer from this that u < v for
every v e B. Consequently,
(9) the formulas u e B, u < y , and u < t always imply u < v,
for every y e X and v e B.
In view o f ( 6) and (9) we can apply (3), replacing V b y
w e thus obtain t < v for every v s B and in particular i < z,
w hich obviou sly contradicts form ula (7). W e m ust therefore
assume that condition (4) is satisfied.
A ccording to ( 2 ) and (4) the elem ent z s B satisfies both
conditions o f D ef. F rom 2?! it is easy to infer that x is
the on ly elem ent w hich satisfies both these conditions; hence
x = y_y. I f in ( 1), ( 2), and (4) we replace V b y ‘ 'w'e

obtain the com plete P ost. 3I8.

F rom D efs. 23? and 23? together it follow s that the
form ulas x — £ and x = y + z are equivalent in case the set

X £ B consists o f only the tw o elem ents y and z, X ~ {y,z};

P ost. 2I3 thus represents a special case o f P ost. 2IS. 9lg is similarly
d eriv ed ; we appiy 2Ta tw ice, it being assumed the first tim e that
X is the em pty set, and the second tim e that X •= 3 . F inally by
means o f an analogous m ethod we prove 2l 4and SI9. F or example,
to obtain 2IS we consider any set X £ B and apply 2igi n ot to
this set X , but to the set Y o f all those elem ents y e B which
satisfy the con dition : y < x for every x e X ; from D ef. 336 it
results that XT* ~
xeX
2 V- ^4 can be derived from ®9 as a special
i/ef*
case b y putting X = {y, z}.
B efore we proceed, it m ay be rem arked that on the basis of
the Posts. ^Kx~314 and 2^ already obtained, P ost. 353 can easily
be transform ed in the follow ing w ay:

( 10) if x } y e B and x non- < y , then there exists an element z e B

such that z < x, z ^ 0, and z .y = 0 .
The condition ( 2 ) o f P ost. 23* m ay be sim ilarly m odified: if
z e B, z < x and if z .y = 0 for every y e X , then 2 = 0 . More­
over, we rem ind ourselves that in establishing 2t8it was proved
X I, s l BOOLEAN ALG EBRA

that the elem ent % which satisfies the conclusion o f P ost. is

identical w ith Y j/. W e can thus state the follow ing:

( 11) %f X s B, z e B, z < Y y */ z .y = 0 for every y e X ,

veX
then 2 = 0 .
W ith the help o f ( 10) and ( 11) we prove 2t10w ithout difficulty.
L et x e B and X £ B. In view o f 2t$ we have z < S y for every
vex
z e X , whence b y 2L and ® 2 it is easy to obtain x.z <. z. V y
v«x
for every z e X, and thus b y 3t|

(12) Y > . y ) < *•

vex iS t
Should the opposite inclusion n ot hold, then b y ( 10) there
w ould be an elem ent z e B such that

(13) 2 < x. T y and z .y (x .y )~ 0;

vex v&
(14) 2 # 0.

A s a consequence o f P ost. 21$ we have x . y < T {x.y) for

every y e X , whence z.(x.y) < z. Y {x.y) and, in view o f (13),

z.(x.y) = (z.x).y = 0 for every y e B\ from the form ula (13) it

follow s that z . x < 2 y . H ence b y applying condition ( 11) we
obtain z.x = 0; but on account o f (13) z < x, and this im plies
z = z.x, so that at last we have z = 0 , in contradiction to
form ula (14).
In this w ay it is established that the inclusion ( 12) actually
holds in the other direction; hence, b y virtue o f P ost. 331, we
im m ediately derive 21J0.
As a special case o f 2If0 we obtain 2b?. From we can
im m ediately derive 2l£. On the other hand 2l £0 m ust be
proved w ith the help o f a m ethod analogous to that used for
P ost. 21J0. (The inclusion x + T lV < TT \x ^rV) follow s easily
yeX yeX
from 2l9, 2Is, and $ 2; to prove the opposite inclusion we em ploy
an indirect m ode o f argument m aking essential use o f ( 10) as
well as P ost. which has already been derived.)
330 ON TH E F O U N D A T IO N S OF X I, §1

9I7 still remains to be considered. On the basis o f the rem ain­

ing postulates we first transform D ef. in the follow ing w ay:
(15) if x ,y e B, then y = x r if and only if both x .y = 0 and
x + y = 1.
(It is superfluous to add here that y is the only elem ent which
satisfies both the equalities m entioned, for from Posts.
it can easily be concluded that tw o elements o f this kind must
be identical.)
Let us consider further any elem ent x e B. W e denote b y the
sym bol ‘X ’ the set o f all elem ents z e B such that x .z = 0 ,
X = E[z g B and x . z ~ 0] ? and finally put y = Y z . B y P ost.
z
3KL we have the form ula x .y = x* Y z = Y (x.z); and since
all terms o f the sum 2 (x .z) are equal to 0 , we also have
zeX
(x .z) = 0 , as can easily be seen from 3I§. Thus

(16) x . y =s 0 .
L et us assume then that x + y + 1 and that *consequently
1 non- < x + y . H ence it follow s b y ( 10) that there exists an
element z e B such that z ^ 0 and z + x + y ) = 0 , whence
z . x + z . y = 0 and z.x = 0 = z.y. I f z.x = 0, then z e X yfrom
which, b y virtue o f we have z < ]£ z = i/ and consequently
ssgX
z = z.y; this equality, together with the form ula z.y = 0 , gives
z = 0 , contrary to the inequality previously stated. W e must
thus assume that
(17) x + y = 1.
From (15)-(17) it is seen at once that y = x', whence %
follow s b y suitable substitution.
In this w ay all postulates o f the system [ 5tj-$I10] can b® derived
from the postulates and definitions © i-S V , the tw o systems are
thus finally seen to be equivalent, q.e.d.
The postulate system [® i- 234] can be form ally simplified:

The postulate system [3 3 * -© 4] is equivalent to the

T h e o b e m 2.
system which consists of © 2 an& the following -sentence:
X I, § 1 BOOLEAN ALGEBRA 331

P ostulate 33*. I f X £ J3, then there is exactly one element

x 6 B which satisfies the conditions ( 1) and ( 2) of Post. 334.
Proof. S 4 was im plicitly derived from the system [® 1- S 4] in
the p roof o f Th. 1 (Part II). In fact, it was shown there that
every elem ent x e B which satisfies the conditions ( 1) and (2 )
o f P ost. ® 4 also satisfies the form ula x = T y\ hence there can
v&
be only one such elem ent. I t remains to be proved that the
Posts. 23x and 33s can be derived from 932 and 33*.
In order to sim plify the exposition we shall define the auxil­
iary sym bol <)(> (the form ula *x)(y’ m ay be read ‘the elements x
and y are disjoint').
( 1) I f x y y e B, then x ) ( y if and only if the formulas u e JS,
u < a?, and u < y always imply u < vfor every v e B.
W e next prove som e lemmas.
( 2) I f x , y , z e B, x < y and y )( z , then x )( z.
This follow s directly from ( 1) and Post. 232.
(3) I f y e B and y )( y, then y < v for every v e B.
T o prove this we use 33*, putting X = {y}. I f we consider
( 1), we see that there is an elem ent x e B which satisfies the con­
ditions: (a) y < x, and (jS) if x e B, z < x and z){y, then z < v
for every v e B. I f in (£) V is replaced b y cy\ then, using (a),
we at once obtain (3).
(4) Ify> z e B ,y < z and y )(z, then y < vfor every v e B.
Consider any element u e B such that u < y \ b y applying
the premiss y < z, we obtain u < z with the help o f 332* B y
definition ( 1) it follow s from the form ulas y)(z, u < y, and
u < z that u < v for every v e B. Thus, if u e B and u < y,
then u < v for every v e B\ if we put in ( 1) ‘y ’ for V , it follow s
that y)(y and hence, according to (3), that y < v for every v e B,
which was to be proved.
( 6) I f x e B , then x < x.
As in the p ro o f o f (3) we put, X = {x} in 93*, and obtain
an elem ent x1 which satisfies the tw o conditions: (a) x < xl9
and (/?) if z e B, z < xx and z )( x, then z < v for every v e B.
882 ON T H E F O U N D A T IO N S OF X I, § I

Further, we consider the set X x = E[y e B and y < *]. W e have

v
(y )y < x for every y e X v and (S) i f z e B , z < x , and z)(y for
every y e X v then z < v for every v e B (for from the hypo­
thesis o f (8) it follow s that z )( z, whence, b y (3), z < v for every
v e B). I f w e now apply 93a, from (a) and (y) we obtain (<r) y < xx
for every y e X x. F inally the follow ing condition is satisfied:
(Qifg e B , z < xLand z )(y for every y e X ls then z < v for every
v e B. In fa ct if u e B, u < z, and u < x, then « e whence,
b y the hypothesis o f condition (£), z )( u, or w hat, according to ( 1),
am ounts to the same thing, u)(z; from (4) and the form ulas
tt < z and u)(z it follow s that u < v for every v e B\ b y ( 1)
we thus have z)(x; thus the elem ent z satisfies the hypothesis
o f condition (j8), and consequently z < v for every v e B, which
was to be proved. B y ( 1) it follow s from (y)-(£ ) that not only x
bu t also x1 satisfies both conditions o f the P ost. 234 as applied
to the set X v From this b y SB* we have x = xx, and this identity
together w ith (a) at once gives (5).
( 6) I f x, y e B, y < x and if for every z e B the formulae z < x
and z)(y imply z < y , then x = y.
In order to prove this we again put X = {y}. In view o f (5)
w e have (a) t < y for every t e X and, m oreover, b y virtue o f
(4), (/3) i f z e B. z < y, and z)(t for t e X , then z < v for every
v e B. On the other hand the hypothesis o f condition ( 6) im­
plies ( y ) t < x for every t e X, as well as (8) if z e B, z < x, and
z)(t for t e X , then z < v for every v e B [for i f z < x and z)(y,
then b y hypothesis we have z < y and hence b y (4), z < v for
every t* e B\. From ( 1) and (a)-(S ) it follow s that y as well as x
satisfies both conditions o f Post. SB4; b y SB* the elements x and
y are therefore identical, q.e.d.
N ow we turn to the p ro o f o f Posts. SBX and SB#.
S8 Xfollow s im m ediately from ( 6), for if x < y and y < x, then
b y SBa the form ula z < x always im plies z < y, the hypothesis
o f ( 6 ) is thus satisfied, and consequently x — y. I t is also not
difficult to derive S x from SB* directly w ith the help o f SBa, by
putting X — E[z e B and z < x] and arguing as in the proof
*
o f Lem m a 5.
X I, J1 BOOLEAN ALGEBRA 333

T o prove 33s we proceed indirectly: we assume that this postu­

late is n ot satisfied, thus that (a) x non-< y and that never­
theless, in view o f ( 1), (/3) for every z e B the form ulas z < x
and z)(y always im ply z < y . N ow 23* is applied to the set
X = we obtain an elem ent t e B such that (y) x < t,
(8) y < ty and (c) for every z e B, the form ulas z < t, z)(x and
z)(y im ply z < v for every v e B. I t is proved that the elem ent
t likewise satisfies the follow ing condition: (£) if z e B, z < t ,
and z)(y then z < y . In order to prove this we consider any
elem ent u € B such that u < z and u < x; according to ( 2) the
form ulas u < z and z){y always im ply u)(y; with the help o f
(jS), from the form ulas u < x and u)(y we obtain the inclusion
u < y> whence, b y virtue o f (4), u < v for every v e B. H ence
it follow s b y ( 1) that z){x, The elem ent z which satisfies the
hypothesis o f condition ( f ) thus also satisfies the hypothesis o f
condition U); consequently z < v for every v e B and in
particular z < y. I f now in ( 6) we replace V b y V and take
note o f (8) and (£), we obtain t ~ y, whence, b y (y), x < y. In
this w ay we com e to a contradiction with assum ption (a), and
hence we m ust assume that Post. 238 is satisfied.
I t is thus proved that the postulate system s [ S j- S J and
[©*» ©J] are actually equivalent .1
1 The formulation of Posts. 93* and 93J as well as some fragments of the
proofs of Ths. 1 and 2 have been influenced by the researches of 8. Le6niewski.
The extended system of Boolean algebra is closely related to the deductive
theory developed by S. Le&niewski and called by him mereology. The founda­
tions of mereology have been briefly discussed in article II, where bibliographi­
cal references to the relevant works of Le6niewski will also be found. The
relation of the part to the whole, which can be regarded as the only primitive
notion of mereology, is the correlate of Boolean-algebraic inclusion. The
postulate system [93a» S3f] has been obtained by a slight modification of the
postulate system for mereology (Posts. I and II) suggested in I I ; regarding
the relation of the latter system to the original postulate system of Le6niewski
see II, p. 25, footnote 2.
The formal difference between mereology and the extended system of
Boolean algebra reduces to one point: the axioms of mereology imply (under the
assumption of the existence of at least two different individuals) that there is
no individual corresponding to the Boolean-algebraic zero, i.e. an individual
which is a part of every other individual. If a set B of elements (together with
the relation of inclusion) constitutes a model of the extended system of
Boolean algebra, then, by removing the zero element from B> we obtain a model
for mereology; if, oonversely, a set C is a model for mereology, then, by adding
a new element to C and by postulating that this element is in the relation of
334 ON T H E FO U N D A T IO N S OF XI, §2

§ 2. T he A tom istic System of B oolean A lgebra

There are m any problem s which, although they are form ulated
exclusively in the term s o f B oolean algebra, yet they cannot be
resolved either in the ordinary or in the extended system o f this
theory. Tw o groups o f questions especially deserve attention:
those o f the first group are connected with the concept o f cardinal
num ber and are especially concerned w ith the cardinal number
o f the set B\ those o f the second group involve the concept o f
an atom or indecomposable element} W e shall deal m ainly with
the problem s o f the second group and proceed at once to define
the concept o f an atom , introducing a special sym bol, ‘ At\ to
denote the set o f all atom s.

D efin itio n (£. x e At ( 1) if and only if x e B and x ^ 0,

and ( 2), far every dement y e B, the formulas y < x and y ^ 0
imply y — x.
Various equivalent transform ations o f this definition are
know n:

T heorem 3. On the basis of Posts. 9Ii~9I10 and Def. (£ the

following conditions are equivalent for ail x e B\ ( 1) x e At; ( 2)
every element y e B satisfies one and only one of the two formulas:
x < y or x < y'\ ( 3) x ^ 0 and the formula x = y-\-z, for all
elements y , z e B, implies x = y or x = z; (4) x ^ 0 and the
formula x < y-\-z,for all elements y , z e B , implies x < y or x < z\
(5) for every set X ^ B the formula x = T y implies x e X \ { § ) for
v£k
every set X Q B i the formula x < S y implies the existence of an
element y e X such that x < y .
The p roof presents no difficulties.
inclusion to every element of <7, we obtain a model for the extended system of
Boolean algebra. Apart from these formal differences and similarities, it should
be emphasized that mereology, as it was conceived by its author, is not to be
regarded as a formal theory where primitive notions may admit many different
interpretations. Regarding Ths. 1 and 2 see Tarski, A. (78).
1 This concept is treated in Schrdder, E . (62) vol. 2, Part 1, pp. 318, 349.
There will be found a definition of the atom which is equivalent to the Def. (£
given here, as well as a proof of several properties of this concept, including
those formulated here in Th. 3. Instead of the expression ‘ atom' SchrSder
uses the term ‘ Individual* (in German).
X I, 5 2 BOOLEAN ALGEBRA 335

In connexion w ith D ef. (£ a whole series o f questions arises:

D o atom s exist at all? I f so, are atom s included in every
elem ent which belongs to the universe o f discourse and is dis­
tin ct from 0 ? Further, is every elem ent the sum o f the atom s
included in it? These questions can be neither affirmed nor
denied on the basis o f the previous assum ptions. B ut all the
above problem s are solved in a positive sense the m om ent we
enlarge the postulate system or b y adding the
follow ing sentence:
P ostulate D. I f x e B and x =£ 0 , then there is an dement
y e At such that y < x.
The tota lity o f postulates [ 9lx-?l10, $>] and the theorem s follow ­
ing from them we call the atomistic system of Boolean algebra.
The Post. ® , although seeming to be logically weak, yet brings
w ith it a whole series o f far-reaching consequences. In what
follow s we shall discuss som e o f these, showing at the same tim e
that they are equivalent to the new postulate. The first group
o f these consequences is form ed b y sentences which in one w ay
or another express the possibility o f representing any elem ent
as the sum o f atom s:
P ostulate 35r (a) 1 = ]£ y;
yeA t
( 6) if x e B, then x — £ y.
yeAtandy<x
P ostulate 2>2. Let x ,y e B\ then
(а) if for every element z e At the formula z < x implies z < y,
then x < y \
(б ) if the formulas z < x and z < y are equivalent for every
element z e A t , then x = y.
T heorem 4. On the basis of Posts. ^-^Ixo and D ef. (£, Post. D
is equivalent to each of the Posts. 35J, £)J, D?, and D^.1
Proof. First D® is derived from D. I f T y # 1, then
/ w velt
( 2,2/1 # 0 ; hence, according to P ost. © , there w ould be a
z e At such that z < ( T y) . B ut at the same tim e b y virtue
'V*At 7
1 Schroder in (62), vol. 2, part 1, pp. 337-8, expresses the conjecture that
Post. $>i cannot be derived from £>.
336 ON T H E F O U N D A T IO N S OF X I, §2

o f 2I§ we should have z < 2 V- From the condition (2) o f

/ veAi \
T h. 3 (where V is replaced b y V and cy ’ b y * 2 y ’ ) it follow s,
how ever, that these last tw o given inclusions contradict one
another. I t m ust thus be assum ed that 1 = Y y, q.e.d.
yejLt
If is now assumed, is proved in the follow ing way.
L et x e B; then b y virtue o f DJ and we have
x = x . l = x . % y = 2 (x.y).
yeA t yeA t

On the other hand we infer from the condition ( 2) o f Th. 3 that

for any y e At either y . x — y o r y . x — Q, according to whether
the inclusion y < x holds or n ot. Consequently the set
X = E[y < x and y e At]
v
differs from the set o f all elements o f the form y . x } with y e At,
at m ost b y the absence o f the null elem ent. H ence w ith the
help o f we easily conclude that

y < x and yeAt yeAt

and finally that
* = y y>
y < x and yeAt
as was to be shown.
F rom we obtain X>® b y the use o f P ost. Sljj in which
we put X = E\z e At and z < * ] and z — y. From by
0“
means o f we can at once derive
In order, finally, to prove D on the basis o f we proceed
as follow s: I f a* e 3 and x ^ 0 , then according to the form u­
las z < x and z < 0 cannot be equivalent for every zeAt\ since
b y D ef. £ we always have z # 0 and consequently z non~< 0
for every z e At, there m ust be an elem ent z e At such that
z < x, w hich was to be proved.
Thus Posts. are in fact equivalent.
B y Th. 1 we can replace in Th. 4 the postulate system [ 3^ -
9t10] b y the system o f postulates and definitions [2 3 i-S 6]- In
this connexion we note that in the system [ S i- 236] enlarged by
D ef. £ and P ost. X>g, Poet. 233 is superfluous, since it can easily
be derived from (the form ula ‘x ^ O’ in £ m ust however
X I, § 2 BOOLEAN ALG EBRA 337

first be replaced b y tbe condition 'there is an element z e B such

that x non-< z\ and sim ilarly for the form ula fy ^ O’), P ost. S *
can also be elim inated from the system if we strengthen
in the follow ing w ay:
I f z, y e B, then x < y if and only if for every
P o stu la te 35*.
z e At the formula z < z implies z < y .
The sentences 25lt 23d, and D* thus form a postulate system
which suffices for the establishm ent o f the atom istic system o f
B oolean algebra.
It should be pointed out that, on the basis o f the usual system
o f B oolean algebra, i.e. Posts. Th. 4 does not lose its
validity: the sentences £>-2)a remain equivalent (if the sym bol
*2* is understood in the sense o f D ef. 23g).f
From Ths. 3 and 4 it follow s that in the B oolean algebra
enriched b y P ost. 35 a certain m ethod o f reasoning can be used
w hich is fam iliar from the calculus o f classes. I f in the calculus
o f classes the inclusion ‘X £ Y\ or the equality ‘X = Y* is to be
proved, it is custom ary to show that every elem ent o f the set X
is also an elem ent o f the set Y, or that the tw o sets consist o f
the same elements* A ccording to postulates and £ § we
can proceed in a com pletely analogous maimer in the atom istic
system o f Boolean algebra b y operating with atom s and m aking
use o f those o f their properties that are established in conditions
( 2), (4), and ( 6) o f Th. 3.
B y means o f the m ethod o f reasoning just described tw o
sentences can be derived from 35 which are generalizations o f
Posts. and 3I10, these are the general distributive laws for
addition and multiplication:
P o stu la te 35g.Lei R be any class of subsets X of the set B
QjT I c i j ; and let £ be the class of all those sets Y which are in­

cluded in the sum of all sets of the class R and have at least one
f In the original o f this article it wae erroneousi3r stated that oja -he basis
o f the ordinary system o f Boolean algebra Poet. T> is essential! w etter than
Posts. Actually, however, the proof that ah these po^i lares are
equivalent can be carried oat along the same lines as the proof outL/u-u *j-bove
for Th. 4. The fact that Posts. 35?~35§ are derivable from Post. 3T also a
direct consequence of Th. 24 in article W ill, § 3.
338 ON THE FOUNDATIONS OF XI, § 2

dement in common with every set of the class Si (Y X ; if

X e Si, then X .Y ^ o ). Then:

(6)
The question now arises whether Post. 2 )3 can be derived
from Posts. w ithout the help o f 2 >. W e have shown in
collaboration with A . Lindenbaum that this is not possible:
2) not only im plies 2 >8, but is actually equivalent to it.
T h e o r e m 5. On the basis of Posts. 9li-9li0 ^nd Def. (£ Posts.
2 ). 2 >3, and 2 >§ are equivalent.
Proof. 2g is derived from 2) with the help of the method
discussed above: it is shown that for every u e At th6 formulas
(a ) u < TT 2 ^ and (/?) ^ < 2 IT 2 are equivalent. In fact, by

9I8 and 9IJ it follow s from (a) that u < T 2 for every l e f t .
ztk
Since ^ e A i, we infer, with the help o f Th. 5, that for every set
X e R there is a corresponding element z e X such that u < z.
Thus, if Ir contains all elements z which belong to at least one
set X e R and satisfy the inclusion u < z, then Y has elements
in com m on with every set X and therefore belongs to the class £.
On the other hand, b y 9^, the form ula u < J ^ z holds. Hence

with the help o f and we obtain the form ula (ft) without
difficulty. The p roof o f the im plication in the opposite direc­
tion is still easier (actually the inclusion T IT z < IT 2 2 can
Y e2 zeY XeSt z e X
be derived from Posts. 91i-9Ii0 alone). Since Post. 2)| follows
from 2), the equivalence of the formulas (a) and (jS) for every
u c: At leads directly to the equation 2g.
In turn we shall show how Post. 2 ) can be derived from 2 >g.
Let x be any elem ent o f the set B which is distinct from 0.
B y the sym bol ‘ 5V we denote the class o f all sets which consist
o f only tw o elements o f the form x .y and x.y\ where y is some
elem ent o f the set B. F or every set X e Si we have
]T z = x . y + x . y f = x . ( y + y f) = x . l = x,
X I, § 2 BOOLEAN ALGEBRA 339

whence we obtain at once = x. Next let us form for the

class St the class £ in the way prescribed by the hypothesis o f
Post. 35a. By 353 we have

whence 2 n z = * # °.
F€JC s e F

From this it is easily inferred that there exists a set F t 2 ,

such that JT z 0 , We consider any element y e B. Since
zg Y

X = {x . y } x.y'} e St, X has at least one element in common

with the set Y, i.e. either x . y e Y or x.y' e Y. Accordingly, by
3I9, either \\z <L x .y or JJ z < x.y' holds; hence
eeT zeY *
z< x n
zeY
and either J"J z < y or n * < y'. However, the last two in-
zeY zeY
elusions cannot both hold, since 2 , It is thus seen that
the element z satisfies the condition (2 ) of Th. 3, whence
n* e At. In this way we have derived 35, in fact for any given
zeY
element x ^ 0 we have constructed an element y e At such that
y < x. As regards Post. it is not difficult to prove that it is
equivalent to 35§ (for this purpose we can use the generalized
De Morgan laws, which can easily be derived from Posts. < &1-
5I10).. Thus the sentences 35, 2)g and 3 >| are finally seen to
be equivalent, as was to be proved.
It should be emphasized that the distributive laws are not
given the sharpest possible form in Post. 353 since the class £
contains in general many superfluous sets. Only in Post. 353 will
these laws be formulated in a completely adequate way.
P ostulate 35*. Let 51 be any class of non-empty subsets of the
set By and let g be the class of all those functions f which correlate
some element f ( X ) e X with every set X € St. Then:

(«> IT 2 v - 2 IT/OT;
X eH yeX /e ff

ib) 2 I l y = /teg
xea vex
11 XeZ
!/(*)•
340 ON TH E FO U N D A T IO N S OF XI, §2

Th. 5 remains valid if X)3 in it is replaced by but when

deriving from X), we have to make use of the axiom of
o f choice.f
We now formulate two sentences which are equivalent, not
only to Post. D (on the basis o f Posts. 21x-21xq or © x -S J , but
to the whole atomistic postulate system of Boolean algebra, i.e.
[ffix~®4> £>]• These sentences are quite different in character
from the postulates so far considered.
P ostulate (£x. There is a set E as well as a function F, which
satisfies the following conditions: (1) if x g B> then F(x) ^ E;
(2) if X £ E, then there exists exactly one element x e B such that
X = F{x); (3) if x , y e B, then x < y if and only if F(x) c F(y).
P ostulate CS2. There is a class R of sets and a function- F
which satisfy the following conditions: (1) if X e R} then also
y Y —X e R: (2) if £ c R} then also y Y a R; (3) if x e B t then
Tek F e£
F(x) e St] (4) if X a St, then there is exactly one element x g B
such that- X = F{x)] (5) if x, y e B , then x < y if and only if
F(x) s F(y).
T heobem 6. On the basis of Defs. 935, 23d; and (E, the postulate
system [® x- $34, X>] {or, what amounts to the same thing, f*2lx-2lx0*$ ])
is equivalent to each of the Posts. and- (S2.
Proof. In order to derive Post. from the system [©x~® 4 >
®] we put E - At and F(x) = E[y g At and y < x] for x g B.
v
The p roof that the set E and the function F determined by
these formulas satisfy the conditions (1)~(3) is based on Ths. 3,
f The result seated hi Th, 5 can be improved in the following wav. Instead
of we consider a somewhat weaker Post. ID*. The only difference
between. T)* and X* consists in the fact that in X)Athe formulae (a) and (b) of
are assumed to hold under certain existential hypotheses- For example,
formula (c) m provided with the hypothesis that the following sums and
products exist (ana heno9 are elements of B ): Y z for every X e Fi, Y z, and
zek Xeft teX
JT z for every Y G £?. 'vj and Xs* are, of course, equivalent within the extended
system of Boolean algebra. On the other hand, by somewhat modifying the
proof of Xh, or we can show that X), X)J, and are equivalent within the
ordinary ny?tom of Boolean algebra, i.e. on the basis o f Posts. 5Zj ~ % and
Dafs. 53* end £'. Theta remarks still hold if we consider, instead of Tq, Post,
obtained in an analogous manner from
X I, § 2 BOOLEAN ALG EBRA 341

5, and 6, and presents no difficulty. I f b y Si we understand the

class o f all subsets o f the set E, we at once obtain ©2 from (£*.
The derivation o f the Posts. 23-t-®4 and D from (g2 is somewhat
m ore com plicated. H owever, only Post. D causes some diffi­
culty. The problem reduces to showing that every set X e Si
distinct from 0 contains an atom ic subset, i.e. another set Y e Si,
likewise distinct from 0 , w hich includes no set o f the class Si
other than 0 and itself. F or this purpose we consider any
elem ent x e X and put Y = XT ^ with the help o f ( 1)
xeZ and ZeR
and ( 2) it is not difficult to prove that the set Y is just such an
atom ic set .1 The theorem is in this way proved.
The sense o f Th. 6 can be expressed in the follow ing w ay:
the atomistic system of Boolean algebra is isomorphic with the
ordinary calculus of classes. M oreover, in this theorem is con­
tained a rather interesting fact belonging to the theory o f sets:
every class Si o f sets which satisfies conditions ( 1) and ( 2) o f
Post. (£2 (i.e. is closed under the operations o f com plem entation
and unrestricted addition) is isom orphic with the class o f all
subsets o f a certain set E with respect to the relation o f in ­
clusion .2

1 Cf. Tarski (75), pp. 235 If.

8 This article was conceived as the first part- of a more comprehensive
paper. The second part (which was never published) included, among ether
things, a discussion of the so-called atoinless system of Boolean algebra, i e.
the system in which Post. 3) is replaced by a postulate stating that there are
no atoms. The atomless system of Boolean algebra can, of course, bo con­
structed as a system in which ‘ B * and 'x < y ' occur as the only primitive
expressions. A model for such a system is provided by the family of ail regular
open sets of a Euclidean space and the relation of set-theoretical inclusion. In
view of the remarks made above in p. 333, footnote, this follows immediately
from the discussion in II, in particular, from Th.
 XII
FOUNDATIONS OF TH E CALCULUS
OF SYSTEM S!

T he results here communicated belong to general metamathe­

matics (which I formerly called the general methodology of the
deductive sciences), and thus to a discipline whose task it is to
define the meaning of general metamathomatical concepts which
appear in the discussion of the most diverse deductive theories
and to establish the basic properties of these concepts. Only
those theories will be considered here, however, whose construc­
tion presupposes a logical basis of a greater or lesser extent,
and at least the whole sentential calculus.
In a previous com m unication (article I II o f the present
volum e) I have sketched a system o f foundations for the general
m etamathematics o f theories with the above characteristics.
Here in § 1 a m odification will be undertaken in this system
whereby, I believe, greater naturalness and sim plicity is reached.
In § 2 I sketch a calculus— 1 call it the calculus of deductive
systems, or sim ply the calculus of systems— which essentially
facilitates m etam athem atical investigations o f the kind con­
sidered here; at the same tim e an interesting analogy will be
apparent between this calculus and intuitionistic logic. In § 3
I extend the calculus o f systems b y introducing the concept o f
an axiom atizable system . In § 4 two further kinds o f deduc­
tive systems are considered, nam ely irreducible and complete
system s, and some rather deep theorems are stated which can
be obtained within the calculus o f systems and which charac­
terize certain classes o f deductive systems with respect to
cardinality and structure. In § 5 the general results will be
t B ib lio g r a ph ic a l N o te . This article originally appeared in two parts
under the title ‘ Grundzuge des SystemenkaJkiil, Erster Toil*, Fundamenta
Mathematical, vol. 25 (1935), pp. 603-26, and ‘ Grunciziigo des Systemen-
kalkiil, Zweiter Teil\ ibid., vol. 26 (1936), pp. 283-301. The first part com­
prised the Introduction and §§ 1-3, the second part §§ 4-6 and the Appendix.
Earlier publications of the author in the same directions are III and V ; ses
also the historical remarks at the end of the article.
X II CALCULUS OF SYSTEM S 343

illustrated b y applying them to some special, and in fact very

elem entary, deductive theories .1
P roofs will not be given here. W ith few exceptions they
are very sim ple.

§ 1. P r im it iv e C o n c e p t s a n b A x io m s
In the postulate system given in I I I four prim itive expres­
sions appeared: ‘S' — ‘the set of all meaningful sentences9, ‘ Cn(X)’
— ‘the set of consequences of the set X of sentences', ‘x '— ‘the nega­
tion of the sentence x\ ‘x y 1-— ‘the implication with the antecedent
x and the consequent y \2 W ith the help o f these terms, the
concept o f deductive system was there defined, i.e. the concept
o f a set o f sentences which contains as elements all its conse­
quences; the class o f all deductive systems was denoted b y the
sym bol ‘ S ’ . It was there rem arked that am ong deductive
system s a smallest exists, i.e. a system which is a sub-system
o f all other deductive system s. It is the system Cn(0), the set
o f consequences o f the em pty set. This system , which here for
brevity w ill be denoted b y ‘ L\ can be interpreted as the set o f
all logically valid sentences (or, m ore generally, as the set o f all
those sentences which from the start we recognize as true when
undertaking the construction o f the deductive theory that is
the object o f our m etam athem atical investigation ).3
I t now appears that the postulate system m entioned takes on a

1 In order to avoid misunderstandings it is important to note that I use the

expressions ‘ deductive theory’ and ‘ deductive system’ in quite distinct
senses. By deductive theories I understand here the models (realizations) of
the axiom system which is given in § 1. Since in this system four primitive
concepts appear, every quadruple of concepts which satisfies all the axioms
of the system is its model. In order not to depart too much from the usual
meaning of the term ‘ deductive theory’, I have in mind only those models
of the axiom system which are constituted by certain sets of expressions and
operations on expressions. On the other hand, deductive systems (in the
domain of a particular deductive theory) are certain special sets of expressions
which I shall characterize precisely at the beginning of § 1 as well as in Def. 5
of § 2 .
2 The symbolism employed here differs slightly from that of I I I ; in fact *x ’
is used instead of ‘ n(a?)’, and i / ’ instead of *c(a?, y)\ Otherwise, just as
in IH , I employ the customary set-theoretical notation; in particular I use
formulas of the type *x, y , . . . e X * as abbreviations of the expressions 'the
elements x, y ,... belong to the set X \
3 Cf. I l l , p. 33 and V , Th. 9 (6), p. 70, of the present work.
344 F O U N D A T IO N S OF TH E x r i, § i

sim pler and, as it seems to me, also m ore natural form if, instead
o f *Cn(X)\ the sym bol *L9 is included in the list o f prim itive
expressions. F or the foundation o f general m etam athem atics
(to the extent to which this task was attem pted in I II ) the
follow ing five axiom s then suffice:

A xiom 1. 0 < 8 < X0.

A xiom 2. I f x , y e S , then x e 8 and x - > y e S.
A xiom 3. L £ S.
A x i o m 4. I f x , y , z e S , then (x -> x) -> x s L, x -> (x -> y) e L
and ( x - > y ) - > ((y z) -> (x > z)) e L.
A xiom 5. I f x, x ->y e L (where y e S), then y e L.

The content o f these axiom s is so clear that it scarcely requires

an explanation. From the intuitive standpoint they m ay never­
theless adm it o f certain doubts which I shall not here discuss.1
The A xs. 2, 4, and 5 are adapted from the system o f prim i­
tive expressions, axiom s and rules o f inference o f Lukasiewicz,
which form s a sufficient basis for the developm ent o f the sen­
tential calculus .2* It is clear that we could equally well use any
other form al system o f the sentential calculus as a starting
point. In particular, if we took N icod ’s sentential calculus as
our m odel, then the num ber o f the prim itive expressions appear­
ing in the axiom s w ould be smaller, and A xs. 2 and 4 w ould be
som ewhat sim pler .8
W e will show how , w ith the help o f the prim itive concepts
adopted, the concept o f consequence can be defined. F or this
purpose we put in the first place:
D efinition 1. x~\~y = x -> y, x .y = x -^ y for aU x , y e 8.
The concepts o f the sum and the product o f sentences can be
extended to an arbitrary finite num ber o f summands and factors
b y recursion:

1 Cf. 3H, p. 31, note 3 ; I X , p. 282, note 2 ; and VUE, footnote, p. 166.
2 Cf. IV , pp. 39 and 4 3 ; III, p. 33, the remarks in connexion with Th. 3*.
* Cf. Nicod, J*. (55).
X II, s 1 CALCULU S OP SYSTEM S 345

D e f in it io n 2. if n = 1 and xx e 8 ;
n —1 n n—1
: JT = J I xi-%n> if n i-s an arbitrary
natural number > 1 and xv x2,..., xn e S.
The definition o f the concept o f consequence runs as follow s:
D efin itio n 3. For an arbitrary set X ^ 8 the set Cn(X) con­
sists of those, and only those, sentences y e 8, which satisfy the
following condition: either y e L >or there exist sentences

such that ( j j j X i\ -*y eL .

This definition can be transform ed in the follow ing w ay:

T heorem 1. For an arbitrary set X £ 8 the set Cn(X) is the
intersection of all sets Y which satisfy the following two conditions:
( 1) L + X g 7 ; ( 2 ) if x, x -> y e Y (where y e 8), then y e Y .
I t is easy to show that the set Cn(X) is one o f the sets Y which
satisfy the conditions ( 1) and (2 ) o f Th. X; it is thus the smallest
o f these sets. Since the operation which correlates the sentence
y w ith tw o given sentences x and x - > y is com m only called the
operation o f detachment, we can, on the basis o f Th. 1, charac­
terize the set Cn(X) as the smallest set which includes the sets
L and X and is closed under the operation o f detachm ent.f
I t m ight appear that the concept o f consequence here in tro­
duced is essentially narrower than that which is generally used
in the construction o f deductive th eories; in the definition o f this
concept we have made use o f only one o f the rules o f inference,
the rule o f detachm ent, as though we had totally forgotten the
existence o f other rules o f this kind, e.g. the rule o f substitution
and the rules o f universal quantification .1 A ll this, how ever,
depends essentially on the interpretation o f the sym bol CL\
1 Cf. for instance VUE, p. 180.

t In connexion with Def. 3, cf. the papers of K . Ajdukiewicz, (2) and (4).
Keeping in mind the meaning of Gn(X) as determined by Def. 3 we recognize
in Th. 1 one of the formulations of the so-called deduction theorem. In fact
this is just the form in which the deduction theorem was first established by
the author in 1921; see HE, p. 32, footnote
346 F O U N D A T IO N S OF TH E XII, § 1

I f the system o f logic on which we base deductive theories is

so constituted that, whenever a given sentence y can be obtained
with the help o f any o f the rules o f inference from another
sentence x . or from other sentences xv ..., xni then the im plica­

tion, x y, or ( JT ^i) V> belongs to the set L o f all logically

valid sentences (actually all known and sufficiently developed
system s o f logic satisfy this condition), then the concept o f con­
sequence is in no w ay narrowed. In fact, the sentence y will
under these conditions be a consequence o f every set o f sentences
which contains the sentence x or the sentences xv >.., xn, not
only in the usual sense, but also in the sense o f D ef. 3 or o f Th. 1 .
W e can thus say that the rule o f detachm ent (even in the special
form in which it appears in D ef. 3) suffices as the only rule o f
inference in constructing deductive theories, provided that these
theories stand on a sufficiently rich logical basis.
The follow ing tw o theorem s can now be proved:
T h eo r em 2. I f X ^ S, then X s Cn(X) c S ;
(a)
(6) i f X ^ S , then €n(Cn{X)) = Cn(X)\
(c) if I c S , then Cn(X) = T Cn{7)\
T Q X a n d T < K«
(d) there exists a sentence x e 8 such that Cn({x}) = 8 ;
(e) if X £ Sy y } z e 8 and y - > z e Cn(X), then z e Cn{X-\-{y})\
( / ) if X £ S, y , z e S and z e C n (X + {y }), then y - > z e Cn(X)\
(g) if x e S, then C?i({x, 5}) = 8 ;
(h) if x e S, then Cn({x}) . Cn({x}) = Gn(0).
T heorem 3. L = Cn(0).
Th. 2 shows that from the A xs. 1-5 given above and from
D ef. 3 all A xs. 1- 10* which were proposed in I II can be derived
(we have neglected the A xs. 1 and 6* o f I II in the form ulation
o f Th. 2, because the first is contained in the new A x. 1, and the
second coincides w ith the new A x. 2 ). It is easy to show that the
converse o f this result also holds: if we regard Th. 3 as the defini­
tion o f the sym bol ‘ L\ then we can derive from this definition
and the axiom system o f I I I the new A xs. 1-5 and D ef. 3 as
theorems. F inally we can thus assert that the tw o systems of
axiom s and prim itive notions are equivalent.
XII, § 2 CALCULUS OF SYSTEM S 347

§ 2 . T h e Ca l c u l u s o f Sy s t e m s
On the basis outlined in the previous paragraph two calculi.
can be constructed which are verv useful in metamathematiea!
%!

investigations, namely the calculus of sentences and the calculus

of deductive systems] the first is a complete and the second a-
partial interpretation of the formal system which is usually
called the algebra of logic or Boolean algebra.
In order to make the further considerations more specific we
shall here formulate in extenso a well-known system of postu­
lates which suffices for the construction of the algebra of logic.
In this system eight primitive concepts appear; ‘B *— 'the uni­
verse of discourse', < y'— cthe element x stands in the relation
of inclusion to the element y\ ‘x = y*— 6the element x stands in
the relation of equality to the element y\ ' x d y ' — the sum of the
elements x and* y\ ‘x .y '— ‘the product of the elements x and y\
‘0’— ‘the zero element* (‘the empty element'), M'— 'the unit element*
( ‘the universal element'), ‘x'— cthe complement of the element x\
The system contains seven postulates:
P o stu lateI. (a) I f x e B, then x < x: (b) ifx , y, z e B ,x < y
and y < z, then x < z.
II. I f x ,y
P o stu late e B, then x = y if and only if both
x < y and y < x.
P o stu lateIII. I f x ,y e B, then (a) x-\~y e B ; (6) x < x + y
and y < #+ ?/; (c) if z e B, x < z and y < z, then x 4~y < z.
P o stu late IV. I f x, y e B, then (a) x .y e B ; (6) x .y < x and
x .y < y\ (c) *4Z € B ,z < x and z < y : then z < x.y.
P o stu late V. I f x, y, z e B, then (a) x . ( y + z ) = x.y-\-x.z
and (b) x+(y, z) •-= (%-j-y).(x~f-z).
P o stu late VI. (a) 0, 1 e- JS; (6) if x e B, then 0 < x and
x < 1.
P o stu late VII. I f x e B, then (a) x e B, (b) x .x = 0, and
(c) x + x =s 1.
The system o f the algebra of logic can be extended by intro­
ducing two infinite operations: *T y *— ‘the sum of all dements
2/e i
348 F O U N D A T IO N S OF THE XII, §2

of the set X 9and 4JT y'— ‘the product of dll elements of the set X\
yeX
It is then necessary to add the follow ing postulates:
P ostulate VEIL I f X cr J5, then (a) T y e B and (b) x < T y
vSt yek
for every x s X ; (c) if moreover z e B, and x < z for every x e X ,

for every x e X\ (c) if moreover z e B, and z < x for every x e X ,

then z < JT y.
l/eiL
P ostulate X . I f x e B and X c B,then (a)x. T y = V (any)
_ veS: 2/eX
onrf (b) x + I i y = IT (* + y )-
j/eX yeX
Posts. I -V I I together with all theorems which can be derived
from them , form the ordinary, and Posts. I - X the contended (or
complete) system of the algebra of logicI
W e pass now to the calculus of sentences, which we shall only
describe briefly. In order to avoid term inological misunder­
standing we shall call it the sentential algorithm instead o f simply
sentential calcu lu s; the term ‘sentential calculus* already has a
definite meaning, other than that intended here.
The set S form s the universe o f discourse o f the sentential
algorithm . Between the elements o f this set we shall define two
relations: ‘a n y * ( ‘a? implies y*) and ‘x ~ y* x is equivalent
to */’ ).
D e f in it io n 4 .(a) x D y if and’ only if x ,y e S and x - > y eL\
(b) x == y if and> only if both xzyy and y o x.
The follow ing can now be proved:
T heorem 4. I f in the ordinary system of the algebra of logic- we
replace in all the postulates the symbols lB\ ‘ < \ and *= * respec­
tively by (S\ (d\ and ( ==’ ; if further ‘O’ is replaced by the variable-
*u* and ‘ 1 9by the variable V , making everywhere in the correspond­
ing postulates the assumptions that u e S and u e L as well as
v e L; and if the remaining symbols are left unchanged) then Posts.
I-V T I are satisfied.
1 Of. XI, pp. 320-33.
X II, § 2 CALCULUS OF SYSTEM S 349

It is to be noted in connexion with this theorem that the

sym bols ‘O’ and ‘ T o f the algebra o f logic cannot be ‘effectively’
interpreted on our foundations, for we cannot define a single
constant which designates an individual sentence .1
Th, 4 shows that all postulates o f the ordinary system o f the
algebra o f logic can be derived from A xs. 1-5 proposed in § 1
on the basis o f suitably chosen definitions (Defs, 1 and 4). It is
also easily seen that A x. 1 plays no part in the p roof o f this
theorem . This result can be reversed: from Posts. I -V I I o f the
algebra o f logic (when the replacem ents indicated in Th. 4 are
carried out) one can derive A xs. 2 -5 provided that *x -> y* is
suitably defined, for exam ple b y putting x -> y = x + y . W e
can thus assert that A xs. 2 -5 form a system o f statements which
is equivalent to the system o f postulates for the ordinary algebra
o f logic. (The problem o f the equivalence o f the system s o f
prim itive concepts which appear in the tw o system s o f state­
m ents w ould require a m ore exhaustive discussion.2)
Since we have chosen as foundation for this w ork the system
o f axiom s o f § 1 we have restricted ourselves in essence to

1 The interpretation of the algebra of logic in sentential algorithm can

clearly be modified so as to avoid the replacement of logical identity by another
equivalence relation. To obtain this modification we consider, instead, of
sentences x e S> the equivalence classes X c S each of which consists of all
sentences y which are equivalent, in the sense of Def. 4 6, to some sentence
x (y ss x). For these equivalence classes we define in an appropriate way the
relations and operations D, -f» •# etc.; for instance, the formula ‘ I d T* will
express the fact that x d y (in the sense of Def. 4 a), for all x e X and y e T.
The elements 0 and 1 are now interpreted *effectively *— in fact 0 as the set
of all sentences x e S such that x £ L, and 1 simply as the set L.
1 As I have already mentioned in § 1, any system of axioms and rules of
inference which suffices for the formal construction of the sentential calculus
can be used as a model in formulating an axiom system for the domain in which
we are interested in this paper. I f we compare this remark with the facts just
discussed we reach the conviction that there exists a general method which
enables us to obtain a system of postulates for the algebra of logic from every
system of axioms and rules of inference for the sentential calculus. It would
be superfluous to describe here more exactly in what this method consists,
especially as it has been developed in recent years in the articles of Bernstein
and Huntington; cf. Bernstein, B. A . (0), Huntington, E. V . (34), (35). I take
this opportunity of mentioning that the system of Axs. 1-5 of § 1 and its
connexion with the algebra of logic were already known to me in the year 1930.
(I mention this system implicitly in article V of the present work, p. 62,
as the basis upon which the unpublished second part of V has been con­
structed.)
350 F O U N D A T IO N S OF THE XII, § 2

characterizing the set 8 as a non-em pty, at m ost denumerable,

set, whose elements w ith respect to the operations defined on
them satisfy the postulates o f the algebra o f logic. Thus, from
a purely form al point o f view , the investigations which rest on
this foundation do not transcend the boundary o f the algebra
o f logic, but they differ from the discussions previously carried
out in that field b y the fa ct that they concern concepts which
have not been dealt w ith hitherto. It is not the elements o f
the system S itself, the relations holding between them, nor
the operations perform ed on them , which form the chief subject
o f the investigation, but certain selected sets o f these elements,
nam ely deductive system s .1
L et us rem ind ourselves o f the definition o f deductive system:
D e f in it io n 5. X e S if and only if Cn(X) £ X £ S.
W ith the help o f the prim itive concepts whioh we have
adopted, one m ay also characterize this concept as follow s:
T heorem 5. X e ® holds if and only if L c X £ 8 holds and
if the formulas x 9x - > y e X and y e 8 always imply y e X .
The caleulus o f deductive system s, w hich we shall call for short
the calculus of systems, represents (as will be seen below) a very
essential extension o f the sentential algorithm . The correlates
o f all prim itive concepts o f the algebra o f logic occur in the
calculus o f system s. The class S form s the universe o f discourse.
L is regarded as the zero system , 8 as the unit system . In­
clusion and equality between systems as well as the product
(the intersection) o f systems preserve their usual sense as laid
dow n in the calculus o f classes (which, as is well known, presents
one o f the com m onest interpretations o f the algebra o f logic);
equality thus coincides w ith logical identity. On the other hand
the addition o f system s, which we call logical addition and denote
b y the sym bol does n ot coincide with set-theoretical addi-
1 In fact the calculus of deductive systems outlined in this paper proves to
coincide with what was somewhat later developed as the calculus of Boolean*
algebraic ideals; in particular axiomatizable deductive systems (discussed in
§ 3) coincide with principal ideals, and complete deductive systems (§ 4) with
prime ideals. Compare the note of the author, (80). See also the papers of
Stone, M. H . (67) and (69).
xn, §2 CALCULUS OF SYSTEM S 351

tion, for this last operation when applied to system s, does not
in general yield a new deductive system ;1 the same holds for
the operation o f com plem entation.
The definition o f the logical sum o f the system s X and 7 is as
follow s:
D efin itio n 6. X + Y == G n (X + Y ) for any given I , Y s
The logical complement or the negation of the system X , sym boli­
cally X , we define as follow s:
D efin itio n 7. X = Cn({x}) for any given X e <3.
xeX
W e shall learn below (Th. 12) other equivalent form ulations
o f this definition.
On the basis o f the definitions just stated the follow ing
theorem can be proved:
T heorem 6. I f in all postulates of the ordinary system of the
algebra of logic we replace the variables V , izi by the variables
lX \ *Y\ ‘ Z ’ which denote deductive systems; if, further, we replace
the constants (B\ * < ’ , ‘ - f \ *Q\and ‘ 1 ’ respectively by the symbols
*L\ and *$’ ; then all postulates are satisfied with
the exception of Post. V ile , which fails ; but the following con-
sequence of Post. V II c preserves its validity:
P ostulate V H {d). I f X ,Y e S and X .Y = L, then f c l
The essential difference between the calculus o f systems and,
e.g., the calculus o f classes thus lies in the fact that, in the
calculus o f system s, instead o f the law o f excluded m iddle
X-\~X = S, on ly a certain weaker consequence o f this law holds.
As will be seen from Ths. 17 and 37 given below , the law o f
excluded m iddle holds in the case when the class 6 is finite (or,
what am ounts to the same thing, when the set S does not con­
tain infinitely m any sentences such that no tw o o f them are
equivalent). If, how ever, the class <5 is infinite— and we en­
counter this case as a rule when we consider special deductive
sciences— then there exist system s which do not satisfy this
law.
The breakdow n o f the law o f excluded m iddle has further
1 Cf. I l l , p. 33, Th, 7+; V, p.. 71, Th. 12 and the remarks added to it.
352 F O U N D A T IO N S OF TH E XII, § 2
consequences. The law o f double negation holds but only in
one direction, and only the law o f triple negation holds in both
directions:
T h eo r em 7, I f X e ®, them J c l and X = X .
O f the two B e Morgan laws only one is true, o f the four laws
o f contraposition tw o drop out and tw o remain valid:
T hsobem 8. I f X , 7 g S,
(a) = (6)
(c) the formula I c y implies Y £ X ;
(d)the formula I g F implies F c l ,
Ths. 7 and 8 can be derived from the system o f postulates
described in Th. 6 b y purely algebraic m ethods from the dom ain
o f the algebra o f logic. The same applies to other theorem s o f
the calculus o f system s which are known to us. provided they
do n ot have an existential character.
The form al resem blance o f the calculus o f system s to the
intuitionistic sentential calculus o f H ey ting is striking :1 we
m ight say that the form al relation o f the calculus o f system s to
the ordinary calculus o f classes is exactly the same as the rela­
tion o f H eytin g's sentential calculus to the ordinary sentential
calculus. In other words, if we construct the calculus o f classes
on the basis o f intuitionistic logic, then this calculus, it seems,
will not form ally differ from the calculus o f system s. W e can
also express this as follow s: the system o f postulates described
in Th. 6 form s (if we abstract from the specific sense o f the
term s appearing in these postulates) a sufficient basis for a
system o f the algebra o f logic which has the intuitionistic cal­
culus o f classes as one o f its interpretations. It is clear that
the last remarks w ould require a m ore exact and detailed
ela boration .!
The calculus o f system s can be extended b y the introduction
o f infinite operations. The product (the intersection) o f systems
o f a class R is understood exactly as in the calculus o f classes
1 Of. Heyting, A . (28).

t For such a more exact and detailed formulation see X and V II, in parti*
cular $ 5 ; of. also Stone, M. H . (70).
X II, §2 CALCULUS OF SYSTEM S 353

(the set S is regarded as the product o f the em pty class). On the

other hand the definition o f the logical sum of systems of the class
ft, sym bolically ^ F , form s the natural generalization o f D ef. 6 :
Fea

D e f in it io n 8. ^ 7 - Cn[ ^ Y j for an arbitrary class i t £ S.

Y*k Yek 1
In the case when the class it consists o f sets r „ r a, F3j... which
form a finite sequence with n terms, or an infinite sequence, we
write respectively: % + Y 2+...-\-Yn\ or % + Y 2+ ...+ Y n+ ...’
instead o f *V F \

T h e o r em 9. I f in all postulates of the extended system of the

algebra of logic— in addition to the changes described, in Tk. 6—we
everywhere replace the, variables ‘X ’ and ‘F ’ by the variables ‘ ft’
and denoting classes of systems, and the cxmstants ‘ V ’ and
‘I T by the constants and then in addition to Posts.
I -V I and V II a, b, also Posts. V III, I X , and X a are satisfied,
on th& other hand Post. X b together with Post. V II c fails .
A ccording to this theorem , finite m ultiplication is distributive
under infinite addition but the distributive law for finite addi­
tion under infinite m ultiplication no longer holds in the calculus
o f system s; a fortiori the general distributive laws for one o f
the infinite operations under the other fails .1 B y comparing
Ths. 17 and 24 given below we see that the distributive law o f
finite addition under infinite m ultiplication preserves its validity
in the same cases as does the law o f excluded middle.
From the postulates described in Ths. 6 and 9 other theorems
o f the calculus o f systems which involve infinite operations are
derivable, e.g. that one o f the tw o De Morgan laws which was
valid for finite operations:

T heorem 10. I f ft £ S , then ^ F — F.

Y zk Tepi

Theorems o f the calculus o f systems are also known which

1 These lews are not even valid in the extended system of the algebra of
logic; cf. article X I of the present work, § 2, especially pp. 337-9.
854 F O U N D A T IO N S OF TH E X II, §2

cannot be derived from the postulates described in Ths. 6 and 9;

as art example we give
Theorem 11. I f Si is an arbitrary class of deductive systems
such that then there is a finite subclass £ of Si such that
. Tek

Ts&
W e now turn our attention to the follow ing theorem :
T heorem 12. I f X e <5, then

and
(6) n r.
X + F -S
The formula (a) of this theorem is easily derivable from those
postulates o f the algebra o f logic which are satisfied in the
calculus o f systems, in particular from Posts. V II b, d; the
formula (6) cannot be derived in this way, and in its proof we
must directly refer to the adopted axioms and to the definitions
o f the symbols appearing in the formula.
Th. 12 shows how the concept o f the logical complement can
be characterized with the help o f other concepts belonging to the
calculus o f systems. We see that the logical complement o f the
system X , as in the ordinary calculus o f classes, is the sum of
all systems disjoint with X , i.e. those for which X .T = L holds,
and likewise it is the intersection o f all systems Y which logically
supplement X : Moreover, by virtue o f the law of
contradiction, X is the largest system which is logically disjoint
with X ; since, however, the law of the excluded middle fails, X
is in general not the smallest system whioh logically supple­
ments X (from Th. 12 it only follows that, if such a smallest
system exists at all then it coincides with X ).
§ 3. A xiomattzable Systems
Since the calculus o f systems deviates from the ordinary
caloulus o f classes, a series o f problems of the following kind
arises: one considers an arbitrary theorem o f the caloulus o f
classes which loses its general validity when carried over into
X II, $3 CALCULU S OF SYSTEM S 356

the oalcnlus o f system s; the problem then is to characterize

in the sim plest possible maimer the class o f those systems for
which the theorem in question preserves its validity. The m ost
im portant o f these problem s concerns the law o f excluded m iddle
and leads to the concept o f the axiomatizable system, already
discussed in I I I .1
A deductive system X is called axiom atizable i f it has at
least one axiom system , i.e. a finite set Y o f sentences such that
X = Cn(Y). Since, how ever, the finite set Y can be replaced b y a
set which contains a single sentence (nam ely the logioal product
o f the sentences o f the set T ordered in any manner), this
definition can be somewhat sim plified; we denote the olass o f all
axiom atizable system s b y the sym bol ‘31’ and set dow n:
D e f in it io n 9. X e 91 if and only if there is an x e 8 such that
X = Cn({x}).
Hence b y A x. 1 it follow s at once:

T heorem 13. 31 < X 0.

The follow ing theorems express im plicitly elem entary proper­
ties o f axiom atizable system s:

T h e o r e m 14. I f x e S, then (a) Cn({x}) — L if and only if

x e L ; (b) Cn({x}) — 8 if and only i f x e L\ (c) Cn({x}) = Cn({x}).
T heorem 15. I f x, y e 8, then
(а) Cn({x}) £ On{{y}) if and only ifyz>z;
(б) Cn({x}) = Cn(fy}) if and only if x= zy,
(c) (^({x-f-y}) =. Cn({x}).Cn(iy});
(d) Gn({x.y}) = On({x,y}) = Cn{{x))+Cn{{y}).
Fram Ths. 14 and 15 it follow s that the algorithm o f sentences is
hom om orphic to that segment o f the calculus o f system s in which
we restrict ourselves to the discussion o f axiom atizable system s.
(This hom om orphism is not an isom orphism for the reason that
a single axiom atizable system is correlated w ith all equivalent
1 Of. m , p. 35, Def. 9 ; also V , p. 76, § 4, ©specially Def. 35. (I use the
concept of axiomatizabiiity in these two articles in a wider sense than here,
sinoe I relate it not only to deductive systems but to arbitrary sets of sen­
tences.)
356 FO U N D A T IO N S OF THE XII, | 3

sentences.)! This also justifies the remark made above that the
calculus o f arbitrary systems— axiom atizable and non-axiom a-
tizable— form s an essential extension o f the sentential algorithm.
E very theorem o f the calculus o f systems involving axiom atiz­
able systems can be translated com pletely or partially (accord­
ing to w it h e r in this theorem we are dealing exclusively with
axiom atizable systems or not) into the language o f the sentential
algorithm , and vice versa. It must be noted, however, that the
tw o calculi are dual w ith respect to one another: to the sum o f
system s corresponds the product o f sentences, to the product
o f systems the sum o f sentences, and so on.
In this connexion the follow ing theorem , which is a translation
o f Th. 4, can be stated:
T heorem 16. I f in the postulates of the ordinary system of the
algebra of logic we carry out all transformations which were
described in the hypothesis of Th. 6, but with the difference that (B i
is replaced not by ‘ S ’ but by ‘2T, then all the Posts. I-V T I are
satisfied.
From this it follow s in particular that the law o f excluded
m iddle holds in the calculus o f axiom atizable systems (as the
correlate o f the law o f contradiction in the sentential algorithm ).
Far m ore interesting, however, is the fact that the axiom atizable
system s are the only systems which satisfy the law o f Excluded
m iddle:
T heorem 17. X eU if and only if X e S and X-\-X = 8 .
Til. 17 is o f interest if only on account o f the analogy discussed
above between the calculus o f systems and intuitionistic logic;
as is to be expected, the law o f excluded m iddle holds when
applied to system s which in a certain sense have a finitistic
character, while losing its validity for the remaining systems.
Th. 17 in view o f its form could be adopted as the definition of
the concept o f axiom atizability; this definition w ould be form u­
lated entirely in terms o f the calculus o f systems. The problem
f If the algorithm of sentences is modified in the way indicated in the first
foot-note on p. 349, it obviously becomes isomorphic to the calculus of axioma­
tizable syBtems.
XII, § 3 CALCULUS OF SYSTEM S 357

o f a simple characterization o f those systems which obey the

law o f excluded m iddle is com pletely solved in Th. 17. Below
(in Ths. 23, 24, and 26) we shall find answers to analogous
questions concerning some other theorem s o f the calculus o f
classes.
In proving Th. 17 it is convenient to make use o f the follow ­
ing theorem, which is the converse o f Posts. I l l a and IV a from
/Th. 16:
T heorem 18.I f X , 7 e S } X + Y e 2 l and X .Y e % then
X y Y e 2I.1 More generally, i f X , 7 e S and X-\-Y e 21, then there
exist systems X v Y± e 21 such that X x £ X , Yx c Y and
X + Y = X ±+T v
In order to com plete the foundations o f the calculus o f
system s we next give a theorem which establishes a certain
connexion between axiom atizable system s and systems o f an
arbitrary character, and which does not follow from the theorems
o f the calculus o f systems hitherto stated:
T heorem 19. I f X , Y e S , and if for every system Z e 21 the
formula Z c J implies Z 7 , then J c f
W e shall now regard the postulates described in Ths. 6 and 9
(obtained b y transform ing and partly b y weakening the postu­
lates o f the algebra o f logic) together with Ths. 11; 13, 17, and
19 as a system of postulates for the calculus of systems, which is
enriched through infinite operations and the concept o f axiom a-
tizability. It can be shown that every theorem o f the calculus
o f systems which can be proved on the basis o f the axiom s o f
§ 1 can also be derived from the postulates just described. W e
include in the calculus o f systems only those theorems which
contain exclusively sym bols and terms o f the follow ing three
kinds: ( 1) terms o f a general logical character, ( 2) constants
which appear in the postulates, as well as sym bols which can be
defined w ith their help, and finally (3) variables which denote
deductive systems, classes o f systems and relations between
1 An analogous theorem also holds in the sentential calculus of Hevting, A.
(28) (cf. p. 514, note 1):
H:oV6. Vl(oV6):A:aA&.Vl(oA6): O.*Vla.\ . bV“|5.
353 F O U N D A T IO N S OF TH E XII, 5 3

system s, classes o f those classes and relations, and so on). This

remark applies in particular to Ths. 12, 16, and 18.1
The postulates o f the calculus o f systems listed above are not
independent o f one another (e.g. Posts. V a, b can be derived
from the remaining sentences); sim ilarly the prim itive concepts
o f the calculus appearing in these postulates are not independent
o f one another, i.e, some o f these concepts can be defined with
the help o f the remaining ones. B y reducing the number o f the
prim itive concepts a significant sim plification o f the system o f
postulates can be reached. E .g., suppose W8 take as prim itive
concepts only the tw o sym bols, ‘ 6 ’ and the inclusion sign c£ 9
(we abstract from the general logical meaning o f this sign). W e
can then define the remaining terms o f the calculus in the
follow ing w ay:

D r For any given class of systems ^ Y is the unique system

Y*k
I e S which satisfies the conditions: ( 1) Y £ X for every f e f t ;
( 2) if Z e <5 and Y £ Z for every Y e St, then also I s Z .
•
Da. For any given class St of systems, T = 2 Z, where 2
reft zefi
is the doss of all systems Z which satisfy the condition Z £ F for
every Y e St.
Da. For any given systems X and Y,

X + Y ^ J 'Z and X .Y ^ Y J Z ,
Zeft Zeft
where St = {X , F }.

D 4.

X)5. For any given system X , X ~ Y\

Ye& andX.Y~L

Finally we can take Th. 17 as the definition o f the sym bol

I t w ill be seen that with the adoption o f the above definitions

x For a simpler, although equivalent, formulation of the system of postu­

lates for the calculus of systems, and for a clarification of the part played by
these postulates in the development of this calculus* see Tarski (80).
x n , {3 CALCULUS OF SYSTEM S 359

the follow ing postulates suifioe for the foundation o f the whole
oalculus o f system s: Posts. la , V ill a , and X a , whioh were
m entioned in Ths. 6 and 9, and The. 11, 13, and 19. (Post.
V i l l a expresses the fact that for any class Si o f systems there
exists exactly one system X which satisfies the tw o conditions
given in D ef. D J .
It is w orth m entioning that within the theory o f sets a variety
o f relatively sim ple interpretations for the calculus o f systems
can be obtained in the follow ing way. W e consider any class 5t
o f sets which satisfies the follow ing conditions: ( 1) 51 is not
em pty and is at m ost denum erable; ( 2) 51 is a field o f sets, i.e.
the form ula X , X e 51 always implies X + X e 51 and X —T e 5T;1
(3) if it is any subclass o f 51 and^£ Y — ^ ?Y , then there is a finite

subclass H o f Si such that V Y — T Y. Then we form the class

rm fez
G o f all sets which are sums o f an arbitrary number o f sets o f
the class 51. In turn we interpret further terms o f the calculus
o f system s as follow s. T o the sym bolio expressions ‘X £ Y ’ and
‘X .X ’ we ascribe the usual set-theoretical meaning. Similarly
•
we regard the sums X 4 -Y and X as the usual sums o f sets
rOk
(in the sense o f set theory). W e put, further, S —^ X and

L — 0 ; finally we interpret the sym bols X* and ‘X ’ (to

which the usual meaning cannot be asoribed) b y follow ing the

lines o f D efs. D a and D 5. It turns out that the system o f con­
cepts obtained in this w ay satisfies all the postulates o f the
oalculus o f systems.
Prom the point o f view o f the problem s in which we are in­
terested the fa ct that the calculus o f systems can be interpreted
in various ways within the calculus o f systems itself is o f great
im portance. In order to form ulate the relevant theorems con­
veniently we introduce the follow ing notation:
D e f in it io n 1 0 . (a) S x is the class of aU systems Y e S such that
X £ X;
(b) <5X is the class of ail systems Y e G such that X £ X.
1 Cf. Hausdorff, F. (25), pp. 78 &
360 F O U N D A T IO N S OF TH E XII, § 3

T h eo rem 20. Let A be any deductive system; let X A = A .X

for every system l e S ; finally let be the class of all systems
X g <5A, such that A s X + X (and in particular let %A = QA.%
in case A e 21). In the postulates of the calculus of systems replace
everywhere the symbols ‘ S ’ , (X \ ‘ S ’ , and ‘ 21* respectively by ‘A\
‘X A\ *&A\ and (yiA\ and leave the remaining symbols unchanged,.
Then all postulates of the calculus of systems remain valid with
the exception of Ths. 11 and 13, either of which holds if and only if
AeW.
T h eorem 21. Let B be any deductive system; further let
XB — T 7
TeGandJC'T£B
for every system I s S (and in particular let X B = in case
X e 21); finally let $lB be the class of all systems of the form B-\-Y
where Y e 21 (and in particular let %B = 3^ .21, in case B e 21).
In the postulates of the calculus of systems replace everywhere the
symbols ‘L\ *X\ ‘ S ’ , and ‘ 2T respectively by *B\ *XB i *<ZB\
and ‘ 2I#\ leaving the remaining symbols unchanged. Then all
postulates of the calculus of systems are satisfied.
The theorems just stated show that all results derived from
the postulates o f the calculus o f system s can be relativized, and
in fact in tw o directions: ( 1) W e can narrow the set 8 o f all m ean­
ingful sentences b y replacing it by any axiom atizable system A
and restricting the discussion to those deductive systems which
are included in A (when we are concerned with results which
can be proved w ithout the help o f Ths. 11 and 13, the assump­
tion o f the axiom atizability o f the system A is superfluous).
( 2) W e can enlarge the set L o f all logical theorems b y replacing
it b y any deductive system B and restricting the discussion to
those systems which include the system JS.1 W e can, o f course,
subject every result to the tw ofold relativization— first in one
direction and then in the other. It is not difficult to formulate
a general theorem which embraces Ths. 20 and 21 as special
cases and makes possible the simultaneous relativization in both
directions.
1 Of. an analogous result in article V of the present work, pp. 67 f., Th. 6
and the accompanying remarks.
XII, 5 3 CALCULUS OF SYSTEM S 361

A ll further theorems o f the present article are consequences o f

the postulates o f the calculus o f systems.
T h eo rem 22. I f X e ®, then
(a) * = I Y
re©1.a
and
(b) 1 = n r.
FeSx .?l
Th. 22 a stands in close connexion w ith Th. 19; in the system
o f postulates for the calculus o f systems Th. 19 could be replaced
b y Th. 22 a. W ith the help o f Th. 22 h we can set up a series
o f necessary and sufficient conditions for those systems which
satisfy the law o f double negation:
T heorem 23. For an arbitrary system X e S the following
conditions are equivalent: ( 1) X = X\ (2) there exists a system
Y e 3 such that X = F ; (3) the formula X s Y always implies
the formula Y G X for Y e ®; (4) the formula I c f always
implies the formula Y ^ X for Y e S ; {5) X = ]"J Y\ (6) there
exists a class R ^ 21, such that X = f j F .
Feft
Consequently the systems which satisfy the law o f double
negation coincide with the intersections o f axiom atizable
systems.
In Th. 17 we already recognized a characteristic property o f
axiom atizable systems which was form ulated entirely in terms
o f the calculus o f system s; we shall now give tw o other pro­
perties o f this kind:
T heorem 24. For an arbitrary system X e & the following con­
ditions are equivalent: ( 1) X e 91; ( 2) every class R c 0 satisfies
the formula X 4- T \ y = T l (J C + n * (3) for every class R £ S
r e* re*
the formula J = ^ F implies the existence of a finite class f i e f t
.Feft
such that X = ^ Y.
Fes
In condition (3) o f this theorem both identity signs can be
replaced b y inclusion signs.
362 F O U N D A T IO N S OF T H E XU, $ 3

N on-axiom atizable system s can be characterized in the follow ­

ing w ay:
T heorem 25. The following conditions are equivalent:
( 1) X e S — 2I; (2) there is an infinite class St £ 6 such that
X = 2 T,where Y L for any Y e Stand Y .Z = L for any two
Yeh
distinct systems Y, Z € St; (3) there exists an infinite sequence
of systems 7n e S , such that X — yi 4-5’i4 -~ .4 -}rn + *-' where
yn s r n+1 and r n # r n+i for every natural number n. It can,
moreover, be assumed in conditions (2) and (3) that all systems of
the class St or all systems Yn are axiomatizabh.
The non-axiom atizable system s are thus sums o f infinite
classes o f logically disjoint system s and also sums o f infinite
series o f strictly increasing systems.
I f we translate the condition ( 2) o f the above theorem into
the language o f the sentential algorithm (after having replaced
‘ S ’ b y ‘ 21’ ), we com e to the conclusion that for every system
l e S a set Y s X can be constructed suoh that (a) X = Cn(Y),
(ft)Y.L = 0 , and (y) x-{-y e L for all x, y e T (or, what amounts
to the same thing, (/S') if x, y e Y, x o z, and y d z, then z e L);
if, m oreover, X e S —21, then (3) the set Y is infinite. The condi­
tions (;3) and (y) express the fa ct that Y is a set o f maximal
independent sentences in the sense o f Shelfer; hence it results
a fortiori that Y is a set o f independent sentences in the usual
sense and therefore a basis o f the system X .1
In an analogous manner .ve easily infer, either from condition
(3) o f Th. 25, or directly, that with every system X 6 S an.
infinite sequence o f sentences yn e X can be correlated which
satisfy the conditions: (a) for every sentence x e X there exists
a sentence y n such that yn o x (thus, if Y is the set o f all mem­
bers o f our sequence, then Cn(Y) = X ); (ft) yn+l D yn for every
natural number n; if, m oreover, X £ S —21, then (y) the formula
yn = yn+z does n ot hold for any natural n. The sequence of
sentences which satisfy the condition (ft) could be called a
logically increasing sequence of sentences; if, m oreover, the condi-
1 Cf. Sheffer, H. M. (63), p. 32; for the concept of bans of. HI, pp. 36 f.,
Def. 7 and Th. 17* and also V, pp, 88 f., especially Def. 6.
X II, I 3 CALCULUS OF SYSTEM S 303

tion (y) is satisfied, then we speak o f a strictly increasing sequence

o f sentences.
B y a logical product of an infinite sequence of sentences yn
we understand a sentence x such that (a) x o y n for every natural
number n, and (/3) if 2 is any sentence such that z z>yn for every
natural num ber n, then z o x (there m ay be m any such sentences
x bu t they are m utually equivalent); in an analogous manner
we define the logical product o f an arbitrary set o f sentences.
It seems natural to call an increasing sequence o f sentences yn
convergent when a logical product o f this sequence exists. Since
such a logical product is also the product o f the whole system
X w ith which the sequence considered was correlated (in the
w ay indicated in the preceding paragraph),we shall in an ana­
logous manner also call the system X itself convergent. W hen
the logical product x o f the system X itself belongs to this
system , the form ula X — Cn({a:}) is easily seen to hold, and the
system X is sim ply axiom atizable.
The class o f convergent system s we shall denote b y the
sym bol translating the definition o f convergence into the
language o f the calculus o f system s we obtain
fim r a iO H 11. I e £ if and only i / I e S and f j Y e 2 l;
v=sx.u
in other words, if there exists a system Tsuch that (1) Y e% and
X £ Y, as well as ( 2) for every system Z the formulas Z 6 31 and
X £ Z imply Y £ Z.
Various transform ations o f this definition are known;
T heorem 26. For any given system l e S the following
conditions are equivalent: (1) I s C , (2) JP e 21, (3) 5 e 21,
(4) = S ’, (5) X .Y = S-\-Y for every system Y e S ;
( 6) x 4~Y = JP-}- y for every system Y e<S.
W e can thus characterize convergent systems m ost sim ply as
these systems whose logical com plem ents are axiom atizable and
which therefore satisfy the law o f excluded m iddle in a weaker
form : JT-f-jP = 8.
I t follow s from Ths. 23, 24, and 26 that some theorems o f the
calculus o f olasses w hich are n ot generally valid in the oaloulus
364 F O U N D A T IO N S OF TH E XII, § 3

of systems regain their validity when we assume for at least one

of the systems to which these theorems refer that it is axioma-
tizable (e.g. the laws of contraposition, the distributive law
under infinite multiplication), or at least that it is convergent
(De Morgan’s law’).
As examples of other theorems which concern convergent
systems we give
T heorem 27. (a) 91 £ C;
(b) X e £ if and only if X e & and l e d ;
(c) i f X , Y e (£, then X + Y e (Z and X .Y e (£.
T heorem 28. I f X , Y e S , X + Y e 91 and X .Y e <£, then
I J e d .
In this theorem, which is an analogue of Th. 18, one cannot
replace £9T by
The question arises whether there exist convergent non-
axiomatizabie systems, in other words whether there exist non-
axiomatizabie systems whose complements are axiomatizable.
We shall see later (in Th. 38) that the answer is in the affirma­
tive assuming that non-axiomatizable systems exist at all, that
;s, assuming that the class S is infinite. On the other hand it
will turn out that on the same assumption there are also non-
con vergent systems. We have thus three classes of systems:
(I) axiomatizable systems, which are correlated with single sen­
tences; (2) non-ayiomatizahle convergent systems, which corre­
spond to infinite convergent strictly increasing sequences of
sentences, and finally (3) divergent systems, which correspond
to infinite divergent sequences. The systems of the first two
classes have in general similar properties; one of the principal
differences consists in the fact that no system of the second
class satisfies the law of double negation.

§ 4. I rreducible and Complete S ystems

On the basis of the calculus of systems certain theorems of
a deeper nature can be established, concerning the cardinality
and structure of various kinds of deductive systems; in the
formulations and proofs of these theorems two concepts occur
XII, § 4 CALCULUS OF SYSTEM S 365

which have not yet been used in this article, nam ely the con­
cepts o f an irreducible and o f a complete system . W e denote the
class o f irreducible systems b y the sym bol ‘ 3 \ that o f com plete
systems b y the sym bol ‘S ’ .1 The definitions o f these two con­
cepts are quite analogous:
D efinition 12 . X e 3 if and only if X ^ L \and if for every
system 7 e S for which 7 £ X , we have Y = L or ¥ = X (in
other wordsy if S x = {L ,X }).
D efinition 13. X e 93 if arid only if X e S, X # S, and if,
for every system Y e S for which X c F, we have Y = X o r Y = S
{in other words, if = {$>X}).
The follow ing are simple consequences o f these definitions:
T heorem 29. The following conditions are mutually equiva­
lent: (1) X 6 3 ; (2) I e S, I ^ L , and the formulas Y, Z e S
and X = Y + Z always imply X = Y or X — Z; (3) I g S,
^ amZ /or every system Y e & we have X c. 7 or X £ F.
T heorem 30, The following conditions are mutually equivalent;
(I) X e 93; (2j X e 3 , X ^ ami formulas Y, Z e 3 am!
X = F .Z aZtmy* X = Y or X = Z\ (o^X e 3 , X ^ /?,
and /or every system Y e 3 we kave Y £ X or Y s.- X.
In the conditions (3) o f these two theorems the formula
T s S ’ can be replaced by ‘F e 2t\
T heorem 31. (a-) j f X , F e 3 ami X # F, eAer, X . F — L;
(6) if X yY e S X ^ F, lAert X - f Y =
T heorem 32, V ; i / ft c 3, £ c: 3 and ^ X" £ T X. lAen

51 c fi;
(b) if ft c s , £ £ * <ro<* f f X £ f j X, 5 . 2£ c ft.
Xest Xr$
1 I have already dealt with the concept of completeness m xir; -y - works
(cf. ITI, p. 34, I)ef. 7; also V, pp. 93 ff., especially Def. 7); t h e :- ,:i. spt is
used here in a somewhat different- sense, because (1) it is aprh-rd evciiuively
t o deductive systems and nut to arbitrary sets of se i iteno-rs, a r d {2} o-ii y t h o s e
deductive system*? are called complete which—in the Ud tennh>.;>k':^j—are
both complete turd c c a s i s x e n t . The concept of an ir r e d u c ib le sy si*- r h ere
introduced for the first time.
see F O U N D A T IO N S OF THE XH, H

Contrary to w jiat m ight be supposed on the ground o f Defs.

12 and 13 as well as Ths. 29 and 30, there exists between the
properties o f the classes 3 and 93 no exact dual correspondence.
A small difference can already be seen in Th. 32; the divergence
appears clearly in the further theorem s.

T h eo rem 33. X e 3 if and only if X e S and X e 93.

In this theorem ‘3 ’ cannot sim ply be replaced b y ‘ S3’ and

vice versa; on the other hand we have:

T h eo rem 34. I f X e S3.2t, then l e 3 ; if X e 93— 21, then

X = L,

One o f the essential differences between the properties o f the

classes 3 and 93 consists in the fact that the class 3 consists
exclusively o f axiom atizable system s, whilst the class 23 oan
contain both axiom atizable and non-axiom atizable system s;
from Th. 34 it follow s, how ever, that com plete systems in all
cases possess axiom atizable com plem ents and are thus con*
vergent:

T e e o b b m 35. 3 £ 2t, 23 £ (£.

The theorem o f Lindenbaum , according to which every system

distinct from S oan be extended to a com plete system 1 is o f
a deeper character. This result oan be im proved in the follow ­
ing w ay:

T h eo rem 36. I f X e S , then X = TT Y.

r«5x . «
Thus every system can be represented as an intersection o f
com plete system s. The corresponding dual theorem for the
class 3 fails; i f the class <3 is infinite, then, as can easily be
seen from Th. 37 below , there certainly are systems which are
not sums o f irreducible systems.
W e now com e to the theorem s announced at the beginning o f
this section.

1 Cf. in , p. 34, Th. 12; V, p. 98, Th. 58.

xn, §4 CALCULUS OF SYSTEM S 367

T heorem 37. The following condition* are equivalent:

( i ) i < K0; (2) SB < N0; (3) there exists a natural number n such
that 5 s= n and § = 2n; (4) there exists a natural number » such
that 1 = n and i = 2»; (5) S £ SK; (6) SB£ SI; (7 ) S » T T‘>
. res
(8) for every I e S , X = jT 3r; (9) for every l e g ,
.3
x - ff r.
I'«Sx •®- S
Theorem 38. The following conditions are equivalent:
(1) 5 > So; (2) 3 = 2 * ; (3) I = N 0; (4) ^ X0;
(5) 1 ^ 1 = 2 *.
B y applying the relativization theorems— 20 and 21— we can
derive consequences o f a m ore general nature from the last tw o
theorem s. These consequences ooncem those deductive systems
which are included in a given axiom atizable system A or which
include a given (not necessarily axiom atizable) system B.
Nevertheless it m ust be noticed that in generalizing Ths. 37 and
38, all kinds o f system s involved in these theorems undergo
relativization with respect to the set A or the set B ; not only
do the classes <3 and SI go over respectively into QA and WA,
or in to SB and %B (cf. Ths. 20 and 21), but an analogous trans­
form ation m ust also be perform ed on classes <£, 3 , and SB. B y
relativizing the definitions o f these classes we easily realize that
(under the assum ption A 6 31) and 3 X — <5^.3,
and that the class SB4 consists o f those and only those sys­
tems X which satisfy the form ulas X ^ A and . <3X = {.<4,X}
(SM can also be characterized as the class o f all systems o f the
form A . Y, where Y is an arbitrary com plete system which does
not include A ). M oreover, the class (ZB consists o f all systems
X e S # which satisfy the condition: am ong the systems o f the
class Ws which include X there is a smallest one (if B e iff, then
<ZB = Qs . <£); the class 3 B consists o f all systems X w hich satisfy
the form ulas X # B and <Bb . < 5 x — { B , X } (the form ulas
X e 3 j - and Y e SBX are thus equivalent); finally we have
— &A •S3- In the case o f relativizations baaed on Th. 20
the assum ption A e ST is essentially necessary. Nevertheless, it
368 FO U N D A T IO N S OF THE xn, § 4

can be shown that the equivalence of the relativized conditions

( 1), (3), and (5) of Th. 37 does not depend on this assumption,
since from each of these conditions it follows that A g SIl; the
same applies to the conditions^?) and (8) if we supplement each
o f them with the words ‘and 3 4 < R f. Similarly the relativized
form ulas (1), (2), (3), and (5) of Th. 38 are equivalent indepen­
dently of the assumption that the system A is axiomatizable.
(A ll this holds on the condition that we interpret the symbol
as and neglect the more general interpretation
indicated in Th. 2G,)
Th. 37 can be extended by including in it various theorems
o f the ordinary cel^ilus of classes which are not generally valid
in the calculus of system nas conditions equi valent to the formulas
(1)—(9). e.g. the law of excluded middle (the equivalence of which
to formula (5) can bo inferred from Th. 17), the law of double
negation, as well as toe laws of contraposition stated in Th. 23,
further, the weaker law of excluded middle and I>e Morgan’s
law which appear in Th. 26. Some of the theorems of the calculus
of classes just mentioned can. be used to supplement- the relati-
vization3 of Th. 37 described above if the range of applications
of these theorems is restricted to systems of the class \oA or <Bb ;
thus, for examph-. under the assumption that At g %, the condi­
tions (] )-($) of Th. C7 relativized wit h respect to the system A
are equivalent to the condition: X --= X for every system X g S T
The power of the class can be called the cardinal degree
of capacity of ike system A . the power of the class S# can be
called the cardinal degree of completeness of the system BA Since
S = — Sr, the power of the class S is both the degree of
capacity of the system S and the degree of completeness of the
system L\ this cardinal number can be called the cardinal degree
of capacity, or the aogrse of completeness, of the deductive theory
discussed« From Ths. 37 and 38 it follows that the power of
the class G is either a natural number of the form 2?I or equals
2K*. By virtue of the relativized theorems this applies to the
degree of capacity and the degree of completeness of an arhi-
1 The second of these concepts I have previously introduced; cf. V of the
present volume, pp. 100 ff., especially Def. 8.
X II, § 4 CALCULUS OF SYSTEM S 369

trary deductive system as well. In case the degree o f com plete­

ness o f a theory is finite, the structure o f the theory is very
sim ple; as Th. 37 shows, all systems are then axiom atizable,
and hence the calculus o f systems is not form ally distinct from
the calculus o f classes; every system can be represented— and
in fact in only one w ay— as the sum o f a finite number o f
irreducible systems or as the intersection o f a finite number o f
axiom atizable com plete systems. The situation is almost equally
clear in the case when non-axiom atizable systems exist but only
one o f them is com plete. This case is treated in the follow ing
theorem :
T heorem 39. The formulas (1) 93— 91 — l and (2) ^ Y e 3S
fS
art mutually equivalent^and imply the following consequences:
(3) S = 1 = 3 = » = l 7 l - H0;J4) i / P l ==» < K0,
then X e 51, X = ^ Y. X — |"J Y, = 2r* and 3 ^ = 2Kj;
F e $ * .3 F e < 3 * .3

(5) if 3 ~ S * = n < Xc, then either X e % X = f[ 7 , and

S ^ = 2», j 7 <m4 C l : 1'7 while in

both cases X ^ .7 and S ^ — 2**: (6) i/ 3 , 3 ~~ 3 = X0>

th en X e& -& ,X ^ J 7, = § 7,and&Z = 2«*.

F e < 5 * .3 F e s -e *

The cases described in Ths. 3? and 39 possess m any points o f

contact. Characteristic for both cases is the role o f irreducible
system s as basic elements out of wiiich arbitrary deductive
system s can be constructed, either with the help o f the operation
o f addition alone or with the help o f two operations— addition
and com plem entation. The logical alternative o f the two case?
considered can be characterized as follows:

T heorem 40. The follmving conditions are equivalent:

(1) < 1; (2) ifX = 2 7, then '2* < (3) for every system
===== r S =3
X e 991 ei£Acr S x < X0 or (4) for every system l e S
either X ^ ^ Y or X
Fe©*.3 FaCS-G*
370 F O U N D A T IO N S OF TH E x n ,§ 4

T o conclude this section we w ill disouse the concept o f the

structural type o f a theory .1 W e say that tw o deductive theories
(or, m ore generally, tw o arbitrary m odels o f the system o f
axiom s and prim itive concepts listed in § 1) possess the same
structural type if the class o f all deductive systems o f the first
theory is isom orphic w ith the class o f all system s o f the second
theory w ith respect to the relation o f inclusion, i.e. i f between
the system s o f the tw o theories a one-one correlation can be set
up which satisfies the follow ing condition: if X x and 1\ are tw o
arbitrary systems o f the first theory, and X 2 and Y? are the
correlated system s o f the second theory, then X x £ if and
only if X a £ Y2. This can also be expressed otherwise: between
the sentences o f the tw o theories a m any-m any correlation can
be set up so that if x1 and yx are tw o arbitrary sentences o f the
first theory whilst xa and y%are any tw o correlated sentences
o f the second theory, then the form ulas x1 d yx and x%o j/a are
equivalent.
It turns out that tw o cardinal numbers exert an essential in­
fluence on the structural type o f a theory: the power a o f the
class o f all com plete aziom atizable systems and the power n
o f the class o f all com plete non-axiom atizable system s. The
ordered pair o f these numbers (a, tt) we oall the characteristic
pair of a deductive theory. The numbers a and n cannot be quite
arbitrary; from The. 37 and 38 it follow s that a < X 0 and
n < 2“*, it also follow s that the case a < X0, 0 < n < X0 and the
case a — X 0, n = 0 are excluded. Other relations between the
numbers a and n have been established b y A . M ostowski*; he has
proved that only the follow ing pairs o f numbers can appear as
characteristic pairs o f deductive theories: ( 1) the pair (a , 0 ),
where a is any natural num ber distinct from 0 ; ( 2) the pair
(X0, n), where tt is any num ber < X 0; (3) the pair (a, 2“ »), where
a is any num ber < X0 (hence it follow s in particular that the
number n as w ell as the number o = a + u — 51 satisfies the
continuum hypothesis). M oreover, it turns out that the num -
1 The importance of this conoept has been recognized, independently of
me, by Lindanbaum.
* The results of Mostowski referred to in this section are published in
Mostowski, A . (63c), pp. 34-68.
xn, M CALCULUS OF SYSTEM S #71

bers in question are subjeot to no further restrictions: given an

arbitrary pair (a, tt) o f one o f the three types just listed we can
always construct a deduotive theory such that 2 0 ! = a and

Theories w ith the characteristic pair (a, 0 ) where a < K0

possess the sim plest structure; follow ing them are the theories
w ith the pair (Kg, 1). From Ths. 37 and 39, in whioh we have
studied theories o f this kind, it follow s that in these sim plest
oases the characteristic pair com pletely determines the structural
type o f a theory: tw o theories w ith the same characteristic pair
possess the same structural type. M ostowski has shown that this
result can be extended to all characteristic pairs (K0, it) so long
as n < K0, bu t that in general the characteristic pair does not
determine the structural type o f the th eory ; am ong the
theories w ith the characteristic pair (K0, N0) non-denum erably
m any different structural types appear.
The results concerning the structural typ e o f an entire theory
can be relativized with respect to individual system s o f the
theory. In that case it is necessary to distinguish between the
inner and the outer structural types of the system: tw o systems
X and 7 (o f one theory or even o f tw o different deductive
theories) possess the same inner (or outer) structural types i f
the classes &x and ® r (or and <Sr ) are isom orphic with
respect to the relation o f inclusion. The structural type o f the
whole theory is both the inner structural type o f the system S
and the outer o f the system L. The m ethod o f relativization
enables us, in the investigation o f structural types, to restrict
ourselves to one single deductive theory w ithout the scope o f
our discussion being thereby essentially restricted: for every
theory w ith the characteristic pair (a, 2**) is, as M ostowski has
shown, universal in the sense that am ong the structural types
o f individual system s o f this theory the structural types o f all
possible theories are already represented.

§ 5. A pplications to Spe c ia l D eductive T heories

We shall now illustrate our general discussion by examples
o f some special deductive theories in whioh the mutual relations
372 F O U N D A T IO N S OF TH E XII, § 5

between various classes o f deductive systems are ordered in a

particularly clear manner.
The first o f the theories w hich we have in m ind can be called
the elementary theory of dense order. In the theorems o f this
theory denum erably m any variables o f different form appear;
let these be, for exam ple, the signs V ’ , *a*\ ca'*\ and so on.
M oreover we have here four constants: the symbol of the order­
ing relation ‘ JR’ , the negation sign C N\ the implication sign ‘C\
and the universal quantifier ‘El’ .1
A n expression which is com posed o f the sym bol tR i and tw o
variables follow ing it, e.g. cRa'a*’ (read ‘a' precedes a"’ or (a*
follows a ” ), we call an elementary formula. The expression con­
sisting o f the negation sign and an arbitrary expression x follow ­
ing it is called the negation of the expression x , sym bolically x;
thus e.g. ‘NRa’a*9(read ‘ a ' does not precede a"9or Ht is false that
a' preced.es a"9) is the negation o f the elem entary form ula ‘Rara
The form ula which is form ed b y the im plication sign and tw o
succeeding expressions x and y is called the implication with the
antecedent x and the consequent y> sym bolically x - * y ; thus, e.g.,
iCRafaNNRa”a” (read cI f af precedes a" then a* does not precede
af’ ) is an im plication w ith the antecedent iRa,arr>and the conse­
quent ‘NRa"a'9. B y putting in front o f an arbitrary expression
x the universal quantifier follow ed b y a variable we obtain the
universal quantification of the expression x with respect to the given
variable\ e.g. ‘II a'NRa’a'9 (read: for every af it is false that a '
precedes itself) is the universal quantification o f the expression
‘N Ra'a'9with respect to the variable V \
Further, we define recursively a certain category o f expres­
sions called sentential functions: sentential functions o f the 1st
order are elem entary form ulas, functions o f higher order arise
out o f functions o f lower order b y applying one o f the three
operations: negation, im plication, and universal quantification.
Am ong the variables which occur in a given sentential function
we distinguish b y a well-known m ethod free and bound variables.2
A sentential function containing no free variables is called a
1 For symbolism see IV , pp. 39 and 5 4 ; V III, p. 168.
* Cf. article TX, Def. 6, and article V III, Def. 11.
X II, § 5 CALCULUS OF SYSTEM S 373

(meaningful) sentence; the set o f all sentences is denoted b y the

sym bol ‘S’ .
Finally we define recursively the concept o f logical theorem.
Am ong the logical theorems o f the 1st order, i.e. am ong the
logical axioms, we include sentential functions o f tw o kinds:
( 1) all expressions o f the type (x -> x) -> x, (x-+ y), and
(x -> y) -> ((y -> z) -> (# -> z)), where a:, y, and z are arbitrary
sentential functions; (2) the follow ing four sentences:
cx = m W M a V ’,
c 2 = ‘ na'na"CRa'af'NRa*a'’ i
c 3 = cn a ,n a ''n a //'C i? a V (7 A i? a 'V 'JRa'a

As is seen, the axiom s o f the 1st kind com prise all sentential
functions which can be obtained b y substitution from the
axiom s o f the sentential calculus. The first o f the axiom s o f
the 2nd kind expresses the fact that the relation R is n ot em pty:
there exist tw o elements o f which one precedes the other. The
next tw o characterize R as a relation which orders all the
elements o f the universe o f discourse .1 Finally, the last asserts
that the order established b y the relation R is dense: i f a ' pre­
cedes a ", then there is an element af" which succeeds a' and
precedes a\2
Logical theorem s o f higher orders are obtained from the
theorems o f lower orders w ith the help o f one o f the four w ell-
known operations: substitution, detachm ent, and the insertion
and deletion o f the universal quantifier in the consequent.8 The
set o f all logical theorem s which contain no free variables is
denoted b y the sym bol ‘L\
W e have thus shown how all prim itive expressions o f general
1 The axioms ca and Cj deviate somewhat from the usual postulates of order­
ing. The reason is that among the signs of the theory discussed the identity
sign does not appear; of. Lukasiewicz, J. (50), pp- 35 ff.
2 The attention of the reader, who may be surprised at the presence of the
sentences cx-c^ among the logical axioms and consequently among the elements
of the set L, should be directed to the fact that the set L can be interpreted
as the totality of sentences which are accepted as true when we begin to develop
the deductive theory under consideration (cf. § 1).
* Cf. IX , p . 285; Y i n , p . 1 8 1 .
374 F O U N D A T IO N S OF TH E X II, $ 5

m etam athem atics are to be interpreted in their application to

the special deductive theory which has been chosen as the sub­
je ct o f investigation. I t can easily be established that the Axs.
1-5 of § 1 preserve their validity in this interpretation. Thus all
consequenoes o f these axiom s are also valid; in particular the
calculus o f system s can be applied in cur discussion.
L et us now consider tw o special sentences:
dx = 'na'N na'N R a’a " and da = 'H a'N Ila'N IlaW ’ .
As can easily be realized these sentences express the fact that
for every elem ent there is another element which follow s it or
w hich precedes i t : in other words, in the ordering o f the elements
w hich is established b y the relation B there is no last and no first
elem ent.
W ith the help o f a frequently used m ethod, which oonsists
in reducing the sentences to norm al form and successively elim i­
nating the quantifiers ,1the follow ing theorem can be established
(we form ulate it in the term s o f the sentential algorithm which
was introduced in §§ 1 and 2):
Every sentence x e 8 satisfies exactly one of the following sixteen
formulas/_ (!) x e L , ( 2) x e L, (3) * = dlt (4 ) x = d1, (5) x j s da,
( 6) x (7) x = d i+ d 2, ( 8) x =s d ^ d ^ (9) x m d j-fd a,
( 10) x = dX“)-da, ( 11) x 32; di.dg, ( 12) x s d^.da, (15) x s d^.dg,
(14) x =3 d^.dnj, (15) x s d1.di+ d i.d 2, (10) x s d^.d^-j— d^.dj.
This theorem im plicitly contains the solution o f all the more
im portant m etam athem atical problem s relating to the theory
o f dense order. Translated into the language o f the calculus o f
system s it states that the class 6 contains sixteen different
system s; if we pu t D x — C n({dJ) and D s = On({d^), then
these system s are L, 8, Dx, Dlt D a, D a, D t .D t,
D1. Z>a, D-y-\~D2t D ^-|-Ha, D±-\-D2, U j-fH j, •-C's-f' H i •D a,
and D 1.D i \-D1.D z. A ll these system s are axiom atizable; the
degree o f content and the degree o f completeness o f each o f them
is obviously finite. F our o f the systems are irreducible: D j .D j,

1 So far as I know this method was first used in Skolem, Th. (64). It was
applied to the theory of dense order by Langford, C. H . (45), pp. 16 ff. The
facts given below concerning this theory supplement Langford’s results.
x n , §5 CALCULUS OF SYSTEM S 375

D x. i> 2» i? i •
and DX.D V Four others, the logical complements
o f the preceding systems, are com plete: I X .D 2 == Dx^ D tt
DX.D 2 = D x+D& Z>t .!>*=== jDx+ D v a n d « 2>X+ A : the
characteristic pair o f the theory is thus (4 ,0), E very system is
the sum o f all the irreducible systems included in it and a!*o
the intersection o f all com plete systems which include it. In a
word, the relations between the systems are exactly o f the kind
anticipated in Th. 37.
The second theory w ith which we shall deal— the elementary
theory of isolated order— has an infinitistio character, and the
variety o f its systems is m uch larger. On a superficial view this
theory differs only slightly from the theory of dense order: the
same constants and variables appear in it, the same expressions
are called sentential functions and sentences, the same opera­
tions are used in the recursive definition o f the concept of logical
theorem . The sole distinction consists in the fact that we om it
from the list o f logical axiom s and replace it by tw o sentences:
cft s ina'na*CRafa*Nna'*CRa'a'*Nna™CRaf<^
ce - tTla,T[a"GRaf,a'NRa,"CRama'NUa™CRa !a*NRaha*f:.
These tw o new axiom s state that every element which, in the
order established b y the relation R> is not a last or a first
element has an im m ediate successor or an im m ediate prede­
cessor .1
In order to study the structure o f deductive systems in the
theory o f isolated order we single out a particular sequence o f
sentences en; translated into English the sentence en expresses
the fact that there exist at m ost n + 1 different elements. W e
put
e, = ‘na'na'IIarCIla'a'N Ba'a"',
e2 - ^ nam arn^ CSaW CBa^ arN Ba^ a199,
and so on ; it is easily seen how the recursive definition o f en
should look . W ith the help o f this sequence and o f die sentences
dx and d2 form ulated in connexion with the theory o f dense
order, we define a new sequence o f sentences in> nam ely:
1 Langford, C. H . (45), pp, 459 ff., is concerned with a similar but more
special theory, namely with the elementary theory of the ord&tr type a>.
376 F O U N D A T IO N S OF TH E X II, § 6

H — ^1'4“^2>^2 — — 61, a n d in — ^n-4~t“ ^n-8

for every natural n 5. The content o f the sentences ilf i 2,
and i 3 is clear (the sentence t3, for exam ple, states that in the
order established b y the relation E either a first or a last element
occurs); the sentences i4> ?'6, state respectively that the
number o f all elements is not exactly equal to 2, 3, 4 ,....
W ith the help o f the above-m entioned m ethod (that o f the
successive elim ination o f quantifiers)1 we can prove the follow ­
ing theorem , which has a decisive significance for the m eta-
m athem atical discussion o f the theory o f isolated order:
Every sentence x e S satisfies one and only one of the four con­
ditions: ( 1) x e L\ ( 2) x e L\ (3) there exists a finite increasing
sequence of natural numbers 7^ , n2>..., np such that x = &
( 4 ) there exists a finite increasing sequence of natural numbers
p —
nv n2,..., np such that x — j? in , The sequences nls n
which occur in (3) and (4) are uniquely determined by the sen­
tence x .
W e state here the m ost im portant consequences o f this theo­
rem in the field o f the calculus o f systems.
Let us put Ir ----- Cn({in)) for every natural number n. It can
be shown that the class o f all systems In is identical with the
class 3 ; since for any tw o different natural numbers n and p the
corresponding system s In and Ip are always different, we have
3 = K
The class S3 consists o f all systems I)l9 where n is an arbitrary
natural number, and o f the system / 14 - / 24 - . - + / n+ * ‘ d this
latter system is not axiom atizable, whilst the remaining ones are
axiom atizable. W e have therefore S3 — 93.SI = K0, SB—91 = 1;
the deductive theory discussed thus has the characteristic pair

The class 91 contains the systems L and S as well as all systems

o f the form 4 1+ 4 , + - + 4 I, and 4 ,- 4 , ...... T^, where the

1 Cf. p. 374, footnote (the application o f this method is however not so simple
in the case now discussed as it was in the previous One).
xn, §s CALCULUS OF SYSTEM S 377

natural numbers n19 n2,.-,n p form a n arbitrary finite increasing

sequence; hence it follow s that 2t = K0. Systems ol the form
4 1+ 4 , - r . . . + 4 , have a finite degree o f capacity equal to 2Py
but an infinite degree o f com pleteness equal to 2*®; it is the other
way round in the case o f the system s o f the form Inx.In$......
It is to be noted that the system S and the systems o f the form
I Ui. 1^ ......are the only system s o f the theory o f isolated order
which cannot be represented as sums o f the irreducible systems
included in them .
The class <5— 31 consists o f system s o f the form

where the numbers nv npi... form an arbitrary infinite in­

creasing sequence; these system s are convergent or divergent
according to whether the num ber o f natural numbers which are
not terms o f the sequence nl9 7t2,..., np>-- finite or not.
H ence 1 = f = ! = K0, while § = = 2**. The
degree o f capacity o f every system o f the class 3 — 21 is equal
to 2K«. I f X — ... is a convergent system ,
then X = and X = 4 ^ -4 ^ ......4^> where
7% , ra2,..., mq are all the distinct natural numbers wlueh do not
occur in the sequence 7^, npi. . . . The degree o f com plete­
ness o f such a system X is equal to 2«+1. If, however,

is a divergent system , then

X ^ 4 h + i mt+ - - - + 4 lg+ - ‘ ->

where mv is the infinite increasing sequence o f all
those natural numbers which do not occur in the sequence
nv n2,..., np,...; m oreover, X = 4 1+ 4 t+ . • • + /„,+ •••, so that
the system X satisfies the law o f double negation. The degree
o f com pleteness o f a divergent system X is infinite, equal to 2K».
In a word, we have here ju st the case which was anticipated
and described in Th. 39.
M any other examples o f deductive theories are known whose
characteristic pair is (X0, 1) and which— in the sense o f the
 378 F O U N D A T IO N S OF TH E x n , §5

concluding remarks o f § 4—have the same structural type as

the elementary theory o f isolated order. An especially simple
theory o f this kind is the elementary theory of identity. Its con­
struction is very similar to both the previous theories; the
difference consists only in that in the system of logical axioms,
instead of the sentences cv ca,..., c8, only two sentences are
included, ‘n o'jK aV ’ and ‘Ila'na’ na"'CRa,a'CRa'amRa‘ a,”\
which together characterize R as a reflexive, symmetrical, and
transitive relation. The symbol {R ’ we can here call the identity
sign, the formula lRa’a is read ‘o ' is identical with a"\ All
xeBults concerning the theory o f isolated order can also be ex­
tended to the theory of identity, provided that the sequence of
sentences t„ is suitably defined. It is not difficult to formulate
the definition o f this sequence: the sentences ip ta, is>... are to
express respectively the idea that the number o f all elements is
not exactly equal to 1, 2, S,...; we can for example put:
‘NUa'Ila’ Ra'a m
i a = 'Ua'Ila’'CIla!nCNRa'a,:'Ra"a"'R<i'a"\

and so on. W e may mention yet another theory o f the same

structural type, namely the atomistic algebra of logic.1 Again,
it differs from the previously considered theories in only one
respeot, i.e. in having a different list of logical axioms. We call
the symbol ‘R ’ in this theory the inclusion sign: we now read
the formula *Ra'a*’ as ‘the dement a' stands in the relation of
inclusion to the element a"* or ‘the element a' is included in the
element a As logical axioms of the theory, apart from the
sentential functions which can be obtained by substitution into
the axioms of the sentential calculus, we adopt sentences which
suffice for the development o f the ordinary system of the algebra
o f logic (which, however, are formulated with the help o f the

x Cf. article X I , p. 334 of the present work. (Nevertheless, we have in mind

here an atomistic theory of the algebra of logic in which infinite operations do
not occur.) The structural properties of the theory in question can be derived
from the results in Skolem (64). In article V III, pp. 199-2081 have given the
most important metaznathematical theorems concerning the atomistic algebra
of logic (although there the problem is treated from quite a different point of
view).
X H , §3 CALCULUS OP SYSTEM S 379
inclusion sign as the only prim itive term )1 and in addition the
follow ing sentence:
l = 'na'CIla'CIta'a'N ptta'Ba'a''
where j9 is an abbreviation o f the sentential function
NCUa//V m a nM V C B a waffE ^ a m a mBaW ff;
‘/J’ expresses the fact that the element d ! is an atom (an in ­
decom posable elem ent), i.e. a non-em pty elem ent which includes
no non-em pty elem ent distinct from itself, and the sentence l
asserts that every non-em pty element includes at least one atom .
The definition o f the sequence i a for the atom istic algebra o f
logic resembles the cxxtTesponding definition for the elem entaiy
theory o f identity; the sentence in states that the num ber o f all
atom s is n ot exactly equal to n.
In conclusion we give examples o f deductive theories o f higher
structural types. I f in the postulate system o f the atom istic
algebra o f logic we replace the sentence l b y the logically weaker
sentence
Jx « <NnarCna%1^EaWNUa/f/GnaffCpM 0amIta'a'/,i9
(where ‘S' has its previous meaning), which states only that
am ong the elements which include all atom s there is a smallest
one, we obtain an exam ple o f a theory with the characteristic
pair (K0, 2). If, on the other hand, we sim ply delete the sentence
lythen the resulting theory, which is clearly the general algebra
o f logic, proves to have (K0, K0) as its characteristic pair. I f we
strike out the sentence c 4 from the axiom system o f the theory
o f dense order (or, what amounts to the same thing, the sen­
tences c6 and cQfrom the axiom system o f the theory o f isolated
order) and pass in this w ay to the general elementary theory of
order, then the structure o f the theory becom es enorm ously m ore
com plicated: while the theory o f dense order has one o f the
sim plest structural types, the general theory o f order has a very
com posite type, and its characteristic pair is the greatest pos­
sible one, nam ely (SK0, 2*«). I f we drop the sentences cl9 Cg, and
Cg from the axiom system o f the general theory o f order, we

: <.-• : K. V. ;32), and ¥111, p. 179.

380 F O U N D A T IO N S OF TH E xn, §5

obtain a new theory with the characteristic pair (X0J 2*»), namely
the general elemerda,ry theory of a binary relation. This theory
em braces in a certain sense all previously considered theories.
W ithin the discussion o f this theory all results sketched in this
section find their place. Thus, for exam ple, all results concern­
ing arbitrary deductive systems o f the theory o f dense order
can be regarded as results relating to those systems o f the. general
theory o f a binary relation which include the system

A. = On{{c^, c%} c3, c4}).

A p p e n d ix
In spite o f their elem entary character, the results stated in § 5
have a series o f interesting consequences extending far beyond
the lim its o f the present work. I will here discuss, although very
briefly and sketchily, some o f these consequences.
L et a be an arbitrary order type.1 Let us consider an arbitrary
set X and an arbitrary relation B which orders this set according
to the type a. L et us assume that the variables V ‘a*\ V ” ,...,
which appear in the elem entary theory o f order, denote the
elements o f the set X exclusively, and confine our attention to
the set o f all those sentences o f the theory o f order which are
true with the given interpretation o f the variables. It is not
difficult to see that this set depends neither on the set X nor on
the relation i?, but exclusively on the order type a; on that
account we denote it b y the sym bol ‘ T (a)\ I t can be shown
that— for any given order type a— the set T(<x) is a com plete
deductive system .
The form ula T(oc) = T(j3) states that all properties o f the
order types a and /? whioh can be form ulated in the language o f
the elem entary theory o f order are identical; for that reason we
call the types a and j8 which satisfy this form ula elementarily
indistinguishable or elementarily equivalent.

1 The remarks given below refer to ordering relations, bub they can be ex­
tended to arbitrary relations. W e must speak, then, not of types of order, but
of relation numbers in the sense of Whitehead, A. N ., and Russell, B. (90),
vol. II, pp. 30 fif.
X II CALCULUS OP SYSTEM S 381

I t follow s from the well-know n theorem o f Skolem-LOwen-

heim 1 that every order type is elem entarily equivalent to an
order type w hich is at m ost denumerable. Prom this it follow s
that every class o f order types o f which no tw o are elem entarily
equivalent can have at m ost the power 2K«, for this is the power
o f the class o f all denumerable types (this can also be proved
m ore sim ply w ithout the theorem o f Skolem -Low enheim ). On
the other hand, it is n ot difficult to construct a class o f elemen­
tarily distinguishable (non-equivalent) order types the power
o f which is actually 2Ko (in the construction we im itate the p roof
o f the set-theoretical theorem just m entioned, according to
which the class o f denumerable order types has the power 2Ko).2
I f we restrict ourselves to dense order types, we reach the
result that only four suoh elem entarily non-equivalent order
types exist; in fact, the set T(a) is always a com plete system ,
and in the elem entary theory o f dense order only four com plete
system s appear. As an exam ple o f four dense elem entarily non­
equivalent order types, the types 77, I + 77, 77+ I , I + tj+ I m ay
serve, where ‘ 77’ denotes as usual the type o f the natural order­
ing o f the set o f all rational numbers. B y using the sym bols
which were introduced in the discussion o f the theory o f dense
order it can be shown that
77) =ss D j-j-JJj, T{ I + 77) = D i+ D j,

T(v+ l ) = Wx+ D z, and r U + 77-H ) = 5 H -5 "2.

E very other type o f dense order is elem entarily equivalent to
one o f these fou r types. Thus, for exam ple, for the type A o f the
natural order o f the set o f all real numbers we have T{ A) = T(y).
Nevertheless, the types 77 and A are distinguished b y a whole
series o f properties: 77 is a denumerable type, A is n ot; A is a con­
tinuous type, 77 is not. From this it follow s that properties o f
order types such as denumerability and continuity are not elemen­
tary properties, in the sense that they cannot be expressed in the
language of the elementary theory of order (as far as denum erability
is concerned this also follow s from the theorem ^ f Skolem -
Lowenheim ).
1 Cf. Skolem, Th. (06), pp. 23 ff.
* Cf. Hausdorff, F.(25), pp. 49 f.
382 F O U N D A T IO N S OF TH E X II

In a similar manner we reach the result at there can be only

a denumerable number o f isolated order types (i.e. order types
having no proper cuts, but only jumps and gaps) o f which no
two are elementarily equivalent. Let us, in fact, consider the
class consisting o f all finite types as well as the types to, to*,
and «+ «> *, where ‘to’ , as usual, denotes the order type
o f the less-than relation in the domain o f natural numbers. It
is easy to see that no two types o f this class are elementarily
equivalent. It can also be shown that
T(to*) = Tlt T(to) = 4 , T(to*+to) = %, T{n) = 1 ^
for every natural n ^ 2 and finally that
T(to-{-<o*) — »
Since in this w ay all com plete system s o f the theory o f isolated
order are exhausted (cf. § 5), every other isolated order type
m ust be elem entarily equivalent to one o f the types considered.
Thus, e.g., T(to) = T{to+to*+toy, on the other hand, to is the
type o f a well-ordered set while w -j-w ^-fw is n ot. H ence it
follow s that well-ordering ia not an elementary property of the
order typee, it cannot be expreaeei in the elementary theory o f order.
It seems that a new, wide realm o f investigation is here opened
up. It is perhaps of interest that these investigations can be
carried out within the framework o f mathematics itself (e.g.
set theory) and that the concepts and methods o f metamathe-
matios are essentially superfluous: all ooncepts which occur in
these investigations (e.g. the concept of an elementary property
o f an order type or the oonoept o f elementarily indistinguishable
order types) can be defined in purely mathematical terms.1

N ote. In the present article I have brought together results

w hich have been obtained at various tim es.
The results in §§ 1 -4 belong, for the m ost part, to the years
1930-2. I reported on them in 1933 to the W arsaw Seotion
o f the Polish M athem atical S ociety in a lecture entitled ‘ On
1 For this purpose it suffices to apply the method which I have developed
in V I. For precise mathematical definitions of the notions involved and for an
outline of the general theory of these notions see Tarski, A . (84a).
xn CALCULUS OP SYSTEM S 883

the foundations o f the m ethodology o f deductive sciences’ (cf.

M azurkiewicz, S. (53) where— m aking use o f the calculus o f
systems— a new m ethod for the foundation o f probability theory
is developed).
The results o f § 5 had already been obtained in the years
1926-8, naturally w ithout applying the apparatus o f the cal­
culus o f systems, which was n ot at m y disposal at that tim e.
(Cf. V III, p. 205, note 2, as well as Presburger, M. (61)).
During the same period I becam e aware o f the facts discussed
in the appendix. B ut I was able to state these facts in a correct
and preoise form only with the help o f the m ethods which were
developed in V I and V III, during the years 1929-30. The facts
discussed in the appendix, together w ith other results o f a re­
lated kind, were com m unicated b y me in 1932 in a lecture at
the W arsaw Section o f the Polish M athem atical Society.
 XIII
ON T H E L IM IT A T IO N S OF TH E
M EANS OF E X P R E S S IO N OF
D E D U C TIV E TH E O R IE S!

T he m etam athem atical results presented in this article are

applicable to various systems o f m athem atical logic and to
various deductive theories based upon these systems o f logic,1
but in order to make the follow ing remarks more specific, we
adopt as logic an arbitrary system which satisfies the follow ing
tw o conditions:
(a) The system includes as a subsystem the logic o f Principia
Mathematica which is assumed to be strictly form alized and to
be m odified as follow s:
(1) a sim ple theory o f types and the axiom o f extensionality
are assumed;
(2) all defined operators, e.g, 'x^x)' or ‘ (?o*){(j>x)\ are elim i­
nated;
(3) on ly those sentential functions which contain no free
variables are regarded as sentences.
(b) The system possibly contains still further logical sentences
(which do n ot occur as axiom s or theorems in Principia Mathe-
matica)> as, for exam ple, the axiom s o f infinity and o f choice,
etc., but no new undefined constants appear in it.
1 I announced Th. 9 to the Second Polish Philosophical Congress in Warsaw
in 1927 in a lecture on the concepts of consistency a n d completeness; the result
is mentioned in Fr&nkel, A . (16), p. 362, note 3. Other results of the present
communication -were first presented at the sessions of the Logical Section of
the Warsaw Philosophical Society on 15 June 1932 (Lindenbaum: On some
methodological questions connected with the foundations of Geometry, I) and on
1 February 1933 (Lindenbaum and Tarski, On some metalogical theorems). To
the same oircle of ideas belong Tarski, A. and Lindenbaum, A. (85) and article
X of the present work.

f B ibxjograjphical N ote . This article is the text of a lecture given by the

author on 12 June 1935 to the Mathematical Colloquium conducted by Karl
Menger in the University of Vienna. The article first appeared under the title
‘ tTber die Beschr&nktheit der Ausdrucksmittel deduktiver Theorien’ in
Jfirgebnisse eines mathematischen Kolloquiums, fascicule 7 (1934—5), pp. 15-22.
For further historical references see footnote 1.
x in E X P R E S S IO N OF D E D U C T IV E T H E O R IE S 385

F or other concepts and sym bols used in the follow ing dis­
cussion see X , pp. 297-8. In particular, we shall em ploy sym ­
bolic expressions o f the form
x', 2',...
B
y\
(explained in X , p. 310), which express the fact that the relation
B maps the class o f all individuals onto itself in one-one fashion,
so that the individuals, classes, relations, etc., x\ y\ 2',... are
m apped on x *, y", z",..., respectively. As a general scheme o f a
sentential function the sym bolic expression 6, c ,...; x, y t 2,...)9
is used; here V , (b\ V ,... are all extra-logical constants and
V , y , are all free variables o f the given sentential func­
tion. The conjunction o f all axiom s o f a deductive axiom atizable
system ,1 in which V , ‘c ’ ,... appear as prim itive extra-logical
constants, is represented in the schem atic form ‘a(a, 6, c,...)\

§1
T heorem 1. Every sentence of the form

(as', as", y ’ , y ", z 2',..., B ) :. B :o-(as',i/',z',...).

y\
s M S ,i f , * ',...)
is logically provable.
R oughly speaking, Th. 1 states that every relation between
objects (individuals, classes, relations, etc.) which can be ex­
pressed b y purely logical means is invariant with respect to
every one-one m apping o f the 'w orld5 (i.e. the class o f all in­
dividuals) onto itself and this invariance is logically provable.
The theorem is certainly plausible and had already been used
as a premiss in certain intuitive considerations, iSTevertheless
it had never before been precisely form ulated and exactly
proved. The p roof o f Th. 1 is carried out by induction with
respect to the form o f the sentential function 'o '(a /,y ', 2',...)\ 2
It should be noted that the sym bol V which occurs in the

1 See article V of the present work.

3 A detailed proof of Th. 1 has recently been published in Mostowski, A.
(53d), pp. 200 £f. The formulation of the theorem in Mostowski’s work is,
however, weaker than that given in the present article.
386 L IM IT A T IO N S OF TH E MEANS OF x m ,§ 1
schemata o f sentential functions must not be regarded as an
independent sym bol (say as a variable predicate). For example,
in contrast to Th. 1, sentences o f the form

x ,y ,z ,...
~ .f(x tt}y \ zn,...)
are, in general, not provable but refutable in logic. On the other
hand, b y virtue o f Th. I sentences o f the follow ing form are
provable:
{x\ x",y'>y* ,z\ z\ ...,R j\ f")-..R x',y', * W '
x\ y\ s ', . . . , / '’
.^ :f\ x \ y \ z f^ ) ^ r { x \ y \ z \ . . . y
I f we apply Th. 1 to sentential functions w ith a single free
variable, and in fact an individual variable, we obtain as an
im m ediate corollary
T h eorem 2. I f 6x ” and ‘x fn are individual variables, then every
sentence of the form
(x',x'') :a(a;'). == .a(ce*)
is provable in logic.
Intuitively interpreted: no tw o individuals are distinguishable
b y purely logical means. B y an easy transform ation we obtain
from Th. 2
T h e o r e m 3. If ‘x ’ is an individual variable, then every sentence
of the form (* ).„ (* ). v ~
is logically provable.
In other words, apart from the universal class and the em pty
class, no class o f individuals can be defined by purely logical
means.1 I t should be noted that the follow ing sharper form ula­
tion o f Th. 3 in general fails: I f ‘ x 9is an individual variable, then
for every sentential function ^o■(a;), either the sentence ((x).cr(x)’ or
1 The words ‘ define’, ‘ definable’, eto., are used in two distinct senses: in
the first sense it is a question of a formal relation of certain expressions to other
expressions of a theory (cf. article X of the present work); in the second sense
of a semantical relation between objects and expressions (cf. article V I). Here
these words are used in the second sense.
XIII, §1 E X P R E S S IO N OF D E D U C T IV E T H E O R IE S 387

the sentence ‘ (a ). ~ cr(x)’ is provable in logic. This holds if and

only if the basic system o f logic is com plete.
I f we now apply Th. 1 to sentential functions with tw o free
variables, we obtain
T h eo rem 4. I f ‘x* and ‘y ’ are individual variables, then every
sentence of the form
{x, y ) . a{x, y ) . V . (x, y ) . ~ a(x, y ) . v : (x, y ) : o{x, y ) .
== .x = y : v : ( x , y ) : a ( x , y ) . == .x ^ y
is logically provable.
Consequently only four binary relations between individuals
are definable b y purely logical means, nam ely the universal
relation, the em pty relation, identity, and diversity (thus the
four m oduli o f Schroder’s algebra o f relations).
This theorem can be extended to m any-term ed relations: for
every natural number n only a specifiable finite number o f n-
term ed relations between individuals can be defined b y purely
logical means, and each o f these relations can be expressed by
means o f identity and the concepts o f the sentential calculus.
Y et another special case o f Th. 1 will be stated here concern­
ing sentential functions with one free variable denoting a class
o f individuals; ‘Nc(x)9 will denote the cardinal number (in the
sense o f Principia Mathemaiica) o f the class x and ‘x 9will denote
the com plem ent o f x.
I f ‘x'9and cx”9are variables ranging over classes of
T h e o r em 5.
individuals, then every sentence of the form
(x',x"):.Nc{x') = Nc(xn).Nc{xf) = Nc[ST) : d :a{x'). == .* (* ')
is logically provable.
Thus, no tw o classes o f individuals which have equal cardinal
numbers and whose com plem ents also have equal cardinal num ­
bers, are distinguishable b y purely logical means.1
It is custom ary to say that our logic is a logic o f extensions
and n ot o f intensions, since tw o concepts with different inten­
sions but identical extensions are logically indistinguishable.
1 That is the solution of a problem which was first formulated in Tarski,
A. (75), p. 182.
388 L I M IT A T IO N S OF TH E MEANS O F x in , { I

In the light o f Th. 5 this assertion can be sharpened: our logic

is not even a logic o f extensions, but m erely a logic o f cardinality,
since tw o concepts w ith different extensions are still logically
indistinguishable i f only the cardinal numbers o f their exten­
sions Are equal and the cardinal numbers o f the extensions o f
the com plem entary concepts are also equal.
W e pass now to the applications o f Th. 1 to special deductive
theories, and to make the discussion m ore speoific we concen­
trate upon Euclidean geom etry. It is well known that Euclidean
geom etry can be constructed b y means o f a single prim itive
concept, e.g. b y means o f the four-term ed relation o f the con­
gruence o f tw o pairs o f points. W e shall denote this relation by
V . The expression ‘a(x, y, z, t) * will thus be read: 'the point x
is the same distance from the point y as the point z is from the
point t\ W e also assume that the axiom system o f Euolidean
geom etry has been supplem ented b y a sentence to the effect
that every individual is a point (so that space is identified with
the class o f all individuals).1 As a special case o f Th. 1 we obtain
at once
T heobem 6. Every sentence of the form
a, z\ y\ s',...
{x',x*>y'>y\s',zV., R ) : . R a, x*, y\ z\...
d :or(q;a',y',z',...).

=== .a {a ;x ”,y ',z ”,...)

is provable in Euclidean geometry.
Intuitively speaking, every relation between objects (points,
sets o f points, classes o f sets o f points, etc.) which can be ex­
pressed in terms o f logic and geom etry is invariant with respect
to every one-one m apping o f space onto itself in which the relation
a is preserved, i.e. with respect to every sim ilarity transform a­
tion (and not only w ith respect to every isom etricai transform a­
tion).
Using the specific properties o f the axiom s o f Euclidean geo­
m etry, it is easy from Th. 6 to derive analogues o f Ths. 2-4.
Thus, b y means o f logic and Euclidean geom etry no tw o points
can be distinguished, neither can any specific set o f points,
1 Cf. the concluding remarks of this article.
X H I, $ X E X P R E S S I O N OF D E D U C T IV E T H E O R IE S 389

apart from the w hole space and the em pty set, be defined, nor
any tw o-term ed relation betw een points, apart from the uni­
versal relation, the em pty relation, identity, and diversity (in
particular, in geom etry no direction on a straight line can be
singled out). The distinction between logic and geom etry com es
to light, how ever, in the discussion o f three-term ed relations;
for, as is well known, there are infinitely m any three- (and
m any-) term ed relations which are definable in Euclidean
geom etry.
In the tw o follow ing corollaries o f Th. 6 the form ulas V ~ x "
(‘the point sets x f and x ” arc, geometrically similar9), ‘x r ~ x "9
( ‘the point sets x* and x* are congruents), and *$(#)’ (*x is a line
segment9) appear, w hich are known to be definable b y means
o f 'a9.

Theorem 7. I f V* and *x*9 are variables ranging over point

setSy then every sentence of the form
(x\x*) \.xr ~ x". d : a(a; xf). == .a(a\ s ')
is provable in Euclidean geometry.

T heorem 8. Under the same hypothesis, every sentence of the

form
(jx') : S(x r) . a(a; x f) . 3 . (3s*). S{x * ). a(a; z ”) . ~ (s ' ~ s*)
is provable in Euclidean geometry.
‘Thus, no tw o similar point sets can be distinguished b y means
o f logic and Euclidean geom etry, and no absolute unit o f length
is definable within E uclidean geom etry.
Various results of this kind can also be reached by investi­
gating other deductive theories, e.g. projective geometry and
topology. Th. 6 holds generally for every theory based on logiCy
because no use is m ade in its p roof o f specific properties o f the
axiom s o f geom etry. N ot only Ths. 2-4, but also the analogue
o f Th. 4 for three-term ed relations, can be carried over to p ro­
jective geom etry; in fa ct there are only finitely m any three-
term ed (although infinitely m any four-term ed) relations between
points which can be defined in this geom etry (in particular the
390 L IM IT A T IO N S OF TH E MEANS OF X H I, § 1

relation o f lying between is not definable in projective geo­

m etry). In topology only those tw o- and m any-term ed relations
between points are at all definable which can be defined by
purely logical means.
§2
W e shall here state tw o general m etam athem atical theorems
as a further application o f Th. 1.
As is well known, an axiom system ‘c ^ a ^ c ,...)’ is said to be
categorical or monomorphic if the sentence
(*',a ', y \ y \ :«(*', y \ .«(*', y \ 2",...)

o.(3
& j y i % >•••
is logically provable. The axiom system is said to be non-rami-
fiable if, for eveiy sentential function ia(xyy }z,...)\ the dis­
junction
{x, y>2,...): <x{x9y %2 ,...). 3 . o{x , y 92,...):
: V : {x,y , 2,...): ot{x9y, 2,...). 3 . ~ o{x ,*/, 2,...)
is logically provable. If, on the other hand, for every sentential
function a member o f this disjunction is logically provable, then
the system is decision-definite or complete} B y the application
o f Th. 1 we can now prove
T heorem 9. Every categorical axiom system is non-ramifioMe.
F or exam ple, the axiom atically constructed arithm etic o f
natural numbers is categorical and thus non-ram ifiable, It is
also incom plete, i.e. there are arithm etical sentences which can
be neither proved nor disproved b y means o f the axioms o f
arithm etic.2 On account o f the non-ram ifiability the indepen­
dence p roof for such sentences cannot be carried out in any
direction b y an interpretation in logic.
The converse o f Th. 9 proves to hold under a supplementary
assum ption. W e define an axiom system *a(a, b, c ,...)’ to be
1 For these concepts cf. Fraenkel, A. (16), pp. 347-54 (especially p. 353).
There are certain points of contact, although of a superficial kind, between
§ 2 of this article and Carnap, R . (9).
2 Cf. Godel, K . (22), pp. 174 f. and 182 ff.
X III, § 2 E X P R E S S IO N OF D E D U C T IV E TH E O R IE S 391

effectively interpretable in logic if there is in logic a sentential

function *r(x, y, z ,...)’ such that the follow ing three formulas are
logically provable:
(x, y , 2,...): t (x , y , z,...). d . <x(x, y ,z,...);
{3x,y,z,...) .r(x,y,z),
( * ', x\y\y% z', J,...):r(x',y\z',...) . t (x " , y \ .
D .x ' = x” .y' = a/'.z' = z "....
I f we take into consideration the rules o f definition usually
adopted in logic, this form ulation can be considerably sim plified:
the axiom system ‘a (a,b,c,...y is effectively interpretable in
logic if and only if there are logical constants V \ ‘6 °, V ’ ,...
(undefined or defined) such that the sentence °b -
tained b y substituting these logical constants for the prim itive
terms in the axiom system discussed is logically provable.
T heorem 10. Every non-ramifiable axiom system which is
effectively interpretable in logic is categorical.
It sometimes happens that an effective interpretation o f an
axiom system in logic can be found only at a higher type (this
applies, for exam ple, to the axiom system o f the arithm etic
o f real numbers). It is then also possible to infer categoricity
from non-ram ifiability, under certain conditions which are
satisfied b y the fam iliar logical systems. W e do not know,
how ever, whether the converse o f Th. 9 holds quite generally.
In all such considerations it is im portant to bear in mind the
relativity o f the three concepts— categoricity, non-ram ifiability,
and completeness, w ith respect to the adopted system o f logic.
J /, for exam ple, the logic is complete, then the concepts of non-
ramifiability and completeness have the same extension for every
axiom system, so that every categorical system is (not only non-
ram ifiable but also) complete. The three concepts discussed
were here introduced as a-concepts in the sense o f Carnap. I f
we consider the corresponding f-con cepts,1 then we find that f-
com pleteness and f-non-ramifiabilifcy have the same extension;
1 Concerning a- mod f-ooncepts, see Carnap, R. (10), pp. 123 ff.
392 L IM IT A T IO N S OF D E D U C T IV E T H E O R IE S XTTT, § 2

the analogue o f Th. 9 proves to hold, bu t the problem o f its

converse is open in this case as w ell.
In conclusion we note the follow ing: when applying Th. 1 to
special deductive theories and to general m etam athem atics we
m ust restrict ourselves to those axiom system s from which it
follow s that there are no individuals outside the dom ain o f dis­
course o f the theory discussed. W e could rem ove this restric­
tion b y means o f an appropriate generalization o f Th. 1. T o
obtain this generalization we should introduce a new notion, in
fa ct the notion o f a sentential function intrinsic under a given
variable (or constant) ‘o ’ or, in other words, the notion o f a
property intrinsic for a given doss a. R oughly speaking, a pro­
perty is intrinsic for a class a i f it involves exclusively elements
o f o, subsets o f o, relations betw een elem ents o f o, eto. (and
not, for exam ple, individuals outside o f o). W e do not give
here a precise definition o f the new notion, nor do we form u­
late the generalization o f Th. 1 in which this notion is essen­
tially involved. W e should like to m ention only that, w ith
the help o f this notion, various new results— o f the same
general character as those previously discussed— can be form u­
lated and established. Thus, for instance, it can be shown that
every property o f a point set x, which is expressed in terms o f
logic and topology, and is intrinsic for x, is invariant under any
hom om orphic transform ation (o f the set x into another set a:').1
1 Of. Kuratowski, C. (41), p. 13.
 XIV
ON E X T E N SIO N S OF IN C O M P LE TE
SYSTEM S OF TH E SE N T E N T IA L CALCULUSf

A ccording to a well-known theorem o f A . Lindenbaum ,1 every

consistent set o f sentences o f an arbitrary deductive theory
can be extended to form a consistent and com plete system . The
question arises how m any such extensions are there; in other
w ords, how great is the number o f all the consistent and com ­
plete system s which include a given set o f sentences ? I shall
here prove a general theorem from the m eta-sentential calculus
b y which under certain assum ptions only a single extension
exists in the dom ain o f the sentential calculus, I shall show
further that m ost o f the incom plete systems o f the sentential
calculus so far treated satisfy these assumptions, so that the
ordinary (tw o-valued) system o f the sentential calculus consti­
tutes the only consistent and com plete extension for all these
system s.
I t is well known that in the expressions o f the sentential
calculus special variables, the so-called sentential variables
p,q, r,...,occu r, and in addition the implication sign G and possibly
still other constants, e.g. the negation sign N> the signs of dis­
junction A , and conjunction K , and so on .2 Correlated with
each constant is a fundamental operation on expressions b y which
com pound expressions can be form ed from simpler ones. F or
exam ple, corresponding to the im plication sign is the operation

1 See Article V„ Th. 60, p. 98 of the present book.

a As regards symbolism of. IV . Concerning terminology and the concepts
involved see the foregoing and article V of the present book. It should be
noted that the signs ‘p \ *q\..., *0*, ‘ jV’ ,... and the expressions composed of
these signs are not used here as expressions of the sentential calculus, but as
designations of the corresponding expressions.

f B iblioqraph icajl N oth . This artiole contains the text of a lecture given
by the author in 1936 to the Mathematical Colloquium conducted by Karl
Monger in the University of Vienna. The article first appeared under the title
‘ tTber die Erweiterungen der unvollst&ndigen Systeme des Aussagenkalktils *
mJUrgebniese eines mathematiecfien Kolloquiums, fascicule 7 (1934-6), pp. 61-07.
394 ON E X T E N S IO N S OF IN CO M PLETE XIV

o f forming an implication, b y means o f which from tw o given

expressions a and b the im plication Gab is form ed. The inter­
section o f all sets o f expressions which contain all the sentential
variables and are closed under each o f the fundam ental opera­
tions is called the set of (meaningful) sentences and denoted
b y 8.
Tw o operations are further distinguished which correlate
certain sentences with other sentences, nam ely the operations
o f substitution and detachment I t would be somewhat tedious
and superfluous here to explain m ore precisely the intuitively
obvious meaning o f the concept o f substitution. The set o f all
sentences which can be obtained from the sentences o f a given
set X b y substitution is denoted b y (Sb(X)\ More generally, we
denote b y ‘Sbr (X )’ the set o f all sentences which are obtained
b y replacing all variables in the sentences o f the set X b y sen­
tences o f the set T (in such a w ay that variables o f the same
shape whioh occur in a given sentence are replaced b y sentences
o f the same shape). The operation o f detachm ent correlates the
sentence b with tw o given sentences o f the form a and Gab.
I f X is any set o f sentences, then the intersection o f all sets o f
sentences which include the set Sb(X) and are closed under the
operation o f detachm ent is called the set of consequences of X ,
in sym bols Cn(X ).
B y means o f the concepts o f sentence and set o f consequences
other m etalogioal concepts can be defined. Thus, for example,
a set X o f sentences is called a deductive system , or sim ply a
system , if Gn(X) = X . The set X o f sentences is called incon­
sistent or consistent according to whether the form ula Cn(X) = S
holds or not. The set X o f sentences is called complete if every
sentence a either is a consequence o f the set X or, when added
to X , yields an inconsistent set More generally we call
the set X o f sentences complete with respect to a set Y of sentences
if for every sentence b o f the set Y either b e Gn(X) holds or
the set X + { 6} is inconsistent.
Various elem entary properties o f the above concepts $re
assumed to be know n .1 In addition to the sym bols listed above
1 Several such properties are established in article V of the present book.
XIV SYSTEM S OP TH E S E N T E N T IA L CALCULUS 395

I shall use the custom ary sym bolism o f set theory. The general
theorem m entioned at the outset is as follow s:
T heorem 1. Let X be a consistent set of sentences, which
satisfies the following condition: there is a set Z of sentences such
that X is complete with respect to the set Sbz (S) and* the set
X + S b z (S) is inconsistent. Then there exists exactly one con­
sistent and complete deductive system Y , which includes the set X .
F or the p roof o f this theorem we require certain lemmas.
L emma 1. For every set X of sentences Sbx {S) is the smallest
set of sentences which includes X and is closed under every funda­
mental operation.
The p roof—based on an exact definition o f the concept o f
substitution— presents no difficulties. As a corollary we easily
obtain:
L emma 2. For every set X of sentences
X s Sbx (S) = S b ^ i S ) .
L emma 3. For all sets X and Y of sentences
SbY(Cn(X)) c: Cn(SbY(Sb(X))).
Proof. Let Z be the set o f all sentences a which satisfy the
form ula $ 6r ({a}) ^ Cn(SbY{Sb{X))). I t is easily shown that
8b(X) £ Z and that Z is closed under the operation o f de­
tachm ent (i.e. if a e Z and Gab e Z } then b e Z). From this it
results im m ediately that Cn(X) ^ Z, in other words that
SbY({a}) £ Cn(SbY(Sb(X))) for every sentence a e Cn{X ); and
thus finally SbY{Cn{X)) c: Cn(Sbr (Sb{X)))9 q.e.d.
I f in Lem m a 3 we put Y = S, then by a simple transform a­
tion we obtain the form ula Sb(Cn(X)) c Gn(X) for every set
X o f sentences. B y virtue o f this form ula the definition o f the
set o f consequences given above can be replaced b y the follow ing:
Gn(X) is the intersection o f all sets o f sentences which include
X and are closed under the operations o f substitution and
detachm ent.

L emma 4. I f the sets X + Y and X-\-Z of sentences are incon­

sistent, then the set X-\-Sbz (Sb(Y)) is also inconsistent.
396 ON E X T E N S IO N S OF IN CO M PLETE X IV

Proof. I f, in Lem m a 3, w e replace 'X ’ b y ‘X -f-F ’ and *F’ b y

‘Z ’ w e obtain 8bz (G n (X + Y )) £ Cn{8bz {Sb{X +Y))). Since,
b y hypothesis, C n (X + Y ) — S, it follow s that
Sbz (8) £ Cn{Sbz {8b{X + Y ))),
whence, b y Lem m a 2 , Z £ Gn(8bz (Sb(X +Y ))) and further
(1) X + Z £ X + C n (8 b z {Sb(X+Y))).
d e a r ly w e have
8bz (8 b (X + 7 )) = Sbz (Sb(X ))+8bz (Sb(Y))
£ Sb{X)+Sbz {8b(Y)) £ C n {X )+ 8 bz (Sb{Y)y,
hence b y applying the form ula Gn(Cn{X)+Y) ~ Cn(X + F )1
we derive
Cn{Sbz {S b (X + Y )) £ Cn{Cn(X)+8bz (Sb(Y)))
= C n(X+Sbz { S b ( m •
Since in addition X £ X + 8 b z (Sb(Y)) £ C n (X +8bz {Sb(Y))),
w e have
( 2) X +<7»(tffez (/S 6 (X + F ))) £ C n (X + 8 bz (Sb(Y))).
The inclusions ( 1) and ( 2) at once give
X + Z £ 0 » (X + S 6 Z(S 6 (7 )));
since the set X + Z is inconsistent, Gn(X+Sbz )(Sb(Y))) is
also inconsistent .3 From this it follow s im m ediately that
X + S b z (Sb(Y)} is inconsistent, q.e.d.
Lemma 5. I f X and Y are consistent deductive systems and if
X .Y is complete with respect to a set Z of sentences, then we have
X .Z = Y .Z .
Proof. L et a be any sentence o f the set X . Z. Then we have
X .F + { a } £ X , and the set X . F + { o } is thus consistent. Since
X .Y is com plete w ith respect to Z and a e Z , it follow s that
a e C n(X.Y). The set X . Y, as the intersection o f tw o systems,
is itself a system , i.e. Gn(X .Y) = X.Y*-, consequently a e X . Y .
Thus the form ula a e X .Z always im plies a e X . Y . A ccord­
ingly, X . Z £ X . Y , X . Z = X .Y .Z . In an analogous maimer
it can be proved that Y .Z — X .Y .Z , whence finally we have
X .Z = Y .Z , q.e.d.
1 Cf. V, Th. 2, p. 66. * Cf. V, Tbu 46, p. 91. * Cf. V, Th. 11, p. 71.
X IV SYSTEM S OP T H E S E N T E N T IA L CALCULUS 397

W e turn now to the p ro o f o f Th. 1. A ccording to the theorem

o f Lindenbaum already m entioned, the consistent set X o f
sentences can he extended to at least one consistent and
com plete system Y. L et ns suppose there is another such system
distinct from Y. Since Y and Yx are com plete, the set T of
sentences m ust be inconsistent.1 B ut since
X + S b z (S) £ Y + S b z (S)
and X + S b z(S ) is inconsistent, Y-\-Sbz (S) m ust also be incon­
sistent. H ence it follow s b y Lem m a 4 that
( 1) the set Y SbSbl^ )(Y1) is inconsistent.
M oreover, w e have
S b ^ iY y ) £ Sb(Yx) £ Gn(Yx) = Yx
since Yx is a deductive system . B y Lemm a 2,
SbmM £ S b ^ S ) = Sbz (S),
whence
^ s* ts> P i) c Yx. Sbz (S) and Y + S b ^ Y J £ Y + Y x. Sbz (S).
The last inclusion, together w ith ( 1), shows that
(2 ) the set 7 + 3 ^ . 8bz (S) is inconsistent.
W e also have X £ Y and X £ 3^; consequently X £ Y.YV
H ence the set Y.YX, like the set X , is com plete with respect to
8bz (S). B y means o f Lem m a 5 we conclude from this that
Y.Sbz (S) — Yx.Sbz (S), and that consequently
(3) Y = Y + Y x.Sbz (S).
From { 2) and (3) it follow s at once that Y is inconsistent,
contrary to our assum ption. H ence the system Yx cannot be
distinct from Y. Th. 1 is thus proved.
In order to apply this theorem to special system s, let us
consider first certain system s o f the so-called restricted senten­
tial calculus, i.e. the theory in w hich the im plication sign is the
only constant .8
T heobem 2 . I f X is a consistent set of sentences which contains
the three sentences CpCqp, CpCOpqq, and CCqrOCpqCpr as
elements, then ihe ordinary (two-valued) system o f the restricted
sentential calculus is the only consistent and complete deductive
system which includes the set X .
1 Cf. V , Th. 63, p. 97. a Cf. IV , pp. 52-4.
398 ON E X T E N S IO N S OF IN CO M PLETE XIV

L emma 6. I f X 0 = { CpCqp, CpCCpqq, GCqrCCpqCpr} and

Z = {jp}, then for every sentence a e 8bz(S) we have either
a e Cn{X0) or both Cap e Cn(X0) and Cpa e Cn{X0).
Proof. L et Y be the set o f all sentences which satisfy the
conclusion o f the lemma. B y applying the usual m ethods o f
the sentential calculus it can be shown w ithout difficulty that the
sentence Cpp is a consequence o f the set X Q. Hence it follows
im m ediately that for a e Z, i.e. for a = p, both Cap e Cn{X0)
and Cpa e Cn(X 0), whence a e Y . Consequently Z <^Y. N ext
consider any tw o sentences a and 6 o f the set Y ; we shall show
that Cab e Y also holds. Here three cases are to be distin­
guished: ( 1) b e C n ( X Q)\ ( 2) Cap e Cn{X0), C p a eC n (X Q)>
Cbp e Cn(X0), and Cpb e Cn(X0); (3) a e Cn(X0), Cbp e Cn(X0),
and Cpb e Cn(X0).
F or the sake o f brevity only the first case will be investigated
in detail. Since CpCqp e X 0, we have CbCab e Sb(X0) and hence
CbCab g Cn(X0); since m oreover b e Cn(X0) and the set Cn(XQ)
is closed under the operation o f detachm ent, we obtain
Cab e Cn(XQ). A nalogously it can be shown that in case ( 2) we
also have Cab e Cn{X 0), whilst in case (3) we have both
CCabp e Cn(X0) and CpCab e Cn(XQ).
Thus in each case Cab e Y. Consequently we have established
that Y includes the set Z and is closed under the operation
o f im plication form ation, i.e. the only fundamental operation
o f the restricted sentential calculus. A ccording to Lemma 1
it follow s that Sbz (8) £ Y ; and thus for every sentence
a e Sbz (S) we have either a e Cn{XQ) or Cap e Cn(XQ) and
Cpa e Cn(X0), q.e.d.
L emma 7. I f X 0 = { CpCqp, CpCCpqq, CCqrCCpqCpr} and
Z = £p}, then (a) the set X Qis complete with respect to the set 8bz (8)
and (b) the set X 0+ 8 b z (S) is inconsistent.
Proof, (a) easily results from the previous lemma. In fact
for every sentence a e 8bz (S) we have either a e Cn(XQ) or
Cap e Cn(X0); thus, if a does not belong to the set Cn(X0), then
the set X 0+ { a} is inconsistent, since its set o f consequences
contains both the sentences a and Cap, hence also the sentential
XIV SYSTEM S OF T H E S E N T E N T IA L CALCULUS 399

variable p , and consequently all possible sentences as elements

(for every sentence can be obtained b y substitution from the
variable p ). F or analogous reasons (6) also holds.
Proof of Theorem 2. I t is a well-known fa ct that the ordinary
system L o f the sentential calculus (which can be defined, for
exam ple, b y means o f the fam iliar tw o-valued m atrix) is con­
sistent and com plete 1 and contains the sentences CpCqp,
CpCCpqq, and CCqrCCpqCpr as elements. H ence it follow s that
the set X 0 which consists o f these three sentences is consistent.
This fa ct, together w ith Lem m a 7, perm its us to apply Th. 1
and in this w ay to conclude that L is the only consistent and
com plete system which includes X 0. I f now X is any consistent
set o f sentences such that X 0 £ X , then, according to Linden-
baum ’s theorem , X can be extended to a consistent and com ­
plete system T. Since X 0 £ Y, Y cannot be distinct from L.
Consequently L is the only consistent and com plete system
which includes X as a sub-set, q.e.d.
From Th. 2 we im m ediately obtain the follow ing

The ordinary system of the restricted sentential cal-

Co r o l l a r y .
culus is the only consistent and complete system which includes the
sentences CpCqp, CpCCpqq, and CCqrCCpqCpr as elements.
Analogous corollaries can be derived from Ths. 3 and 4 given
below.
I t should be noted in connexion w ith Th. 2 that the class
o f all system s o f the sentential calculus which contains the
sentences CpCqp, CpCCpqq, and CCqrCCpqCpr is fairly
com prehensive. F or exam ple, to this class belong all many­
valued systems o f the sentential calculus which originate with
J. Lukasiewioz ,2 as well as the positive system o f the sentential
calculus originated b y D . H ilbert .3 Each o f these systems can
therefore be extended in only one way to a consistent and com ­
plete system .
1 Cf. rv , pp. 43 ff.
a Cf. IV , pp. 47 ff.
* The set of all consequences of those axioms of Hilbert for the sentential
calculus which contain no negation sign is called the positive system of the
sentential calculus, see (29a), p. 153.
 400 ON E X T E N S IO N S OF IN C O M P LE TE SYSTE M S XIV

W e turn now to those system s o f the sentential calculus in

w hich tw o constants oooux, the im plication sign and the negation
sign. F or these we have
Theobem 3. I f X is a consistent set of sentences which contains
the four sentences CpNNp, CqCpq, GNpCpq, and CpCNqNGpq
as elements, then the ordinary (two-valued) system o f the sentential
calculus is the only consistent and complete system which includes
the set X .
The p ro o f is com pletely analogous to that o f Th. 2 and is based
on the follow ing lem m a w hich corresponds exactly to Lemmas
6 and 7 :.
Lemma 8 . Z /X 0 = {CpNNp, CpCqp, GNpGpq, CpGNqNGpq}
and Z — { CpNNp }, then: (a) for every sentence a e Sbz (S) either
a e Cn(X0) or Na e Cn(X0), (b) X 0 is complete with respect to the
set 8bz (S), and (c) the set X 0- r Sbz (S) is inconsistent.
F inally w e oonsider system s o f the sentential calculus in
which, in addition to the signs o f im plication and negation, tw o
other constants occur, the signs o f disjunction and conjunction:
Theobem 4. I f X is a consistent set of sentences which contains
thefollowing ten sentences as elements: CpNNp, CqCpq, CNpCpq,
CpCNqNGpq, GpApq, CqApq, CNpCNqNApq, CpCqKpq,
CNpNKpq, and CNqNKpq, then the ordinary (two-valued) system
of the sentential calculus is the only consistent and complete system
which includes the set X .
T o the system s X which satisfy the hypothesis o f Th. 4 (or 3)
belong again all m any-valued system s o f Lukasiewicz and also,
am ong others, the intuitionistic system o f the sentential calculus
o f A . H eyting .1 F or each o f these system s the ordinary tw o­
valued system constitutes the only consistent and com plete
extension .8
1 Heyting, A . (28).
a This result was previously obtained for the many-valued systems of the
sentential calculus by another method, cf. IV , p. 49, Th. 21 and the accom­
panying footnote.
 XV
THE E STA B LISH M E N T OF
SC IE N T IF IC SE M A N TIC S!

The w ord ‘semantics ’ is used here in a narrower sense than usual.

W e shall understand b y semantics the totality o f considerations
concerning those concepts which, roughly speaking, express
certain connexions between the expressions o f a language and
the objects and states o f affairs referred to b y these expressions.
A s typical exam ples o f semantical concepts we m ay m ention
the concepts o f denotation, satisfaction, and definition, which
appear, for exam ple, in the follow ing statem ents:
The expression ‘the victor o f Jena ’ denotes Napoleon; snow
satisfies the condition ‘x is white'; the equation ‘a? — 2 ’ defines
(determines uniquely) the cube root o f the number 2.
The concept o f truth also— and this is not com m only recognized
— is to be included here, at least in its classical interpretation,
according to w hich ‘tru e’ signifies the same as ‘corresponding
w ith reality’ .
Concepts from the dom ain o f semantics have traditionally
played a prom inent part in the disoussions o f philosophers,
logicians, and philologists. Nevertheless they have long been
regarded with a certain scepticism . From the historical point
o f view this scepticism is well founded; for, although the con­
tent o f the semantical concepts, as they occur in colloquial
language, is clear enough, yet all attem pts to characterize this
content m ore precisely have failed, and various discussions in
w hich these concepts appeared and which were based on quite
plausible and seem ingly evident premisses, have often led to
paradoxes and antinom ies. I t suffices to m ention here the anti-
t B ibliographical N ote . This article is a summary of an address which
was given at the International Congress of Scientific Philosophy in Paris, 1935.
The article first appeared in Polish under the title *O ugruntowaniu naukowej
semantyki*, in Przeglqd FUozoftczny, vol. 39 (1936), pp. 60-67, and later in
German under the title 'Grundlegung der wissenschaftlichen Semantik’, in
Acted du Congrla International de Philosophic Scientiflque, vol. 3 (Actuality
Scientifiques et Induatrielles, vol. 390), Paris, 1936, pp. 1-8.
402 E S T A B L ISH M E N T OF SC IE N T IF IC SEM AN TICS XV

nom y o f the liar, the Grelling-N elson antinom y o f heterological

terms and the R ichard antinom y o f definability.
The main source o f the difficulties m et with seems to lie in
the follow ing: it has not always been kept in m ind that the
semantical concepts have a relative character, that they must
always be related to a particular language. People have not
been aware that the language about which we speak need b y no
means coincide w ith the language in which we speak. They
have carried out the semantics o f a language in that language
itself and, generally speaking, they have proceeded as though
there was only one language in the w orld. The analysis o f the
antinomies m entioned shows, on the contrary, that the seman­
tical concepts sim ply have no place in the language to which they
relate, that the language which contains its own semantics, and
within which the usual logical laws hold, must inevitably be
inconsistent. Only in recent years has attention been given to
all these facts (as far as I know Lesniewski was the first to
becom e fully aware o f them ).
As soon as we give full recognition to the above circumstances,
and carefully avoid the errors so far com m itted, the task o f
laying the foundations o f a scientific semantics, i.e. o f character­
izing precisely the semantical conoepts and o f setting up a
logically unobjectionable and m aterially adequate way o f using
these concepts, presents no further insuperable difficulties.
N aturally, in doing this we must proceed cautiously, making
full use o f the apparatus which m odem logic provides and care­
fu lly attending to the requirements o f present-day m ethodology.
In the solution o f this problem we can distinguish several
steps. W e must begin with the description o f the language
whose semantics we wish to construct. In particular we must
enumerate the prim itive term s o f the language and give the
rules o f definition by which new terms distinct from the
prim itive ones can be introduced into the language. N ext we
must distinguish those expressions o f the language which are
called sentences, separate the axiom s from the totality o f sen­
tences, and finally form ulate the rules o f inference by means
o f which theorems can be derived from these axiom s. The
XV E S T A B L IS H M E N T OF SC IE N T IF IC SEM ANTICS 403

description o f a language is exact and clear only if it is purely

structural, that is to say, i f we em ploy in it only those concepts
which relate to the form and arrangement o f the signs and
com pound expressions o f the language. N ot every language can
be described in this purely structural manner. The languages
for which such a description can be given are called formalized
languages. Now, since the degree o f exactitude o f all further
investigations depends essentially on the clarity and precision
o f this description, it is only the semantics of formalized languages
which can be constructed by exact methods.
The next step is the construction o f the language on the basis
o f which the semantics o f the given language is to be developed
and which, for the sake o f brevity, we shall call the metalanguage,
The m ost im portant point in this construction is the problem o f
equipping the metalanguage w ith a sufficiently rich vocabulary.
B ut the solution o f this problem is prescribed b y the particular
nature o f the semantical concepts. In faGt, semantical concepts
express certain relations between objects (and states o f affairs)
referred to in the language discussed and expressions o f the
language referring to these objects. Hence the statements which
establish the essential properties o f semantical concepts must
contain both the designation o f the objects referred to (thus
the expressions o f the language itself), and the terms which
are used in the structural description o f the language. The
latter terms belong to the dom ain o f the so-called morphology of
language and are the designations o f individual expressions o f
the language, o f structural properties o f expressions, o f structural
relations between expressions, and so on. The metalanguage
which is to form the basis for semantical investigations must thus
contain both kinds o f expression: the expressions o f the original
language, and the expressions o f the m orphology o f language,
In addition to these, the metalanguage, like every other language,
must contain a larger or smaller stock o f purely logical expres­
sions. The question now arises whether a metalanguage which
is equipped exclusively with expressions o f the kinds m entioned
form s a sufficient basis for sem antical investigations. W e shall
return to this question later.
404 E S T A B L IS H M E N T OF S C IE N T IF IC SEM ANTICS XV

Our next task is to determ ine the conditions under which

we should be inclined to regard a w ay o f using sem antical con­
cepts as m aterially adequate and in accordance with ordinary
usage. W e shall explain this m ore exactly b y reference to the
concept o f truth. W e regard the truth o f a sentence as its ‘corre­
spondence w ith reality’ . This rather vague phrase, which can
certainly lead to various misunderstandings and has often done
so in the past, is interpreted as follow s. W e shall regard as valid
all such statements as .
the sentence ‘it is snowing9is true if and only if it is snowing;
the sentence ‘the world war will begin in the year 1963’ is true if
and only if the world war will begin in 1963.
Quite generally we shall accept as valid every sentence o f the
^orm the sentence x is true if and only if p
where ‘p ’ is to be replaced b y any sentence o f the language
under investigation and V b y any individual name o f that
sentence provided this name occurs in the metalanguage. (In
colloquial language such names are usually form ed b y means
o f quotation marks.) Statem ents o f this form can be regarded
as partial definitions o f the concept o f truth. They explain in
a precise w ay, and in conform ity with com m on usage, the sense
o f all special expressions o f the ty p e : the sentence x is true. Now, if
we succeed in introducing the term ‘true’ into the metalanguage
in such a- w ay that every statem ent o f the form discussed can
be proved on the basis o f the axiom s and rules o f inference o f
the metalanguage, then we shall say that the way o f using the
concept o f truth which has thus been established is materially
adequate. In particular, if we succeed in introducing such a
concept o f truth b y means o f a definition, then we shall also
say that the corresponding definition is m aterially adequate.
W e can apply an analogous m ethod to any other semantical
concepts as well. E or each o f these concepts we form ulate a
system o f statem ents, which are expressed in the form o f equiva­
lences and have the character o f partial definitions; as regards
their contents, these statem ents determine the sense o f the
concept concerned with respect to all concrete, structurally
XV E S T A B L IS H M E N T OF S C IE N T IF IC SEM ANTICS 405

described expressions o f the language being investigated. W e

then agree to regard a w ay o f using (or a definition o f) the se­
m antical concept in question as m aterially adequate if it enables
us to prove in the m etalanguage all the partial definitions just
m entioned. B y w ay o f illustration we give here such a partial
definition o f the concept o f satisfaction:

John and PtUr satisfy the sentential function ‘X and Y are

brothers9 if and only if John and Peter are brothers.
I t should also be noted that, strictly speaking, the described
conventions (regarding the m aterial adequacy o f the usage o f
sem antical concepts) are form ulated in the m etametalanguage
and not in the metalanguage itself.
A ll the problem s so far have been o f a preparatory and
auriliary nature. Only after these preliminaries can we ap­
proach our ch ief problem . This is the establishm ent o f a m a­
terially correct w ay o f using the sem antical concepts in the
metalanguage. A t this point tw o procedures com e into con ­
sideration. In the first the semantical concepts (or at least some
o f them ) are introduced into the metalanguage as new prim i­
tive concepts and their basic properties are established by
means o f axiom s. Am ong these axiom s are included all state­
ments which secure the m aterially adequate use o f the concepts
in question. In this w ay semantics becom es an independent
deductive theory based upon the m orphology o f language. But
when this m ethod, which seems easy and simple, is worked out
in detail various objections present them selves. The getting up
o f an axiom system sufficient for the developm ent o f the whole
o f semantics offers considerable difficulties. F or certain reasons,
which we shall not go into here, the choice o f axiom s always
has a rather accidental character, depending on inessential fa c­
tors (such as e.g. the actual state o f our knowledge). Various
criteria which we should like to use in this connexion prove to
be inapplicable. M oreover, the question arises whether the
axiom atically constructed semantics is consistent. The problem
o f consistency arises, o f course, whenever the axiom atic m ethod
is applied, bu t here it acquires a special im portance, as we see
406 E S T A B L ISH M E N T OF S C IE N T IF IC SEM ANTICS XV

from the sad experiences we have had with the semantical con­
cepts in colloquial language. A part from the problem o f con­
sistency, a m ethod o f constructing a theory does not seem to
be very natural from the psychological point o f view if in this
m ethod the role o f prim itive concepts— thus o f concepts whose
meaning should appear evident— is played b y concepts which
have led to various misunderstandings in the past. Finally,
should this m ethod prove to be the only possible one and not
be regarded as m erely a transitory stage, it would arouse certain
doubts from a general philosophical point o f view . It seems to
me that it w ould then be difficult to bring this m ethod into
harm ony with the postulates o f the unity o f science and o f
physicalism (since the concepts o f semantics w ould be neither
logical nor physical concepts).
In the second procedure, which has none o f the above dis­
advantages, the sem antical concepts are defined in terms o f the
usual concepts o f the metalanguage and are thus reduced to
purely logical concepts, the concepts o f the language being
investigated and the specific concepts o f the m orphology o f
language. In this way semantics becom es a part o f the m orpho­
logy o f language if the latter is understood in a sufficiently wide
sense. The question arises whether this m ethod is applicable
at all. I t seems to me that this problem can now be regarded
as definitely solved. I t proves to be closely connected with the
theory o f logical types. The chief result relevant to this question
can be form ulated as follow s:

It is possible to construct in the metalanguage methodologically

correct and materially adequate definitions of the semantical con-
cepts if and only if the metalanguage is equipped with variables of
higher logical type than all the variables of the language which is
the subject of investigation.

I t w ould be im possible to establish here, even in general out­

line, the thesis just form ulated. I t can only be pointed out that
it has been found useful, in defining the semantical concepts,
to deal first with the concept o f satisfaction; both because a
definition o f this concept presents relatively few difficulties, and
XV E S T A B L ISH M E N T OF SC IE N T IF IC SEM ANTICS 407

because the remaining semantical concepts are easily reducible

to it. It is also to be noted that from the definitions o f seman­
tical concepts various im portant theorems o f a general nature
concerning the concepts defined can be derived. F or example,
on the basis o f the definition o f truth we can prove the laws o f
contradiction and o f excluded midtile (in their m etalogical
form ulation).
W ith this the problem o f establishing semantics on a scientific
basis is com pletely solved. Only the future can definitely say
whether further investigations in this field will prove to be fruit­
ful for philosophy and the special sciences, and what place
semantics will win for itself in the totality o f knowledge. But
it does seem that the results hitherto reached ju stify a certain
optim ism in this respect. The very fact that it has been possible
to define the semantical concepts, at least for form alized lan­
guages, in a correct and adequate manner seems to be not
entirely without im portance from the philosophical standpoint.
The problem o f the definition o f truth, for example, has often
been emphasized as one o f the fundamental problem s o f the
th eoiy o f knowledge. Also certain applications o f semantics,
and especially o f the theory o f truth, in the dom ain o f the
m ethodology o f the deductive sciences, or m etamathematics,
seem to me to deserve attention. Using the definition o f truth
we are in a position to carry out the p roof o f consistency for
deductive theories in which only (m aterially) true sentences
are (form ally) provable: this can be done under the condition
that the metalanguage in which the p roof o f consistency is
carried out is equipped with variables o f higher type than all
the variables which occur in the sentences o f the theory dis­
cussed. A dm ittedly such a p roof has no great cognitive value
since it rests on logically stronger premisses than the assump­
tions o f the theory whose consistency is proved. Nevertheless,
the result seems to be o f some interest for the reason that it
cannot be im proved. For it follow s from the investigations o f
Godel that the p roof o f consistency cannot be carried out if the
metalanguage contains no variables o f higher type. The defini­
tion o f truth has yet another consequence which is likewise
408 E S T A B L IS H M E N T OF S C IE N T IF IC SEM AN TICS XV

connected with the investigations o f Godel. As is well known,

G odel has developed a m ethod which makes it possible, in every
theory which includes the arithm etic o f natural numbers as a
part, to construct sentences which can be neither proved nor
disproved in this theory. B ut he has also pointed out that the
undecidable sentences constructed b y this m ethod becom e de­
cidable if the theory under investigation is enriched b y the
addition o f variables o f higher type. The p roof that the sen­
tences involved actually in this w ay becom e decidable again
rests on the definition o f truth. Similarly-—as I have shown
b y means o f the m ethods used in developing semantics— for
any given deductive theory it is possible to indicate concepts
which cannot be defined in this theory, although in their con­
tent they belong to the theory, and becom e definable in it if
the theory is enriched b y the introduction o f higher types.
Summarizing, we can say that the establishm ent o f scientific
semantics, and in particular the definition o f truth, enables us
to m atch the negative results in the field o f m etamathematics
w ith corresponding positive ones, and in that w ay to fill to
some extent the gaps which have been revealed in the deductive
m ethod and in the very structure o f deductive science .1
1 More detailed information about many of the problems discussed in this
artiole can be found in V m . Attention should also be called to my later
paper, Tarski, A. (82). While the first part of that paper is close in its content
to the present article, the second part contains polemical remarks regarding
various objections which have been raised against my investigations in the
field of semantics. It also includes some observations about the applicability
of semantics to empirical sciences and their methodology.
 XVI
ON THE CONCEPT OF L O G IC A L
CONSEQUENCEf

T he concept o f logical consequence is one o f those whose intro*

duction in to the field o f strict form al investigation was n ot a
m atter o f arbitrary decision on the part o f this or that investi­
gator; in defining this concept, efforts were m ade to adhere to
the com m on usage o f the language o f everyday life. B u t these
efforts have been confronted with the difficulties which usually
present them selves in such oases. W ith respect to the clarity
o f its content the com m on concept o f oonsequence is in no w ay
superior to other ooncepts o f everyday language. Its extension
is not sharply bounded and its usage fluctuates. A n y attem pt
to bring into harm ony all possible vague, sometim es contra­
dictory, tendencies whioh are connected with the use o f this
concept, is certainly doom ed to failure. W e must reconcile our­
selves from the start to the fa ct that every precise definition o f
this concept will show arbitrary features to a greater or less
degree.
E ven until recently m any logicians believed that they had
succeeded, b y means o f a relatively meagre stock o f concepts,
in grasping alm ost exactly the oontent o f the com m on concept
o f consequence, or rather in defining a new concept which coin­
cided in extent with the com m on one. Such a belief could easily
arise am idst the new achievem ents o f the m ethodology o f de­
ductive science. Thanks to the progress o f m athem atical logio
we have learnt, during the course o f recent decades, how to
present m athem atical disciplines in the shape o f form alized
deductive theories. In these theories, as is w ell known, the
f B ibxjogbaphioax N ote . This is a summary of an address given at the
International Congress of Soientifio Philosophy in Paris, 1935. The article
first appeared in print in Polish under the title ‘ O pojoiu wynikania logics-
nego* in Przeglqd Filozoflczny, vol. 89 (1936), pp. 68-68, and then in German
under the title ‘ tlber den Begriff der logisohen Folgerung’, Aetes du Oongrit
International de PhUoeophio Soiantiftque, vol. 7 (Aotualitis Scientifiques et
Industrielles, vol. 894), Paris, 1936, pp. 1-11.
410 ON TH E CONCEPT OF LOG ICAL CONSEQUENCE XVI

p roof o f every theorem reduces to single or repeated application

o f some simple rules o f inference— suoh as the rules o f substitu­
tion and detachm ent. These rules tell us what transformations
o f a purely structural kind (i.e. transformations in which only
the external structure o f sentences is involved) are to be per­
form ed upon the axiom s or theorems already proved in the
theory, in order that the sentences obtained as a result o f such
transform ations m ay themselves be regarded as proved. L ogi­
cians thought that these few rules o f inference exhausted the
content o f the concept o f consequence. W henever a sentence
follow s from others, it can be obtained from them— so it was
thought— in more or less com plicated ways b y means o f the
transform ations prescribed b y the rules. In order to defend this
view against sceptics who doubted whether the concept o f conse­
quence when form alized in this w ay really coincided in extent
with the com m on one, the logicians were able to bring forward
a w eighty argument: the fact that they had actually succeeded
in reproducing in the shape o f form alized proofs all the exact
reasonings which had ever been carried out in mathematics.
Nevertheless we know today that the scepticism was quite
justified and that the view sketched above cannot be main­
tained. Some years ago I gave a quite elem entary example o f
a theory which shows the follow ing peculiarity: among its
theorems there occur such sentences as:
A 0. 0 possesses the given property P ,
A v 1 possesses the given property P ,
and, in general, all particular sentences o f the form
A n. n possesses the given property P ,
where V represents any sym bol which denotes a natural number
in a given (e.g. decim al) number system . On the other hand the
universal sentence:
A . Every natural number possesses the given property P ,
cannot be proved on the basis o f the theory in question b y means
o f the norm al rules o f inference .1 This fact seems to me to speak
1 For a detailed description of a theory with this peculiarity see I X ; for
the discussion of the closely related rule of infinite induction see V III, pp. 258 ff.
XVI ON TH E CONCEPT OF LOG ICAL CONSEQUENCE 411

for itself. I t shows that the form alized concept o f consequence,

as it is generally used b y mathematical logicians, b y no means
coincides with the com m on concept. Y et intuitively it seems
certain that the universal sentence A follows in the usual sense
from the totality o f particular sentences A Qi A v „>, A n>.... P ro­
vided all these sentences are true, the sentence A must also be
true.
In connexion with situations o f the kind just described it has
proved to be possible to form ulate new rules o f inference which
do not differ from the old ones in their logical structure, are
intuitively equally infallible, i.e. always lead from true sentences
to true sentences, but cannot be reduced to the old rules. An
exam ple o f such a. rule is the so-called rule o f infinite induction
according to which the sentence A can be regarded as proved
provided all the sentences A 0, A v ..., A n,... have been proved
(the sym bols ‘A 0’ , (A ±\ etc., being used in the same sense as
previously). B ut this rule, on account o f its infinitistic nature,
is in essential respects different from the old rules. It can only
be applied in the construction o f a theory if we have first suc­
ceeded in proving infinitely m any sentences o f this theory— a
state o f affairs which is never realized in practice. But this
defect can easily be overcom e b y means o f a certain m odifica­
tion o f the new rule. For this purpose we consider the sentence
B which asserts that all the sentences A 0i A lf..., A nt... are
provable on the basis o f the rules o f inference hitherto used
(not that they have actually been proved). W e then set up
the follow ing rule: if the sentence B is proved, then the corre­
sponding sentence A can be accepted as proved. But here
it m ight still be objected that the sentence B is not at all a
sentence o f the theory under construction, but belongs to the
so-called m etatheory (i.e. the theory of the theory discussed)
and that in consequence a practical application o f the rule in
question will always require a transition from the theory to the
m etatheory .1 In order to avoid this objection we shall restrict

1 For the concept of metatheory and the problem of the interpretation of a

metatheory in the corresponding theory see article V H I, pp. 167 ff., 184, and
247 ff.
412 ON TH E CONCEPT OF LO G IC A L CONSEQUENCE XVI

consideration to those deductive theories in which the arith­

m etic o f natural numbers can be developed, and observe that
in every such theory all the concepts and sentences o f the
corresponding m etatheory can be interpreted (since a one-one
correspondence can be established betwreen expressions o f a
language and natural num bers ).1 W e can replace in the rule
discussed the sentence B b y the sentence B\ which is the arith­
m etical interpretation o f 2?. In this w ay we reach a rule which
does n ot deviate essentially from the rules o f inference, either
in the conditions o f its applicability or in the nature o f the
concepts involved in its form ulation or, finally, in its intuitive
infallibility (although it is considerably m ore com plicated).
N ow it is possible to state other rules o f like nature, and even
as m any o f them as we please. A ctually it suffices in fact to
notice that the rule last form ulated is essentially dependent
upon the extension o f the concept ‘sentence provable on the
basis o f the rules hitherto used\ B ut when we adopt this rule
we thereby widen the extension o f this concept. Then, for the
widened extension we can set up a new, analogous rule, and
so on ad infinitum . I t w ould be interesting to investigate
whether there are any objective reasons for assigning a special
position to the rules ordinarily used.
The conjecture now suggests itself that we can finally succeed
in grasping the full intuitive content o f the concept o f conse­
quence b y the m ethod sketched above, i.e. b y supplem enting
the rules o f inference used in the construction o f deductive
theories- B y m aking use o f the results o f K . G odel2 we can
show that this conjecture is untenable. In every deductive
theory (apart from certain theories o f a particularly elem entary
nature), however m uch we supplem ent the ordinary rules o f
inference b y new purely structural rules, it is possible to con­
struct sentences which follow , in the usual sense, from the
theorem s o f this theory, but which nevertheless cannot be
proved in this theory on the basis o f the accepted rules o f
1 For the concept of metatheory and the problem of the interpretation of a
metatheory in the eorrosponding theory see article V III. pp. 167 £f., 184, and
247 ff.
* Cf. G<>dei, K . (22), especially pp. 190 f.
XV I ON TH E CONCEPT OF LO G IC AL CONSEQUENCE 413

inference ,1 In order to obtain the proper concept o f consequence,

which is close in essentials to the com m on concept, we must
resort to quite different m ethods and apply quite different con­
ceptual apparatus in defining it. It is perhaps not superfluous
to point out in advance that— in com parison with the new—-the
old ooncept o f consequence as com m only used by mathematical
logicians in no w ay loses its im portance. This ooncept will
probably always bare a decisive significance for the practical
construction o f deductive theories, as an instrum ent which
allows us to prove or disprove particular sentences o f these
theories. It seems, however, that in considerations o f a general
theoretical nature the proper concept o f consequence must be
placed in the foreground .3
The first attem pt to form ulate a precise definition o f the
proper concept o f consequence was that o f R , Catnap .3 B ut this
1 In order to anticipate possible objections the range of application of the
result jost formulated should be determined more exac Jy and the logical nature
o f the rules o f inference exhibited more clear]^ , iu yartioulftr it should be
exactly explained what is meant by the structural character of those rules.
* An opposition between the two concept i i question is clearly pointed cut-
in article IX , pp. 2fl3 ff. Nevertheless, in contrast tc m y present standpoint,
I have there expressed myself in a decidedly negative manner about the possi­
bility of setting up an exact formal definition lor the proper concept of conse­
quence. My position at that time is explained by the fact that, when I was
writing the article mentioned, I wishtr.i tc any means of construction
which went beyond the theory of logical types tu any of its classical forms;
but it can be shown that it is impossible to define the proper concept of
consequence adequately whilst using exclusively the means admissible in the
classical theory of types; unless we should thus limit our considerations
solely to formalized language? of au elementary and fragmentary character (to
be exact, to the so-called languages of Unite order; of. article V III, especially
pp. 268 if.). In his extremely interesting book, Carnap, K. (10), the term
{logical) derivatioti or dericabiUty Ls applied to old ooncept of consequence
as commonly used in the construction cf deductive theories, in order to distin­
guish it from the concept of eof*$eqvmce as the proper concept. The opposition
between the two concepts is extended by Carnap to the most diverse derived
conoepts (*f-concepts'' and *Sk~c<jrxej)ts\ cf. pp. 88 ff., and 124 ff.); he tlso
emphasizes— to my mind correctly— the importance of the proper concord of
consequence and the concepts derived from it, for general theoretical discus­
sions (cf. e.g. p. 128).
* Cf. Carnap, R. (10), pp. 88 £, and Carnap, R. (11) especially p. 181.
In the first of these works there is yet another definition of consequence which
is adapted to a formalized language of an elementary character. This definition
is not considered here because it cannot be applied to languages of a more com­
plicated logical structure. Carnap attempts to define the concept of logical conse­
quence not only for special languages, but also within the framework of what he
calls *general syntax \ We shall have more to say about this on p. 416, note 1.
414 ON TH E CONCEPT OF LOG ICAL CONSEQUENCE XVI

attem pt is connected rather closely w ith the particular proper­

ties o f the form alized language which was chosen as the subject
o f investigation. The definition proposed by Carnap can be
form ulated as follow s:
The sentence X follow s logically from the sentences of the class
K if and only if the class consisting of all the sentences of K and
of the negation of X is contradictory.
The decisive element o f the above definition obviously is the
concept 'con tradictory', Carnap s definition o f this concept is
too com plicated and special to be reproduced here w ithout long
and troublesom e explanations .1
I should like to sketch here a general m ethod which, it seems
to me, enables us to construct an adequate definition o f the
concept o f consequence for a com prehensive class o f form alized
languages, I emphasize, however, that the proposed treatm ent
o f the concept o f consequence makes no very high elaim to
com plete originality. The ideas involved in this treatm ent will
certainly seem to be something well known, or even something
o f his own, to m any a logician who has given close attention to
the concept o f consequence and has tried to characterize it
more precisely. Nevertheless it seems to me that only the
m ethods which have been developed in recent years for the
establishment o f scientific semantics, and the concept's defined
with their aid, allow us to present these ideas in an exact form .2
Certain considerations o f an intuitive nature will form our
starting-point. Consider any class K o f sentences and a- sentence
X which follow s from the sentences o f this class. From an in­
tuitive standpoint it can never happen that both the class K
consists only o f true sentences and the sentence X is false.
M oreover, since we are concerned here with the concept o f
logical, i.e. formal, consequence, and thus with a relation which is
to be uniquely determ ined b y the form o f the sentences between
which it holds, this relation cannot be influenced in any way by
em pirical knowledge, and in particular b y knowledge o f the
1 See footnote 3 on p. 413.
* The methods and concepts of semantics and especially the concepts of
truth and satisfaction are discussed in detail in article V II I ; Bee also article X V .
XV I ON TH E CONCEPT OF LOGICAL CONSEQUENCE 415

objects to which the sentence X or the sentences o f the class

K refer. The consequence relation cannot be affected b y re­
placing the designations o f the objeots referred to in these
sentences b y the designations o f any other objects. The two
circumstances just indicated, which seem to be very charac­
teristic and essential for the proper concept o f consequence, m ay
be join tly expressed in the follow ing statem ent:

(F) If, in the sentences of the class K and in the sentence X , the
constants—apart from purely logical constants—are replaced by
any other constants (like signs being everywhere replaced by like
signs), and if we denote the class of sentences thus obtained from
K by iK ri, and the sentence obtained from X by 6X'\ then the
sentence X ' must be true provided only that all sentences of the
class K f are true.
[F or the sake o f sim plifying the discussion certain incidental
com plications are disregarded, both here and in what follow s.
They are connected partly with the theory o f logical types, and
partly with the necessity o f eliminating any defined signs which
m ay possibly occur in the sentences concerned, i.e. o f replacing
them by prim itive signs.]
In the statem ent (F ) we have obtained a necessary condition
for the sentence X to be a consequenoe o f the class K . The
question now arises whether this condition is also sufficient.
I f this question were to be answered in the affirmative, the
problem o f form ulating an adequate definition o f the concept
o f consequence w ould be solved affirm atively. The only diffi­
culty would be connected with the term ‘true’ which occurs in
the condition ( F ). B ut this term can be exactly and adequately
defined in sem antics .1
U nfortunately the situation is not so favourable. I t m ay, and
it does, happen— it is not difficult to show this b y considering
special form alized languages— that the sentence X does not
follow in the ordinary sense from the sentences o f the class K
although the condition ( F ) is satisfied. This condition m ay in
fa ct be satisfied only because the language with which we are
1 See footnote 2 on p. 414.
41# ON T H E CONCEPT OF LOG ICAL CONSEQUENCE X V I

dealing does not possess a sufficient stock o f extra-logical con­

stants. The condition (F ) could be regarded as sufficient for
the sentence X to follow from the class K only if the designations
o f all possible objects occurred in the language in question. This
assum ption, how ever, is fictitious and can never be realized .1
W e m ust therefore look for some means o f expressing the in­
tentions o f the condition (F ) which will be com pletely inde­
pendent o f that fictitious assum ption.
Such a means is provided by semantics. Among the funda­
mental concepts o f semantics we have the concept o f the satis­
faction of a sentential function by single objects or by a sequence
<5f objects. It would be superfluous to give here a precise ex­
planation o f the content o f this concept. The intuitive meaning
o f such phrases ae: John and Peter satisfy the condition 'X and
7 are brothers’, or the triple of numbers 2, 3, and 5 satisfies the
equation :£C-fy = z\ can give rise to no doubts. The concept o f
satisfaction—like other semantical concepts—must always be
relativized to some particular language. The details o f its pre­
cise definition depend on the structure o f this language. Never­
theless, a general method can be developed which enables us to
construct such definitions for a comprehensive class o f forma­
lized languages. Unfortunately, for technical reasons, it would
be impossible to sketch this method here even in its general
outlines.*
One o f the concepts which can be defined in terms o f the
concept o f satisfaction is the concept o f model. L et us assume
that in the language we are considering certain variables corre­
spond to every extra-logical constant, and in such a w ay that
every sentence becom es a sentential function if the constants
in it are replaced b y the corresponding variables. L et L be any
class o f sentences. W e replace all extra-logical constants which
1 These last remarks constitute a criticism of some earlier attempts to
define the concept of formal consequence. They concern, in particular, Car­
nap’s definitions of logical consequence and a series of derivative concepts
(L-consequences and L-conoepts, of. Carnap, R . (10), pp. 137 & ). These defini­
tions, in so far ae they are set up on the basis of ‘general syntax’, seem to me
to be materially inadequate, just because the defined concepts depend essen-
tially, in their extension, on the richness of the language investigated.
* See footnote 2 on p. 414.
XV I ON TH E CONCEPT OF LO G IC A L CONSEQ UENCE 417

ocour in the sentences belonging to L by corresponding variables,

like constants being replaced b y like variables, and unlike b y
unlike. In this w ay we obtain a class & o f sentential functions.
A n arbitrary sequence o f objects whioh satisfies every sentential
function o f the class L r w ill be called a model or realization of
the class L of sentences (in ju st this sense one usually speaks o f
m odels o f an axiom system o f a deductive theory). If, in parti­
cular, the class L consists o f a single sentence X , we shall also
call the m odel o f the class L the model of the sentence X .
In term s o f these concepts we can define the concept o f logical
consequence as follow s:

The sentence X follows fisca lly from the. sentences of the class
K if and only if every model of the class K is also a model of the
sentence X .t
I t seems to me that everyone who understands the content
o f the above definition must adm it that it agrees quite well
with com m on usage. This becom es still clearer from its various
consequences. In particular, it can be proved, on the basis o f
this definition, that every consequence o f true sentences must
be true, and also that the consequence relation which holds
between given sentences is com pletely independent o f the sense
o f the extra-logical constants which occur in these sentences.
In brief, it can be shown that the condition (F ) form ulated
above is necessary if the sentence X is to follow from the sen­
tences o f the class K . On the other hand, this condition is in
general not sufficient, since the concept o f consequence here
defined (in agreement with the standpoint we have taken) is-
independent o f the richness in concepts o f the language being
investigated.
Finally, it is not difficult to reconcile the proposed definition
with that o f Carnap. F or we can agree to call a class o f sentences

t After the original of this paper had appeared in print, K . Scholz in his
article ‘ Die Wistenschaftslehxe Bc-lzanos, Eine Jahrhundert-Betrachtung’,
AbhancUungen der Fri&s'schen Schule, new series, vol. 6, pp. $99-472 (see in
particular p. 472, footnote 58) pointed out a far-reaching analogy between this
definition of consequence and the one suggested by B. Bolzano about a
hundred years earlier.
418 ON TH E CONCEPT OP LO G IC AL CONSEQUENCE XVI

contradictory i f it possesses no m odel. A nalogously, a class o f

sentences can be called analytical i f every sequence o f objects
is a m odel o f it. B oth o f these ooncepts can be related not only
t o classes o f sentences but also to single sentences. L et us
assume further that, in the language with which we are dealing,
for every sentence X there exists a negation o f this sentence,
i.e. a sentence 7 which has as a m odel those and only those
sequences o f objeots which are n ot m odels o f the sentence X
(this assum ption is rather essential for Carnap’s construction).
On the basis o f all these conventions and assum ptions it is easy
to prove the equivalence o f (he iwo definitions. W e can also show
— ju st as does Carnap— that those and only those sentences are
analytical w hich follow from every class o f sentences (in parti­
cular from the em pty class), and those and on ly those are
oontradiotory from whioh every sentence follow s .1
I am not at all o f the opinion that in the result o f the above
discussion the problem o f a m aterially adequate definition o f
the concept o f consequence has been com pletely solved. On the
contrary, I still see several open questions, on ly one o f whioh—
perhaps the m ost im portant— I shall point out here.
U nderlying our whole construction is the division o f all terms
o f the language discussed in to logical and extra-logical. This
division is certainly n ot quite arbitrary. If, for exam ple, we
were to include am ong the extra-logioal signs the im plication
sign, or the universal quantifier, then our definition o f the con ­
cept o f consequence w ould lead to results which obviously
contradict ordinary usage. On the other hand, no objective
grounds are know n to m e whioh perm it us to draw a sharp

1 Cf. Carnap, B . (10), pp. 135 ff., especially Tbs. 52.7 and 62.8; Carnap, B .
(11), p. 182, Ths. 10 and 11. Incidentally I should like to remark that the
definition of the concept of oonsequence hare proposed does not exceed the
limits of syntax in Carnap's conception (cf. Carnap, R. (10), pp. 6 £f.). Ad­
mittedly the general concept of satisfaction (or of model) does not belong to
syntax; but we use only a special case of this oonoept— the satisfaction of
sentential funotions whioh contain no extra-logioal constants, and this special
case can be characterized using only general logical and specifio syntactical
eonoepts. Between the general concept of satisfaction and the special case of
this concept used here approximately the same relation holds as between the
semantical concept of true sentence and the syntactical oonoept of analytical
sentence.
X V I ON TH E CONCEPT OF LOG ICAL CONSEQUENCE 419

boundary between the tw o groups o f terms. It seems to be

possible to include am ong logical terms some which are usually
regarded b y logicians as extra-logical without running into
consequences whioh stand in sharp contrast to ordinary usage.
In the extrem e case we could regard all terms o f the language
as logical. The concept o f formal consequence w ould then co ­
incide with that o f material consequence. The sentence X w ould
in this case follow from the class K o f sentences if either X were
true or at least one sentence o f the class K were false .1
In order to see the im portance o f this problem for certain
general philosophical views it suffices to note that the division
o f terms into logical and extra-logical also plays an essential
part in clarifying the concept ‘analytical*. B ut according to
m any logicians this last concept is to be regarded as the exact
form al correlate o f the concept o f tautology (i.e. o f a statem ent

1 It will perhaps be instructive to juxtapose the three concepts; 1deriva-

bility* (of. p. 413, note 2), ‘ formal consequence\ and ‘ material consequence\
for the special case when the class K , from whioh the given sentence X follows,
consists of only a finite number of sentences: Ylt Y&..., Yn. Let us denote
by the Bymbol 'Z* the conditional sentence (the implication) whose antecedent
is the conjunction of the sentences Ylt Yn and whose consequent is the
sentence X . The following equivalences can then be established i
the sentence X is (logically) derivablefrom the sentences of the class K if and only
if the sentence Z is logically provable (i.e. derivable from the axioms of logic) ;
the sentence X follows formally from the sentences of the class K if and only if
the sentence Z is analytical;
the sentence X follows materially from the sentences of the class K if and only if
the sentence Z is true.
Of the three equivalences only the first can arouse certain objections; of.
article X H , pp. 342-64, especially 346. In connexion with these equivalences
cf. also Ajdukiewicz, EL. (2), p. 19, and (4), pp. 14 and 42.
In view of the analogy indicated between the several variants of the con*
cept of consequence, the question presents itself whether it would not be useful
to introduce, in addition to the special concepts, a general concept of a relative
character, and indeed the concept of consequence with respect to a class L of
sentences. I f we make use again of the previous notation (limiting ourselves
to the case when the class K is finite), we can define this concept as follows:
the sentence X follows from the sentences of the doss K with respect to the
class L of sentences if and only if ike sentence Z belongs to the class L.
On the basis of this definition, derivability would ooineide with consequence
with respect to the class of all logically provable sentences, formal consequences
would be consequences with respect to the class of aQ analytical sentences, and
material consequences those with respect to the class of all true sentences.
420 ON TH E CONCEPT OF LO G IC AL CONSEQUENCE XVI

which ‘says nothing about reality’ ), a concept which to me

personally seems rather vague, but which has been o f funda­
m ental im portance for the philosophical discussions o f L. W itt­
genstein and the whole Vienna Circle .1
Further research will doubtless greatly clarify the problem
which interests us. Perhaps it w ill be possible to find im portant
objective arguments which w ill enable us to ju stify the tradi­
tional boundary between logical and extra-logical expressions.
B ut I also consider it to be quite possible that investigations
will bring no positive results in this direction, so that we shall
be com pelled to regard such concepts as ‘logical consequence’ ,
‘analytical statem ent’, ‘and ‘tau tology’ as relative concepts
which must, on each occasion, be related to a definite, although
in greater or less degree arbitrary, division o f terms into logical
and extra-logical. The fluctuation in the com m on usage o f the
concept o f consequence w ould— in part at least— be quite
naturally reflected in such a com pulsory situation.
1 Cf. Wittgenstein, L. (91), Carnap, R. (10), pp. 37-40.
 XVII
SE N T E N T IA L CALCULUS AND
TOPOLOGYf
I s this article I shall point out certain form al connexions between
the sentential calculus and topology (as well as some other
m athem atical theories). I am concerned in the first place with
a topological interpretation o f tw o systems o f the sentential
calculus, nam ely the ordinary (tw o-valued) and the intuitionistic
(B rouw er-H eyting) system . W ith every sentence o f the sen­
tential calculus we correlate, in one-one fashion, a sentence 21*
o f topology in such a w ay that 21 is provable in the tw o-valued
calculus if and only if 2lx holds in every topological space. An
analogous correlation is set up for the intuitionistic calculus.
The present discussion seems to me to have a certain interest
not only from the purely form al point o f view ; it also throws
an interesting light on the content relations between the two
system s o f the sentential calculus and the intuitions underlying
these systems.
In order to avoid possible misunderstandings I should like to
emphasize that I have not attem pted to adapt the m ethods o f
reasoning used in this article to the requirem ents o f intuitionistic
logic .1 F or valuable help in com pleting this work I am indebted
to Professor A, M ostowski.
1 Most results o f this article were obtained in the year 1935. The connexion
between the intuitionistic calculus and Boolean algebra (or the theory of
deductive eystems, see § 5) was discovered by me still earlier, namely in 1931.
Some remarks to this effect can be found in article X II of the present book and
in Tarski, A. (80). Only after completing this paper did I become acquainted
with the work, then newly published, of Stone, M. H. (70). Ir- spite of an
entirely different view of Brouwerian logic there is certainly some connexion
between particular results o f the two works, as can easily be seen comparing
Stone’s Th. 7, p. 22, and my Th. 4.11. In their mathematical content these
two theorems are closely related. But this does not at all apply to the two
works as wholes. In particular, Th. 4.24, in which I see the kernel of this paper,
tends in quite a different direction from Stone’s considerations.
t B ibliographical N ote . This article is the text of an address given by
the author on 30 September 1937 to the Third Polish Mathematical Congress in
Warsaw (see Annales de la Societi polonaise ds matkernatiexie, vol. 16 (1937),
p. 192). The article first appeared under the title ‘ Der Aussagenkalkul und die
Topologie’ , in Fundamenta Mathematical, vol. 31 (1928), pp. 103-34.
422 S E N T E N T IA L CALCULUS AN D TOPOLOG Y XVII, § 1

§ 1. T h e T w o -v a l u e d and the I n t u it io n ist ic S e n t e n t ia l

Calcu lu s

As is well known, the m ost elem entary part o f m athem atical

logic is the sentential calculus. In the expressions o f the sen­
tential calculus variables o f only one kind occur, nam ely senten­
tial variables, which represent whole sentences. As sentential
variables the letters ‘ X T ’ , will be used. In addition to
the variables, four constants occur in the sentential calculus:
the implication sign the disjunction sign *v the conjunc­
tion sign ‘ a ’ , and the negation sign ‘ -v. ’ (a fifth constant, the
equivalence sign ‘o ’ , will not be in volved here).
A rbitrary expressions which are com posed o f sentential
variables, the four constants m entioned, and possibly brackets,
w ill be denoted b y the letters ‘ 31’ , ‘ S ’ , . It is assumed that
the sentential variables are ordered in an (infinite) sequence with
distinct term s: 5B1( SS2,..., $»>••• • The sym bols MS-*- ® ’ , ‘ 31V SB’ ,
and ‘91A 83’ denote the implication, the disjunction, and the
conjunction o f Sff and SB respectively, that is the com pound ex ­
pressions which are form ed when Sit and SB are com bined b y the
corresponding constant ‘ v ’ , or *A ’ . Analogously the
negation o f Sit is denoted b y ‘ -vSJT.1
In addition to the signs just listed the usual set-theoretioal
sym bolism will be used.
The only expressions o f the sentential calculus with which we
shall deal will be sentences:

D efin ition 1.1. An expression 21 is called a sentential func­

tion, or (for shyrt) a sentence, if 3t belongs to every system which
(i) contains all sentential variables among its elements, (ii) is closed
under the operations of forming implications, disjunctions, con­
junctions, and negations; in other words, the system of all
sentences is the smallest system which has the properties (i)
and (ii).

1 In principle we thus use the signs \ * V \ ‘ A \ and * * in two senses:

in the logical and in the metalogical sense* In practice only the second sense is
involved* In }§ 4 and £ yet another, quite different sense will be given to these
signs.
XVII, § 1 S E N T E N T IA L CALCULUS A N D TO P O LO G Y 423

In the system o f all sentences two subsystems are distin­

guished: the provable sentences o f the two-valued calculus and
those o f the intuitionistic calculus.
D efin itio n 1.2. The sentence 21 is called an axiom o f the tw o­
valued calculus, or o f the intuitionistic calculus, if there are
sentences S , G, and £> such that 21 satisfies one of the following
formulas (i)-(x ) in the first case, or one of the formulas (i)-(ix ) and
(xi) in the second case:
(i) 21= 3 -> (G -> 3 ),
(ii) 21 =
(iii) 2 l= 3 -* (3 v G ),
(iv) 21 = G -* (23 VG),
(v) 21= ( 3 -*2>)->{(<E-► 3 ) -* [(» V < £ )-► »]},
(vi) 2t = ( 3 A G) -► 3 ,
(vii) 21 = (3 A G) -* G,
(viii) 21= (D -* 3 )-► {($ -* G )- * [ $ - * ( 3 A G)]},
(ix) 21= 3 - > ( 3 - * G),
(x) 21 = ( ^ 3 - » 3 ) - * S ,
(xi) 21= ( 3 - * ^ 3 ) - * ^ 3 1
D efintiton 1.3. I f 21, 3 , and G are three sentences and if
21 = 3 -* G, then G is said to be the result o f the detachment
o f the sentence 3 from the sentence 21.
D efin itio n 1.4. The system of provable sentences o f the
two-valued calculus is the smallest system of sentences which con­
tains all the axioms o f the two-valued calculus and is closed under
the operation of detachment. The system of provable sentences
o f the intuitionistic calculus is the smallest system which contains
all the axioms of the intuitionistic calculus and is closed under
the same operation *
It is known that from these definitions the following two
theorems are derivable:
1 Cf. similar axiom systems: for the two-valued calculus in Hilbert, D. and
Bemays, P. (31) and for the intuitionist calculus in Gontzen, G. (18)* An axiom
system for the intuitionistic calculus was first given in Heyting, A. (28). The
problem of the mutual independence of the axioms (or, more exactly, axiom-
schemata) of 1.2 is not here discussed.
8 The operation of substitution need not be considered in 1.4 since the
axiom system defined in 1.2 is already closed under this operation.
424 S E N T E N T IA L CALCULUS AN D TO P O LO G Y XVH , § 1

T heorem 1.5. 3ft £ (but not conversely: if, e.£., 35 is a sen­

tential variable, tfAen 93V -v- 93 ami -v 23V ^ -v 93 belong to 3#
but not to 3ft).

T heorem 1.6. For every sentence 91thefollowing conditions are

equivalent: (i) 91e 3ft* (ii) ^"^91 e 3ft> (iii) ^91 e 3ft.1
An exact proof o f these theorems (as well as of all the results
given below) must of course be based upon a suitable axiom-
system for the meta-sentential calculus and not only upon the
definitions of the notions involved. But it would be superfluous
here to formulate such an axiom system explicitly.2

§ 2. T he Ma t r ix M ethod
The definitions in § I afford no criterion that would enable
us to decide in each particular case whether a giv^en sentence 21
is provable in the two-valued or in the intuitionistic calculus.
Such a criterion is provided by the so-called ^matrix method.3

D e f in it io n Let there be given a set Hr of arbitrary

2 .1 .
elements, an dement A e ■#', three binary operations »->, , and
Jo, and a unary operation It is assumed that iF* is closed
under these operations and that the following holds:
if Y e iT and A *-> F — A, then Y — A.
Under these assumptions, the ordered sextuple
M= [iT, A , »->, cp . j* , nJ]
is called a (normal logical) matrix.
Note 2.2. I f M= , A, »->, T, th, ~ ] is a matrix, then iT
is sometimes called the value system, A the designated element,
and *-►, T , Jb , ^ the fundamental operations (first, second, etc.)
of M.
1 The first part of Th. 1.5 follows easily from 1.2-1.4 (it suffices to show that
every sentence 91 of the form 1.2 (xi) belongs to For the second part of.
Heyting, A. (28), p. 56. Th. 1.6 was stated in Glivenko, T. (19).
2 In this connexion see V III, p. 173, IX , p. 282.
3 Cf. IV, p. 41. The concept of matrix i3 there rather more widely con­
ceived than here, since matrices with more than one designated element are also
considered (cf. 2.1, 2.2).
XVH , § 2 S E N T E N T IA L CALCULU S AN D TO P O LO G Y 426

2.3. Two matrices Mx = .[iT l9 A 19 *->A, cpv cA>i,

D e f in it io n
and M2 = [ ^ , J.2, »+2, ^ 2? cb2> ^ 2] are said t° isom orphic
if there is a function F which maps on in a one-one
fashion and satisfies the following formulas: F(Af) = A 2)
F ( X *+x 7 ) - 7 (X ) ~ 2 F(Y), F ( X t . 7 ) = F(X) t 2 7 (7 ),
7 (X dhi 7 ) - 7 (Z ) db2 7 (7 ), and 7 ( ~ t X ) = ~ 2Z (X ) for
aU X , 7 e
C orollary 2.4, Every matrix M is isomorphic with itself; if
M2 is isomorphic with M2, then M2 is isomorphic icith M-^* if MA is
isomorphic with M2 and M2 is isomorphic with M3, then is iso­
morphic with M3. [By 2.3]

2.5* Let M = \ ifr \ A ,

D e f in it io n <L , ~ ] 6 e a matrix and
21a sentence. The following formulas define (recursively) a function
7$tMwhich correlates an element
7m
>M(Xlv..,Xn,...)e ^
with every infinite sequence of elements X v ..., X n,.,. g If/'\
(i) 7*-M(X l5...,X n,...) = X p if 21 = ©p (p = 1, 2,...).
(ii) 72lM(X l5..,JX n,...)

i / 21 — © -> £ (where © and £ are any sentences).

(iii), (iv) analogously for the operations T and v , or db and A .
(v ) Z ^^X ^ .... X n,...) = ~ 7 ^ M(X lf..., X w,...) if 2{ — ©
(where S is any sentence).
T7csay tfiesentence 21is satisfied by £Ae matrix M, in symbols
21 6 C(M). i f Z «fM(X lf..., X *,...) = 4 /o r aB ^ ..... X n,... e

iVo^e 2.6. It is sometimes said that a matrix M is adequate for

the system S of sentences if ffi(M) = 6 .
2.7. I f the matrices
Co r o t j a r y and M2are isomorphic, then
®(MX) - ®(M2). “ [By 2.1 2.3, 2.5]
1 In the formulation of 2.5, we could use functions with finitely many
variables, but that would create certain technical difficulties in our further
considerations. Another, although equivalent, definition of the set <s(M) was
given in IV .
42C S E N T E N T IA L C A L C U L U S A N D T O P O L O G Y X V II, § 2

If
D e f in it io n 2 . 8.
M = [ i f , A , *■>, T , Jo, ~ ] and Mx = [ ify, A , »-*•, T , J>,
aretwomatrices a n d ifif £ s IT , then Mxi,s called a subm atrix 0/ M.
I f M is any matrix and Mx is a submairix of
Co b o l la e y 2.9.
M, then ®(M) £ «(M X). [B y 2 . 1, 2.5, 2 . 8]
2 . 10. We denote by ZK the ordered sextuple
D e f in it io n
[ i f , 1, *-*, Jo, where i f = ( 0 , l), x = l-x -\ -x .y ,
s ! T y = x-\-y—x .y ,x J o y ~ x .y ,a n d ~ x = 1—xfor aUx,y e i f .
The follow ing result is well known:
T h e o r e m 2.11. ZK is a matrix and ®(ZK) = 3 ft .1
F or the system 3 ft there is, in contrast to 3 ft. no adequate
m atrix with a finite value system .2 W e can, however, con­
struct an infinite sequence o f m atrioes IK*,..., iK „,... with finite
00
value system s such that JX ®(IKn) = 3 ft. W e shall now de-
1
scribe the construction o f this sequenoe .8
Let M — [#^ , B, »-*, t , JU,
D e f in it io n 2 . 12. be a matrix
and A any element which does not belong to if* We put:
(i) i f * = if+ {A }\
(ii) l H * r = I n 7 if X t Y e i f and X +*Y B;
X *+ *Y — A if X , Y e i f and X ++ Y =* B; X *+ *A ~ A for
X e i f * ; A *+*Y * 7 for Y e i f * ;
(iii) X y *Y = X r Y f o r X , Y e i f ; Z r * A = A r * Z = A
for Z e i f * ;
(iv) X L * 7 = X j * Y for X , Y e i f ; Z ^ * A = A<L*Z =-- Z
for Z e H r*;
(v) ~ * X = ~ X if X e i f and ~ X j* B; ~ * X — A if
X e i f and ~-5T = B; r*J*A — ■ —‘B.
The ordered sextuple [ i f *, A , »-**, T * , Jo *, <--'*] is denoted
by M*.
1 2.11 easily results from the well-known theorem according to which the
system 3& is complete. The first completeness proof for is due to Poet, E.
(60), pp. 180
3 Cf. Gddel, K , (23), p. 40.
4 This result is due to Jaskowski, S, (36). (Our account deviates only in in­
essentials from that of Jankowski. The operation T of Ja&kcwski is replaced
by the operation * defined in 2.12, which serves the same purpose.)
X V n, f 2 SENTENTIAL CALCULUS AND TOPOLOGY 427

Note 2.13. The operation on matrices just defined is not

unambiguous beoause its result depends on the choice o f the
element A. This fact does not affect the subsequent considera­
tions because all matrices whioh can be obtained from the given
matrix M by means o f the operation * are isomorphic with one
another. The ambiguity o f ‘M*’ is, moreover, avoidable by
constructing a set-theoretical function F which correlates an
element F ( i f ) non-e i f with every system i f , and then replacing
'A ’ in 2.12 by *F ( i f y (certain difficulties which may arise in
connexion with the theory o f logical types will here be ignored).
2.14. J/M is a matrix, then M* is also a matrix and
Co b o lla b y
£ ®(M); if the matricea Mx and Ma are isomorphic, then
M* and M* are also isomorphic. [By 2.1, 2.3, 2.6, 2.12]
D e f in it io n 2.15. Let n be a natural number and
M = [ i f , A , »-*, T , Jo,

a matrix. We put:
(i) i f n = the system of ordered n-tuples [X lv .., X n] with
X t , . . . , Xn Gi f ;
(ii) A n — [X j,..., X n], where X x — ... — X n = A ;
(Hi) [X » ,...,X J » » [ 7 x,...,Yn] = [X x » Y x,...,Xn » Y n]for
X x,..., X n, !F],..., Yn g ;
(iv), (v) analogously for T and ;
(vi) [X x,..., X n] — [ ~ X 1,..., ~ X n] for X x,...,Xn g i f .
[ i f n, A n, cA, ” , ^ n] is called the ?ith power of the matrix
M and is denoted by ‘ Mn’ .
2.16. I f n is a natural number and M is a matrix,
Co b o l la b y
then Mn is also a matrix and we have ®(Mn) — G(M); if Mx and
Ma are isomorphic matrices, then MjJ and Ma are also isomorphic.
[B y 2 . 1, 2.3, 2.5, 2,15]

D e f in it io n 2.17. IKX = ZK, IKn+1 = ((IKn)n)* for every

natural number n.
On the basis of this definition the following theorem can be
proved:
428 S E N T E N T IA L CALCULUS A N D TO P O LO G Y XVII, § 2

T heorem 2 .IS. In order that 2 I e 3 it , it is necessary and

sufficient that 21 e G(IKn)/or every natural number n; in other words

n « ( ik j -
n=-l
Note 2.19. It is known that this theorem can be improved:
with every sentence 21 a well-determined natural number can
be correlated (depending exclusively on the structure of this
sentence) such that the formulas 2Ie©(IKrJ and 21 e are
equivalent.1
The decision criterion mentioned at the beginning of this sec­
tion is provided by 2.11 for the system 3 & and by 2.IB in its
improved form just mentioned for the system

§ 3. T opological Spaces 3
We first recall some familiar topological concepts:
D e f in it io n3.1. A non-empty set 8 is called a topological space
(with fundamental operation ~), if the follovnng conditions arc
satisfied:
(i) if X c S, then X = X c S;
(ii) if X c S and X consists of at most one dement, then X = X\
(iii) i f X c= S and ¥ c S, then X + Y = X + 7 .
Note 3.2. Let S he a topological space and A a non-empty
subset ot S. We now define an operation ~~'A) relative to A by
the formuia X U} A . X for every X c A . On the basis of
3.1 it can then easily be shown that A is a topological spa.ee with
the fundamental operation ~(A'}: such a space is called a sub­
space of S.
D efin itio n 3.3. I f 8 is a topological space, , a set X is said to
be open (in S), in symbols X e 0(8), if X = S—8 —X.
D efin itio n 3.4. A subset X of a topological space 8 is said to
be dense (in 8) if X = 8 .
1 See p. 420, note X
3 Another decision criterion for the intuitionistic calculus was given in
Gentzen, G. (18).
3 For what follows ei\ Kuratowski, C. (41), in particular pp. 15 ff., 38, 40,
82 f., 95, and 101 ff.
X V n, § 8 S E N T E N T IA L CALCULUS AN D TO P O LO G Y 429

D efin ition 3.5. The topological space 8 is said to be isolated

if x non-e $ —{#}, and dense-in-itself if x e S—{x}for every x e S.
Corollary 3.6. A topological space 8 is isolated if arid only if
X = X for every X £ 8. [By 3.1, 3.5]
D efin ition 3.7. A topological space S is called norm al if,
for any two sets X x ^ 8 and X 2 ^ 8 such that X x. X 2 == 0, there
exist two disjoint sets Yx, Y2 e (9(8) such that, X x s Yx and X 2 S Y2.
D efin ition 3.8. We say thaJt the topological space 8 is a space
with a countable basis if there is an infinite sequence of non-empty
sets X x,.t., e @(S) such that every non-empty set Y s (9(8}
can be represented in the form T ~ (where
il3..., in9... is a sequence of natural numbers).
For later applications we shall distinguish another special
class of topological spaces:
3.9. A topological space 8 is called an S'-space if ii
D e f in it io n
satisfies the following condition:
For every natural number n and every non-empty set A f= (9(8).
there exist non-empty, pair-wise disjoint sets
B n e (9(8)
for which
(i) I?i + . . . + l ?n S A and B x+ . . . + B n # A,
(ii) A —(Bx-\-...A-Bn) 3 A —A,
(iii) B x. ..... Bn S A —
T heorem 3.10. Every normal and dense-in-itself topological
space 8 with a countable basis is an E-space.
Proof.1 Aceording to 3.8 there exists an infinite sequence of
sets < v ~ , Ck,... with the following properties:
(1) the sets Cx...., Gkt... are open and non-empty,
(2) every open and non-empty set X can be represented in the form
x = cr<I+ —+c<fc+ —•
Since, by hypothesis, the space 8 is normal and dense-in-itself,
1 Originally I had proved the theorem for the Euclidean straight line (and
its subspaces dense-in-theroselves). I am indebted to Professor S. Eilenberg
for the general proof.
480 S E N T E N T IA L CALCULUS AN D TO P O LO G Y XVII, 5 3

we easily obtain from (2), b y means o f 3.5 and 3.7,

(3) for every non-empty set X e 0(8) there exists a set Ok such that
@k £ X and Ck ^ X .
W e now consider any natural num ber n and a set A for which
(4) A e 0(8) and A ^ 0 ;
we shall construct sets Bv ..., Bn whioh satisfy the follow ing
conditions:
(5) B ,,..., Bn are open, non-enypty, and pair-wise disjoint',
(6 ) Bi-\-...-{-Bn c A a n d B x-\-...+B n ^ A ',
(7) A — (JBj-J-... 4 *Bn) 2 A —A ",
( 8) S j . ..., B n 2 A — (jB1-f-...-|-JBn).
W e carry this out first for the case » = 1 ; that is to say we
construct a set B such that
(9) B e 0(8), B 0 , B £ A , and B ^ A,
(10) A —B 2 S.— A and B 2 A — B.
A Bet B o f this kind is most easily obtained in the following way.
W e consider the sets A k = Ck.(A—A) and choose from each
such set (if it is not empty) a point ak, let D = {d j,...,
I t is easy to see that J5 2 A —A (for otherwise, according to ( 2 ),
there w ould be for X = 8 —D a set C* S S—B c S—D, having
an elem ent in com m on w ith A —A, and this w ould contradict
the definition o f D). The set D is at m ost denum erable and not
empty (apart from the trivial case A —A = 0 ). W e order the
elem ents o f D in an infinite sequence dx,..., dk)... in such a way
that in this sequence every elem ent x e D occurs either only
once or infinitely m any tim es, according to whether x e D —{x)
or not. M oreover, since 8 is a norm al space w ith a countable
basis it is metricizable: with every pair o f points x , y e S a real
num ber \x— y\ can be correlated, the so-called distance between
x and y, in such a w ay that the follow ing conditions are satisfied:
(i) for aU x , y e S the formulas \x—y\ — 0 and x — y are
equivalent-,
(ii) |x —y\-{- \x—z\ > \y—z\ for x , y , z e S;
(iii) if X £ S, then x e X if and only if for every r > 0 there is
an dement y e X with \x— y\ < r.
XVn, § 3 S E N T E N T IA L CALCULU S A N D TO P O LO G Y 431

N ow , b y (iii), since D £ A , for every x e D there are points

y e A whioh lie as near to * as w e please. A ccordingly, we can
choose for every natural h a poin t ek e A w ith
I**—«*l < l/*»
and we put B — e*,...}; it can be shown w ithout diffi­
culty that B satisfies form ulas (9) and ( 10) (the fa ct that & is
dense-in-itself plays an essential part in this).
The case n — l has thus been dealt with. W e turn now to
the general case, m aking use o f the set B just uennad.
W e first pu t, for every X £ 8,
(11) X+ =
Prom (11), 3.1, and 3.3, w e easily obtain the follow ing rules
for calculating w ith the operation +:
(12) (X +)+ — X+ e <D(S) for every X £ S-,
(13) X £ X+ and X =* X * for every X e ^ ( jS);
(14) if X e 0{8), T s S and X s T , then X s
(15) if X 1+ . . . + X n £ 8 and the sets X ^ ..., X H are pair-wise
disjoint, then (X 1-K » -i-X * )+ = ,
Further, we define b y recursion an infinite sequence o f sets

(16) Gx = ( 7 j, where k is the smallest natural number for which

Ck £ B and Gk =£ B\
(17) Ol+l - - C£ where k is the smallest natural- number for which
Gk £ B — f l f h Z + G , and Gk ■-/. B -G ^ fZ T -%
On iho basis o f ( 1), (3), ( 9 ), ( 12), and (13) it is u e s i shown
(by means o f an easy induction) that, b y (16) and (17) a set G>
is aotually correlated w ith every natural number l and also that
these sets Gx satisfy the follow ing conditions:
(18) G, is open, non-empty, and such that G f *= £ o*, £ B for
every l;
(19) the sets Gls..., Gt,... (and hence also the secs (2X,..., (?/,...) are
pair-wise disjoint.
L et G s»s B — Gr14 - ...+ ( ? j+ .... In view o f (9) we have C e G{8).
I f G were n ot em pty, then b y (9) there w ould be a number
432 S E N T E N T IA L CALCULUS AN D TOPOLOG Y XVTL § 3

k for which Ck ^ C and Ck ^ C . W e should then a fortiori have

Ck c B — and Gk ^ B~~G1+ ... + Ql for every l, and
we could infer from that by (16) and (17) that Ck is identical
w ith one o f the Gh whence, in consequence o f ( 1) and (13),
C* = G, £ B - G
The set G, must thus be em pty, in contradiction to (18). Con­
sequently C ~ 0 , i.e. B £ (vr r ~ ..-f &r'r • A ccordingly, by
( 10) we have i f £ cl B c =Gxy - G r I -.... O ntheother
hand, from : lh ; aad (9) we get
fr^-j-... -f- ... £1 B S/ A.
so that finally
( 20) j - g; .
W e shall now show the follow ing.
(21) I f X c 5;!iS‘b X c A , and X non- c; ' J , +
there exist infinitely many numbers l for which X .0 , ^ 0.
F or this purpose we consider an open set X £ A which has
elements in com m on with only a fixate num ber o f sets G,. There
is thus an lX} such that X . G-. =~ 0 for every l > Z0. The set
Y = X — 6V r has then no elem ent in com m on with the
whole sum (?1+ ..-4-0^,-}-... . Since in addition Y e $ ( 8 ) , we
also have —- r ^ 0+ ... =— 0 ; thus. on account o f ( 20),
Y ,A —- 0 a ad j fortiori Y ,A ~ 0 . On the other hand.
f c l c i ;

consequent s/ o, i.e, X ^ 6S4 ...4 From this, by means

of (14), (15 / 4 9}. we obtain
X 4 — Gx + ...+<?£,
whence, in view of (18),
X 4 <?*+...+ (A £ + *
We have thus- shown that every open set X 4* A, which has
elements In common with only finiSSy many secs Gh is included
in Gi+...~H-o * Hence by contraposition we obtain (21),
Let
(22) GPl..... Cw ... be those sets of the sequence Cy,..., Gk.... which are
included in A but not in # !+ ... + #*4“ — *
XV H , § 3 S E N T E N T IA L CALCULU S AN D TO P O LO G Y 433

The set Gv ..., m ay be divided into n system s o f sets

u f x,..., in such a w ay that the follow ing conditions are satis­
fied:
(23) = {GXi-* >
(24) the systems are non-empty and pair-wise disjoint;
(26) for every set CPk. k = 1, 2 ,..., awd! /o r every number j }
l j ?C n, there is a set X such that
CPk. X # 0 .
T o prove this we apply the follow ing procedure. In view o f ( 1),
( 22) , and ( 21), there certainly exist n sets G^,..., G^ which have
no elements in com m on with GPl. W e include in the system
the sets 1 < j < n, Sim ilarly, there exist n sets Oti n.
w hich are distinct from G^,..., Gln and have no elem ent in com ­
m on w ith CPi: the set , 1 < j < n, is again included in
This procedure can obviously be continued w ithout end. The
sets Gt possibly then remaining are subsequent]y arbitrarily in ­
cluded in. the system s (e.g. all m ay be included in
the system ^ x).
W e now put

(26) Bj = 2 i /o r j = 1, 2,..., n.
XeJT*
From (18), (19), (23), (24), and (26) it is seen at once that the
sets 2^ ,,.., Bn just defined satisfy the condition (6 ). B y (IS).
(23) , and (26) we have ■ # * + „ . — Gj1- f ^ B and
consequently A — (1 ?!+ ... ~r~Bn) 5 A —B. H ence b y means o f (9)
and ( 10) we obtain ( 6) and (7). Finally, let us suppose that the
form ula B} 2 A — ...+ i ? n) does not hold for a given
j , 1 < j < n. W e thus have

A - B j n o n - ^ B x-\-...+B n = 6 ^ + ...+ G^+...,

whence, b y virtue o f (4), A —Bi is open. B y means o f ( 2 ) and

( 22) we infer from this the existence o f a set CPkwhich is included
in A —JBy and has no elem ent in com m on with B3, B ut this is in
obvious contradiction to (26) and (26). A ccordingly, our supposi
tion is refuted and ( 8) holds.
434 S E N T E N T IA L CALCULUS A N D TO P O LO G Y XVII, § 3

W e have thus constructed (for every natural n and every non­

em pty open set A ) sets Bv ..., Bn w hich satisfy the conditions
(5 )-(8 ). H ence, b y 3.9, $ is an J?-space, w hich was to be proved.
Note 3.11. It is clear from 3.10 that the E uclidean spaces (of
an y num ber o f dim ensions) belong to the jEJ-spaces. A lso, every
subspace o f a Euclidean space w hich is dense-in-itself is an E-
apace.

§ 4. T opological I n terpretatio n op th e T w o -valu ed and

op THE iN TU m O N ISTIO SENTENTIAL CALCULUS
W e now define, for subsets o f an arbitrary topological space,
four operations w hich we denote b y the same sym bols as were
used for the operations on sentences whioh were discussed at the
beginning o f § 1.
D e f in it io n 4.1. I f S i* a topological space, we put, for all sets
X<=8undYzS,
(i) X - + Y «= 8 - X ^ Y ,
(ii) X v Y — X + Y (the ordinary set-theoretical sum),
(iii) I a Y — X .Y (the ordinary set-theoretical product),
(iv) ^X » X 0 (= 8 - X ) .
C o r o llar y 4.2. (i) I f S is a topological space, X £ 8 and
Y £ 8, then X - + Y , e 0 (5 ), and in fact X -^ Y is the largest
open set Z for which X .Z £ T, and - v l is the largest open set
disjoint from X .
(ii) I f in addition X , Y e <9(8)thenwe alsohaveX V i e 0(S)and
X a Y e 0(8), and in fact I v 7 is the smallest open set which
includes X and Y, and X A Y is the largest open set which is in­
cluded in X and Y. [B y 3.1, 3.3, 4.1]
Co r o llar y 4.3.I f 8 is a topological space, X £ 8 and Y £ 8,
then X -> Y — 8 if and only if X £ Y ; in particular S - + Y = 8
if and only i f Y — 8. [B y 3.1, 4.1 (i)]
C orollary 4.4. I f 8 is a topological space and X £ 8, then
X is dense in 8 if and only if -v -v-X = 8. [B y 3 .1 ,3 .4 ,4 .1 (iv)]
D efin itio n 4.5. The ordered sextuple [0 (5 ), 8, v , A,
where 8 is a topological space, is denoted by 0 (5 ).
3CVH. § 4 SENTENTIAL CALCULUS AND TOPOLOGY 43*

T h eorem 4.6. For every topological space 8, O(S) is a matrix.

[B y 2 . 1, 3.1, 3.3, 4.2, 4.3, 4.5]

The m atrix 0(8), like every other m atrix, uniquely determines

a system o f sentences, nam ely <E(0(8)). W e shall investigate the
relation o f this system to the system s 3 ft and 3 ft in detail.
L emma 4.7. I f 8 is a topological space and SP (<S, 0), then
M — \$P, 8, v , A, is a evJbmatiix of 0 ( 8 ) which is iso­
morphic with ZK.
The p ro o f (b y 2 . 1, 2.3, 2 . 8, 2. 10, 3.1, 3.3, and 4.1) presents no
difficulties.
T heobem 4.8. For every topological space 8 we have
® ( 0 (fif)) s 3 ft.
[B y 2.7, 2.9, 2 .1 1 ,4 .7 ]

L emma 4.9. I f 8 is any topological space and 91 on axiom of the

intuitionistic sentential calculus, then 91 e ® ( 0 (8)).
Proof. In accordance w ith 1.2 it is necessary to distinguish
in the p ro o f ten cases according to the form o f the axiom 91.
Sinoe the m ethod o f arguing is in all oases nearly the same,
we shall consider only one case in detail, say 1.2 (ix).
L et then
1
( ) 21 — ^- 93-> (© -* $ ),
where © and <£ are any sentences. W e construct in accordance
w ith 2.6 (and w ith the help o f 4.5, 4.6) the functions F^,o<s>>
Fse.om> ^ c,o <sy W e consider further an arbitrary sequence o f
sets X j,..., X n,... e 0(8) and put

( 2) F%,o(S)(X ^,..., X M,,..) = X , -^8 ,o(<s)(Xi,..*, X n,...) — X ,

B y 2.5 (ii), (v),£rom ( 1) and (2) we obtain X = ->• (F - » Z),

whence, b y 4.1 (i), (iv),

(3) X = ( S - 7 ) -r (Y -*■ Z) — S - ( S - Y ) - ( S - Y - Z ) .

Since, b y virtue o f 3.1 (iii), Y ~ Z Y and consequently

436 S E N T E N T IA L CALCULUS AN D TO P O LO G Y XVII, { 4

S—Y c S—Y —Z , it results from (3) and 3.1 (ii) that X = 8.

B y ( 2) we thus have
F k,o(£)(-^i>*••> = S for all X n9... & ${&)•
H ence, b y 2.5, we obtain 91 e ® ( 0 (S))9 q.e.d.
L em m a 4.10. Let S be any topological space and 31, 93, ® three
sentences such that 91 = 93 -*> ®. I f 91, 23 e ® ( 0 ($ )), aiteo
® e ® ( 0 ($ )); in other words, the system ® ( 0 (# )) is cZosed wnder
tfAe operation of detachment.
Proof. In accordance with 2.5 (and with the help o f 4.5, 4.6)
we construct the functions F%oiS), F%0{S)9 and F$tQ(ig); we
then have, for all sets X l9..., X n,... e @(S)9
( 1)
^ FfB,o(S)(Xl9: ; X n,..>) -> F€%0(S)(Xl9.>.9x n,...),
and, since 91, 93 e ® (0 ($ )),
(2) %n9...) = S — F#t0(S)(Xl9...9X n,..) .
B y virtue o f 4.3 the form ulas ( 1) and ( 2 ) yield
-^ ip ***) ^ & f w -*£«>••• g &{S)9
whence, b y 2.5, (£ <=® (0 ($ )). The system ® (0 (S)) is thus closed
under the operation o f detachm ent (cf. 1.3), q.e.d.
T heorem t,11. F o r e v e r y to p o lo g ic a l s p a c e S w e have

35t c ® ( 0 (S)).
[B y 1.4, 4.9, 4.10]
T heorem 4.12. For every topological space S and every sen­
tence 91 the conditions 91 e 3$t cwd ^ ^91 e ® ( 0 (£ )) are equiva­
lent. [B y 1.6 , 4.8, 4.11]
In view o f Ths. 4.8 and 4.11, the double inclusion
35t £ ® (0 ($ )) £ 3*t holds for every topological space S. W e
shall now show that there exist spaces 8 for which (£(0(S)) = 3&
and also spaces S for whioh ® ( 0 ($ )) = 351. In fa ct the first
equality holds if and only if S is an isolated (and thus, so to
speak, a degenerate) space. The second holds for all i?-spaces,
and thus in particular for all norm al spaces which are dense-
in-them selves and have a countable basis (cf. 3.9-3.11).
XVII, § 4 S E N T E N T IA L CALCULUS AND TO P O LO G Y 437

L em m a 4.13. Let 8 be a topological space. In order that every

sentence 21 of the form 3 1 = (-v-23 -> 23) 23 (where 23 is any
sentence) should belong to (£(0(8)), it is necessary and sufficient
that S be isolated.
Proof. I f 8 is an isolated space, we easily obtain from 3.6 and
4.1 (i), (iv) the form ula ( ^ 1 -*> X ) X = S for every X £ S,
and hence b y means o f 2.6, 4.5, and 4.6 (just as in the p roof
o f 4.9) we conclude that every sentence 2 1 = (-v-23 23) -> 23
belongs to (£(0(8)).
Suppose now , conversely, that (£(0(8)) contains all sentences
21 = ('v®->»)-► 33.
B y 1.1 we can, in particular, assume that 23 is a sentential
variable, say S = 93x. A ccording to 2.6 and on account o f 4.5,
4.6 we then have for every sequence o f open sets X x,..., X n...

and F^>0{S)(X1,..., X n,...) = (^-Xx ->■ X x) -> X v

Since 21 e (£(0 (<S)), we obtain
( ^ X 1- r X l) - + X l = S,
whence, on account o f 4.3, ( X x -> X x) £ X x and further, b y
virtue o f 3.1 and 4.1, 8 —8 —X x £ X x for every X x e 0(S).
If, in particular, x is an elem ent o f 8, we conclude from 3.1 and
3.3 that 8 —{x} e 0{S ) ; consequently

8 - 8 - 8 ^ { x } £ S-{x},

and therefore S—S—{r } 2 {x}\ from this we see at once that

8 —{* } # S and 8 —{a:} = 8 —{*}. W e thus have
xn on -e S — {x} for every x e 8;
bu t this means that the space 8 is isolated (cf. 3.6). Lemma
4.13 thus bolds in both directions.
T heorem 4.14. Let S be a topological space. In order that
® (0 (S)) = 3& , it is necessary and sufficient that 8 be isolated.
Proof. L et 8 be an isolated space. Then, according to 1.2 ,
4.9, and 4.13, © (O (8)) contains all axiom s o f the tw o-valued
438 S E N T E N T IA L CALCULU S AN D TO P O LO G Y 2001, § 4

sentential calculus and, b y 4.10, is d osed under the operation

o f detachm ent. Consequently, b y 1.4 W6 have

3* s *(om,
whence, on account o f 4.8, <E(0(8)) *= If> conversely, this
latter equality is satisfied, then the system <E(0(iS)) contains
in particular all sentences 91 o f the form 91— ( ^ 93 93) -> 93
(cf. 1.2 (x) and 1.4); in view o f 4.13 the space S is therefore
isolated, q.e.d.
B efore continuing we shall subject the operations -> and -a.
to a relativization (cf. 3.2).
I f 8 is a topological space and A £ 8, then
D e f in it io n 4.16.
for all sets X £ 8 and 7 £ 8 we put:
(i) 7 = A -X -7 ,

(ii) X = X 0 ( — A~~Xj\
A A

Co r o llar y 4.16. I f 8 is a topological space, A £ 8, X £ 8,

and 7 £ 8 , then I - > F s i and -yX s A ; if moreover A e 0(8),
then X -> 7 , ^ X e 0(8). [B y 3.1, 3.3, 4.15]

Co b o l la r y 4.17. Let 8 be a topological space and A £ 8.

(i) I f X £ A and 7 £ 8, then we have X -*■ 7 — A if and
only if X £ 7 .
(ii) I f 7 £ A , then A - > 7 — A if and only if 7 — A .
A
(iii) X -*■ A = A for every I s i .
Al
(iv) A -* Y = Y for every open set Y c A .
A
(v ) = A 0 = 0 and ^-0 = 0 0 — A.
A A A A
[B y 3.1, 3.3, 4.15]

Co e o l la b y 4.18. I f 8 is a topological space, then

X->y = X -*7 and -v X = -wX
8 S

for all sets I s 8 and 7 £ S. [B y 4.1 (i), (iv), 4.16]

The follow ing lem m a is a generalization o f 4.7:
XV n, §4 S E N T E N T IA L CALCULU S AN D TO P O LO G Y 439

L emma 4.19. I f S is a topological space, A £ S, A ^ 0, and

ST — {A , 0}, then N = [3~, A , V , A , •*.] is a matrix which is
isomorphic with ZK .
[B y 2 . 1, 2.3, 2 . 10, 3.1, 4.1 (ii), (iii), 4.17 (iii), (y)]
L e m m a 4.20. Premisses:
(a) 8 is a topological space',
(fi) A , B e 0(8), B e A, B ^ A , T ^ B 2 I - A and B 2 A - B ;
(y) SP is a system of open sets X £ B ;
(S) if X , Y e ? and X - Y ^ 0 , then X - Y ^ S - B )
(e) M sb= [SP, B>-+> V , A , ~ ] is a matrix-,
(£) 3T « . SP+{A) and N = [3Tt A , v , A,
Conclusions:
(i) 3 " is a system of open sets I s i ;
(ii) if X ,Y e3T and X —Y 0 , then X —Y 2 A —A ;
(iii) N is a matrix and N = M*.
Proof. A coording to the premisses (a) and (/?), and in view
o f 3.1, w e have B 2 B and B 2 A —B. whence B 2 A; and
since A 2 B, we get
( 1) A —B and A —A s B —B.
From (y) and ({) w e obtain at once the conclusion (i):
( 2) 3* is a system of open sets X £ A .
W e shall next prove the oonolusion (ii):
(3) if X ,Y e 3 " and X —Y ^ 0, then X —Y 2 A —A.
In fa ct, i f X —Y 0 , then Y # A b y (2) ; thus Y e SP on
account o f (£). I f also X e S P , then it follow s from (8) and ( 1) that
X -Y 2 B -B 2 A -A .
B ut, if X non- e SP, then X — A , and, if Y ^ B, then b y (y)
we have B —Y ^ 0 and hence b y (8) and (1) we obtain
X -Y 2 B -Y 2 B -B 2 A - A
(since B e SP b y virtue o f («) and 2. 1). F or X — A and Y = B
the form ula X —Y 2 A —A follow s directly from (/9). (3) thus
holds in all oases.
440 8 E N T E N T IA L C A L C U L U S A N D T O P O L O G Y XVU, § 4

The operation -> has the follow ing properties:

(4) if X ,Y e y , then either X -> Y — A or — X -+Y , according
A JB
to whether X -> Y — B or ^ B:
B
(5) X -> A — A for every l e J ;
( 6) A - > Y = Y for every Y e f .
A
If, in fact, X , Y e and X -*■ Y — B, then b y 4.17 (i) we have
jb

I s f and X -» Y — A (since X £ B £ A on account o f (/5)

A _____
and (y)). B ut if X ->• Y = B — X —Y ^ B. then it follow s from
JB

(8) that X —Y 2 B —B, and thus, in view o f ( 1), we have

A - X - Y £ A -(B -B ) = (A -B )+ B = B
and therefore A — X —Y £ B — X — Y ; on the other hand, since
A 2 B and henoe A — X —Y 2 B —X —Y. we dually get
A -X -Y = B -X -Y , i.e. X Y == X - > 7 .
A B

In thiB w ay (4) is proved. (5) and ( 6) result im m ediately from

4.17 (iii), (iv), and (2).
From (2), b y 4.1 (ii), (iii), we also obtain
(7) Z\J A — A y Z — A and Z a A — A A Z — Z for every
Z ef.
The operation -v- satisfies the follow ing conditions:
(8) ^A “ 'v l? 0;
A B
(9) if X e £?. then ^ 1 = 4 or — -^X, according to whether
A JB
= B or # B.
B
( 8) follow s directly from 4.17 (v). W ith the help o f 2.1 and
(«) we infer from ( 8) that 0 e ; in view o f this we put in
(4) 7 = 0 and b y means o f 4.15 (ii) we im m ediately obtain (9).
I f now we com pare the premisses (e), (£) and the form ulas
(4 )-(9 ) w ith 2.12 (i)-(v ), we see at once that
(10) N = M*,
whence, b y 2.14,
(11) N is a matrix.
W ith ( 2 ), (3), ( 10), and ( 11) the p roof is com plete.
XVn, § 4 S E N T E N T IA L CALCULUS A N D TO P O LO G Y 441

L e m m a 4.21. Premisses:
(a) S is a topological space;
05) Bu ...,Bn e<9{8),Bx,...,Bnare non-empty, pair-un.se disjoint
sets, Bx-{- **.~\~Bn “ B and Bx. ... *Bn 2 5
(y) /o r = 1, 2,..., n, SPp is a system of open sets X £ Bp;
(8) if X , Y eSfp (p = 1, 2 ,..., n) and X —T ^ 0 , then
X = Y 2 Bp- B p ,
(e) M = [if''', A , »-», T , JL>, ~ ] is a matrix-,
(£) forp = 1 , 2 , . . . , M„ = [<5% V, A , ^ a matrix
and is isomorphic with M;
(>j) i f is the system of sets X = X x+ ...-| -X n, where
X xe X ti e S?n, and P = [$?, B , A , V , 'A -
x» £>
Conclusions:
(i) Se' is a system of open sets X £ B-,
(ii) if X , Y e 6? and X - Y # 0 , then X ^ Y 2 B - B ;
(iii) P is a matrix and is isomorphic with M77.
Proof. The assertion (i), i.e.
( 1) 9 is a system of open sets X c B,
results easily from (j8), (y), and ( 77); since, b y 3.1 and 3.3, every
sum o f open sets is itself an open set.
In, order to prove (ii) we consider tw o arbitrary sets X ,Y e 9
such that X —Y ^ 0. B y ( 17),
X = X x+ . . . + X n and Y = T1+ ...+ Y n9
where X p, Yp e 9 p for p = 1, 2 ,..., n. Since, in view o f (y) and
{fi), we have X p s Bpi Yp ^ Bp and the sets Bn are pair­
wise disjoint we obviously have
x - y - ( z 1+ . . . + x j - ( i i + . . . + r n)
= ( x 1- r 1) + ...+ ( x n- r n);
if, therefore, X —Y ^ 0 , there m ust be aj>, 1 < p ^ n, such that
Xp—Yp # 0 . H ence b y (8) w e get X p—Yp 2 Bp—Bp and
consequently X —Y 2 Bp—Bp, because b y (a) and 3.1 (iii) '

z -r = z1-:r1-f...-fz„-r,,
442 S E N T E N T IA L C A L C U L U S A N D T O P O L O G Y XVH, § 4

But according to (fi)

Bp- B p 2 (Br-B)~-Bp = B - { B + B P) = B—B,
so that- finally X —T 2 B—B. Thus we have
(2) * f X ,7 e & a n d X -Y 0 then~X-F B -B .
We turn now to (iii) and in accordance with 2.15 (and making
use of 2.16 and (<?)) we construct the matrix
Mn =s [iFn, A n, *-+n, r n, Jb", <->].
By (i) (and 2.4), M is isomorphic with Mi,..., M„; hence by 2.3
there exist functions Flt..., Fn>which produce this isomorphism
and which in particular map iFonto Sv ..., Snin one-one fashion.
We now put
(3) F(U) = F1(U1)+ ...+ F n(Un)for U - {Uv ..., Un] elT*.
Since here FP(XJP) &Sfp (p = l , 2,...,n), whence by (iii)
Fp(Up) c B p, and the sets Blt...,Bn are pair-wise disjoint we
obtain, with the help of (17),
(4) the function F maps oFn onto £? in one-one fashion.
By means of 2.3, 2.15, and (/3) we conolude that
(6) F(A*) — Fx(A)-f- ...-}-Fn(A) — = B.
Further, let
(6) U « [Uv ..., Un.] g 1F Wand V — g tF*.
By 2.15, (6), (3), 2.3, (0, and 4.16 (i) we get
(7) F(U F) = F([Ux» Vv ..., Un~ Fn])

- % Flm + - + ( F n(un) f m )

= (B1- F 1{U1) - F 1{V1))+ ...+ (B n- F n(Un) - F n(Vn)).

For p ,q — 1 ,..., n and p =£ q, FP(UP) e Sfp, F^?') g SFv and
thus by (iii) and (ii) we have
Fp(Up) c Bp> Ffl(Fe) £ B „ Fp(Up).B a = Fp(Up) . F ^ ) =» 0;
since in addition Ba e 0(8), we also have
F,{UP) - F p(Vp).B t = 0.
XVn, § 4 S E N T E N T IA L C A L C U L U S A N D T O P O L O G Y 443

From this with the help o f (7) and 3.1 (iii) we obtain

(8) F ( l 7 ^ F ) = s (B1+ ...+ J 5 J -

- ( F 1(C71) - JF1(F1) + ...+ F n(C7n) - F n(Fn))

= 5 - ( F 1(C 4 )-F 1(F1) ) + ...+ ( F n(C7ft) - F n(FJ)

^ B - ( F 1(U1) + . . .+ F n{Un) ) - ( F 1(V{)+...+Fn( V ^

by virtue o f 4.16 (i), formulas (3), (6), and (8) yield
(9) F(U*+ *F ) « B - F ( V ) - F ( V ) - F(U) F(V) for aU
U , V e 1T*
In an analogous but more simple manner we obtain the
formulas

(10) F {U y »V) = F(U)VF(V)<mdF(U<k»V) - F (U ) a F(V)

fo rU ,V e1 T « ;
(11) F ( ~ nU) « - F(U) for U e #*».
B

B y means o f 2,1, 2.15, and 2.16 we infer from (4), (5), (9)—(11)
that B e S? and is dosed under V , A , -v ; hi view o f
B B
(*))> (1)* aJid 4.17 (ii), P is thus a matrix. We also oondude—
again from (4), (6), (9), (10), (11)—that the matrices P and Mn
satisfy the conditions o f the definition 2.3; consequently we have
(12) P is a matrix isomorphic with Mn.
B y (1), (11), and (12), all the conclusions o f the lemma are
satisfied.

L em UA 4.22. Premisses:
(a) 8 is an E-space)
(ft) M is a matrix;
(y) for every non-empty Set Be 0(8) there is a system $P of open
sets X £ A with the following properties:
(Yi) if X, Y e S? and X —Y ^ 0* then X — Y 2 B —B;
(y*) M' *** [ y , B, V , A , ■+*] is a matrix which is isomor-
Jo JB
phic with M.
444 8 E N T E N T IA L CALCULU S AND T O P O LO G Y X V II, § 4

Conclusion: for every natural number n and every non-empty set

A e <P{S) there is a system 2T of open sets X £ A with the following
properties-.
(i) if X , Y e f and X - Y ^ 0,then X - F A -A ;
(ii) N = \ST, A , V , A , -v ) is a matrix isomorphic with
(Mn)*. A
Proof. L et n be a natural num ber and A a non-em pty open set.
B y virtue o f 3.9 and in view o f (a) there exist set3 B lt..., Bn, w hich
satisfy the follow ing conditions:
( 1) Bx,...,B nare non-empty,pair-wise disjoint open sets;
( 2) = B £ A and B ^ A ;
(3) A ^ B sA -A ;
W B i ...... Bn 2 A —B.
From (2 )-(4 ) we easily obtain
(5) Bx ...... Bn 2 (A - A ) + ( A - B ) = A - B 2 B —B.
B y (jS) and (y) there exist system s o f sets SPX..... £Pn with the
follow ing properties:
( 6) for p = 1, 2 ,..., n,£Pp i s a system of open sets X 2 Bp ;
(7) if X , Y e SPp (p = 1, 2 ,..., n) and X —Y =£ 0, then
X = Y 2 Bp- B p;
(8) fo r p — 1, 2,..., n, Mp = \SPp, B p, - r , V , A , -y ] is a m a trix
Up Up
which is isomorphic with M.
L et us now pu t
(9) = [the system of sets X = . . . + X n, where

P = [ < ? ,£ , - + , V, A, - ] . ,
B B

B y (a), (/?), ( 1), ( 2 ), and (5 )-(9 ) the premisses o f 4.21 are satis­
fied. Consequently we have
( 10) SP is a system of open sets X £ A ;
( 11) if X , Y e S* and X —Y ^ 0, then X —Y 2 B —B;
(12) P is a matrix which is isomorphic with Mn.
XVH , $ 4 S E N T E N T IA L C A L C U L U S A N D T O P O L O G Y 446

L et
(13) and N = A,->, v : A ,
A -cL
B y (a), (l)-(4 ), and (9)—(13), the premisses o f 4.20 hold. Conse­
quently,
(14) F is a system of open sets X Q A ;
(15) i f X , Y e J T and X - Y ^ 0, t h e n X - Y 2 A ~ A \
(16) N is a matrix and N = P*.
From (12) and (16) b y means o f 2.14 we obtain
(17) N is a matrix which is isomorphic with (M7*)*.
In view o f (14), (15), and (17) the p roof o f Lem m a 4.22 is
com plete.
L em m a 4.23. I f 8 is an E-space, then for every natural number
n the matrix 0 ( 8 ) contains a submatrix which is isomorphic
with IKn.
Proof. B y means o f an inductive procedure we shall establish
a logically stronger conclusion, nam ely
( 1) for every non-empty set A e d)(8) there is a system 3T of open
sets X c; A with the following properties:
(i) if X 9 Y e 3T and X —Y ^ 0, then X —Y 2 A —A;
(ii) N - [AT, A , - k V , A , -v-] is a matrix isomorphic with
jA a
IKn-
In fa ct b y 2.17 and 4.19, ( 1) holds for n = 1. Assuming that
( 1) is satisfied for a given natural n we apply 4.22 (with M = IKn)
and w ith the help o f 2.17 we easily see that (1) also holds fo r n -j-l.
I f we now put A = 8 in ( 1) we at once obtain the conclusion o f
the lem m a from 2.8, 3.1, 3.2, 4.5, and 4.18.
T heobbm 4.24. I f 8 is an E-space, then G (Q (S )) = 3 ft.
Proof. B y 4.23 there is for every natural n a subm atrix Nn
w hich is isom orphic w ith IKn. N ow , if 31 e © ( 0 (# )), then b y 2.9
and 2.7 we have 31 e ® (N „) = ®(IKn) for n = 1, 2,... and con­
sequently, b y 2.18,31 e 3 ft. A ccordingly, © ( 0 ($ )) £ 3 ft; hence
b y 4.9 we at on ce obtain © (O (8)) = 3 ft, q.e.d.
446 S E N T E N T IA L CALCULUS AN D T O P O L O G Y X V II, § 4

Note 4.25. The follow ing theorem is easily established:

Let 8 be a topological space, A a non-empty open set £ 8,&~ the
system of open setsX s A , and N = [ ST, A , V, A, Then
A A
N is a matrix and we have (£(0(8)) £ ®(N ). If, in particular,
« (N ) = 3 ft, then tt(0 (S )) = 3 ft.
From this we see that the converse o f Th. 4.24 does not h o ld ;
the form ula ® ( 0 (8)) — 3 ft applies, for exam ple, to all norm al
spaces w ith a countable basis, w hich include a non-em pty open
set that is dense-in-itself, and, m ore generally, to every space 8
w hich includes an open .E-spaoe (of. 3.2), independently o f
whether 8 itself is an .E-space.
On the other hand there are also spaces 8 such that the m atrix
0 ( 5 ) is adequate neither for the system 3 ft, nor for the system
3 ft, bu t for an interm ediate system . Exam ples can be found
b oth am ong norm al spaces with a countable basis and am ong
those whioh have no countable basis. F or instance, norm al spaces
are know n whioh are dense in them selves and satisfy the oondi-
titm if X e 0(8), then e 0(8).
F or every such space 8 the system (£(0(S)) is, b y 4.14, distinct
from 3 ft. I t is also easily shown that this system contains all
sentences 'll o f the form 31 = - v i B V 'v ^ S and consequently
oannot coincide w ith 3 ft (cf. 1.5). The problem o f setting up
an exact correlation between the topological properties o f a
space 8 and the logical (or rather m etaldgical) properties o f
the corresponding system (£(0(8)) is still b y n o means com ­
pletely solved.
W e shall now put Some o f the results obtained into a more
intuitive and m ore lucid form .
L et $ be a sentential function o f the sentential calculus in
Which, in addition to the constants ‘ v ’ , etc., the sentential
variables ‘X \ 'Y*, ‘Z\... ocour. W e give the expression 31 the
follow ing schem atic form :
31 = (<f>(X,Y,Z,...)\
N ow let uS suppose that the variables ‘ X\ ‘Y ’ , ‘Z\... do n ot
represent sentences bu t denote sets o f points o f a topological
XVH . $ 4 S E N T E N T IA L CALCULU S AND TO P O LO G Y 447

space 8. W e give to the constants ‘v etc., the meaning

explained in 4.1. W ith this interpretation 21 is no longer a sen­
tential funotion, bu t a designatory function, which (exactly like
‘X ’ , T ’ ,..,) denotes a set o f the space 8. In view o f this we can
construct the follow ing sentences ^ and 21,:
2IX = ‘For any open sew X , Y, Z,... of the space 8 , the set
<f>(X, T, Z,...) is dense in 8 .’
21, = ‘For any open sets X , T, Z,... of the space 8,
<KX,Y,Z,...) = S ’ .
[S trictly speaking, 2lx and SC, are n ot sentences bu t sentential
functions, because free variables, e.g. ‘ 8 ’ , occur in them .]
W e now consider tw o expressions: 'Six holds (or is valid or is
satisfied) in the space S ’ and ‘ 21, holds in the space 8 ,’ The in­
tuitive m eaning o f these expressions seems to be com pletely
dear. Nevertheless, certain difficulties are encountered when
one tries to explain their m eaning in a strictly form al w ay .1 In
connexion w ith 2.5 the easiest w ay is to interpret the second
expression as being synonym ous w ith the expression: ‘21 is satis­
fied by the matrix 0 (8 )\ In order to construct a definition for
the first expression, we note that the sentence 2lx, b y virtue o f
4.4, adm its o f the follow ing equivalent transform ation:
‘For any open sets X , Y, Z,.,. of the space 8,
^^<j>(X,Y,Z,...)=-S’ .
H ence w e can say that the expression ‘ 2lx holds in the space
8 ’ means the same as ‘ -^-^21 is satisfied by the matrix 0(8)'.
On the basis o f these stipulations we obtain from 4.11,4.12, and
4.24 the follow ing form ulations:
First Principal Theorem. Let 21be a sentence of the sentential
calculus and 8 any topological space. The following conditions are
then equivalent'.
(i) 21 is provable in the two-valued calculus-,
(ii) 2lx holds in the space 8;
(iii) 2Ij holds in every topological space.
1 The oonoept of Yalidity (and of satisfiability) of a sentence belongs to
semantics. F op the problem o f an exact definition of this concept cf. article
VH 1 of the present work, in particular pp. 189 and 199.
448 S E N T E N T IA L CALCULU S A N D TO P O LO G Y XVH , § 4

Let S&bea sentence of the senten­

S econd P r in c ip a l T h e o b e m .
tial calculus and S any E-spate. The following conditions are
then equivalent:
(i) 91 is provable in the intuitionistic calculus \
(ii) 9l2 holds in the space 8\
(iii) 9l2 holds in every topological space.
Ths. 4.8 and 4.14 can also be brought into an analogous
form
In connexion with the second principal theorem it is worth
remembering that, in particular, all Euclidean spaces are 22-
spaces (of. 3.11).
With the theorems just stated, the decision criteria whioh
were mentioned in § 2 (cf. 2.19) can now be applied to topo­
logical sentences of the form 9IXor 9l2 (and even to somewhat
more extensive classes o f topological sentences). We are in a
position to decide, in each particular case, whether a sentence
of this form is generally valid in topology.
In conclusion it should be noted that the sentential calculus
can he interpreted in topology in various ways; the interpretation
discussed above is obviously not the only possible one. I f we
are dealing, for example, with the two-valued calculus, we
derive a quite trivial and in fact a general set-theoretical (not
especially topological) interpretation from 4.14; every set S can
in fact be made into an isolated topological space by putting
X = X for every c; 8 (cf. 3.6). A less trivial interpretation
of this calculus is obtained in the following way. We consider
the so-called regular oven sets of a topological space 8, i.e. sets
X cz 8 for which X — S-~S—X ; let &'(S) be the system of all
these sets. We define for the sets X, Y e @'(S) the operations
/., and *v- exactly as in 4.1, but we put

Iv T = S —S—X —Y
(X v Y is thus the smallest regular open set which includes X
and Jr). It can then be shown that the matrix
O '(S ) = [ 0 '( £ ) , 8, V A , -V]
XVn, $ 4 S E N T E N T IA L CALCULU S A N D T O P O L O G Y 44«

is adequate for the system 3 ft -1 In view o f this we correlate

with the sentence
« = ‘<f>{X,Y,Z,.„y
the follow ing sentence:
= ‘ For any regular open sets X , Y, Z,... of the space 8,
<j>'(X,Y,Z,...) = S',
where ‘tf>'{X,Y,Z,...y is obtained from ‘<f>(X,Y,Z,...)’ b y re­
placing the sign ‘ v ’ b y the sign ‘ V I t then appears that
the first principal theorem remains valid if “Hy in it is replaced
by % ’•

§ 5. I n terpretatio n op the Sen ten tial Calculus in

B oolean A lg ebr a an d in R elated M ath em atical
T h eo ries ®
Generalized Boolean algebra is here regarded as a part o f
abstract algebra, nam ely as the theory o f B oolean rings:
D efin itio n 5.1. A set B with at least two elements is called a
B oolean ring {with the fundamental operations + and .) if the
following formulas hold for all elements x, y, z e R:
(i) x-yy, x . y e B ,
(ii) x - f (y -fz ) = (x -f y )+z,
(iii) x = y+{x+y),
(iv) x.ly.z) = {x.y).z,
(v) x.ty+z) — x .y + x .z,
(vi) x . y = y.x,
(vii) x .x = x.
1 This follows from the fact that the family 0'{B) together with the opera­
tions V \ A , and *v- satisfy the postulates of Boolean algebra. This fact was
noticed by me as far back as 1927, and was implicitly stated in Th. B of II,
where, however, a different terminology was used; cf. X I, p. 341, footnote 2.
Compare also Tarski, A . (81), Th. 7.23, p. 178, and footnote 25, p. 181, where
a reference to an earlier paper of von Neumann is given.
3 For varioiia topics discussed in this section see the following papers: article
X I of this book (the foundations of Boolean algebra, the concept of atom) ;
article X II of this book (the theory of deductive systems); Stone, M. H . (07)
(the relation of Boolean algebra to general abstract algebra and to the theory
of fields of sets); Stone, M. H . (68) (the relation of Boolean algebra to general
topology); Tarski, A . (80) (operations on Boolean-algebraic ideals).
450 S E N T E N T IA L CALCULUS AN D TO P O LO G Y X V II, § 5

Note 5 . 2* I t can be shown that form ulas (iii) and (vi) in 5.1
can be replaced b y the follow ing equivalent conditions:
(iii') there is an element u e R such that

x = y+u = u + y;

(v i') (y+z).x = y .x + z.x .

H ence we see that B oolean rings coincide with those rings
(in the sense o f abstract algebra) in which every element x
satisfies 5.1 (vii).

D e f in it io n 5.3. Let R b e a Boolean ring.

(i) We shall say that x is divisible b y y, or that y is a divisor o f
x , in symbols y\x (or x < y), if x, y e R and if there is a z e R
such that x = y.z.
(ii) We denote by 0 that element x e R which is divisible by every
yeR.
The tw o sym bols ‘ |’ and ‘0 ’ can be defined in various other
(equivalent) ways.

5.4. The Boolean ring R is called atom istic or

D e f in it io n
atom less according to whether, for every element y e R distinct
from 0 , there are finitdy or infinitely many elements x >which are
divisible by y.
Note 5.5. E very elem ent y o f a Boolean ring R is called an
atom if there exist exactly tw o elements x which are divisible
b y y (in fact x = 0 and x = y). A ring R is atom istic if and
only if every elem ent y ^ O of R can be represented as a sum o f
a finite number o f atom s1;* R is atom less if and only if there are
n o atom s in R .

5.6. I f R is a Boolean ring, then a non-empty set

D e f in it io n
I ^ R is called an ideal, (in R), in symbols I e<f(R), if for any two
elements x and y of I their sum x + y is in I, and for every dement
y in I aU the elements x such that, y |x are in I .

1 In article X I of the present work the term 4atomistic ’ is used in a wider

sense.
X V II, § 6 S E N T E N T IA L CALCULUS AN D TO P O LO G Y 451

D e f in it io n 6.7. IfRisaB<xdeanring,th&nforanyI, J e S { R )
we put
(i) y x-,
XeJ'(,R?J.Xcj

(ii) / V J = TT Xi
XeSa$)J+J£X
(Hi) l A J — I .J (the intersection of I and J):
(iv) ^ i = i-+{o} (= 2 x).
' M K I X X -W '
Prom 6.1, 5.6, and 5.7 we easily obtain
C o r o l l a r y 5 .8 . For every Boolean ring R we have
(i) {0}, R e J(R), and in fact {0} is the smallest and R the
largest ideal in R;
(ii) if I, J e*f(R) then I -> J , I v J , I A «/, - v / are ideaZs jR,
in fact I J is the largest ideal X for which I .X £ J and ^-1
the largest ideal X for which I .X = {0}, moreover, J V J is the
smallest ideal which includes I and J> finally IA J is the largest
ideal which is included in I and J ♦
D efin ition 5.9. The ordered sextuple [,/(2 ?), R, V , A , ^ ],
where R is a Boolean ringy is denoted by I(i?).
The follow ing well-known theorem exhibits a close form al
connexion between B oolean algebra and topology1:
T heorem 5.10. A normal topological space JBX, which satisfies
thefollowing conditions, can be correlated with every Boolean ring R :
(i) there is a function F which maps the system S(R) onto the
system &(RX) in such a way that the formulas
Ic:J and F{I)czF{J)
are equivalent for all / , J e*f(R) (in particular J^({0}) = 0 and
F(R) = B*);
(ii) R x is isolated if and only if R is atomistic;
(iii) R* is dense in itself if and only if R is atomless;
(iv) R x is a space with a countable basis if and only if R is
denumerable.
1 See Stone, M. H . (68).
462 S E N T E N T IA L CALCULUS A N D TO P O LO G Y XVU, f 6

T heorem 5.11. For every Boolean ring B, l(_R) is a matrix,

if B x is the topological space correlated with the ring B according to
6 . 10, then the matrices l(JS) and 0 ( B X) are isomorphic.
Proof. W e first show b y means o f 4.2 and 5.8 that the function
F , which b y 5.10 (i) maps the system l(J?) on to 0 (B X) in one-
one fashion, satisfies the follow ing form ulas:
F ( I -> J) = F(I) F{J), F ( I vJ) = F(I) v F{J),
F { I a J) = F ( I ) a F{J), and F (^ I)= ^ F (I)
for all I , J eJ'(B) (where the signs ‘ v ’ , eto., are to be
interpreted on the left-hand side o f each form ula in the sense o f
Boolean algebra, and on the right-hand side in the topological
sense). H ence we easily conclude, with the help o f 2. 1, 2.3,
4.5, 4.0, and 5.9, that I(B) is a m atrix and is in fa ct isom orphic
w ith 0 { B X), q.e.d.
W ith the help o f the last tw o theorems all the results o f § 4
can be oarried over to Boolean algebra:
T h e o r e m 5.12. For every Boolean ring B we have
3 ft £ « (l(i? )) c 351.
[B y 2.7, 4 .8 , 4.11, 5.11]
T heorem 5.13. For every Boolean ring B and every sentence 51
the conditions 51 e 3 ft and -v -a- 51 e ®(l(iJ)) are equivalent.
[B y 2.7, 4.12, 5.11 or 1. 6, 5.12]
T h e o r e m 5.14. Let B b e a Boolean ring. In order that
« ( !( * ) ) = ZSt
it is necessary and sufficient that B be atomistic.
[B y 2.7, 4.14, 5.10 (ii), 5.11]
T h e o r e m 5.15. I f B is an atomless Boolean ring, then
« ( !( £ ) ) = 3 ft.
[B y 2.7, 3.10, 4.24, 5.10 (iii), (iv), 5.11]
Note 5.16. N ote 4.25 can mutatis mutandis be applied to
Th. 5.15. The converse o f this theorem does not hold. The
form ula ®(l(B)) = 3 ft, applies for exam ple, to every denumer­
able B oolean ring B which, although not atomless itself, in­
cludes an atom less subring (i.e. an atomless ring B1 £ B with
X V II, | 5 S E N T E N T IA L CALCULUS AN D TO P O LO G Y 453

the same fundam ental operations as B). On the other hand,

m any exam ples o f Boolean rings can be given for which neither
® (l(R)) = 3 & n or ® (l(i?)) = holds. T o such rings belong, in
particular, all infinite completely additive rings, i,e. Boolean rings
R which satisfy the follow ing condition:
for every set X c R there exists the greatest common divisor of
all elements x e X (i.e. an element y which is a common divisor of
all x e X and is divisible by every other common divisor of these
elements).
F or the com pletely additive rings R the follow ing property is
characteristic: = R holds for every I e*f(R)\ in
other words, the system ® (l(i?)) contains all sentences 31 o f the
form 3 1 = 'v ® V and cannot therefore be identical with
S it (cf. 1.5). On the other hand, as is easily shown, no infinite
com pletely additive ring R is atom istic, whence, b y virtue o f
5.14, «(»(H )) *£= 3 it.
Ths. 5.12-5.15 can be expressed in a form analogous to that
o f the first and second principal theorems o f § 4. In particular,
tw o sentences 31j and 3l2 o f Boolean algebra can be correlated
with every sentence 31 o f the sentential calculus in such a way
that 31 is respectively provable in the tw o-valued or in the in-
tuitionistic calculus if and only if 31xor 312holds for every Boolean
ring. From this we obtain a decision criterion for sentences o f
the form 3t2 or 3l2.
The remarks at the end o f § 4 regarding other possible inter­
pretations o f the sentential calculus can also be extended to
B oolean algebra. The regular open sets in that case are to be
replaced b y the ideals I which satisfy the form ula = J.
A ll these results hold not only for the form al system o f
Boolean algebra, bu t also for every realization o f this system .
The best known o f these realizations is the theory o f fields o f
sets, i.e. o f systems o f sets which are closed under the operations
o f addition and subtraction. E very field o f sets with at least
tw o elements is, as is easily seen, a Boolean ring with the so-
called sym m etric subtraction T © Y = ( X— X) and
the ordinary set-theoretical m ultiplication as the fundamental
454 S E N T E N T IA L CALCULUS AND TOPO LO G Y XV II, § 6

operations. The sim plest exam ples o f special Boolean rings,

which were m entioned in 5.14-5.16, can be drawn directly from
the theory o f fields o f sets. E very field o f sets, for instance,
which consists o f all finite subsets o f a given (finite or infinite)
set is an exam ple o f an atom istic ring. In order to obtain a
denumerable atomless ring, we consider the set X o f positive
rational numbers x < 1 and form the field o f sets which consists
o f all sums o f finitely m any intervals £ X (the right-hand end­
point being included in the interval, but not the left-hand one).
As examples o f com pletely additive rings (in the sense o f 5.16) the
fields o f sets which consist o f all subsets o f a given set m ay serve.
Another im portant realization o f Boolean algebra is general
metamathematics, i.e. the theory of deductive systems.1
Finally, some o f the results obtained can be extended to a
m ore general theory, nam ely to the theory of lattices. In fact
we consider lattices in which the operation S o f infinite addition
(infinite join operation) is always perform able and in which
finite m ultiplication is distributive under both finite and infinite
addition .2 As examples o f such lattices the system o f all open
sets o f a topological space and the system o f all ideals o f a
Boolean ring m ay be m entioned. Let L be a lattice o f this type
with at least tw o different elements; we put
1 = S z, 0 = Sz (0 denoting the em pty set);
zeL zeo
x v y = x+y, xa y~x.y,
£->!/= S z, ^x —X 0
(xAz)\/V^V
for all x, y e L. It then results that M(L) = [L, 1, v , A,
is a m atrix, and that Ths. 5.12 and 5.13 remain valid if *M(L)’
is substituted for M(J2) ’ in them . Ths. 5.14 and 5.15 can also be
carried over to the theory o f lattices. In this way we obtain an
interpretation o f the tw o-valued and o f the intuitionistic cal­
culus in the theory o f lattices.
1 See X II.
2 More generally, we could consider here arbitrary distributive lattices
(without infinite addition) in which for any elements x, y s L there is an ele­
ment z e L such that (i) x.z-\-y = y t and (ii) u + z = z whenever u e L and
x .u + y = y. Such lattices are intimately related to the so-called Brouwerian
algebras (or Brouwerian logics). Cf. Birkhoff, G. (7a) and McKinsey, J. C. C.
and Tarski, A . (53a).
 ABBREVIATIONS

Akad. der Wiss. Ahademie der Wissenschaften

Amer. J. Math. American Journal of Mathematics
-dmer. math. Soc. American Mathematical Society
Ann. $oc. polon. Math. Annates de la SocUU Polonaise de Mathd•
matique
Bull. Bulletin de la or Bulletin of the
G.R. Comptes Rendus du or Oomptes Rendus des
Stance# de la
Cambridge phil. Soc. Cambridge Philosophical Society
ed. edition
JSrgr. math. KoU. Ergebnisse eines mathematischen Kollo-
quiums
Fund. Math. Fundamenta Mathematical
J. Journal of or Journal of the
London math. Soc. London Mathematical Society
Mat.-not. Masse Matematishmaturvidenshapelig Masse
Math. Ann. Mathematische Annalen
Math. Z. Mathemati8che Zeitschrift
Mh. Math. Phys. Monatshefte fur Mathematik und Physik
Nat. Acad. Sci. National Academy of Sciences
Phys.-math. Klasee Physikalisch-fnalhemaUsche Klasse
Proc. Proceedings of the
Roy. Irish Acad. Royal Irish Academy
Shrift. Videnshapsselshapet SJcrifter utgit av Videnskapsselskapet i
Kristiania
Soc. Sci. Lett. SocidU des Sciences et des Lettres de
Trans. Transactions of
Trav. Travaux de la
 B IB L IO G R A P H Y

(1) W. A ckehmann, ‘ tJber die Erfullbarkeit gewisser Zahlaus-

drfioke’, Math. Ann. o (1928), 638-49.
(2) K. A jdukebwicz, Z metodologji nauk dedukcyjnych (From the
methodology of the deductive sciences) (Lw6w, 1921).
(3) K. Ajduxucwicz, Oldume zasady metodologji nauk i logikiformalnej
(Main principles o f the methodology of the sciences and of formal
logic) (Warszawa, 1928).
(4) K. Ajdttkcewioz, Logiczne podstawy nanczania (The logical
foundations of teaching), Encyclopidja Wychowania, ii (War­
szawa, 1934), 1-75.
(5) P. Bern ay s , ‘ Axiomatische Untersuchung des Aussagenkalkuls
der Prinoipia mathematical, Math. Z., xxv (1926), 305-20.
(5a) P. Born-ays and M. SchOkpinkjeel, ‘ Zum Entecheiduiigsproblem
der mathematischen Logik’, Math. Ann., xcix (1928), 342-
72.
(6) B. A. B ernstein , ‘ Whitehead and Bussell’s theory o f deduction
as a mathematical science’, Bull. Amer. math. Soc., xxxvii
(1931), 480-8.
(6a) E. W. B oth, ‘ On Padoa’s method in the theory o f definition*,
Jndagationea Mathematicoe, xv (1953), 330-9.
(7) G. B ibxhopp , ‘On the combination o f subalgebras’, Proo. Cam­
bridge phU. Soc., xxix (1933), 441-6.
(7a) G. B irkhotf, Lattice Theory, Amer. Math. Soc. Colloquium Publi­
cations X X V (revised ed.). (New York, 1948).
(8) B. Carnap , Abrifi der LogisHk (Wien, 1929).
(9) B. Carnap , ‘ Bericht fiber Untersuchungen zur allgemeinen Axio-
matik*, Erleenntnis, i (1930), 303-7.
(10) B. Carnap , Logisohe Syntax der Spraehe (Wien, 1934).
(11) B. Carnap , ‘ Ein Gfiltigkeitskriterium fur die S&tze der klassi-
schen Mathematik’, Mh. Math. Phys., xlii (1935), 163-90.
(11a) A. Church, ‘ Special cases of the decision problem*, Revue Philo-
sophique de Louvain, xlix (1951), 203-21.
(12) L. Chw istek , ‘ The theory of constructive types (Principles of
logic and mathematics)’, Part I, Ann. Soc. polon. Math.,
ii (1924), 9-48.
(13) L. Chw istek , *Neue Grundlagen der Logik und Mathematik’,
Math. Z „ xxx (1929), 704-24.
(14) L. Chw istek , ‘ tJber die Hypothesen der Mengenlehre’, Math.
Z., xxv (1926), 439-73.
(14a) L. Couturat , L ’alg&bre de la logique (Paris, 1905).
(15) B. D ed ek in d , Was sind und was soUen die Zahlen? 5th. ed.
(Braunschweig, 1923).
(15a) A. H. D iamond and J. C. C. M oK in se y , ' Algebras and their
subalgebras’, BuU. Amer. math. Soc., liii (1947), 959-62.
 B IB L IO G R A P H Y 457

(16) A . F raenkel , Eirdeitung in die Mengenlehre, 3rd. ed. (Berlin,

1928).
(17) M. FBjfccHET, *Des families et fonctions additives d’ensembles
abstraits’, Fund. Math., iv (1923), 328-66.
(18) G. Gentzbn , ‘ Untersuchungen liber das logische SchlieBen’,
Math. Z., xxxix (1934), 176-210 and 406-31.
(19) T. Glivenko, ‘ Sur quelques points de la logique de M. Brouwer’,
Academie Royale de Belgique, Bull. Glasse dee Sciences, xv (1929),
183-8.
(20) K. Godel , ‘ Die Vollstandigkeit der Axiome dee logischen Funk-
tionenkalkulsMh. Math. Phys., xxxvii (1930), 349-60.
(21) K . Godel , ‘ Einige metamatbematische Resultate iiber Ent-
scheidungsdefinitheit und Widerspruchsfreiheit’, AJead. der
Wise, in Wien, Mathematisch-natururisserischaftliche Klasse,
Akademischer Anzeiger, lxvii (1930), 214-16.
(22) K . G odel , ‘ Ober formal unentscheidbare Satze der Prindpia
mathematica und verwandter Systems I ’, Mh. Math. Phys.,
xxxviii (1931), 173-98.
(23) K. G5d e i,, ‘Eine Interpretation des intuitionistisohen Aussagen*
kalkuls’, Erg. math. Kott., iv (1933), 39-40.
(24) K . G relu ng and L. N elson , ‘ Bemerkungen zu den Paradoxien
von Russell und Burali-Forti’, Abhandlungen der Fries'schen
Schule neue Folge, ii (1908), 301-34.
(26) F. H ausdobff , Mengenlehre (Berlin und Leipzig, 1927).
(26) J. H erbrand , ‘ Recherohes sur la th^orie de la demonstration’,
Trav. Soc. Sd. Lett. Varsovie, Classe I I I , Nr. 33 (1930).
(26a) H . H erm es , ‘ Zur Theorie der aussagenlogischen Matrizen’,
Math. Z., liii (1961), 414-18.
(27) P. H ertz, ‘ t)ber Axiomensysteme fur beliebige Satzsysteme’,
Math. Ann., ei (1929), 467-614.
(28) A. H e y t in g , ‘ Die formalen Regain der intuitionistischen Logik’,
Sitzungsberichte der Preufiischen Akad. der Wise., Phys.-math.
Klasse (1930), 42-66.
(29) D. H ilbert , ‘Die Grundlegung der elementaren Zahlenlehre ’,
Math. Ann., civ (1931), 486-94.
(29a) D. H ilbert , ‘ Die logischen Grundlagen der Mathematik’, Math.
Ann., lxxxviii (1923), 161-66.
(30) D. H ilbert and W . A ceermann , OrundzUge der theoretischen
Logik (Berlin, 1928).
(31) D. H ilbert and P. B er n ays , Grundlagen der Mathematik, i (Berlin,
1934).
(31a) H. H ii, ‘ On primitive terms o f logic’, J. Symbolic Logic, xvii
(1962), 156.
(32) E. V . H untington , ‘ Sets o f independent postulates for the
algebra o f logic’, Trans. Amer. math. Soc., v (1904), 288-309.
(33) E. V. H untington , ‘ A set of postulates for abstract geometry-
exposed in terms of the simple relation o f inclusion ’, Math. Ann.,
lxxiii (1916), 522-59.
458 B IB L IO G R A P H Y

(34) E . V. H untington , ‘ New sets of independent postulates for the

algebra of logic*, Trans. Amer. math. Soc., xxxv (1933), 274-
304, 551-8, and 971.
(35) E. V. H untington , ‘ Independent postulates for the “ informal”
part of Principia Mathematdca\ Bull. Amer. math. Soc., xl
(1934), 127-36.
(36) S. J ankowski, ‘ Recherches sur le systdme de la logique intuition -
iste Actes du Congrks International de Philosophic Sdentiflque,
vi (Paris, 1936).
(36a) S. Ja&kowski, ‘ Une modification des definitions fondamentales
de la g6ometrie des corps de M. A. Tarski’, Ann. Soc. polon.
Math., xxi (1948), 298-301.
(365) S. J ankowski, ‘ Trois contributions au calcul des propositions
bivalent*, Studia Sodetatis Scientiarum Torunensis, Section A,
i (1948), 3-15.
(36c) B. J6 ns80n and A. T arski, ‘ Boolean algebras with operators’,
part I, Amer. J. Math., horiii (1951), 891-939.
(37) T. K otarbu^ski, Elementy teorji pozncmia, logihi Jormalnej i
metodologji nauk (Elements of the theory of knowledge,
formal logic, and the methodology of the sciences) (Lw6w,
1929).
(38) C. K uratow ski, ‘ Sur la notion de l’ordre dans la th4orie des
ensembles’, Fund. Math., ii (1921), 161-71.
(39) C. K uratow ski, ‘ Evaluation de la classe borelienne ou projective
d’un ensemble de points h l’aide des symboles logiques *, Fund.
Math., xvii (1931), 249-72.
(40) C. K uratowski, 4tJber eine geometrische Auffassung der Logistik
Ann. Soc. polon. Math, ix (1931), 201.
(41) 0. K uratow ski, Topologie I (Warszawa-Lw6w, 1933).
(42) C. K uratow ski, ‘ Sur reparation X de l’analysis situs’, Fund.
Math., iii (1922), 192-6.
(43) C. K uratow ski, *Les ensembles projectifs et l’induction trans-
finie*, Fund. Math., xxvii (1936), 269-76.
(44) C. H. L angford , ‘ Analytical completeness of postulate sets Proc.
London math. Soc., xxv (1926), 115-42.
(45) C. H. L angford , ‘ Some theorems on deducibility*, Annals of
Mathematics, xxviii (1927), 16-40; ‘ Theorems on deducibility
(Second paper)*, ibid., 459^-71.
(45a) S. Le &nthjwski, Podstawy ogdlnej teorii mnogoid, I (Foundations of
general set theory I) (Moscow, 1916)*
(46) S. L e Sntewski, ‘ Grundzuge eines neuen Systems der Grundlagen
der Mathematik’, §§ 1-11, Fund. Math., xxv (1929), 1-81.
(47) S. Le ^neewski, ‘ tTber die Grundlagen der Ontologies C. R. Soc.
Sd. Lett. Varsovie, Classe I II , xxiii (1930), 111-32.
(47a) S. L e&niewski, ‘ O podstawach matematyki* (On the foundations
of mathematics, in Polish) Przeglqd Filozoficzny, xxx (1927),
164-206; ibid., xxxi (1928), 261-91; ibid., xxxii (1929), 60-105;
ibid., xxxiii (1930), 77-105; ibid., xxxiv (1931), 142-70.
 B IB L IO G R A P H Y 459

(476) S. L e §n ie w s k i , ‘ Grundziige eines neuen Systems der Grundlagen

der Mathemafik *, § 12, Collectanea Logica, i (1938), 61-
144.
(48) C. I. L e w is , A survey of symbolic logic (Berkeley, Calif.. 1918).
(48a) J. LoS, ‘ O matrycach logicznych ’ (On logical matrices) Trav.
Soc. Sci. Lett. Wroclaw, Seria B , Nr. 19 (1949).
(49) L. L 6 w en h eim , ‘ tJber Moglichkeiten im Relativkalkul*, Math.
Ann., Ixxvi (1915), 447-70.
(50) J. L u k a siew ic z , ‘ O poj^ciu wielko6ci> (On the concept o f m agni­
tude), Przeglqd FUozoflczny, xix (1916), 35 ff.
(50a) J. L u k a siew ic z , ‘ Logika dwuwarto^ciowa’ (Two-valued logic),
Przeglqd FUozoflczny, xxiii (1921), 189-205.
(51) J. L u k a s ie w ic z , Elementy logiki matematycznef (Elements o f
mathematical logic) (1929).
(51a) J. L u k a siew ic z , fO znaezeniu i potrzebach logiki matematyeznej *
(On the significance and needs of mathematical logic), Ncvuka
Polska,, x (1929), 604-20.
(52) J. L u k a siew ic z , ‘ Logistyka a filozofia’ ('Logistic and Philosophy),
Przeglqd FUozoflczny, xxxix (1936), 115-31.
(52a) J. L u k a siew ic z , ‘ The shortest axiom of the impiicational cal­
culus of propositions’, Proc. Boy. Irish Acad., Section A,
lii (1948), 25-33.
(53) S. M azubktew icz , (tJber die Grundlagen der Wahrscheinlich-
keitsrechnung,, Mh. Math. Pkys.. xl (1934), 343-52.
(53a) J. C. C. M oK in rey and A, T a r sk i , ‘ Ou closed elements in closure
algebra \ Annals of Mathematics, xivii (1946), 122—62.
(536) J. C. C. M cK in s e y , ‘ A new definition of truth’. Synthesc, vii
(1948-9), 428-33.
(53c) A. Mostowski, ‘ Abzahlbare Boolesche K$rper und ihre An-
wendung auf die allgemeine Metamathematik\ Fund. Math.,
xxix (1937), 34-53.
(53d) A. M ostow ski , Logilca matematyezna (Mathematical Logic)
(Warszawa-Wroclaw, 1948),
(53c) A. M osto w sk i , ‘ Some improdicative definitions in the axiomatic
set-theory’, Fund. Math., xxxvii (1950), 111-24, and xxxviii
(1951), 238.
(53/) A. M osto w sk i , Sentences undccidable in formalized arithmetic
(Amsterdam, 1952).
(54) J. v. N eumaistn, ‘ Zur Hilbertschen Beweistheorie’, Math. Z ., xxvi
(1927), i-46.
(55) J. Nioon, ‘A reduction in the number of the primitive propositions
of logic’, Proc. Ccmibridge pml. Soc., xix (1917), 32-41.
(56) J. N ic o d , La g&metrie dans le monde sensible (Paris, 1924).
(57) A . Padoa , *Un nouveau systems irr&iuctible de postulats pour
l’algebre’, C. R. Deuxi&me Congrks International des Mathi-
maticiens (Paris, 1902), 249-56.
(58) A. P adoa, ‘ Le probl6me No. 2 de M.David Hilbert', L'Enseigne-
ment Math., v (1903), 85-91.
460 B IB L IO G R A P H Y

(58a) C. S. P e ir c e , ‘ On the algebra o f logic: a contribution to the

philosophy o f n ota tion ’ , Amer. J. Math., vii (1885), 180-202.
(59) M. F i e r i , ‘ L a geom etria elementare instituita sulle nozioae di
‘ ‘ punto ’ ’ e sfera ’5’, Memorie di Mctternatica e di Fisica deUa
Societd Italicma d-elle Scienze, Serie Terza, xv (1908), 345-450.
(60) E. L . P ost , ‘ Introduction to a general theory o f elementary
propositions’ , Amer. J. Math., xliii (1921), 163-85.
(61) M. P resjbtctrger, ‘ tjb er die Vollstaxidigkeit eine3 gewissen Systems
der Ariihm etik ganzerZahlen, in welchem die A ddition ala einzige
Operation hervortritt C.R. Premier Congr&s den Mathimaticiens
des Pays Slaves, Warszawa- 1929 (W arszawa, J930), 92-3 01.
(61a) A . R ose . ‘ The degree o f completeness o f M - valued Lukasiewicz
propositional calculus*, J. London math. Soc.. xx v ii (1952),
92-102.
(616) J . B. R osser and A . R . T ttrqttette, ‘ A xiom schemes for m*
valued propositional calcu li’ , J. Symbolic Logic, x (1945), 61-82.
(61c) J. B. R osser and A . R. T u r q u e tte , ‘ A note on the deductive
completeness o f m -valued propositional calcu li’ , J. Symbolic,
Logic, x iv (1950), 219-25.
(6 Id) B . A . W . R u s s e ix , The principles of mathematics (London, 1903).
(61c) B. A . W . R,tjsssix , Introduction to mathematical philosophy (London,
1920).
(62) E. S ch roder , Vorlesuzzgen iiber die Algebra der Logik (exakte
Logik) (L eipzig; i, 1890; iis part 1, 1891; ii, part 2, 1905: iii,
part L 1S95).
(62a) K . S chroter , ‘ D eduktive abgesohlossene Mengen ohne B asis’ ,
Mathematische Nachrichten, vii (1952), 293-304.
(63) H. M. Sk e f f k r , The general theory of notationed relativity (Cambridge,
Mass,, 1921).
(63a) H. M. Sh e f p e r , cA set o f five postulates for Boolean algebras
with application to logical constants’ . Trans. Amer. math. Soc.,
x v (1913), 481-8.
(64) T h . Sxojjbm, ‘ Untersuchungen iiber die A xiom e des Kiassen-
kalkiils und iiber Produktatious- und Sumniationsproblem e,
welche gewisse Klassen von Aussagen betreffen’ , Shrift Viden*
shapsseUJcapet, I . Mat.-nal. hlasse 1919, N o. 3 (1919).
(65) T h . Seotjsm, ‘ Logisch-kom binatorische Untersuchungen iiber
die Erfullbarkeit oder Beweisbarkeit mathematisoher Satze
nebst einem Theorem iiber diehte M engen’ , Shrift Yidenshaps-
selskapet, I . Mat.-nat. hlasse 1920, N o. 4 (1920).
(66) T h . Sk o ie m , ‘ tJber einige Grundlagenfragen der M athem atik’ ,
Shrifter utgitt av Det Norske Videnshaps-Alcadezni i Oslo, I.
Mat.-nat. hlasse 1929* N o. 4 (1929).
(66a) J. Settpecki, ‘ J3ow6d &ksjomatyzow&lno6ci pelnych systembw
wielwarto6ciowych rachunku zdah’ (A p roof o f the axiomatiza-
bility o f full systems o f m any-valued sentential calculus, in
Polish), C. R. Soc. Sci. Lett. Varscroit, Classe III, x x xii (1939),
110-28.
 B IB L IO G R A P H Y 461

(666) J. SatnpECKi, ‘ Pefaiy tr6jwarto6ciowy rachunek zdan* (The full

three-valued sentential calculus), Anncdes Unwersitatis Marias
Curie-Sklodowska, i (1946), 193-209.
(06c) B. SOBOOnfrsxi, *Z badah nad teorj$ dedukcji’ (Some investigations
upon the theory of deduction) Przeglqd FUozofi^zny, xxxv
(1932), 171-93.
(60d) B. Soboct&sxi, *Aksj omatyzacj a pewnych wielowarfcoScicwych
system6w teorji dedukcji’ (Axiomatization of certain many­
valued systems of the theory of deduction, in Polish), Roczniki
Prac Naukowych Zrzeszenia Asystentcw Uniwersytetu Jozefa
PUsudskiego w Warszawie, i (1936), 399-419.
(67) M. H. Sto n e , ‘The theory of representations for Boolean algebras’ ,
Trans. Amer. math. S o c xi (1936), 37-111.
(68) M. H . Sto n e , ‘ Applications of the theory of Boolean rings to
general topology*, Trans. Amer. math. Soc., xii (1937), 370-481,
(69) M. H. Sto ne , ‘ Applications of the theory of Boolean rings to
general topology’, Ftmd. Math., xxix (1937), 223-303.
(70) M. H. Stone, ‘ Topological representations of distributive lattices
and Brouwerian logics’, Casopis pro P&stovdnl Matematiky a
Fysilcy, Ixvii (1937-8), 1-26.
(71) A, T ab sk i, ‘ Sut les ensembles finis’, Fund. Math., vi (1925),
45-96.
(72) A. T ab sk i, ‘ Remarques but les notions fondamentales de la
m6thodologie des math6matiques ’, Ann. Soc. polon. Math., vii
(1929) 270-2.
(73) A. T ab sk i, ‘ O poj^ciu prawdy w odniesieniu do sformalizowanych
nauk dedukcyjnvch ’ (On the notion o f truth in reference to
formalized deductive sciences), Ruch FUozoficznyt xii (1930-1),
210- 11.
(74) A. T abski, ‘ tTber definierbare Mengen reeller Zahlen’, Ann. Soc.
poUm. Math., ix (1931), 206-7.
(76) A. T abski, ‘ Sur les classes d’ensembles closes par rapport &
eertaines operations 616mentaires’, Fund. Math., xvi (1930),
181-304.
(70) A. T abski, ‘ Der Wahrheitsbegriff in den Sprachen der deduktiven
Bisziplinen’, A had. der Wiss. in Wien, M athematisch-natur-
wissenschaftliche Klasse, Akademischer Anzeiger, Ixix (1932),
23-26.
(77) A. T abski, Cardinal algebras (New York, 1949).
(78) A. T abski, ‘ Einige Bemerkungen zur Axiomatik der Boole’schen
Algebra’, C. JR. Soc. Soi. Lett. Varsovie, Classe I I I , xxxi (1938),
33-36.
(79) A. T abski, ‘ On undecidable statements in enlarged systems of
logic and the concept of truth’, J. Symbolic Logic, iv (1939),
105-12.
(80) A. T abski, ‘ Ideal© in den Mengenkftrpem *, Ann. Soc. polon. Math.,
xv (1937), 186-9.
(81) A. T abski, ‘ t)ber additive und multiplikative MengenkOrper und
462 B IB L IO G R A P H Y

Mengenfunktionen*, C . R . Soc. Sci. Lett. Varsovie, Classe I I I ,

xxx (1937), 151-81.
(82) A . T arski, ‘ The semantic conception of truth and the founda­
tions of semantics’, Philosophy and Phenomenological Research,
iv (1944), 341-76.
(83) A. T arski, ‘ A problem concerning the notion of definability’,
J . Symbolic Logic, xiii (1948), 107-11.
(84) A . T arski, A decision method fo r elementary algebra and geometry,
2nd ed. (Berkeley, Calif., 1961).
(84a) A. T arski, ‘ Some notions and methods on the borderline of
algebra and metamathomatics *, Proc. International Congress of
Mathematicians, Cambridge, M ass., 1950 , i (Providence, Rhode
Island, 1952), 705-20.
(85) A. T arski and A. L indenbaum , ‘Sut 1’ independence des notions
primitives dans les syst&mes math6matiques’, A n n. Soc. polon.
math., v (1927), 111-13.
(86) O, V eblen , ‘ A system of axioms for geometry Trans. Am er. math.
Soc., v (1904), 343-81.
(87) O. V eblen , ‘ The square root and the relations of order’, Trans.
Am er. math. Soc., vii (1906), 197-9.
(87a) M. W ajsberg , ' Aksj omatyzacj a tr6warto6ciowego raohunku
zdan’ (Axiomatization of the three-valued propositional cal­
culus)’, C. R . Soc. Sci. Lett. Varsovie, Classe I I I , xxiv (1931).
(876) M. W ajsberg , ‘ Beitrage zum Metaaussagenkalkul I ’, M h . Math.
P h ys., xlii (1935), 221-42.
(87c) H . W ang , ‘ Remarks on the comparison of axiom systems’, Proc.
N at. Acad. Sci„ Washington, xxxvi (1960), 448-53.
(88) A . N . W hitehead , A n enquiry concerning the principles of natural
knowledge (Cambridge, 1919).
(89) A . N. W h ite h e a d , The concept of nature (Cambridge, 1920).
(90) A. N . W hitehead and B. A . W . R ussell, Principia Mathematica
2nd ed., i-iii (Cambridge, 1925-7).
(91) L. W ittgenstein , Tractatus logico-phUosophicus (London, 1922).
(92) E . Zermelo , ‘ tJber den Begriff der Definitheit in der Axiomatik’ ,
Fund. M ath., xiv (1929), 339-44.
 SUBJECT IN D E X
Abstraction, method of extensive, 29. cardinal degree of capacity (of a
addition, logical, 114, 144, 175. system), 368.
-------- of systems ( - f ), 351. --------of completeness, 368.
algebra, Boolean, atomistic, 334, 378. categorical, 390.
— — extended, 323, 348. — (intrinsically and absolutely), 311.
--------ordinary, 320, 347. — strictly, 313.
— of logic, 168, 320, 347. categoricity, 34, 100, 309, 313, 391.
algorithm, sentential, 348, 374, 355. category, semantical, 215, 218, 268.
analytical, 418. -------- first principle of, 216.
antecedent, 39. --------order of, 218.
antinomy of heterological words — unifying, 230.
(Grelling-Nelson), 165 n., 402. complementation, 123, 320, 347.
— of Richard, 110, 119, 402. — double, law of, 124.
— of the liar, 157. complete (defined), 93, 185, 287, 394.
atom ( A t ) , definition of, 334. — absolutely, 93.
axiom, definition of, 179. — relatively, 94.
— logical, 284, 373. completeness, 34, 93.
— metamathematical, 31. — degree of, 35, 100.
— o f infinity, 243, 289. — of concepts, 308.
— of intuitionistic calculus, 423. — with respect to sets of terms, 313-
— of sentential calculus, 423. 14.
— system, finite, 36, 40, 76. — with respect to specific terms, 311.
axiomatizability of system L n, 49. co-completeness, 288.
1 * ,, 61. concentric, 26.
axiomatizable, finitely, 35, 76. concepts, completeness of, 308 f.
— systems, 354. — semantical, definitions of, 252,406,
conjunction (defined), 176, 422.
consequence, 30,40, 63,182, 252, 285,
Basis, of a set X of sentences, 35, 88.
345, 409 et seq.
— ordered, 76.
— formal, 419.
— formalized concept of, 294.
Calculus of classes, language of, 165, — material, 419.
168, 320. —•of the nth degree (defined), 181.
— of systems, 342, 347, 360. consequent, 39.
-------- interpretations of, in set theory, consistency, 34, 90.
359. consistent (defined), 185, 287, 394.
— sentential, 38, 64, 348. to-consistent, 288.
-------- and topology, 421, 434. constants, 168.
-------- interpretation of, in Boolean contradiction, principle of, 197, 236,
algebra, 449. 257.
-------- incomplete system of, 393. contradictory, 418.
-------- interpretation of, in Boolean convention T , 187, 236, 247.
algebra, 449. convergence of an increasing se­
-------- many valued, 47. quence of sentences, 363.
-------- ordinary, 42. — of systems, 363.
-------- of Heyting, 362, 400. counter domain, 121.
--------positive system of, 399.
-------- restricted, 62. Decision domain, 93.
-------- topological interpretation of deductive sciences, methodology of,
Two-valued and Intuitionistic, 434. 30.
464 SUBJECT IN D E X
deductive system (defined), 33, 185,
350.
— systems, theory of, 454.
definability, 110, 128, 296 f., 301.
— elementary, 128.
— of concepts of mechanics, 318.
definition, 401.
— implicit, 150.
— materially adequate, 152,188,404.
— recursive, 177 n.
— semantical, 155, 162.
— structural (of true sentence), 163-
4.
De Morgan, law of, 124, 339.
denotation, 401.
dense (subset), 428.
detachment, rule of, 40, 345, 423.
development, law of, 12, 17.
diametrical, externally and internally,
26.
disjoint, 25.
disjunction (defined), 175, 283, 422.
distance, 430.
divisor of (in Boolean ring), 450.
domain (of a relation), 121.
— decision (of a set of sentences), 93.
— individual, concept of correct
sentence in, 266.
— sentential, 69.
duality, logico-mathematical, 145.
Element, designated, 41, 424.
— empty (or zero), 320, 347.
— unit, 320, 347.
elementarily indistinguishable or
equivalent (of order types), 380.
equidistant, 27.
equivalence, logical (of two sets of
sentences), 34, 72.
— of sentences, 185, 283, 348.
Ift-spaoe, 429.
excluded middle, principle of, 197,
236, 257, 351.
expressions, structural-descriptive,
167, 172.
extenaionality, 161 n.
Family, regular, 137.
follows logically from, 414 (see conse­
quence).
formalism, intuitionistic, 62.
formalization, 166, 403.
function, bounds of, 13, 17.
function, projective, 143, 148.
— quotation-, 161.
— sentential, 114, 148, 176-7, 422.
under a given variable or con­
stant, 392.
— truth, 5.
— value, of a logical matrix, 41.
functions, law of number of, 9.
functor, 161.
Geometry, 317.
— Euclidean, 306, 388.
— projective, 389.
greatest common divisor (for Boolean
rings), 453.
Ideal (in a Boolean ring), 450.
identity (in Boolean algebra), 323,347.
— elementary theory of, 378.
implication, 39, 283, 348, 422.
inclusion (definition), 175.
— relation of, 347.
indecomposable, 45.
independence, 35.
— complete, 38.
— maximal, 36.
— of concepts, 298.
independent, 35, 83.
— ordinally, 75.
induction, principle of, 173.
— infinite, rule of, 259 f., 294.
inference, rules of, 30, 63, 116.
infinity, axiom of, 243, 289.
interior to, 27.
interpretable, effectively, in logic,
391.
interpretation, topological, of sen­
tential calculus, 434.
— finitistie, 293.
intervals, 133.
isolated, 429.
isomorphic (of matrices), 425, 452.
Language, finitistie character of, 253*
— colloquial, and philosophical pro­
blems, 267.
— formalized, 165, 403.
-------- definition of true sentence for,
265.
-------- semantics of, 256, 266, 403.
— morphology of, 251, 403.
— natural, semantics of, 267.
— of general theory of classes, 242.
— semantics of, 252.
 SUBJECT IN D E X
languages, of finite and infinite order,
220, 265.
— of infinite order, 262 f.
lattices, theory of, 454.
law of number of functions, 9.
— of development, 12.
— of substitution, 5.
logic, the, of a given science, 252.
logical product, 2, 114, 144, 176, 283.
— sum, 114, 144, 175, 283.
Materially adequate- (of definitions),
152, 188, 404.
matrix, logical, 41, 424.
-------- nth power of, 427.
mechanics, concepts of, 318.
mereology, 24, 333.
metacalculus of classes, 169.
metalanguage, 167, 280, 403.
— axioms of, 173, 282.
metamathematics, 30, 60.
— general, axioms of, 344.
metatheory, 167, 210, 251.
method, axiomatic, 40, 43.
— deductive, 295.
— matrix, 41, 424.
— of extensive abstraction, 29.
— of many-rowed sequences, 227.
— of Padoa, 305.
— of semantical unification of the
variables, 227.
methodology of deductive sciences,
60.
metricizable, 430.
model, 416,
monomorphy (see categoricity), 100,
390.
monotransformability, 313.
morphology of language, 251, 403.
Names, quotation-mark, 156.
— structural-descriptive, 150.
negation, 39, 114, 144, 175, 283.
— of a system, 351.
non-ramifiability, 35, 100, 390.
normal (topological space), 429.
number, ordinal, 269 f.
numbers, real, definable sets of, 110.
Open set, 428.
operations, fundamental (of a logical
matrix), 424.
465
operations, logical, 143.
— of insertion, and deletion of quan­
tifiers, 55, 181.
operators, 218.
order, dense, elementary theory of,
372.
— general elementary theory of, 379.
— isolated, elementary theory of,
375.
— of a semantical category, 218.
— of a sign, 218.
— of a language, 220.
— type,. 381.
ordinal number, 269.
Padoa, A ., method of, 299, 305.
part, proper, 25.
— of, the relation of, 24.
point, 27.
postulates of Boolean algebra, 321 et
seq., 347.
— of calculus of systems, 357.
product, logical, 2, 114, 144, 283.
— (in Boolean algebra), 347.
provable sentence (definition of), 182.
pseudodefinitions, 223> 285.
Quantification, existential (definition
of), 114, 176, 283.
— universal (definition of), 114, 176,
283.
quantifiers, 54.
quotation-functions, 161.
Relations, many-termed, logic of, 231.
— two-termed, logic of, 222.
relativization theorem, 69.
ring, Boolean, 449.
--------atomistic and atomless, 450.
--------completely additive, 453.
Satisfaction, 189, 214, 224, 232, 239,
244, 264, 401. 416.
— semantical ambiguity of, 225.
satisfied by (a logical matrix), 41,
425.
satisfy (a given function), 116, 193,
200.
— by infinite sequence, 191, 193.
semantical unification of the variables,
method of, 227.
semantics of language, 252, 273, 401,
403.
466 SUBJECT IN D E X

sentence, 176, 178. sequences, set of homogeneous, 122.

— asserted, 166, 162. set, basic, 69.
— atomic, 284. — definable, of real numbers, 111,
— basic, 67. 118.
— (meaningful), 62, 178, 284. «-----------------of order 1, 120.
— primitive, 179. — elementary, linear, 132.
— provable, 186, 182, 206, 286, 263 — membership, 114.
n., 411, 423. — open, 428.
— — in language of the calculus of — power of , 61,
classes, 186, 195, 208. — projective, 143.
— — in languages of finite order, 209, — regular, open, 24 n., 448.
235. sets, fields of, 453.
--------infinite order, 241, 246. solid, 27.
--------in everyday language, 165. space, metricizable, 430.
— true, in every individual domain, — topological, 428.
201, 206. --------open, 428.
sentences, 30, 39, 62, 116, 176. --------dense (in S)> 428.
— calculus of, 347. --------isolated, 429.
— complete set of, 287. --------normal, 429.
— consistent set of, 287. --------with oountable basis, 429.
— equiform, 31 n. submatrix, 426.
— logical product of infinite sequence substitution, 40, 127, 180, 285,
of, 363. — l&'P of, 5, 6,
— logically increasing sequence of, sum (in Boolean algebra), 347.
362. — logical, 114, 144, 175, 283.
— logically provable, 385-7. — — of systems, 351.
— maximal independent, 362. — (of elements of a class), 25.
— provable in Euclidean geometry, system, axiom, 40, 76.
388-9. --------categorical, 390.
— quantitative, 201. --------effectively interpretable in
— strictly increasing sequence of, logic, 391.
363. -------- finite, 35.
— system of all logically true, 83. --------non-ramifi&ble, 390.
sentential calculus, extended, 54. — axiomatizable, 35, 354-5.
--------intuitionistio, 352, 400. — convergent, 363.
-------- n-valued system of, 47. — complete 364-5.
— —- number of possible systems of, — deductive, 33, 40, 69, 125, 287.
51, 63. structural type of, 371.
-------- ordinary system of, 42, 399. — divergent, 304.
-------- restricted, 52, 899. — irreducible, 364-5.
— function, see function, — non -axiomatixable, 362, 364.
sequence, 171, 116, 121. — value, 424.
— finite, 121, 171. systems, calculus of, 347.
— infinite, 171.
— of sentences, logically increasing, Tangent, externally and internally,
362. 26.
-------- strictly increasing, 363. tautology, 419.
*— prolongation of, 122. theorem, 116, 182, 373.
— two-rowed, 227. — provable, 221.
sequences, infinite, satisfying a given —»relativization, 69.
sentential function, 191. theory, deductive, cardinal degree of
-------- o f sentences, logical product of, capacity of, 368.
363. --------degree of completeness of, 388.
 SUBJECT IN D E X 46?

theory, structural type of, 370. type, semantical. 210.

— — characteristic pair of. 370.
*—•— limitation* of the means of ex­
pression. of, 384,
— elementary, of dense order, 372,
--------of isolated order, 375.
--------of identity, 378,
general elementary, of order, 379,
------- of a binary relation, 380,
topology, 380.
true sentence (definition), 195.
------------ (in every individual do­
main). 201, 206,
—— ----- in languages of finite order,
209, 235,
— — —~ — of infinite order, 241, 246.
conditions for definition of, 273.
'— semantical definition of, 237.
truth, concept of, 152,
--------in colloquial language, 154.
— < — summary, 265,
—•concept of, conditions for con­
structing definition;. 273.
— structural definition of, 254.
type, logical, 215.
Unification, semantical, of the vari­
ables, 227, 233.
unit (universa]) element, 347.
universality, of oolloouiaJ l&nguag.,
164.
universe of discourse 335, 139 n., 320,
347.
Value function {of a logical matrix),
41, 57.
>— system (of a logical matrix), 424.
variable. 168, 233.
— bound, 214, 372.
designated, 290.
— free, 214, 284, 372.
-----— of & sentential function (of the
calculus of classes), 178.
variable®, method of semantical unifi­
cation of the, 227.
Well-ordering, SS2.
Zero (empty) element, 347.
 IN D E X OF NAM ES OF P E R SO N S
Ackermann, W ., 241.
Ajdukiewicz, K ., 32, 161, 216. 237.
419.
Aristotle, 165.
Bemays, P ., 40, 199, 241, 423-4.
Bernstein, B. A ., 349.
Beth, E. W ., 300.
Birkoff, G., 464.
Carnap, R .f 113. 161, 170, 216-16,
218, 270, 277, 297, 311, 390-1, 413,
416, 418.
Church, A ., 32, 241.
Chwistek, L ., 113, 215, 242.
Couturat, L ., 12, 14, 18, 322.
Dedekind, R ., 175-6.
Diamond, A. H ., 47.
Eilenberg, S., 429.
Fraenkel, A ., 174, 234, 248, 270, 390.
Frege, G., 175.
Gentzen, G., 423, 428,
Glivenko, T ., 424.
Godel, K ., 240,247, 251, 252, 254, 258,
260, 274, 277-9, 288, 295, 390, 412,
426.
Grelling, K ., 166, 402.
Hausdorff, F ., 359, 381.
Herbrand, J., 200, 241.
Hertz, P ,, 70.
Heyting, A ., 352, 357, 400, 421, 423-4.
Hilbert, D ., 60, 260, 295, 399, 423,
Hilbert, D ,. and Ackermann, W ., 54,
55,116,178, 199, 221, 231, 240, 254.
Hiz, H ., 2.
Huntington, E. V., 26, 179, 186, 349,
379.
Husserl, E ., 215.
Jankowski, S., 26, 47, 58, 426.
Jbnsson, B ., 323.
Knaster, B ., 26.
KotarbMflki, T., 54, 153, 155-6.
Kuratowski, C., 24, 113, 150, 242,
392, 428.
Langford, C. H ., 374-5.
Le&niewski, S., 8, 24, 25, 31, 44, 53,
54, 155, 180, 215, 218, 221, 223,
242, 263, 333, 402.
Lewis, C. I., 115.
Lindenbaum, A ., 32, 33, 36, 38, 48,
71, 255, 297, 307, 366, 384, 393.
Lowenheim, L ., 241.
Lukasiewicz, J., 4, 8, 14, 21, 30, 33,
38, 40, 44, 47, 49, 53, 65, 115, 157,
168, 181, 221, 373, 399, 400.
Lusin, N .f 144, 148.
Mazurkiewicz, S., 383.
McKinsey, J. C. C., 47, 194, 246, 454.
Meredith, C. A ., 45, 51.
Moore, E. H ., 36.
Mostowski, A ., 237, 254, 370, 385.
421.
Neumann, J. v., 113, 218, 223, 242.
Nelson, L., 165, 402.
Nicod, J., 24, 29, 44.
Nikodym and Sierpiriski, 149.
Padoa, A ., 299? 300, 305.
Peano, G., 313.
Peirce, C. S,, 40, 54.
Fieri, M., 27.
Post, E. L., 40, 426.
Presburger, M., 205, 221, 383.
Quine, W . V ., 32. 53.
Rose, A ., 49.
Rosser, J. B ., 50.
Scholz, H ., 417.
Schonfinkel, M., 199y 241.
Schroder, E ., 9, 12, 14, 40, 115, 168,
186, 207, 222, 334-5.
Schroter, K ., 44, 51, 90.
Sheffer, H . M., 2, 36, 362.
Skolem, Th., 241, 271, 374, 378, 381.
Shipecki, J., 50.
Sobocixiski, B ., 38, 44, 45.
Stone, M, H ., 360, 352, 421, 449, 461.
 IN D E X OF NAMES OF PERSONS
Tarski, A ., 25, 32, 40, 44, 53-54, 58,
200, 248, 277, 279, 297, 323, 341,
368, 382, 384, 387, 408, 421, 449,
454.
Turquette, A. R ., 60.
Veblen, O., 174, 282, 307, 309, 311.
Wajsberg, M ., 38, 45-47, 49, 50-51,
53, 59.
469
Wang, H ., 237.
Whitehead, A. N ., 24, 29.
Whitehead, A. N ., and Russell,
B. A. W ., 1, 2, 5, 8, 22, 113, 156,
161, 168, 170, 176, 178, 183, 185,
215, 223, 233, 242, 279, 297, 380.
Wittgenstein, L ., 420.
Woodger, J. H ., 25.
Zermelo, E ., 70, 271.
 IN D E X OF SYMBOLS

A, 281. Z£, 56-67.

Apq, 188. L+, 52.
X ei.
a *(p )t 3. H *. 426.
A t , 334. M ", 427.
91, 43, 76, 366. M*,i, 123.
9(l(X), 86, 76.
9IqU), 72. Np, 39, 280.
n(x), 31, 40.
B, 320, 347. ng(p)> 3.
58(X), 35, 88. Nc(x), 387.

C p g , 39, 280. C(B), 428.

®(®, y)» 31, 39. 434.
0 (B ),
(E, 363.
C n ( X ) , 30, 40, 63, 182, 285, 346. SP( A ) , 61.
O n x( X ) , 56. Pr, 182, 275.
On(0), 70-71. «*(y). 65.
C tk, 200.
J l, 1, 54, 61, 114, 280.
D (S ), 122.
a { S ) , 122. ^ ’ 12S'
2 , 128.
2 ( F ) , 136.
2}, 128. J ^ y . 322, 325.
2 f ( & ) , 136.
J g /M , 322.
3 [...], 61, 321.
m TT y , 353
<E, 61. reft
®(M), 41, 425. Dfc, 176. 232.
«n t(4 ), 94. *• ««•
e, 61, 114, 171.
fljU . 232, 283.
6, 171.
PI, 9. f i i ’4 ) , 6 1 .
P, 284.
^VtK* 425- R k, 172.
3. Rs , 123.

0 (4 ). 10 1. 5 , 30, 63, 178, 284.

S x , 65.
IKb, 426. -S/, 121-2.
I{jR), 451. 123.
J ( R ) , 450.
tut, 423. ~S, 127.
3 ,3 6 5 .
B b (X ),394.
•*,r> 1 75 .
-S&jf,, 285.
£ , 284. £ , 128;
i .,4 7 . 2 , 1, 61, 114.
 IN D E X OF SYM BOLS 471

SB, 34, 91, 287.

9B„, 288.

J g tk, 176, 346. X, 114, 347, 361, 387.

S, 343.
J jy , 322, 325.
ZK, 426.
2 # * > . 144. 3 » , 423.
j g /(*)> 322. to,61.
T F, 353. Q, 61.

S , 33, 70, 2»7, 360. { }, 42 fh., 61.

© *, 359. (negation) 2, 424.
<5X , 359. (equivalence) 34.
be?, 2, 348.
Ta , 289. D, 2, 348.
T m, 292. 422, 434, 451.
!T{a), 380. 7 * , 438.
Tr, 9, 196, 250, 267, 276.
tr {p ), 3.
*38-
283, 422, 434, 451.
., 2, 6, 114, 325, 347.
Ui, 178. 123.
Ui7. 283. + , 114, 324, 344, 347.
O*, 123. + , 351.
U, 85, 83.
•f» 123.
V, 135, 283. 171.
S , 34, 287, 386. £ , 170, 321 fh.
stTy, 172, 281.
vr2,2283, 422, 434, 461. = , 171, 323, 347.
A, 283, 422, 434, 461. 171.
 PRINTED IN
GREAT BRITAIN
AT THB
UNIVERSITY PRBS8
OXFORD
BY
GHARLK8 BATEY
PRINTS*
TO THB
UNIVERSITY