UNDECIDABLE THEORIES BY
ALFRED TARSKI Professor of Mathematics, University of California, Berkeley
IN COLLABORATION WI...

Author:
Lev D. Beklemishev

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!

UNDECIDABLE THEORIES BY

ALFRED TARSKI Professor of Mathematics, University of California, Berkeley

IN COLLABORATION WITH

ANDRZEJ MOSTOWSKI Professor of Mathematics, University of Warsaw AND

RAPHAEL M ROBINSON Professor of Mathematics, University of California, Berkeley

I

1971

NORTH-HOLLAND PUBLISHING COMPANY AMSTERDAM

© North-Holland Publishing Company -1953 All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, without the prior permission of the copyright owner

ISBN: 0720422469

First edition 1953 Second printing 1968 Third printing 1971

PRINTBD IN TBB NKTIIBltLAM>S

PREFACE This monograph consists of three papers: A general method in proofs of undecidability, Undecidability and essential undecidability in arithmetic, Undecidability of the elementary theory of groups. While the first and the third papers have been written by the undersigned alone, the second paper is a joint work of A. Mostowski, R. M. Robinson, and the undersigned. The three papers are referred to throughout the monograph by Roman numerals I, II, III. The introduction to paper I is thought of as an introduction to the whole work and gives an idea of the scope of the problems discussed and the results obtained. The notations and symbolic conventions introduced at any place of the monograph are applied in the whole subsequent discussion. The work contains results obtained over a long period of time, 1938-1952. The first draft of paper II was prepared by Mostowski in 1949 and contained exclusively results found by him and Tarski in the pre-war period. However, its publication was postponed and the text was considerably modified, so as to embody the results and simplifications subsequently obtained by Robinson and partly by Tarski; the final draft was written jointly by these two authors in 1951-52. The other two papers were prepared for publication during the same period. It was originally planned to publish the work as a series of connected papers in some regular mathematical periodical. However, the Editors of the series Studies in Logic found the material integrated enough to appear as a separate volume in this series; their offer was accepted with appreciation. A large part of the technical work on the monograph was done during the period when Tarski and, for a shorter time, also Robinson were engaged in a research project in the foundations of mathematics sponsored by the Office of Ordnance Research, U.S. Army. VIII

l'REFAOE

IX

In this work the authors received much valuable advice and assistance from several friends-Mr. C. C. Chang, Professor Leon Henkin, Professor J. C. C. McKinsey, Dr. Julia Robinson, and Mr. R. L. Vaught. Professor E. W. Beth and Mr. F. W. J. Marx have greatly obliged the authors with their help in reading proofs. University of California Berkeley, April 1953

Allred Tarski

I

A GENERAL METHOD IN PROOFS OF UNDECIDABILITY BY

ALFRED TARSKI

I A GENERAL METHOD IN PROOFS OF UNDECIDABILITY 1.1. Introduction. 1 By a decision procedure for a given formalized theory T we understand a method which permits us to decide in each particular case whether a given sentence formulated in the symbolism of T can be proved by means of the devices available in T (or, more generally, can be recognized as valid in T).2 The decision problem for T is the problem of determining whether a decision procedure for T exists (and possibly of exhibiting such a procedure). A theory T is called decidable or undecidable according as the solution of the decision problem is positive or negative. As is well known, the decision problem is one of the central problems of contemporary metamathematics. Since only few theories turn out to be decidable 3, most endeavors are directed toward a negative solution. In the attacks on the decision problem for various special theories and, more specifically, in the attempts at obtaining a negative solution of this problem two different methods are applied. The first, direct method is essentially based upon ideas which originated with Godel and were used for the first time in [7], in the proofs of his well-known incompleteness theorems. When applying this method we have to make use of some deep properties of notions which are involved in the precise statement of the decision problem; in fact, of the notions of general recursive functions and sets. The method is rather involved and can be applied only to those theories in which a sufficiently strong number-theoretical apparatus 1 The observations contained in this paper were made in 1938-1939; they were presented by the author to a meeting of the Association for Symbolic Logic in 1948, and were summarized in [34]. (The numbers in square brackets refer to Bibliography.) 2 The meanings of various terms used in this introduction will be explained in a more detailed and precise way in later sections of the paper. S Some decidable theories are discussed or at least mentioned in [30].

3

4

A GENERAL METHOD IN PROOFS OF UNDECIDABILITY

can be developed. With the help of this method some fundamental results concerning the decision problem have been obtained. In fact, it was shown by Church in [3] that the solution of the decision problem for Peano's arithmetic (and some fragments of it) is negative, and Rosser proved later in [24] that the same applies to every consistent theory which is an extension of Peano's arithmetic. We express these results briefly by saying that Peano's arithmetic is, not only undecidable, but also essentially undecidable. - The theoretical foundations of the direct method will be outlined in II, in particular in II.2 and 11.4. The second, indirect method consists in reducing the decision problem for a theory T1 to the decision problem for some other theory T2 for which the problem has previously been solved. In the original form of this method, to establish the undecidability of a theory T1 one tried to show that either (i) T1 can be obtained from some undecidable theory T2 by deleting finitely many axioms from the axiom system of T2 (but without removing any constant from the symbolism of T2 ) , or else that (ii) some essentially undecidable theory T2 is interpretable in T1 • By applying the procedure (i) and by taking a fragment of Peano's arithmetic (suitably modified) for T2 , Church proved in [2] that the first-order predicate logic is undecidable. When applying the procedure (ii), Peano's arithmetic is usually taken for T2 ; in this way, e.g., various axiomatic systems of set theory have turned out to be undecidable. The indirect method in its original form was rather restricted in applications. Only in exceptional cases can a theory for which the decision problem is discussed be obtained from another theory, which is known to be undecidable, simply by omitting finitely many sentences from the axiom system of the latter. On the other hand, one could hardly expect to find an interpretation of Peano's arithmetic in various simple formalized theories, with meager mathematical contents, for which the decision problem was open. With regard to theories of this kind both the direct and the indirect methods seemed to fail. However, it has proved to be possible to extend and modify the indirect method (by combining some features of the two procedures indicated above) so as to widen

TltEORlES

wrra

STANDARD FORMALIZATION

5

considerably its range of application. In fact, a simple argument shows that, in order to establish the undecidability of a theory TI , it suffices to show that some essentially undecidable theory T2 can be interpreted, not necessarily in TI , but (what is much easier) in some consistent extension of TI - provided only that T2 is based upon a finite axiom system. As a consequence of this last condition, Peano's arithmetic cannot any longer be used as T2 since it is not based upon a finite axiom system. On the other hand, examples of essentially undecidable theories which are based upon finite axiom systems and are readily interpretable in other theories have been found (by the direct method) among fragments of Peano's arithmetic. Using this fact and applying the extended indirect method, many formalized theories-like the elementary theories of groups, rings, fields, and lattices-have recently been shown to be undecidable. The aim of the present paper is to set up theoretical foundations for the general method just described. The paper is conceived as a framework for later publications in this field, with the idea of permitting the authors to avoid the repetition of some lengthy, though elementary, discussions. With this in view, we give in 1.2 a rather detailed description of formalized theories to which the method applies. In 1.3-5 we define the notions involved in the description of the method, and we explicitly state and prove a few elementary theorems upon which this method is based. Finally, 1.6 contains a short survey of the results obtained so far with the help of the method discussed. Throughout the whole paper the discussion has an informal character. 1.2. Theories with standard formalization. The theories discussed in this paper will be referred to as theories with standard formalization. They can be briefly characterized as theories which are formalized within the first-order predicate logic (with identity, without variable predicates). ' & For a. detailed discussion of predicate logic and theories formalized within this logic consult, e.g., [9]; the discussion is spread over vol. 1. of this work, and is summarized and supplemented in vol. 2, pp. 375-391. The reader

6

A GENERAL :ME THOD IN PROOFS OF UNDECIDABILITY

The symbols which occur in expressions of a given theory Tare divided into variables and constants. The set of variables is assumed to be denumerable and hence infinite; the set of constants is either finite or denumerable. All the variables are treated as ranging over the same set of elements. The constants are divided into logical and non-logical ones. The logical constants are the sentential connectives-the negation sign ,..." the implication sign ~, the equivalence sign ~, the disjunction sign v, and the conjunction sign 1\; the quantifiers-the universal quantifier A and the existential quantifier V; and, finally, the identity symbol =. 5 The non-logical constants are the predicates (or relation symbols), the operation symbols, and the individual constants. With every predicate and every operation symbol a positive integer is correlated which is called the rank of the symbol. Thus, we may have in T unary predicates and operation symbols (i.e., symbols of rank 1), binary predicates and operation symbols (symbols of rank 2), etc. The identity symbol, though regarded as a logical constant, is included in the set of binary predicates. In practice, in addition to variables and constants, the so-called technical symbols, like parentheses and commas, are also used in constructing expressions; theoretically, however, these technical symbols can be dispensed with. Among expressions (i.e., finite concatenations of symbols) we will find there elaborations of certain points which have been disregarded in the present account (e.g., the definitions of free and bound variables, the specification of logical axioms and operations of inference, and the proofs of the deduction theorems). 5 In view of the general character of our present discussion we have not considered it necessary to maintain a strict distinction between expressions of formalized theories and their metamathematical designations. No inconsistency will arise, however, if we agree to regard all the symbolic expressions used in this paper, not as expressions of the theories discussed, but as metamathematical denotations of such expressions. In this case the letters "x", "y", "z", ... should be regarded as metamathematical variables which range over variables of the theory under discussion, and usually it should be assumed that two different letters represent two distinct variables of the theory.

THEORIES WITH STANDARD FORMALIZATION

7

distinguish terms and formulas. The simplest, so-called atomic, terms are the variables and the individual constants; a compound term is obtained by combining n simpler terms by means of an operation symbol of rank n. Similarly, an atomic formula is obtained by combining n arbitrary terms by means of a predicate of rank n; compound formulas are built from simpler ones by means of sentential connectives and quantifier expressions (i.e., quantifiers followed by variables, like Ax or Vy). An occurrence of a variable in a formula may be either free or bound; a formula in which no variable occurs free is called a sentence. Two further notions, those of logical derivability and logical validity, are involved in the metamathematical discussion of any theory T. They are usually introduced in the following way. First, we single out certain sentences of T which are referred to as logical axioms. Secondly, we describe certain (finitary) operations, the so-called operations of inference, which when performed on sentences yield new sentences. Usually the set of logical axioms is infinite while the set of operations of inference is finite. The most important operation of inference is that of detachment (modus ponens), which when applied to two sentences f/J and f/J --+ P yields the sentence P. In fact, it proves to be possible, by selecting a suitable set of logical axioms, to use the operation of detachment as the only operation of inference in formulating adequate definitions of derivability and logical validity. 6 A sentence is now said to be logically derivable or simply derivable from a set Aof sentences if it can be obtained from sentences of A and from logical axioms by performing operations of inference an arbitrary number of times. A sentence is called logically valid (or logically provable) if it is derivable from the set of logical axiomsor, what amounts to the same thing, from the empty set of sentences. Still another method of defining these two notions is available which, however, essentially involves the use of some semantieal 8 Cf. in this connection the remarks in [33], pp. 507 f. A suitable system of logical axioms is given in [19], pp. 80-85; the system is to be supple. mented by axioms involving the identity symbol.

8

A GENERAL :METHOD IN PROOFS OF UNDECIDABILITY

notions and, in particular, of the notion of satisfaction. 7 We assume that all the non-logical constants of T have been arranged in a (finite or infinite) sequence Co, ... , C.., ... ), without repeating terms. We consider systems ffi formed by a non-empty set U and by a sequence 00' .. . ,0.., ... ) of certain mathematical entities, with the same number of terms as the sequence of non-logical constants. The mathematical nature of each 0.. depends on the logical character of the corresponding constant C... Thus, if C.. is a unary predicate, then 0 .. is a subset of U; more generally, if C.. is an m-ary predicate, then 0.. is an m-ary relation the field of which is a subset of U. If C.. is an m-ary operation symbol, 0 .. is an m-ary operation (function of m arguments) defined over arbitrary ordered m-tuples Xl' ••• , xm ) of elements of U and assuming elements of U as values. If, finally, C.. is an individual constant, 0 .. is simply an element of U. Such a system (sequence) ffi = 11 which are logical axioms or members of A and for which the sentence (11)1 i\

••• i\

11>11) -+ lJF

is tautological. Hence, by including all tautological sentences in the set of logical axioms, we obtain a simple characterization of derivability in terms of logical axioms. It may be mentioned that Deduction Theorem II has sometimes been used as a definition of the notion of derivability. See [33], p. 507, and [38], pp. 10-11, footnote 10 (where references to earlier papers of Ajdukiewicz can also be found). 9 We can of course transform this definition by means of Deduction Theorem II so as to obtain a characterization of valid sentences in terms of logically valid sentences and non-logical axioms. Under conditions stated in footnote 8 a further simplification is possible; we obtain a simple characterization of valid sentences in terms of arbitrary (logical and non-logical) axioms, a characterization in which the notions of derivability and logical validity do not appear and no repeated application of operations of inference is involved.

THEORIES WITH STANDARD FORMALIZATION

II

the notion of validity has been defined in one way or another. We also assume that under this definition every sentence which is logically derivable from a set of valid sentences is itself valid, and that consequently every logically valid sentence is valid; this is the only condition imposed upon the definition of validity. A possible realization in which all valid sentences of a theory Tare satisfied is called a model of T. In all the theories with standard formalization the same symbols are assumed to be used as variables and logical constants; apart from differences in non-logical constants, the same expressions are regarded as formulas, sentences, logical axioms, and logically valid sentences. However, the notions of validity in these theories may of course exhibit essential differences. A theory is uniquely determined by the set of all its valid sentences; two theories are regarded as identical if their sets of valid sentences coincide. An axiomatic theory is uniquely determined by its non-logical constants and non-logical axioms. A theory T1 is called a subtheory of a theory T2 if every sentence which is valid in T1 is also valid in T2 ; under the same conditions T2 is referred to as an extension of T1 • An extension T2 of T1 is called inessential if every constant of T2 which does not occur in T1 is an individual constant and if every valid sentence of T2 is derivable in T2 from a set of valid sentences of T1 • (By saying that a sentence (/J is derivable in a theory T from a set A we stress the fact that, in deriving (/J, we may use both sentences of A and logical axioms of T. It is easily seen that, whenever (/J is derivable from A in some theory T, it is also derivable from A in every theory T' which contains all the non-logical constants occurring in (/J and in sentences of A.) If T1 is axiomatic, then an inessential extension of T1 ' is obtained by adding some new individual constants, but without adding any new non-logical axioms. An extension T2 of T1 is referred to as a finite extension if there is a finite set A of valid sentences of T2 such that every valid sentence of T2 is derivable from a set of sentences which are valid in T10r belong to A. Clearly every inessential extension is a finite extension.

12

A GENERAL METHOD IN PROOFS OF UNDECIDABILITY

Among the extensions common to two given theories T1 and T2 there is always a smallest one, which is a subtheory of any other common extension; this smallest common extension is referred to as the union of the given theories. The union T of T1 and T2 is fully characterized by the following two conditions: (i) the set of all non-logical constants of T is the (set-theoretical) union of the sets of all non-logical constants of T1 and T2 ; (ii) a sentence is valid in T if and only if it is derivable in T from a set of sentences which are valid in T10r T2' (Notice that condition (i) unambiguously determines the notions of a sentence of T and of a logical axiom of T, and hence also the notion of derivability in T.) If the theories T1 and T2 are axiomatic, we can construct T by postulating, in addition to (i), the analogous condition for the set of non-logical axioms. A theory T is called consistent if not every sentence of T is valid in T; or, in an equivalent formulation, if there is no sentence (/J such that both ' is the formula obtained from 4"> by replacing x everywhere by u. Intuitively speaking, sentence (iii) expresses the fact that, in every possible realization of T, for any two elements represented by y and z there is just one element represented by x such that the three elements satisfy the formula 4">. We proceed analogously with predicates and operation symbols of arbitrary ranks; individual constants are treated in this context as operation symbols of rank O. Let now T1 and Tll be any two theories. First assume that T1 and T. have no non-logical constants in common. In this case we say that Tll is interpretable in T1 if we can extend T1, by including in

INTERPRETABILITY AND WEAK INTERPRETABILITY

21

the set of valid sentences some possible definitions of the nonlogical constants of T2' in such a way that the resulting extension of T1 turns out to be an extension of T2 as well. Speaking precisely, T2 is interpretable in T1 if and only if there is a theory T and a set D satisfying the following conditions: (l¥) T is a common extension of T1 and T2 , and every constant of T is a constant of T1 or T2 ; (fJ) D is a recursive set of sentences which are valid in T and which are possible definitions in T1 of non-logical constants of T2 ; (y) each non-logical constant of T2 occurs in just one sentence of D; (15) every valid sentence of T is derivable (in T) from a set of sentences each of which is valid in T1 or belongs to D. In case T2 has only finitely many non-logical constants, the term "recursive" in condition (fJ) can clearly be omitted. In the general case, when T1 and T2 may have some common non-logical constants, we first replace the non-logical constants in T2 by new constants not occurring in T1 (different symbols by different symbols), without changing the structure of T2 in any other respect; if the resulting theory T~ proves to be interpretable in T1 , we say that T2 is also interpretable in T1 • It is easily seen that, in order to prove the interpretability of T2 in T1 in the general case, it is sufficient, though not necessary, to construct a theory T and a set D satisfying conditions (l¥)- (15), with the difference that D is assumed to contain possible definitions of only those constants of T2 which do not occur in T1 • A theory T2 is said to be weakly interpretable in T1 if T2 is interpretable in some consistent extension of T1 which has the same constants as T1 • In order to prove the weak interpretability of T2 in T1 it suffices to show that T1 and T2 have a common consistent extension T satisfying the following condition: (e) there is a recursive set D of valid sentences of T such that, for every non-logical constant C of T2 which does not occur in T1 , some possible definition of C in T1 belongs to D. In case T2 has only finitely many nonlogical constants, condition (e) can clearly be replaced by a simpler one: (C) for every non-logical constant C of T2 which does not occur in T1 , some possible definition of C in T1 is valid in T. Moreover, in both (e) and (C) the notion of a possible definition as applied

22

A GENERAL METHOD IN PROOFS OF UNDECIDABILITY

to operation symbols and individual constants can be understood in a wider sense than the original one. If, e.g., the operation symbol does not occur in T1 , then as a possible definition of in T1 we can regard any sentence of the form

+

+

Ax Ay Az(x

= y +z ~

... , lJIl I are valid in T. It is known that, whenever a sentence e is logically valid, then the sentence e* is also logically valid. (This can be proved, e.g., by showing that, if e holds in every possible realization, the same applies to e*.) e* clearly coincides with ((lJIt»)* 1\ ••• 1\ (lJI~P»)*] -+ if>*. From the definitions of lJICP) and lJI* it is easily seen that every sentence lJIof T is logically equivalent to (lJIcP»)*. Hence the sentences (lJIt»)*, ... , (lJI;;»)* are valid in T, and therefore if>* is valid in T. Consequently every sentence of A' is valid in T. If if> is any sentence valid in T, then if>CP) is valid in TCP) and therefore derivable from B'. By again applying Deduction Theorem II, we show that (if>{P»)*, and hence also if>, is derivable from A'. We conclude that T is axiomatizable. We have thus established conclusion (i) of our theorem. Moreover we notice that, whenever the set A (in the argument outlined above) is finite, the set B is also finite, provided there are only finitely many individual constants and operation symbols occurring in T. Furthermore, if B' is finite, then A' is always finite. Consequently, conclusion (ii) also holds, and the proof is complete. If T is an axiomatic theory, then, by Theorem 9, T{P) can also be represented as an axiomatic theory; from the proof of this theorem we learn how to construct an adequate axiom system for T{P) if the axiom system of T is known. Supplementing Theorem 9 (ii), we can show that, whenever T{P) is finitely axiomatizable and consistent, T can contain only finitely many individual constants and operation symbols. 10. Let T be any theory and P a unary predicate which is not a constant of T. Then T(P) is essentially undecidable if and only if T is essentially undecidable. PROOF: We want first to show that THEOREM

(1)

TCP) is interpretable in T.

For this purpose we construct an extension T* of T by adding P to the set of constants of T and by stipulating that a sentence is

28

A GENERAL METHOD IN PROOFS OF UNDECIDABILITY

valid in T* if and only if it is derivable from the set of all valid sentences of T supplemented by the following sentence e: Ax(Px ~ x

=x).

Obviously, e is a possible definition of PinT. C/J being any sentence of T, it is easily seen that C/J(IX) and 4>(IX,fJ), which would also be suitable for our purposes. In fact, we could agree to use 4>(IX) as an abbreviation for the expression I\u(u

=IX

~

4»

in case u does not occur in IX, and for the expression I\w[w

=IX

~ I\u(u

=w

~

4>)]

in case u occurs in IX while w is a variable which does not occur either in 4> or in IX. We may assume that all the variables have been arranged in an infinite sequence, and choose as w the first variable in this sequence occurring neither in 4> nor in IX. 4>(IX,fJ) is defined analogously. Under this alternative definition the sentence is logically valid for every formula 4> and every term IX. 4 II.2. Definability in arbitrary theories. Let T be any theory with standard formalization. Ii We assume that in this theory an infinite sequence of terms .110,.111> •.. , .11", • •• containing no variables is available. A subset P of N, i.e., a set P of natural numbers, is said to be definable in T (relatively to the sequence of terms .110' .11 1 , ••. ) if there is a formula 4> of T such that 4>(.11,,) is a sentence valid in T whenever n E P, and", 4>(.11,,) is a sentence valid in T whenever n EN and n ¢ P. 6 Such a formula 4> is said to define P. From the fact that 4>(.11,,) is a sentence we conclude (looking back at the , Compare [35]. fi Actually the discussion in this section applies to arbitrary formalized theories whose logical basis includes at least the first-order predicate logic with identity. With small changes, the discussion could even be extended to theories in which no quantifiers occur. S The notion of definability was first discussed in [36]. A more general notion, which comprehends both the notion of definability in the sense of [36] and that of general recursiveness as particular cases, was introduced in [15], pp. 72 ff.; it is still not as general as the notion discussed in this pa.per.

DEFINABILITY IN ARBITRARY THEORIES

45

definition of lP(lX) in II. 1) that the variable u does not occur bound in lP and that no variable different from u occurs free in lP. A function F is said to be definable in T if there is a formula lP such that (i) for any n,p E N with Fn = p, lP(LI", Ll Il ) is a sentence valid in T; (ii) for any n,p EN with Fn =1= p, --lP(LI",Ll Il ) is a sentence valid in T; (iii) for every n EN, AuAv[lP(LI",u) 1\ lP(LI", v) -+ u

=v]

is a sentence valid in T. As before, such a formula lP is said to define F; the variables u and v do not occur bound in lP, and no variable different from u and v occurs free in lP. It is easily seen that conditions (i) and (iii) can be replaced by the following condition: (i') for every n E N, the sentence

=

Av[lP(LI",v) B-V LI.II'''] is valid in T. Under the assumption that, for all n,p E N with n =1= p, the sentences LI" '# Ll Il are valid in T, condition (ii) proves to be a consequence of (i) and (iii), and hence can be dispensed with. Even without this assumption, however, the omission of condition (ii) would not invalidate the proof of the essential results of this section, i.e., Theorem 1 and Corollary 2. By modifying in an obvious way the definition of definable sets, we arrive at the notion of definable binary relations (sets of ordered couples) and, more generally, of definable n-ary relations. Since functions (with one argument) can be treated as special binary relations, we obtain in this way a new definition of definable functions. The new definition is weaker than the original one and differs from the latter by the absence of condition (iii). It suffices for the proof of Theorem 3 and Corollary 4 outlined below; however, condition (iii) will be essentially involved in the proof of Theorem 1 and Corollary 2. When using the term "definable

46

UNDECIDABILITY IN ARITHMETIC

functions" in the present article, we shall have in mind the original definition of this term. On the other hand, the definability of sets reduces under certain conditions to the definability of functions. In fact, let P be any set of natural numbers, and let Op be its characteristic function defined by the formulas: Opn = 0 if n Opn = 1 if n

E E

P, Nand n ¢ P.

The definability ofOp (even in the weaker sense) implies that of P; in fact, if a formula C/J defines Op, then the formula C/J(u,L1 o) is easily seen to define P. Under the assumption that, for all n,p EN with n =1= p, the sentences ,1,. #- L1 p are valid in T, it turns out that, conversely, the definability of P implies that of Op; for, if a formula lJ' defines P, then the formula [lJ'(u)

1\

v

=

,10] v [,...., lJ'(u)

1\

V

=

,11]

proves to define 0 rConsider now any fixed one-to-one correspondence between natural numbers and expressions of T. For the time being we make no other assumptions regarding the nature of this correspondence. Given a natural number n, the correlated expression of T will be denoted by En; conversely, given an expression C/J, we shall denote the correlated natural number by Nr(C/J). Let D (the diagonal function) be the function defined by the formula Dn = N r(E,.(L1,.)) ; thus Dn is the number correlated with the expression which is obtained from E,. by replacing everywhere the variable u by the term ,1,.. Let V be the set of all natural numbers n such that E,. is a sentence valid in T. Using this notation we can establish the following result: THEOREM 1. If the theory T i~ consistent, then the function D and the set V are not both defiruJ1Jle in T.

DEFINABILITY IN ARBITRARY THEORIES

47

Assume, to the contrary, that both D and V are definable in T. Thus there are formulas (/) and 1JI satisfying, for every natural number n, the following conditions: PROOF:

(1) the sentence Av[(/)(L1",v)

B-

v

is valid in T; (2) (3)

if n if n

E

=L1

Dn ]

V, then the sentence 1JI(L1,,) is valid in T;

f# V, then the sentence '" 1JI(L1,,) is valid in T.

The variable v does not occur free in 1JI. Without any loss of generally we can assume that v does not occur in 1JI at all and that, consequently, 1JI(v) is a regular substitution of 1JI.

Let m = Nr(Av[(/)(u,v)

-i>- ' "

1JI(v)])

so that and consequently, (4)

Em(L1m ) = Av[(/)(L1 m,v)

-i>- ' "

1JI(v)].

If the sentence Em(L1 m ) is valid in T, then, by (1) (with n = m) and (4), the sentence", 1JI(L1.vm) is valid. If E m(L1m ) is not valid, then Nr(Em(L1m }} f# V; and since, by the definition of the function D, (5)

Dm

=

Nr(Em(L1m }} ,

we conclude by (3) that in this case '" 1JI(L1.vm) is valid as well. Thus, (6)

the sentence '" 1JI(L1.vm) is valid in T.

By (1) and (6), the sentence Av[(/)(L1 m,v) -i>- ' " 1JI(v)] is valid. Consequently, by (4) and (5), Dm E V; and therefore, by (2), (7)

the sentence 1JI(L1 Dm ) is valid in T.

(6) and (7) imply that the theory T is inconsistent, contrary to the hypothesis of the theorem. This completes the proof. Theorem 1 and its proof represent a metamathematical reconstruction and generalization of arguments involved in various

48

UNDECIDABILITY IN ABITHMETIC

semantical antinomies and, in particular; in the antinomy of the liar. The idea of this reconstruction and the realization of its farreaching implications is due to Godel [7]. The present version of this reconstruction is distinguished by its generality and simplicity. It applies to arbitrary formalized theories, and not only to those in which a comprehensive fragment of the arithmetic of natural numbers can be developed; to a large extent it is independent of the way in which the notion of validity has been defined for a given theory, and in particular it does not involve the notion of a formal proof within this theory; it does not use the apparatus of recursive functions-although this apparatus will play a fundamental role in applications of Theorem 1 to the decision problem. From now on we assume that the correspondence between natural numbers and expressions of T satisfies the usual conditions concerning the recursiveness of certain functions and sets under this correspondence (cf. 1.3). The most fundamental of these conditions can be expressed as follows: Let (/> ~ lJI denote the concatenation of two arbitrary expressions (/> and lJI; then the function G on NX N to N defined by the formula G(n,p) = Nr(En ~ E p )

is recursive. We also assume that the terms Ll o, Ll 1 , • •• form a recursive sequence, i.e., that the function H (on N to N) defined by the formula Hn

=

Nr(Ll n )

is recursive. From these assumptions it follows, in particular, that the function D is also recursive. 7 7 Actually the recursiveness of D is the only assumption involved in our further discussion. A proof of this assumption can be obtained, e.g., from the discussion in [15], pp. 73 ff. It may be interesting to notice that, under the alternative definition of IP(lX) mentioned at the end of 11.1, the function D is defined by the formula

Dn = Nr[Au(u = LIn -+ En)]' Using this formula and applying some elementary properties of recursive functions we can immediately derive the recursiveness of D from that of the functions G and H defined above in the text.

DEFINABILITY IN ABBITRARY THEORIES

49

There are close connections between the notions of recursiveness and definability. In 11.4 we shall see that many theories Tare known such that all the functions and sets which are recursive are definable in T (and conversely). The importance of this fact for our discussion is seen from the following corollary which is an easy consequence of Theorem 1: COROLLARY 2. 1fT is a consistent theory in which all recursive functions are definable, then T is essentially undecidable. PROOF: As has been stated above, the function D is recursive and therefore, by the hypothesis of the corollary, it is definable in T. Hence, by Theorem 1, the set V is not definable in T. If the set V were recursive, its characteristic function Oy (Oyn = 0 for n E V, Oyn = 1 for n ¢ V) would also be recursive and hence definable in T; as was mentioned before, this would imply the definability of the set V itself. Thus, the set V is not recursive; in other words, the set of all sentences which are valid in T is not recursive, and the theory T is undecidable. Since the definability in T implies the definability in every extension of T (relatively to the same sequence of terms Ll o, Ll1 , .•• ), our argument shows that every consistent extension of T is undecidable and that, consequently, T is essentially undecidable.

To conclude this section, we give two results, Theorem 3 and Corollary 4, concerning those theories in which every, or not every, definable function is recursive. These results follow in a simple way from the definitions of the notions involved and are less deep than Theorem 1 and Corollary 2; also they will find no essential applications in our further discussion. Nevertheless it will be interesting to compare the content of these results with that of Corollary 2. Every function which is definable in a con8i8tent and axiomatizable theory T is recursive. The same applies to every set 0/ natural numbers. PROOF: Using the intuitive notion of recursiveness, we argue as follows. Since T is axiomatizable, there is a recursive set A of THEOREM

3.

50

UNDECIDABILITY IN ARITHMETIC

sentences of T such that a sentence is valid in T if and only if it is derivable from A. Hence, as is well known, we can arrange all valid sentences of T in a recursive infinite sequence lJ'o, lJ'1' .. . . (In other words, there is a recursive function the range of which is the set V.) Let now F be any function definable in T, and let rp be a formula defining F. Then, for any given n EN, the sentence rp(Lt", Lt.Fto) is valid in T, and therefore it must occur somewhere in the sequence lJ'o, lJ'I> .... Hence, by checking the successive terms of this sequence, we must arrive in finitely many steps at a term lJ'm which is of the form rp(Lt.., Lt,,) for some pEN. If the numbers p and Fn were different, then the sentence rp(Lt.., Lt,,) would be valid in T, and therefore T would be inconsistent; hence p = Fn. We thus have a procedure which enables us, for any given n EN, to determine the function value Fn in finitely many steps; in other words, F is recursive. If, instead of a function, we consider a set of natural numbers, the proof remains practically unchanged. 1'.1

The converse of Theorem 3 is by no means true. In fact, as is seen from Corollary 2, if T is any consistent theory which is decidable or, more generally, which is not essentially undecidable, then there are recursive functions which are not definable in T. COROLLARY 4. 1fT is a consistent theory such that there is a function (or set of natural numbers) which is definable in T but not recursive, then T is essentially undecidable. PROOF: By Theorem 3, the theory T is not axiomatizable and hence a fortiori undecidable (cf. 1.3). For the same reason every consistent extension of T is undecidable, and hence T is essentially undecidable.

Corollaries 2 and 4 seem to provide us with two widely different methods of constructing essentially undecidable theories. Actually, however, we know no single case in which Corollary 4 would be of real help in a proof of essential undecidability. This is explained partly by the rather trivial character of Corollary 4 and partly by the fact that a theory T which satisfies the hypothesis of this corollary must have a much stronger property than essential

FORMALIZED ARITlIMETIO OF NATURAL NUMBERS

51

undecidability; in fact, by analyzing the proof of Theorem 3, we conclude that, in every consistent extension of T, the set of all valid sentences not only is not recursive, but cannot even be recursively enumerable. 8

II.3. Formalized arithmetic of natural numbers and its subtheories. The formalized arithmetic of natural numbers in which we are interested here will be referred to as Theory N. The set of all constants of N is assumed to consist of four symbols: an individual constant 0, a unary operation symbol S, and two binary operation symbols, + and -, The same assumption applies to all other theories discussed in this and the following two sections. Hence every possible realization of each of these theories is a system 9i = in which U is an arbitrary set, c is an element of U, F is a unary operation (function) on U to U, and + and· are two binary operations on UX U to U (cf. 1.2). To define the validity in N we consider a special realization 91 = N, 0, S, +, . >where all the symbols N, 0, S, ... have their usual arithmetical meaning (cf. 11.1); a sentence is said to be valid in N if and only if it holds in 91. We shall be interested in some axiomatic subtheories of N referred to as Theories P, Q, and R. The axiom system of Q consists of the following seven sentences :

for

all n,p

E

N

are shown to be valid in Q by induction on p. (All the inductive arguments in this proof are of course metamathematical inductions, and not inductions within Theory Q.) In fact, for p = 0, (1) is a particular instance of 8 4 ; when passing from p to p + 1, we apply 8 0 , Similarly, by an induction on p based upon 8 8 , 8 7 , and (1), we derive the sentences D2 : (2)

L1,. •L1"

=L1,.." for all n,p EN.

Consider now the sentences Da: (3)

L1,. #;. L1" for all n,p

We can clearly assume that n

E

< p.

N with n =1= p. For n

=

0 (and every p

> n)

54

UNDECIDABILITY IN ARITHMETIC

(3) is a particular case of e2 ; when passing from n to n apply e1 • e s and e5 imply z x 0 II X '# 0 ~ Vy[x Sy II S(z y) 0];

=

+

hence, by

e

=

+

+1

we

=

2,

(4)

Similarly, for every natural number n we have byes and z

+ x =LI"+1

hence, by (5)

'# 0 ~

II X

=Sy

Vy[x

e1> x

where N, 1, 0,8, + have their usual meaning while 8', $, and· are defined by the conditions

OTHER ARITHMETICAL THEORIES AND VARIOUS THEORIES OF RINGS

65

B'O = 0, B'n = Bn if n =I=- 0; nEf)p= 0 if n= 0 or p= 0; nEf) p= n +p-l if n=l=-O and p =1=-0; n· p = 0 for all n,p EN.

.. .

Let T+ be a theory with three constants: 0, S, +; a sentence of T+ is valid in T+ if it is valid in T6 , i.e., is satisfied in the system (N, 0, B, + where all the symbols have their usual meaning. The following sentences are clearly valid in T6 :

>

= Sy (x. =0 .Y =0) v (x. =Sy. . Y #= 0),. x =y + z (x =0 Y =0) (x =0 . =. 0) . x

. 0= SO,.

~

~

A

A

A

v

A Z

.

(Sx

= y +Z

v

A

Y

#= 0 A

Z

.

#= 0),

Y·Z = O. With the help of these sentences, arguing as in the case of T4 , we reduce the decision problem for T6 to that for T+. The latter theory is known to be decidable (see [17]); hence T6 is also decidable. (Another, perhaps more natural, example of a decidable extension of Q 6 is provided by a theory T~ which has the same constants as Q. A sentence is valid in T~ if it is satisfied in the system (U, 0, B, +, . where (i) U is the set consisting of the natural numbers and an additional element 00; (ii) 0, B, + have their usual meaning when applied to natural numbers, and in addition Boo = x + 00 = 00 + x = 00 for every x E U; (iii) x . y = 00 for all x,y E U. The decision problem for T~ can again be reduced to that for T+; however, the reduction is somewhat more involved than in the case of T6 . ) A sentence is valid in T7 if it is satisfied in the system (N, 0, B, +, . > where n . p = 0 for all n,p EN. We see at once that the decision problem for the theory T7 again reduces to that for the theory T+ described above in connection with T6' (The symbols

>

...

0, S, + are now replaced in T+ by 0, S, and the proof is complete.

+.) Hence T7 is decidable,

II.6. Extension of the results to other arithmetical theories and to various theories of rings. The results obtained in the preceding

66

UNDECIDABILITY IN ARITHMETIC

sections and in particular in n.5 can be carried over to formalized systems of the arithmetic of natural numbers with different sets of constants. Consider, e.g., the theory N* which differs from N only in that the unary operation symbol S has been replaced by the individual constant 1 denoting the integer one; the definition of validity in N* is entirely analogous to that in N. We can construct axiomatic subtheories P*, Q*, R* of N* closely related to the subtheories of N described in n.3-simply by replacing in the axioms of P, Q, R all the terms of the form Sex. by ex. 1. It is easily seen that Theorem 9 and Corollary 10 remain valid if we replace in them N, P, Q, R by N*, P*, Q*, R*, respectively. To prove this, it suffices to show that each of the theories N, P, '" is interpretable in the corresponding theory N*, P*, ... in the sense of 1.4, and then to apply Theorems 7 and 8 of 1.4. For instance, if we add the symbol S to the constants of Theory Q* and the sentence

+

x

=Sy

~x

=y

+1

to its axioms, we clearly obtain an extension of Q; and since the new axiom is a possible definition of S in Q*, Q proves to be indeed interpretable in Q*. Incidentally, it is easily seen that, conversely, Q* is interpretable in Q. On the other hand, it may be interesting to notice that Theorem 11 partly fails when applied to Q*. In fact, by removing from the axiom system of Q* we obtain a subtheory of Q* in which Q* proves to be interpretable and which therefore is essentially undecidable. If, however, we omit any of the remaining six axioms of Q*, then the resulting subtheory will no longer be essentially undecidable. Instead of replacing S by 1, we can eliminate S as well as 0 from the list of constants of N altogether. The system N+ of the arithmetic of natural numbers thus obtained contains and. as the only constants. N is interpretable in N+ since every valid sentence of N is derivable from the set of valid sentences of N+ supplemented, e.g., by the following two possible definitions of o and S:

e:

+

x

= Sy ~

x=O~x=x+x,

Vz(x

= y + z s z '# z + z « z = z· z).

OTHER ARITHMETICAL THEORmS AND VARIOUS THEORmS OF RINGS

67

Therefore Q is also interpetable in N+, by Theorems 8(i) and 7(ii) of lA, N+ and all its subtheories are undecidable, and there is a finitely axiomatizable subtheory Q+ of N+ which is essentially undecidable. It is easy actually to construct such a subtheory Q+ by appropriately transforming the axioms of Q. When we analyze the arguments in the preceding sections and specifically the proof of Theorem 6, we notice that all our results remain valid if, instead of Theory R, we use a weaker theory R'in which the axiom scheme Q 2 is replaced by a more special scheme:

Q;.

L1" •L1"

=L1". for every natural n.

Making use of this observation, we arrive at the conclusion that in all the theories discussed above the symbol. can be replaced by the unary operation symbol K denoting the operation of squaring a number. In particular we obtain a finitely axiomatizable and essentially undecidable theory Q' by replacing Axioms 6 and 7 by the following two axioms (which constitute together a recursive definition of K):

e

e; e;

K(Sx)

=

e

KO =0. Kx

+ S(x + x).

17

The fundamental results of II.5 can be extended to the arithmetic of positive integers and to that of arbitrary integers. As a simple example of a finitely axiomatizable and essentially undecidable subtheory of the arithmetic of positive integers we may mention a theory QX closely related to Q and Q*; it contains 1, and . as the only constants, and is based upon the following six axioms:

+,

QX

17

1

QX

17

2

•

,

QX 17 • 8 QX

17, •

QX , 0 QX

17 17

6

•

x

+ 1 =y + 1 -+ x =y. l¢y+1.

=y + 1). x + (y + 1) =(x + y) + 1. x·l =x. x ¢ 1-+ Vy(x

x.(y

+ 1) =(x.y) + x.

17 For the extension of these results to formalized theories of natural numbers with still other sets of constants see [20], pp. 112 f.

68

UNDECIDABILITY IN ARITHMETIC

As regards the arithmetic ot arbitrary integers, we shall consider two formalized systems of this arithmetic, J and J

ALFRED TARSKI Professor of Mathematics, University of California, Berkeley

IN COLLABORATION WITH

ANDRZEJ MOSTOWSKI Professor of Mathematics, University of Warsaw AND

RAPHAEL M ROBINSON Professor of Mathematics, University of California, Berkeley

I

1971

NORTH-HOLLAND PUBLISHING COMPANY AMSTERDAM

© North-Holland Publishing Company -1953 All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, without the prior permission of the copyright owner

ISBN: 0720422469

First edition 1953 Second printing 1968 Third printing 1971

PRINTBD IN TBB NKTIIBltLAM>S

PREFACE This monograph consists of three papers: A general method in proofs of undecidability, Undecidability and essential undecidability in arithmetic, Undecidability of the elementary theory of groups. While the first and the third papers have been written by the undersigned alone, the second paper is a joint work of A. Mostowski, R. M. Robinson, and the undersigned. The three papers are referred to throughout the monograph by Roman numerals I, II, III. The introduction to paper I is thought of as an introduction to the whole work and gives an idea of the scope of the problems discussed and the results obtained. The notations and symbolic conventions introduced at any place of the monograph are applied in the whole subsequent discussion. The work contains results obtained over a long period of time, 1938-1952. The first draft of paper II was prepared by Mostowski in 1949 and contained exclusively results found by him and Tarski in the pre-war period. However, its publication was postponed and the text was considerably modified, so as to embody the results and simplifications subsequently obtained by Robinson and partly by Tarski; the final draft was written jointly by these two authors in 1951-52. The other two papers were prepared for publication during the same period. It was originally planned to publish the work as a series of connected papers in some regular mathematical periodical. However, the Editors of the series Studies in Logic found the material integrated enough to appear as a separate volume in this series; their offer was accepted with appreciation. A large part of the technical work on the monograph was done during the period when Tarski and, for a shorter time, also Robinson were engaged in a research project in the foundations of mathematics sponsored by the Office of Ordnance Research, U.S. Army. VIII

l'REFAOE

IX

In this work the authors received much valuable advice and assistance from several friends-Mr. C. C. Chang, Professor Leon Henkin, Professor J. C. C. McKinsey, Dr. Julia Robinson, and Mr. R. L. Vaught. Professor E. W. Beth and Mr. F. W. J. Marx have greatly obliged the authors with their help in reading proofs. University of California Berkeley, April 1953

Allred Tarski

I

A GENERAL METHOD IN PROOFS OF UNDECIDABILITY BY

ALFRED TARSKI

I A GENERAL METHOD IN PROOFS OF UNDECIDABILITY 1.1. Introduction. 1 By a decision procedure for a given formalized theory T we understand a method which permits us to decide in each particular case whether a given sentence formulated in the symbolism of T can be proved by means of the devices available in T (or, more generally, can be recognized as valid in T).2 The decision problem for T is the problem of determining whether a decision procedure for T exists (and possibly of exhibiting such a procedure). A theory T is called decidable or undecidable according as the solution of the decision problem is positive or negative. As is well known, the decision problem is one of the central problems of contemporary metamathematics. Since only few theories turn out to be decidable 3, most endeavors are directed toward a negative solution. In the attacks on the decision problem for various special theories and, more specifically, in the attempts at obtaining a negative solution of this problem two different methods are applied. The first, direct method is essentially based upon ideas which originated with Godel and were used for the first time in [7], in the proofs of his well-known incompleteness theorems. When applying this method we have to make use of some deep properties of notions which are involved in the precise statement of the decision problem; in fact, of the notions of general recursive functions and sets. The method is rather involved and can be applied only to those theories in which a sufficiently strong number-theoretical apparatus 1 The observations contained in this paper were made in 1938-1939; they were presented by the author to a meeting of the Association for Symbolic Logic in 1948, and were summarized in [34]. (The numbers in square brackets refer to Bibliography.) 2 The meanings of various terms used in this introduction will be explained in a more detailed and precise way in later sections of the paper. S Some decidable theories are discussed or at least mentioned in [30].

3

4

A GENERAL METHOD IN PROOFS OF UNDECIDABILITY

can be developed. With the help of this method some fundamental results concerning the decision problem have been obtained. In fact, it was shown by Church in [3] that the solution of the decision problem for Peano's arithmetic (and some fragments of it) is negative, and Rosser proved later in [24] that the same applies to every consistent theory which is an extension of Peano's arithmetic. We express these results briefly by saying that Peano's arithmetic is, not only undecidable, but also essentially undecidable. - The theoretical foundations of the direct method will be outlined in II, in particular in II.2 and 11.4. The second, indirect method consists in reducing the decision problem for a theory T1 to the decision problem for some other theory T2 for which the problem has previously been solved. In the original form of this method, to establish the undecidability of a theory T1 one tried to show that either (i) T1 can be obtained from some undecidable theory T2 by deleting finitely many axioms from the axiom system of T2 (but without removing any constant from the symbolism of T2 ) , or else that (ii) some essentially undecidable theory T2 is interpretable in T1 • By applying the procedure (i) and by taking a fragment of Peano's arithmetic (suitably modified) for T2 , Church proved in [2] that the first-order predicate logic is undecidable. When applying the procedure (ii), Peano's arithmetic is usually taken for T2 ; in this way, e.g., various axiomatic systems of set theory have turned out to be undecidable. The indirect method in its original form was rather restricted in applications. Only in exceptional cases can a theory for which the decision problem is discussed be obtained from another theory, which is known to be undecidable, simply by omitting finitely many sentences from the axiom system of the latter. On the other hand, one could hardly expect to find an interpretation of Peano's arithmetic in various simple formalized theories, with meager mathematical contents, for which the decision problem was open. With regard to theories of this kind both the direct and the indirect methods seemed to fail. However, it has proved to be possible to extend and modify the indirect method (by combining some features of the two procedures indicated above) so as to widen

TltEORlES

wrra

STANDARD FORMALIZATION

5

considerably its range of application. In fact, a simple argument shows that, in order to establish the undecidability of a theory TI , it suffices to show that some essentially undecidable theory T2 can be interpreted, not necessarily in TI , but (what is much easier) in some consistent extension of TI - provided only that T2 is based upon a finite axiom system. As a consequence of this last condition, Peano's arithmetic cannot any longer be used as T2 since it is not based upon a finite axiom system. On the other hand, examples of essentially undecidable theories which are based upon finite axiom systems and are readily interpretable in other theories have been found (by the direct method) among fragments of Peano's arithmetic. Using this fact and applying the extended indirect method, many formalized theories-like the elementary theories of groups, rings, fields, and lattices-have recently been shown to be undecidable. The aim of the present paper is to set up theoretical foundations for the general method just described. The paper is conceived as a framework for later publications in this field, with the idea of permitting the authors to avoid the repetition of some lengthy, though elementary, discussions. With this in view, we give in 1.2 a rather detailed description of formalized theories to which the method applies. In 1.3-5 we define the notions involved in the description of the method, and we explicitly state and prove a few elementary theorems upon which this method is based. Finally, 1.6 contains a short survey of the results obtained so far with the help of the method discussed. Throughout the whole paper the discussion has an informal character. 1.2. Theories with standard formalization. The theories discussed in this paper will be referred to as theories with standard formalization. They can be briefly characterized as theories which are formalized within the first-order predicate logic (with identity, without variable predicates). ' & For a. detailed discussion of predicate logic and theories formalized within this logic consult, e.g., [9]; the discussion is spread over vol. 1. of this work, and is summarized and supplemented in vol. 2, pp. 375-391. The reader

6

A GENERAL :ME THOD IN PROOFS OF UNDECIDABILITY

The symbols which occur in expressions of a given theory Tare divided into variables and constants. The set of variables is assumed to be denumerable and hence infinite; the set of constants is either finite or denumerable. All the variables are treated as ranging over the same set of elements. The constants are divided into logical and non-logical ones. The logical constants are the sentential connectives-the negation sign ,..." the implication sign ~, the equivalence sign ~, the disjunction sign v, and the conjunction sign 1\; the quantifiers-the universal quantifier A and the existential quantifier V; and, finally, the identity symbol =. 5 The non-logical constants are the predicates (or relation symbols), the operation symbols, and the individual constants. With every predicate and every operation symbol a positive integer is correlated which is called the rank of the symbol. Thus, we may have in T unary predicates and operation symbols (i.e., symbols of rank 1), binary predicates and operation symbols (symbols of rank 2), etc. The identity symbol, though regarded as a logical constant, is included in the set of binary predicates. In practice, in addition to variables and constants, the so-called technical symbols, like parentheses and commas, are also used in constructing expressions; theoretically, however, these technical symbols can be dispensed with. Among expressions (i.e., finite concatenations of symbols) we will find there elaborations of certain points which have been disregarded in the present account (e.g., the definitions of free and bound variables, the specification of logical axioms and operations of inference, and the proofs of the deduction theorems). 5 In view of the general character of our present discussion we have not considered it necessary to maintain a strict distinction between expressions of formalized theories and their metamathematical designations. No inconsistency will arise, however, if we agree to regard all the symbolic expressions used in this paper, not as expressions of the theories discussed, but as metamathematical denotations of such expressions. In this case the letters "x", "y", "z", ... should be regarded as metamathematical variables which range over variables of the theory under discussion, and usually it should be assumed that two different letters represent two distinct variables of the theory.

THEORIES WITH STANDARD FORMALIZATION

7

distinguish terms and formulas. The simplest, so-called atomic, terms are the variables and the individual constants; a compound term is obtained by combining n simpler terms by means of an operation symbol of rank n. Similarly, an atomic formula is obtained by combining n arbitrary terms by means of a predicate of rank n; compound formulas are built from simpler ones by means of sentential connectives and quantifier expressions (i.e., quantifiers followed by variables, like Ax or Vy). An occurrence of a variable in a formula may be either free or bound; a formula in which no variable occurs free is called a sentence. Two further notions, those of logical derivability and logical validity, are involved in the metamathematical discussion of any theory T. They are usually introduced in the following way. First, we single out certain sentences of T which are referred to as logical axioms. Secondly, we describe certain (finitary) operations, the so-called operations of inference, which when performed on sentences yield new sentences. Usually the set of logical axioms is infinite while the set of operations of inference is finite. The most important operation of inference is that of detachment (modus ponens), which when applied to two sentences f/J and f/J --+ P yields the sentence P. In fact, it proves to be possible, by selecting a suitable set of logical axioms, to use the operation of detachment as the only operation of inference in formulating adequate definitions of derivability and logical validity. 6 A sentence is now said to be logically derivable or simply derivable from a set Aof sentences if it can be obtained from sentences of A and from logical axioms by performing operations of inference an arbitrary number of times. A sentence is called logically valid (or logically provable) if it is derivable from the set of logical axiomsor, what amounts to the same thing, from the empty set of sentences. Still another method of defining these two notions is available which, however, essentially involves the use of some semantieal 8 Cf. in this connection the remarks in [33], pp. 507 f. A suitable system of logical axioms is given in [19], pp. 80-85; the system is to be supple. mented by axioms involving the identity symbol.

8

A GENERAL :METHOD IN PROOFS OF UNDECIDABILITY

notions and, in particular, of the notion of satisfaction. 7 We assume that all the non-logical constants of T have been arranged in a (finite or infinite) sequence Co, ... , C.., ... ), without repeating terms. We consider systems ffi formed by a non-empty set U and by a sequence 00' .. . ,0.., ... ) of certain mathematical entities, with the same number of terms as the sequence of non-logical constants. The mathematical nature of each 0.. depends on the logical character of the corresponding constant C... Thus, if C.. is a unary predicate, then 0 .. is a subset of U; more generally, if C.. is an m-ary predicate, then 0.. is an m-ary relation the field of which is a subset of U. If C.. is an m-ary operation symbol, 0 .. is an m-ary operation (function of m arguments) defined over arbitrary ordered m-tuples Xl' ••• , xm ) of elements of U and assuming elements of U as values. If, finally, C.. is an individual constant, 0 .. is simply an element of U. Such a system (sequence) ffi = 11 which are logical axioms or members of A and for which the sentence (11)1 i\

••• i\

11>11) -+ lJF

is tautological. Hence, by including all tautological sentences in the set of logical axioms, we obtain a simple characterization of derivability in terms of logical axioms. It may be mentioned that Deduction Theorem II has sometimes been used as a definition of the notion of derivability. See [33], p. 507, and [38], pp. 10-11, footnote 10 (where references to earlier papers of Ajdukiewicz can also be found). 9 We can of course transform this definition by means of Deduction Theorem II so as to obtain a characterization of valid sentences in terms of logically valid sentences and non-logical axioms. Under conditions stated in footnote 8 a further simplification is possible; we obtain a simple characterization of valid sentences in terms of arbitrary (logical and non-logical) axioms, a characterization in which the notions of derivability and logical validity do not appear and no repeated application of operations of inference is involved.

THEORIES WITH STANDARD FORMALIZATION

II

the notion of validity has been defined in one way or another. We also assume that under this definition every sentence which is logically derivable from a set of valid sentences is itself valid, and that consequently every logically valid sentence is valid; this is the only condition imposed upon the definition of validity. A possible realization in which all valid sentences of a theory Tare satisfied is called a model of T. In all the theories with standard formalization the same symbols are assumed to be used as variables and logical constants; apart from differences in non-logical constants, the same expressions are regarded as formulas, sentences, logical axioms, and logically valid sentences. However, the notions of validity in these theories may of course exhibit essential differences. A theory is uniquely determined by the set of all its valid sentences; two theories are regarded as identical if their sets of valid sentences coincide. An axiomatic theory is uniquely determined by its non-logical constants and non-logical axioms. A theory T1 is called a subtheory of a theory T2 if every sentence which is valid in T1 is also valid in T2 ; under the same conditions T2 is referred to as an extension of T1 • An extension T2 of T1 is called inessential if every constant of T2 which does not occur in T1 is an individual constant and if every valid sentence of T2 is derivable in T2 from a set of valid sentences of T1 • (By saying that a sentence (/J is derivable in a theory T from a set A we stress the fact that, in deriving (/J, we may use both sentences of A and logical axioms of T. It is easily seen that, whenever (/J is derivable from A in some theory T, it is also derivable from A in every theory T' which contains all the non-logical constants occurring in (/J and in sentences of A.) If T1 is axiomatic, then an inessential extension of T1 ' is obtained by adding some new individual constants, but without adding any new non-logical axioms. An extension T2 of T1 is referred to as a finite extension if there is a finite set A of valid sentences of T2 such that every valid sentence of T2 is derivable from a set of sentences which are valid in T10r belong to A. Clearly every inessential extension is a finite extension.

12

A GENERAL METHOD IN PROOFS OF UNDECIDABILITY

Among the extensions common to two given theories T1 and T2 there is always a smallest one, which is a subtheory of any other common extension; this smallest common extension is referred to as the union of the given theories. The union T of T1 and T2 is fully characterized by the following two conditions: (i) the set of all non-logical constants of T is the (set-theoretical) union of the sets of all non-logical constants of T1 and T2 ; (ii) a sentence is valid in T if and only if it is derivable in T from a set of sentences which are valid in T10r T2' (Notice that condition (i) unambiguously determines the notions of a sentence of T and of a logical axiom of T, and hence also the notion of derivability in T.) If the theories T1 and T2 are axiomatic, we can construct T by postulating, in addition to (i), the analogous condition for the set of non-logical axioms. A theory T is called consistent if not every sentence of T is valid in T; or, in an equivalent formulation, if there is no sentence (/J such that both ' is the formula obtained from 4"> by replacing x everywhere by u. Intuitively speaking, sentence (iii) expresses the fact that, in every possible realization of T, for any two elements represented by y and z there is just one element represented by x such that the three elements satisfy the formula 4">. We proceed analogously with predicates and operation symbols of arbitrary ranks; individual constants are treated in this context as operation symbols of rank O. Let now T1 and Tll be any two theories. First assume that T1 and T. have no non-logical constants in common. In this case we say that Tll is interpretable in T1 if we can extend T1, by including in

INTERPRETABILITY AND WEAK INTERPRETABILITY

21

the set of valid sentences some possible definitions of the nonlogical constants of T2' in such a way that the resulting extension of T1 turns out to be an extension of T2 as well. Speaking precisely, T2 is interpretable in T1 if and only if there is a theory T and a set D satisfying the following conditions: (l¥) T is a common extension of T1 and T2 , and every constant of T is a constant of T1 or T2 ; (fJ) D is a recursive set of sentences which are valid in T and which are possible definitions in T1 of non-logical constants of T2 ; (y) each non-logical constant of T2 occurs in just one sentence of D; (15) every valid sentence of T is derivable (in T) from a set of sentences each of which is valid in T1 or belongs to D. In case T2 has only finitely many non-logical constants, the term "recursive" in condition (fJ) can clearly be omitted. In the general case, when T1 and T2 may have some common non-logical constants, we first replace the non-logical constants in T2 by new constants not occurring in T1 (different symbols by different symbols), without changing the structure of T2 in any other respect; if the resulting theory T~ proves to be interpretable in T1 , we say that T2 is also interpretable in T1 • It is easily seen that, in order to prove the interpretability of T2 in T1 in the general case, it is sufficient, though not necessary, to construct a theory T and a set D satisfying conditions (l¥)- (15), with the difference that D is assumed to contain possible definitions of only those constants of T2 which do not occur in T1 • A theory T2 is said to be weakly interpretable in T1 if T2 is interpretable in some consistent extension of T1 which has the same constants as T1 • In order to prove the weak interpretability of T2 in T1 it suffices to show that T1 and T2 have a common consistent extension T satisfying the following condition: (e) there is a recursive set D of valid sentences of T such that, for every non-logical constant C of T2 which does not occur in T1 , some possible definition of C in T1 belongs to D. In case T2 has only finitely many nonlogical constants, condition (e) can clearly be replaced by a simpler one: (C) for every non-logical constant C of T2 which does not occur in T1 , some possible definition of C in T1 is valid in T. Moreover, in both (e) and (C) the notion of a possible definition as applied

22

A GENERAL METHOD IN PROOFS OF UNDECIDABILITY

to operation symbols and individual constants can be understood in a wider sense than the original one. If, e.g., the operation symbol does not occur in T1 , then as a possible definition of in T1 we can regard any sentence of the form

+

+

Ax Ay Az(x

= y +z ~

... , lJIl I are valid in T. It is known that, whenever a sentence e is logically valid, then the sentence e* is also logically valid. (This can be proved, e.g., by showing that, if e holds in every possible realization, the same applies to e*.) e* clearly coincides with ((lJIt»)* 1\ ••• 1\ (lJI~P»)*] -+ if>*. From the definitions of lJICP) and lJI* it is easily seen that every sentence lJIof T is logically equivalent to (lJIcP»)*. Hence the sentences (lJIt»)*, ... , (lJI;;»)* are valid in T, and therefore if>* is valid in T. Consequently every sentence of A' is valid in T. If if> is any sentence valid in T, then if>CP) is valid in TCP) and therefore derivable from B'. By again applying Deduction Theorem II, we show that (if>{P»)*, and hence also if>, is derivable from A'. We conclude that T is axiomatizable. We have thus established conclusion (i) of our theorem. Moreover we notice that, whenever the set A (in the argument outlined above) is finite, the set B is also finite, provided there are only finitely many individual constants and operation symbols occurring in T. Furthermore, if B' is finite, then A' is always finite. Consequently, conclusion (ii) also holds, and the proof is complete. If T is an axiomatic theory, then, by Theorem 9, T{P) can also be represented as an axiomatic theory; from the proof of this theorem we learn how to construct an adequate axiom system for T{P) if the axiom system of T is known. Supplementing Theorem 9 (ii), we can show that, whenever T{P) is finitely axiomatizable and consistent, T can contain only finitely many individual constants and operation symbols. 10. Let T be any theory and P a unary predicate which is not a constant of T. Then T(P) is essentially undecidable if and only if T is essentially undecidable. PROOF: We want first to show that THEOREM

(1)

TCP) is interpretable in T.

For this purpose we construct an extension T* of T by adding P to the set of constants of T and by stipulating that a sentence is

28

A GENERAL METHOD IN PROOFS OF UNDECIDABILITY

valid in T* if and only if it is derivable from the set of all valid sentences of T supplemented by the following sentence e: Ax(Px ~ x

=x).

Obviously, e is a possible definition of PinT. C/J being any sentence of T, it is easily seen that C/J(IX) and 4>(IX,fJ), which would also be suitable for our purposes. In fact, we could agree to use 4>(IX) as an abbreviation for the expression I\u(u

=IX

~

4»

in case u does not occur in IX, and for the expression I\w[w

=IX

~ I\u(u

=w

~

4>)]

in case u occurs in IX while w is a variable which does not occur either in 4> or in IX. We may assume that all the variables have been arranged in an infinite sequence, and choose as w the first variable in this sequence occurring neither in 4> nor in IX. 4>(IX,fJ) is defined analogously. Under this alternative definition the sentence is logically valid for every formula 4> and every term IX. 4 II.2. Definability in arbitrary theories. Let T be any theory with standard formalization. Ii We assume that in this theory an infinite sequence of terms .110,.111> •.. , .11", • •• containing no variables is available. A subset P of N, i.e., a set P of natural numbers, is said to be definable in T (relatively to the sequence of terms .110' .11 1 , ••. ) if there is a formula 4> of T such that 4>(.11,,) is a sentence valid in T whenever n E P, and", 4>(.11,,) is a sentence valid in T whenever n EN and n ¢ P. 6 Such a formula 4> is said to define P. From the fact that 4>(.11,,) is a sentence we conclude (looking back at the , Compare [35]. fi Actually the discussion in this section applies to arbitrary formalized theories whose logical basis includes at least the first-order predicate logic with identity. With small changes, the discussion could even be extended to theories in which no quantifiers occur. S The notion of definability was first discussed in [36]. A more general notion, which comprehends both the notion of definability in the sense of [36] and that of general recursiveness as particular cases, was introduced in [15], pp. 72 ff.; it is still not as general as the notion discussed in this pa.per.

DEFINABILITY IN ARBITRARY THEORIES

45

definition of lP(lX) in II. 1) that the variable u does not occur bound in lP and that no variable different from u occurs free in lP. A function F is said to be definable in T if there is a formula lP such that (i) for any n,p E N with Fn = p, lP(LI", Ll Il ) is a sentence valid in T; (ii) for any n,p EN with Fn =1= p, --lP(LI",Ll Il ) is a sentence valid in T; (iii) for every n EN, AuAv[lP(LI",u) 1\ lP(LI", v) -+ u

=v]

is a sentence valid in T. As before, such a formula lP is said to define F; the variables u and v do not occur bound in lP, and no variable different from u and v occurs free in lP. It is easily seen that conditions (i) and (iii) can be replaced by the following condition: (i') for every n E N, the sentence

=

Av[lP(LI",v) B-V LI.II'''] is valid in T. Under the assumption that, for all n,p E N with n =1= p, the sentences LI" '# Ll Il are valid in T, condition (ii) proves to be a consequence of (i) and (iii), and hence can be dispensed with. Even without this assumption, however, the omission of condition (ii) would not invalidate the proof of the essential results of this section, i.e., Theorem 1 and Corollary 2. By modifying in an obvious way the definition of definable sets, we arrive at the notion of definable binary relations (sets of ordered couples) and, more generally, of definable n-ary relations. Since functions (with one argument) can be treated as special binary relations, we obtain in this way a new definition of definable functions. The new definition is weaker than the original one and differs from the latter by the absence of condition (iii). It suffices for the proof of Theorem 3 and Corollary 4 outlined below; however, condition (iii) will be essentially involved in the proof of Theorem 1 and Corollary 2. When using the term "definable

46

UNDECIDABILITY IN ARITHMETIC

functions" in the present article, we shall have in mind the original definition of this term. On the other hand, the definability of sets reduces under certain conditions to the definability of functions. In fact, let P be any set of natural numbers, and let Op be its characteristic function defined by the formulas: Opn = 0 if n Opn = 1 if n

E E

P, Nand n ¢ P.

The definability ofOp (even in the weaker sense) implies that of P; in fact, if a formula C/J defines Op, then the formula C/J(u,L1 o) is easily seen to define P. Under the assumption that, for all n,p EN with n =1= p, the sentences ,1,. #- L1 p are valid in T, it turns out that, conversely, the definability of P implies that of Op; for, if a formula lJ' defines P, then the formula [lJ'(u)

1\

v

=

,10] v [,...., lJ'(u)

1\

V

=

,11]

proves to define 0 rConsider now any fixed one-to-one correspondence between natural numbers and expressions of T. For the time being we make no other assumptions regarding the nature of this correspondence. Given a natural number n, the correlated expression of T will be denoted by En; conversely, given an expression C/J, we shall denote the correlated natural number by Nr(C/J). Let D (the diagonal function) be the function defined by the formula Dn = N r(E,.(L1,.)) ; thus Dn is the number correlated with the expression which is obtained from E,. by replacing everywhere the variable u by the term ,1,.. Let V be the set of all natural numbers n such that E,. is a sentence valid in T. Using this notation we can establish the following result: THEOREM 1. If the theory T i~ consistent, then the function D and the set V are not both defiruJ1Jle in T.

DEFINABILITY IN ARBITRARY THEORIES

47

Assume, to the contrary, that both D and V are definable in T. Thus there are formulas (/) and 1JI satisfying, for every natural number n, the following conditions: PROOF:

(1) the sentence Av[(/)(L1",v)

B-

v

is valid in T; (2) (3)

if n if n

E

=L1

Dn ]

V, then the sentence 1JI(L1,,) is valid in T;

f# V, then the sentence '" 1JI(L1,,) is valid in T.

The variable v does not occur free in 1JI. Without any loss of generally we can assume that v does not occur in 1JI at all and that, consequently, 1JI(v) is a regular substitution of 1JI.

Let m = Nr(Av[(/)(u,v)

-i>- ' "

1JI(v)])

so that and consequently, (4)

Em(L1m ) = Av[(/)(L1 m,v)

-i>- ' "

1JI(v)].

If the sentence Em(L1 m ) is valid in T, then, by (1) (with n = m) and (4), the sentence", 1JI(L1.vm) is valid. If E m(L1m ) is not valid, then Nr(Em(L1m }} f# V; and since, by the definition of the function D, (5)

Dm

=

Nr(Em(L1m }} ,

we conclude by (3) that in this case '" 1JI(L1.vm) is valid as well. Thus, (6)

the sentence '" 1JI(L1.vm) is valid in T.

By (1) and (6), the sentence Av[(/)(L1 m,v) -i>- ' " 1JI(v)] is valid. Consequently, by (4) and (5), Dm E V; and therefore, by (2), (7)

the sentence 1JI(L1 Dm ) is valid in T.

(6) and (7) imply that the theory T is inconsistent, contrary to the hypothesis of the theorem. This completes the proof. Theorem 1 and its proof represent a metamathematical reconstruction and generalization of arguments involved in various

48

UNDECIDABILITY IN ABITHMETIC

semantical antinomies and, in particular; in the antinomy of the liar. The idea of this reconstruction and the realization of its farreaching implications is due to Godel [7]. The present version of this reconstruction is distinguished by its generality and simplicity. It applies to arbitrary formalized theories, and not only to those in which a comprehensive fragment of the arithmetic of natural numbers can be developed; to a large extent it is independent of the way in which the notion of validity has been defined for a given theory, and in particular it does not involve the notion of a formal proof within this theory; it does not use the apparatus of recursive functions-although this apparatus will play a fundamental role in applications of Theorem 1 to the decision problem. From now on we assume that the correspondence between natural numbers and expressions of T satisfies the usual conditions concerning the recursiveness of certain functions and sets under this correspondence (cf. 1.3). The most fundamental of these conditions can be expressed as follows: Let (/> ~ lJI denote the concatenation of two arbitrary expressions (/> and lJI; then the function G on NX N to N defined by the formula G(n,p) = Nr(En ~ E p )

is recursive. We also assume that the terms Ll o, Ll 1 , • •• form a recursive sequence, i.e., that the function H (on N to N) defined by the formula Hn

=

Nr(Ll n )

is recursive. From these assumptions it follows, in particular, that the function D is also recursive. 7 7 Actually the recursiveness of D is the only assumption involved in our further discussion. A proof of this assumption can be obtained, e.g., from the discussion in [15], pp. 73 ff. It may be interesting to notice that, under the alternative definition of IP(lX) mentioned at the end of 11.1, the function D is defined by the formula

Dn = Nr[Au(u = LIn -+ En)]' Using this formula and applying some elementary properties of recursive functions we can immediately derive the recursiveness of D from that of the functions G and H defined above in the text.

DEFINABILITY IN ABBITRARY THEORIES

49

There are close connections between the notions of recursiveness and definability. In 11.4 we shall see that many theories Tare known such that all the functions and sets which are recursive are definable in T (and conversely). The importance of this fact for our discussion is seen from the following corollary which is an easy consequence of Theorem 1: COROLLARY 2. 1fT is a consistent theory in which all recursive functions are definable, then T is essentially undecidable. PROOF: As has been stated above, the function D is recursive and therefore, by the hypothesis of the corollary, it is definable in T. Hence, by Theorem 1, the set V is not definable in T. If the set V were recursive, its characteristic function Oy (Oyn = 0 for n E V, Oyn = 1 for n ¢ V) would also be recursive and hence definable in T; as was mentioned before, this would imply the definability of the set V itself. Thus, the set V is not recursive; in other words, the set of all sentences which are valid in T is not recursive, and the theory T is undecidable. Since the definability in T implies the definability in every extension of T (relatively to the same sequence of terms Ll o, Ll1 , .•• ), our argument shows that every consistent extension of T is undecidable and that, consequently, T is essentially undecidable.

To conclude this section, we give two results, Theorem 3 and Corollary 4, concerning those theories in which every, or not every, definable function is recursive. These results follow in a simple way from the definitions of the notions involved and are less deep than Theorem 1 and Corollary 2; also they will find no essential applications in our further discussion. Nevertheless it will be interesting to compare the content of these results with that of Corollary 2. Every function which is definable in a con8i8tent and axiomatizable theory T is recursive. The same applies to every set 0/ natural numbers. PROOF: Using the intuitive notion of recursiveness, we argue as follows. Since T is axiomatizable, there is a recursive set A of THEOREM

3.

50

UNDECIDABILITY IN ARITHMETIC

sentences of T such that a sentence is valid in T if and only if it is derivable from A. Hence, as is well known, we can arrange all valid sentences of T in a recursive infinite sequence lJ'o, lJ'1' .. . . (In other words, there is a recursive function the range of which is the set V.) Let now F be any function definable in T, and let rp be a formula defining F. Then, for any given n EN, the sentence rp(Lt", Lt.Fto) is valid in T, and therefore it must occur somewhere in the sequence lJ'o, lJ'I> .... Hence, by checking the successive terms of this sequence, we must arrive in finitely many steps at a term lJ'm which is of the form rp(Lt.., Lt,,) for some pEN. If the numbers p and Fn were different, then the sentence rp(Lt.., Lt,,) would be valid in T, and therefore T would be inconsistent; hence p = Fn. We thus have a procedure which enables us, for any given n EN, to determine the function value Fn in finitely many steps; in other words, F is recursive. If, instead of a function, we consider a set of natural numbers, the proof remains practically unchanged. 1'.1

The converse of Theorem 3 is by no means true. In fact, as is seen from Corollary 2, if T is any consistent theory which is decidable or, more generally, which is not essentially undecidable, then there are recursive functions which are not definable in T. COROLLARY 4. 1fT is a consistent theory such that there is a function (or set of natural numbers) which is definable in T but not recursive, then T is essentially undecidable. PROOF: By Theorem 3, the theory T is not axiomatizable and hence a fortiori undecidable (cf. 1.3). For the same reason every consistent extension of T is undecidable, and hence T is essentially undecidable.

Corollaries 2 and 4 seem to provide us with two widely different methods of constructing essentially undecidable theories. Actually, however, we know no single case in which Corollary 4 would be of real help in a proof of essential undecidability. This is explained partly by the rather trivial character of Corollary 4 and partly by the fact that a theory T which satisfies the hypothesis of this corollary must have a much stronger property than essential

FORMALIZED ARITlIMETIO OF NATURAL NUMBERS

51

undecidability; in fact, by analyzing the proof of Theorem 3, we conclude that, in every consistent extension of T, the set of all valid sentences not only is not recursive, but cannot even be recursively enumerable. 8

II.3. Formalized arithmetic of natural numbers and its subtheories. The formalized arithmetic of natural numbers in which we are interested here will be referred to as Theory N. The set of all constants of N is assumed to consist of four symbols: an individual constant 0, a unary operation symbol S, and two binary operation symbols, + and -, The same assumption applies to all other theories discussed in this and the following two sections. Hence every possible realization of each of these theories is a system 9i = in which U is an arbitrary set, c is an element of U, F is a unary operation (function) on U to U, and + and· are two binary operations on UX U to U (cf. 1.2). To define the validity in N we consider a special realization 91 = N, 0, S, +, . >where all the symbols N, 0, S, ... have their usual arithmetical meaning (cf. 11.1); a sentence is said to be valid in N if and only if it holds in 91. We shall be interested in some axiomatic subtheories of N referred to as Theories P, Q, and R. The axiom system of Q consists of the following seven sentences :

for

all n,p

E

N

are shown to be valid in Q by induction on p. (All the inductive arguments in this proof are of course metamathematical inductions, and not inductions within Theory Q.) In fact, for p = 0, (1) is a particular instance of 8 4 ; when passing from p to p + 1, we apply 8 0 , Similarly, by an induction on p based upon 8 8 , 8 7 , and (1), we derive the sentences D2 : (2)

L1,. •L1"

=L1,.." for all n,p EN.

Consider now the sentences Da: (3)

L1,. #;. L1" for all n,p

We can clearly assume that n

E

< p.

N with n =1= p. For n

=

0 (and every p

> n)

54

UNDECIDABILITY IN ARITHMETIC

(3) is a particular case of e2 ; when passing from n to n apply e1 • e s and e5 imply z x 0 II X '# 0 ~ Vy[x Sy II S(z y) 0];

=

+

hence, by

e

=

+

+1

we

=

2,

(4)

Similarly, for every natural number n we have byes and z

+ x =LI"+1

hence, by (5)

'# 0 ~

II X

=Sy

Vy[x

e1> x

where N, 1, 0,8, + have their usual meaning while 8', $, and· are defined by the conditions

OTHER ARITHMETICAL THEORIES AND VARIOUS THEORIES OF RINGS

65

B'O = 0, B'n = Bn if n =I=- 0; nEf)p= 0 if n= 0 or p= 0; nEf) p= n +p-l if n=l=-O and p =1=-0; n· p = 0 for all n,p EN.

.. .

Let T+ be a theory with three constants: 0, S, +; a sentence of T+ is valid in T+ if it is valid in T6 , i.e., is satisfied in the system (N, 0, B, + where all the symbols have their usual meaning. The following sentences are clearly valid in T6 :

>

= Sy (x. =0 .Y =0) v (x. =Sy. . Y #= 0),. x =y + z (x =0 Y =0) (x =0 . =. 0) . x

. 0= SO,.

~

~

A

A

A

v

A Z

.

(Sx

= y +Z

v

A

Y

#= 0 A

Z

.

#= 0),

Y·Z = O. With the help of these sentences, arguing as in the case of T4 , we reduce the decision problem for T6 to that for T+. The latter theory is known to be decidable (see [17]); hence T6 is also decidable. (Another, perhaps more natural, example of a decidable extension of Q 6 is provided by a theory T~ which has the same constants as Q. A sentence is valid in T~ if it is satisfied in the system (U, 0, B, +, . where (i) U is the set consisting of the natural numbers and an additional element 00; (ii) 0, B, + have their usual meaning when applied to natural numbers, and in addition Boo = x + 00 = 00 + x = 00 for every x E U; (iii) x . y = 00 for all x,y E U. The decision problem for T~ can again be reduced to that for T+; however, the reduction is somewhat more involved than in the case of T6 . ) A sentence is valid in T7 if it is satisfied in the system (N, 0, B, +, . > where n . p = 0 for all n,p EN. We see at once that the decision problem for the theory T7 again reduces to that for the theory T+ described above in connection with T6' (The symbols

>

...

0, S, + are now replaced in T+ by 0, S, and the proof is complete.

+.) Hence T7 is decidable,

II.6. Extension of the results to other arithmetical theories and to various theories of rings. The results obtained in the preceding

66

UNDECIDABILITY IN ARITHMETIC

sections and in particular in n.5 can be carried over to formalized systems of the arithmetic of natural numbers with different sets of constants. Consider, e.g., the theory N* which differs from N only in that the unary operation symbol S has been replaced by the individual constant 1 denoting the integer one; the definition of validity in N* is entirely analogous to that in N. We can construct axiomatic subtheories P*, Q*, R* of N* closely related to the subtheories of N described in n.3-simply by replacing in the axioms of P, Q, R all the terms of the form Sex. by ex. 1. It is easily seen that Theorem 9 and Corollary 10 remain valid if we replace in them N, P, Q, R by N*, P*, Q*, R*, respectively. To prove this, it suffices to show that each of the theories N, P, '" is interpretable in the corresponding theory N*, P*, ... in the sense of 1.4, and then to apply Theorems 7 and 8 of 1.4. For instance, if we add the symbol S to the constants of Theory Q* and the sentence

+

x

=Sy

~x

=y

+1

to its axioms, we clearly obtain an extension of Q; and since the new axiom is a possible definition of S in Q*, Q proves to be indeed interpretable in Q*. Incidentally, it is easily seen that, conversely, Q* is interpretable in Q. On the other hand, it may be interesting to notice that Theorem 11 partly fails when applied to Q*. In fact, by removing from the axiom system of Q* we obtain a subtheory of Q* in which Q* proves to be interpretable and which therefore is essentially undecidable. If, however, we omit any of the remaining six axioms of Q*, then the resulting subtheory will no longer be essentially undecidable. Instead of replacing S by 1, we can eliminate S as well as 0 from the list of constants of N altogether. The system N+ of the arithmetic of natural numbers thus obtained contains and. as the only constants. N is interpretable in N+ since every valid sentence of N is derivable from the set of valid sentences of N+ supplemented, e.g., by the following two possible definitions of o and S:

e:

+

x

= Sy ~

x=O~x=x+x,

Vz(x

= y + z s z '# z + z « z = z· z).

OTHER ARITHMETICAL THEORmS AND VARIOUS THEORmS OF RINGS

67

Therefore Q is also interpetable in N+, by Theorems 8(i) and 7(ii) of lA, N+ and all its subtheories are undecidable, and there is a finitely axiomatizable subtheory Q+ of N+ which is essentially undecidable. It is easy actually to construct such a subtheory Q+ by appropriately transforming the axioms of Q. When we analyze the arguments in the preceding sections and specifically the proof of Theorem 6, we notice that all our results remain valid if, instead of Theory R, we use a weaker theory R'in which the axiom scheme Q 2 is replaced by a more special scheme:

Q;.

L1" •L1"

=L1". for every natural n.

Making use of this observation, we arrive at the conclusion that in all the theories discussed above the symbol. can be replaced by the unary operation symbol K denoting the operation of squaring a number. In particular we obtain a finitely axiomatizable and essentially undecidable theory Q' by replacing Axioms 6 and 7 by the following two axioms (which constitute together a recursive definition of K):

e

e; e;

K(Sx)

=

e

KO =0. Kx

+ S(x + x).

17

The fundamental results of II.5 can be extended to the arithmetic of positive integers and to that of arbitrary integers. As a simple example of a finitely axiomatizable and essentially undecidable subtheory of the arithmetic of positive integers we may mention a theory QX closely related to Q and Q*; it contains 1, and . as the only constants, and is based upon the following six axioms:

+,

QX

17

1

QX

17

2

•

,

QX 17 • 8 QX

17, •

QX , 0 QX

17 17

6

•

x

+ 1 =y + 1 -+ x =y. l¢y+1.

=y + 1). x + (y + 1) =(x + y) + 1. x·l =x. x ¢ 1-+ Vy(x

x.(y

+ 1) =(x.y) + x.

17 For the extension of these results to formalized theories of natural numbers with still other sets of constants see [20], pp. 112 f.

68

UNDECIDABILITY IN ARITHMETIC

As regards the arithmetic ot arbitrary integers, we shall consider two formalized systems of this arithmetic, J and J

Our partners will collect data and use cookies for ad personalization and measurement. Learn how we and our ad partner Google, collect and use data. Agree & close