You are on page 1of 20

An Unsolvable Problem of Elementary Number Theory

Author(s): Alonzo Church


Source: American Journal of Mathematics, Vol. 58, No. 2 (Apr., 1936), pp. 345-363
Published by: The Johns Hopkins University Press
Stable URL: http://www.jstor.org/stable/2371045 .
Accessed: 24/11/2014 19:34

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.

The Johns Hopkins University Press is collaborating with JSTOR to digitize, preserve and extend access to
American Journal of Mathematics.

http://www.jstor.org

This content downloaded from 155.97.178.73 on Mon, 24 Nov 2014 19:34:23 PM


All use subject to JSTOR Terms and Conditions
AN UNSOLVABLE PROBLEM OF ELEMENTARY NUMBER
THEORY.1
By ALONZOCHURCH.

1. Introduction. There is a class of problemsof elementarynumber


theorywhichcan be statedin the formthatit is requiredto findan effectively
calculablefunction f of n positiveintegers,suchthatf (xl, X2p,22. . . Xn)
is a necessaryand sufficient conditionforthe truthof a certainpropositionof
elementarynumbertheoryinvolvingxl, x2, *, xn as freevariables.
An example of such a problemis the problemto find a means of de-
terminingof any givenpositiveintegern whetheror not thereexist positive
integersx, y,z, suchthatxn + yn = zn. For thismaybe interpreted, required
to findan effectively calculablefunctionf, such thatf(n) is equal to 2 if and
only if thereexist positiveintegersx, y, z, such that Xn + yn = zn. Clearly
the conditionthat the functionf be effectively calculableis an essentialpart
,ofthe problem,sincewithoutit the problembecomestrivial.
Anotherexampleof a problemof this class is, for instance,the problem
of topology,to finda completeset of effectively calculableinvariantsof closed
three-dimensional simplicialmanifoldsunderhomeomorphisms.This problem
can be interpretedas a problemof elementarynumbertheoryin view of the
fact that topologicalcomplexesare representableby matricesof incidence.
In fact,as is well known,the propertyof a set of incidencematricesthat it
representa closed three-dimensional manifold,and the propertyof two sets
of incidencemnatrices that theyrepresenthomeomorphic complexes,can both
be describedin purelynumber-theoretic terms. If we enumerate,in a straight-
forwardway, the sets of incidence matriceswhich representclosed three-
dimensionalmanifolds,it will thenbe immediatelyprovablethat the problem
underconsideration(to finda completeset of effectively calculableinvariants
of closed three-dimensional manifolds) is equivalentto the problem,to find
an effectively calculablefunctionf of positiveintegers,such that f(m, n) is
equal to 2 if and onlyif the m-thset of incidencematricesand the n-thset
of incidencematricesin the enumerationrepresenthomeomorphic complexes.
Otherexampleswill readilyoccurto the reader.

IPresented to the American Mathematical Society, April 19, 1935.


2
The selection of the particular positive integer 2 instead of some other is, of
course, accidental and non-essential.
345

This content downloaded from 155.97.178.73 on Mon, 24 Nov 2014 19:34:23 PM


All use subject to JSTOR Terms and Conditions
346 ALONZO CHURCH.

The purposeof the presentpaper is to proposea definitionof effective


calculability3 whichis thoughtto correspondsatisfactorily
to the somewhat
vagueintuitivenotionin termsof whichproblemsof this class are oftenstated,
and to show,by means of an example,that not everyproblemof this class
is solvable.

2. Conversionand X-definability.We select a particularlist of sym-


bols,consisting {, }, (,), A,[,], and an enumerably
ofthesymbols infinite
set of symbolsa, b, c, to be called variables. And we definethe word
formulato mean any finitesequenceof symbolsout of this list. The terms
well-formed formula,free variable,and bound variable are then definedby
inductionas follows. A variablex standingalone is a well-formed formula
and the occurrenceof x in it is an occurrenceof x as a freevariablein it;
if the formulasF and X are well-formed, {F} (X) is well-formed,and an
occurrenceof x as a free (bound) variablein F or X is an occurrenceof x
as a free (bound) variablein {F} (X); if the formulaM is well-formed and
containsan occurrence ofx as a freevariablein M, thenAx[M] is well-formed,
any occurrenceof x in Ax[M] is an occurrenceof x as a bound variablein
Ax[M], and an occurrenceof a variabley, otherthan x, as a free (bound)
variablein M is an occurrenceof y as a free (bound) variablein Ax[M].

s As will appear, this definition of effectivecalculability can be stated in either


of two equivalent forms, (1) that a function of positive integers shall be called
effectivelycalculable if it is X-definablein the sense of ? 2 below, (2) that a function
of positive integers shall be called effectivelycalculable if it is recursive in the sense
of ? 4 below. The notion of X-definabilityis due jointly to the present author and
S. C. Kleene, successive steps towards it having been taken by the present author in
the Annals of Mathematics, vol. 34 (1933), p. 863, and by Kleene in the American
Journal of Mathematics, vol. 57 (1935), p. 219. The notion of recursiveness in the
sense of ? 4 below is due jointly to Jacques Herbrand and Kurt Godel, as is there
explained. And the proof of equivalence of the two notions is due chiefly to Kleene,
but also partly to the present author and to J. B. Rosser, as explained below. The
proposal to identify these notions with the intuitive notion of effectivecalculability
is firstmade in the present paper (but see the firstfootnote to ? 7 below).
With the aid of the methods of Kleene (American Journal of Mathematics, 1935),
the considerations of the present paper could, with comparatively slight modification,
be carried through entirely in terms of X-definability, without making use of the notion
of recursiveness. On the other hand, since the results of the present paper were
obtained, it has been shown by Kleene (see his forthcomingpaper, "General recursive
functions of natural numbers") that analogous results can be obtained entirely in
terms of recursiveness, without making use of X-defilnability.The fact, however, that
two such widely differentand (in the opinion of the author) equally natural definitions
of effectivecalculability turn out to be equivalent adds to the strength of the reasons
adduced below for believing that they constitute as general a characterization of this
notion as is consistent with the usual intuitive understanding of it.

This content downloaded from 155.97.178.73 on Mon, 24 Nov 2014 19:34:23 PM


All use subject to JSTOR Terms and Conditions
AN UNSOLVABLE PROBLEM OF NUMBER THEORY. 347

We shall use heavy type lettersto stand for variable or undetermined


formulas. And we adopt the conventionthat, unless otherwisestated,each
heavytypelettershall representa well-formed formulaand each set of symbols
standing apart which contains a heavy type letter shall representa well-
formedformula.
When writingparticularwell-formed formulas,we adopt the following
abbreviations.A formula{F} (X) maybe abbreviatedas F (X) in anycasewhere
F is or is represented by a singlesymbol. A formula{{F} (X) }(Y) may be
abbreviatedas {F} (X, Y), or,if F is or is represented by a single symbol,as
F(X, Y). And {{{F} (X) } (Y) } (Z) maybe abbreviatedas {F} (X, Y, Z), or
as F(X, Y,Z), and so on. A formulaAX1[XX2[. * x,n[M] ]] may be
abbreviatedas Xx1x2 xn* M or as Ax1x2 * xnM
We also allow ourselvesat any time to introduceabbreviationsof the
form that a particularsymbolac shall stand for a particular sequence of
symbolsA, and indicatethe introduction of such an abbreviationby the nota-
tion a --> A, to be read,"'o standsforA."
We introduceat once the followinginfinitelist of abbreviations,

1 ->ab a(b),
2 ->Aab a(a(b)),
3 ->Aab a(a(a(b)))

and so on, each positiveintegerin Arabic notationstandingfor a formula


of the formXab*a (a(. *a (b) . . . ).
The expressionSxMI is used to stand for the result of substitutingN
forx throughout M.
We considerthe threefollowingoperationson well-formed formulas:

I. To replaceany part Xx[M] of a formulaby Xy[SxMi], wherey is


a variablewhichdoes not occurin M.
II. To replaceany part {Xx[M]} (N) of a formulaby S>lI I,provided
that the bound variablesin M are distinctboth fromx and fromthe free
variablesin N.
III. To replace any part S^M i (not immediatelyfollowingA) of a
formulaby {Akx [M] } (N), providedthatthe bouandvariablesin M are distinct
bothfromx and fromthefreevariablesin N.

Any finitesequenceof these operationsis called a conversion,and if B


is obtainablefromA by a conversionwe say that A is convertible into B, or,
"A convB." If B is identicalwith A or is obtainablefromA by a single

This content downloaded from 155.97.178.73 on Mon, 24 Nov 2014 19:34:23 PM


All use subject to JSTOR Terms and Conditions
348 ALONZO CHURCH.

applicationof one of the operationsI, II, 1II, we say that A is immediately


convertible into B.
A conversionwhichcontainsexactlyone applicationof OperationII, and
no applicationof OperationIII, is called a reduction.
A formulais said to be in normalforn if it is well-formed and contains
no part of the form{Ax[M] }(N). And B is said to be a normalformof A
if B is in normalformand A convB.
The originallygiven order a, b, c, of the variablesis called their
naturalorder. And a formulais said to be in principalnormalformif it is
in normalform,and no variableoccursin it both as a freevariableand as a
bound variable, and the variables which occur in it immediatelyfollowing
the symbolA are,whentakenin the orderin whichtheyoccurin the formula,
in naturalorderwithoutrepetitions, beginningwitha and omittingonlysuch
variablesas occur in the formulaas freevariables.4 The formulaB is said
to be the principalnormalformof A if B is in principalnormalformand
A convB.
Of the threefollowingtheorems,proofof the firstis immediate,and the
secondand thirdhave been provedby the presentauthorand J. B. Rosser:
THEOREMI. If a formulais in normal form,no reductionof it is
possible.
THEOREMII. If a formulahas a normal form,this normal form is
unique to withinapplicationsof OperationI, and any sequenceof reductions
of the formulamust (if continued)terminatein the normalform.
THEOREMIII. If a foirmula has a normalform,everywell-formed part
of it has a normalform.
We shall call a functiona functionof positiveintegersif the range of
each independentvariable is the class of positiveintegersand the range of
the dependentvariable is containedin the class of positiveintegers. And
when it is desiredto indicatethe numberof independentvariableswe shall
speak of a functionof one positiveinteger,a functionof twopositiveintegers,
and so on. Thus if F is a functionof n positiveintegers,and a,, a2,*, an
are positiveintegers,thenF(al, a2,. . ., an) mustbe a positiveinteger.

4For example, the formulas Xab . b (a) and Xa . a (Xc . b (c) ) are in principal normal
form,and Xac . c(a), and Xbc. c(b), and Xa . a (Xa . b (a) ) are in normal form but not
in principal normal form. Use of the principal normal form was suggested by S. C.
Kleene as a means of avoiding the ambiguity of determination of the normal form
of a formula, which is troublesome in certain connections.
Observe that the formulas 1, 2, 3, . ..are all in principal normal form.
B Alonzo Church and J. B. Rosser, "Some properties of conversion,'?forthcoming
(abstract in Bulletin of the American Mathematical Society, vol. 41, p. 332).

This content downloaded from 155.97.178.73 on Mon, 24 Nov 2014 19:34:23 PM


All use subject to JSTOR Terms and Conditions
AN UNSOLVABLE PROBLEM OF NUMBER THEORY. 349

if it is
A function F of one positive integer is said to be A-definable
possible to find a formula F such that, if F (n) = r and m anid r are the
formulasforwhichthepositiveintegersm and r (writtenin Arabicnotation)
stand accordingto our abbreviationsintroducedabove,then {F}(m) conv r.
Similarly,a functionF of two positive integers is said to be A-definable
if it is possible to find a formulaF such that, wheneverF (m, n) - r, the
formula{F} (m, n) is convertible into r (in, n, r beingpositiveintegersand
m, n, r the corresponding formulas). And so on for functionsof threeor
morepositiveintegers.6
It is clear that, in the case of any A-definablefunctionof positive
integers,the process of reductionof formulasto normal formprovidesan
algorithmfor the effective calculationof particularvalues of the function.

of a formula. Adaptingto the formal


3. The Godel representation
notation just described a device which is due to Giidel,7 we associate with
every formula a positive integer to represent it, as follows. To each of the
symbols{, (, [ we let correspondthe number11, to each of the symbols
}, ), ] the number13, to the symbolA the number1, and to the variables
a, b, c, * the prime numbers17, 19, 23, respectively. And with a
formulawhichis composedof the symbolsr1, * * * , T,,in orderwe associate
n -2y
the number2t132 . .* pntn,wheretUis thenumbercorresponding to the symbol
,r, and wherep,n standsforthe n-thprimenumber.
This number 2t3t2. . . p,tnwill be called the Godel representationof the
formula '1r72 * *n

Two distinctformulasmaysometimeshave thesame G6delrepresentation,


because the numbers11 and 13 each correspondto three different symbols,
but it is readilyprovedthat no two distinctwell-formed formulascan have
thesame Godelrepresentation.It is clear,moreover, thatthereis an effective
methodbywhich,givenanyformula,its Godelrepresentation can be calculated;
and likewisethat thereis an effective methodby which,given any positive
integer,it is possibleto determinewhetherit is the GMdelrepresentation of a
well-formed formula and, if it is, to that
obtaini formula.
In this connectionthe Godel representation plays a role similarto that

" Cf. S. C. Kleene, " A theoryof positive integers in formal logic," American Journal
of Mathematics, vol. 57 (1935), pp. 153-173 and 219-244, where the X-definabilityof a
number of familiar functionsof positive integers, and of a number of important general
classes of functions,is established. Kleene uses the term definable,or for;mallydefinable,
in the sense in which we are here using X-definable.
7Kurt Oidel, " tber formal unentscheidbare Siitze der Principia Mathematica und
verwandter Systeme I," Monatshefte fi4r Mathematik und Physik, vol. 38 (1931),
pp. 173-198.

This content downloaded from 155.97.178.73 on Mon, 24 Nov 2014 19:34:23 PM


All use subject to JSTOR Terms and Conditions
3 5O ALONZO CHURCH.

of the matrixof incidencein combinatorialtopology(cf. ? 1 above). For


thereis, in thetheoryof well-formed
formulas,an importantclass of problems,
eachofwhichis equivalentto a problemof elementary
numbertheoryobtainable
by meansof the Giidelrepresentation.8
4. Recursive functions. We definea class of expressions,which we
shall call elementaryexpressions,and whichinvolve,besidesparenthesesand
commas,the symbols1, S, an infiniteset of numericalvariablesx, y, z, - *,
and, for each positiveintegern, an infiniteset fn,g%n, hn, of functional
variableswith subscriptn. This definitionis by inductionas follows. The
symbol1 or any numericalvariable,standingalone, is an elementaryexpres-
sion. If A is an elementaryexpression,then S(A) is an elementaryexpres-
sion. If A1,A2,- * *, Anare elementaryexpressionsand fnis any functional
variablewithsubscriptn, thenf (A,, A2,* * *, An) is an elementary expression.
The particularelementaryexpressions1, S (1), S(S(1) ), * * * are called
numerals. And the positiveintegers1, 2, 3, * are said to correspondto the
numerals1, S(1), S(S(1)), * *.
An expressionof the formA B, whereA and B are elementaryex-
pressions,is called an elementary equation.
The derivedequationsof a set E of elementaryequationsare definedby
inductionas follows. The equationsof B themselvesare derivedequations.
If A B is a derivedequation containinga numericalvariable x, then the
result of substitutinga particularnumeralfor all the occurrencesof x in
A B is a derivedequation. If A B is a derived equation containing
an elementaryexpressionC (as part of eitherA or B), and if eitherC D
or D = C is a derived eqluation,then the result of substitutingD for a
particularoccurrenceof C in A B is a derivedequation.
Suppose that no derivedequationof a certainfiniteset E of elementary
equationshas the formk I wherek and I are different numerals,that the
functionalvariableswhichoccur in E are fn11, fn22,* *, fnr$with subscripts
n1,n2,* *, nr respectively, and that,foreveryvalue of i from1 to r inclusive,
and foreveryset of numeralskit,k2 **, 1 7c,, thereexistsa uniquenumeral1ck
such that fn$j(k1c5k215
.
..mkn) 1ctis a derivedequation of E. And let
F15P2
. . .
* nFr be the functionsof positive integersdefinedby the con-

8 This is merely a special case of the now familiar remark that, in view of the
G4idel representation'and the ideas associated with it, symbolic logic in general can
be regarded, mathematically, as a branch of elementary number theory. This remark
is essentially due to Hilbert (cf. for example, Verhandlungen des dritteni-nternationalen
Mathematiker-Kongresses in Heidelberg, 19X)4, p. 185; also Paul Bernays in Die
Naturwissenschaften, vol. 10 (1922), pp. 97 and 98) but is most clearly formulated
in terms of the G,6delrepresentation.

This content downloaded from 155.97.178.73 on Mon, 24 Nov 2014 19:34:23 PM


All use subject to JSTOR Terms and Conditions
AN UNSOLVABLEF PROBLEM OF NUiMBER THEORY. 351

ditionthat,in all cases,Fl (m,1,m2iy* * *, mj.l) shall be equal to ml, where


ml, m2ty m**n
, to
and ml are the positiveintegerswhichcorrespond
the numeralskli,k2i,. , knt , and kl respectively.Then the set of equa-
tions E is said to define,or to be a set of recursionequcations for, any one
of the functionsFt, and the functionalvariable fnitis said to denote the
functionFi.
A functionof positiveintegersforwhicha set of recursionequationscan
be givenis said to be recursive.9
It is clear thatforany recursivefunctionof positiveintegersthereexists
an algorithmusing whichany requiredparticularvalue of the functioncan be
effectively calculated. For the derivedequationsof the set of recursionequa-
tions E are effectively enumerable,and the algorithmfor the calculationof
particular values of a functionFi, denoted by a functionalvariable fnt i,

consistsin carryingout the enumerationof the derivedequationsof E until


the requiredparticularequation of the form fn,q(1c1,12t, l n,l) kt- is ,

found.'0
We call an infinitesequenceof positiveintegersrecursiveif the function
P such thatF(n7) is the n-thtermof the sequenceis recursive.
We call a propositionalfunctionof positive integersrecursiveif the
functionwhosevalue is 2 or 1, accordingto whetherthe propositionalfunction
is true or false,is recursive. By a recursivepropertyof positiveintegerswe
shall mean a recursivepropositionalfunctionof one positiveinteger,and by
a recursiverelation betweenpositive integerswe shall mean a recursive
propositionalfunctionof two or morepositiveintegers.

9 This definitionis closely related to, and was suggested by, a definitionof recursive
functions which was proposed by Kurt Godel, in lectures at Princeton, N. J., 1934, and
,credited by him in part to an unpublished suggestion of Jacques Herbrand. The
principal features in which the present definitionof recursiveness differsfrom G8del's
are due to S. C. Kleene.
In a forthcomingpaper by Kleene to be entitled, " General recursive functions of
natural numbers," (abstract in Bulletin of the American Mathematical Society, vol. 41),
several definitions of recursiveness will be discussed and equivalences amongfthem
obtained. In particular, it follows readily from Kleene's results in that paper that
every function recursive in the present sense is also recursive in the sense of G6del
(1934) and conversely.
10The reader may object that this algorithm cannot be held to provide an effective
calculation of the required particular value of Fi unless the proof is constructive that
the required equation f,,i (kli, k2, . . , k i) = ki will ultimately be found. But if so
this merely means that he should take the existential quantifier which appears in our
definitionof a set of recursion equations in a constructive sense. What the criterion
of constructiveness shall be is left to the reader.
The same remark applies in connection with the existence of an algorithm for
calculating the values of a X-definablefunction of positive integers.

This content downloaded from 155.97.178.73 on Mon, 24 Nov 2014 19:34:23 PM


All use subject to JSTOR Terms and Conditions
352 ALONZO CHURCH.

A functionF, forwhichthe range of the dependentvariableis contained


in the class of positiveintegersand the range of the independentvariable,
or of each independentvariable,is a subset (not necessarilythe whole) of the
class of positiveintegers,will be called potentiallyrecursive,if it is possible
to finda recursivefunctionF' of positiveintegers(for whichthe rangeof the
independentvariable,or of each independentvariable,is the wholeof the class
of positiveintegers),such that the value of ' agreeswiththe value of F in
all cases wherethe latteris defined.
By an operationon well-formed formulaswe shall mean a functionfor
whichthe range of the dependentvariable is containedin the class of well-
formedformulasand the range of the independentvariable,or of each in-
dependentvariable,is the wholeclass of well-formed formulas. And we call
such an operationrecursiveif the corresponding functionobtainedbyreplaci-ng
all formulasby theirGoidelrepresentations is potentiallyrecursive.
Similarlyany functionforwhichthe range of the dependentvariableis
containedeitherin the class of positiveintegersor in the class of well-formed
formulas,and for whichthe range of each independentvariable is identical
eitherwith the class of positiveintegersor with the class of well-formed
formulas(allowing the case that some of the ranges are identicalwith one
class and somewiththe other),will be said to be recursiveif the corresponding
functionobtainedby replacingall formulasby their Godel representations is.
potentiallyrecursive. We call an infinitesequenceof well-formed formulas
recursiveif the corresponding infinitesequence of G6del representations is
recursive. And we call a propertyof, or relation between,well-formed
formulasrecursiveif the corresponding propertyof, or relationbetween,their
G6del representations is potentiallyrecursive. A set of well-formed formulas
is said to be recursivelyenumerableif thereexistsa recursiveinfinitesequence
whichconsistsentirelyof formulasof the set and containseveryformulaof
the set at least once."1
In termsof the notionof recursiveness we may also definea proposition
of elementarynumbertheory,by induction as follows. If c is a recursive
propositionalfunctionof n positiveintegers(definedby giving a particular
set of recursionequationsfor the corresponding functionwhosevalues are 2
and 1) and if xl, x2, , xn are variableswhichtake on positiveintegersas
values,then p(xi,,x, , xn) is a propositionof elementary numbertheory.
If P is a propositionof elementarynumbertheoryinvolvingx as a free

11 It can be shown, in view of Theorem V below, that, if an infinliteset of formulas


is recursively enumerable in this sense, it is also recursively enumerable in the sense
that there exists a recursive infiniitesequence which consists entirely of formulas of
the set and contains every formula of the set exactly once.

This content downloaded from 155.97.178.73 on Mon, 24 Nov 2014 19:34:23 PM


All use subject to JSTOR Terms and Conditions
AN UNSOLVABLE PROBLEM OF NUMBER THEORY. 353

variable,then the resultof substitutinga particularpositiveintegerfor all


occurrencesof x as a freevariablein P is a propositionof elementary number
theory,and (x)P and (3x)P are propositionsof elementarynumbertheory,
where(x) and (3x) are respectively the universaland existentialquantifiers
of x overthe class of positiveintegers.
It is then readilyseen that the negationof a propositionof elementary
numbertheoryor the logical productor the logical sum of two propositions
of elementary numbertheoryis equivalent,in a simpleway,to anotherproposi-
tion of elementary numbertheory.
of the Kleene p-function.We provetwotheorems
5. Recursiveness
whichestablishthe recursivenessof certainfunctionswhich are definablein
words by means of the phrase, "The least positiveintegersuch that," or,
"The n-thpositiveintegersuch that."
THEOREM IV. If F is a recursivefunctionof twopositiveintegers,and
if for everypositiveintegerx there exisis a positiveinteger y such that
P(x, y) > 1, then the functionF*, such that, for everypositiveintegerx,
F* (x) is equal to the least positive integer y for which F(x, y) > 1, is
recursive.
For a set of recursionequationsforF* consistsof the recursionequations
forF togetherwiththe equations,

i2(1, 2) =- 2, g2(x,1)
-X i2(f2(x,1), 2),
i2(S(X)5 2) 15 g2(X., S(y) ) i2(ft(X5 S(y) ), g2(X5 Y) ),
i2(X, ) ='3, h2(S(x), y) x,
i2(x,S(S(y))) == 3, h2(g2(x,y),x) j2(g2 (X, y), y),
j2 (1, y) y, f,(x) == h2 (1, x),
j2(S(x), y) = x,

wherethe functionalvariablesf2 and f, denotethe functionsF and F* re-


forS(1) and S(S(1)) respectively.12
and 2 and 3 are abbreviations
spectively,
THEOREM V. If F is a recursivefunctionof one positiveinteger,and
if thereexist an infinitenumberof positiveintegersx for whichF(x) > 1,
then the functionF0, such that,for everypositiveintegern, FO(n) is equal
to the n-thpositiveintegerx (in orderof increasingmagnitude)for which
F(x) > 1, is recursive.

12Since this result was obtained, it has been pointed out to the author by S. C.
Kleene that it can be proved more simply by using the methods of the latter in American
Journal of Mathematics, vol. 57 (1935), p. 231 et seq. His proof will be given in his
forthcomingpaper already referredto.
8

This content downloaded from 155.97.178.73 on Mon, 24 Nov 2014 19:34:23 PM


All use subject to JSTOR Terms and Conditions
354 ALONZO CHURCH.

For a set of recursionequationsforFO consistsof the recursionequations


forF togetherwiththe equations,
g2(1, y) = g2(fI(S(y) ), S(y)),
g2(S(X),y) =y,
g1(1) = k,
91(S (y) ) = g2(l, 91 (y) )
where the functionalvariables g1 and f1 denote the functionsFO and F
and where7cis thenumeralto whichcorresponds
respectively, the least positive
integerx forwhichF(x) > 1.13
6. Recursiveness of certain functionsof formulas. We list now a
numberof theoremswhichwill be provedin detail in a forthcoming paper
by S. C. Kleene 14 or followimmediately fromconsiderationstheregiven. We
omit proofshere,exceptforbriefindicationsin some instances.
Our statementof the theoremsand our notationdifferfromKleene's in
that we employthe set of positiveintegers(1, 2, 3,* ) in the role in which
he employsthe set of natural numbers(0, 1, 2, ). This differenceis, of
course,unessential. We have selectedwhat is, fromsomepointsof view,the
less natural alternative,in orderto preservethe convenienceand naturalness
of the identificationof the formula)ab a (b) with 1 ratherthan with0.
VI. The propertyof a positiveinteger,that there exists a
THEOREM
formulaof whichit is the Godel representation
well-formed is recursive.
THEOREM VII. The setof well-formed
formulasis recursively
enumerable.
This followsfromTheoremsV and VI.
THEOREM VIII. The functionof twovariables,whosevalue,whentakcen
of the well-formed
formulasF and X, is the formula{F} (X), is recursive.
THEOREM IX. The function,whosevalueforeach of thepositiveintegers
1, 2, 3, formula1, 2, 3,
is the corresponding *, is recursive.
THEOREM X. A function, whosevaluefor eachof theformulas1, 2, 3, .*
is the correspondingpositiveinteger,and whosevalue for otherwell-formed
formulasis a fixedpositiveinteger,is recursive. Likewvise
thefunction,whose
value foreach of theformulas1, 2, 3,. is the corresponding
positiveinteger

13 This proof is due to Kleene.


14 S. C. Kleene, " X-definability and recursiveness," forthcoming (abstract in
Bulletin of the American Mathematical Society, vol. 41). In connection with many
of the theorems listed, see also Kurt Gvdel, Monatshefte fur Mathemditikund Physik,
vol. 38 (1931), p. 181 et seq., observing that every function which is recursive in the
senise in which the word is there used by G6del is also recursive in the present more
general sense.

This content downloaded from 155.97.178.73 on Mon, 24 Nov 2014 19:34:23 PM


All use subject to JSTOR Terms and Conditions
AN UNSOLVABLE PROBLEM OF NUMBER THEORY. 355

vlus one, and whose value for other well-formed


formulas is the positive
integer1, is recursive.
THEOREMXI. The relationof immediateconvertibility,
betweenwell-
formedformulas,is recursive.
THEOREM XII. It is possibleto associatesimultaneously witheverywell-
formedformulaan enumerationof the formulasobtainablefromn it by con-
version,in such a way that the functionof two variables,whosevalue, when
takenof a well-formed formulaA and a positiveintegern, is then-thformula
in theenumeration of theformulasobtainablefromA byconversion, is recursive.
THEOREMXIII. The propertyof a well-formed
formula,that it is in
principalnormalform,is recursive.
THEOREM XIV. The set of well-formed
formulaswhichare in principal
normalformis recursively
enumerable.
This followsfromTheoremsV, VII, XIII.
THEOREM XV. The set of well-formed formulaswhichhave a normal
formnis recursivelyenumerable.'5
For by TheoremsXII and XIV this set can be arrangedin an infinite
square arraywhichis recursivelydefined(i. e. definedby a recursivefunction
of two variables). And the familiarprocessby which this square array is
reducedto a single infinitesequenceis recursive(i. e. can be expressedby
means of recursivefunctions).
THEOREMXVI. Every recursive function of positive integers is
A-definable.'6

THEOREMXVII. Every A-definablefunction of positive integers is


recursive.'7
For functionsof one positiveintegerthis followsfrom TheoremsIX,
VIII, XII, XIII, IV, X. For functionsof more than one positiveinteger

15This theorem was firstproposed by the present author, with the outline of proof
here indicated. Details of its proof are due to Kleene and will be given by him in his
forthcomingpaper, " X-definabilityand recursiveness."
16 This theorem can be proved as a straightforward application of the methods
introduced by Kleene in the American Journal of Mathematics (loc. cit.). In the form
here given it was first obtained by Kleene. The related result had previously been
obtained by J. B. Rosser that, if we modify the definitionof well-formedby omitting
the requirement that M contain x as a free variable in order that Xx[M] be well-
formed,then every recursive function of positive integers is X-definablein the resulting
modifiedsense.
17 This result was obtained independently by the present author and S. C. Kleene

at about the same time.

This content downloaded from 155.97.178.73 on Mon, 24 Nov 2014 19:34:23 PM


All use subject to JSTOR Terms and Conditions
356 ALONZO CHURCH.

it followsby the same method,using a generalizationof Theorem IV to


functionsof morethantwopositiveintegers.
7. The notion of effectivecalculability. We now definethe notion,
alreadydiscussed,of an effectively calculablefunctionof positiveintegersby
identifyingit with the notion of a recursivefunctionof positiveintegers18
(or of a A-definable functionof positiveintegers). This definition is thought
to be justifiedbythe considerations whichfollow,so far as positivejustification
can everbe obtainedforthe selectionof a formaldefinitionto correspondto
an intuitivenotion.
It has already been pointed out that, for every functionof positive
integerswhichis effectively calculable in the sense just defined,thereexists
an algorithmforthe calculationof its values.
Converselyit is true,underthe same definitionof effective calculability,
that everyfunction,an algorithmfor the calculationof the values of which
exists,is effectivelycalculable. For example,in the case of a functionP of
one positiveinteger,an algorithmconsistsin a methodby which,given any
positiveintegern,a sequenceofexpressions (in somenotation)EB,, 2 ..**Enr,,
can be obtained; whereE., is effectively calculablewhenn is given; where
En*is effectively calculablewhenn and the expressionsEnj, j < i, are given;
and where,whenn and all the expressionsEn* up to and includingEnrnare
given,the fact that the algorithmhas terminatedbecomeseffectively known
and the value of F(n) is effectively calculable. Suppose that we set up a
systemof Godel representations forthe notationemployedin the expressions
En*, and that we then furtheradopt the methodof &Mdelof representing a
finitesequenceof expressionsEn1,En2, * , En* by the single positiveinteger
2enl3en2. * pion, wheree%,n,e n2., *, en* are respectively
* the MMdel representa-
-
tions of E,,,,-En2 * * En* (in particularrepresenting a vacuous sequenceof
expressionsby the positiveinteger1). Then we may definea functionG
of two positive integers such that, if x representsthe finite sequence
EnlnEn2, . . . E,, then G (n, x) is equal to the Godel representation of EM,
wherei = lk+ 1, or is equal to 10 if kI rn (that is if the algorithmhas
terminatedwith E?*), and in any other case G(n, x) is equal to 1. And
we may definea functionH of two positiveintegers,such that the value of
H(n, x) is the same as that of G(n, x), exceptin the case that G(n, x) = 10,
in which case Hf(n,x) = F(n). If the interpretation is allowed that the

18 The question of the relationship betwen effectivecalculability and recursiveness

(which it is here proposed to answer by identifyingthe two notions) was raised by


Godel in conversation with the author. The correspondingquestion of the relationship
between effectivecalculability and X-definabilityhad previously been proposed by the
author independently.

This content downloaded from 155.97.178.73 on Mon, 24 Nov 2014 19:34:23 PM


All use subject to JSTOR Terms and Conditions
AN UNSOLVABLE PROBLEM OF NU.M-BER THEORY. 357

requirementof effective calculabilitywhichappears in our descriptionof an


algorithmmeans the effective calculabilityof the functionsG and H,19 and
if we take the effectivecalculabilityof G and H to mean recursiveness
(X-definability), then the recursiveness(A-definability)of F followsby a
straightforward argument.
Supposethatwe are dealingwithsomeparticularsystemof symboliclogic,
which contains a symbol,-, for equality of positive integers,a symbol
{ } ( ) for the applicationof a functionof one positiveintegerto its argu-
ment,and expressions1, 2, 3, * to stand for the positiveintegers. The
theoremsof the systemconsist of a finite,or enumerablyinfinite,list of
expressions,the formalaxioms,togetherwith all the expressionsobtainable
fromthemby a finitesuccessionof applicationsof operationschosenout of
a givenfinite,or enumerably list of operations,the rulesof procedure.
infinite,
If the systemis to serveat all the purposesfor whicha systemof symbolic
logic is usually intended,it is necessarythat each rule of procedurebe an
effectively calculable operation,that the completeset of rules of procedure
(if infinite)be effectivelyenumerable,that the completeset of formalaxioms
(if infinite)be effectively
enumerable,and thatthe relationbetweena positive
integerand the expressionwhich stands for it be effectively determinable.
Suppose that we interpretthis to mean that,in termsof a systemof G6del
representations for the expressionsof the logic, each rule of proceduremust
be a recursiveoperation,20 the completeset of rules of proceduremust be
recursively enumerable(in the sense that thereexists a recursivefunction-
such that b (n, x) is the representation of the resultof applyingthe n-thrule
of procedureto the orderedfiniteset of formulasrepresentedby x), the
completeset of formalaxiomsmustbe recursively enumerable, and therelation
betweena positiveintegerand the expressionwhich stands for it must be
recursive.21 And let us call a functionF of one positiveinteger22 calculable
withinthelogicif thereexistsan expressionf in thelogic suchthat {f} (jA) = v
is a theoremwhen and onlywhenF(m) -= n is true, ,uand v being the ex-
pressionswhich stand for the positiveintegersm and n. Then, since the

19 If this interpretation or some similar one is not allowed, it is difficultto see

how the notion of an algorithm can be given any exact meaning at all.
20 As a matter of fact, in known systems of symbolic logic, e. g. in that of Principia

Mathematica, the stronger statement holds, that the relation of immediate consequence
(unmittelbare Folge) is recursive. Cf. G,8del,loc. cit., p. 185. In any case where the
relation of immediate consequence is recursive it is possible to find a set of rules
of procedure, equivalent to the original ones, such that each rule is a (one-valued)
recursive operation, and the complete set of rules is recursively enumerable.
21 The author is here indebted to Godel, who, in his 1934 lectures already referred

to, proposed substantially these conditions, but 'in terms of the more restricted notion

This content downloaded from 155.97.178.73 on Mon, 24 Nov 2014 19:34:23 PM


All use subject to JSTOR Terms and Conditions
358 ALONZO CHURCH.

completeset of theoremsof the logic is recursivelyenumerable,it followsby


Theorem IV above that every functionof one positive integer which is
calculablewithinthe logic is' also effectively
calculable (in the sense of our
definition).
Thus it is shownthat no moregeneraldefinitionof effective calculability
than that proposedabove can be obtainedby eitherof two methodswhich
naturally suggest themselves(1) by defininga functionto be effectively
calculableif thereexistsan algorithmforthe calculationof its values (2) by
defininga functionF (of one positiveinteger) to be effectively calculableif,
for every positive integer m, there exists a positive integer n such that
F(m) =- n is a provabletheorem.

8. Invariants of conversion. The problemnaturallysuggestsitselfto


findinvariantsof that transformation of formulaswhichwe have called con-
version. The only effectively calculableinvariantsat presentknownare the
immediately obviousones (e. g. thesetof freevariablescontainedin a formula).
Othersof importanceveryprobablyexist., But we shall prove (in Theorem
XIX) that, under the definitionof effectivecalculabilityproposed in ? 7,
no completeset ofeffectively calculableinvariantsof conversionexists(cf. ? 1).
The resultsof Kleene (AmericanJournalof Mathematics,1935) make
it clear that,if the problemof findinga completeset of effectively calculable
invariantsof conversion weresolved,mostof thefamiliarunsolvedproblemsof
elementary numbertheorywould therebyalso be solved. And fromTheorem
XVI aboveit followsfurtherthatto finda completeset of effectively calculable
invariantsof conversion wouldimplythesolutionof the Entscheidungsproblem
for any systemof symboliclogic whatever(subject to the verygeneral re-
strictionsof ? 7). In thelightof thisit is hardlysurprisingthatthe problem
to findsuch a set of invariantsshouldbe unsolvable.
It is to be remembered, however,that,if we consideronlythe statement
of the problem(and ignorethingswhichcan be provedabout it by moreor
less lengthyarguments),it appears to be a problemof the same class as the
problemsof numbertheoryand topologyto which it was comparedin ? 1,
havingno strikingcharacteristic by whichit can be distinguishedfromtbem.
The temptationis strongto reasonby analogythat otherimportantproblems
of this class may also be unsolvable.

of recursiveness which he had employed in 1931, and using the condition that the
relation of immediate consequence be recursive instead of the present conditions on the
rules of procedure.
22 We confine ourselves for convenience to the case of functions of one positive

integer. The extension to functions of several positive integers is immediate.

This content downloaded from 155.97.178.73 on Mon, 24 Nov 2014 19:34:23 PM


All use subject to JSTOR Terms and Conditions
AN UNSOLVABLE PROBLEM OF NUMBER THEORY. 359

LEMMA. The problem,to find a recursivefunctionof two formulas


A and B whosevalue is 2 or 1 accordingas A conv B or not, is equivalent
to the problem,to finda recursivefunctionof one formulaC whosevalue is
2 or 1 accordingas C has a normalformor not.23
For, by TheoremX, the formulaa (the formulab), whichstandsforthe
positive integerwhich is the Godel representationof the formulaA (the
formulaB), can be expressedas a recursivefunctionof the formulaA (the
formulaB). Moreover,by TheoremsVI and XII, there exists a recursive
functionF of twopositiveintegerssuch that,if m is the G6del representation
of a well-formedformulaM, thenF (n, n) is the Godel representation of the
n-thformulain an enumerationof the formulasobtainablefromr M by con-
version. And,byTheoremXVI, F is A-definable, by a formulaf. If we define,
*
Z, --S (Akxx$(I), I),
Z2- ?9>,i*(XxyS(X) -y,I),

where~2is the formuladefinedby Kleene (AmericanJournalof Mathematics,


vol. 57 (1935), p. 226), thenZ1 and Z2 A-define the functionsof one positive
integerwhosevalues,for a positiveintegern, are the n-thtermsrespectively
oftheinfinitesequences1, 1, 2, 1, 2, 3, ... and 1, 2, l, 3, 2, 1, *. By Theorem
VIII the formula,
{Xxy * (An *8 (f(x, Z, (n) , (Y,Z2 (n) ) )) }(an, b)
where4 and 8 are definedas by Kleene (lOc. cit., p. 173 and p. 231), is a
recursivefunctionof A and B, and this formulahas a normalformif and
onlyif A convB.
Again, by Theorem X, the formulac, which stands for the positive
integerwhichis the G6del representationof the formulaC, can be expressed
as a recursivefunctionof the formulaC. By TheoremsVI and XIII, there
exists a recursivefunctionG of one positiveintegersuch that G(m) = 2
if m is the G6del representation
of a formulain principalnormalform,and
G(m) = 1 in any othercase. And, by TheoremXVI, G is A-definable, by a
formulag. By TheoremVIII the formula,
-
{Xx (n- ( x,n . 1, ) c
^; 1)\w

23
These two problems,in the forms, (1) to find an effectivemethod of determining
of any two formulas A and B whether A conv B, (2) to find an effectivemethod of
determining of any formula C whether it has a normal form, were both proposed by
Kleene to the author, in the course of a discussion of the properties of the p-function,
about 1932. Some attempts towards solution of (1) by means of numerical invariants
were actually made by Kleene at about that time.

This content downloaded from 155.97.178.73 on Mon, 24 Nov 2014 19:34:23 PM


All use subject to JSTOR Terms and Conditions
360 ALONZO CHURCH.

wheref is the formulaf used in the precedingparagraph,is a recursivefunc-


tion of C, and this formulais convertibleinto the formula1 if and only if
C has a normalform.
Thus we have proved that a formulaC can be found as a recursive
functionof formulasA and B, such that C has a normalformif and only
if A convB; and that a formulaA can be foundas a recursivefunctionof a
formulaC, suchthatA conv1 if and onlyif C has a normalform. From this
the lemma follows.
THEOREMXVIII. There is no recursivefunctionof a formulaC, whose
value is 2 or 1 accordingas C has a normalformor not.
That is, thepropertyof a well-formed formula,thatit has a normalform,
is not recursive.
For assumethe contrary.
Then there exists a recursivefunctionH of one positiveintegersuch
that H (n) = 2 if m is the Go-delrepresentation of a formulawhichhas a
normal form,and H (in) 1 in any other case. And, by TheoremXVI,
H is A-definable by a formula j.
By TheoremXV, thereexistsan enumeration ofthe well-formed formulas
whichhave a normalform,and a recursivefunctionA of one positiveinteger
such that A (n) is the G6del representationof the n-th formula in this
enumeration. And, by TheoremXVI, A is X-definable, by a formulaa.
By TheoremsVI and VIII, thereexists a recursivefunctionB of two
positiveintegerssuch that, if m and n are WMdel representations of well-
formedformulasM and N, then B (in, n) is the G6del representationof
{M} (N). And, by TheoremXVI, B is A-definable, by a formulab.
By TheoremsVI and X, thereexistsa recursivefunctionC of one positive
integersuch that, if m is the Go6delrepresentation of one of the formulas
1, 2, 3, . . , then C(m) is the corresponding positiveintegerplus one, and
in any othercase C (n) = 1. And, by TheoremXVI, C is X-definable, by a
formulac.
By Theorem IX there exists a recursivefunctionZ-1 of one positive
integer,whosevalue for each of the positiveintegers1, 2, 3, is the GO6del
representation of the corresponding formula1, 2, 3, . And, by Theorem
XVI, Z-1 is A-definable, by a formula3.
Let f and g be the formulasf and g used in the proofof the Lemma.
By Kleene 15III Cor. (loc. cit.,p. 220), a formulab can be foundsuch that,
b (1) convXx x (1)
b(2) convXu c(f(U.,4(Am g(f(U, m)),))).

This content downloaded from 155.97.178.73 on Mon, 24 Nov 2014 19:34:23 PM


All use subject to JSTOR Terms and Conditions
AN UNSOLVABLE PROBLEM OF NUMBER THEORY. 361

We define,
e-> An b(4 (b (a (n), 3(n) )),( (n), 3(n))).
Then if n is one of the formulas1, 2, 3, , e(n) is convertible into one
of the formulas1, 2, 3, * in accordancewith the followingrules: (1) if
b(a (n), 3(n) ) conva formulawhichstandsforthe G6del representation of a
formulawhichhas no normalform,e(n) conv1, (2) if b(an(n), 3(n) ) conva
formulawhichstandsforthe Giidelrepresentation of a formulawhichhas a
principalnormalformwhichis not oneoftheformulas1, 2, 3, *, e(n) conv1,
(3) if b(a (n), 3(n) ) conva formulawhichstandsforthe Gozdelrepresentation
of a formulawhichhas a principalnormalformwhichis one of the formulas
2, 3, . -.-, e(n) conv the next following formula in the list 1, 2, 3,* - - .
By TheoremIII, since e(1) has a normal form,the formulae has a
normalform. Let 0, be theformulawhichstandsforthe G,6delrepresentation
of e. Then, if n is any one of the formulas1, 2, 3, * * , ($ is not convertible
into the formulaa (n), because b (6Y,3(n)) is, by the definitionof Ii, con-
vertibleinto the formulawhichstands for the Goidelrepresentation of e(n),
while b(a (n), 3(n) ) is, by the precedingparagraph,convertibleilto the
formulastandsforthe Godel representation of a formuladefinitely not con-
e
vertibleinto (n) (TheoremII). But, by our definition of a, it must be true
.
of one of the formulasn in the list 1, 2, 3, * that a(n) conv
Thus, sinceour assumptionto the contraryhas led to a contradiction, the
theoremmustbe true.
In order to presentthe essential ideas withoutany attemptat exact
statement,the precedingproofmay be outlinedas follows. We are to deduce
a contradictionfromthe assumptionthat it is effectively determinableof
everywell-formedformulawhetheror not it has a normal form. If this
assumptionholds,it is effectively determinableof everywell-formed formula
whetheror not it is convertibleinto one of the formulas1, 2, 3, . . *; for,
givena well-formed formulaR, we can firstdeterminewhetheror not it has
a normal form,and if it has we can obtain the principalnormal formby
enumeratingthe formulasinto whichR is convertible(Theorem XII) and
pickingout the firstformulain principalnormalformwhich occursin the
enumeration,and we can then determinewhetherthe principalnormalform
is one of the formulas1, 2, 3, . Let A1,A2,A3, be an effectiveenumera-
tion of the well-formed formulaswhichhave a normalform(TheoremXV).
Let E be a functionof one positiveinteger,definedby the rule that,where
m and n are the formulaswhich stand for the positiveintegersm and n
respectively, E (n) = 1 if {An}(n) is not convertible into one of the formulas
1,2,3< *, and E(n) =m + 1 if {An}(n) conv m and m is one of the
formulas1, 2, 3, . The functionE is effectively calculable and is there-

This content downloaded from 155.97.178.73 on Mon, 24 Nov 2014 19:34:23 PM


All use subject to JSTOR Terms and Conditions
362 ALONZO CHURCH.

foreA-definable,by a formulae. The formulae has a normalform,since


e(1) has a normalform. But e is not anyone of theformulasA1,A2,A3, -
because,foreveryn, e (n) is a formulawhichis lot convertible
into {An} (n).
And this contradictsthe propertyof the enumerationA1,A2,A3, that it
containsall well-formed formulaswhichhave a normalform.
COROLLARY 1. The set of well-formed
formulaswhichhave no normal
formis not recursively
enumerable.24
For, to outlinethe argument,the set of well-formed formulaswhichhave
a normalformis recursively enumerable, by TheoremXV. If the set of those
whichdo not have a normalformwere aslo recursively enumerable,it would
be possibleto tell effectivelyof any well-formed formulawhetherit had a
normalform,by the processof searchingthroughthe two enumerations until
it was foundin one or the other.This, however,is contraryto TheoremXVIII.
This corollarygives us an exampleof an effectively enumerableset (the
set of well-formed formulas)whichis dividedinto two non-overlapping sub-
sets of whichone is effectively
enumerableand the othernot. Indeed, in view
of the difficultyof attachingany reasonablemeaningto the assertionthat a
set is enumerablebut not effectivelyenumerable,it may even be permissible
to go a step furtherand say that here is an example of an enumerableset
whichis dividedinto two non-overlapping subsetsof whichone is enumerable
and the othernon-enumerable.25
2. Let a functionF of one positiveintegerbe definedby
COROLLARY
the rule that F(n) shall equal 2 or 1 accordingas n is or is not the Godel
representation ofa formulawhichhas a normalform. Then F (ifits definition
be admittedas valid at all) is an exampleof a non-recursive
functionof posi-
tiveintegers.26
This followsat oncefromTheoremXVIII.

24
This corollary was proposed by J. B. Rosser.
The outline of proof here given for it is open to the objection, recently called to
the author's attention by Paul Bernays, that it ostensibly requires a non-constructive
use of the principle of excluded middle. This objection is met by a revision of the
proof, the revised proof to consist in taking any recursive enumeration of formulas
which have no normal form and showing that this enumeration is not a complete
enumeration of such formulas, by constructing a formula e (n) such that (1) the
supposition that e(n) occurs in the enumeration leads to contradiction (2) the sup-
position that e(n) has a normal form leads to contradiction.
25 Cf. the remarks of the author in The American Mathematical Monthly, vol. 41

(1934), pp. 356-361.


26 Other examples of non-recursive functions have since been obtained by S. C.

Kleene in a differentconnection. See his forthcomingpaper, " General recursive func-


tions of natural numbers."

This content downloaded from 155.97.178.73 on Mon, 24 Nov 2014 19:34:23 PM


All use subject to JSTOR Terms and Conditions
AN UNSOLVABLEPROBLEMOF NUMBERTHEORY. 363

Considertheinfinitesequenceofpositiveintegers,F (1), F (2), F (3), * - .


It is impossibleto specifyeffectively a methodby which,given any n, the
n-thtermof this sequencecould be calculated. But it is also impossibleever
to select a particulartermof this sequenceand prove about that termthat
its value cannot be calculated (because of the obvioustheoremthat if this
sequencehas termswhosevalues cannotbe calculatedthen the value of each
of thoseterms1). Thereforeit is naturalto raise the questionwhether,in
spite of the fact that thereis no systematicmethodof effectively calculating
the termsof this sequence,it mightnotbe trueof each termindividuallythat
thereexisteda methodof calculatingits value. To this questionperhapsthe
best answeris thatthe questionitselfhas no meaning,on the groundthatthe
universalquantifierwhichit containsis intendedto expressa mere infinite
successionof accidentsratherthan anythingsystematic.
There is in consequencesome roomfor doubtwhetherthe assertionthat
the functionF existscan be given a reasonablemeaning.
THEOREM XIX. There is no reciursive functionof twoformulasA and
B, whosevalue is 2 or 1 accordingas A convB or not.
This followsat once fromTheoremXVIII and the Lemma precedingit.
As a corollaryof Theorem XIX, it followsthat the Entscheidungs-
problemis unsolvablein the case of any systemof symboliclogic which is
)-consistent in the sense of Godel (loc. cit.,p. 187) and
(w-widerspruchsfrei)
is strongenoughto allow certaincomparatively simplemethodsof definition
and proof. For in any such systemthe propositionwill be expressibleabout
two positive integers a and b that they are Gtdel representations of formulas
A and B such thatA is immediatelyconvertible into B. Hence, utilizingthe
factthata conversion is a finitesequenceofimmediateconversions, theproposi-
tion I(a, b) will be expressiblethat a and b are Godel representations of
formulasA and B such that A convB. Moreoverif A convB, and a and b
are the Goidelrepresentations of A and B respectively, the propositionI(a, b)
will be provablein thesystem, bya proofwhichamountsto exhibiting, in terms
of Godelrepresentations, a particularfinitesequenceof immediateconversions,
leading fromA to B; and if A is not convertibleinto B, the 0-consistency
of the systemmeansthat*(a, b) will not be provable. If the Entscheidungs-
problem for the system were solved, there would be a means of determining
of everypropositionI(a, b) whetherit was provable,and hence
effectively
a meansof determining of everypair of formulasA and B whether
effectively
A convB, contraryto TheoremXIX.
In particular,if the systemof Principia Mathematicabe 0-consistent,
its Entscheidungsproblem is unsolvable.
PRINCETONUNIVERSITY,
PRINCETON,N. J.

This content downloaded from 155.97.178.73 on Mon, 24 Nov 2014 19:34:23 PM


All use subject to JSTOR Terms and Conditions

You might also like