On the Nature of Mathematical Truth

Author(s): C. G. Hempel
Source: The American Mathematical Monthly, Vol. 52, No. 10 (Dec., 1945), pp. 543-556
Published by: Mathematical Association of America
Stable URL: http://www.jstor.org/stable/2306103
Accessed: 19/07/2010 22:43

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at
http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at
http://www.jstor.org/action/showPublisher?publisherCode=maa.

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.

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.

Mathematical Association of America is collaborating with JSTOR to digitize, preserve and extend access to
The American Mathematical Monthly.

http://www.jstor.org
 ON THE NATURE OF MATHEMATICAL TRUTH
C. G. HEMPEL, QueensCollege
1. The problem. It is a basic principleof scientificinquirythat no proposi-
tion and no theoryis to be accepted withoutadequate grounds. In empiricalsci-
ence, whichincludesboth the natural and the social sciences,the groundsforthe
acceptance of a theoryconsist in the agreementof predictionsbased on the the-
ory with empirical evidence obtained either by experimentor by systematic
observation. But what are the groundswhichsanction the acceptance of mathe-
matics? That is the question I propose to discuss in the presentpaper. For rea-
sons which will become clear subsequently, I shall use the term "mathematics"
here to referto arithmetic,algebra, and analysis-to the exclusion,in particular,
of geometry[1].
2. Are the propositionsof mathematicsself-evidenttruths?One of the sev-
eral answers which have been given to our problem asserts that the truthsof
mathematics, in contradistinctionto the hypotheses of empirical science, re-
quire neitherfactual evidence nor any otherjustificationbecause they are "self-
evident." This view, however, which ultimately relegates decisions as to
mathematical truthto a feelingof self-evidence,encountersvarious difficulties.
First of all, many mathematical theoremsare so hard to establish that even to
the specialist in the particular fieldthey appear as anythingbut self-evident.
Secondly, it is well known that some of the most interestingresults of mathe-
matics-especially in such fieldsas abstract set theoryand topology-run coun-
ter to deeply ingrained intuitions and the customary kind of feeling of self-
evidence. Thirdly, the existence of mathematical conjectures such as those of
Goldbach and of Fermat, which are quite elementaryin content and yet un-
decided up to this day, certainlyshows that not all mathematical truthscan
be self-evident.And finally,even if self-evidencewere attributed only to the
basic postulates ofmathematics,fromwhichall othermathematicalpropositions
can be deduced, it would be pertinentto remark that judgments as to what
may be consideredas self-evidentare subjective; they may vary frompersonto
person and certainlycannot constitutean adequate basis fordecisions as to the
objective validity of mathematical propositions.
3. Is mathematicsthe most general empiricalscience? Accordingto another
view, advocated especially by JohnStuart Mill, mathematicsis itselfan empiri-
cal science which differsfromthe other branches such as astronomy,physics,
chemistry,etc.,mainly in two respects: its subject matter is more general than
that of any other field of scientificresearch, and its propositions have been
tested and confirmedto a greater extent than those of even the most firmly
established sections of astronomyor physics.Indeed, according to this view, the
degree to which the laws of mathematics have been borne out by the past ex-
periencesof mankind is so overwhelmingthat-unjustifiably--we have come to
think of mathematical theorems as qualitatively differentfromthe well-con-
firmedhypothesesor theoriesof otherbranches of science: we considerthem as
543
544 ON THE NATURE OF MATHEMATICAL TRUTH [December,

certain, while other theoriesare thoughtof as at best 'very probable" or very

highlyconfirmed.
But this view, too, is open to serious objections. From a hypothesiswhich is
empirical in character-such as, forexample, Newton's law of gravitation-it
is possible to derive predictionsto the effectthat under certain specifiedcondi-
tions certain specifiedobservable phenomena will occur. The actual occurrence
of these phenomena constitutesconfirmingevidence, their non-occurrencedis-
confirming evidence forthe hypothesis.It followsin particularthat an empirical
hypothesis is theoreticallydisconfirmable;i.e., it is possible to indicate what
kind of evidence, if actually encountered,would disconfirmthe hypothesis.In
the light of this remark,consider now a simple "hypothesis" fromarithmetic:
3 + 2 =5. If this is actually an empiricalgeneralizationof past experiences,then
it must be possible to state what kind of evidence would oblige us to concede
the hypothesiswas not generally true after all. If any disconfirming evidence
forthe given propositioncan be thoughtof,the followingillustrationmightwell
be typical of it: We place some microbeson a slide, puttingdown firstthree of
them and then another two. Afterwardswe count all the microbes to test
whetherin this instance 3 and 2 actually added up to 5. Suppose now that we
counted 6 microbesaltogether.Would we consider this as an empiricaldiscon-
firmationof the given proposition,or at least as a proofthat it does not apply
to microbes? Clearly not; rather,we would assume we had made a mistake in
countingor that one of the microbes had split in two between the firstand the
second count. But under no circumstancescould the phenomenonjust described
invalidate the arithmeticalpropositionin question; forthe latterasserts nothing
whateverabout the behavior ofmicrobes;it merelystates that any set consisting
of 3+2 objects may also be said to consist of 5 objects. And this is so because
the symbols "3+2" and "5" denote the same number: they are synonymousby
virtue of the fact that the symbols "2," "3," "5," and a+" are defined(or tacitly
understood) in such a way that the above identityholds as a consequence of the
meaning attached to the concepts involved in it.
4. The analyticcharacterof mathematicalpropositions.The statementthat
3+2 = 5, then, is true forsimilar reasons as, say, the assertion that no sexage-
narian is 45 years of age. Both are true simply by virtue of definitionsor of
similar stipulations which determinethe meaning of the key terms involved.
Statementsof this kind share certainimportantcharacteristics:Their validation
naturallyrequiresno empiricalevidence; theycan be shownto be trueby a mere
analysis of the meaning attached to the termswhich occur in them. In the lan-
guage of logic, sentencesof this kind are called analytic or true a priori,whichis
to indicate that theirtruthis logicallyindependentof, or logically priorto, any
experientialevidence [2]. And while the statementsof empiricalscience, which
are syntheticand can be validated only a posteriori,are constantlysubject to
revisionin the light of new evidence, the truthof an analytic statementcan be
established definitely,once and forall. However, this characteristic"theoretical
1945] ON THE NATURE OF MATHEMATICAL TRUTH 545

certainty"of analytic propositionshas to be paid forat a highprice: An analytic

statementconveys no factual information.Our statementabout sexagenarians,
for example, asserts nothingthat could possibly conflictwith any factual evi-
dence: it has no factual implications,no empirical content; and it is precisely
forthisreason that the statementcan be validated withoutrecourseto empirical
evidence.
Let us illustrate this view of the nature of mathematical propositionsby
referenceto another, frequentlycited, example of a mathematical-or rather
logical-truth, namely the propositionthat whenevera = b and b = c then a = c.
On what groundscan this so-called "transitivityof identity"be asserted? Is it
of an empiricalnature and hence at least theoreticallydisconfirmableby empiri-
cal evidence? Suppose, forexample, that a, b, c, are certainshades of green,and
that as far as we can see, a=b and b=c, but clearly a0#c. This phenomenon
actually occurs under certain conditions;do we considerit as disconfirming evi-
dence for the proposition under consideration? Undoubtedly not; we would
argue that ifa 0 c, it is impossiblethat a = b and also b = c; between the termsof
at least one of these latter pairs, theremust obtain a difference, thoughperhaps
only a subliminalone. And we would dismiss the possibilityof empiricaldiscon-
firmation,and indeed the idea that an empiricaltest should be relevanthere,on
the groundsthat identityis a transitiverelationby virtue of its definitionor by
virtue of the basic postulates governingit [3]. Hence, the principlein question
is truea priori.
5. Mathematics as an axiomatized deductive system. I have argued so far
that the validity of mathematicsrestsneitheron its alleged self-evidentialchar-
acter nor on any empiricalbasis, but derives fromthe stipulationswhich deter-
mine the meaning of the mathematical concepts, and that the propositionsof
mathematics are thereforeessentially "true by definition."This latter state-
ment, however,is obviously oversimplifiedand needs restatementand a more
careful justification.
For the rigorousdevelopmentof a mathematicaltheoryproceeds not simply
froma set of definitionsbut rather froma set of non-definitionalpropositions
whichare not proved withinthe theory;these are the postulates or axioms of the
theory [4]. They are formulatedin termsof certain basic or primitiveconcepts
forwhich no definitionsare providedwithinthe theory.It is sometimesasserted
that the postulates themselvesrepresent"implicitdefinitions"of the primitive
terms. Such a characterizationof the postulates, however, is misleading. For
while the postulates do limit,in a specificsense, the meaningsthat can possibly
be ascribed to the primitives,any self-consistentpostulate systemadmits,never-
theless, many differentinterpretationsof the primitiveterms (this will soon be
illustrated),whereasa set ofdefinitionsin the strictsense ofthe worddetermines
the meaningsof the definiendain a unique fashion.
Once the primitivetermsand the postulates have been laid down, the entire
theoryis completelydetermined;it is derivable fromits postulational basis in
546 ON THE NATURE OF MATHEMATICAL TRUTH [December,

the followingsense: Every termof the theoryis definablein termsof the primi-
tives, and every propositionof the theoryis logically deducible fromthe postu-
lates. To be entirelyprecise,it is necessaryalso to specifythe principlesof logic
whichare to be used in the proofof the propositions,i.e. in theirdeductionfrom
the postulates. These principlescan be stated quite explicitly.They fall into
two groups: Primitivesentences, or postulates, of logic (such as: If p and q is
the case, then p is the case), and rules of deduction or inference(including,for
example, the familiar modus ponens rule and the rules of substitutionwhich
make it possible to infer,froma general proposition,any one of its substitution
instances). A more detailed discussion of the structureand content of logic
would, however,lead too farafieldin the contextof this article.
6. Peano's axiom systemas a basis formathematics.Let us now considera
postulate system fromwhich the entire arithmeticof the natural numberscan
be derived. This systemwas devised by the Italian mathematicianand logician
G. Peano (1858-1932). The primitivesof this system are the terms "0," 'num-
ber," and "successor." While, of course, no definitionof these terms is given
withinthe theory,the symbol "0" is intended to designate the number 0 in its
usual meaning,while the term "number" is meant to referto the natural num-
bers 0, 1, 2, 3 - * * exclusively. By the successor of a natural numbern, which
will sometimes brieflybe called n', is meant the natural number immediately
followingn in the natural order. Peano's systemcontains the following5 postu-
lates:
P1. 0 is a number
P2. The successor of any numberis a number
P3. No two numbershave the same successor
P4. 0 is not the successor of any number
PS. If P is a propertysuch that (a) 0 has the propertyP, and (b) whenevera
numbern has the propertyP, then the successor of n also has the property
P, then every number has the propertyP.
The last postulate embodies the principle of mathematical induction and
illustratesin a very obvious mannerthe enforcementof a mathematical "truth"
by stipulation. The constructionof elementaryarithmeticon this basis begins
with the definitionof the various natural numbers. 1 is definedas the succes-
sor of 0, or brieflyas 0'; 2 as 1', 3 as 2', and so on. By virtue of P2, this process
can be continuedindefinitely;because of P3 (in combinationwith P5), it never
leads back to one of the numbers previouslydefined,and in view of P4, it does
not lead back to 0 either.
As the next step, we can set up a definitionof addition which expresses in a
precise formthe idea that the addition of any natural number to some given
number may be considered as a repeated addition of 1; the latter operation is
readilyexpressibleby means of the successor relation.This definitionofaddition
runs as follows:
1945] ON THE NATURE OF MATHEMATICAL TRUTH 547

Dl. (a) n+O=n; (b) n+k'=(n+k)'.

The two stipulationsof this recursivedefinitioncompletelydeterminethe sum
of any two integers. Consider, for example, the sum 3+2. According to the
definitionsof the numbers2 and 1, we have 3+2=3+1'=3+(0')'; by DI (b),
3+ (0') = (3+0')' = ((3+0)')'; but by D1 (a), and by the definitionsofthe num-
bers 4 and 5, ((3 +0)')' = (3')' =4'=5. This proofalso rendersmoreexplicitand
precise the commentsmade earlierin this paper on the truthof the proposition
that 3 + 2=5: Withinthe Peano systemof arithmetic,its truthflowsnot merely
fromthe definitionof the concepts involved, but also fromthe postulates that
govern these various concepts. (In our specificexample, the postulates P1 and
P2 are presupposedto guarantee that 1, 2, 3, 4, 5 are numbersin Peano's system;
the generalproofthat Dl determinesthe sum ofany two numbersalso makes use
of P5.) If we call the postulates and definitionsof an axiomatized theory the
"stipulations"concerningthe concepts ofthat theory,thenwe may say now that
the propositionsof the arithmeticof the natural numbersare true by virtue of
the stipulations which have been laid down initiallyfor the arithmeticalcon-
cepts. (Note, incidentally,that our proofof the formula "3+2 = 5" repeatedly
made use of the transitivityof identity;the latteris accepted here as one of the
rules of logic which may be used in the proofof any arithmeticaltheorem;it is,
therefore,included among Peano's postulates no more than any other principle
of logic.)
Now, the multiplicationof natural numbersmay be definedby means of the
followingrecursivedefinition,whichexpressesin a rigorousformthe idea that a
productnkof two integersmay be consideredas the sum ofk termseach ofwhich
equals n.
D2. (a) n 0=0; (b) n*k'=n*k+n.
It now is possible to prove the familiargeneral laws governingaddition and
multiplication,such as the commutative, associative, and distributive laws
(n+k=k+n, n*k=k n; n+(k+l)=(n+k)+l, n.(k.l)=(n-k).l; n.(k+l)
= (n. k) + (n l)).-In terms of addition and multiplication,the inverse opera-
.

tions of subtractionand divisioncan thenbe defined.But it turnsout that these

"cannot always be performed";i.e., in contradistinctionto the sum and the
product, the differenceand the quotient are not defined for every couple of
numbers; forexample, r- 10 and 7 10 are undefined.This situation suggests
an enlargementof the number system by the introductionof negative and of
rational numbers.
It is sometimes held that in order to effectthis enlargement,we have to
"assume"yor else to "postulate" the existence of the desired additional kinds of
numberswith propertiesthat make them fitto fillthe gaps of subtractionand
division. This method of simply postulatingwhat we want has its advantages;
but, as BertrandRussell [5 ] puts it, theyare the same as the advantages of theft
over honest toil; and it is a remarkablefact that the negative as well as the ra-
548 ON THE NATURE OF MATHEMATICAL TRUTH [December,

tional numbers can be obtained fromPeano's primitivesby the honest toil of

constructingexplicit definitionsforthem,without the introductionof any new
postulates or assumptionswhatsoever. Every positive and negative integer-in
contradistinctionto a natural numberwhich has no sign-is definableas a cer-
tain set of orderedcouples of natural numbers;thus, the integer+2 is definable
as the set of all orderedcouples (m, n) of natural numberswherem =n+2; the
integer -2 is the set of all ordered couples (m, n) of natural numbers with
n ==m+2.-Similarly, rational numbersare definedas classes of orderedcouples
of integers.-The various arithmeticaloperationscan thenbe definedwithrefer-
ence to these new types of numbers,and the validity of all the arithmeticallaws
governingthese operationscan be provedby virtueofnothingmorethan Peano's
postulates and the definitionsof the various arithmeticalconcepts involved.
The much broader systemthus obtained is still incompletein the sense that
not every number in it has a square root, and more generally,not every alge-
braic equation whose coefficientsare all numbersof the system has a solution
in the system. This suggests furtherexpansions of the number system by the
introductionof real and finallyof complex numbers.Again, this enormous ex-
tension can be effectedby mere definition,without the introductionof a single
new postulate [6]. On the basis thus obtained, the various arithmeticaland alge-
braic operationscan be definedfor.the numbersof the new system,the concepts
of function,of limit,ofderivative and integralcan be introduced,and the famil-
iar theoremspertainingto these concepts can be proved,so that finallythe huge
system of mathematics as here delimited rests on the narrowbasis of Peano's
system: Every concept of mathematics can be defined by means of Peano's
three primitives,and every proposition of mathematics can be deduced from
the fivepostulates enrichedby the definitionsof the non-primitiveterms [6a].
These deductions can be carried out, in most cases, by means of nothingmore
than the principlesof formallogic; the proofof some theoremsconcerningreal
numbers,however,requiresone assumptionwhichis not usually includedamong
the latter. This is the so,-calledaxiom of choice. It asserts that given a class of
mutuallyexclusiveclasses, none of whichis empty,thereexistsat least one class
which has exactly one element in common with each of the given classes. By
virtue of this principleand the rules of formallogic, the contentof all of mathe-
matics can thus be derived fromPeano's modest system-a remarkableachieve-
mentin systematizingthe contentof mathematicsand clarifyingthe foundations
of its validity.
7. Interpretationsof Peano's primitives.As a consequence of this result,the
whole systemof mathematicsmightbe said to be true by virtue of mere defini-
tions (namely, of the non-primitivemathematical terms) provided that the five
Peano postulates are true. However, strictlyspeaking, we cannot, at this junc-
ture,referto the Peano postulates as propositionswhich are eithertrue or false,
fortheycontain threeprimitivetermswhichhave not been assigned any specific
meaning.All we can assert so faris that any specificinterpretationof the primi-
1945] ON THE NATURE OF MATHEMATICAL TRUTH 549

tives which satisfiesthe fivepostulates-i.e., turnsthem into true statements-

will also satisfyall the theoremsdeduced fromthem. But for Peano's system,
there are several-indeed, infinitelymany-interpretations which will do this.
For example, let us understandby 0 the originof a half-line,by the successor of
a point on that half-linethe point 1 cm. behind it, counting fromthe origin,
and by a numberany point which is eitherthe originor can be reached fromit
by a finitesuccessionof steps each ofwhichleads fromone point to its successor.
It can then readily be seen that all the Peano postulates as well as the ensuing
theoremsturn into true propositions,although the interpretationgiven to the
primitivesis certainly not the customary one, which was mentioned earlier.
More generally,it can be shown that every progressionof elements of any kind
providesa true interpretation,or a "model," of the Peano system.This example
illustratesour earlier observation that a postulate system cannot be regarded
as a set of "implicitdefinitions"forthe primitiveterms:The Peano systemper-
mits of many differentintetpretations,whereas in everyday as well as in sci-
entificlanguage, we attach one specificmeaning to the concepts of arithmetic.
Thus, e.g., in scientificand in everyday discourse, the concept 2 is understood
in such a way that fromthe statement "Mr. Brown as well as Mr. Cope, but no
one else is in the office,and Mr. Brown is not the same personas Mr. Cope," the
conclusion "Exactly two personsare in the office"may be validly inferred.But
the stipulations laid down in Peano's system for the natural numbers,and for
the number 2 in particular,do not enable us to draw this conclusion; they do
not "implicitlydetermine" the customary meaning of the concept 2 or of the
other arithmeticalconcepts. And the mathematician cannot acquiesce at this
deficiencyby arguing that he is not concerned with the customarymeaning of
the mathematicalconcepts; forin proving,say, that every positive real number
has exactly two real square roots,he is himselfusing the concept 2 in its custom-
ary meaning,and his very theoremcannot be proved unless we presuppose more
about the number 2 than is stipulated in the Peano system.
If thereforemathematicsis to be a correcttheoryof the mathematical con-
cepts in their intended meaning, it is not sufficientfor its validation to have
shown that the entiresystem is derivable fromthe Peano postulates plus suit-
able definitions;rather,we have to inquirefurtherwhetherthe Peano postulates
are actually true when the primitivesare understoodin theircustomarymean-
ing. This question, of course,can be answeredonly afterthe customarymeaning
of the terms '0," "natural number," and "successor" has been clearly defined.
To this task we now turn.
8. Definition of the customary meaning of the concepts of arithmeticin
purelylogical terms. At firstblush, it mightseem a hopeless undertakingto try
to definethese basic arithmeticalconcepts withoutpresupposingothertermsof
arithmetic,which would involve us in a circular procedure. However, quite
rigorousdefinitionsof the desired kind can indeed be formulated,and it can be
shown that forthe concepts so defined,all Peano postulates turninto true state-
550 ON THE NATURE OF MATHEMATICAL TRUTH [December,

ments. This important result is due to the research of the German-logician

G. Frege (1848-1925) and to the subsequent systematic and detailed work of
the contemporary English logicians and philosophers B. Russell and A. N.
Whitehead. Let us consider brieflythe basic ideas underlyingthese definitions
[7].
A natural number-or, in Peano's term,a number-in its customarymean-
ing can be consideredas a characteristicof certain classes of objects. Thus, e.g.,
the class of the apostles has the number 12, the class of the Dionne quintuplets
the number5, any couple the number2, and so on. Let us now expressprecisely
the meaningof the assertion that a certain class C has the number2, or briefly,
that n(C) = 2. Briefreflectionwill show that the followingdefiniensis adequate
in the sense of the customarymeaning of the concept 2: There is some object x
and some object y such that (1) xeC (i.e., x is an element of C) and yEC,
(2) x$y, and (3) if z is any object such that zEC, then either z=x or z=y.
(Note that on the basis of this definitionit becomes indeed possible to inferthe
statement "The numberof persons in the officeis 2" from"Mr. Brown as well
as Mr. Cope, but no one else is in the office,and Mr. Brown is not identicalwith
Mr. Cope"; C is herethe class ofpersonsin the office.)Analogously,the meaning
of the statement that n(C) =1 can be definedthus: There is some x such that
xEC, and any object y such that yXC,is identical with x. Similarly, the cus-
tomarymeaning of the statementthat n(C) =0 is this: There is no object such
that xeC.
The general pattern of these definitionsclearly lends itselfto the definition
of any natural number. Let us note especially that in the definitionsthus ob-
tained, the definiensnever contains any arithmeticalterm,but merelyexpres-
sions taken fromthe fieldof formal logic, includingthe signs of identity and
difference.So far, we have defined only the meaning of such. phrases as
"n(C) =2," but we have given no definitionforthe numbers0, 1, 2, - * * apart
fromthis context.This desideratumcan be met on the basis of the consideration
that 2 is that propertywhich is common to all couples, i.e., to all classes C such
that n(C) =2. This common propertymay be conceptually representedby the
class of all those classes which share this property.Thus we arrive at the defini-
tion: 2 is the class of all couples, i.e., the class of all classes C for which
n(C) = 2.-This definitionis by no means circular because the concept of
couple-in other words, the meaning of "n(C) = 2"-has been previously de-
finedwithout any referenceto the number 2. Analogously, 1 is the class of all
unit classes, i.e., the class ofall classes C forwhichn(C) = 1. Finally, 0 is the class
of all null classes, i.e., the class of all classes withoutelements. And as there is
only one such class, 0 is simply the class whose only element is the null class.
Clearly, the customarymeaning of any given natural numbercan be definedin
this fashion [8]. In orderto characterizethe intendedinterpretationof Peano's
primitives,we actually need, of all the definitionshere referredto, only that of
the number0. It remains to definethe terms "successor" and "integer."
The definitionof "successor," whose precise formulationinvolves too many
1945] ON THE NATURE OF MATHEMATICAL TRUTH 551

niceties to be stated here,is a carefulexpressionof a simple idea which is illus-

trated by the followingexample: Consider the number 5, i.e., the class of all
quintuplets. Let us select an arbitraryone of these quintupletsand add to it an
object which is not yet one of its members.5', the successor of 5, may then be
definedas the numberapplying to the set thus obtained (which,of course, is a
sextuplet). Finally,it is possible to formulatea definitionof the customarymean-
ing of the concept of natural number; this definition,which again cannot be
given here,expresses,in a rigorousform,the idea that the class of the natural
numbersconsistsof the number0, its successor,the successor of that successor,
and so on.
If the definitionshere characterizedare carefullywrittenout-this is one of
the cases where the techniques of symbolic,or mathematical,logic prove indis-
pensable-it is seen that the definiensof every one of them contains exclusively
termsfromthe fieldof pure logic. In fact, it is possible to state the customary
interpretationof Peano's primitives,and thus also the meaningofevery concept
definableby means of them-and that includes everyconcept of mathematics-
in termsof the following7 expressions,in addition to variables such as "x" and
"C": not,and, if-then; for everyobjectx it is the case that . . ; thereis some
objectx such that ... ; x is an elementof class C; theclass of all thingsx such
that.... And it is even possible to reduce the numberof logical concepts needed
to a mere four: The firstthree of the concepts just mentionedare all definable
in termsof "neither-nor,"and the fifthis definableby means of the fourthand
"neither-nor." Thus, all the concepts of mathematicsprove definablein terms
of fourconcepts of pure logic. (The definitionof one of the more complex con-
cepts of mathematics in terms of the fourprimitivesjust mentionedmay well
fillhundredsor even thousands of pages; but clearly this affectsin no way the
theoreticalimportanceof the result just obtained; it does, however,show the
great convenienceand indeed practical indispensabilityformathematicsof hav-
ing a large system of highlycomplex definedconcepts available.)

9. The truthof Peano's postulates in their customaryinterpretation.The

definitionscharacterizedin the precedingsection may be said to renderprecise
and explicit the customarymeaning of the concepts of arithmetic.Moreover-
and this is crucial for the question of the validity of mathematics-it can be
shown that the Peano postulates all turninto true propositionsif the primitives
are construed in accordance with the definitionsjust considered.
Thus, P1 (O is a number) is true because the class of all numbers-i.e., natu-
ral numbers-was definedas consistingof 0 and all its successors. The truthof
P2 (The successor of any numberis a number) followsfromthe same definition.
This is true also of P5, the principleof mathematical induction.To prove this,
however,we would have to resortto the precise definitionof "integer" rather
than the loose descriptiongiven of that definitionabove. P4 (O is not the succes-
sor of any number) is seen to be true as follows: By virtue of the definitionof
"successor,' a number which is a successor of some number can apply only to
552 ON THE NATURE OF MATHEMATICAL TRUTH [December,

classes which contain at least one element; but the number 0, by definition,
applies to a class ifand only ifthat class is empty.-While the truth of PI, P2,
P4, P5 can be inferredfromthe above definitionssimplyby means of the prin-
ciples of logic, the proofof P3 (No two numbershave the same successor) pre-
sents a certain difficulty.As was mentionedin the precedingsection,the defini-
tion of the successor of a numbern is based on the process of adding, to a class
of n elements,one element not yet contained in that class. Now if thereshould
exist only a finitenumber of thingsaltogether then this process could not be
continued indefinitely,and P3, which (in conjunction with P1 and P2) implies
that the integersforman infiniteset, would be false. Russell's way of meeting
this difficulty[9] was to introducea special "axiom of infinity,"which stipu-
lates, in effect,the existence of infinitelymany objects and thus makes P3
demonstrable.The axiom of infinitycan be formulatedin purely logical terms
and may thereforebe considered as a postulate of logic; however, it certainly
does not belong to the generallyrecognizedprinciplesof logic; and it thus intro-
duces a foreignelement into the otherwiseunexceptionable derivation of the
Peano postulates frompure logic. Recently, however, it has been shown [to]
that a suitable systemof logical principlescan be set up which is even less com-
prehensivethan the rules of logic which are commonlyused [It], and in which
the existence of infinitelymany objects can be proved without the need fora
special axiom.

10. Mathematics as a branch of logic. As was pointed out earlier,all the the-
oremsof arithmetic,algebra, and analysis can be deduced fromthe Peano postu-
lates and the definitionsof those mathematical termswhich are not primitives
in Peano's system. This deduction requires only the principlesof logic plus, in
certain cases, the axiom of choice. By combiningthis result with what has just
been said about the Peano system,the followingconclusion is obtained, which
is also known as thethesisof logicismconcerningthenatureof mathematics:
Mathematics is a branch of logic. It can be derived fromlogic in the follow-
ing sense:
a. All the concepts of mathematics,i.e. of arithmetic,algebra, and analysis,
can be definedin terms of four concepts of pure logic.
b. All the theoremsof mathematics can be deduced fromthose definitions
by means of the principlesof logic (includingthe axiom of choice).
In this sense it can be said that the propositionsof the system of mathe-
matics as heredelimitedare trueby virtueof the definitionsof the mathematical
concepts involved,or that they make explicitcertain characteristicswith which
we have endowed our mathematicalconcepts by definition.The propositionsof
mathematicshave, therefore,the same unquestionablecertaintywhichis typical
of such propositionsas "All bachelors are unmarried,"but they also share the
complete lack of empiricalcontentwhich is associated with that certainty:The
propositionsof mathematics are devoid of all factual content; they convey no
informationwhatever on any empiricalsubject matter.
1945] ON THE NATURE OF MATHEMATICAL TRUTH 553

to empiricalsubjectmatter.This
of mathematics
11. On the applicability
result seems to be irreconcilablewith the fact that after all mathematics has
proved to be eminentlyapplicable to empiricalsubject matter,and that indeed
the greater part of present-day scientificknowledge has been reached only
through continual reliance on and application of the propositions of mathe-
matics.-Let us try to clarifythis apparent paradox by referenceto some ex-
amples.
Suppose that we are examininga certain amount of some gas, whose volume
v, at a certainfixedtemperature,is found to be 9 cubic feetwhen the pressurep
is 4 atmospheres.And let us assume furtherthat the volume of the gas forthe
same temperatureand p =6 at., is predicted by means of Boyle's law. Using
elementaryarithmeticwe reason thus: For correspondingvalues of v and p,
vp= c, and v = 9 when p = 4; hence c = 36: Therefore,when p = 6, then v= 6. Sup-
pose that this predictionis borne out by subsequent test. Does that show that
the arithmeticused has a predictivepower of its own, that its propositionshave
factual implications? Certainlynot. All the predictivepower here deployed, all
the empirical content exhibited stems fromthe initial data and fromBoyle's
law, which asserts that vp= c forany two correspondingvalues of v and p, hence
also forv = 9, p = 4, and forp = 6 and the correspondingvalue ofv [12 ]. The func-
tion of the mathematicshere applied is not predictiveat all; rather,it is analytic
or explicative: it rendersexplicit certain assumptions or assertions which are
included in the contentof the premisesof the argument (in our case, these con-
sist of Boyle's law plus the additional data); mathematical reasoning reveals
that those premises contain-hidden in them, as it were,-an assertion about
the case as yet unobserved. In accepting our premises-so arithmeticreveals-
we have-knowingly or unknowingly-already accepted the implication that
the p-value in question is 6. Mathematical as well as logical reasoningis a con-
ceptual technique of making explicit what is implicitlycontained in a set of
premises. The conclusions to which this technique leads assert nothingthat is
theoreticallynew in the sense of not being contained in the content of the
premises. But the resultsobtained may well be psychologically new: we may not
have been aware, beforeusing the techniquesof logic and mathematics,what we
committedourselves to in accepting a certain set of assumptions or assertions.
A similar analysis is possible in all other cases of applied mathematics,in-
cluding those involving,say, the calculus. Consider,forexample, the hypothesis
that a certainobject, movingin a specifiedelectricfield,will undergoa constant
acceleration of 5 feet/sec2.For the purpose of testingthis hypothesis,we might
derivefromit, by means of two successive integrations,the predictionthat ifthe
object is at rest at the beginningof the motion,then the distance covered by it
at any time t is 4t2feet. This conclusion may clearlybe psychologicallynew to
a person not acquainted with the subject, but it is not theoreticallynew; the
contentof the conclusionis already contained in that of the hypothesisabout the
constantacceleration.And indeed, here as well as in the case of the compression
of a gas, a failureof the predictionto come true would be consideredas indica-
554 ON THE NATURE OF MATHEMATICAL TRUTH [December,

tive of the factual incorrectnessof at least one of the premisesinvolved (f.ex., of

Boyle's law in its application to the particulargas), but never as a sign that the
logical and mathematical principlesinvolved mightbe unsound.
Thus, in the establishmentof empiricalknowledge,mathematics (as well as
logic) has, so to speak, the functionof a theoretical juice extractor: the tech-
niques of mathematical and logical theorycan produce no morejuice of factual
informationthan is contained in the assumptionsto which theyare applied; but
they may produce a great deal morejuice of this kind than mighthave been an-
ticipated upon a firstintuitiveinspectionof those assumptions which formthe
raw material forthe extractor.
At thispoint,it may be well to considerbrieflythe status of those mathemat-
ical disciplines which are not outgrowthsof arithmeticand thus of logic; these
include in particulartopology,geometry,and the various branches of abstract
algebra, such as the theoryof groups, lattices, fields,etc. Each of these disci-
plines can be developed as a purelydeductive systemon the basis of a suitable
set of postulates. If P be the conjunction of the postulates fora given theory,
thenthe proofof a propositionT of that theoryconsistsin deducing T fromP by
means of the principlesof formallogic. What is established by the proofis there-
forenot the truthof T, but ratherthe fact that T is trueprovided that the postu-
lates are. But since both P and T contain certain primitivetermsof the theory,
to whichno specificmeaningis assigned, it is not strictlypossible to speak of the
truthof eitherP or T; it is thereforemoreadequate to state the point as follows:
If a propositionT is logicallydeduced fromP, then every specificinterpretation
of the primitiveswhich turns all the postulates of P into true statements,will
also renderT a true statement.-Up to this point,the analysis is exactly analo-
gous to that of arithmeticas based on Peano's set of postulates. In the case of
arithmetic,however,it proved possible to go a step further,namely to definethe
customarymeaningsof the primitivesin termsof purelylogical concepts and to
show that the postulates-and thereforealso the theorems-of arithmeticare
unconditionallytrue by virtue of these definitions.An analogous procedure is
not applicable to those disciplineswhich are not outgrowthsof arithmetic:The
primitivesof the various branches of abstract algebra have no specific"custom-
ary meaning"; and ifgeometryin its customaryinterpretationis thoughtof as a
theory of the structureof physical space, then its primitiveshave to be con-
strued as referringto certain types of physical entities,and the question of the
truthof a geometricaltheoryin this interpretationturnsinto an empiricalprob-
lem [13]. For the purpose of applying any one of these non-arithmeticaldisci-
plines to some specificfieldof mathematics or empirical science, it is therefore
necessary firstto assign to the primitivessome specificmeaning and then to
ascertain whetherin this interpretationthe postulates turninto truestatements.
If this is the case, then we can be sure that all the theoremsare true statements
too, because they are logically derived fromthe postulates and thus simplyex-
plicate the contentof the latterin the given interpretation.-In theirapplication
to empiricalsubject matter,therefore,these mathematical theoriesno less than
1945] ON THE NATURE OF MATHEMATICAL TRUTH 555

those which grow out of arithmeticand ultimatelyout of pure logic, have the
functionof an analytic tool, which brings to light the implicationsof a given
set of assumptions but adds nothingto theircontent.
But while mathematicsin no case contributesanythingto the contentofour
knowledgeof empiricalmatters,it is entirelyindispensableas an instrumentfor
the validation and even for the linguisticexpression of such knowledge: The
majorityof the morefar-reachingtheoriesin empiricalscience-including those
whichlend themselvesmosteminentlyto predictionor to practicalapplication-
are stated withthe help ofmathematicalconcepts; the formulationofthese theo-
ries makes use, in particular,of the numbersystem,and of functionalrelation-
ships among different metricalvariables. Furthermore,the scientifictest ofthese
theories,the establishmentof predictionsby means of them, and finallytheir
practical application, all require the deduction,fromthe general theory,of cer-
tain specific consequences; and such deduction would be entirelyimpossible
without the techniques of mathematics which reveal what the given general
theoryimplicitlyasserts about a certain special case.
Thus, the analysis outlined on these pages exhibits the system of mathe-
matics as a vast and ingeniousconceptual structurewithout empirical content
and yet an indispensable and powerfultheoreticalinstrumentforthe scientific
understandingand masteryof the worldof our experience.

References
1. A discussionofthestatusofgeometry is givenin myarticle,Geometry and EmpiricalSci-
ence,AmericanMathematicalMonthly,vol. 52, pp. 7-17, 1945.
2. The objectionis sometimes raisedthatwithoutcertaintypesofexperience, suchas encoun-
teringseveralobjectsof the same kind,the integersand the arithmetical operationswiththem
wouldneverhave beeninvented, and thattherefore thepropositionsofarithmeticdo havean em-
piricalbasis.This typeofargument, however, involvesa confusion ofthelogicaland thepsychologi-
ical meaningof the term"basis.' It mayverywellbe the case thatcertainexperiences occasion
psychologically theformation ofarithmetical ideas and in thissenseforman empirical"basis"for
them;but thispointis entirely irrelevantforthelogicalquestionsas to thegrounds on whichthe
propositions of arithmetic may be acceptedas true.The pointmade above is thatno empirical
"basis"orevidencewhateveris neededto establishthetruthofthepropositions ofarithmetic.
3. A preciseaccountof thedefinition and theessentialcharacteristicsoftheidentity relation
maybe foundin A. Tarski,Introduction to Logic,New York,1941,Ch. III.
4. For a lucidand conciseaccountoftheaxiomaticmethod,see A. Tarski,I.c., Ch. VI.
5. BertrandRussell,Introduction to MathematicalPhilosophy, NewYorkand London,1919,
p. 71.
6. For a moredetailedaccountoftheconstruction ofthenumbersystemon Peano'sbasis,cf.
BertrandRussell,l.c.,esp. Chs. I and VII.-A rigorous and concisepresentationofthatconstruc-
tion,beginning, however, withthesetofall integers ratherthanthatofthenaturalnumbers, may
be foundin G. Birkhoff and S. MacLane,A SurveyofModernAlgebra,NewYork1941,Chs. I, II,
III, V.-For a generalsurveyof the construction of the numbersystem,cf.also J. W. Young,
Lectureson theFundamentalConceptsofAlgebraand Geometry, New York, 1911,esp. lectures
X, XI, XII.
6a. As a resultofverydeep-reaching carriedout by K. Godelit is knownthat
investigations
arithmetic,and a fortiori
mathematics,
is an incompletetheoryin the followingsense:Whileall
thosepropositions whichbelongto theclassicalsystemsofarithmetic,algebra,and analysiscan
556 ON THE NATURE OF MATHEMATICAL TRUTH

indeedbe derived,in thesensecharacterized above,fromthePeano postulates,thereexistnever-

thelessotherpropositions whichcan be expressedin purelyarithmetical terms,and whichare
true,but whichcannotbe derivedfromthePeano system.Andmoregenerally:For any postulate
systemof arithmetic(or ofmathematics forthat matter)whichis not self-contradictory, there
existpropositions whichare true,and whichcan be statedin purelyarithmetical terms,but which
cannotbe derivedfromthat postulatesystem.In otherwords,it is impossibleto constructa
postulatesystemwhichis not self-contradictory, and whichcontainsamongits consequencesall
truepropositions whichcan be formulated withinthelanguageofarithmetic.
This factdoes not,however,affectthe resultoutlinedabove, namely,thatit is possibleto
deduce,fromthePeano postulatesand theadditionaldefinitions ofnon-primitive terms,all those
propositions whichconstitute theclassicaltheoryofarithmetic, algebra,and analysis;and it is to
thesepropositions thatI referabove and subsequently as thepropositionsofmathematics.
7. For a moredetaileddiscussion,cf.Russell,l.c.,Chs. II, III, IV. A completetechnicalde-
velopmentof the idea can be foundin the greatstandardworkin mathematical logic,A. N.
Whiteheadand B. Russell,PrincipiaMathematica,Cambridge, England,1910-1913.-Fora very
preciserecentdevelopment of the theory,see W. V. 0. Quine,MathematicalLogic, New York
1940.-A specificdiscussionof the Peano systemand its interpretations fromthe viewpointof
semanticsis includedin R. Carnap,Foundationsof Logic and Mathematics,International En-
cyclopediaofUnifiedScience,vol. I, no. 3, Chicago,1939;especiallysections14, 17, 18.
8. The assertionthatthedefinitions givenabove statethe"customary" meaningofthearith-
meticaltermsinvolvedis to be understood in thelogical,not thepsychological senseof theterm
"meaning."It wouldobviouslybe absurdto claimthattheabove definitions express"whatevery-
bodyhas in mind"whentalkingabout numbersand thevariousoperationsthatcan be performed
withthem.Whatis achievedby thosedefinitions is rathera "logicalreconstruction"oftheconcepts
ofarithmetic in thesensethatifthedefinitions are accepted,thenthosestatements in scienceand
everydaydiscoursewhichinvolvearithmetical termscan be interpreted coherently and systemati-
callyin sucha mannerthattheyare capableofobjectivevalidation.The statement about thetwo
personsin theoffice providesa veryelementary illustrationofwhatis meanthere.
9. Cf.BertrandRussell,I.c.,p. 24 and Ch. XIII.
10. This resulthas beenobtainedby W. V. 0. Quine;cf.his MathematicalLogic,New York,
1940.
11. The principlesoflogicdevelopedin Quine'sworkand in similarmodernsystemsofformal
logicembodycertainrestrictions as comparedwiththoselogicalruleswhichhad beenrathergen-
erallyacceptedas sounduntilabouttheturnofthe20thcentury. At thattime,thediscovery ofthe
famousparadoxesoflogic,especiallyof Russell'sparadox(cf.Russell,I.c.,Ch. XIII) revealedthe
factthatthelogicalprinciples implicitin customarymathematical reasoninginvolvedcontradic-
tionsand therefore had to be curtailedin one manneror another.
12. Note thatwe maysay "hence"by virtueof theruleof substitution, whichis one of the
rulesoflogicalinference.
13. For a moredetaileddiscussionofthispoint,cf.thearticlementioned in reference 1.