Sie sind auf Seite 1von 14

Set-Theoretic Foundations for Logic Author(s): W. V. Quine Reviewed work(s): Source: The Journal of Symbolic Logic, Vol. 1, No.

2 (Jun., 1936), pp. 45-57 Published by: Association for Symbolic Logic Stable URL: http://www.jstor.org/stable/2268548 . Accessed: 10/10/2012 01:32
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.

Association for Symbolic Logic is collaborating with JSTOR to digitize, preserve and extend access to The Journal of Symbolic Logic.

http://www.jstor.org

T Hz JOURNAL OF SyYBouc LOcIC Volume 1, Number 2, June 1936

SET-THEORETIC

FOUNDATIONS FOR LOGIC


W. V. QUINE'

uses the variables"x", "y", etc. Zermelo2 In 1. Introduction. his set theory Amongthese thingshe includes of forthe representation "things" generally. "e" classes.He adopts theconnective of memsets, as I shall say henceforth, or, of thus the elementary as bership his sole special primitive; formulae his system "xey",withany thing-variables of as simply expressions theform are describable of are "x" "tx",''y'', ''z'', etc. supplanting and "y". The postulates his system so use of as fashioned to avoid the logicalparadoxeswithout of the theory types. may One of the postulates,the so-called Aussonderungsaxiom, be stated in logicalnotationas familiar
"( x) (yM((ye) ((y(-z).eY(y)W)

in about y whichis definite a ceras where"e(y)" is understood any statement for informally the purpose. Skolem has tain sense whichZermelointroduces as statements empointedout that it is adequate here to construe"definite" by and all formulae thenceconstructible formulae bracingjust the elementary A with respectto thing-variables.3 and by quantification the truthfunctions this asserts second of Zermelo'spostulatesis the principleof extensionality; of i.e. inclusiveclasses are identical, are members just the same that mutually of postulateswhichprovideforthe existence the null classes.There are further of class, the class of all subclassesof any givenclass, the class of all members and of members anygivenclass,theunitclassofanygiventhing, theclass whose axiom (Ausare sole members any two giventhings.Finallythe multiplicative are adopted. and wahlprinzip) the axiomof infinity the of In the expression Zermelo'spostulatesand the proofof his theorems as sublogic are presupposed an implicit notionsand techniqueof elementary are The onlylogicalnotionswhichare hererequired the truthfuncstructure. and with respectto thing-variables, the only logical tions and quantification and To functions quantification. the is required thatgoverning truth technique in then,we must system its ownright, as system a deductive the frame Zermelo to to extendits foundations includethispartoflogic,in application theelemenfuncand This can be done by adoptingquantification the truth taryformulae. and the of primitives, providing tions(or an adequate selection these)as further of with rules of inference such kind as to provide all the deductive system and functions quantification. of technique truth of It is usual, in the fullpresentation a deductivesystem,to begin metaReceivedJuly17, 1936. l Societyof Fellows,Harvard University, 2 Zermelo, I, der Grundlagen Mengenlehre Mathematische Annalen,vol.65 (1908),pp. 261-281. in 3 See Skolem,Einige Bemerkungen der Abhandlung vonE. Zermdo: "Ober die Definitheit zu Fundamenta mathematicae, vol. 15 (1930), pp. 337-341. derAxiomatik",

45

46

W. V. QUINE

mathematically by specifyingwhat expressionsare to be formulaeof the system; then to use this notion of formulain a metamathematical account of the rules of inference the system. In the present case, since the primitivesare just memberof ship, quantification,and the truthfunctions,clearly the formulae of the system will embrace just the "definite" statements as construed by Skolem. As to the one will be the familiarlogical rule which allows the inference rules of inference, of a formula f as a theorem whenever the result of substituting f for "Q" and some theorem for "P" in "Pm Q" is a theorem; and there will be other such rules, of familiar kind, providing in the aggregate for the whole technique of truth functions and quantification. Moreover, since a "definite" statement is now simply any formula of the system, we can express the Aussonderungsaxiom which we may call the Aussonderungsitselfas another ordinaryrule of inference, regel: Each resultof puttingaformulafor "P" in "( Bx) (y) ((yCx) -z).P)

is a theorem.(Strictly, we should add explicitly that "x" must not he a free variable of the substituted formula.) The Aussonderungsaxiomthus ceases to depend on the notion of "definiteness" and acquires the same formal status as the logical rules of inference. Finally the rest of Zermelo's postulates can be translated into the primitive notation and retained as formal postulates. Zermelo's system is thus transformedinto a formal deductive system containing rules of inferenceand postulates. The purpose of this paper is the presentation of a system r which resembles the above system but is more economical. r has the same rules of inferenceas the above system,including the Aussonderungsregel; latter,however,is modithe fied to the extent of supplanting "yez" by the inclusion "y cz", which we may suppose definedin the familiarfashion. By way of formalpostulates, F contains only the extensionality principle; the described modificationof the Aussonderenables us to abandon all the other postulates and still derive the whole ungsregel of standard mathematical logic, as measured e.g. by P.M.4 This does not mean that r, so constituted, is adequate to the derivation of all Zermelo's postulates; it means that r is adequate to the derivation of as many consequences of Zermelo's postulates as are needed for standard logic in the adopted sense. In particular the multiplicative axiom and the axiom of infinity are as requisite to r as they are to Zermelo's system; I follow P.M.," however, in suppressing these as postulates and requiring their statement rather as explicit hypotheses wherevernecessary. There is a furtherrespect in which r is to be contrasted with Zermelo's system and others: it contains no null class. The null class presupposes a distinction between classes and individuals (non-classes) which is inexpressiblein terms of membership, since the null class is like an individual in lacking members. Repudiation of the null class enables us to explain the variables of r as representing objects generally,classes and otherwise,and then to defineclasses simply as objects y such that ( 3x) (xey).
4 Whitehead and Russell,Principia mathematica,2nd. edition.

'Vol. 2, p. 183; vol. 1, p. 481.

SET-THEORETIC

FOUNDATIONS

FOR

LOGIC

47

of Thoughr lacks two chieffeatures the standardlogic,viz. typesand the a definitionallyderivative nullclass,we shallsee thatwithin we can construct r A into stratification types theory in whichboth the null class and the familiar of are ostensibly restored and all theorems the standardlogicare forthcoming. of or of 2. Formal construction r. The r-formulae, formulae r, are describof able recursively thus: expressions the form"(xey)", with any r-variables "x" and "y", are r-formulae; also are the "x", "y", "z", etc. supplanting so resultsof putting any r-formulae "P" and "Q" and any r-variablefor"x" for in ' P)P", "(P. Q)", and "(x)P". A r-variableXis said to be boundat a given q point(say at thekthsign) of a r-formula ifthatpointlies witha P-formula p of occurrence X (as uniwhichforms part of p and beginswitha parenthesized of versalquantifier). occurrence Xis freein p ifx is not boundat thatoccurAn of rencein p. Xis said to be freein p, or to be a freeP-variable p, ifXhas a free occurrence p. in r-formulae abbreviated are the familiar definitions. through following D 1. D3.
(P.Q) =df - (PD-~Q)(aX)P =df - (X) - P-

D 2. (P -Q) D4.
(X Cy)

-,df((P 3 Q) -(Q3 P))


= d f(Z) ((ZL'X) D (2UY))

are Definitions viewed as conventions extraneousto the formalsystem.For formal of abbreviations expandedalways as purposeswe may think definitional intofullprimitive expression. techniqueof the propothe Amongvarious ways of compressing familiar of of sitionalcalculusand theory quantifiers a fewrulesforthe generation into in underany theorems r, one way is to declarethata r-formula is a r-theorem f conditions: ofthefollowing for Fl. fis formed putting by r-formulae thecapitalsin one oftheschemata (a) "((PD Q)n ((Qn R): (Pn R)))",
(c "( P 3P) 3P)"

(b)

"(P

D (-

P. Q))",

for of r2. The result putting for"Q" and somer-theorem "P" in "(P Q)" f is a -theorem. r r3. There are r-formulae and q and a r-theorem such that fand r are p from"(P n (x)Q)" and "(Pm Q)" by replacing"P" by p, formed respectively "Q" by q, and "x" by a r-variablewhichis not freein P. r r4. There are F-formulae and q and a P-theorem such that f and r are p from"(P. Q)" and "(P. (x)Q)" by replacing"P" by I, formed respectively "Q" by q, and "x" by a r-variable. of is formedfroma r-theorem replacingall freeoccurrences a r5. by r-variableS therein a r-variable whichis not boundat any of thoseoccurby p rencesofX. of systematizations thepropor1-2 are an adaptationofone ofLukasiewicz's which instances are all calculus.6 sitional Theyprovideas r-theorems r-formulae
6

p. Sci. et Let. de Varsovie, vol. 23 (1930), Classe III, p. 34 (reprint 6), Satz 6.

Comptes rendus Soc. fiber See Lukasiewiczand Tarski,Untersuchungen den Aussagenkalkid,

48

W. V. QUINE

of calculus.7 r2-5 answerto Tarski's definition of valid forms the propositional of calculus as Given the machinery the propositional of logical consequence.8 of all technique quantification. provided rl-2, r3-5 provide thefamiliar by in of As intimated ?1, the generaltheory deductionas embodiedin ri-5 in is supplemented r by one special postulateand one special rule as follows:
r6.
< ((x C y) D ((y cz x):D *fX) D ((xew) D (yew)))))'

underD4, of] a r-theorem. is [a transcription, in in by-a r-formula r7. If "P" is supplanted "(ax)(y)((yex)-((ycz).P))" the is underD1-3, of] a r-theorem. which "x" is notfree, result [a transcription, r6 to principle is inserted restrict The clause "(zex)" in the extensionality of theprinciple classesand thusleave roomforthe existence morethan one to individual. universal implication, quantifiAs framed above, r involvesdenial,material we as Actuallydenial is superfluous; could cation,and membership primitive. define thus: it
P = d f(PD (X) (y) (Xey)) .

in (a)-(c) could thenbe supplanted rl by "(((Pm Q) D ((Rn S) n T)) n ((Un ((Rn S) z T)) n ((P m U) m(S 3 T))))", implication.' whichanswersto Wajsberg'spostulateforthecalculusof material is equivalentto r. If on theotherhand it is The resulting system demonstrably of indethoughtdesirableto providefor the properties the truthfunctions as an of pendently thoseof membership, is done in r, thereremains alternative of stroke function instead namelytheadoptionof Sheffer's reduction primitives, and to of implication denial.'0r2 wouldthenbe refashioned matchNicod's rule of inferencesand (a)-(c) wouldbe supplantedin rl by

"'((PI(QIR))I((SI (SIS))I ((SIQ)I((PIS)I (PIS)))))",


form" Nicod's postulate.Throughout of to which answers Lukasiewicz'sreduced it to thispaper,however, will be convenient retainr in its originalredundant form. A with 3. The systemA. A system due to Tarski" willbe identified "standard of nullity logic" and shownto be derivablefromr. A, unliker, countenances kindofvariablesis of the classesand imports simpletheory types.A distinctive
7 See Quine, Truth Philosophical essays for A. N. Whitehead (0. H. Lee, edibyconvention, tor; New York, 1936),pp. 107-112,esp. note 18. undderw-Vollstandiguber der 8 Tarski,EinigeBetrachtungen die Begriffe c.-Widerspruchsfrcikeit Monatsheftefur Mathematik und Physik, vol. 40 (1933), p. 103, Def. 9. Note that"Fr(e)" keit, for hereis a misprint "Fr(r)". 9 See Lukasiewiczand Tarski,op. cit.,p. 42 (reprint 14), Satz 30. p. 10 See P.M., vol. 1, pp. xiii,xvi. u See P.M., vol. 1, p. xvi. Nicod'a i o "dedukcyi 12 See Lukasiewicz, Uwagio aksyomacie uog6lniajqcej," Ksirga pamiqtkowa Polskiego TowarzystwaFilozowicznego we Lwowie (Lw6w, 1931), pp. 2-7. 1 Op. cit., pp. 97-103. My formulation largely departs fromTarski in inessentialrespects, notational.

SET-THEORETIC

FOUNDATIONS

FOR

LOGIC

49

to appropriated each type. All the A-variablesare r-variableswith positive as indicesattached: those withthe index "1" take individuals values, integral thosewith"3" take classes of such thosewith"2" take classes of individuals, classes,and so on. The P-variableto whichthe indexis attachedwillbe called the base of the A-variable. "x" and "y" in "(xey)" by are A-formulae formed replacing The elementary whose indicesare consecutiveand ascending.These formulae by A-variables and functions quantificaones,as in r, by the truth are elaboratedintofurther it of later matters willbe demandedthat no A-formula tion.For simplification does no withlike bases and unlikeindices;this restriction containA-variables choiceofletters. sinceit touchesonlythearbitrary violenceto Tarski'sscheme, as is in then,in termsof the r-formulae, any A A-formula describable general, and integral, (positive in expressionsuchthateveryP-variable fbearsa numeral f in as all hereafter) index, like -variables fbear like as willalwaysbe understood ascendinginindices,"e" occursonly betweenr-variablesbearingconsecutive are A-formulae all f of dices,and the result deleting indicesfrom is a r-formula. as thenabbreviated in D1-2. jedes the Despite its simplicity, languageof A is adequate "zum Ausdrucke Gedankens."'4 Relations,e.g., can be formulierbaren in Principia Mathematica or of of in introduced terms classesby themethod Wiener'5 Kuratowski.'s conunderany of the following A A-formulais declaredto be a A-theorem f ditions: A1-4. (Same as rl-4, but with"A" put for"P' throughout.) for A-variables by underD1-2, of] f is formed putting AS. [A transcription, A-variables and p for"x" by of] A6. [A transcription f is formed putting X in "F" by a A-formula which and "y" in "-(x)-(y)((yex) -P)" and supplanting Xis notfree. For purposesofA1-6 the words"bound" and "free"are of courseexplained "rP. just as in ?2, but with"A" supplanting rS. is there a further A-rule, paralleling He needsthis In Tarski'sformulation inof the because in his formulation A5-6 he specifies bases of the A-variables here,on the otherhand,to phraseA5-6 without volved. It has been convenient no sets of variables.Since thereis consequently such referto reference specific of in ence anywhere Al-6, the substitution one freevariable foranotherin a of the can by theorem be accomplished repeating wholederivation thattheorem parallelto rS initialchoiceof variables.A ruleof substitution undera different from be rS could similarly eliminated Note that becomessuperfluous. therefore variables.Conversely, to r6-7 as to avoid reference specific so rephrasing r, by in also, the systemA admits of reduction ways parallel to those sketchedfor P in ?2.
11 Tarski,op.

the letters in "((z)((zex)

(zey)) n ((xew) 3 (yew)))".

16Kuratowski, la notion l'ordre Fundamenta mathematical, dans la thdorie ensembles, des de Sur vol. 2 (1920), p. 171.

vol. cal Society, 17 (1914), pp. 387-390.

16Wiener, simplification thelogicofrelations, Proceedings of the Cambridge Philosophiof A

cit.,p. 97.

50

W. V. QUINE

from P not only in depending upon 4. Construction of A within r. A differs the theory of types but also in involving some extra elements, viz. the null class in each type. The problem of derivingA fromr thus resembles the problem of deriving the system of rational numbers fromthat of integers,or the system of integers from that of natural numbers. In the latter derivation the natural numbersare not directlysupplemented by the negative integers;ratherthe whole series of integers is developed anew, in such fashion that the non-negative integers do not coincide with the natural numbers with which they are intuitively associated. A similar course is adopted in derivingA fromr; rather than directly supplementing P somehow with null classes, the procedure is to present a general .reconstructionof the elements of A in terms of r. recursivelythus: the correFor each element of A let us definea correspondent of an individual is its unit class, and the correspondentof a class x is the spondent class whose memberscomprise all individuals and in addition the correspondents A of the members (if any) of x. Now I incorporate A into P by reinterpreting to the extent of supposing all its elements replaced by theircorrespondents.Thus a A-element of firsttype, represented in A by a variable with "1" as index, and construed as an individual, is reconstrued as a class (a unit class, as it formerly happens) of individuals. A A-element of n+ 1st type, formerlyconstrued as a class of A-elementsof nth type, is reconstrued in two respects: its members are reconstruedin the manner appropriate to the nth type, and it is then assigned all individuals as additional members.In particular,.thus,the null class of each type is reconstrued simply as the class of all individuals. All A-elements are thus reconstruedas classes; and this in the strictsense of F, viz. classes with members. Now "1 Ix", read "x is of firstr-type," is definedin P thus: D5. I Ix
= df(( Sy) (yeX) . (Y) ((yeX) D (W) (wy))).-

Thus 1 Ix if and only if x has members but none of them have members in turn; i.e., the firstr-type comprises all and only the classes of individuals. Hence the elements of A, reconstrued as they are as certain "individuals" or first-type of individuals, become classes of firstP-type. classes Next, given the nth P-type, the n+lst is defined thus: D6.
n+ I I x =W~Y) (((YeX)-

aw) (wey))n n ly).

Where n+ lIx, D6 obviously provides that if yex and y has members then nty; furtherD6 can be shown to provide that yex if y has no members.'7 In effect, then, the n+ 1st r-type is definedas comprisingjust those classes x which have all individuals and in addition none but classes of nth r-type as members. From of the described reinterpretation A it thus follows that if all A-elements of nth type are reconstruedas classes of nth r-type then all A-elementsof n+ 1st type are reconstrued as classes of n+1st r-type. Since all A-individuals are recon-

of that the A-elements every by r-type,it follows induction struedas of first


type n are reconstruedas of nth r-type. In particular the null class of any type n, reconstruedas it is as the class of all individuals, turns out to belong at once to all P-types.'8
17

of in See thenextto thelast formula theproof Lemma 3, below. " This is the basis of theproof Lemma 5, below. of

SET-THEORETIC

FOUNDATIONS

FOR LOGIC

51

Formally the incorporationof A into r will consist in regardingeach A-variable, having say the nth numeral as index, as restricted in range to the nth r-type. Thus indexed quantifiersare introduced into P through this definition:

D7.

(X")P =df(X)(nIx D P).

This convention is added: Cl. An index which is attached to a P-variable within an indexed quantifier may be attached also at will to any recurrenceof that variable within the scope of that quantifier. Strictly, D5-6 should be regarded not as a definitionof the general case "In series of specificabbreviix" but as a convention forthe adoption of an infinite series of abbreviations. ations " lx", "21ix", etc. D7 then introduces a further A A-formulawill be called closed if it contains no freeA-variables. Now let 0 be a A-formulawith just k(>O) freeA-variables. Any result of putting P for "p" in "(xi(')) * * (Xk(k))P", and putting the free A-variables of p in any order for itxl(l)'"to "Xk(k)", will be called a c.g. (closed generalization)of p. Now D1-7 and of as C1 explain every closed A-formula a conventional transcription a F-formula. In particular, all closed A-theoremsare thus translated into r. The A-theorems each as a closed theorem in general may then be translated into r by interpreting in the familiar fashion, viz. by supposing each supplanted by one of its c.gs.19 (If a A-theoremis closed to begin with, it is of course its own c.g.) To show that the system A proceeds fromr, then, we need only show that every c.g. of a A-theorem is [a transcription,under D1-7 and Cl, of] a P-theorem. 5. Proof that every c.g. of a A-theorem is a r-theorem The proof will only be sketched here; in rigorousformit would involve elaborate metamathematical developments and run to greaterlength. All the familiar technique of the propositional calculus and theory of quantifierswill be taken for granted, since there in is no essential difficulty showing that the technique accrues to P through r1-5. Such clauses as the bracketed ones in r6-7 and A5-6 and at the end of ?4 will be suppressed entirely,as in the title of ?5; in effect, thus, definitionaltranscriptions of r-formulae or of A-formulaewill be treated as -formulaeor A-formulae in theirown right.

the are LEMMA 1. If P-formulae putforthecapitalsin the following schemata, are results P-theorems:
(a) "((P D Q) D (((RP) D (((R-P) D Q) D (R D Q)))" P) D Q) D (P D R)))"

(b) "((Q: (c) "(((P.

Q) D (R D S)) D ((P D R) D ((P. Q) D S)))" _= Q) z R). ((S =_ Q) D R)) D ((R D (P -_S)) D ((P D S). (S:

(d) "((((P (e) "((Pa

P))

Q)

((Rn '-

P):

((R:

Q)

((S

(Q.(-'

(((S -

P 3 (R. )))) P) * R)-(R D (S -T)))))))

in A 19 See P.M., vol. 1, p. xiii. Indeed, Tarski has revised in just thisfashion Der Walvkeilsin Sprache, Studia phluosophica,vol. 1 (1935), pp. 365-366,291-299. begriff denformalisierlen

52

W. V. QUINE

Proof.All these are valid formsof the.propositional calculus, as can be checked thetruth-table by method otherwise. or
LEMMA

2. Thefollowing r-theorems: are


(b) "(ax)(y)((yex) (d) "( 3x) ((yex) (y) (ycz))" (W) (Wey) " (Wey))

(a) " ((w) -(wey) 3(y cZ))" (c) "(By) (W)

(e) "( 3x)l x"


By r7,

(f) "( ax)(z')(z c x)"

Proof.The case of (a) is clear fromD4. (b)-(d) are derivableas follows.


(ax) (y)((yex)_ (ax)(y)((yex) ((y c Z).(y c z))), ((y cz).(w) ((w cz).'

(1) (2)

(wey))),

and

( 3y)(w)((wey)

(wcz)).

(3)

(b) proceeds from (1), (c) from and (d) from and (a). (e) and (f), finally, (3), (2) are derivable follows. (d), as By ( ax) ((y) ((W) Hence, by (c), Again,by (d), Hence,by DS,
LEMMA

3 (wey) (yex)) -(y)((yex) 3 (W)

(Wey)))

( 3x) (( y) (yeX)(Y)((yEX) 3 (W) '-' (wey))). (3x)(z)((y)((yCez) (ax)(z)(1I (w) - (wey)) 3 (y)((yez) 3 (yex))).

(4) (5)

z (y)((yez) 3 (yex))).

In viewof DS, D4, and D7, (4) and (5) are (e) and (f). 3. A r-theorem results any for fromputting numeral "n" in " (n ly. ( aw) (wey)) ". (a)

Proof.Let the nth and n+1st numerals represented "n" and "n+1", be as and suppose (a) to be a r-theorem. Then we can deriveanotherr-theorem as follows. Lemma 1(b) and D3, By ((n ly:3 ( aw) (wey)) Hence, by (a), Hence,by D6,
3 ((((yEx)

a(w)(wey))3 n 1y) ((w) -(wey)

(yex)))).

((((yex) (n+1Ix

a(w)(wey)) 3 nly) ((w) (y)((w) 3 (wey) (yex))). (n+1lx3 (ay)(yex)).

3 (wey) (yex))).

Hence,by Lemma 2(c),

orem, the result of putting the n+ 1st numeral for "n" in (a) is derivable as a immediately fromDS. Lemma 3 then followsby induction.

In thisway,iftheresult putting nthnumeral "n" in (a) is a r-theof the for

"1" for"n" in (a) is a r-theorem, is seen r-theorem. theresult putting But of as LEMMA A r-theorem 4. results from putting numeral "n" in any for (a) "P(rtz)t(y")m(y CZ)r".
Proof. Let the nth and n+1st numerals be representedas "n" and "n+1",

SET-THEORETIC

FOUNDATIONS

FOR LOGIC

53

as Then we can deriveanotherr-theorem and suppose (a) to be a r-theorem. By follows. D6-7,


((Y") (Y C Z) 3 (X) (n+ I Ix D (y) (((yex)aw) (Wey))D (y C z)))).

Hence,by (a), (((w)

(Sz)(x)(n+ 1x D (y)(((yex)
D

(w)(wey)) D (yC Z)))

(1)

By Lemma 1(a) and D3,


'

(wey)D (y C z))

((((yex)

(aw)(wey)) * (yC z)) 3 ((yex)* (y C OM)


S(w) (wey))3 (yC z)) D ((yex)* (y C z))).

Hence, by Lemma 2(a), Hence, by (1), Hence,by Lemma 2(b), I.e., by D7 and D4,

((((yx) ( Sz)(x)(n+

lIx D (y)((yex)D (yC z))).


(y) ((yex) * (yet))).

( at) (x) (n+ 1Ix ( at)(x"+')(x c t).

In this way, if the resultof puttingthe nth numeralfor"n" in (a) is a for the the r-theorem, resultof putting n+1st numeral "n" in (a) is derivable by FromLemma 2(f), then,Lemma 4 follows induction. as a r-theorem.
LEMMA

nfor"n" in "( 3x)n Ix". any results putting numeral S. A F-theorem from

Proof.Lemma 2(e) covers the case wheren is "1". If n is not "1", then, thus.By rI(b), n representingas "m+ 1", we can derivethe r-theorem
(((yex) (w) (wey)) D ('((yex) (w) - (wey)) D mly)). (w)
-

Hence, by Lemma 2(d),


ILe.,

(x) (y)(-((yEx)
a x) (y)W(yex)-

(wey)) 3 mIy).

(w) - (wey)) 3 mly).

I.e., by D3 and D6,


DEFINITION.

(ax)m+1Ix.

is by which obtained dropping that If p is a A-formula, r-formula of the from freeA-variables p willbe called I the indices result puttingp| for"Q", of and indices attached, q is the respective n, tonikarethe I in for for g1toXk "x1" to"Xk", and ni tonk "(1)" to "(k)" in
LEMMA

6. If g1 to Sk

(O<

of freeA-variables a A-formula k) are thebasesofthe

"(((I) Ixi.

3 (k)IXk) Q)" I

c.g. if then is a r-theorem and onlyifevery ofp is a F-theorem. q


k)P'', with "P" replaced (ykl Proof. Every c.g. r of p has the form"(yi11) * and "[1]" to by p, and "yi" to "yk" replacedby the Xiin some permutation, By permutation. D7 and C1, "[k]" replacedby the n, in the corresponding from wheret is formed if r will be a r-theorem and only if t is a r-theorem, . z ...**(yk)( [k ]Iyk Z Q) )" by putting I pI for "Q" and replacing 11(yl)( [I] y calculusand But, by thepropositional "yi" to "yk" and "[1]" to "[k]" as before. t if i ofquantifiers, willbe a r-theorem and onlyiftheresult of thesame theory into But in replacements "(([1] lyi ... [k]IyA)D Q)" is a r-theorem. t is carried calculus. are which allowedby thepropositional viceversabypermutations q and

54

W. V. QUINE

But if r, Hence t, and therefore willbe a r-theorem and onlyifq is a r-theorem. if r was any c.g. of p. Hence q is a r-theorem and only if everyc.g. of p is a r-theorem, q.e.d. and LEMMA 7. If p is a A-formula P-theorem.

every c.g. of p is a is a P-theorem,

as 6 the Proof.If j is a P-theorem, q of Lemnma followsimmediately a q.e.d. By r-theorem. Lemma 6, then,everyc.g. of p is a r-theorem, and 8. If p, q, r, and 6 are A-formulae,is a A-variable, r and 6 are X respectively "(P v Q)" and "(Pm (x)Q)" by putting for "P", q for from p formed so "Q", and Xfor"x", then(i) if all c.gs.ofp and ofr are P-theorems, are all c.gs. then in free p,and all c.gs.ofr are P-theorems, so areall c.gs.of6; ofq; (ii) ifXis not then (iii) if all c.gs.of9 are P-theorems, so are all c.gs.ofr.
LEMMA

as be represented "Xh(h)"X to "xk(k) (h< k+ 1), and let thosewhichare freein q as but not in p be represented "X(*$1)'' to iXm(m)) (kim). Let p and q be repreas sented accordingly "4(x1('), - - X(k)" and "41(Xh(h), * * X(m)". Let X be n A as as represented "X"W", wheren is 0, m, 1, or h according X is freein neither p norq (Case 1), in q but not in p (Case 2), in p but not in q (Case 3), or in both p and q (Case 4). Consequently
+O (X1, -Xk)",

representedas "xi(l)) to "Ax-1)') (1 < h), let those which are free in both p and q

Let whichare freein p but not in q be Representations. those A-variables

"'/(xh, *

Xm)",

and "(4(x1,

* Xk) 3 k(Xh,

X))

to answerrespectively j
" ( (XI

, q |, and

as , and j6I is represented


Xm))" Xm_1,Xm(m)))" -Xn))"

* - * Xk) D (Xnf))#(xh,**
.
.

(n<h; Cases 1 and 3) (Case 2) (Case 4).

or or

"(4(x1,
"(I(x1,

Xk) 3 (Xm(m))44(Xh,

- - Xk)D (xh(h))4ti(xh(h), Xh+1,

of is By Proofof (i). A r-theorem derivableas follows. the hypothesis (i), withLemma 6, together (((1) Ixi. and
I (((1)xi. X

- (k) IXk)
(m) Ixm)

1(Xl, (1(xI,

Xk))
Xk) X D 1(Xh,

(1) *
Xm))).

(2)

By Lemma 1(c),
((((1) I XI-. (m) I X.)
D((X1 1O(X1, *

Xk) 1{(Xh,
D (((I) I X1- *

X.)))

2
D CAx* X.,))))

((((I) Ixi.

(k) I Xk) D

**Xk))

(m) I X.)

Hence, by (2) and (1),

(((1) Ix1. *

(m) X) D O(xh,

Xm)).

I e.,
(( x,)(1) Ix X, (( 3
_)(k -, 1) I Xh-1 D (((h)
I Xh -

(m) I Xn) D

SET-THEORETIC

FOUNDATIONS

FOR LOGIC X.))

55

Hence,by Lemma 5,

(((h) IXh.

(m) IXm)D 4P(Xh,

By Lemma 6, then,everyc.g. of q is a r-theorem, q.e.d. of withLemma6, (2) above is a Proofof(ii). By thehypothesis (ii), together r-theorem. hypothesis, By also, Cases 3 and 4 are excluded.In Cases 1 and 2, r-theorems derivableas follows. are Case 1. By (2),
(((I) I xi.-1(m) Ixm,):o (,0(xi, * * * * (i)
Im)

A, X n (X,) ((n)l lX.


D (ck(xi,

(Xh,

X,,))))*

(3)

ILe.,by D7,

(((1) Ix.

x-k)

3 (Xfl,() W(XA,

Xm)))- (4)

By Lemma 6, then,everyc.g. of 0 is a r-theorem, q.e.d. Case 2. By (2),


(((I) I X1 (m -1) I n- 1) D (10(Xi,* Xk) 3 (n) ((M) IlX,, D 4(Xh, * Xn-a)))).(S)

I.e., by D7 and C1,


, (((1) Ix1. - - - (m- 1) IxM-i) 3 (O (x1i
- Xi)
D

(Xm(m))4,(Xh,*

Xm-1, Xm(m)))).

(6)

By Lemma 6, then,everyc.g. oft is a r-theorem, q.e.d. of Proof (iii). Cases 1 and 3. By thehypothesis (iii), together of withLemma6, (4) is a r-theorem. D7, then,so is (3). Hence so is By
"(((I)

I X1.-

(m)

I Xm) D (4)(X12

* *

Xk) D (( qXn)(n)

lXn34'(Xh,

*X

m))))",

and consequently, Lemma 5, so is (2). by and Lemma 6, (6) is a r-theorem. D7 and C1, Case 2. By hypothesis By then,(5) is a r-theorem. Hence so is (2). and Case 4. (2) is derivableas follows. hypothesis Lemma 6, By
((1 X1.
(m)
I Xm) 3 (qS(XI, ***Xk) 3 (Xh( ))(Xh(^ Xh+1, **S))

I.e., by D7 and CI.,


X(1IX1

(m) I Xm) ZI (+(XI,* (m) lm) = ( I(X1,*

Xk) D (Xh)

A ZO 4t5(Xh,

Xm))))

Hence

(((1) Ix1.

Xk)

((I) lXh A

4(XhA, * *

Xm)))).

Deletionof the repeatedpremise reducesthisto (2). (2) is thusa r-theorem all fourcases. By Lemma 6, then,all c.gs. ofr are in r-theorems, q.e.d.
LEMMA 9. If the A-formulais formedby puttingA-formulae thecapitals in f for (a), (b), or (c) of rP, thenevery c.g. off is a r-theorem.

Proof.If fresultsfrom puttingp, q, and r forthe capitals in (a), then If I resultsfromputtingthe r-formulae IpI, Iql,and IrI forthe capitals in (a). Then,by rt, J is a r-theorem. fJ Consequently, Lemma 7, everyc.g. offis a by The r-theorem. cases of (b) and (c) are analogous.
LEMMA

is a P-theorem.

"x,

10. If a A-formulafis formedby puttingA-variablesX, I, j, and tt for "y", "z", and "w" in "((z)((zex) (zey)) D ((xew) D (yew)))", everyc.g. of f

56

W. V. QUNE

Proof.Let us represent t, j, and )o as "x"1+","y.+l",I"Z", and "w"+2'. X, (The indicesmustbe consecutive indicated, as sincefis a A-formula.) can We derivea r-theorem follows. Lemma 1(d), as By
(M((((Zx)Sw) (we))
n n Iz).
((n

n nl2 )) n aw) (wez)) (((zey)-- (zey))) n (((zex) 3 (zey)) z n ((zex) 3 ((zey)

D (zex)))))

Hence,by D6-7, C1, and D4,


((n + I Ix. n+ I ly) n ((zn) ((znex) =_(zney)) 3 ((x C y). (y C x)))).

Hence, by Lemma 3,
((n + I Ix. n+ I Ily) D ((zn) ((znex) _= 2^ey)) 3 ((x C y). (y C x). ( az) (zex)))).

Hence, by r6, Hence ((n+2

((n+ 1I x.n+ 1 Iy) n ((z")((znex) Iw.n+ 1 Ix.n+ 1 ly) n ((zn)((zREx)

(zney)) 3 ((xew) 3 (yew)))). (zney)) Z ((xew) 3 (yew)))).

Then,since"((z1I)((zaex) (z"ey))2 ((xew)3 (yew)))" is | fI,itfollows from Lemma 6 that everyc.g. of f is a r-theorem, q.e.d.
LEMMA11. If a A-formula is formedbyputtingA-variablesr and qfor "x" and f "y" in CCs(x)#-(y)((yex) =P)" and supplanting "FP" by a A-formula in whichZ p is notfree,thenevery c.g. offis a r-theorem.

Proof.Let 0 and F be represented "yk and CCx(k)+l, and let the free as ()?" 0 of A-variables p be represented "y as (i)" to Ity&(h) (Oh? <h+1). k If 0 is free in h=k- 1. Finally,let us represent as 94+(yi),...** QW) p, h=k; otherwise p Now a r-theorem derivableas follows. is Let
((YkeX) -((Yk

C 2).-((

aW) (Weyk)

C ((k)

O Iy.(Y1,-

yh)))))

be abbreviated "c1". By Lemma 1(e) and D3, as


(((W) -( (weYk) 3
(yk C Z)) 3

(4) 3 ((((yk'EX) _

Yk3 ( W)(Weyk)) Cyk (y C z)) D (((k) I Y Z W 3 ((ykCX) ( W)(eyk)) 2 (k) I Y*)k) O(y1, ***yet)))))))) IYk

Hence, by Lemmas 2(a) and 3,


Y(k lyD (Yk C )) D (4+

((((ykefx) -

D 3 aw) (Weykc)) (k) I Yk)- ((k) IYkc ((ykeX) - +(yI, $ D ((k) + 1 I x (yk()) ((ykex)

***yh))))))

Hence,by D6-7,
((y(k)) (yk C Z) D ((yk)

(Y,

But, by r7, (ax)(yk)

$.

Hence

c ((yk(k))(yk z) D (3x)((k)

+ 1I x.(yk (k))((yYkex) + 1Ix.(yk(k))((ykex) -

(yl, t(yI,

yh)))). *Y))). (yi, . . . yh))). yh)).

Hence, by Lemma 4, I.e., I.e., by D7 and Cl,

(ax)((k)

(x)((k) + 1Ix #s (yk(k))((ykex) (x(M+1)-*' (yk(k))(yk(kex) -= (y,*

The r-theorem thusderivedis

IfI if h=

k-1. If h= k, it becomes Ifj when

SET-THEORETIC

FOUNDATIONS

FOR LOGIC

57

is to (k)". In either from case, then,it follows according CI as "Yk Ityh" rewritten Lemma 7 thateveryc.g. offis a r-theorem, q.e.d.
THEOREM.

Every ofa A-theorema r-theorem. c.g. is

Proof. A A-theoremf will be said to be of rank 0 if it satisfiesAl, A5, or A6. A A-theoremfwill be said to be of rank n+ 1 if there are A-theoremsp and r, one of rank n and the other of rank n or less, such that r is the result of putting p for "P" and f for "Q" in "(P Q)"; or if there are A-formulaep and q and a A-theorem of rank n such that fand r are formedrespectivelyfrom"(P 3 (x)Q)" r and "(P n Q)" by replacing "P" by p, "Q" by q, and "x" by a A-variable which is not freein p; or ifthereare A-formulae and q and a A-theorem of rank n such 6 p that f and * are formed respectively from "(PF Q)" and "(P n (x)Q)" by replacing "P" by p, "Q" by q, and "x" by a A-variable. of If we assume that every c.g. of any A-theorem rank n or less is a r-theorem, of and take f as a A-theorem rank n+ l, then Lemma 8 shows that every c.g. of f of is a r-theorem. Then, since by Lemmas 9-11 every c.g. of any A-theorem rank 0 is a r-theorem,it followsby induction that every c.g. of any A-theoremof rank m is a r-theorem, for every natural number m. But it is clear from the above definitionof rank, together with A1-6, that every A-theoremhas at least one natural numberas its rank. Thereforeevery c.g. of any A-theoremis a r-theorem, q.e.d.
HARVARD UNIVERSITY

Das könnte Ihnen auch gefallen