You are on page 1of 31

A Set of Axioms for Logic

Author(s): Theodore Hailperin


Reviewed work(s):
Source: The Journal of Symbolic Logic, Vol. 9, No. 1 (Mar., 1944), pp. 1-19
Published by: Association for Symbolic Logic
Stable URL: http://www.jstor.org/stable/2267307
Accessed: 07/06/2012 00:22

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
TwE JOURNAL Or SYMBOLIC Looxc
Volume 9, Number 1, March 1944

A SET OF AXIOMS FOR LOGIC1


THEODORE HAILPERIN

One of the preeminentproblemsconfrontinglogiciansis that of constructing a


systemof logicwhichwill be adequate formathematics. By a system'sbeing
adequate formathematics,we mean that all mathematicaltheoremsin general use
can be deduced within the system. Several distinctlogical systems,all having this
end in view, have been proposed. Among these perhaps the best knownare the
systemsreferredto as "PrincipiaMathematica" and "set theory." In both of these
systems(we referto the revisedand simplifiedversions)there is a nucleus of
propositionswhichcan be derivedby using only the axioms and rules of the
restrictedpredicatecalculus. However,if anythinglike adequacy for mathematicsis
to be expected,additional primitivesand axioms must be added to the
restrictedpredicatecalculus. It is in theirtreatmentof the addi-tional primitivee,
denotingclass or set membership,that the above-mentioned systemsdiffer.

In additionto thesetwo, a thirdand a strongersystemhas been proposedby W.


V. Quine in his paper New foundationsfor mathematicallogic.2 It is with this
systemof Quine's that our workis concernedand of whichwe now give a
briefdescription.
Like the previouslymentionedsystems,New foundationsassumes only one
predicate,that of class membership. One formalaxiom is presented,the prin-ciple
of extensionality,accordingto whicha class is determinedby its members. In
additionQuine has a metalogicalrulestatingthat all formulasofa certainkind are
theorems. The explicitstatementof this rule requiresan explanationof
stratification, conceptdue to Quine. (The followingversionis forexpository
purposesonly,since it is not generallyadequate-an accurate definitionis given in
section7).
All formulasof the systemare of the form xey, or combinationsof this form with
quantifiersand connectivesof the propositionalcalculus. If, now, to the variables
in a formula4 can be attached non-negativeintegerssuch that (1) to each
occurrenceof a variable is attached the same integer,and (2) the integers
attached to variables separatedby e are consecutivelyascending,then0 is said to
be stratified. There is a constructiveprocedurefordeterminingwhetheror not a
formulais stratified.
Received November 24, 1943. Presented to the American Mathematical Society Sep-tember
13, 1943.
1
The material of this paper is contained in the author's doctoral thesis presented to the
Graduate Faculty of Cornell University, May 1943. The author wishes to express his thanks to
Professor J. B. Rosser for helpful guidance in the development of this work.
2 The American mathematical monthly, vol. 44 (1937), pp. 70-80. This is not to be confused with
the logical system presented in Quine, Mathematical logic (New York, 1940)
which Rosser and, independently, Lyndon have shown to be inconsistent. (See this
JOURNAL, VII 1). On the other hand, despite a critical examination by Rosser, no incon-
sistencies have been found in New foundations. (See this JOURNAL, IV 15, as well as the cited VII
1).
1
2 THEODORE HAILPERIN

The metalogicalrule of Quine whichwe have referredto, denotedby R3' in


Newfoundations,is as follows:
R3'. If 4,is stratifiedand doesnotcontainx, then
4
(Ex)(y)[ye 3

is a theorem.
Intuitivelyinterpreted,the rule states that foreach stratifiedformula 4,there is a
class x whosemembersare just those values of y whichmake 4,true. Some such
restrictionon 4,is necessary,forif one wereallowedto take the unstratified
formula-%.(yey)for4, one could readilydeduce the Russell contradictionabout
non-selfmemberedclasses.
It is to be observedthat R3' is not a formalpostulateof the system,but a
statementabout the systemassertingthat all formulasof a certain kind are
theorems. Inmacertainsense, then,R3' is an infinitestringofformalpostulates,
one for each stratifiedformula. The primaryobjective of this paper is the
demonstrationthat this essentiallyinfinitestringof axioms may be replaced by a
finitenumber. The demonstrationwill be achieved by derivingQuine's R3'
fromour axioms,and then showingthat these axioms can, in turn,be de-
rivedfromQuine'ssystem. We nowturnto a preciseformulationofoursystem.
1. Abstractions. A propositionalsymbolprecededby a circumflexedvariable will
be called an abstraction. Abstractionsmay be consideredas incomplete
symbolsand be givenmeaningonlyin context-theywould be eliminablefrom any
propositionwithinwhichthey occur. Yet it is convenientto work with
propositionshaving abstractionspresent,ratherthan with the unabbreviated
formswhichresultwhenthe abstractionshave been removed. To this end we
shall set up two formallogics,L1 and L2, havingthe following~propertiesThe.
logic L1 uses the symbolsI, e, (, ), and a denumerablenumberof variables de-
notedby the lowercase Latin lettersu to z, the Greeklettersa, (, y, and all of
these with numericalsubscripts. The use of the Greek lettersis merelyfor
typographicalclarity and has no logical significance. We assume that the
variables have been arrangedin some definiteorder.
The secondlogic L2 will be like L1 but havingan additionalsymbol,the cir-
cumflex,applied to variables,and additionalpostulates. To each proposition in
the "augmented"logic L2 willcorresponda unique propositionnot containing any
circumflexes,called its abstractionlesstransform. We shall provethat if p is a
theoremin In and p' is its abstractionlesstransform,thenp' is a theoremof L1 .
Thus, in view of thistheorem,as soon as we have proveda theoremof L2 we
can immediatelyset down its abstractionlesstransformas a theoremof L1 . In
thisway we shall have the convenienceof workingwithformulascontaining the
circumflexwithoutpayingthe pricein additionalprimitives.
The followingdefinitionis to holdforbothlogics-the partsin square brackets,
however,referonly to L2.
DEFiNITION 1.01. Noun, proposition,and freeand boundoccurrenceof a
variable:
A SET OF AXIOMS FOR LOGIC 3

1. If x is a variableit is a noun,and thesingleoccurrenceofx is free.


2. If A andB axenouns,then(AeB) is a proposition,and anyfreeor bound
occurrenceofa variablein A or B is a freeor boundoccurrenceofthevariable
in (AEB).
3. If p and q are propositionsthen(p I q) is a proposition,and any freeor
boundoccurrenceof a variablein p or q is a freeor boundoccurrenceof
the variablein (p I q).
4. If p is a propositionand x is a variable,then((x)p) is a proposition[and
1p is a noun]. Anyoccurrenceofx in ((x)p) [or?p] is a boundoccurrence.For
a variabledifferentfromx, any occurrencewhichis freeor boundin p willbe
freeor boundin ((x)p) [and 4]. Anypartofp is said to be in the scopeof
(x) [or x] in the formula((x)p) [orXp].
We pointoutthat,as a consequenceof2, unstratifiedformulassuchas (xex)
are propositions,contraryto the practiceofPrincipixmathematicswheresuch an
expressionis meaningless.In 3 thesymbol" I " is theShefferstrokefunc-tion,in
termsof whichall otherconnectivesof the propositionalcalculuscan
be defined. The prefix"(x)" means,as usual,universalquantificationThe.
noun "4A," as alreadymentioned,willbe referredto as an abstraction.The
parenthesesabouttheformulasare presentto insureuniquenessofinterpreta-
tionand willmostlybe omitted,or replacedby Peanodots,whenno ambiguity
could result. We also use the otherwell-knownparentheticalsignssuch as
squarebrackets.
By convention,thelettersp, q, r,and s willbe usedforpropositionsand X,
Y. and Z willbe usedfornouns.
DEFINITION 1.02. If
z is the firstvariablein the list of variableswhichis
differentfromeachvariablein X or Y, thenwe employtheabbreviation:
X = Y for (z)(zeX zeY).

The particularchoiceof the variablez in theabove definitionisimmaterial


providedthatit is differentfromeachvariableinX or Y. By choosinga definite
onewe havemadethetranslationofX = Y unique.
We nowlist the axiomsof L2. These willbe labeledR1 to R4 (theseare
reallyaxiomforms),PId, and P1 to P9. The precisecontentof axiomsPId
and P1-9 is not relevantat present,and a statementof theseaxiomswillbe
deferreduntillatersections. The existentialquantifier,and thelogicalconnec-
tivesotherthantheprimitivestroke,are definedin theusualfashion.
RI. ((p (q I r)) I ((8 I (s 8)) I ((8 I q) I((p I s) I (p 18))))).
I

R2. Let P(x) be a propositionand P(y) theresultof replacingeachfreeoccur-


renceofx (if any) in P(x) byy. If no boundoccurrenceof y in P(y) is theresult
of such a replacement,then
(z)P(x). D . P(y)
is an axiom.
4 THEODORE HAILPERIN

R3. Let p be any proposition,x and y any variables,and a thefirstvariablein


thelist of variableswhichis differentfromeach variablein yap. Then
ye.p .-- : (Ea) . yea. (x)(xea- p)
is an axiom.
R4. Let p, x, and X be respectivelyany proposition,variable,and noun, and let
w be thefirstvariablein thelistof variableswhichis differentfromeach variable in
IpeX. Then
xpeX. :(Ew).weX.w=tp
is an axiom.
Both systemsL1 and L2 use the followingtwo rulesof inference:
RULE I. If p and p I (q I r), thenq.
RULE II. If p ) q and thereare no freeoccurrencesofx in p; thenp ) (x)q.
Similarly,the followingdefinitionapplies alike to L1 and L2.3
DEFINITION 1.03. Let S be a set of propositionsand let ql, *, q, be a se
.

quence of propositions. We say that this sequence is a demonstrationin n steps


that qn is a consequenceof S if foreach i from1 to n inclusiveeither
(a) qj is in the set S, or
(b) thereare numbersj and k, both less than i, and a propositionr such that
qj is (q%I (qj I r)), or
(c) thereis a numberj, less than i, a variable x and propositions and s, such
that thereare no freeoccurrencesofx in r, q, is (r D s), and q1is (r D ((x)s)).
If the set S is a set of axioms,then qn is said to be a theoremof the system.
THEOREM 1.1. For each proposition in L2, thereis a unique propositionp'
(called the abstractionlesstransformof p) whichcontainsno circumflexes,and
whichis equivalentto p.
Proof. If p has no abstractionswe take p' to be p. If p has abstractionswe
obtainp' as the resultofthe followingoperations. Firsteliminateeach abstrac-tion
occurringimmediatelyto the leftof an e by beginningat the right-handend ofp
and successivelyreplace each part of the formxqeX by (Ew) . weX. w = 1q,
wherew is the firstvariable in the list whichis differentfromeach variable in :
9qeX. Then, beginningalso at the right-handend of p, replace each part
ofthe formyex.qby (Ea): yea . (x)(xea q), wherea is the firstvariablein the
list differentfromeach variable in yeq. In virtueof R3 and R4, each of the
replacementsyieldsa propositionequivalentto p.
2. Eliminabilityof abstractions.4 Having definedabstractionlesstransform, we
now specifythat the axioms of L1 are to consistof the following:
(1) the axioms R1 and R2, and
(2) the abstractionlesstransformsof PId and Pl-P9.

' Cf. J. B. Rosser, in this JOURNAL, IV 17.


4
The ideas for this proof were taken from an unpublished paper of C. D. Firestone
on a similar treatment of descriptions.
A SET OF AXIOMS FOR LOGIC 5
From the definitionof abstractionlesstransformwe haves

THEOREM 2.1. If p' and q' are theabstractionlesstransformsof p and q, then


p' I q' and (x)p' are theabstractionlesstransformsof p I q and (x)p.
THEOREM 2.2. The abstractionlesstransformof R1 is a theoremof L1.
Proof. The abstractionlesstransformof Ri is of the same formas RI with
p', q', r', and s' replacingp, q, r, and s.
THEOREM 2.3. The abstractionlesstransformof R2 is a theoremof L1.

Proof. Let
(1) (x)P'(x) D P'(y)

be the abstractionlesstransformof R2, whereby P'(x) and P'(y) we mean the


abstractionlesstransformsof P(x) and P(y). If, apart fromsimple changes of
bound variables,P'(y) is like P'(x) except in having freeoccurrencesof y wher-
ever P'(x) has freeoccurrencesof x, then (1) is an axiom of L1. To prove this we
show that the successivereplacementsin P(x) are "like" those of P(y) apart
frompossible differencesin the bound variables, and that no bound y results
fromreplacementsin P(y). We considerthe various possibilities.
Case 1. uezp(x) is replaced,wherep may have free occurrencesof x. This part
is replacedby
(2) (Ea) . uea . (z)(zea 5 p(x)), while the
correspondingpart uezp(y) is replacedby
(Ed) * uep. (z)(ze-p(y)),
whichis "like" (2) except that a and ,3mightbe different.We note that y does not
become bound since z, and each quantifieror circumflexedvariable within whose
scope y falls in p(y), must be differentfromy, otherwisey would have
become bound in goingfromuezP(x) to uezp(y).
Case 2. xezp(x) is replaced.
Case 3. &p(x)eZ(x) is replaced. These
are treatedas was uezp in case 1.
THEOREM 2.4. The abstractionlesstransformof R3 is a theoremof L1. Proof. By
2.1 the abstractionlesstransformof (Ea) . yea. (x) (xea p) is
(Ea) . yea . (x) (xea -p'). But thisis also the abstractionlesstransformofyexip.
Hence the abstractionlesstransformof
yedp. (Ea) . yea . (x)(xea =p) is of
the formq =q, whereq containsno abstractions.
THEOREM 2.5. The abstractionlesstransformof R4 is a theoremof L1.
Proof. Similar to that of 2.4.
THEOREM 2.6. If p is a theoremof L2 and p' is its abstractionlesstransform, then
p' is a theoremof L1.
6 TEMODORE HAILPERIN

Proof. By inductionon the numberofstepsin a demonstrationofp, using


theorems2.2,2.3,2.4 and 2.5,as wellas.theinferentialiules I and II.
This completesthe proofof the eliminabilityof the abstractions.
3. Identity. In thissectionwe provethatif two nounsare identical,they
are inter-substitutable(withsuitableprovisionsto avoid confusionof bound
variables). The proofrequirestheuse ofaxiomPId, whichwe nowstate:
PId.. x = y . D . (a)(xea D yea).
The followingthreetheoremsare immediateconsequencesof the definition
ofidentity(1.02) and well-knownpropertiesofequivalence.
THEOREM 3.1. X = X. = X.
THEOREM 3.2. X= Y . D .Y
THEOREM3.3. X= Y. :X = Z. . Y = Z.
By meansof PId and thepropertiesofabstractionsgivenin R3 and R4
one derivesthe followingseriesof theorems.
THEOREM 3.4. If w doesnotoccurin X orZ, then
XeZ. .(EW)(weZ .w = X).
THEOREM 3.5. X = Y. D.XeZ YeZ.
THEORY 3.6. X = Y. D.ZeX ZeY.
THIEORE (x)(p 3 q) .3D . p = 1q.
3.61.
THEOREM 3.7. LetF(X, X) bea propositioncontainingoneormore occurrences
ofX and sch thatat leastoneoccurrenceofX is notcoveredby-thatis, is notin
thescopeof-anyoneofoil thevariablesthatoccurin X or Y. LetF(Y, X) bethe
propositionresultingfromF(X, X) bysimultaneouslyreplacingsomeor all ofthe
X's notso covered,by Y. Then
X = Y. D. F(X, X) F(Y, X)
i8 a theorem.
The proofofthistheorem(forthereplacementofan arbitraryX) is by induc-
tionon the maximumnumberof circumflexedvariableswithinwhosescopean
X lies. Withthistheoremwe have obtainedinter-substitutabilityofidentical
abstractions.
4. Abstractionsas classes. In this sectionwe investigatethe conditions
which enable us to use abstractionsas classes.
The followingtwo theoremsare easilyproved.
THEOREMi 4.1. = (x).
THEOREMi 4.2. (x)(xef dp.
p) * * =
We turnourattentionnowto provingthatifa propositionistrueforall values.
ofa variable,thenit is trueforanyabstraction.For this,however,we require
the followingadditionalaxiom.
P1. (u, v)(EB)(x)(xeft =
xeU Jxev].
A SET OF AXIOMS FOR LOGIC 7

We shall prove a series of consequencesof this axiom; some of these will not
be used until a later section.
THEOREM 4.31. (u)(E(3)(x)[xe,8 -=(xeu)].
Proof. In P1 take u and v to be the same.
THEOREM 4.32. (u, v)(Ef3)(x)[x - (xeu . xev)].
Proof. By P1 thereis a (,3 such that forany x,
xef3i. XEu xeV

and by 4.31 thereis a (2 such that forany x,


Xe12.

Whenceby substitution
Xe,02. 3 '.(XeU .
I XeV). (XeU . XeV).

THEOREM 4.33. (u)(E(3)(x)[xeO3. xeu . (xeu)].

Proof. From 4.31 we have a f,3such that forany x,


and from4.32 we have a (32 such that
XIE02 .-.XeU * XC-1e

Whence by substitution
X 2. 'x'(xEU)..

This theoremgives us a null class.


THEOREM4.34. (u)(E3)(x)[xe3. -. xeuv--(xeu)].
Proof. Theorems4.31 and 4.33.
This theoremgives us a universalclass.
THEOREM 4.35. (u, v)(E3)(x)[x0-. xeu D xev].
THEOREM 4.36. (u, v)(E,3)(x)[xqe. - xeU xev].
THEOREM4.37. (u, v)(Ej)(x)[xej3. * xeu v xev].
Proofsof the above threetheoremsare easy and will be omitted.
THEOREM 4.4. If (3doesnotoccurfreein p, then
= up)
(E#))(ft
is a theorem. p).
Proof. Case 1. (Ea)(x)(xea =
In thiscase the theoremfollowsdirectlyfrom4.2.
Case 2. --'(Ea)(x)(xea _ p).
From this we obtain
,-[(Ea) . yea . (x)(xea_ p)]
so that
xeu. -(xeu) (Ea) . yea . (x)(xea p).
8 THEODORE HAILPERIN
From this and 4.33

(El) (y)[y4 (Ea) . yea. (x)a(x p)]s

fromwhich,by R3 and 1.02, we obtain our theorem.


THEOREM 4.5. Let F(Ap) be thepropositionwhichresultsfromF(u) byreplacing all
freeoccurrencesofu in F(u) by1p. If no freevariableof1p becomesboundbyj
theprocessthen
(u)F(u) . .F(xp)
is a theorem.
Proof. Assume j3 = 1p, where j3 does not occur in ip, (u)F(u), or F(fp).
Since (u)F(u) D F(C8)we have, by 3.7,
=xp. .(u)F(u) DF(xp).
Whence
(EMP( = 1p). D. (u)F(u) ) F(ip),
and therefore,by 4.4, (u)F(u) ) F(ip).
THEOREM4.6. If j doesnotoccurfreein p, then
(E#)(x)[xe# p] * D * (i)[xicp- p].
Proof. Assume we have a j3 such that (x) (xo p), wherej3 does not occur
freein p. Then by 4.2, j3 = Ap. Whenceby 3.7 and assumption,(x)[xdp = p].
THEOREM 4.7. (x, y)(Ef)(z)[zO xey].
Proof. In case xey, 4.34 gives us a j3 of the desiredkind,and in case (xey),
4.33 givesus the desiredP. Hence in eithercase we have the theorem.
5. The ordered n-tuple. This section introducesthe orderedn-tuple-the
definitionis not the usual one, but one whichis especiallydesignedforour pur-
poses.
DEFINITION 5.01. IX for a(u = X),
whereu is the firstvariablein the list whichis differentfromeach variable in X.
DEFINITION 5.02. &OXforX
'X for &(&'1X) (n>0).
This gives aX for &'X. u Y for4(ueX . v. uY),
DEFINITION 5.03. X
whereu is the firstvariable in the list differentfromeach variable in X and Y.
DEFINITION 5.04. (X) forX
(X, Y) for &2X u &(tXu &Y).
(X1, X2, ... , X.) for ... PX")) (n>2).
(92z -2)X1,(X2,
This definesthe orderedn-tuple; the firstabbreviationis merelyfor con-
venience. The definitiontells us that the orderedn-tupleis the orderedpair
whosefirstelementis 2("-2)X, and whosesecondelementis the (n- I)-tuplI
A SET OF AXIOMS FOR LOGIC 9
2(2) is to insure"homogeneity,"that is, if inte-
(X2, *--, X,"). The use of
gers wereto be attachedto the variablesin any formulacontaining - **, x"t) (xi,

in accordance with the definitionof stratification,then to each of the variables xi,


--
*, xnwould be attachedthe same integer.
THEOREM 5.1. (L x1, 2 X2, ... ** , Yk+1)) is
, 2YXjI (Y1,
(XI12, *" I Yk+l) (j>01 k_0).
I Xi$ Ye I ..
This propositionis frequentlyused. We run the subscripton Y up to k+1,
ratherthan k, to avoid the cumbersomeexponent2k-2.
Proof. If k=O, the proofis trivial (see definitions5.02 and 5.04). Suppose
then that k>0. For j= 1 the theoremreproducesthe definitionof an n-tuple.
As hypothesisof inductionassume the theoremtrue forj-1 and considerthe
... , I ... Yk+l)) is
case forj. By definition(a2X1, X,, (Y1, *,
(L 2k+2(j-l) xi (2X2 ***..., I TXj (Y1, ,Ykl)))

whichin turn,by hypothesisof induction,is


(t2k+2(jil)X1, ,
(X2, ... Xi, Y1, ..., Yk+l))

,
and this, by definition,is (X1,I*. Xi, Y1, Yk+l)

THEOREM 5.1 1. a = a U a.
.
Proof. uea u a. ux(xea v xea)
*
-- . ud_(xea) [R3]

.- *Uea [4.1, 3.7]


THEOREM 5.2. (X, X) = 2X. &X and
Proof. From 5.11 and 4.5 we readily obtain &X = LX u &2X =
I2Xu 2X. Thus (X, X) = 2X U L(X u X) =2X U L2X = &2X.

6. Preliminarytheorems. Our furtherprogressnow dependsupon additional


axioms. The results of sections 3 and 4 have shown that if two abstractions are
identical they are inter-substitutable,and that if a propositionis true for every
value of a variable,it is true for any abstraction. One propertyis yet
lackingbeforeabstractionscan be treatedas classes, namely the propertythat the
elementsofan abstraction,say Ap,are just thosevalues whichmake p true-that is,
we do not have
(1) (x(xxexp =p).
If this could be proved in general, it would be very unfortunate-forby takingp
to be -%-(xex)we could deduce the Russell contradiction. On the other hand, in
view of 4.6 we can prove (1) wheneverwe have
(2) (E3)(x)[xe#= p].
Our goal is to prove that (2) is a theoremwheneverp is stratified. To do so we
require the followingadditional axioms. For convenienceof reference,we have
repeated P1.
10 THEODORE HAILPERIN
P1. (u, v)(E13)(x)[xO
e xeulxev].
P2. (a)(Eft)(x,y%(E, ty) _(x,y)ea].
P3. (a)(EP)(x, y, z)[(x, y, z)amn(x, y)ea]r
an. (a)(Exo)(x, y,z)[(x,z, y) g (X,y)ue].
P5. (a)(En)l(x, y)[(y,x)e taxea].d
P6. (a)(E,#)(x)[xe# (u)((u, &x)ea)].
Tn. 6.1
(a)(Et) ()(E)(x, Y)ea].

y)[(3(X,
P8. (Ej6)(x)[x0 3 (Ey)(x = y)].
P9. (EB)(x,[(inductiy) o y3k.Fx,
These are the siomswhich tie to replace Quine's R3'. A few woromof
descriptionare in order. AnyomP1 has alreadybeen treated. For any relation a,
P2 providesus with a correspondingrelationwhich is of one highertype. A()xns
P3, P4, and P5 enable us, amongotherthings,to "raise" the degreeof a relation.
Axiom P6 eventuallygives us the domain of a relation,P7 gives us the
converseof a relation,and P8 gives us the class of all unit cleats, that is the
cardinal 1. Finally,P9 state that xeydeterminesa relation-but one-whichis
inhomogeneous.t The use ofh in P9 insurespr.thatositionthis is stratified.
The followingtheoremgeneralizesP2.
THoREM 6.11. (a)(E P)(x, y)[( ?x, ?y)c,#-(x, y)ea (k20).

Proof. By inductionon k. For k=O the theoremfollowsfrom(x, y)ea 3= (x,


y)ea. As hypothesisof inductionassume the theoremtrue fork. From P2, thereis
a #Isuch that forany x and y,
(1) (LX, Y)01) (x.,Y)ea.
By hypothesisofinduction,thereis a such that forany x and y,
(2) ( ?X, fly)e2 - (X, Y) et,-

Since this is true forany x and y, it is also true for as and ty (see 4.5). On mIn g
the substitution we obtain

When combinedwith (1) this gives


l;,
(91;+IX ^Y)C#2 -(X2 Y)Ea2

whichtelLsus thereis a P of the desiredkind.


The next theoremgeneralizesP3.
TE`HoREm 6.12.
(a)(E.8)(xi, X2,yi, * ,Yk+l)[(XI) A) YI, *** -
Yk+l)4 ($xi xt)Ca] (k20:).

Proof. By 6.11 thereis a ,BIsuch that


(1) (tzl tXIXZ2)ef1- (1XI X2)'Ea.

16
The termis used heremerelyto rendersome aid to the intuitiveunderstandingof the.
a&cdome.
A SET OF AXIOMS FOR LOGIC

FromaxiomP3 and Theorem4.5, thereis a #2such that


(ux1, Z&X2, (yi, * * *, A/+1)) 2 . =. (& 1, LX,) 01.

By combiningthis equivalence with (1) and using 5.1, we have our theorem.
The next theoremgeneralizesP4.
THEEOREM6.13.

(a) (Eft) (y, *y Yk, Xi,* X X+l) [(ZI yl Y1X Yk *2 X%+i)ef -


(xI, * , x,.+I)ea] (k>O, n>O).
Proof. From P4 and 4.5 we know thereis a Pi such that
.
( 2(n-)X l,2(n-1) Yi, (Xa2, X,x"+l))fl _ (L2(n"l1x, (X, xn+i))ea.

But by 5.1 thisis


(Xi, y1) X2,* X a (Xi, X2) ... ) Xn+l)ea.
n41)01l

A k-folditerationof this processgives us our theorem.


The next theoremgeneralizesP5.
THEOREM 6.14.
(a)(Efl)(yi, * * *,kyk Xi, ** *, XnAR)[Y1, ***Yk, Xi, ***XXn+l)e -

Proof.6 For k (X, . (k >0, n2~t0).


* * *I Xn+l)ea]
1, the theoremfollowsfromP5 by replacingy by y and

'n
x by (xi, * , x, +1). For k>1, by iteration.
We now prove a seriesof theoremsleadingup to 6.3, the generalizationof P6.
THEOREM 6.21. (u)[(Ey)(u = t y) D (u, x)ea] (z)((kz, x)ea) (k20).
Proof. Assume the left-handside of the equivalenceand substitutetkzforu.
This gives
(Ey) (4kz= tLy) D (Lhz,X)ea.
Droppingthe truehypothesis,we have (tkz,x)ea; whence(z)((thz,X)ea). Now
assume the right-handside of the equivalence. From 3.7
(t ;y, X)ea . u = shy )
. (u, x)ea.
Whence fromthis and assumption,
U= Rky. D (U, x)ea,

and therefore (Ey)(u =ky). D. (u, x)ea.


THEOREM 6.22. (a, y)(Ef)(x, u)[(u, x)O. uea D (u, x)ey].
Proof. By P5 and P7 thereis a j31such that
(u, x)6-i uea.
S From here on we shall omit references to
4.5.
12 THEODORE HIAILPERIN
Whence

(u, x)epi D (u, X)e- uea D (u, x)ey.


From this and 4.35, we obtain our theorem.
THEOREM 6.23. (a)(E#)(x)[xO = txea].
Proof. From P5 we have a 6, such that
(y. tx)e8i LZEa.

Whence
(1) (y)((y, &X)e1) LZea.

By P6 we have a j2 such that


X0 - (y) ((y, &X)?1).

When combinedwith (1), this gives us a is of the desiredkind.


THEOREM 6.231. (a)(Efl)(x)[tzo XEaJ.
Proof. In P5 take y to be x and use 5.2-this gives us a gI such that
(1) XeIli xea.

From 6.23 we have a A2such that


&Xe?2 -2XE.

Whence,with (1),
&Xe12 C-Xa.

THEOREM 6.24. j (u) ((u, x)ea)].


(a) (EB) (x) [xep

Proof. For the reasons cited on the right-handside, we have O's such that:
(l) (LU, t)Ol P21
-(u X)ea

(2) (u)((u, &X)e2) (u)[(Ey)(u = ty) D (u, &x)epi] [6.22, P8]


-
(3) (u)((u, Lx)epl) [(2) 6.21]
(4) Xe3 (U))((U, &X)E02)
[
Xe#3-(U) ((U. XPea)
P61
THEOREM 6.25. (a)(Ed)(x)[xed =- (u)((&u, x)ea)].
[(4)
Proof. For the reasonscited,we have the a's:
(3)
(1) (u)((u, x)01) )[(Ey)(u=&y)
(u) (1) ]
Xe#2 (u)((Lu, X)ea)
whichgivesus our theorem.
THEOREM 6.251. (a)(Efl)(x)[xe =- x C a].
[6.22,
Proof. We have the following P8J
jO's:
(1) (LU, x)el Uex [(1) 6.24,
6.211

[IP91
A SET OF AXIOMS FOR LOGIC 13
(2) LUEJ% Uea [6.2311

(3) (LU, X)#3 IU02 [P5,P71


(4) Uea [(2)(3)]
(5) (LU, x)efl uex D uea [(1)(4) 4.351
(6) (u)((tU, X)e#4) x C a [(5)]

xeflb xC a [(6) 6.251


THEOREM6.252. (z) (EB) (X)[xefl. - . x = z].
Proof. Exactly lke that of 6.251, using 4.36 instead of 4.35.
THEOREM 6.253. (u)(uez. . u=z). [6.252,4.61
Thus &zis a class.
DEFINITION6.01. 1 forx (Ey) (x = ty).
THEOREM 6.26. (x)(xel . e . (Ey)(x=y)). [6.01, P8, 4.61
DEFINITION6.02. U1 for 1
Uk1) (k > 1),
Uk for X(XFl . X C
THEOREM 6.261. (X)(XeUk. . Xel * X C Uk-1) (k> 1).
Proof. That Uk is a class followsfrom4.32, 6.251 and 4.6.
THEOREM6.27. (u)(L UeUk) (k > 0)
Proof. By inductionon k. For k= 1, the theoremis an easy consequenceof
6.26. Assume the theoremtrue for k and considerthe case of k+ 1. From
assumption,
= LU*D * VUk).
(V)(V
Whence by 6.253, tUk+u C Uh. But by 6.26 we also have tk+ uel. Thus
k+uel .L u C U1).

(u) (t
THEOREM6.28. (v)[veUi. D . (Ew)(v = kW)] (k>0).
Proof. For k= 1 the theoremfollowsfrom6.26. As hypothesisof induction
assume the theoremtrue fork. Startingwithv = LU . v C Uk, we have
(UEV)(UEV D UEUk),

so that ueUk. Hence by hypothesisof induction


(Ew)(u = L*W).

From this and v = Lu we obtain


(Ew) (v = L
+1W).
Thus
(Eu)(v = LU) . v C Uk D. (Ew)(v =A;+ w),

which,in view of 6.26 and 6.261, is the desired conclusion. THEOREM6.29.


(u)(ueUk D (u, x)ey) . - (u)((L u, x)ey) (k>0).
14 THEODORE HAILPERIN

Proof.Assumetheleft-handsideoftheequivalence. Fromthis,by replac-


ing u by &ku,
D (&.ku,x)ey).
(u)(kuVEUt
Whence,by 6.27,
(u)((&ku,x)ey).
Now assumetheright-handside oftheequivalence. Fromthis
= ?u) D (v,x)ey,
(Eu)(v
and therefore,by 6.28,
veUb D (v, x)ey,
fromwhichwe obtaintheright-handsideoftheequivalence.
THEOREM 6.291. (a)(E#) (x)[xei (u)((l u, x)ea)] (k2O).
Proof. For k=0 and k= 1 thisreproduces6.24 and 6.25. We considernow
k> 1. In 6.22takea to be Ukand changeyto a. We thenhavea ,1 suchthat
(u, X)?i1i- UeUk D (u, x)ea.

Whence
(U) ((U. X)?1) .-- (U) (UzekA D) (U. x)ea)
(u)((Lku,X)ea). [6.29]
Whencombinedwith6.24,thisgivesus ourtheorem.
THEOREM 6.3. (a) (Et) (xi, * *,)[( n &kjX, *,lX")

(U)(( Iu, k1, * * * , &kx,)ea)] (jO0, n>0).

Proof. In 6.291take k to be j+2(n-1) and replacex by (l i,


XI)
Thenby 5.1 we havethetheorem.
THEOREM 6.4. (a)(EP)(x)[xeP --lxea] 6.23. (k10).
Proof. Forkl=0,takeP to be a; forkl>0 byinduction,using
THEOREM 6.5. (a)(EP)(x)t[laxed zeal (x2O)
.
Proof. Similarto 6.4, using6.231.
THEOREM 6.6. (x)(Ef)(y)[ye z ey].
Proof. We have thefollowingP's suchthat:
(1) =Uey [P9]
(lu, yOA-
(2) Lue2 U=x [6.252,6.2311
(3) (lu, y)Of U = X [(2) P5, P7]

(4) (lu, y)e-(uey .u= x) [(1)(3) 4.32]


(5) (Eu)((tu, y)34) (Eu) (uey . u = x) [(4)]
(6) = jy [(5) 3.4]
A SET OF AXIOMS FOR LOGIC 15
(7) yeam (Eu)((&u, y)e#4) [4.31,6.251

-xey [(6)(7)]

7. Stratification.In this section we derive some results about stratified


propositions. These resultsare needed forthe proofof our generalmetalogical
theoremabout the existenceof classes.
DEFINITION 7.01. A propositionp is said to be stratified non-negative
integralsuperscriptscan be attached to the variables in p in such a way as to
satisfythe followingconditions-7
1. If x is a variablead q is a part of p, thenall freeoccurrencesofx in q have
the same superscriptattached.
2. If q containsany freeoccurrenceof x, then forany occurrenceof (x)q or xq
the explicitlyindicatedoccurrenceof x is to have the same superscriptas the
freeoccurrencesofx in q.
3. In each occurrenceof the formxqy,Xeyq,xqeyor xqeyrin p, if the super-
scriptattachedto the explicitlyindicatedoccurrenceof x is n, then the super-
scriptattachedto the explicitlyindicatedoccurrenceof y is to be n+ 1, n, n+2,
and n+1 respectively.
DEFINITION 7.02. Inverselystratified.This is the same as the preceding
definitionexceptthat in 3, the superscriptattachedto x is to be attachedto y and
vice versa.
T`HOREm 7.1. If a proposition is stratified,thenso is any partofp.
Proof. The same alignment of integersto the variablesin p showsalso that
the part is stratified.
THE~oRY 7.2. If p is stratified,thenso X8 its abstractionlesstransformp'.
Proof. It is sufficientto show that the abstraction-removingtransformof p
is stratified.
Case 1. xeyqis replaced.
Here, in place of xzyq,we substitute
(1) (Ea) . xea. (y)(yea--q),
wherea is differentfiomeveryvariablein xeytq. If, now,to the variablesin the
part (1) we attach the same integersas in xeyqand to a the nextgreaterinteger
to that assignedto x, and in the rest of the transformthe same subscriptsas
before,thenthe transformwillsatisfythe conditionsforstratification,
Case 2. qEy is replaced.
Case 3. xqe4ris replaced.
These are treatedsimilarlyto case 1.
THEOREM 7.3. If a propositionis stratified,thenit is inverselystratified,and
conwersely.
Proof. If p is stratifiedthen we may attach integersto the variables as statedin
the definition. Let N be the largestsuperscript;in place ofeach super-
I
CompareRosser,this JOURNAL, IV 17.
16 THEODORE HAILPERIN

script i write N-i. This resultingassignmentshows that p is inversely stratified.


This same proofalso does service for the converse-merelyinter-changethe
words"stratified"and "inverselystratified."
DEFINITION7.03. In attachingintegersto the variablesin accordancewith the
definitionof "stratified,"the integerattachedto a variablewill be called its
type,relativeto thisparticularalignment of integers.
DEFINITION 7.04. If xi, - - *, x, is a set of variablesand if p is any inwersely
stratifiedproposition,we say the set of integersk, ** *, k. is a typeassignment
--
toxi, *, x, relativetop if,whenthe integerskl, **, k. are signed to the free
occurrencesofxi, s,Xx, as typesin p (someor all ofthex's may be absentfrom
p), the remainingvariables (and the bound occurrencesof xi, x,,) may be
aasignedtypesin accordancewiththe definitionof inversestratification. As
a consequenceof these definitionsand 7.3 we have:
THEOREM 7.4. If p is stratified,thenthereexistsa typeassignmentforany set
ofvariablesrelativetop.
THEOREM 7.5. If ki, ***, k,t 8 a typeassignmentto yi, ***, y, relaive to
- - k is a typeassignmenttoy, ,y. relativeto p.
(u)p, thenki,
THEOREM7.6. If ki, , ka is a typeassignmentto yi, ***, y, relativeto
- knis a to I relativeto or r.
Or, thenki, *-
, typeassignment yj,I *, q

Proofsforthe precedingtwo theoremsare easily obtainedfromthe definition of


type alignment.
THEOREM 7.7. If ki, *,k is a typeassignmentto yi, * y relativeto p, thenit is also
a typeassignmentto yj, .**, y, relativeto p', theab8tractiondes
transformofp.
Proof. In the proofof 7.2, we observethat the variables in the transform have
the same typesas in the originalproposition.
8. General existencetheorems. We now provea metalogicaltheoremto the
effectthat a stratifiedpropositiondeterminesan n-adicrelation-but one which is
not necessarilyhomogeneous. The inhomogeneityis representedby the
numberof L's precedingthe elementsin the orderedn-tupleswhichconstitute the
relation. This inhomogeneityis already illustratedby axiom P9, which may be
written
(E#)(x, y)[<Lx, L~y) x
emy]
Notice that the exponentsof the iotas run counterto the stratification.This is
the reasonfordefiningtypeassignmentin termsof inversestratificationrather
than stratification.
GENERAL EXISTENCETHEOREM8.1. If p is stratified,contains no ,3,and among
xl, **,*, x ,,
y , y, are containedall freevariablesin p, then
Y,
(XI, * , Xm)(E3)(yi >O. n> O) .
I p.(m.,

is a theorem,whereki, * *, k,.is any typeassignmentto yi, *** , y. relativeto p.


A SET OF AXIOMS FOR LOGIC 17

Proof. We shall provethe theoremwithp replaced by p', the abstractionless


transformof p. From this the theoremwill then be an immediateconsequence of p
p' (see 1.1) and Theorems7.2 and 7.7.
The proof proceeds by inductionon the numberof logical operatorsin p', where
by a logical operatorwe mean a stroke or a quantifier. We consider firstthe case
wherep' has no logical operators-in thiscase p' can only be of the formuev.

Assumingthe hypothesis-namelythat p' is stratified;containsno #; xi, ,


Xm, ...
yi, , yY contain all freevariables in p', and ki, ** *, knis a typeassign-
ment to Yi, * y, relative to p'-we consider the various possibilitiesfor
u and v. u and v are both x's.
Case 1.
By Theorem4.7 we have a , such that forany z,
(1) zef uev.

Replacingz by ( * *.. Xkflyn) in (1) gives us a iswiththe desiredproperty.


Case 2. u is some y, y, say, and v is an x.
By 6.5 thereis #Isuch that
(2) k ye131 yiev uev.

From here on the proofdivides into two cases, dependingon whether i<n or
i=n. In case i=n, Theorem6.14 gives us a #2 such that UC

Y...
v,2 ue
and in case i <n, P5 and P7 give us a 2 such that

& -UeV.

k NY tlYi+l)e#2
Whence by 6.12 and 6.14 we have a 3 such that
(k ")43 - uev.

In eithercase we have the requisiteB.


Case 3. v is some y,, u is an x.
From 6.6 we have a ,3such that
y,8 Uev,

and the restofthe proofnow followsthat of case 2.


Case 4. u and v are both y's.
Here uev must be of the formyieyjor yjeyiwherei<j; here i and j must be
distinct,otherwisep' would not be stratified. For yey,, P9 providesfor a ,
such that
(EyE,y,)6 uev, while foryjey,,

P9 followedby P7 providesfora ,3such that


18 THEODORE HAILPERIN

In eithercase, by 6.11, we have a ,1 such that


hi
(shi, f,?-u
whereki and k1are integerssuch that ks-kj is + 1 or -1 dependingupon which
ofthetwo cases we have. By 6.12, thereis a 2 such that
(4) h *s1 ^lyr)?2 -o hi
(4hiy, skfyj& hfjy+ls (&Yi, & OA
By 6.13 thereis a Ps such that
(5) h hYx)3 _ (&,";z, ,kiyi,...
& sI+ly,+i, ...&, ,abys0a

Finallyby 6.14 thereis a ,B such that


(6) (4 kyi,1 .. - ofkiy, * * * , k"Y.)
(y)04

Combiningequivalences(3), (4), (5), and (6), we have a ,B such that


(V

As hypothesisof induction,assume the theoremtrueforany propositionhav-ing h or


less logical operators,and considera stratifiedp' whichhas h+ 1 logical operators.
Clearlyp' is eitherof the formqlr or (u)&.
Case 1. p' is qjr. yn relativeto qlr. Then
Let ki, -*. ,. k. be any type assignmentto yi, ***,
by 7.6, k, - - - ka is a type alignment to yi, **, y. relativeto q and to r.
*

Since q and r are both stratified(7.1), we have by hypothesisof inductiona


0i and a 02 such that

(7) Y, ..., &4ys)% q,


Whenceby (7), (8) and P1 we have a jBsuch that

( X , &y*) qjr.
Y,

Case 2. p' is (u)&.


Let ki, - -
.

* , k.
be any type assignmentto yi, , y. relativeto (u)&. Then
by 7.5, k, **.*. kn is a type assignmentto yi, ***, y. relativeto s, and we can
finda j so thatj, *,**
ki,
k. is a typeassignmentto u, yi, , y. relativeto S.
Hence by hypothesisof inductionthereis a #Isuch that
&k & - 8

(MU, lY1, . ..*ayn)?


Whence
...,
(U)((Uu, ?wYi, &y)l)-- (U)8.

From this and 6.3 we obtain a #2such that


( Y,, -, '"Yt)e2 = (U)8.
, X
This completesthe proofof 8.1.
A SET OF AXIOMS FOR LOGIC 19
GNERALEXISTENCE THEOEM 8.2. If p is and contain no A, then
stratified
(EP)(y)[yO- p]
is a theorem.
This is Quine's R3'.
Proof of 8.2. Since p is stratifiedthereis a type alignment ki, ***, k,,to
V*I-, y. relativeto p (see 7.4.). In 8.1 take n to be 1 and take yi to be y.
Then we have a #Isuch that
* p
lye1--

and 6.4 gives us a 62 such that


Y012 - 01Jefll

Thus we have a 12 such that

9. Equivalence withQuine's system. With the demonstrationof8.2 we have,


exceptfora minordetail, shownthat Quine's New foundations is derivablefrom our
system. The minordetail is the differentpostulateforidentity. However Quine's
postulate is seen to be an immediateconsequence of our definitionof identity.
Turningnow to the derivationof our systemfromQuine's, we note that our PId is
his definitionof identityand that our axioms P1, P6, and P8 are
immediateconsequencesof his R3'-in each case the formula(or ratherthe
abstractionlesstransformof the formula)is easily verifiedto be stratified. For
the remainingaxioms P2-P5, P7, and P9, the followingsix classes provide,
respectively,for the existenceof a '3 mentionedin these axioms-in each case
stratificationcan be verified:
xt[(Eu,v) . x = (&u,w) . (u, v)ea]
x[(Eu, v, t) . x = (u, v, t) . (u, v)ea]
=
1[(Eu, v, t) . z (u, t, v) . (u, v)eaj
=
'[(Eu, v) . x (v, u) . uea]
x[(Eu, v) . x = (u, v) . (v, u)eaJ
x[(Eu, v) . z = (Lu,V) . Uew]
This completesthe proofof the equivalenceof our set of axioms withQuine's
Newfoundations.
CORNELL UNIVERSITY

You might also like