Algebraic Analysis of Many Valued Logics

Author(s): C. C. Chang
Source: Transactions of the American Mathematical Society, Vol. 88, No. 2 (Jul., 1958), pp.
467-490
Published by: American Mathematical Society
Stable URL: http://www.jstor.org/stable/1993227 .
Accessed: 06/08/2013 01:14

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.

American Mathematical Society is collaborating with JSTOR to digitize, preserve and extend access to
Transactions of the American Mathematical Society.

http://www.jstor.org

This content downloaded from 147.26.11.80 on Tue, 6 Aug 2013 01:14:03 AM

All use subject to JSTOR Terms and Conditions
 ALGEBRAIC ANALYSIS OF MANY VALUED LOGICS(')
BY
C. C. CHANG
This paper is an attemptat developinga theoryof algebraic systemsthat
would correspondin a natural fashion to the No-valued propositional cal-
culus(2). For want of a better name, we shall call these algebraic systems
MV-algebras where MV is supposed to suggest many-valued logics. It is
known that the classical two-valued logic gives rise to the study of Boolean
algebras and, as can be expected, every Boolean algebra will be an MV-
algebra whereas the converse does not hold. However, many results for
Boolean algebras can be appropriatelycarriedover to MV-algebras,although
in some cases the proofsbecome more subtle and delicate. The motivation
behind the present study is to find a proof of the completenessof the No-
valued logic by using some algebraic results concerningMV-algebras; more
specifically,it is known that the completenessof the two-valued logic is a
consequence of the Boolean prime ideal theorem and we wish to exploit
just some such correspondingresult for MV-algebras(3). It will be seen that
our effortin duplicating this result is only partially successful. In the first
foursections of this paper we presentvarious theoremsconcerningboth the
arithmeticin MV-algebras and the structureof these algebras. In the last
section we give some applications of our resultsto the study of completeness
of NO-valuedlogic and some related topics. We point out here that the treat-
ment of MV-algebras as given here is not meant to be complete and exhaus-
tive.
1. Axioms of MV-algebras and some elementaryconsequences. An MV-
algebra is a system (A, +, *, -, 0, 1) whereA is a nonemptyset of elements,
Oand 1 are distinctconstant elementsof A, + and are binary operations
on elementsof A, and - is a unary operation on elementsof A obeying the
followingaxioms. (We assume here, of course, that A is closed under the
operations+, , and-.)
Received by the editors January 8, 1957.
(1) The researchand preparationof this paper were supported in part by the United States
Navy under Contract No. NONR 401 (20)-NR 043-167 monitored by the Officeof Naval
Research; reproductionin whole or in part is permittedforany purpose of the United States
Government.
(2) We have in mind specificallythe Ho-valued propositional calculus developed in [3]
and [5 ]; the axioms forthis calculus are known as the Lukasiewicz axioms.
(3) It is of course known that the completeness of the Lukasiewicz axioms for 9o-valued

logic has been proved in [5 ]. This resultwas also stated on p. 240 in [8]; however,it was never
published.

467

This content downloaded from 147.26.11.80 on Tue, 6 Aug 2013 01:14:03 AM

All use subject to JSTOR Terms and Conditions
468 C. C. CHANG [July

Ax. 1. x+y=y+x. Ax. 1'. x.y=y x.

Ax. 2. x+(y+z)=(x+y)+z. Ax. 2'. x.(y.z)=(x.y) z.
Ax. 3. x+x=1. Ax. 3'. x2=0.
Ax.4. x+1=1. Ax.4'. x0=0.
Ax.5. x+0=x. Ax.5'. x 1=x.
Ax. 6. [x+y] =xty. Ax. 6'. [x-y]-=x+y.
Ax. 7. x= [x]-. Ax. 8. 0=1.
In order to write the remainingaxioms in a compact formwe introducethe
followingdefinition.
DEFINITION 1.1. XVY=(X.Y)+Y.
XAY = (X+Y) *Y.
Ax. 9. xVy=yVx. Ax. 9'. xAy=yAx.
Ax. 10. xV(yVz)=(xVy)Vz. Ax. 10'. xA(yAz)=(xAY)Az.
Ax. 11. x+(yAz)=(x+y)A(x+z). Ax. 11'. x.(yVz)=(x y)V(x-z).
It is clear that this axiom systemis not the most economical one; theyare
given in the above formfortheirintuitivecontents.It is also clear that, just
as in the case of Boolean algebras, there is a duality involvingthe elements
0 and 1, the operations + and *, and the operations V and A. Thus any
theoremstated will have as an easy consequence fromthe axioms its dual.
We make the convention,as in the case of ordinaryarithmetic,that * shall
be more binding than +, and, by the associative laws, we shall omit the
usual parenthesesin expressionsof the formx+(y+z) etc. Since the mean-
ings of the axioms are clear we shall usually use them without mentioning
themspecifically;this is especially true forthe firstten axioms. In the follow-
ing theoremswhenever the variables x, y, z, * * * occur they are assumed
to be the elementsof some fixedMV-algebra A.
THEOREM 1.2. (i) xVO=x=xAl, xAO=O, and xVl=1.
(ii) xVx=X=XAx.
(iii) [xVy]-=fxAyand [xAy]-= xVy.
(iv) xA(xVy)=x=xV(xAy).
(v) If x+y=O, thenx=y=O.
(vi) If xy=l,thenx=y=l.
(vii) If xVy = 0, thenx = y= 0.
(viii) If x AY= 1,Pthenx = y= 1.
Proof. (i)-(iii) are obvious from1.1 and the axioms. By the same token
xA(XVy) = (yVx) Ax= (y*x+x) Ax = (y*x+x+x) x= (y x+1) x= 1 x=x.
The other equality of (iv) follows fromduality. If x+y=O, then O=XAO
=xA(x+y)=(x+O)A(x+y)=x+(OAy)=x+O=x. Thus, also y=O.
(vi) followsfrom(v) by duality. If xVy=O, then by 1.1, (x.y)+y=O hence
by (v) y = 0 and also x = 0. (viii) again followsfrom(vii) by duality.
We introduce an inclusion relation : among the elements of an MV-
algebra:

This content downloaded from 147.26.11.80 on Tue, 6 Aug 2013 01:14:03 AM

All use subject to JSTOR Terms and Conditions
19581 ALGEBRAIC ANALYSIS OF MANY VALUED LOGICS 469

DEFINITION 1.3. x y if, and only if, xVy=y. As usual the notation
x<y shall mean x?y and x y.

THEOREM 1.4.
(i) O<x<1.
(ii) x <x.
(iii) If x ? y and y < z, thenx < z.
(iv) If x?y and y<x, thenx=y.
(v) x?y if and onlyif xAy=x.
(vi) x<y if and onlyif y<x.
Proof. (i)-(iv) are obvious by 1.1-1.3 and the axioms. (v) follows im-
mediatelyfrom1.2 (iv), and (vi) followsfrom(v).
THEOREM 1.5 (MONOTONY OF V AND A). If x_y, thenxVz?yVz and
xAz -<yAZ.
Proof. If x_y, then by 1.1 and 1.4(v), xAy=x and xVy=y. Thus
(xVz)V(yVz)=xVyVzVz=yVz and (xAz)A(YAz)=xAyAzAz=xAz.
This implies xVz?yVz and xAz?YAz by 1.1 and 1.4(v).
THEOREM 1.6. XAY?<x?<xVY.
Proof. By 1.2(i), 1.4(i), and 1.5.
THEOREM 1.7. If x<y and x'<y', thenxVx'<yVy' and xAx'<YAY'.
Proof. By 1.5, xVx'<xVy'<yVy'. SimilarlyxAx'?xAy'<yAy'.
THEOREM 1.8 (MONOTONY OF + AND *). If x<y, thenx+z<y+z and
x*z_?y z.
Proof. We note again that if x?y then xAy=x and xVy=y. By Ax. 11,
(x+z)A(y+z)=(xAy)+z=x+z. Hence x+z<y+z. Similarly,by Ax. 11',
x.z<y.z.
THEOREM 1.9. X.Y?_X?<X+Y.
Proof. By Ax. 5, Ax. 5', 1.4(i), and 1.8.
THEOREM 1.10. If x?y and x'_y', thenx+x'?y+y' antdx.x'<y.y'.
Proof. By 1.8.
THEOREM 1.11. The relation ? is a partial orderingrelationamong the
elementsof A. The elementsxVy and xAy are respectively the l.u.b. and the
g.l.b. oftheelementsx and y withrespectto theordering<.
Proof. By 1.4, 1.6, and 1.7.
Putting 1.6, 1.9, and 1.11 together,we obtain
THEOREM 1.12. X.Y?<XAY <x 5xVY ;x+Y.

This content downloaded from 147.26.11.80 on Tue, 6 Aug 2013 01:14:03 AM

All use subject to JSTOR Terms and Conditions
470 C. C. CHANG [July

The next theoremwill be very useful.

THEOREM 1.13. Thefollowingconditionsare equivalent:
(i) x<y, (ii) y+x=1, and (iii) xy-=O.
Proof. Clearly (ii) and (iii) are equivalent. If x y0=, then by 1.1, xVy
=xjy+y=O+y=y and hence x<y. If x<y, by 1.10, 1=x+x<y+x. But
y+? < 1, hence y+x = 1.
We have the followinginterestingcancellation law.
THEOREM 1.14. If x+z=y+z, x<z, and y<z, thenx=y.
Proof.By Ax. 9', x=x 1=x (xt+z)=z (x+z)=z (y+z)=y (y+z)=y1i
=Y.
The followingtheoremis motivated by 1.4(v) and 1.13.
THEOREM 1.15. Thefollowingfour conditionsare equivalent:
(i) x?+y=y.
(ii) x.*y=x.
(iii) yvjc=l.
(iv) xA9=O?
Proof. We shall prove the equivalence of (ii) and (iii). Then by duality
the equivalence of (i) and (iv) will follow.Clearly (iii) is equivalent with (iv).
If x.y=x, then xVy=x+(x y)=x+x=1. If xVy=1, then by Ax. 11',
x=x* 1 =x* (xVy)=x Vxy=OVx y=x y.
THEOREM 1.16. The followingconditionsare equivalent:
(i) x+x=x.
(ii) xc x=x.

(iv) x *x = x.
(v) xVx= 1.
(vi) xAX=O.
Proof. By 1.15.
1.16 points out the interestingfact that in an MV-algebra the set of
elementsB which are idempotentwith respect to the operations + or * are
preciselythose elements which satisfythe law of the excluded middle with
respect to the operations V or A. Furthermore,1.16 leads to the following
THEOREM1.17. Let B be theset of elementsx ofA such thatx+x =x. Then
B is closed underthe operations+, *, and - and wherex+y=xVy and x y
=xAy forx, yCB. Furthermore, thesystem(B, +, ., -, O, 1) is notonlya sub-
algebraof A butis also thelargestsubalgebraof A whichis at thesame timea
Boolean algebrawithrespectto thesame operations+, *, and -.
Proof. The fact that B is closed under the three operations followsim-

This content downloaded from 147.26.11.80 on Tue, 6 Aug 2013 01:14:03 AM

All use subject to JSTOR Terms and Conditions
1958] ALGEBRAIC ANALYSIS OF MANY VALUED LOGICS 471

mediatelyfrom1.16. Clearly B is also closed under the operations V and A.

Now, in orderto prove that B is a Boolean algebra we notice that ifx, ycB
then, since x?xVyC B and y?_xVyCB, x+y?<(xVy)+(xVy)=xVy.
But by 1.12 xVy<x+y. Thereforex+y=xVy and, similarly,x y=xAy.
Hence the elementsof B satisfyAx. 11 and Ax. 11' with V and A replaced
by + and * respectivelyand, as it is known, B thus becomes a Boolean
algebra. If C is a subalgebra of A which is also a Boolean algebra withrespect
to the operations +, , and -, then every element of C must satisfy the
identityx+x=x. Hence CCB and the theoremis proved.
From 1.16 and 1.17 it is seen that withrespectto the operations +, , and
- the distinguishingfeaturebetween an MV-algebra (A, +, , -, 0, 1) and a
Boolean algebra is the lack of the idempotent law x+x=x, whereas with
respect to the operations V, A, and - the differencebetween the system
(A, V, A, -, 0, 1) and a Boolean algebra is the lack of the law of theex-
cluded middle xV5?= 1. We mightmentionhere that, while various general-
izations of the Boolean algebra which do not satisfythe law of the excluded
middleare known (e.g., all kind of lattices), thereare veryfewgeneralizations
of the Boolean algebra where the idempotentlaw does not hold. It is also
knownthat the study of Boolean algebras can be subsumed underthe general
theoryof rings,i.e., the study of the so-called Boolean rings.This transforma-
tion is due to certain nice propertiesof the symmetricdifferenceoperator in
Boolean algebras. We shall see fromthe later sections that there will be, in
general,no such resultsfor MV-algebras.
To conclude this section we introducesome general proceduresto obtain
new MV-algebras fromthose already known(4).
Given an MV-algebra (A, +, *, -, 0, 1), we say that B is a subalgebra
of A if BCA, 0, 1CB, B is closed under the operations of +, *, and -. A
system(B, +, *, -, 0, 1) is a homomorphic imageof A (or A is homomorphic to
B) if thereis a mappingf of A onto B such that f(O) = 0, f(1) = 1, and f pre-
serves the three operations +, *, and -. WVesay that the functionf is a
homomorphism of A onto B. If the functionf is one-to-one,thenf if an iso-
morphismof A onto B. In this case we say that the systems(A, +, , -, 0, 1)
and (B, +, *, -, 0, 1) are isomorphic.Given a collectionof MV-algebras Ai,
iC I, we denote by Pie,Ai the cartesian (or direct) product of the sets Ai.
We denote by (PiE1Ai, +, ., -, 0, 1) the cartesian (or direct)productof the
algebras A i, i C I, where the element0 is the functionf such thatf(i) = 0 for
each iGI, the element 1 is the functionf such thatf(i) = 1 foreach iCI, the
addition of two functionsf and g shall be the functionh such that h(i) =f(i)
+g(i) for each iCI, and the product and converse of functionsare defined
analogously. Due to the formof the axioms Ax. 1-Ax. 11, we see at once that
the followingis true.

(4) can be foundin theirmostgeneralformin [7].

The notionswe are aboutto introduce

This content downloaded from 147.26.11.80 on Tue, 6 Aug 2013 01:14:03 AM

All use subject to JSTOR Terms and Conditions
472 C. C. CHANG [July

THEOREM 1.18. A subalgebraof an MV-algebrais an MV-algebra,a homo-

morphicimage of an MV-algebra is an MV-algebra,and thedirectproductof
MV-algebrasis an MV-algebra.
2. Examples of MV-algebras. The firstand most importantexample of an
MV-algebra is the algebra L obtained by consideringthe Mo-valuedproposi-
tional calculus. We use here extensivelythe results and notation in the first
three sections of [5]; the formulasand theoremsof [5] are referredto by
their numbers in parentheses. Let us recall that the formulasin this par-
ticular logic are built up of denumerably many statement variables
X1, X2, . . ., Xn, . . . with the two operations C and N in the following
manner:
(i) Every statementvariable is a formula.
(ii) If P is a formulathen NP is a formula.
(iii) If P and Q are formulasthen CPQ is a formula.
The particularfouraxiom schemas are listed below(5):
A.1. CPCQP.
A.2. CCPQCCQRCPR.
A.3. CCCPQQCCQPP.
A.4. CCNPNQCQP.
The symbol f-P is introducedto mean that the formulaP is provable from
A.1-A.4 using only modus ponens, i.e., if P and CPQ then Q. The elements
of our MV-algebra L shall be equivalence classes of formulasdeterminedby
the equivalence relation _; we let P- denote the equivalence class one of
whose representativesis the formulaP. It is clear that [P/--] = [Q/-] if
and only if F-P=Q. The operations +, , - and the elements0, 1 are defined
as follows:
[P/-] + [Q/3]] [BPQ/I=J.
[P/-I [Q/_ ] = [LPQ/I].
[P/-] = [NP/I].
[P/I] = 1 if and only if F P.
[P/-] = 0 if and onlyif F NP.
It can be proved without difficultythat these operations on equivalence
classes of formulasare independentof the representativesand, hence, are
well-defined.In orderto see that the system(L, +, , -, O, 1) is an MV-alge-
bra, we note that Ax. 1 and Ax. 2 are given by the commutativityand asso-
ciativityof B, Ax. 3 by (3.1) and (1.8), Ax. 4 by (3.32), Ax. 5 by (3.45), Ax. 6
by (3.8) and (3.4), and Ax. 7 and Ax. 8 by (3.4). We note also that by (3.4),
(1.6), and (1.9), the definitionsof the operations A and V are such that
(5) The proofthatA.5 is derivablefromA.1-A.4 can be foundin [2]. This factwas also
noticedindependentlyby C. A. Mereditha fewyearsago, and was publishedin [4].

This content downloaded from 147.26.11.80 on Tue, 6 Aug 2013 01:14:03 AM

All use subject to JSTOR Terms and Conditions
19581 ALGEBRAIC ANALYSIS OF MANY VALUED LOGICS 473

[P/I] V [IQ/-]- [APQ/-]

and
[P/I] A [Q/-] = [KPQ/-].
Thus Ax. 9, Ax. 10, and Ax. 11 are given by the commutativityof A, the
associativity of A, and the distributivelaw (3.44) respectively.This takes
care of the unprimed axioms; as for the primed axioms, they follow easily
fromthe duality. Hence L is an MV-algebra. In particular,we see that Ax. 1-
Ax. 11 followfromA. 1-A. 4 upon a propersubstitutionof the operations C
and N by the operations +, *, and -. On the otherhand, makingthese same
substitutions,we see that A. 1-A. 4 can be transformedequivalentlyinto the
following(subject to the condition that [x]-=x, i.e., Ax. 7):
A. 1'. x+(y+x)=1.
A. 2'. (x +y)-+[(y+z)-+G(x+z)]=1.
A. 3'. (xVy)-+ (yVx) = 1.
A. 4'. (x+y)-+(y+x) = 1.
Clearly A. 1' and A. 3' are derivable fromAx. 1-Ax. 11. A. 2' is simply,after
a transformationusing the axioms and 1.1, (jVy)+(yVz)=1, which is de-
rivable from1.6 and 1.8. A. 4' can be transformedinto (xVy) +y=1 which
is again derivable from1.6 and 1.8. Hence we conclude that the two sets of
axioms A. 1-A. 4 and Ax. 1-Ax. 11 are equivalent underan appropriaterela-
tionshipbetween C and N and +, , and -.
Anotherclass of examples of MV-algebras is obtained by consideringany
set S of real numbersbetween 0 and 1 where S satisfiesthe following:
(i) OESand1ES.
(ii) If x, yES then min (1, x+y) CS.
(iii) If x, yES then max (0, x+y-1)ES.
(iv) If xCS then 1-xES.
If we now define for the elements x, yCS, x+y=min (1, x+y), x y
-max (0, x+y-1), and x=1 - x, then therewill be no difficulty in checking
that the system (S, +, *, , 0, 1) is an MV-algebra. We point out here that
the operations V and A on S are simply
x V y = max (x, y) and x A y = min (x, y),

and the relation < is simply the natural orderingof real numbers.
Various special sets S may be taken which satisfy conditions (i)-(iv),
e.g., S= {0, 1 }, S=the interval [0, 1], S=the set of all rational numbersbe-
tween 0 and 1, and S= the set of all rationalsof the formn/mforsome fixed
positive integerm and 0?<n<?rm.For each positive integerm we let S(m)
denote this last set of numbers and we notice that the operations +, *,
and - in the system (S(m), +, , I, 0, 1) are respectively

This content downloaded from 147.26.11.80 on Tue, 6 Aug 2013 01:14:03 AM

All use subject to JSTOR Terms and Conditions
474 C. C. CHANG [July

[p/m]+ [q/m]= min (1, [(p + q)/m]),

[p/m] * [q/m] = max (0, [(p + q -)/m]),
and

[p/m]- = [(m - p)/m].

Finally, a somewhat special model of MV-algebras is obtained by conI-

sideringthe followingset C of formalsymbols:
O, c,c + c, c + c+c, .

1, 1 - c, 1 - c-c, 1 - c - C - C

For abbreviation, we define Oc=0 and n.c=c+c+ +c n-times and

1-0 c = 1 and 1-n *c = 1-c-c- * --c n-times.In terms of this conven-
tion the rules of addition, multiplication,and complementationare written
down as follows:
(i) Ifx=nc and y=mc thenx+y= (m+n) c; ifx=n*c and y=l -mc,
then x+y=1 if m<n and x+y=1-(m-n).c if n<m; if x=1-m c and
y=n.c, then x+y=1 if m<vnand x+y=1-(m-n)-c if n<m; if x=1-nc
and y = 1-m c, then x+y = 1.
(ii) If x = n *c then x = 1-n *c; if x = 1-n cc then x = n *c.
(iii) xy = [+]
One can check easily that the system (C, +, , -, 0, 1) thus defined is an
MV-algebra by showing that each of the axioms Ax. 1-Ax. 11 is satisfied.
We can see this fact more simply if we realize that any finitesubset of C
can be embedded isomorphicallyinto some MV-algebra S(m) for a suffi-
cientlylarge m. The inclusionrelation < in the algebra C can be described
as follows: x<y if, and only if, one of the conditions below is satisfied:
(i) x=n*candy=1-m.c. (ii) x=n candy=m.cwhere n<m. (iii) x=1-n c
and y = 1 - m*c wherem < n. It will be seen in the followingsections that this
MV-algebra C will provide a crucial counterexample.
As has already been mentionedat the end of ?1, any subalgebra, homo-
morphicimage, or direct product of the above given examples will again be
an MV-algebra.
3. Some deeper arithmeticalconsequences of the axioms.
In this section we continue the investigationstarted in ?1. It turned out
that the identityx.9+y=y.x+x, i.e., Ax. 9, will be used quite frequently
in the followingtheoremsand we shall not cite Ax. 9 every time it is used.
THEOREM 3.1. y (x+z) <x+y z.
Proof. [y (x + z)] + x + yz = 9 + xz + x + yz = (9 + y.z)
+(x+x z) = (z+9 z) +(z+z x) = (z+z) + Y+zx = 1.
Hence by 1.13 the theoremfollows.

This content downloaded from 147.26.11.80 on Tue, 6 Aug 2013 01:14:03 AM

All use subject to JSTOR Terms and Conditions
1958] ALGEBRAIC ANALYSIS OF MANY VALUED LOGICS 475

THEOREM3.2 (6). If xA9-O, thenx+y z=y. (x+z).

Proof. Letting w= x (9+z) +y (x+z), we see that by 3.1 it is sufficient
to prove that w=1. Now,
x+ w= (x+ x(9 + z))+ y (x+ z)
(1) - + z + x*y*z+ y (x + z)
- + x y z + (y + y (x + z))
- + x.y.z + x + z + 9 x z

and
'1+ w = 9+ y. (x + z)) + x (9+ z
= x + z + Y9 z + + xyz
(2) Z
s+ 9 + (x + x.((+Z

= 1. z9z+

From (1) and (2) we obtain

(3) (x + w) A ( + w) =1.
By Ax. 11, (3), and the hypothesis
1 = (x A 9) + w = w
and the proofis complete.
THEOREM 3.3(7). (x+y)V(y+x) =1.
Proof.
(x+9)V(y+ x) =
(x[+(+))[(y+?)-]+y+x
= [(x + 9) 9x + xt]+ y
= (x + 99+ xc[(x+ 99]+ y
= ((x + y+y) + x (xy + y)
= x + y + y.(x + 9) + x (x y + y)
= y+yx y+ (x+x (x.y+y))
= 9 + yxy + xgy+ y + x.(x y + y)-

This is simply Theorem 3.2 of [5].

(6)

This is an algebraic version of the proofof A.5 in [2 ]. We point out here that most of
(7)
the resultsof [5] can be given algebraic proofsas in Theorems 3.2 and 3.3. Similarly,most of
our resultshere can be translatedinto the notation and language of [5 ].

This content downloaded from 147.26.11.80 on Tue, 6 Aug 2013 01:14:03 AM

All use subject to JSTOR Terms and Conditions
476 C. C. CHANG [July

THEOREM3.4. If xVy=1, thenx xVy y=1.

Proof.
(1) x xVy y = x*x+ (x+ x) yy.
By using the hypothesisx?Ay=O and 3.2 twice,
(2) (x+ x) yy= (x+ xy) y=-+x y y.
(1) and (2) yield
xxV + ) +
y-y= (xXxyy = (xKf + x) + y y
(3)= x + (x + x yyy = -x + y.y + x.(9 + 9).
Again by the hypothesisand 3.2, x (9+9) =x 9+. Hence (3) leads to
x.x V yy = x x +y y+ x9 +
= x x+ x9+(y y+Y)
= X-X + x5Y + 9Y- + y
= x + Yy + (x.Y + y)
-txx +9+ 9+ (x V y)
= 1,
which is the desired conclusion.
We introducethe followingdefinitions:
DEFINITION 3.5. (i) O x=O and (n+1) x=n x+x.
(ii) x? = 1 and xn+l = (x8) x.
DEFINITION 3.6. The order of an element x, in symbols ord(x), is the
least integerm such that m x= 1. If no such integerm existsthenord(x) = oo.
We obtain easily from3.5 by inductionthat
[n.x]- =
tn' [Xn]- = n x, m (n.x) = (m n). x,
X(m+n) = (xm).(xn), and X(m n) = (Xm)n.
THEOREM 3.7. If xVy=1, then xnVyn=1 for each n.
Proof. By an easy induction from3.4, we see that the hypothesisleads
to x(2m)Vy(2m)= 1 for each m. Since for each n there exists an m such that
n < 2m,we see that by 1.9, X2?< Xn and y2m<yn. Using thisand 1.8 we obtain
immediatelyxnVyn= 1.
THEOREM 3.8. If ord(x y) < oo, then x+y = 1.

Proof. By the hypothesis,forsome n_ 1, n. (x y) = 1. Hence (x+y) n=o.

By 3.3
(x+y)V(+y) = 1

and by 3.7

This content downloaded from 147.26.11.80 on Tue, 6 Aug 2013 01:14:03 AM

All use subject to JSTOR Terms and Conditions
1958] ALGEBRAICANALYSISOF MANY VALUED LOGICS 477

(X + y)n V (XC + y)n = 1.

Thus (x+y)n=1. By an easy induction using 1.2 (vi) we see that this last
leads to x+y = 1.
THEOREM 3.9. If ord(x) >2, thenord(x x) = oo.
Proof. By 3.8.
DEFINITION3.10. An MV-algebra A is locally finiteif,and only if,every
elementof A differentfrom0 has a finiteorder.
DEFINITION 3.11. An MV-algebra A is linearly ordered if, and only if,
forevery x, yEA, eitherx<y or y_x.
THEOREM3.12. EverylocallyfiniteMV-algebra is linearly ordered.
Proof. Let A be locally finiteand let x, yEA. We wish to show that
eitherx +y = 1 or x +y = 1. If x +y ? 1, then x*y 5O. Hence, by the assump-
tion on A, ord(x.y) < oo. Now it followsfrom3.8 that x+y= 1.
The example C given in ?2 furnishesa counterexampleto the converse
of 3.12. Clearly C is linearlyorderedbut ord(c) = oo. It is also easily seen that
the only Boolean algebra whichis linearlyorderedis the two-elementBoolean
algebra consistingof 0 and 1 alone.
THEOREM 3.13. If A is linearly ordered,then x+z=y+z and x+zo1
impliesx=y.
Proof. Since x+zO1 and y+zO1, we see that z$x and z$y. Since A is
linearly ordered, this means x<z and y<z. We obtain the conclusion im-
mediatelyfrom1.14.
In the remainingpart of this section we shall study intensivelysome
properties of linearly ordered MV-algebras. We shall firstintroduce the
functiond(x, y) which plays the role of a distance function.The functiond
is definedas follows:
d(x, y) = -y + y*x.
The reader will recognizethat in Boolean algebras the elementd(x, y) is
simply the familiarsymmetricdifferenceof x and y. We note here that in
Boolean algebras there is another way of definingthe symmetricdifference
of x and y, namely,d'(x, y) = (x+y) (x+y). However, the operation d' thus
definedforMV-algebraswould not have the desiredpropertythat d'(x, x) = 0.
For in the algebra S(2) we easily see that d'(1/2, 1/2) 50.
THEOREM3.14. (i) d(x, x) = 0, d(x, y) = d(y, x), d(x, y) =-d(, y), d(x, 0)
=x, and d(x, 1)=x.
(ii) If d(x, y) = 0, thenx = y.
(iii) d(x+u, y)+dd(y,)).
(iv) d(x+u, y+v) -<d(x, y) +d(u, v).
7 _ 7
(v d( u, y v /5d(x yI 7du v

This content downloaded from 147.26.11.80 on Tue, 6 Aug 2013 01:14:03 AM

All use subject to JSTOR Terms and Conditions
478 C. C. CHANG [July

Proof. (i) is obvious fromthe definition.(ii) followsfrom1.2 (v). In order

to prove (iii), we have to show that
(1) x-2 + x z < x y + x y + Y z + 2-y.

Now
[z.-fd + fo?y+ Y.z = 2 + x + X- + Y*z
= (x + -y) + ( + yz) = y + x-Y + y + y- 1
hence
(2) z. _xy
? + yz.
In an entirelysimilarfashion,
(3) x-2 < x-y+yY.
(2) and (3) yield (1) and (iii). Condition (iv) is essentially
(4) (x + u) (y + v) + (x + u) (y + v) <? xy + y x + u v + v u
Now
(x + u + y v) + x y + uiv = (x + x y) + y v + (ut+ ui v)
= (y+x Xy)+y v-~+ (v+u v-) =x- y+ u v+ v+ (Y + Y-
= x9 + u- + v+ + v= 1
which implies
(5) (x+u)- (y+v) x y+u v.
Similarly,we derive
(6) (x + u) (y + v)- x + ~3u.
(5) and (6) give (4) and (iv). As forcondition (v), we note that
d(x u, y v) = d([Xt+ U]-, [y + -) = d(.t + ut,y + t)
< d(x, y) + d(ui,v)= d(x, y) + d(u, v).
The theoremis proved.
3.14 tellsus that many of the propertiesof the symmetricdifference opera-
tion in Boolean algebras can be carriedover to MV-algebras. However, two
of the most importantpropertiesof symmetricdifference,i.e., associativity
and the distributionof multiplicationover symmetricdifference,fail. This
can be seen, for instance, in the MV-algebra S(3), where d(1/3, d(2/3, 1))
#d(d(1/3, 2/3), 1), and in the MV-algebra S(2), where (1/2).d(1/2, 1)
7z?d(1/2. 1/2, 1/2. 1). This is the main reason why we pointed out in ?1 that
the study of MV-algebras cannot be subsumed underthe theoryof ringswith
the operationd interpretedas the ringaddition. However, this does not pre-
clude the possibilitythat some other definitionof ring addition may work.

This content downloaded from 147.26.11.80 on Tue, 6 Aug 2013 01:14:03 AM

All use subject to JSTOR Terms and Conditions
1958] ALGEBRAICANALYSISOF MANY VALUED LOGICS 479

In this connection,we easily see that the operation d' introducedearlier is

also such that d'(1/3, d'(2/3, 1)) $d'(d'(1/3, 2/3), 1) in S(3) and (1/2)
*d'(1/2, 1)ed'(1/2.1/2, 1/2. 1) in S(2).
THEOREM 3.15. If x<y, thend(x, y)=x y and x+d(x, y)=y.
sy. Also, ifx<y, theny=xVy
Proof. If x<y thenx y=0. Thus d(x, y) =x
=x+x y=x+d(x, y).
THEOREM 3.16. (i) If n y ? x ? (n + 1).y, then d(x, n y) < y and
d(x, (n+l) y)<-y
(ii) If n y<x<(n+1) y, thend(x, n y)<y.
(iii) If n y < x < (n + 1) .y, thend(x, (n + 1) *y) <y.
Proof. If n.y<x<(n+1).y, then (n.y).x=O and x.(y(n+l))=O. Also
d(x, n. y) =x. yn and d(x, (n+1)*y)=fx ((n+t).y). Fromx. (y(n+1)) = 0, we
obtain (x. yn). y=0, thus d(x, n.y)<y. From x.(n.y)=O and 3.1, we see
that d(x, (n+l).y)=xc.(y+n.y)<y+x.(n.y)=y. Hence (i) is proved. If
d(x, n.y)=y, then x.yn=y and x=x+O=x+(n.y).C=n.y+yn x=n.y+y
= (n + 1) .y. This proves (ii). If d(x, (n + 1) .y) = y, then x *((n + 1) y) = y and
+ 0 = x + x y(n+l) = 9(n+1) + x ((n + 1) .y) = yn+l + y = yn. y + Y
=yn+y*(ny) yn. By 1.4(vi), this means x<n*y. This proves (iii). The
proofis complete.
THEOREM 3.17. If A is linearly ordered and ord(y)= m < oo, then
d([n y]-, (m-n) y) <y for each n < m.
Proof. The theorem is clearly true if n = m. Therefore let us assume
n<m. In this case (m-n).y+n y=1, hence [n.y]-<(m-n).y. On the
otherhand, if [n.y]-<(m-n-1) y, then 1=(m-n-1) y+n y=(m-1) y
which contradictsord(y) =m. Thus we have (m-n-1) .y< [n.y]-?< (m-n)
*y. The conclusion now followsfrom3.16.
DEFINITION 3.18. An element x of an MV-algebra A is an atom if, and
only if,xO0 and whenever0O y <x, then eithery =0 or y =x.
The followingtheoremdescribescompletelythose MV-algebras whichare
locally finiteand which contain an atom.
THEOREM 3.19. If A is linearlyorderedand containsan atom of orderm,
thenA is isomorphicto S(m).
Proof. Let y be an atom of A of orderm and we see by 3.13 that
0 < y < 2.y < ... < (m-l) y < my = 1.
Since A is linearly ordered it follows that any element x $1 is such that
n y?x < (n+ 1) .y forsome n <m-1. From 3.16(ii) we have that d(x, n y) <y
and, since y is an atom, d(x, n y) =0. This, by 3.14(ii), implies x=n y. We
have now proved that any elementmust be a multipleof y. By 3.17, d((n y)-,

This content downloaded from 147.26.11.80 on Tue, 6 Aug 2013 01:14:03 AM

All use subject to JSTOR Terms and Conditions
480 C. C. CHANG [July

(m-n) y)<y and, again since y is an atom, [n y]-=(m-n).y. Clearly

n y+l y= (n+l) .y. It is now evident that the functionf definedby f(n y)
=n/m will map A isomorphicallyonto S(m) and the theoremis proved.
There now remainsthe moredifficult situationof MV-algebras containing
no atoms.
THEOREM 3.20. If A is linearlyorderedand containsno atoms,thenfor any
x 0 and for any n thereexistsa yO 0 for whichn y? x.
Proof. We shall firstprove the following
LEMMA. If A is linearlyorderedand x<y, theneitherx+x?y or d(x, y)
+d(x, y) <y.
Assume that not x+x <y. Since A is linearly ordered, y <x+x and
1=y+x+x= [y4x]-+x. Thus d(x, y)=x y<x. By 3.15 we see that d(x, y)
+d(x, y) <x+d(x, y) =y. The lemma is proved.
Now the theoremwill be proved by inductionon n. Clearly it holds for
n = 1. Assume that the theoremholds forn and x, y are such that y $0 and
n.y<x. Since A contains no atoms, there exists a z such that O<z<y. By
lettingw=z or w=d(z, y), we see fromour lemma that w#0 and w+w<y.
It is now easy to see that (n+1) w?(2 n) w?n y<x and w is the desired
element.
THEOREM 3.21. Let A be locallyfiniteand containno atoms. Thenfor any
two elementsx, yEA for whichx<y, thereexists an elementzGA such that
x<z<y.
Proof. Suppose that x<y, then, by 3.15, x+d(x, y) =y where d(x, y)0.
If d(x, y)=1, then,again by 3.15, x y=1 and, by 1.2(vi), x=0 and y= 1. In
this case by our hypothesis,clearly there will exist an element z such that
x<z<y. Thus, let us assume that Od(x, y) and lXd(x, y). By 3.12 and
3.20 we see that there exists an element w such that Ow, w<d(x, y), and
w+w<d(x, y). Since A is locally finite,w<d(x, y). For otherwised(x, y)
=d(x, y)+d(x, y) and ord(d(x, y))= c*. Thus we have that 0<w<d(x, y).
Consider now the element z=x+w. Clearly x<z?y. Now, if z=y, then
x+w?x+w+w<x+d(x, y)=y_x+w. Hence, x+w=x+w+w. If y=1,
then d(x, y) =xt, w<x, w+w<x, and, by 1.14, w=w+w. This, of course is a
contradictionto the fact that ord(w) < c*. If y1, then, by 3.12 and 3.13,
again we arrive at the contradictionw=w+w. Thus we see that z<y. Now,
if x=z, then x=x+w. Since we already know that z<y and z$1, by 3.13,
we obtain the contradictionw=0. Thus x<z. The theoremhas been proved.
From 3.21 we see that if A is locally finiteand contains no atoms, then
A is densely ordered. Now a theoremwhich would correspondto 3.19 for
this case might go somewhat like this: If A is denumerable, locally finite,
and contains no atoms, then A is isomorphicto a subalgebra of the MV-

This content downloaded from 147.26.11.80 on Tue, 6 Aug 2013 01:14:03 AM

All use subject to JSTOR Terms and Conditions
19581 ALGEBRAIC ANALYSIS OF MANY VALUED LOGICS 481

algebra of all real numbersbetween 0 and 1. So far we have been unable to

prove or disprove this conjecture. It seems to be a difficultquestion. In this
connectionthe followingexample suggestedby J. B. Rosser is of some inter-
est. It is essentiallyan MV-algebra which is a generalizationof the algebra
C discussed in ?2 and which is densely orderedbut not locally finite.Let D
be the set of all formalsymbols of the form
n*co, 1 - *lco,
n.ci, 1 - n.cl,

n cp, 1
n-n1 ,

with the stipulation that for O0p<q, we have the relation cp-2 (q-p) Cq.
Now, given any two elementsx and y of D, we firstreduce these elementsby
the above given relations to formswhere they both involve a common cp
and thenwe imitatethe definitionof +, ., and - as forthe case of the algebra
C. It is again easily checked that D under these definitionsof +, *, and - is
an MV-algebra. Furthermore,D is obviouslydensely ordered. However, the
elementco has infiniteorder. This example,of course, shows that our conjec-
ture stated above would be false if the phrase "locally finite"is replaced by
"densely ordered."
The followingis simplyan attempt to throwthe whole situation of MV-
algebras whichare locally finiteand atomless back into the finitecases S(m).
For the subsequent discussion, we introducethe notion of a polynomial
functionof s variables v1,V2, * * *, vs. Since, by the duality, the operation is
definablein termsof + and -, we shall restrictour discussion to polynomials
built up from+ and - only. We point out here that in the followingdefinition
two polynomialfunctionsare equal if and only if they are identical.
DEFINITION 3.22. (i) P(v1, * - , v8) is a polynomialfunctionof rank 1 if
P(vi, . ,v8)=vjforsome j, 1j<s.
.

(ii) P(vi, . , v,) is a polynomial function of rank (m+1) if either

P = Q where Q is a polynomialfunctionof rank m, or else P = Q+R where Q
and R are polynomial functionsof rank ? m and at least one of the poly-
nomials Q or R is of rank m.
(iii) P is a polynomial functionif, and only if, for some m P is a poly-
nomial functionof rank m.
It is quite clear that if xi, , x, are elementsof A, then P(x1, * * *, x8)
is also an elementof A.
THEOREM 3.23. Let x1i , x8 be elementsofA, let n1, , nS be an arbi-
of
trarysequence integers, and let =
yi ni*yfor some fixedy and 1 <i < s. Suppose
furtherthat ?
d(x?, yi) y for 1 < i < s, then,for whatever polynomialfunctionP
ofs variablesand rank<(m+1), d(P(xi, * , x8),P(yY, * * *Y))_2Y.

This content downloaded from 147.26.11.80 on Tue, 6 Aug 2013 01:14:03 AM

All use subject to JSTOR Terms and Conditions
482 C. C. CHANG [July

Proof. By inductionon m. The theoremis clearly true form =0. Assume

now that the theoremholds form. Let P be a polynomial functionof rank
<m+2 and we have two cases:
CASE 1. P = Q where Q is of rank< m+ 1. By the inductive hypothesis
d(Q(xi, , x8), Q(y', * * *, ys)) < 2m*y. By 3.14(i), d(P(xi, x, ),
P (yl *, ys,))= d(Q(xi, * , x,) , Q(yi, y,,))= d(Q(xi, ,x),
Q(y,* y,y)) 2my < 2(m+l).y.
CASE 2. P=Q+R where Q and R are of ranks<m+1. By the inductive
hypothesisand 3.14(iv),
d(P(xi, ... , x,), P(yi, ... , ys)) = d(Q(xi, , x8)+ R(x*,x.. ,
Q(y1, * * * , y,) ? R(yi,. , y,)) - d(Q(xi,* * , x8),Q(yi,* , ys))
+ d(R(xi, ... , x8), R(yi, * , ys)) < 2m y + 2m y = 2(m+l).y.

The inductionis complete and the theoremis proved.

From 3.12, 3.16, 3.20, and 3.23 we see readily that:
THEOREM3.24. Let A be locallyfiniteand containno atoms. Then,forevery
polynomialfunctionP of s vyariables,
for any elementsxl, , x8 zA, and for
any elementx ofA differentfrom0, thereexistsan elementy ofA and a sequence
of integersni, , n8 such that
*

d(P(xi, ... , x8), P(ni y, * * , n8 y)) ? x.

Consider an element y of finiteorder m and the set of multiples of y,

n y, forO<n<m. We definethe operation-' for this set of elements as fol-
lows:
(n y)-' = (m - n).y.
Now, for any polynomial functionP we let P' be the resultingpolynomial
functionbuiltup of + and -' wherewe replace each - in P by -'. If ni, * * *, n8
is a sequence of integers,then clearly P'(ni y, n8 -y) will again be an * *
,

element of the formn y for some n. Also, it follows readily that (Q+R)'
= Q' +R'and (Q)' = [Q' ]'. Notice thatwe are essentiallytryingto pretendthat
the element y in some way behaves like the generator 1/m in the algebra
S(m). We have the followingapproximationtheorem:
THEOREM3.25. If A is linearlyorderedand ord(y) = m, thenfor any poly-
nomial P of s variables and rank< (n +1) and for any sequence of integers
nl,*** n8,

d(P(ni.y, , n8.y), P'(ni y, * ,,i y)) < 2n.y.

Proof. By inductionon n. We shall let yj=ni y for 1 <i<s. Clearly the

theoremholds forn =0. Assume that the theoremholds forn and let P be a
polynomialfunctionof rank< n + 2.

This content downloaded from 147.26.11.80 on Tue, 6 Aug 2013 01:14:03 AM

All use subject to JSTOR Terms and Conditions
1958] ALGEBRAIC ANALYSIS OF MANY VALUED LOGICS 483

CASE 1. P=Q where Q has rank?<n+l. In this case by 3.14(i) and (iii)
d(P(yi, ... *, ye)y P'(y * * * y,))
8

= d(Q(yi, ...* ye), [Q']'(yi, * * * , ye))

(1) _~~~~
d (Q (y,2 .. *XY8)2 [Q,]- (Y2 .
.
*.X y8))

..
+ d([Q']-(y, * * ,ys)2 [Q']-'(yll, * ,ys))

By 3.14(i) and the inductivehypothesis,

d(Q(yi, ye), [Q'f1(yi2 . . .) y8))
= d(Q(yi, ..., ys), Q'(yi, y8)) < 2n.y.
On the other hand, by 3.17,
(3) d( [Q' ]-(y1, .. * * ys)X [Q' ]-' (Y1,*** ys)) < y.
Hence, it followsfrom(1), (2), and (3) that
. * ys)) < + y _ 2(n+l)y.
d(P(yi, * ys)2 P'(yi, ,
2n_y

CASE 2. P=Q+R where Q and R have ranks?(n+l). By a straight-

forwardcalculation using 3.14(iv) and the inductive hypothesis,
d(P(yi, ..., ys), P'(yi, ... *, y))
= d(Q(yi, * * *, y,) + R(yi, * * *, ys), Q'(y', ... *, y) + R'(yi, *, y
Ye))
< d(Q(yi, * , y,), Q'(y', * * ys)) + d(R(y, * y), R'(y, * *, y)))
< 2n y + 2n. y = 2 (n+i) .

Thus the inductionis complete and the theoremis proved.

Combining3.24 and 3.25 and with the use of 3.14(iii), we obtain
3.26. Let A be locallyfiniteand containno atoms. Then,for every
THEOREM
polynomialP of s variables,for any elementsx1, , x8of A, and for any ele-
mentx of A different from0, thereexistsan elementy of A and a sequenceof
integersni, * *, n, such thatd(Pxl, * * *, x8),P'(ni y, * * *, n. y)) ? x.
At firstsight these approximation theorems seem very crude and in-
elegant. However, without a proof of the conjecture mentioned after 3.21
we see at present no other way of obtaining our results 5.3 and 5.4 in ?5.
4. Ideals, congruencerelations, and the problem of representation.
DEFINITION 4.1. A subset I of A is an ideal of A if,and only if, (i) 0-I,
(ii) if x, yzI, then x+yzI, and (iii) if xzI and y_x, then yzI.
An ideal I is said to be proper if IFA. Clearly an ideal I is proper if,
and only if, 1 EEI.
DEFINITION 4.2. R is a congruencerelationover A if,and only if (i) R is
an equivalence relation over A and (ii) if x R y and u R v, then xR ,
x+uRy+v, and x uRy v.
The definitionof a congruencerelation R is such that R automatically

This content downloaded from 147.26.11.80 on Tue, 6 Aug 2013 01:14:03 AM

All use subject to JSTOR Terms and Conditions
484 C. C. CHANG [July

preservesthe operations V and A. As usual we let x/R denote the coset

of the equivalence relation R determined by x; we let AIR denote the
set of all cosets of R. It may also be pointed out here that, just as in the case
of Boolean algebras, 4.1(iii) may be replaced by 4.1(iii'): If xEI and yEA,
then xyEI. It is clear that 4.1(iii) implies 4.1(iii'). On the other hand,
4.1 (iii) followsfrom4.1 (iii') because if y < x, then y = x Ay = x -(x +y). The
followingis a theoremthat we would expect concerningthe connectionbe-
tween ideals, homomorphisms,and congruencerelations.
THEOREM 4.3. (i) If f is a homomorphism of A ontoanotherMV-algebra,
thentheset of elementsx EA such thatf(x) = 0 is an ideal and therelationR
definedbyx R y if and onlyiff(x) =f(y) is a congruencerelation.
(ii) If R is a congruencerelation,thentheset of elementsof O/R is an ideal
and theset of all cosetsof R formsan MV-algebra (A/R, +, ., -, O/R, 1/R)
under the definitions(x/R) + (y/R) = (x +y)/R, (x/R) (y/R) = (x . y)/R, anld
[(x/R) ]- =x/R, whereA/R is an homomorphic image of A underthemapping
=
f(x) x/R.
(iii) If R is a congruencerelation,thenx R y if and onlyif d(x, y) R 0.
(iv) If R and S are congruencerelations,thenR = S if and only if O/R
=0/S.
(v) If I is an ideal, thenthe relationR definedby x R y if and only if
d(x, y) CI is a congruencerelation.
Proof. (i) and (ii) are obvious. (iii) is obtained as follows.Assume that
x Ry, then since x R x and y R y, we obtain x y R 0, x y R 0, and
(x *y+y *x) R 0. Hence d(x, y) R 0. Assume now d(x, y) R 0. Since (x y+0)
R (x.y+O),
[(xY + 0) A (x.y + yx)] R [(xY + 0) A o]
and, by Ax. 11,
[xy + (GA (y.x))]R 0.
This impliesx y R 0. Similarlyy *x R 0. Now, (y+0) R (y+x y), (x+ysx) R
(x+0), y+x y=x+y g, and xRy. (iv) follows directly from (iii). As for
(v), it is simplya consequence of 3.14.
4.3 tells us that there is a one-to-onecorrespondencebetween the ideals
of A and the congruencerelationsover A. Thus in our discussionwe can use
either notion interchangeably.In particular,given an ideal I we write x/I
=x/R and A/I=A/R where R is the unique congruencerelationassociated
with I.
The familiartheoremon principal ideals in Boolean algebra reduces to
the following:
THEOREM 4.4. Thereexistsa properideal I such thatxEI if, and only if,
ord(x)= oo.

This content downloaded from 147.26.11.80 on Tue, 6 Aug 2013 01:14:03 AM

All use subject to JSTOR Terms and Conditions
19581 ALGEBRAIC ANALYSIS OF MANY VALUED LOGICS 485

Proof. Clearly if x belongs to a proper ideal, then ord(x) = o. On the

otherhand, if ord(x) = co, then we take I to be the set of all elementsy such
that y < m x forsome m. It then is obvious that I is a properideal.
Thus we see that, in general,not every elementx $1 can be a memberof
a properideal.
DEFINITION 4.5. (i) M is a maximal ideal of A if,and only if,M is a proper
ideal and wheneverI is an ideal such that MCICA, then either M=I or
I=A.
(ii) R is a maximal congruencerelation of A if,and only if,R is not the
trivialcongruencerelationover A (i.e., R A2) and wheneverS is a congru-
ence relationsuch that RCSCA2, then eitherR=S or S=A2.
It can easily be seen that the previously mentioned one-to-one cor-
respondencebetween ideals and congruencerelations is such that maximal
ideals correspondto maximal congruencerelationsand vice-versa.
THEOREM 4.6. Everyproperideal can be extendedto a maximal ideal.
Proof. Usual proofusing the Axiom of Choice and the fact that the ele-
ment 1 never belongs to a proper ideal.
THEOREM 4.7. Let M be an ideal of A, thenthefollowingconditionsare
equivalent:
(i) M is a maximal ideal.
(ii) For everyelementx E M, nE Al for some n.
(iii) A/M is locallyfinite.
Proof.Assume (i) and let x be such that xE M. Let I be definedas the set
of all t's such that for some yEM and for some n, t<y+n x. Clearly I is
an ideal and MCI. Since xEI and xEM, we see that I=A and 1EI. This
means that forsome yE M and some n, 1 = y+n x. This last leads to [n. x ]
=<y, xr_y, and xnEM. Assume (ii) and let x/MEA/M be such that O/M
F#x/M,i.e., xEJM. Thus, forsome n, xnEM and (O/M) = (fon/M)= n
= ([x/M]-)j. This last implies that n (x/M) = 1/M. Assume (iii) and let I
be any ideal forwhich MCI and there exists an element x such that xEI
and xEEM. Thus, forsome n, n*(x/M) =1/M. Since xI, n x/I=0/I, and
since MCI, n x/I = 1/I. This gives 0/1= 1/I and I=A.
Due to 4.7 we see that the class of locally-finiteMV-algebras which we
introducedin ?3 correspondsto preciselythe class of so-called simple MV-
algebras, i.e., those MV-algebras A in which the only ideals are the sets {I0
and A. Also this fact can be seen easily and directlyfromthe definitions.It
is knownthat Boolean algebras can always be representedas algebras of sets,
i.e., where the operations +, *, and - can always be interpreted as the union
of set, the intersectionof sets, and the complementationof sets with respect
to a given unit(8). Here forMV-algebras we are at a loss to see what, if any-
(8) Cf., [6].

This content downloaded from 147.26.11.80 on Tue, 6 Aug 2013 01:14:03 AM

All use subject to JSTOR Terms and Conditions
486 C. C. CHANG [July

thing,a natural representationmeans. However, algebraically the Boolean

representationtheoremcan be equivalently stated in the following:
(*) Every Boolean algebra is isomorphic to a subalgebra of a direct
product of simple Boolean algebras (i.e., the two-elementBoolean algebra).
Taking our cue from (*) we now define representableMV-algebras as
follows:
DEFINITION4.8. An MV-algebra is representableif,and only if, it is iso-
morphicto a subalgebra of a direct product of locally finiteMV-algebras.
It easily followsfromour discussion of ideals and congruencerelations
and from [1, Theorem 9, p. 92], adapted to our special case of MV-algebras
that the class of representableMV-algebras can be characterizedas follows:
THEOREM4.9. An MV-algebra A is representable if, and onlyif, theinter-
sectionofall maximalideals ofA is theset { 0 }.
The question now remainingis whetherevery MV-algebra is representa-
ble. The answer is no. Take verysimplythe MV-algebra C discussed in ?2. It
is easily seen that C has only one maximal ideal whichis the set of all elements
of the formn c forsome n; thus it followsfrom4.9 that C is notrepresentable
in the sense of 4.8. It should be mentioned here that the characterization
given in 4.9 is the same as the statementforevery x O, thereis a maximal
ideal M such that x E M. This latter conditionis of course satisfiedin every
Boolean algebra.
If we now let K denote the class of representableMV-algebras, then by
the example C mentionedabove we see that the class K is a propersubclass
of the class of all MV-algebras. The natural question to ask here is whether
the class K is an equationalclass, i.e., whetherK can be characterizedby a set
of equations in addition to those already given by the axioms. The answer
is again no. Using the terminologyof [7], we state and prove the following
which is a strongerstatementthan the fact that K is not equational.
THEOREM4.10. The class K is nota universalclass.
Proof. It has already been remarked that the MV-algebra C is not
representable.Also, everyfinitesubset of C can be isomorphicallyembedded
in some representableMV-algebra S(m) forsome m. Thus it followsfromthe
characterizationgiven in [7] that K is not a universalclass.
5. Applicationsto the MV-algebra L and some related results. Let us
now returnto the algebra L introducedin ?2. For the sake of uniformityof
notation,we now let x, y, z, * * * denote the elementsof L. The generatorsof
L shallbe denotedby capitallettersX1, * . , X., * . . . We understand that
Zp or I W1, * * *, Wq}
any set ofcapital letterslike { Y1, . . . , Ys, {Z1, . . Z,
shall denote a subset of the set of generatorsof L.
For the discussionin the succeedingparagraphs,we let the symbol S( c0)
denote the set of rational numbers between 0 and 1, and the system

This content downloaded from 147.26.11.80 on Tue, 6 Aug 2013 01:14:03 AM

All use subject to JSTOR Terms and Conditions
1958] ALGEBRAIC ANALYSIS OF MANY VALUED LOGICS 487

(S( cc), +, *, , , 1) shall denote the MV-algebra formedfromthe set S( oo).

Given the generators Y1, * * *, Y8 of L, an assignmenth of Yi, * * *, fY into
S( oc) is simplya mapping h such that h( Y) EzS( so) for1? ii<s. An element
x =P(Yi, * * *, Y,) of L is valid in case P(h(Y1), * * *, h(Y8))=1 for every
assignmenth. In termsof these notions, the completenessof the No-valued
logic can then be stated in the following
THEOREM 5.1. Let { Y1, * * *, Y8} be anyfinitesubsetofthegeneratorsofL.
Then for any polynomialfunctionP(vi, v), P( Y1,
v*, , Y8)= 1 if and
onlyif P is valid.
Since L is generatedby the generatorsX1, * X, , we see without
that everyelementx of L can be representedas x =P( Y1, * * *, Y8)
difficulty
forsome polynomialP and some set of generators{ Yi, * * , Y8} . The next
theoremexpresses a uniqueness propertyof this representation.
THEOREM 5.2. If an elementxEL can be represented as x=P(Y,.*
Y , Y8)
and x = Q( Yi, * * * , Y8), thenfor everyassignmenth, P(h( Y1), , h( Y))
Q(h(Yi), * * *, h(Y,)).
Proof. By the hypothesis,we see that
[(P + Q)(Y1, . .. , Ys)].[(P + Q)(Y1, . .. , Y8)j = 1.
Let h be any assignment,theii by 5.1 we see that

[(P + Q)(O(Y1), . .. , h(Y8))] [(P + Q)(h(Yl), . .. , h(Y8))] 1.

By 1.2(vi), this means that

(1) (P + Q)(h(Y1), * * , h(Ys)) 1

and
(2) (P + Q)(h(Y1), . ..
* h(Ys)) 1.

(1) and (2) lead to

P(hv(Yl) .. * *, ( Ys)) _ Q(h1(Yi) .. * *,g(Ys))
and
Q(h( Yi), .. ** h(YJs))_ P(h( Yi), .. ** h( Ys)))
which give the desired conclusion.
We now give an interestingtheoremwhich connects algebra with logic.
THEOREM 5.3. In orderthatthe Mo-valuedlogic be completeit is necessary
and sufficient
thatthe MV-algebra L be representable.
Proof. (NECESSITY). By 4.9 it is sufficientto prove that every element

This content downloaded from 147.26.11.80 on Tue, 6 Aug 2013 01:14:03 AM

All use subject to JSTOR Terms and Conditions
488 C. C. CHANG [July

x5O in L is not included in some maximal ideal. Let an element x be given

with a representationx=P(Yi, * * *, Y8). Since x#O, by 5.1 (i.e., our as-
sumption) there exists an assignmenth forwhich
(1) P(h(Yi), , h(Y8)) 0.
We now extend the assignmenth to the set of all generatorsby arbitrarily
assigning the value 0 to every generator not among the set { Yi, , Y8}.
We definea set M of elementsof L as follows: M is the set of all elementsy
of L such that with respect to some representationy = Q(Z1, * * * Zp),
Q(h(Zi), * * *, h(Z,)) =0. It is evident that M is an ideal of L. Let zEM,
this means that in some representation z = R(Wi, * * *, W,) where
R(h(Wj), * * *, h(Wq))#0o. Since the algebra S(oo) is locally finite,there
exists an n forwhich
n R(h(Wi), , h(Wq)) = 1
and
[R ]n(h(W1), ***, h(Wq))= 0.
Thus the elementzn withthe representationzn = [R]n(Wl, * W.) belongs ,

to Al. By 4.7, M is a maximal ideal. By 5.2, (1), and the definitionof M,

xEE I. Hence the necessityhas been proved.
(SUFFICIENCY.) First of all, by the definitionof L, there is no difficulty
in showing that if P( Yi, . . *, Y8) = 1, then P( Y1, * * * , Y8) is valid. Let us
now assume that an element x = P( Y1, * * , Y8) of L is valid. We prove that
x = 1 by contradiction.Assume that x 5 1, then,by 4.9, thereexistsa maximal
ideal M such that xM. This means that x/M#1/M. Let us now consider
the algebra L/M which, by 4.7, is locally finite.There are two cases:
CASE 1. L/M contains an atom. In this case, by 3.19, L/M is isomorphic
to S(m) for some m. Let the isomorphismbe g. Clearly the mapping h(x)
=g(x/M) is an assignment of Y1, * * * , Y8, and, furthermore, h is a homo-
morphismof L onto S(m). Since x/M# 1 Alf,we see that
1 # h(x) = h(P(Yl, * * *, YV)) = P(h(Yl), h(Y))

This is a contradiction.
CASE 2. L/M contains no atoms. In this case, since x/M
=P(Y,IM, * * * , Y81M) F61, we see by 3.26 that, fora suitable choice of the
elementx (of 3.26), thereexists an elementyEL/M, an integerm such that
ord(y) =m, and a sequence of integers n1, * * , n8 such that

(2) P'(nl*y, , n,*y) # 1.

On consideringthe definitionof the polynomialP', we see that the formula
in (2) simply means that, in the algebra S(m),
(3) P(ni/m, * * , n./m) # 1.

This content downloaded from 147.26.11.80 on Tue, 6 Aug 2013 01:14:03 AM

All use subject to JSTOR Terms and Conditions
1958] ALGEBRAIC ANALYSIS OF MANY VALUED LOGICS 489

It now follows from (3) that the mapping h defined by h(YJ)=ni/m for
1_i_s is an assignmentforwhich P(h(Y1), , h(Y8))-1. This is also a
contradiction.
Since we have arrived at a contradictionin both cases, we see that the
sufficiencyis proved.
5.3 provides the algebraic method of which we spoke and by whichwe
hoped to findan algebraic proofof the completenessof N0-valued logic. Un-
fortunately,we see at presentno simple and directproofthat L is representa-
ble. The essential differencethen between the two-valued case and the 0o-
valued case is that not every MV-algebra is automatically representable
while every Boolean algebra is representable.We offeran algebraic proofof
the followingtheoremwhich states, in the notation of [5], that if P is valid
then F-BPP. This, of course,is a much weaker resultthan the completeness;
however,the proofwe offeris very differentin spirit fromthe proofin [5].
THEOREM5.4. If x=P(Yi, ***, Y)EL and P( Y, *, Y8)is valid,then
x+x=1.

Proof. By CONTRADICTION.Suppose that x+x$-41, then by 3.9, 4.4, and

4.6 there exists a maximal ideal M such that x x G M. This clearly means
that xJM. Now the argument followsexactly as in the proofof the suffi-
ciency of 5.3.
We conclude this section by consideringsome consequences when the
axiom
Ax. 12(m). m*x = (m + 1) x
(where m is some fixed positive integer) is added to our original axioms
Ax. 1-Ax. 11. If we now let A.5(m) be -the natural translationof Ax. 12(m)
into the notation of [5] and definethe algebra L(m) accordingly,then we
obtain the followingtheorem:
THEOREM 5.5. For everym_1 and for everyelement x=P(Y, *Y * , Y8) of
L(m), (i) x = 1, if, and onlyif, (ii) for everyn, 1< n< m, and for everyassign-
ment h of Y1, * * * , Y8 into S(n), P(h(Y1), * . . , h(Ye))=1.

Proof. Again, there is no difficultyin proving (ii) from (i). Let us now
assume (ii) and, as usual, assume that (i) does not hold, i.e., x$1. This
implies that xmn1. From our axiom Ax. 12(m), we see that m x+m x
=m x. By 1.16, we see that xm+xm=xm and ord(xm)= oo. By 4.4 and 4.6,
thereexists a maximal ideal M such that xm M. Clearly, E M. This means
that the element x/MP 1/M. Let us now consider the algebra L(m)/M. If
yEM, then by 4.7 y~nM forsome n. Due to Ax. 12(m), it is not difficultto
see that the integer n can always be taken so that n _ m. Thus, not only
L(m)/M is locally finite,but everyelementof L(m)/M must have order < m.
By 3.20, we see that L(m)/M must contain an atom which is of order rni.

This content downloaded from 147.26.11.80 on Tue, 6 Aug 2013 01:14:03 AM

All use subject to JSTOR Terms and Conditions
490 C. C. CHANG

Now, 3.19 tells us that L(m)/M is isomorphicto S(n) for some n<m. Let
the isomorphismbe g, and let h(x) =g(x/M) forxEL(m). Then h is an assign-
ment of the elements Y1, * * *, Y8 into S(n) and, furthermore,
h is a homo-
morphismof L(m) onto S(n). Since x/M#7/M, we see that
1 # h(x) = h(P(Y1, . . ., Y8)) = P(h(yl), . , h(Ys))

which is a contradiction.Thus, the theoremis proved.

In termsof many-valued logics, 5.5 asserts that a formulaP is provable
fromthe Axioms A. 1-A. 4 and A. 5(m) if, and only if, P is valid in every
n-valued logic forn < (m+1). By an easy analysis on the notions of validity
forNo-valuedand n-valued logics,we see that the followingis true which we
shall simplystate without proof.
THEOREM 5.6. The completeness of theNo-valuedlogic is also equivalentto
thefollowing:If P(vi, , v8) is a polynomialfunction,thenP( Y1, * * , Y8)
* * *

= 1 in L if and only if P( Yi, * * *, Y8)=1 in each L(m) with 1 _ m.

REFERENCES
1. G. Birkhoff, Latticetheory,
Rev. ed., 1948.
2. C. C. Chang,Proofofan axiomofLukasiewicz, Trans.Amer.Math. Soc. vol. 87 (1958)
pp. 55-56.
3. J. Lukasiewiczand A. Tarski, Untersuchungen uiberden Aussagenkalkil,Comptes
Rendusdes Seancesde la Societedes Scienceset des Lettresde Varsovie,Classe III, vol. 23
(1930) pp. 30-50.
4. C. A. Meredith,The dependence ofan axiomofLukasiewicz, Trans.Amer.Math. Soc.
vol. 87 (1958) p. 54.
5. A. Rose and J. B. Rosser,Fragments of manyvaluedstatement calculi,Trans. Amer.
Math. Soc. vol. 87 (1958) pp. 1-53.
6. M. H. Stone, The theory ofrepresentationsforBooleanalgebras,Trans. Amer.Math.
Soc. vol. 40 (1936) pp. 37-111.
7. A. Tarski,Contributions tothetheory ofmodelsI, Indag. Math. vol. 16 (1954) pp. 572-
581.
8. M. Wajsberg,Beitrdge zum Metaaussagenkalk il I, MonatsheftefurMathematikund
Physikvol. 42 (1935) pp. 221-242.
CORNELL UNIVERSITY,
ITHACA, N. Y.
UNIVERSITY OF SOUTHERN CALIFORNIA,
Los ANGELES, CALIF.

This content downloaded from 147.26.11.80 on Tue, 6 Aug 2013 01:14:03 AM

All use subject to JSTOR Terms and Conditions