Professional Documents
Culture Documents
TRENDS IN LOGIC
Studia Logica Library
VOLUME7
Managing Editor
Ryszard W6jcicki, Institute of Philosophy and Sociology,
Polish Academy of Sciences, Warsaw, Poland
Editors
Daniele Mundici, Department of Computer Sciences, University of Milan, Italy
Graham Priest, Department of Philosophy, University of Queensland,
Brisbane, Australia
Krister Segerberg, Department of Philosophy, Uppsala University,
Sweden
Alasdair Urquhart, Department of Philosophy, University of Toronto, Canada
Heinrich Wansing, Institute of Philosophy, Dresden University ofTechnology,
Germany
Assistant Editor
Jacek Malinowski, Box 61, UPT 00-953, Warszawa 37, Poland
Trends in Logic is a bookseries covering essentially the same area as the journal
Studia Logica - that is, contemporary formal logic and its applications and rela-
tions to other disciplines. These include artificial intelligence, informatics, cogni-
tive science, philosophy of science, and the philosophy of language. However, this
list is not exhaustive, moreover, the range of applications, comparisons and SOUf-
ces of inspiration is open and evolves over time.
The titles published in this series are listed at the end afthis valurne.
ROBERTO L.O. CIGNOLI
Department of Mathematics.
University of Buenos Aires. Argentina
and
DANIELE MUNDICI
Department of Computer Science.
University of Milan. Italy
ALGEBRAIC
FOUNDATIONS OF
MANY-VALUED
REASONING
Introduction 1
3 Free MV-algebras 51
3.1 McN aughton functions . . . . . . . . 51
3.2 The one-dimensional case . . . . . . . 56
3.3 Decomposing McNaughton functions 62
3.4 Ideals in free MV-algebras 64
3.5 Simple MV-algebras .. 70
3.6 Semisimple MV-algebras 72
3.7 Bibliographical remarks 75
vii
viii CONTENTS
Bibliography 203
Index 225
Introduction
The aim of this book is to give self-contained proofs of all basic results
concerning the infinite-valued proposition al calculus of Lukasiewicz and
its algebras, Chang's MV-algebras. This book is for self-study: with
the possible exception of Chapter 9 on advanced topics, the only prere-
quisite for the reader is some acquaintance with classical propositional
logic, and elementary algebra and topology.
In this book it is not our aim to give an account of Lukasiewicz's
motivations for adding new truth values: readers interested in this
topic will find appropriate references in Chapter 10. Also, we shall
not explain why Lukasiewicz infinite-valued propositionallogic is a ba-
sic ingredient of any logical treatment of imprecise notions: Hajek's
book in this series on Trends in Logic contains the most authorita-
tive explanations. However, in order to show that MV-algebras stand
to infinite-valued logic as boolean algebras stand to two-valued logic,
we shall devote Chapter 5 to Ulam's game of Twenty Questions with
lies/errors, as a natural context where infinite-valued propositions, con-
nectives and inferences are used. While several other semantics for
infinite-valued logic are known in the literature-notably Giles' game-
theoretic semantics based on subjective probabilities-still the transi-
tion from two-valued to many-valued propositonallogic can hardly be
modelled by anything simpler than the transformation of the familiar
game of Twenty Questions into Ulam game with lies/errors.
This book is mainly addressed to computer scientists and mathe-
maticians wishing to get acquainted with a compact body of beautiful
results and methodologies-that have found applications in the treat-
ment of uncertain information, (e.g., adaptive error-correcting codes)
as weIl as in various mathematical areas, such as toric varieties, lattice-
1
R. L. O. Cignoli et al., Algebraic Foundations of Many-Valued Reasoning
© Springer Science+Business Media Dordrecht 2000
2 INTRODUCTION
ordered groups and C"-algebras. As the title indicates, the main em-
phasis is on algebraic methods. Thus, reversing the historical order,
we shall make the reader familiar with MV-algebras before introducing
Lukasiewicz's propositional calculus in Chapter 4. This will allow us to
get neat and elementary proofs of several deep results, using much less
symbolism and detail than in traditional syntax-oriented approaches.
The definition-theorem-proof style adopted throughout this book will
hopefully result in time saving for the reader who wishes to get the
proofs of all main theorems on the infinite-valued calculus as quickly as
possible, without embarking on a potentially unbounded search through
a scattered literature on ordered groups, lattices, algebraic logic, poly-
hedra, geometry of numbers, model theory, linear inequalities, et cetera.
By definition, an MV-algebra A is a set equipped with an associat-
ive-commutative operation EB, with a neutral element 0, and with an
operation -, such that -,-,x = x, x EB -,0 = -,0, and, characteristically,
These six equations are intended to capture some properties of the real
unit interval [0, 1] equipped with negation -,x = 1 - x and truncated
addition x EB y = min(l, x + y). For instance, once interpreted in [0,1],
the left hand term in the last equation coincides with the maximum
of x and y; thus the equation states that the max operation over [0,1]
is commutative. The fundamental theorem on MV-algebras is Chang's
completeness theorem, stating that every valid equation in [0, 1] is auto-
matically valid in all MV-algebras. A new proof of this theorem is given
in Chapter 2. As a preliminary step, in Chapter 1 we prove Chang's
subdirect representation theorem, stating that an equation is valid in
every MY-algebra iff it is valid in every totally ordered MV-algebra.
As in the classical case, one may ask for an effective procedure
to decide when an equation is valid. Rather than working in "MY-
algebraic equationallogic", it is more convenient to give the Lukasiewicz
infinite-valued calculus the same role that the classical propositional
calculus has for the boolean decision problem. Accordingly, one may
write
-,x ~ y instead of x EB y
INTRODUCTION 3
***
While the first germ of this book is our monograph [58], the proofs
of many fundamental theorems are given here in a more general and
self-contained form, using results that have appeared in the literature
after the publication of [58]. Several sections on advanced topics have
been added, and the bibliography has been considerably expanded. The
book also contains a wealth of previously unpublished material.
This book is didactic in its spirit: preliminary versions have been
tested in several graduate courses in Bahfa Blanca, Barcelona, Buenos
Aires, Campinas, Merida, Milan, Patras, and in the general context of
the European Project known as Action COST number 15 on "Many-
valued Logic for Computer Science Applications". We are grateful to
Stefano Aguzzoli, Agata Ciabattoni, Vincenzo Marra and Claudia Pi-
cardi for their valuable comments on earlier drafts of this book. For
valuable hints and discussions we are indebted to many more students
and colleagues: we ask them to forgive us for not listing them here. We
also thank the anonymous referee for his careful and competent reading
and for suggesting several improvements.
We gratefully acknowledge partial support from the National Re-
search Councils of Argentina (CONICET), Brazil (CNPq) and Italy
6 INTRODUCTION
Basic nations
1.1 MV-algebras
Definition 1.1.1 An MV-algebra is an algebra (A, tf), -', 0) with a bi-
nary operation tf), a unary operation -, and a constant 0 satisfying the
following equations:
7
R. L. O. Cignoli et al., Algebraic Foundations of Many-Valued Reasoning
© Springer Science+Business Media Dordrecht 2000
8 CHAPTER 1. BASIC NaTIONS
MV9) x Ee -,x = 1.
(i) -,xEey=l;
(ii) x0-,y=0;
(iii) y = x Ee (y 9 x);
(iv ) there is an element z E A such that x Ee z = y.
Proof: (i) =? (ii) By MV4) and MV7). (ii) =? (iii) Immediate from
MV3) and MV6'). (iii) =? (iv) Take z = y 9 x. (iv) =? (i) By MV9),
-,x EI1 x EI1 z = 1. 0
Let A be an MV-algebra. For any two elements x and y of A let
us agree to write
x~y
iff x and y satisfy the above equivalent conditions (i)-(iv). It follows that
~ is a partial order, called the natural order of A. Indeed, reflexivity
10 CHAPTER 1. BASIC NOTIONS
(1.4) {aEBx = 1
a0x = 0
Proof' By Lemma 1.1.2, these two equations amount to writing -,a ::;
x ::; -,a. 0
Lemma 1.1.4 In every MV-algebra A the natural order::; has the fol-
lowing properties:
Proof: (i) This follows from Lemma 1.1.2(i), since -,xEBy = -'-'yEB-,x.
(ii) The monotonicity of EB is an easy consequence of Lemma 1.1.2(iv);
using (i), one immediately proves the monotonicity of 0. (iii) It is
sufficient to note that x 0 y ::; z is equivalent to 1 = -,(x 0 y) EB z =
-,x EB -'y EB z. 0
(1.5) x V y = (x 0 -,y) EB y = (x e y) EB y,
(1.6) x 1\ y = -,( -,x V -,y) =x0 (-,x EB y).
1.1. MV-ALGEBRAS 11
-,((x e y) E9 y) E9 z = (-,(x e y) e y) E9 y E9 (z e y)
= (ye -,(x e y)) E9 -,(x e y) E9 (z e y)
= (y e -,(x e y)) E9 -,x E9 y E9 (z e y)
= (y e -,(x e y)) E9 -,x E9 z = 1.
It follows that (xey) E9y:::; z, which completes the proof of (1.5). We
now immediately obtain (1.6) as a consequence of (1.5) together with
Lemma 1.1.4(i). 0
(i) x0(yVz)=(x0y)V(x0z),
(ii) x E9 (y!\ z) = (x E9 y) !\ (x E9 z).
(1.7) (xey)!\(yex)=o.
Proof: By making repeated use of MV6) and its variants, together with
the basic properties of the operations E9 and 0 we obtain:
(xey)!\(yex)
12 CHAPTER 1. BASIC NOTIONS
= (x e y) 8 (-,(x e y) E9 (y e x»
= x 8 -,y 8 (y E9 -,x E9 (y e x»
= x 8 (-,x E9 (y e x» 0 (-,(-,x E9 (y e x» E9 -,y)
Hl) h(O) = 0,
H2) h(x E9 y) = h(x) E9 h(y),
o
In particular, for each element z of an MV-algebra A, the ideal
(z) = ({ z }) is called the principal ideal generated by z, and we have
(1.8) (z) = {x E AI nz ? x for some integer n ? O}.
Note that (0) = {O} and (1) = A. Further, for every ideal J of an
MV-algebra A and each z E A we have
(1. 9) (J U {z}) = {x E A Ix ~ nz ffi a, for some n E N and a E J}.
An ideal 1 of an MV-algebra Ais proper Hf 1 =1= A. We say that 1
is prime Hf it is proper and satisfies the following condition:
14 CHAPTER 1. BASIC NaTIONS
(ii) for each x E A, x rt. J iff -mx E J for some integer n 2:: 1.
We denote by I(A), P(A) and M(A) the sets of ideals, prime ideals
and maximal ideals of A, respectively.
In the next lemma we summarize, for furt her reference, some easy
relations between ideals and kernels of homomorphisms.
Prao!" The set 1i. =def {I E I(A) I I =I- A and J ~ I} is totally ordered
by inclusion. Therefore, M =def UIE'H I is an ideal of A. Furt her , M is
a proper ideal, because 1 ~ M; we conclude that M is the only maximal
ideal containing J. 0
and
1.4 MV-equations
As we shall see, an important consequence of Chang's Subdirect Rep-
resentation Theorem is that in order to prove that an equation holds in
1.4. MV-EQUATIONS 21
°
(Tl) The elements and Xi, for i = 1, ... , t, considered as one-element
strings, are MV-terms.
Those strings Ti that belong to every formation sequence for T are said
to be the sub terms of T.
The following result is known as the unique readability theorem; its
proof is precisely the same as for the classical propositional calculus,
and is left as a routine exercise.
Theorem 1.4.2 Every term Ti in the variables Xl, ... ,Xn satisfies pre-
cisely one of the above conditions (i)-(iii). Moreover, both term Tj of
case (ii) and the pair (Tj, Tk) 01 case (iii) are uniquely determined. 0
22 CHAPTER 1. BASIC NaTIONS
A Fr = a,
x 1\ (y V z) = (x 1\ y) V (x 1\ z)
and
x V (y 1\ z) = (x V y) 1\ (x V z).
1.5. BOOLEAN ALGEBRAS 25
B(L)
(i) xE B(A);
(ii) x V -,x = 1;
(iii) x /\ -,x = 0;
(iv) x E9 x = x;
(v) x0x=x;
1.6 MV-chains
In this section we collect several results on totally ordered MV-algebras,
to be used in the next chapter.
(1.19) (x 0 y) $ z = (x 0 z) $ y.
Note that equation (1.19) holds in case x0y = 0 or x0z = O. Indeed,
suppose x 0 y = O. Since x $ y = 1, it follows from Lemma 1.1.3 that
x = 'y, whence from y = -.x :S z we obtain (x 0 y) $ z = z = y V z =
1.7. BIBLIOGRAPHICAL REMARKS 29
(Equivalents of) MV-algebras are known in the literat ure under sev-
er al names. As an example, following [249],[121], let us say that a
bounded commutative ECK-algebra is an algebra (A, *, 0,1) with a bi-
nary (bounded subtraction) operation * and two constants 0 and 1
satisfying the following equations:
Y1) (x * y) * z = (x * z) * y
Y2) x*(x*y)=y*(y*x)
Y3) x*x = 0
Y4) x*o= x
Y5) x *1 = O.
Bounded commutative BCK-algebras have been considered by several
authors. See, e.g., [233], [213], [214], [116], [120].
A tedious but straightforward verification yields the following result,
first proved in [86] (also see [164]):
Theorem 1.7.1 If (A, EB,', 0) is an MV-algebra then (A, e, 0,1) is a
bounded commutative ECK-algebra. Moreover, we have the identities
.x = 1 ex and x EB y = 1 e ((1 e x) e y)).
Conversely, for any bounded commutative ECK-algebra (A, *,0,1),
upon dejining .x =dej 1 * x and x EB y =dej 1 * ((1 * x) * y), then
(A, EB,', 0) is an MV-algebra, and x e y = x * y.
There exist several other equivalent counterparts of MV-algebras, in-
cluding Bosbach's bricks [31], Buff's S-algebras [33], Komori's CN-
algebras [130], Lacava's L-algebras [131], Rodriguez's Wajsberg algebras
[212], [86]. We will return to the latter in subsequent chapters. As
we shall also see, up to categorical equivalence, MV-algebras are the
same as abelian lattice-ordered groups with a distinguished strong unit.
Thus MV-algebras provide an equational formulation of the theory of
magnitudes with an archimedean unit.
Among all associative commutative continuous operations on the
unit interval [0,1], the EB operation and its dual 0 have a special status.
For details we refer, e.g., to the papers by Menu and Pavelka [153], [201],
and to the relevant chapters ofthe book by Butnariu and Klement [34].
Chapter 2
31
R. L. O. Cignoli et al., Algebraic Foundations of Many-Valued Reasoning
© Springer Science+Business Media Dordrecht 2000
32 CHAPTER 2. CHANG COMPLETENESS THEOREM
and
(2.3) t+(xAy)=(t+x)A(t+y).
In every o-group xVy = max{x,y} and xAy = min{x,y}; o-groups
are particular cases of f-groups.
For each element x of an f-group G, the positive part x+, the
negative part x-, and the absolute value lxi of x are defined as folIows:
(2.4) x+ =d,eJ 0 V x;
= y E9 -,(y E9 -,x)
= u A (y + (u - (u A (y + u - x))))
2.1. THE FUNCTOR r 33
= y V x = x Vy.
This shows that x and y are interchangeable. 0
(2.7) X ED··· ED X = u A (x
n.x =def "-----v---" + ... + x), = u A nx.
n times n ti~es
34 CHAPTER 2. CHANG COMPLETENESS THEOREM
Proposition 2.1.5 Let Adenote the category whose objects are pairs
(G, u) with G an f-group and u a distinguished strang unit of G, and
whose morphisms are uni tal f-homomorphisms. Then r is a functor
from A into the category MV of MV-algebras. 0
(b + a) +c
= (lP+q+r, aEBbEBc, (a0b)EB«aEBb)0c), a0b0c)
= (lP+q+r, a EB bEB c, (a 0 c) EB «a EB c) 0 b), a 0 b 0 c)
=b+(a+c).
Similarly, to prove cancellation, avoiding trivialities, assume that a,
band c are different from 1. If q = r, then by Lemma 1.6.1(ii), b = c,
38 CHAPTER 2. CHANG COMPLETENESS THEOREM
and we are done. If q < r-1 then from the identity (1 p +q , aEBb, a0b) =
(1 P+r ,a EB c,a 0 c) we get a 0 b = 1, Le., a = b = 1, which is a
contradiction. If q = r - 1 then a 0 b = a and a EB b = 1, which is
impossible because these two equalities imply that b = 1. The cases
corresponding to r < q are similarly shown to lead to contradiction. 0
Proposition 2.3.2 Let a = (al, ... , an) and b = (bI, ... , bm) be good
sequences. Recalling (2.8) assume, without loss of generality, m = n.
Then the following are equivalent:
Proposition 2.3.5 Let a = (al, . .. , an, ... ) and b = (bI, ... ,bn , ... )
be good sequences of an MV-algebra A.
(i) The sequence
[a, b].
There will be no danger of confusion with the notation for unit intervals
in f-groups.
Let GA = (GA, 0, +, -) be the set of equivalence classes of pairs of
good sequences, where
the zero element 0 is the equivalence class [(0), (0)],
addition is defined by [a, b] + [c, d] =def [a + c, b + d], and
subtraction is defined by -ra, b] =def [b, a].
Then by direct inspection one easily sees that GA is an abelian
group. GA is called the enveloping group of A.
We shall now equip GA with a lattice-order. Let (a, b) be a pair of
good sequences of the MV-algebra A. By Proposition 2.3.2(i), (a, b)
has an equivalent pair of the form (e, (0)) if and only if a 2: b. Let
M~ be the submonoid of GA given by the equivalence classes of pairs
(e, (0)), for all good sequences e. Since the map e 1---+ (e, (0)) induces
an isomorphism of the monoid M A onto M A, we shall freely identify
the two monoids M A and M A.
= 2q + a + c + x + Y
= 2q + a + c + z + (x V y)
= z + q + ((x + q + a + c) V (y + q + a + c))
= z + q + ((p + b + c) V (p + d + a))
= p + z + q + ((b + c) V (d + a)).
One similarly proves (ii). Finally, (iii) is an immediate consequence of
the definitions of the partial orders $ and ::S. 0
Definition 2.4.3 The f-group GA with the above lattice-order is
called the Chang f-group of the MV-algebra A.
Proposition 2.4.4 The element UA =del [(1), (0)] is a strong unit 0/
the f-group GA.
Proof: As a matter of fact, any element of GA + can be represented
by [a, (0)], for some good sequence a = (al, a2,' .. ) in A. Let the
integer m ~ 1 be so chosen that an = 0 for all n ~ m. By Definition
2.4.1, mUA = [1 m , (0)] dominates [a, (0)], whence the desired conclusion
immediately follows. 0
A crucial property of the f-group GA is given by the following result:
Theorem 2.4.5 The correspondence
a I-t 'PA(a) = [(a), (0)]
defines an isomorphism from the MV-algebra A onto the MV-algebra
r(GA,UA)'
Proof: By definition, [(0), (0)] ::S [a, b] ::S UA iff there is c E A such
that (a, b) is equivalent to ((c), (0)). Thus, 'PA maps A onto the unit
interval [[(0), (0)], UA] of GA' It is easy to see that this map is one-one.
By (2.16), 'PA(aEBb) = ('PA(a)+'PA(b)) /\UA, and by (2.15), 'PA(--,a) =
UA -'PA(a). Therefore, 'PA is a homomorphism from A to r(GA,UA).
o
Remark: An MV-algebra A is a chain if and only if GA is totally or-
dered. Indeed, if A is totally ordered, then it follows from Proposi-
tion 2.3.2(i) that M A is totally ordered, and this implies that GA is a
totally ordered group. The converse is an immediate consequence of
the Theorem 2.4.5 above.
2.5. CHANG COMPLETENESS THEOREM 43
Ö=def 0,
=:;a =def (y - a),
(p EB(1) =def (y A (,0 + a)).
Since unique readability also holds for f-group terms, the mapping
T I--t f is weIl defined; indeed, a moment's reflection shows that this
map is computable by a Turing machine. We then have a purely syn-
tactic counterpart of the mappings (G,u) I--t r(G,u) and A I--t GA,
in a sense that is made precise by the following two propositions:
°: ;
Proposition 2.5.1 If G is a totally ordered abelian group,
G,
°
< u E
gl, ... gn ::; u and A = r(G,u), then for every MV-term
T(XI, .. . , x n ) we have TA(gl'" ., gn) = fG(gl'" ., gn, u).
= -'((1A(gl"" ,gn))
44 CHAPTER 2. CHANG COMPLETENESS THEOREM
h ~p k iff k - h E P.
Let us display the subterms 0"0,0"1, ... ,O"t of f as follows:
We can safely assurne that the list contains the zero term. The map
y ~ h o, Xl ~ h l ,· .. , X n ~ h n uniquely extends to an interpretation
O"j ~ h j (j = 0, ... ,t) of subterms of f into elements of the totally
ordered group T = (zr, ~p). In particular, by hypothesis we have
(2.21) o ~p h l , ... , h n ~p h o, 0 ~p h t ~p h o,
(2.22) o=f. h t = fT(h l , ... , h n , h o).
Let w be apermutation of {O, ... , t} such that
For any vectors ko, ... , kn E zr, the map y t-+ ko, Xl t-+ k I , ... , X n t-+
kn uniquely extends to an interpretation O'j t-+ k j , j = 0, ... ,t of
all subterms O'j of f into elements k j of T'. In the particular case
when ko = h o, ... , kn = h n , arguing by induction on the number of
operation symbols occurring in O'j, from (2.28) we obtain k j = h j for
all j = 0, ... , t; moreover, all inequalities in (2.21) are still valid with
respect to the new total order relation '5:.p' over zr. In symbols, 0 '5:.p'
h I , ... ,hn '5:.p' h o, and 0 '5:.p' h t '5:.p' h o, 0 =1= h t = fT' (h I , ... ,hn , h o)·
As an effect of the independence of the I"S over Q, T' is isomorphie,
as a totally ordered group, to the subgroup U = ZI'I + ... + Zl'r of R
generated by 1'1,"" I'r, with the natural order. An isomorphism is
given by the map
Free MV-algebras
for each ordinal a < /'i,. Then by definition, each MV-term r in the
51
R. L. O. Cignoli et al., Algebraic Foundations of Many-Valued Reasoning
© Springer Science+Business Media Dordrecht 2000
52 CHAPTER 3. FREE MV-ALGEBRAS
where X o E A for each a < K. The ath projection 7r0 : A 11: ~ A is the
function 7ro (x) = X O ' We let
Definition 3.1.1 For each term T in the variables {Xo}o<1I: the term
function
TA: All: ~ A is given by induction on the number of connectives in
T as follows:
Term~
denote the set of all term functions over All:. By construction, each
element of Term~ is a function only depending on a finite number of
variables. Further, Term~ is a subalgebra of the MV-algebra AAl< of all
A-valued functions over All:, with pointwise defined operations. More
precisely, we have the following result, whose proof immediately follows
by definition:
3.1. MCNAUGHTON FUNCTIONS 53
(bi, mitE Z), such that for each point Y = (Yo, ... , Yn-l) E [0, l]n there
is an index j E {I, ... , k} with 1(Y) = Pj(Y)·
is a McNaughton 1unction over [0, ll~ Hf there are ordinals a(O) < ... <
a(m - 1) < >. and a McNaughton function 1 over [0, 1jm such that for
each x E [0,1],\ g(x) = f(xo;(o) , ... , Xo;(m-I))'
all i = 1, ... , m and j = 1, ... , n,) together with the constant function
1. We conclude that the McN aughton functions form a subalgebra of
[O,l][O,lJ". By Lemma 3.1.2, all term functions are McNaughton func-
tions. Now apply Proposition 3.1.4. 0
McNaughton's theorem, to be proved in a subsequent chapter, states
the converse of Proposition 3.1.8. A short proof of the one-variable case
shall be given in the next section. For most applications, however, one
does not need the full strength of McNaughton's theorem, but only
Lemma 3.1.9 below-a much simpler result.
For each real-valued function " we let
Lemma 3.1.9 Let 9 : [0, l]n --+ R be a linear function with integer co-
efficients, say, g(x) = moxo+ ... +mn-lxn-l +b, with mo, ... ,mn-I, bE
Z. Then gtt E Free n.
The identity trivially holds whenever x is such that h(x) > 1, or h(x) <
-1. If x is such that h(x) E [0,1], then htt(x) = h(x), and (h(x) + l)U =
1. Since Xo E [0,1], (h(x) + xo)tt = h(x) EB Xo, which establishes (3.3).
°
Finally, if h(x) E [-1,0}, then M(x) = and (h(x) + l)U = h(x) + 1,
whence (3.3) follows from the identities (h(x)+xo)U = max(O, h(x)+xo)
= max(O, Xo + h(x) + 1 - 1) = x00 (h(x) + 1). Thus, in case mo > 0,
56 CHAPTER 3. FREE MV-ALGEBRAS
identity (3.3) holds for each x = (xo, ... , Xn-l). By induction, together
with Proposition 3.1.4, we have (h+xo)~ = g~ E Free n . In case mo < 0,
applying the same argument to the function 1 - g, one shows that
(1 - g)~ E Freen- Since 1 - (1 - g)~ =g~, we conclude that gtt E Free n
in all possible cases. 0
(ii) Every irreducible fraction p/q E [O,IJ occurs in FareYm, for some
index m.
Proof: (i) Let (1,1) and (0,1) be the vectors in Z2 respectively giv-
ing the homogeneous correspondents of 1 and 0. Then the fact that
the closed interval [O/I,I/IJ satisfies unimodularity has the following
equivalent reformulations:
(a) the determinant of the matrix whose rows are given by (1,1) and
(0,1) is equal to one;
(b) every vector in Z2 is a linear combination, with integer coefficients,
ofthe vectors (1, 1) and (0, 1)-for short, the pair ((1, 1), (0, 1)) is a basis
in Z2.
One then immediately sees that if an interval [u/v,u'/v'J ~ [O,IJ
satisfies the uni modular law, then so do the two intervals ru/v, u" /v"J
and [u"/v",u'/v'J where u"/v" =def (u + u')/(v + v'). For short,
unimodularity is preserved under the operation of taking mediants.
The rest is clear.
°
To prove (ii), by way of contradiction assurne < p/q < 1 to be
an irreducible fraction not occurring in any FareYn- The coordinates
of vector (p, q) E Z2 in the basis B o =def ((1,1), (0, 1)) are given by
p and q - p. Note that their sum is strictly less than the sum p + q
of the coordinates of the same vector in the initial basis ((1,0), (0, 1)).
°
Taking the mediant 1/2 of 0/1 and 1/1 (the latter two fractions giving
the inhomogeneous correspondents of and 1), and passing to homo-
geneous coordinates in Z2, we obtain two new bases ((1,1), (1,2)) and
((1,2), (0, 1)). Precisely one of them, denoted BI, encapsulates the vec-
tor (p, q), in the sense that the coordinates al and ßl of (p, q) with
respect to basis BI are integers 2: 1. By direct inspection, the sum of
these coordinates is strictly less than q. Proceeding inductively, assurne
we are given an encapsulating basis B j = (Vj, Wj), for suitable integer
vectors Vj and Wj in Z2, and let aj and ßj be the coordinates of (p, q)
with respect to B j . Upon taking the mediant of the inhomogeneous
correspondents of Vj and Wj, we obtain two bases (Vj, Vj + Wj) and
(Vj + Wj' Wj); precisely one of them, denoted Bj+I, encapsulates (p, q);
further, the coordinates aj+l and ßj+l of (p, q) with respect to B Hl
satisfy the inequality 2 :::; aj+l + ßj+l < aj + ßj. Thus, by our absur-
dum hypothesis, the sum of the coordinates of (p, q) decreases infinitely
58 CHAPTER 3. FREE MV-ALGEBRAS
lid
w E
°
Prao!" Let = cd d 1 < C2/ d2 < ... < cu/du = 1 be the ascending
sequence of elements of FareYn. Then di = J.-li. Direct inspection, in
the light of the unimodularity property in Proposition 3.2.1 (i), shows
that the function hi 1\ hi+1 attains its maximum value 1/(di + di + 1 )
at the mediant point p = (Ci + ci+l)/(di + di+d. Since the nodes of
h i 1\ hi+1 are given by the five rational numbers 0, ci/di , p, Ci+l/di+1, 1,
and the function coincides with k 2i at each node, then k2i coincides with
hi 1\ hi+l over an of [0, 1]. Similarly, the function hi - (h i 1\ (h i - 1+ hi+l))
is constantly equal to zero over both intervals [0, (Ci-l + ci)/(di- 1 +
60 CHAPTER 3. FREE MV-ALGEBRAS
Proof: Let q be the smallest integer 2:: max( m, n) such that all the nodes
ofthe function fRffifs = min(1, fR+ fs) occur in FareYq. The existence
3.2. THE ONE-DIMENSIONAL GASE 61
from Propositions 3.2.5 and 3.2.6 it follows that the set of functions of
the form Ir, for T a subsystem T of some Schaudern is closed under
applications of -, and EB. This yields the desired conclusion. 0
Corollary 3.2.8 The free MV-algebra mel over one generator is iso-
morphie to the MV-algebra 0/ McNaughton junctions %ne variable.
display the elements of FareYq. Note that the value of / at each node
mj = aj/bj is a multiple of l/bj , for some integer 0 ~ t ~ bj . It
follows that a suitable linear combination g (with integer coefficients)
of the hats in Schauderq will coincide with / at all nodes of mj. Since
both g and / are linear over each interval [mi, mi+l], then g and / will
coincide over the interval [0,1]. There remains to be shown that every
Schauder hat is an element of Freel' This is an immediate consequence
of Theorem 3.2.7 and Proposition 3.1.4. 0
holds for each point x E H+, while q(x) ~ p(x) holds for each point
x E H-. More generally, for every permutation p of the set {I, ... , k},
let Pp ~ [0, l]n be defined by
(3.4) Pp =def {x E [0, l]n I Pp(l) (x) ~ Pp(2) (x) ~ ... ~ Pp(k) (x)}.
As an intersection of the cube [0, l]n with a finite set of closed half-
spaces, each Pp is, by definition, a (possibly empty) convex compact
polyhedron.
By definition, the vertices of Pp are those points of Pp that cannot
be expressed as nontrivial convex combinations of points of Pp. By the
fundamental theorem on polyhedra, Pp can be visualized as the convex
hull of the finite set of its vertices. Since all linear constituents of f
have integer coefficients, each vertex v of Pp is rational, in the sense
that the coordinates of v are rational numbers.
Let us agree to denote by C the set of n-dimensional polyhedra of
the form Pp, for some permutation p, in the above decomposition of the
domain of f. Then C is a finite set of compact convex n-dimensional
polyhedra with rational vertices, having the following additional prop-
erties:
(i) the union of all polyhedra in C coincides with the cube [0, l]n,
(ii) any two polyhedra in C are either disjoint or they intersect in a
common face, and
(iii) for each polyhedron P E C there is an index u = Up E {I, ... , k}
such that upon restriction to P, the two functions fand Pu. coincide.
For furt her applications, it will be convenient to replace C with a
family of n-dimensional simplexes satisfying (i)-(iii). To this purpose,
one can simply triangulate every polyhedron in C, generalizing the fa-
miliar triangulation of convex polygons in the two-dimensional case. In
this way one does not even need to add new vertices.
As an alternative construction, for each d = 0, ... ,n let us denote
by :;:(d) the set of d-dimensional faces of polyhedra in C. Thus, for
instance, :;:(0) is the set of singletons given by the vertices of polyhedra
in C, :;:(1) is the set of edges of polyhedra in C, ... , :;:(n) = C.
For every j = 2,3, ... , n and every polyhedron P E :;:(j) let us
select, once and for all , a rational point b p in the relative interior of
P. In other words, b pEP n Qn and b p does not belong to any
64 CHAPTER 3. FREE MV-ALGEBRAS
(i) The union of all simplexes in S coincides with the cube [0, l]n ,.
(iii) For each simplex WES there is an index u E {I, ... , k} such
that, upon restriction to W, the two functions fand Pu coincide.
o
°
f(x) > min(b1 , ... , bk ) > for each x E S. Hence for some integer
n ~ 1, we must have -.nf E J s and -.nf(z) = 1, thus showing that
z rt VJs · This yields the desired conclusion VJs ~ S.
(iii) For each S ~ X, Js = nXES Jx . On the other hand, J ~ Jx if
and only if xE VJ . The desired result now follows from (i). 0
Example: With reference to formula (3.1), for each k = 1,2, ... , let the
McNaughton function !k : [0,1] - [0,1] be defined by
fk(X) = (k - (k + l)x)~.
Let J be the ideal of Freel generated by the functions JI, 12, .... By
direct inspection we easily obtain VJ = {1}, and (1 - x)~ E J l \ J.
Therefore, J =f JVr
70 CHAPTER 3. FREE MV-ALGEBRAS
(i) A is simple;
(ii) A is nontrivial and for every norizero element x E A there is an
integer n > 0 sueh that 1 = x EB " . EB x (n times);
Proof: In the light of Proposition 1.2.2, (ii) states that {O} is a maximal
ideal of A. Therefore, (i) and (ii) are equivalent. It is obvious that
(ii) is satisfied by all subalgebras of the MV-algebra [0,1]. Finally, to
prove (i) -+ (iii), assume A to be simple. If the cardinality of A is K"
then by Proposition 3.1.5, we can identify A with the quotient MV-
algebra Free"./ J for some ideal J of Free lt • By Proposition 1.2.10,
since A is simple, J must be a maximal ideal of Free lt • Therefore, by
Proposition 3.4.7, J = Jx , for a uniquely determined point x E [0,1]1t.
Applying now Proposition 3.4.5, we obtain that A is isomorphie to
the MV-algebra Freeltl{x} = 7fx (Free lt ), where 7rx: Free lt -+ [0,1] is
the projection given by 7rx (f) = f(x). Hence A is isomorphie to a
subalgebra of [0,1].0
1 2 n- 2
(3.8) Ln =def {O, - - , - - , ... , - - , I}.
n-1 n-1 n-1
3.5. SIMPLE MV-ALGEBRAS 71
° °
Proof: In case a = 0, let < z ::; 1 and < € < z/2. By assumption
°
there is b E A + such that < b < €. Letting n be the smallest integer
such that nb;::: z, noting that n > 2, it follows that Z-€ < (n-l)b < z.
Thus A is dense. In case a > 0, if a = 1 then A = L 2 = {O, I}. If a < 1,
since A is closed under the operation 1 - x, we have a::; 1/2 and
a E A+. For otherwise, there exist two elements x, y E A+ such that
a < x < y < 2a, whence a > y - x = y 0 -,x E A+, a contradiction.
Having proved that a E A +, let m be the unique integer such that
(m - l)a < 1 ::; ma. Note that m ;::: 2. Let
(i) A is semisimple;
(ii) A is isomorphie to a separating MV-algebra 01[0, 1]-valued eontin-
uous funetions on some nonempty eompaet Hausdorff spaee, with
pointwise operations. 0
Lukasiewicz oo-valued
calculus
Free w ? These are respectively known as the word problem for free MV-
algebras, and the word problem for finitely presented MV-algebras.
In this chapter we shall reformulate these problems in purely logi-
cal terms, within the infinite-valued sentential calculus of Lukasiewicz.
We shall regard MV-terms as propositions, valid equations as tautolo-
gies, ideals as theories, word problems as decision problems in this
calculus-in the traditional sense. Adopting this viewpoint, we shall
need an equivalent reformulation of Chang's completeness theorem to
the effect that all tautologies are obtainable from a certain set of initial
tautologies (corresponding to the MV-axioms) by a finite number of
applications of modus ponens. Free MV-algebras shall be re-obtained
as algebras of propositions up to logical equivalence. Using the results
of Chapter 3 we shall finally obtain an effective procedure to decide
whether a proposition is a tautology, thus automatically giving a posi-
tive solution to the word problem for free MV-algebras. Moreover, the
77
R. L. O. Cignoli et al., Algebraic Foundations of Many-Valued Reasoning
© Springer Science+Business Media Dordrecht 2000
78 CHAPTER 4. LUKASIEWICZ oo-VALUED CALCULUS
(4.3) x - Y = -,x EB Y
whence,
Thus, the operations ..." Ei' and -+ on [0,1] have the same interdefin-
ability properties as negation, disjunction and implication in classical
logic. It is natural to consider Ei' as a disjunction connective. Accord-
ingly, equation (1.2) suggests to consider <:) as conjunction.
Given an arbitrary MV-algebra A, let us define the binary operation
-+ on A by formula (4.3). That is, for all x,y E A, x -+ Y =def ...,xEi'y.
Then, triviallY, equation (4.4) holds in A.
In the next section· we shall consider an equivalent reformulation of
MV-algebras in terms of the operations -+, ..., and the constant 1. This
will more closely correspond to Lukasiewicz's original presentation of
the infinite-valued calculus, which we shall now introduce formally as
follows:
Definition 4.1.1 As in the classical case, one starts from the finite
alphabet
Notation: For each formula a, Var(a) will denote the set of all propo-
sitional variables occurring in a.
Theorem 4.1.4 A formula a is a [0, 1]-tautology ij, and only ij, for
every MV-algebra A, a is an A-tautology. Thus, for any two formulas
a and ß, we have a[O,I) = ß[O,I) iff a A = ßA for all MV-algebras A.
o
vi(Xm ) = 0 for all remaining variables. Whenever a pair (i, 'ljJt) is such
that vi('ljJt) < 1, put 'ljJt in the list of nontautologies. By Proposition
4.1.3 together with Theorem 4.1.4 and our analysis of free MV-algebras
in Section 3, the desired conclusion now follows from the continuity of
McNaughton functions: indeed, if a formula 'ljJ is not a tautology then
there is a valuation v whose values are all rational, such that v( 'ljJ) < 1.
o
In the next sections we shall obtain an effective procedure to enu-
merate all tautologies.
W1) 1-x=x
W3) (x - y) - y = (y - x) - x
W5) x - x = 1
W11) x - (y - z) = y - (x - z).
84 CHAPTER 4. LUKASIEWICZ 00- VALUED CALCULUS
(iii) By (ii), (W3), (i) and (W1) we can write -,-,x = (x - -,1) -
-,1 = (-,1 - x) - x = 1 - x = x.
(iv) By (W4) and (iii) we have 1 = (-,-,x - -,-,y) - (-,y - -,x) =
(x - y) - (-,y - -,x). Nowapply (W4) and (W6). 0
Theorem 4.2.5 Let (A, -, -,,1) be a Wajsberg algebra. Upon defining
x EB y =dej -,x - y and 0 =dej -,1, the system (A, EB, -', 0) is an MV-
algebra.
(Fl) 1 E F;
(F2) For all x, y in A, if x E Fand x ---.. y E F, then y E F.
An implicative filter F of A is said to be proper iff F =I- A; F is called
maximal iff F is proper and A is the only implicative filter strictly
containing F.
Lemma 4.2.7 The following are equivalent conditions for each subset
F of an MV-algebra A:
Praof (i) and (ii) can be easily derived from (1.8) and (1.9), re-
spectively. To prove (iii), let nl,"" nk ;::: 1 be integers, and n =
max(nl," ., nk). Then the desired conclusion follows from the inequal-
ity (Xl 0 ... 0 Xk)n ~ X~l 0 ... 0 X~k. 0
4.3 Provability
Definition 4.3.1 An axiom of the Lukasiewicz infinite-valued propo-
sitional calculus is a formula that can be written in any one of the
following ways, where a, ß and I denote arbitrary formulas:
(A2) (a ~ ß) ~ «ß ~ ,) ~ (a ~ ,));
(A3) «a ~ ß) ~ ß) ~ «ß ~ a) ~ a);
(A4) (-,a ~ -,ß) ~ (ß ~ a).
Note that one can effectively decide whether a given string of sym-
bols over the alphabet E of (4.5) is an axiom.
(i) ai is an axiom, or
(ii) ai E e, or
(iii) there are j, k E {I, ... , i-I} such that ak coincides with the
formula (aj ~ ai).
(iii / ) there are j, k E {I, ... , i-I} such that ai follows by modus
ponens from aj and ak.
(4.9) f- (Ci --+ (ß --+ ,)) --+ (ß --+ (Ci --+ ,))
(4.10) f- Ci --+ Ci
I- (a -+ (ß -+ a)) -+ (ß -+ (a -+ a)).
((ß -+ ,) -+ (( a -+ ß) -+ (a -+ ,))).
(Al)
0:2 : (··ß --+ • • 0:) --+ (.0: --+ ·ß) (A4)
0:3 : ("0: --+ ( • •ß --+ • • 0:)) --+ (( ( • •ß --+ • • 0:) --+
(,0: --+ .ß)) --+ (.-;'0: --+ (.0: --+ ·ß))) (A2)
0:4 : ((··ß --+ • • 0:) --+ (.0: --+ ·ß)) --+
(••0: --+ (.0: --+ .ß)) (MP-0: 1,0:3)
(MP- 0:2,0:4)
0:6 : (,0: --+ ·ß) --+ (ß --+ 0:) (A4)
0:7: ("0: --+ (.0: --+ ·ß)) --+ (((,0: --+ ·ß) --+ (ß --+ 0:)) --+
(••0: --+ (ß --+ 0:))) (A2)
0:8 : ((,0: --+ ·ß) --+ (ß --+ 0:)) --+ ("0: --+ (ß --+ 0:)) (MP- 0:5,0:7)
0:9 : ••0: --+ (ß --+ 0:) (MP- 0:6,0:8).
This settles (4.12).
To prove (4.13), combining (4.9) and (4.12) we easily obtain f- ß-
("0: --+ 0:). Since ß is arbitrary, letting ß be an axiom we immediately
get the desired proof.
To prove (4.14) it is enough to verify that a new proof is obtained
upon adding to any proof 0:1, ... , O:n of ••0: --+ 0: the following for-
mulas:
O:n+1: (••0: --+ 0:) --+ ((0: --+ .ß) --+ ( • •0: --+ ·ß)) (A2)
O:n+2 : (0: --+ ·ß) --+ ("0: - ·ß)
O:n+3: ((0: --+ ·ß) --+ (,,0: --+ ·ß)) --+ ((( • • 0: --+ ·ß) --+
92 CHAPTER 4. LUKASIEWICZ 00- VALUED CALCULUS
{'Y - o} f- (a - ß) - ('Y - ß)
and
{o - 'Y} f- ('Y - ß) - (0 - ß).
Therefore,
Prool By (4.18) and (4.19) in Theorem 4.4.1, it follows that the two
identities (4.23) and (4.24) yield weIl defined operations on the quotient
set Form/ =:. Moreover, by Lemma 4.4.2, 0f- E Form/ ==. Then it
remains to verify that the operations defined by (4.22) and (4.23) and
the constant 1 as defined by (4.24) satisfy equations (W1) - (W4) in
Definition 4.2.1. To prove (W1), note first that 1 -+ lai = 0f- -+ lai =
94 CHAPTER 4. LUKASIEWICZ oo-VALUED CALCULUS
Iß al,
where ß E 0f-. By (Al) we have f- a --+ (ß --+ a). On the
=
--+
other hand, by (A3), (ß --+ a) --+ a (a --+ ß) --+ ß and, since f- ß,
it follows from (4.17) that f- (a --+ ß) --+ ß. Hence, by Lemma 4.4.2
we have f- (ß --+ a) --+ a, and then ß --+ a = a. Consequently,
1 --+ lai = 0f- --+ lai = Iß --+ al = lai, and this proves (W1). In the
light of Lemma 4.4.2, conditions (W2), (W3) and (W4) followat once
from (A2), (A3) and (A4), respectively. 0
The MV-algebra
.c =def (Form/=, 0", Ee)
is called the Lindenbaum algebra 0/ Lukasiewicz infinite-valued propo-
sitional calculus.
defines an isomorphism from the lattice Theo onto the lattice of im-
plicative filters of the Lindenbaum algebra C. The inverse isomorphism
is given by
(4.29) F 1-+ {a E Form Ilal E F}. 0
An important result of classical propositional calculus is the Deduc-
tion Theorem. The next proposition gives aversion of this theorem
for the Lukasiewicz infinite-valued propositional calculus.
e~ = n
iEI
Vi- 1 ( {1 }).
Proo/: Let g: {IXol, lXII, ... } -+ A be an arbitrary map from the set of
logical equivalence classes of propositional variables, onto A. By Propo-
sition 4.5.5, 9 can be extended to a (unique) surjective homomorphism
h:.c -+ A. Let F = {lai E.c I h(laJ) = I}. Then by Theorem 4.6.3, F
corresponds to a theory 8, and from A ~ .cl F we conclude A = .c(8).
o
Remark: A straightforward generalization of the above theorem shows
that every MV-algebra arises as the Lindenbaum algebra of so me the-
ory, provided sufficiently many variables are available in the alphabet.
Proo/: By (4.8) and Theorem 4.6.7 we can safely assurne that 8 only
contains a single formula () = (}(XI , ... , X n ). In light of Proposition
4.2.9(ii), we can effectively enumerate all pairs ((), 'I/J) such that 'I/J is a
consequence of ().
Conversely, we shall now describe an effective enumeration of all
pairs ((), 'I/J) such that formula 'I/J is not a consequence of (). Arguing as in
the proof of Proposition 4.1.5, first of all we shall effectively enumerate
all triplets (x, (), <p) such that
(i) x is an m-tuple of rational numbers in [0,1] (m 2: n), and,
4.7. BIBLIOGRAPHICAL REMARKS 101
Ulam's game
103
R. L. O. Cignoli et al., Algebraic Foundations of Many-Valued Reasoning
© Springer Science+Business Media Dordrecht 2000
104 CHAPTER 5. ULAM'S GAME
(5.1) Sb = So n D,
precisely,
(5.4) r(y) =
1, if y falsifies no answer .
Any function
1 m
r: S -+ {0'--1""'--1,1}
m+ m+
is called astate 0/ knowledge in Ulam game over S with m lies/errors.
In accordance with this notation, the initial state is the constant func-
tion 1 over S. At the other extreme, the constant function 0 is the
incompatible state, in which every element of S falsifies m + 1 answers,
or more.
For every quest ion D ~ S, the positive answer
Dyes : S -+ {m/(m + 1), 1}
is naturally defined by the following stipulation:
1, if z E D
(5.5) Dyes(z) = {
1 - 1/(m + 1), if z E D.
We similarly define the negative answer Dno by the stipulation
Dno = YJYes. In other words, for each z E S:
1, if z rt. D
(5.6) Dno(z) = {
1 - 1/(m + 1), if z E D.
The proof of the following proposition is an immediate consequence of
(5.3)-(5.6), recalling from (1.2) the definition of Lukasiewicz conjunc-
tion:
5.3. OPERATIONS ON STATES OF KNOWLEDGE 107
r = Db(l)
1
0 ...
0 Db(t)
t,
Lattice-theoretical properties
111
R. L. O. Cignoli et al., Algebraic Foundations of Many-Valued Reasoning
© Springer Science+Business Media Dordrecht 2000
112 CHAPTER 6. LATTICE-THEORETICAL PROPERTIES
(IL1) 0 E I;
(IL2) If x E land y ~ x then y E I;
(IL3) If x, y E I then x V y E I.
(6.1) ( z 1= {x E L I x ~ z}.
We say that I is proper iff I =I L. We say that I is prime iff I is
proper and for any two x, y E L, if x Ay E I then either x E I or y E I.
Every ideal of an MV-algebra A is also an ideal of the underlying
lattice L(A). To see that the converse does not hold in general, let
z be a nonboolean element of A. Then (z 1 is an ideal of the lattice
L(A), but is not an ideal of the MV-algebra A. As a matter offact, by
Corollary 1.5.6, an MV-algebra A is a boolean algebra iff every ideal of
the lattice L(A) is an ideal of A.
Proof: (i) => (ii): If 1 = -'x V n.x = -,(x E9 n.x) E9 n.x then n.x ~
(n + l).x = x E9 n.x ~ n.x.
(ii) => (iii): If n.x = (n + l).x, then by induction we get n.x =
(n + k).x, for all integers k ~ O. In particular, n.x = 2n.x and by
Theorem 1.5.3(iv), n.x E B(A).
(iii) => (iv): Let no ~ 1 be an integer such that no.x E B(A). By
Theorem 1.5.3(iv), no.x = 2no.x = kno.x for all k ~ 2. Assume Y E (x).
Then Y ~ n.x for some n, whence a fortiori, for suitably large k, we get
Y ~ kno·x E B(A) n (x). Thus (x) is a stonean ideal of L(A).
(iv) => (i): By assumption, since x E (x), there is z E B(A) n (x)
such that x ~ z. Since z E (x), there is an integer n ~ 1 such that
z ~ n.x, whence, by Theorem 1.5.3(vi), 1 = -,z E9 n.x = -,z V n.x ~
-,x V n.x ~ 1. 0
(6.3) A = II Ln·
n~2
(i) A is hyperarchimedean;
(ii) Every ideal of A is a stonean ideal ofL(A);
(iii) Every prime ideal of A is maximal;
(iv) Every prime ideal of A is minimal;
X(B)
Proposition 6.4.1 For each nonzero element bE B(A), (( b], EB, -'b, 0)
is an MV-algebra and hb is a homomorphism of A onto (b] with
Ker(h b ) = (-,b].
= -'b ( -'bX EB y) EB y,
(x 1\ b) EB (y 1\ b) = (x EB y) 1\ (b EB y) 1\ b = (x EB y) 1\ b.
Definition 6.4.2 For any partially ordered set X with minimum ele-
ment 0, by an atom of X we mean an element a E X such that a >
and whenever x E X and x ~ athen either x = or x = a. °
°
The following is an immediate consequence of the definition of (b J:
°
Remark: For each < a E A, upon defining x EBa Y = (x EB y) 1\
a and -'aX = a 8 -,x, it follows that (( a], EB a , -'a, 0) is an MV-algebra.
However, if a is not a boolean element of A, then in general (aJ is not
a homomorphic image of A. For instance, let n 2: 3 and < k < n - 1.
Then k/(n-1) is an element ofthe Lukasiewiez chain Ln and (kl(n-1) 1
°
is isomorphie to L k +1' Since k + 1 < n and Ln is simple, there is
no homomorphism of Ln onto (k/(n - 1)]. On the other hand, the
existence of a homomorphism of A onto ( a] need not imply that a is a
boolean element of A. As a matter of fact, for each < a E [0, 1J, the °
map x ~ ax is an isomorphism of [0,1] onto (a]. However, a is not a
boolean element of [0,1], unless a=l.
Notation: Given a nonempty family {ai hEl of elements of an MV-
algebra A, we write ViEl ai = 1 iff 1 is the only upper bound of the
family.
6.4. DIRECT PRODUCTS 123
8.( ") = { 1 E Ai if j =i
~ J 0 E Ai if j =1= i.
Then 8i E B(P), 8i /\8j = 0, for all i =1= j, and ViEl 8i = 1. Let 7l"( P ---+ Ai
be the canonical projection. Then the kernel of h6i : P ---+ (8i ] coincides
with the kernel of 7l"i. As a matter of fact, by Corollary 6.4.1 we can
write Ker(h 6J = (,8d = {f E P I f(i) = O} = Ker(7l"i)· Thus, by
Lemma 1.2.7, Ai is isomorphie to (bd. 0
The above lemma has the following partial converse:
Al or A 2 is trivial.
124 CHAPTER 6. LATTICE-THEORETICAL PROPERTIES
weak boolean product of the {AX}XEX' For each z E X let the function
fz: X ---. UXEX A x be defined by stipulating that, for each x E X,
if x = z,
if x # z.
Since f E A and [fz = 0] = {z}, then, for condition (i) to hold it is
necessary to endow X with the discrete topology. While this topology
is Hausdorff and has a basis of dopen sets, it fails to make X into a
compact space.
the latter being open by condition (i) in Definition 6.5.1. Hence, inverse
images of basic open sets of X(C) are open in X. Now (i) follows
from the weIl known fact that continuous bijections between compact
Hausdorff spaces are homeomorphisms. Arguing as in the proof of
Theorem 1.3.2, from the assumption that A is a subdirect product of
the family {AX}XEX, letting for each x E X, Px = {I E AI I(x) =
O} = Ker(7rx ), it is easy to see that A x ~ A/Px. Hence to prove (ii) we
need to show that Px = (Qx) for each x E X. Suppose 1 E Px. Then
x E [/ = 0], and since x ~ Px is a homeomorphism from X onto X( C),
there is an element 9 E C such that x E [g = 1] = [-,g = 0] ~ [I = 0l
Therefore, -,g E Px n C = Qx and 1 ~ -'g. Hence Px = (Qx), as
required. If C = B(A), for each x E X, Px = (Qx) is a maximal stonean
ideal of A, and by Corollary 6.4.8, A/(Qx) is directly indecomposable.
On the other hand, iffor some x E X, A x is not directly indecomposable
then Qx cannot be a prime ideal of B(A). Since Qx is a prime ideal of
C, the latter must be a proper subalgebra of B(A). Hence (iii) holds.
Conversely, assurne that C is a subalgebra of B(A). Then, in the
light of Theorem 1.3.2, in order to prove that A is a subdirect product
of the family {(Q) }QEX(C), it suffices to prove the following:
Claim. The intersection of all the ideals of A that are generated by
prime ideals of C coincides with the ideal {O}.
Let 0 =1= a E A. Applying Corollary 1.2.14 to the ideal {O}, we get
a prime ideal P of A such that a fj. P. Since P n C is a prime ideal
of C, to complete the proof it suffices to show that a fj. Q = (P n C).
Suppose that a E Q (absurdum hypothesis). Then there is c E P n C
such that a ~ c. Moreover, a 1\ -,c = 0 E P, and since a fj. P, by
Lemma 6.1.1 we obtain -,c E P. Hence, 1 = cE9 -,c E P, and P = A, a
contradiction. This proves the claim.
To simplify the notation, we can safely identify A with its corre-
sponding subalgebra of TIQEX(C) A/(Q), and regard the elements of A
as functions
I:X -+ UQEx(c)A/(Q).
Suppose 1 E A, and let Q E [I = 0]. Then / E (Q), and there is gE Q
such that 1 ~ g. It follows that
Q E ac(-,g) = [g = 0] ~ [J = 0],
128 CHAPTER 6. LATTICE-THEORETICAL PROPERTIES
6.6 Completeness
By definition, a lattice L is complete iff every subset {Xi I i E I} of L
has a supremum and an infimum, which we respectively denote by
VXi , ( or V{Xi I i E I} )
iEI
and
AXi , (or A{Xi I i EI}).
iEI
Any complete lattice has a minimum element 0 = V 0 and a maximum
element 1 = 1\ 0.
Definition 6.6.1 We say that an MV-algebra A is eomplete iff its
underlying lattice L(A) is complete. We say that A is a-eomplete iff
suprema and infima exist for all finite or denumerable subsets in L(A).
By Theorem 3.5.1, the only complete and simple MV-algebras are
[0, 1] and the finite chains Ln.
Proposition 6.6.2 Every a-complete MV-algebra A (whence, a for-
tiori, every eomplete MV-algebra) is semisimple.
Proof: Assuming X E Rad(A), we shall show that X = o. As a matter
of fact, by Proposition 3.6.4 we have
(6.6) n.x ~ -'X, Le., X 0 n.x = O.
Let s = VnEN(n.X). Then for all n = 0,1, ... , we have (n + l).x ~ s.
From the monotonicity ofthe 0 operation it follows that (n+1).xex ~
sex. From (6.6) we now get
(6.7) 1\ Xi = -, V -'Xi
iEI iEI
and
and
(6.10) X V 1\ Xi = 1\ (X V Xi).
iEI iEI
(6.11) X8VXi=V(x8xi)'
iEI iEI
To this purpose, let a = ViEl Xi' Since Xi :::; a, for every i E I, X 8 Xi :::;
X 8 a. Let us assume that for each i EI, X 8 Xi :::; z. By Lemma
1.1.4(iii), Xi :::; -'x EB z for each i E I. Hence a :::; -'x EB z, and by the
same lemma, we get X 8 a :::; z. Thus, X 8 a = ViEI(X 8 Xi), which
settles (6.11). Now by (6.8), for every i E I we can write -,a :::; -'Xi;
hence, by (6.11)
iEI iEI
and
(6.13) 1\ bi E B(A);
iEI
and
From Lemmas 6.4.4 and 6.6.4 we get that for every complete MV-
algebra A, there is a one-one correspondence between the direct prod-
uct decompositions of A and the sets {bi I i E I} ~ B(A) such that
ViEl bi = 1 and bi 1\ bj = 0 whenever i =J j. More generally, given
an infinite cardinal K and a K-complete MV-algebra A (as defined by a
natural extension of 6.6.1), an easy adaptation of the proofs of Lemmas
6.4.4 and 6.6.4 yields a one-one correspondence between direct product
decompositions of A into K many factors, and sets {ba I Q' E K} ~ B(A)
such that VaEIt ba = 1 and ba 1\ bß = 0 whenever Q' =J ß. The crucial
point is that in Lemma 6.6.4 the supremum on the right hand side ex-
ists Hf so does the supremum on the left hand side, and if this is the
case the suprema coincide.
Proof: It is easy to check that the two lattices L(P) and I1 iEI L(A i )
coincide. Clearly, the direct product of complete lattices is a complete
lattice. Thus, if each Ai is complete, then so is P. The converse is
an immediate consequence of Corollary 6.6.5(ii), together with Lemma
6.4.4. 0
(6.17) x $ xoOoO;
Sinee, by Lemma 6.7.2, both x** and y** are elements of B(A), we
must either have x** ::; -.z or y** ::; -.z. By (6.17), either x::; -.z or
y ::; -.z. Therefore, (-.z] is a prime ideal of L(A) and, by Corollary
1.5.6, it is also a prime ideal of A. Thus the quotient MV-algebra
AI (-.z] is totally ordered. By Corollary 6.4.1, AI (-.z] is isomorphie
to (z]; by Corollary 6.6.5(ii), (z] is eomplete. Sinee a is an atom of
(z] iff ais an atom of A and a::; z, the desired result now follows
from the observation that [0,1] is the only simple eomplete atomless
MV-algebra. 0
x = x A S A = x A Va;*
iEI iEI iEI
(MI) -,0 = 1;
(M2) -,-,x = x;
(M3) -,(x V y) = -,x A -'y.
Proposition 6.6.2 is due to [132]. In the same paper the author also
proved Lemma 6.6.4. Also see [49]. The direct algebraic proofpresented
here is due to [14]. Equations (6.17)-(6.20) hold in every distributive
pseudocomplemented lattice (see [103], or [12]).
For a proof of the c1assical result of Tarski referred to in the proof
of Theorem 6.8.1 see, for instance, [25], Chapter 5, §5, Theorem 17.
Corollary 6.8.2 is due to Bosbach (see [31]). In the same paper one can
also find a proof of Corollary 6.8.3 (see also [49]).
Complete MV-algebras, convergence properties, various kinds of
topological and order completions are considered, e.g., in [91], [89],
[124], [14], [15], [128], [125], [126], [69], [224], [119] and [127]. For
model-completions see [135] and [136].
Chapter 7
In our present notation, Theorem 2.4.5 states that the map a.- 'PA(a)
= [(a), (0)] defines an isomorphism ofthe MV-algebras A and r(E(A)).
Using the maps 'PA (A E MV), we obtain
Lemma 7.1.3 Suppose Gis an f-group with order unit u, and let A =
r(G, u) ~ G. For each 0 :s; a E G there is a unique good sequence
g(a) = (al,"" an) 0/ elements 0/ A such that a = al + ... + an.
a - al = a - (a 1\ u) = 0 V (a - u) = (a - u)+.
For the induction step, assume the identity to be true for k = m. Then
We have used the inequality t+ 2:: ((2t - u) 1\ u), which follows from
(t V 0) - (( 2t - u) 1\ u)
= (u - t) V (t - u) V (u - 2t) V -u
2:: (u - t) V (t - u) = It - ul 2:: o.
Hence (al,"" an) is a good sequence of elements of A = [0, u], and
n
a - Lai = an+l = an +2 = ... = O.
i=l
Finally, to prove uniqueness, note that if (bI, ... ,bm ) is a good sequence
of elements of [0, u] such that a = :E~l bi , then by definition of good
sequences together with Lemma 2.1.3(ii), we have
= (a - bl - ... - bi-d 1\ u
as required. 0
It follows from the above lemma that the correspondence a ~ g(a)
defines an injective mapping from the positive cone G+ of G, onto the
monoid Mr(G,u) of good sequences of r(G, u). In order to show that
this mapping is both a monoid-isomorphism and a lattice-isomorphism,
we prepare the following
7.1. INVERTING THE FUNCTOR r 143
GrCG,u) be defined by
In the light of Corollary 7.1.6, using the maps '1jJ(G,u) (for all (G, u) E A),
we have the following
(G,u) (H,v)
'1jJ(G,u) ! ! 'IjJ(H,v)
S(r(G,u)) S(r(H, v))
Proof" By Lemma 7.1.3 we can write g(a+) = (al, ... , an), for a
uniquely determined good sequence (al' ... ' an) E MrCG,u). Letting
h = r(j), we then obtain
n n
f(a)+ = f(a+) = L f(ai) = L h(ai),
i=l i=l
'IjJ(H,v) (j(a))
146 CHAPTER 7. MV-ALGEBRAS AND f-GROUPS
= [g(f(a)+), g(f(a)-)]
= [h*(g(a+», h*(g(a-»]
= h#([g(a+), g(a-m
= 8(r(f» ([g(a+), g(a-m
= 8(r(f» ('I/l<G,u) (a».
o
From Theorems 7.1.2 and 7.1.7 we immediately get
7.2 Applications
Lemma 7.2.1 Let h: r(G, u) -+ reH, v) be a homomorphism 01 MV-
algebras.
(i) There is a unique unital f-homomorphism I: (G, u) -+ (H, v) such
that h = r(f);
(ii) 11 h maps r(G,u) onto r(H, v) then 1 maps G onto Hj
(iii) 11 h maps r(G, u) one-one into reH, v) then 1 maps G one-one
into H.
Prool: (i) In the light of Corollary 7.1.6 and Theorem 7.1.7, let 1 be
defined by
1 = 'I/lw,v) (8(h»'I/l<G,u)'
By (7.5), (7.6) and (7.10), for each a E G we ean write
I(a) = 'I/lw,v) ([h*(g(a+», h*(g(a-»]).
In the partieular esse when a E [0, u), sinee g(a+) = g(a) = (a) then
[h*(g(a+», h*(g(a-))] = [(h(a», (0)], whenee
I(a) = 'I/lw,v) ([(h(a», (0)]) = h(a).
7.2. APPLICATIONS 147
= ((lxi /\ u) + Iyl) /\ u
= (lxi + Iyl) /\ u
~ Ix - yl /\u,
Then it is not hard to see that, for any irrational 0 < ß < 1, Sa = Sß
° °
if and only if = ß or = 1 - ß.
As a matter of fact, assume Sa = Sß and write 0 < 0 < ß < 1
°
without loss of generality. By definition, = m + nß and ß = p + qa.
It follows that °= m + np + nqa, whence m + np = 0 and nq = 1.
In case n = q = 1, we get the contradiction 0 < ß - a = p < 1. In
the remaining case when n = q = -1, we get a = m - ß < 1, whence
m = 1, as required.
By Corollary 7.2.6, we then have
Proo/: Put G(J) =def 4>(<PA(J». From Lemmas 7.3.2 and 7.3.3(i) it fol-
lows that <PA is both a monoid-isomorphism and an order-isomorphism
from J onto G(J)+. Uniqueness follows from the fact that any f-group
is determined by the ordered monoid of its positive elements. 0
Remark: Suppose that J is a nilpotent ideal of an MV-algebra A. If
xE J, then nx E J, and since J is nilpotent, nx0nx = O. This implies
that nx ::; ...,nx ::; ""x. Therefore, J ~ Rad(A). By Lemma 7.3.3(ii),
J ~ Rad(A) iff J is nilpotent. In other words, Rad(A) is the largest
nilpotent ideal 0/ A.
(x,a) if x< y
(7.12) (x, a) 1\ (y, b) = { (y, b) if y <x
(x,al\b) if x = y
(7.13) A(h)((m, a)) = (m, h(a)) for each pair (m, a) E ZoG.
and
-,(i, a) = (1 - i, -a).
By definition of lexicographic order it follows that
Gase 1: x EB Y E Rad(A).
Then both x and Y are in Rad(A), and equation (7.14) follows at
once from Lemma 7.3.3(i).
Gase 2: x EB Y E -,Rad(A).
Subcase 2a: x E -,Rad(A) and Y E -,Rad(A).
Then by Lemma 7.3.3(ii), aA(xEBy) = (1,-<PA(-,x0-,y)) = (1,0),
and by definition oflexicographic order, aA(x)EBaA(Y) = (2, -(<PA(X) +
<PA(Y))) A (1,0) = (1,0).
Subcase 2b: x E Rad(A) and Y E -,Rad(A).
From <PA(-'(X EB y)) = (u - <PA(X) + <PA(-'Y) - u) V 0 = (<PA(""Y)-
<PA (X)) VO, we get aA(xEBy) = (1, (<PA(X) -<p(-,y)) AO) = ((0, <PA(X)) +
(1, -<PA(""Y))) A (1,0) = aA(x) EB aA(Y)·
Subcase 2c: x E -,Rad(A) and Y E Rad(A).
One argues as in the previous case, by interchanging the roles of x
and y. 0
o
Conversely, the following lemma shows that the composite functor
is naturally equivalent to the identity functor of the category of
~.6.
MV-algebras:
Varieties of MV-algebras
157
R. L. O. Cignoli et al., Algebraic Foundations of Many-Valued Reasoning
© Springer Science+Business Media Dordrecht 2000
158 CHAPTER 8. VARIETIES OF MV-ALGEBRAS
°
function r[O,I): [0, l]n -+ [0,1] is eontinuous. Therefore, sinee by Propo-
sition 3.5.3, Ais dense in [0,1], the equation r(xb .. . ,xn ) = holds in
[0, 1] iff it holds in A. The desired result is now a eonsequenee of the
Completeness Theorem. o.
(i+j-n+l,x+y-b) ifi.+.J>.n-l
(8.3) (i x) 0 (' ) = { or If 1, + ) = n - 1
, ), Y and x + y > b.
(0,0) otherwise.
A straight forward verification shows that r(A(G), (n - 1, b)) is a
nonsimple MV-chain of rank n. Indeed, the projection 7l"1: A(G) - Z,
defined by 7l"l((j, t)) = j, for each (j, t) E A(G), is a surjective f-group
homomorphism, with 7l"l((n - 1, b)) = n - 1. Hence
we have
Kn = r(A(Z), (n - 1,0))
and
Rn = r(A(Z), (n - 1,1)).
Proof: Immediate from Propositions 8.1.1 and 8.1.2, upon noting that
A E C implies AI Rad(A) E C. 0
In the next sections we shall establish that, conversely, if the ranks
of all MV-chains in a variety C are bounded, then C 1= MV. We first
consider simple MV-algebras of finite rank, i.e., the algebras Ln. The
simplicity assumption shall be removed in a subsequent section.
Proof: The first result easily follows from Theorems 8.2.2 and 1.5.3(iv).
For the second result, by the remark following Proposition 3.5.3, L 2 is
a subalgebra of L 3 • Thus, an equation is simultaneously satisfied by L 2
and by L3 iff it is satisfied by L3 . Thus, C( {L2 , L3 }) = C(L3 ), and the
desired conclusion now follows from Theorem 8.2.2. 0
Lemma 8.3.2 Por each integer n 2:: 2 and each nonsimple MV-chain
A of order n, Rn E C(A).
is an embedding, whence
Proposition 8.3.3 Let G be an l-group and 0< bEG. Then for each
integer n 2:: 2, r(A(G), (n - 1, b)) E K n .
fm(aKn(b l , ... , bk)) = fm(b i ) = (v, mw) = aHn (fm(bd, ... , fm(b k)),
thus establishing (8.6) for g(a) = 0.
Proceeding by induction, pick an integer d > 0, and suppose that
(8.6) holds for all MV-terms ~ in the variables Xl! .. . , Xk such that
g(~) < d.
Let a(xI, . .. , Xk) be an MV-term such that g(a) = d. From Sec-
tion 1.4 we know that precisely one of the following two cases must
occur:
Gase 1: There is an MV-term 7(XI, ... , tk) such that a = ""7;
Gase 2: There are MV-terms /-l(XI, . .. , Xk) and V(XI, . .. , Xk) such that
a=/-lEBv.
For notational convenience, and without any essential loss of gener-
ality, we shalllimit ourselves to considering MV-terms in one variable.
In Case 1, we can write g(7) = g(a) - 1 < d, whence, by induction
hypothesis,
166 CHAPTER 8. VARIETIES OF MV-ALGEBRAS
= (n - 1,1) - THn(fm(b))
::; (n - 1,1) - fm(TKn(b)) + (0, g(7))
= (n - 1,0) - fm (T Kn (b)) + (0, 1 + g(T))
= fm (u Kn (b)) + (0, g(T))
and
uHn(fm(b)) ~ (n - 1,1) - fm(TKn(b)) - (0, g(T))
~ fm(TKn(b)) - (O,g(u)).
= ((n - 1,1) A (fm (J.lK n(b)) + fm (lIK n(b))) - (0, g(J.l) + g(lI))
8.4. KOMORl'S CLASSIFICATION 167
If m > g(7r), then ... , /m(b k )) > (0,1) and Rn does not
7r Hn (fm(bt},
satisfy the equation 7r(Xll ••• ' Xk) = O. We conclude that K n must
satisfy all equations that are satisfied by Rn. 0
From Theorem 8.3.1, Proposition 8.3.3 and Lemmas 8.3.2 and 8.3.4,
we obtain
and
(kj - (k - 1)(n - 1), ka) if .k~ > (~- 1)(: - 1) or
(810)(. )k = { kJ - (k 1)(n 1) and
. J,a a> 0
(0,0) otherwise.
(ii) A E C( {K 2 , K 3 ,· •. , K n });
Proof: The equivalence between (i) and (ii) easily follows from the
above Lemma, together with Theorems 1.3.3 and 1.4.6. To obtain the
equivalence between (ii) and (iii) note that AI Rad(A) is semisimple
and that L 2 , ... ,Ln is the complete list of all simple algebras of rank
at most n. 0
Div(n)
be the set of all divisors d ~ 1 of n.
and
and
(Enp ) (p.xp-1)n = n.xp.
a-<b
we understand that b covers a, in the divisibility order. Stated other-
wise, a divides band there is no c with c =f. a and c =I b, such that a
divides c and c divides b. For each c E Div( n - 1) we define
n-I
{i E {I, ... , nr } I gcd(ail,"" air, n - 1) = -d-}
n-I n-I
= H(-d-) \ (U{H(c) leE Min(-d-' n - In)·
As a consequence, we can also write
n-I n-I
a(n, T, d) = ~H(-d-) - ~U{H(c) I cE Min(-d-' n -ln
= ~) _I)#X nH(c),
cEX
where
n-I
o=J x ~ Min(-d-' n - 1).
For each nonempty X ~ Min(ndl, n - 1) let lcm(X) be the least
common multiple ofthe elements of X. Then nCEX H(c) = H(lcm(X)),
whence
~ n H(c)
cEX
= (1
n-I
cm
(X) + Ir·
Note that a E Min(n~l, n - 1) iff n;l E MaxDiv(c). Defining now
n-I
X* = { - leE X},
c
we get X ~ Min(ndl, n-I) iff X* ~ MaxDiv(d). We have just proved
that 1c~(~) = gcd(X*), for all X=/: 0. Let us stipulate that gcd(0) = d.
176 CHAPTER 8. VARIETIES OF MV-ALGEBRAS
Free~ = rr
dEDiv(n-l)
La:(n,r,d)
d+l ,
where
• and so on.
rv T
11 l$i$s L o(n,T,d;)
d; = erm (Ln, r ) .
is an isomorphism of D onto E. 0
Advanced topics
The first part of this chapter deals with disjunctive normal forms in
the infinite-valued calculus of Lukasiewicz. We shall generalize the
Farey-Schauder machinery of Chapter 3 to formulas in any number of
variables. Disjunctive normal forms will be the key tool to prove Mc-
Naughton's theorem, generalizing the proof given in 3.2.8 for functions
of one variable. We shall also discuss the relationships between normal
form reductions and toric desingularizations, and the correspondence
between MV-algebras and AF C· -algebras. Strengthening Corollary
4.5.3, we shall show that the tautology problem in the infinite-valued
calculus is in fact co-NP-complete, thus having the same complexity as
it boolean counterpart. We shall give a proof of Di Nola's representa-
tion theorem for all MV-algebras.
With the possible exception of the first section, the pace in this chap-
ter is usually fast er than in the previous chapters. Prerequisites ranging
from l-group theory, polyhedral topology, NP-completeness theory, al-
gebraic geometry, functional analysis, model theory, may be necessary
for a complete understanding of the results in the following sections.
In a final section appropriate references shall be given to the interested
readers.
179
R. L. O. Cignoli et al., Algebraic Foundations of Many-Valued Reasoning
© Springer Science+Business Media Dordrecht 2000
180 CHAPTER 9. ADVANCED TOPICS
(V) Ms be the (n + 1)
x (n + 1) matrix whose ith row coincides
with vfom. We say that the list of vertices va, VI, ... , V n has the
positive orientation iff 1 ~ det(Ms).
Let c be the set of all n-dimensional simplexes of the form [VT' F],
where F ranges over the (n - 1)-dimensional faces of T not containing
the point VT. Replacing in S the simplex T by the new simplexes of
c, we obtain a rational subdivision of S, Le., a set SI of n-dimensional
simplexes with rational vertices such that
(i) each simplex of S is a union of simplexes of SI! and
182 CHAPTER 9. ADVANCED TOPICS
Proof: (i) Let Wo, ... ,Wn be a positively oriented list of vertices of S.
By hypothesis, v = WI, for some l = 0, ... , n. Then the sequence of
coefficients ao, . .. ,a(n-l), b coincides with the lth column of the inverse
9.1. MCNAUGHTON'S THEOREM 183
(9.6) Po' = {x E [0, ljn I 90'(1) (x) ~ 9u(2)(X) ~ ••• ~ 9u(q) (x)}.
(9.7) v E nP
TrEa
Tr , [0, Ir = U PTr •
TrEa
For any J.1,11 E {} the polyhedra PJ.I. and P" are either disjoint, or else
they intersect in a common face.
Fix now an arbitrary permutation 7r E {}. Then, however we choose
apermutation p E {} and i =1= j, the difference 9Tr(i) - 9T(0) will never
vanish in the interior of Pp- Moreover, there is a unique index i Tr E
{I, ... , q} such that
pieces. By our ass um pt ions about i and y, there is y' =f. y such that TJ
coincides with g7r(i) over the interval N = [y', y] ~ [x, y]. Thus by our
assumption about x, over the half-open interval N\ {y} the graph of TJ
lies strictly below the segment [X, Y]. Among all W E [x, y] different
from y and such that (w, TJ(w)) E [X, Y], there is a point z nearest
to y (It is quite possible that z = x). Let j E {I, ... , q} be such
that TJ( z) = g7r(j) (z), and TJ coincides with g7r(j) on a small open interval
N' = [z, z'] ~ [z, y], for some z' =1= z. Then the restriction of the graph
of g7r(j) to the half-open interval N'\ { z} lies strictly below the segment
[X, Y]. Moving from z to y, we get g7r(j)(Y) < hv(Y). Moving from z to
x, we get g7r(j) (x) > hv(x) = g7r(i,..) (x), whence j < i 7r . Thus, g7r ~ g7r(j)
over all of Rn, which settles our claim.
To conclude the proof, let gfl = V7rEfl g7r. By our claim, together
with (9.8), since for each ()" E n, gfl = gU = hv over pun U star(v), then
gfl = hv over star(v). Trivially, gfl V 0 = h v = Oover [0, l]n\ Ustar(v).
Since gfl V 0 is a (V A)-combination of the gi U and the latter, by (i)
and Lemma 3.1.9, are elements of Free n , the desired conclusion follows
from Proposition 1.1.5. 0
Theorem 9.1.5 FOT each cardinal K, the free MV-algebra FreeK, is
given by the McNaughton functions over [0, Ir, with pointwise oper-
ations.
Praof: In the light of Propositions 3.1.4 and 3.1.8 it suffices to show
that every McNaughton function f : [0, l]n ---+ [0,1] belongs to Free n .
To this purpose, let U be as in Theorem 9.1.2. Let U E Qn be an
arbitrary vertex of U, and let d be the least common denominator of
the coordinates of u. Then f(u) = mu/d for some integer 0 ~ m u ~ d.
Moreover, f is linear over each n-dimensional simplex of U. Let hu E
'Hu be as in Definition 9.1.3. Then the two continuous functions fand
Eumuh u coincide over all of [0, l]n. Replacing sum by truncated sum
EB, we have
(9.10) f = EB u mu·hu = h u EB h u EB ... EB h u (mu times).
Since, by Proposition 9.1.4, h u E Free n , then f E Freen. a required.
o
Remark: The set 1iu in the above proof is a DNF (Disjunctive Normal
Form) reduction of f.
9.2. NONSINGULAR FANS AND NORMAL FORMS 185
Theorem 9.2.2 Any two Schauder sets I and.c in [0, 1]n have a com-
mon star refinement.
Proof: Readers familiar with toric varieties will recognize this state-
ment as a reformulation of the strong form of Oda's conjecture. While
for several years only the one-dimensional case of the conjecture was
known to be true, (by Danilov's decomposition theorem) it appears
that Morelli has finally settled the conjecture in the affirmative for the
general case. 0
Thus, for any unimodular triangulation U we can explicitly con-
struct H u starting from any Schauder set H w , and then applying only
one (deduction) rule, namely the one-step star refinement. This method
is more efficient than the inductive procedure given by Lemma 3.1.9.
Theorem 9.2.3 For any two Schauder sets Hand I there is a star
refinement H* ofH such that every element ofI is a (truncated) sum of
elements ofH*, as in the above formula (9.10). Moreover, all one-step
star refinements leading /rom H to H* may be assumed to be binary.
Corollary 9.2.4 Every set Q = {/I, ... , fk} ~ Freel has aleast DNF
reduction H, i.e., a Schauder set H satisfying the following conditions:
• (i) Each fi is a (truncated) sum ofthe hats in H as in (9.10);
9.3. COMPLEXITY OF THE TAUTOLOGY PROBLEM 187
Corollary 9.3.2 Let P(XI, ... , Xn) = C + mIX + ... + mnXn be a linear
polynomial with integer coejJicients c, ml,' .. ,mn' Let fl/>(XI,' .. ,xn) be
the McNaughton function associated to a formula </>. Suppose fl/> eoin-
eides with p over an n-simplex T ~ [0, l]n. Then we have
(9.12) max(lmll,···, ImnD ::; 1</>1.
o
Proposition 9.3.3 Let fl/>(XI, ... , Xn) be the MeNaughton funetion as-
sociated to a formula </>. Assume fl/> does not eoineide with the zero
funetion over [0, l]n. Then there exists a point
x = (aI/b, ... , an/b) E [O,I]n
with ai, b E Z and 0 ::; ai ::; b (i = 1, ... , n) such that fl/>(x) > 0 and
o < b< 2(411/>1 2 ).
Definition 9.3.5 For eaeh integer n 2: 1 and t 2: 2 we define the [0, 1]-
valued function fn,t by stipulating that for all x = (Xl, ... , X n ) E [O,l]n
t times
Further, for eaeh integer i 2: 1 the formulas 4>i, 'l/Ji,t, and Pn,t are
defined by
190 CHAPTER 9. ADVANCED TOPICS
enumerate the edges oJ[O, ~]n adjacent to Vj. Por each i = 1, ... , n and
t 2:: 2 let Yji be the point lying on edge Cji at a distance 1ft from Vj.
Let Tj be the n-simplex with vertices Vj, Yjb"" Yjn' Then we have
(i) fn,t{Vj) = 1;
(ii) fn,t{Yji) = 0;
(iii) fn.t is linear over each simplex Tj ;
(iv) fn.t vanishes in [0, l]n outside U;:l Tj. 0
Lemma 9.3.7 Adopt the above notation. Let 4> = 4>(X1 , ••• ,Xn) be a
formula. Let t = 14>1, and say without loss of generality, t 2: 2. Then 4>
is a tautology in the boolean calculus iff Pn.t ~ 4> (i. e., "'Pn.t EB 4» is a
tautology in the infinite-valued calculus.
By our analysis, together with Proposition 9.3.1, for each point y lying
in the interval [Vj, x] we have
f~,t(Y; u) :5 -t = -14>1·
Further,
f~(Y; u) ~ -14>1·
Thus, for all Y E [Vj, x],
(*) For every projection pE A, among all classes [q] such that [P] +
[q] = [lA] there is a smallest one, denoted ...,[p], namely the class
[lA - p].
Remarks.
1. Classes of AF C"-algebras A whose L(A) is a lattice include com-
mutative, finite-dimensional, continuous trace, liminary with Hausdorff
spectrum, as well as all AF C"-algebras with comparability of projec-
tions in the sense of Murray-von Neumann.
2. Intuitively, part (iii) in the above theorem suggests that MV-algebras
are a noncommutative generalization of boolean algebras. Since, by
Theorem 4.6.9, every countable MV-algebra is the Lindenbaum alge-
bra of a theory e in the infinite-valued calculus of Lukasiewicz with
denumerably many variables, any set ofaxioms for e is a presentation
of a unique AF C"-algebra A e . The complexity of the word problem
(in the sense of Theorem 4.6.10) of eis a faithful measure of the com-
binatorial complexity of A e . While most AF C"-algebras existing in
the literature have polynomial time complexity, if the word problem e
happens to be Gödel incomplete, then necessarily Ae has a nontriv-
ial ideal. This shows the incompatibility of two equally imprecise and
interesting conjectures: (a) that the C*-algebraic mathematizations of
physical systems existing in nature should have no quotient structures,
and (b) that Gödel incomplete AF C"-algebras might exist in nature.
3. Readers familiar with Grothendieck's group will recognize in the
f-group Q(K(A)) the group Ko(A) equipped with the order induced by
the image Ko(A)+ of the generating monoid of Ko(A).
r(G(I),u(I)) C::! Bll. Let us embed G(I) into a totally ordered divis-
ible abelian group K(I) with the same strong unit u(I). Let D(I) =
r(K(I),u(I)). Then from Lemma 7.2.1 it follows that Bll is embed-
dable into D(I). Since any totally ordered divisible abelian group is
elementarily equivalent to the additive group R of real numbers with
natural order, it follows that D(l) is elementarily equivalent to the
MV-algebra [0,1]. By Frayne's theorem in model theory, each D(I) is
elementarily embeddable in a suitable ultrapower [0,1]*/ of [0,1]. The
joint embedding property of first-order logic now yields an ultrapower
[0,1]* (only depending on the cardinality of B), such that each MV-
algebra [0,1]*/ is elementarily embeddable into [0,1]*. Thus every Bll
is embeddable into [0, 1]*, whence the desired conclusion immediately
follows. 0
Further Readings
197
R. L. O. Cignoli et al., Algebraic Foundations of Many-Valued Reasoning
© Springer Science+Business Media Dordrecht 2000
198 CHAPTER 10. FURTHER READINGS
Questions with lies, links the two interpretations. See [92) for a first
exploration. Other semantics for infinite-valued logic are given, e.g., in
[223}, [236) and [238).
Although many-valued logic was originally introduced in algebraic
form by Lukasiewicz hirnself (see [139] and references therein), its defini-
tive algebraization may be ascribed to Chang [36}, [38}, [40) who in-
troduced MV-algebras and used the model theory of totally ordered
abelian groups to prove the completeness of the Lukasiewicz axioms.
The main ideas of Chang's completeness theorem are presented by
Rosser in his early survey [219], and compared with the techniques used
by Rose and Rosser for their own proof [216], "involving a frightening
amount of detail" ([219, p.140]).
In 1940 and 1941, Moisil introduced another dass of algebraic struc-
tures for his study of Lukasiewicz n-valued propositional calculi. These
algebras are distributive lattices with a negation operation and some ad-
ditional unary operations expressing modality (see [157)). While Moisil
named these structures Lukasiewicz n-valued algebras, Rose proved
that for n 2:: 5, it is impossible to define Lukasiewicz's n-valued im-
plication building on Moisil's modal operations (see [44, p.2J). In the
papers [46) and [47) it is shown how Moisil's constructions can be mod-
ified so as to provide adequate algebraic counterparts of Lukasiewicz
n-valued propositional calculi.
Independently of Lukasiewicz, Post in 1921 introduced his n-valued
propositional calculi, see [203]. The algebraic counterparts of Post's
calculi were described by Rosenbloom in 1942, [217) and named by hirn
Post algebras of order n. In 1960 Epstein [79] investigated Post algebras
from the lattice-theoretic viewpoint, (see [76) for Epstein's theory and
its subsequent developments). In [44] it was shown that Post algebras
of order n are obtainable from (Moisil's) Lukasiewicz algebras of order
n by adding n - 2 constant operators.
The monograph [28] presents a detailed study of Moisil algebras,
Post algebras and (Moisil's) Lukasiewicz n-valued algebras.
Among the texts concerned with many-valued Lukasiewicz logics let
us mention Rosser-Thrquette [220}, Ackermann [1}, Rescher [206},
Gottwald [102], Bolc-Borowik [29}, and Malinowski [143]. The books by
W6jcicki [244] and Hajek [112] contain chapters devoted to Lukasiewicz
10.2. CURRENT RESEARCH TOPICS 199
10.2.1 Product
Generalized conjunction connectives over the unit real interval (also
known as T-norms) are interesting objects of study from various view-
points. The reader may consult the monograph by Butnariu and Kle-
ment [34] and the relevant chapters in Hajek's book [112] for back-
ground. One of the merits of T-norm theory is to show that a
substantial portion of the expressive power needed for applications of
infinite-valued logic to control theory, probability theory, and game the-
ory with variable coalitions would be provided by a logic incorporating
a product connective jointly with Lukasiewicz disjunction and negation.
Many people are actively pursuing this line of research, including Di
Nola, Dvurecenskij, Esteva, Georgescu, Godo, Leustean, Panti, Riecan.
We refer, e.g., the reader to the papers [81], [207] and to Montagna's
analysis [159] of the relationships between MV-algebras "with product"
and various categories of lattice-ordered rings.
A different approach is taken in [187], using tensor products-the
latter perhaps being the bare minimum needed for if-then-else approx-
imations of continuous real-valued functions. According to this ap-
proach, the Lukasiewicz calculus is as basic as groups are in algebra-
and (tensor) multiplication naturally appears as the fulfillment of the
following desideratum: Having a "conjunction connective" that dis-
200 CHAPTER 10. FURTHER READINGS
10.2.3 Deduction
By contrast with finite-valued logic-and notwithstanding its rich alge-
braic structure-infinite-valued Lukasiewicz logic lacks a natural prooj
theory. Currently used proof techniques spuriously range from variants
of integer programming [108] and Fourier-Motzkin elimination [240} to
the calculation of level sets of McNaughton functions [189], [2]. The
paper [4] is a first attempt to introduce an analytic calculus for the
infinite-valued propositional logic of Lukasiewicz. Tight estimates are
given for the complexity of the consequence relation, thus strengthening
earlier results in [167].
Methodologies for automated deduction in infinite-valued Lukasiew-
icz logic are currently investigated by several people, including Aguz-
zoli, Baaz, Ciabattoni, Escalada Imaz, Fermüller, Hähnle, Lehmke,
Manyia Serres, Olivetti, PauHk, Salzer, Vojtas, Wagner. See for in-
stance [239], [137], [109], [80], [189], [2], [3], [240], the survey paper
[111] and the handbook chapter [9]. Various not ions of literal, clause,
resolution are being considered, and various types of deduction pro ce-
dures for "easy cases" are implemented, by analogy with the Horn case
and 2-CNF case in the classical propositional calculus.
Altogether, much work is still to be done before proof theory and
automated deduction in infinite-valued propositional logic reach a ma-
ture stage.
203
204 BIBLIOGRAPHY
[10] BAAZ, M., FERMÜLLER, C.G., SALZER, G., ZACH, R. (1998) La-
beled ealeuli and finite-valued logics. Studia Logica. Special issue on Many-
valued logics, (Mundici,D., Ed.,) 61, p. 7-33.
[11] BAAZ, M., HAJEK, P., KRAIJCEK, J., SVEIDA, D. (1998) Embedding
logics into produet logie. Studia Logica. Special issue on Many-valued
logics, (Mundici,D., Ed.,) 61, p. 35-47.
[22J BELLUCE, L.P., DI NOLA, A., SESSA, S. (1994) The prime spectrum
of an MV-algebra. Mathematical Logic Quarterly, 40, p. 331-346.
[23J BERLEKAMP, E.R. (1968) Block coding for the binary symmetrie chan-
nel with noiseless, delayless feedback. In: Error-correcting Codes.
(Mann, H. B., Ed.) Wiley, New York, p. 330-335.
[26J BIRKHOFF, G., VON NEUMANN, J. (1936) The Logie of Quantum Me-
chanics. Annals of Mathematics, 37, p. 823-843.
[27J BLOK, W.J., FERREIRIM, I.M.A. (1993) Hoops and their implicational
reducts. In: Aigebraic Methods in Logic and in Computer Science,
Institute of Mathematies. Polish Academy of Sciences, Warsaw, p. 219-
230. (Banach Center Publications, v. 28)
[35] CASARI, E. (1989) Comparative logics and abelian f-groups. In: Logic
Colloquium 1988, Padova. (Ferro, R. , Bonotto, C., Valentini,S., Za-
nardo, A, Eds.), North-Holland, Amsterdam, p. 161-190. (Studies in Logic
and the Foundations 0/ Mathematics)
[39] CHANG, C.C. (1963) Logic with positive and negative truth values.
Acta Philosophica Fennica, 16, p. 19-39.
[40] CHANG, C.C. (1998) The writing of the MV-algebras. Studia Logica,
special issue on Many-valued logics, (Mundici,D., Ed.,) 61, p. 3-6.
[43] CICALESE, F., MUNDICI, D. (199?) Optimal binary search with two
unreliable tests and minimum adaptiveness, In: Proceeding of the Eu-
ropean Symposium on Algorithms, ESA'99. Lecture Notes in Com-
puter Science. To appear.
[45] CIGNOLI, R. (1971) Stone filters and ideals in distributive lattices. Bul-
letin M athematique de la Societe des Sciences M athematiques et Physiques
de la Republique Populaire Roumaine, 15, p. 131-137.
BIBLIOGRAPHY 207
[57] CIGNOLI, R., DI NOLA, A., LETTIERI, A. (1991) Priestley duality and
quotient lattices of many-valued algebras. Rendiconti Circolo Matematico
di Palermo, Serie II, 40, p. 371-384.
[62] CZYZOWICZ, J., MUNDICI, D., PELC, A. (1989) Ulam's searching game
with lies. Journal of Combinatorial Theory, Series A, 52, p. 62-76.
[70J DI NOLA, A., GEORGESCU, G., SESSA, S. (199?) Closed ideals ofMV-
algebras. In: Proceedings XIth Brazilian Logic Meeting, Special
issue of Contemporary Mathematics, American Math. Society. To appear.
[72J DI NOLA, A., LIGUORI, F., SESSA, S. (1993) Using maximal ideals in
the classification of MV-algebras. Portugalia Mathematica, 50, p. 87-102.
[74J DUNN, J. M., EpSTEIN, G., (Eds.) (1977) Modern Uses ofmultiple
valued Logics. D. Reidel, Dordrecht.
[81] ESTEVA, F., GODO, L., MONTAGNA, F. (199?) The LP and LP1/2
logics: two complete fuzzy systems joining Lukasiewicz and product logic.
Archive for Math. Logic. To appear.
[107J GUZICKI, W. (1990) Ulam's searching game with two lies. Journal 01
Combinatorial Theory, Series A, 54, p. 1-19.
[118] Hoo, C.S. (1995) Maximal and essential ideals of MV-algebras. Math-
ware and Soft Computing, 2, p. 181-196.
[119] Hoo, C.S. (1997) Topological MV-algebras. Topology and its Appli-
cations 81, p. 103-12l.
[120J HOo, C.S., RAMANA MURTY, P.V. (1987) The ideals of a bounded
commutative BCK-algebra. Mathematica Japonica, 32, p. 723-733.
[140] MAC WILLIAMS, J.F., SLOANE, N.J.A. (1986) The Theory ofEr-
ror-correcting Codes. 5th ed. North-Holland, Amsterdam. (Mathem-
atical Library)
[144] MANGANI, P. (1973) Su certe algebre connesse con logiche a phI valori.
Bollettino Unione Matematica Italiana (4), 8, p. 68-78.
[146] MARTfNEZ, N.G. (1990) Una dualidad topol6gica para estructuras al-
gebraicas omenadas. Universidad de Buenos Aires. (Ph.D. Thesis)
[147] MARTfNEZ, N.G. (1990a) The Priestley duality for Wajsberg algebras.
Studia Logica, 49, p. 31-46.
[149J MCCALL, S. (Ed.) (1967) Polish Logic 1920-1939. Cl aren don Press,
Oxford.
[153J MENU, J., PAVELKA, J. (1976) A note on tensor produets of the unit
interval. Commentarii Math. Univ. Carolinae, 17, p. 71-83.
[165] MUNDICI, D. (1986b) Mapping abelian l-groups with strong unit one-
one into MV-algebras. Journal 0/ Algebra, 98, p. 76-81.
[169} MUNDICI, D. (1987c) Every abelian l-group with two positive genera-
tors is ultrasimplicial. Journal 0/ Algebra, 105, p. 236-241.
[1751 MUNDICI, D. (1992) The logic of Ulam's game with lies. In: Knowl-
edge, Belief and Strategie Interaetion. (Bicchieri, C., Dalla Chiara,
M.L., Eds.,) Cambridge University Press, p. 275-284. (Cambridge Studies
in Probability, Induction and Decision Theory)
218 BIBLIOGRAPHY
[185J MUNDICI, D. (1998) Nonboolean partitions and their logic, In: First
Springer-Verlag Forum on Soft Computing. Prague, August 1997,
Soft Computing 2, p. 18-22.
[188] MUNDICI, D. (1999b) Ulam game, the logic of Maxsat, and many-
valued partitions. In: Fuzzy Sets, Logics, and Reasoning ab out
Knowledge, (Dubois, D. , Prade, H., Klement, E.P., Eds.), Kluwer,
Dordrecht, 1999. To appear.
[201] PAVELKA, J. (1979) On fuzzy logic, I, 11, 111. Zeitschrift für math.
Logik und Grundlagen der Mathematik, 25, p. 45-52, 119-134, 447-464.
[238J VASYUKOV, V.L. (1993) The completeness of the factor semantics for
Lukasiewicz's infinite-valued logics. Studia Logica, 52, p. 143-167.
225
226 INDEX
Belluce, L. P., 156, 202 Di Nola, A., 156, 178, 193, 195, 199, 201,
Berlekamp, E. R., 109 202
Birhkoff, G., 19, 137, 143 Direct product, 19
Blok, W. J., 202 Directly indecomposable MV-algebra, 123
Boolean Disjunctive normal form, 184
element, 25 Distance function, 15
product, 125 Distributive lattice, 24
space, 120 Divisible o-group, 194
Bosbach, B., 30, 138, 202 DNF reduction, 184
Brick, 30 Dvurefenskij, A., 199
Buff, H. W., 30
Butnariu, D., 30, 199 Elliott, G. A., 201
c1assification theory, 194
C*-algebra, 191 partial addition, 192
Cancellation, 37 Embedding, 13
Cantor set, 120 Enveloping group, 40
Casari, E., 202 Epstein, G., 198
Cauchy, A., 56 Equation, 22
Chang, C. C., 20, 29, 44, 49, 76, 82, 101, Equivalence
156, 198, 202 logical, 96
of projections, 192
l-group, 42
Ciabattoni, A., 201 Equivalent
formulas, 80
Classical propositional calculus, 78
good sequences, 40
Clopen set, 119
Error-correcting code, 109
CN-algebra, 30
Escalada Imaz, G., 201
Communication with feedback, 109
Esteva, F., 199
Complement, 25
Complemented element, 25
Fan, 185
Complete
nonsingular, 185
lattice, 129
simplicial, 185
MV-algebra, 129
Farey
Completely distributive
partition, 56
lattice, 134 sequence, 56
MV-algebra, 134 Farkas lemma, 49
Completion of an MV-algebra, 138 Fermüller, C., 201, 202
Cone, simplicial, 185 Ferreirim, I. M. A., 202
Congruence relation, 15 Filipoiu, A., 202
Conjunction of states in Ulam game, 106 Filter
Connective implicative, 86
bi-implication, 79 in a lattice, 114
implication, 78 Font, J. M., 30, 101
negation, 78 Formula
product, 199 2-CNF, 201
Consequence Horn, 201
semantic, 80 equivalent, 80
syntactic, 88 provable, 88
Coproduct, 202 satisfied by a valuation, 80
Covers, 174 Fourier-Motzkin elimination, 201
Frayne's embedding theorem, 194
Danilov's decomposition theorem, 186 Free
De Morgan algebra, 137 MV-algebra, 53
Decidability of word problem, 95 MVn -algebra, 173
Decision problem, 77 product, 202
Deduction theorem, 98
Desingularization of a toric variety, 185 Gaitan, H., 202
INDEX 227
underlying, 31 Rational
Order unit, 32 subdivision, 181
vertex, 63
Panti, G., 49, 178, 199, 199, 201, 202 Rescher, N., 198
Partially ordered abelian group, 31 Residuation in Elliott's addition, 192
Parsing sequence, 21 Resolution, 201
Partition in an MV-algebra, 200 Ri~an, B., 199, 200
Pasquetto, M., 49 Rodr(guez, A. J., 30, 76, 101, 137
Patzig, G., 197 Romanowska, A., 30,202
Paulfk, L., 201 Rose, A., 49, 75, 101, 198, 202
Pavelka, J., 30 Rosenbloom, P. C., 198
Perfect MV-algebra, 152 Rosser, J. B., 49,75, 197, 198
Polyhedron, 63 Routley, R., 197
Poset of prime ideals, 202
Positive Saeli, D., 138
answer, 106 S-algebra, 30
cone, 31 Salzer, G., 201
orientation, 180 Schauder hat, 182
part, 32 multiplicity, 59
Post, E., 198 of a Farey partition, 58
algebra of order n, 198 Schauder set, 185
Precedence laws, 9 Schwartz, D., 202
Priest, G., 197 Search space in Ulam game, 103
Priestley, H. A., 202 Semantic
Prime consequence, 80
ideal of a lattice, 112 equivalence, 81
ideal of an MV-algebra, 13 Semisimple MV-algebra, 72
vector, 185 Separating MV-algebra, 66
Principal Sessa, S., 138, 202
ideal, 13 Sharper state of knowledge, 107
ideal of a lattice, 112 Simple MV-algebra, 70
Probability measure on an MV-algebra, 200 Simplex, 63
Product in infinite-valued logic, 199 unimodular, 180
Projection, 52 Simplicial
function, 19 cone, 185
in a C'"-algebra, 191 fan, 185
Proof in the infinite-valued calculus, 88 Spectral space of prime ideals, 202
Proper Stachniak, Z., 202
equational dass, 157 Star
ideal, 13 in a triangulation, 182
ideal of a lattice, 112 refinement, 186
variety, 157 State
Propositional of knowledge in Ulam game, 106
formula, 79 on an MV-algebra, 200
variable, 79 Stone, M. H., 137
Provable formulas, 88 Stonean ideal in a lattice, 115
= tautologies, 95 String of symbols, 21
Thring enumeration of-, 89 Strong order unit, 32
Pseudocomplement, 133 Subalgebra generated by a subset, 8
Subdirect product, 19
Question in Ulam game, 104 of i-groups, 143
Quotient algebra, 16 Subsystem, 60
Subterm, 21
Radical of an MV-algebra, 72 Sum of good sequences, 36
Ramana Murty, P. V., 30 Surjective homomorphism, 13
Rank of an MV-chain, 158 Symbols of an alphabet, 21
230 INDEX
Vertices of a polyhedron, 63
Vojtas, Po, 201
Yutani, Ho, 29