Cignoli, Mundici-Algebraic Foundations of Many-Valued Reasoning PDF

ALGEBRAIC FOUNDATIONS OF MANY-VALUED REASONING
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
SCOPE OF THE SERIES
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
ITALA M.L. D'OTfAVIANO

Department of Philosophy and The Centre for Logic.
Epistemology and the History of Science.
State University of Campinas. Brazil
and
DANIELE MUNDICI
Department of Computer Science.
University of Milan. Italy
ALGEBRAIC
FOUNDATIONS OF
MANY-VALUED
REASONING
Springer-Science+Business Media, B.V.

A c.I.P. Catalogue record for this book is available from the Library of Congress.
ISBN 978-90-481-5336-7 ISBN 978-94-015-9480-6 (eBook)

DOI 10.1007/978-94-015-9480-6
Printed on acid-free paper
All Rights Reserved

© 2000 Springer Science+Business Media Dordrecht
Originally published by Kluwer Academic Publishers in 2000.
Softcover reprint of the hardcover 1st edition 2000
No part of the material protected by this copyright notice may be reproduced or
utilized in any form or by any means, electronic or mechanical,
incIuding photocopying, recording or by any information storage and
retrieval system, without written permission from the copyright owner.
To the memory 0/
ROLANDO CHUAQUI
ENNIO DE GlORGI
ANTONIO MONTEIRO
great Scientists and Teachers
Contents
Introduction 1
1 Basic not ions 7

1.1 MV-algebras . . . . . . . . . . . . 7
1.2 Homomorphisms and ideals '" 12
1.3 Subdirect representation theorem 19
1.4 MV-equations . . 20
1.5 Boolean algebras . . . . 24
1.6 MV-chains . . . . . . . . 27
1.7 Bibliographical remarks 29
2 Chang completeness theorem 31

2.1 The functor r . . . . . . . . 31
2.2 Good sequences . . . . . . . 34
2.3 The partially ordered monoid M A 37
2.4 Chang's f-group GA' . . . . . 40
2.5 Chang completeness theorem. 43
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
vii
viii CONTENTS
4 Lukasiewicz oo-valued calculus 77

4.1 Many-valued propositional calculi 78
4.2 Wajsberg algebras. . 82
4.3 Provability.......... 87
4.4 Lindenbaum algebra . . . . 92
4.5 All tautologies are provable 94
4.6 Syntactic and semantic consequence . 97
4.7 Bibliographical remarks . . . . . . . · 101
5 Ulam's game 103

5.1 Questions and answers . . . . . . . · 103
5.2 Dynamics of states of knowledge .. .104
5.3 Operations on states of knowledge . · 107
5.4 Bibliographical remarks .. · 109
6 Lattice-theoretical properties 111

6.1 Minimal prime ideals . . . . . . . . . . . . .112
6.2 Stonean ideals and archimedean elements . .115
6.3 Hyperarchimedean algebras . . . .116
6.4 Direct products . . . . . . . . . . · 121
6.5 Boolean products of MV-algebras · 124
6.6 Completeness . . . . . . . . . . · 129
6.7 Atoms and Pseudocomplements · 132
6.8 Complete distributivity . · 134
6.9 Bibliographical remarks · 137
7 MV-algebras and f-groups 139

7.1 Inverting the functor r · 139
7.2 Applications . . . . . · 146
7.3 The radical . . . . . . · 150
7.4 Perfeet MV-algebras . · 151
7.5 Bibliographical remarks · 156
8 Varieties of MV-algebras 157

8.1 Basic definitions. . . . . · 157
8.2 Varieties from simple algebras · 160
8.3 MV-chains of finite rank . . . · 161
CONTENTS ix
8.4 Komori's c1assification . . . . . . . . · 167

8.5 Varieties generated by a finite chain . · 171
8.6 The cardinality of Free~ .173
8.7 Bibliographical remarks . . . . . . . · 177
9 Advanced topics 179

9.1 McNaughton's theorem . . . . . . . . . 180
9.2 Nonsingular fans and normal forms . . 185
9.3 Complexity of the tautology problem . 187
9.4 MV-algebras and AF C*-algebras . 191
9.5 Di Nola's representation theorem . 193
9.6 Bibliographical remarks . . . . . . 194
10 Further Readings 197

10.1 More than two truth values . 197
10.2 Current Research Topics . . . 199
10.2.1 Product . . . . . . . . 199
10.2.2 States, observables , Probability, Partitions . . 200
10.2.3 Deduction . . . . . . . . 201
10.2.4 Further constructions . . . . . . . . . . . .. . 201
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
and transform valid MV-equations into tautologies by writing p ~ q

instead of p = q. The main theorem proved in Chapter 4 then states
that the rule of modus ponens is sufficient to obtain all tautologies
in the infinite-valued calculus of Lukasiewicz starting from four ba-
sic tautologies (originally given by Lukasiewicz) corresponding to the
above defining equations for MV-algebras. These equations are thus
"complete", in the sense that every equation that is valid for [0, 1] is
obtainable from them by substituting equals for equals. As another
corollary, in Chapter 5 we shall show that tautologies in the infinite-
valued calculus of Lukasiewicz coincide with those formulas that are
true in every Ulam game, independently of the number of errors/lies.
Having thus acquired a unified view of valid MV-equations and
infinite-valued tautologies, we can handle logical notions using stan-
dard algebraic methods.
Thus, e.g., from the logic-algorithmic viewpoint, free MV-algebras
over n generators consist of all equivalence classes f of formulas in n
variables. On the other hand, from the algebraic-geometric viewpoint,
free MV-algebras consist of all continuous [0, IJ-valued piecewise linear
functions f with integer coefficients defined over the cube [0, 1Jn. (This
is McNaughton's representation theorem. See Chapter 3 for the case
of one variable, and Chapter 9 for the general case). The multiple
nature of f is useful in the study of normal form reductions, and it
yields a concrete visualization of the not ion of consequence and its fine
structure. Further , one can strengthen the completeness theorem and
give an algorithm to decide whether a formula is a tautology-having no
greater complexity than its analog for boolean tautologies (see Chapter
9).
Generalizing the relationship between the interval [0, 1] and the nat-
urally ordered additive group of real numbers, every MV-algebra A can
be realized as the unit interval A = [0, u] = r(G, u) of a unique abelian
lattice-ordered group G with a strong unit u equipped with negation
u - x and truncated addition x EB y = (x + y) /\ u. Specifically, as
proved in Chapter 7, r is a categorical equivalence between abelian
lattice-ordered groups with strong unit, and MV-algebras. Among the
many important consequences of this equivalence, one can unambigu-
ously say, e.g., that elements al, ... , an E A "sum up to one", or that
they are " linearly independent". These two conditions allow one to
4 INTRODUCTION
give an MV-algebraic definition of "partition of unit", thus generalizing

the basic notion of boolean partition.
Since abelian lattice-ordered groups are so weH established-their
roots going back to the time-honored theory of magnitudes-one might
wonder, why MV-algebras should be given special attention. One main
reason is that, while a strong unit u in a group G of magnitudes is no
less important than the zero element, the property of u being a strong
unit in Gis formalized by
for aH x E G there is n = 0, 1,2, ... such that x::; nu.
This archimedean-like property is not only beyond the expressive power

of equations, but, by Gödel's incompleteness theorem, is undefinable
even in first-order logic.
Remarkably enough, up to categorical equivalence, lattice-ordered
abelian groups with a strong unit can be defined by equations-the
equations of MV-algebras. Since these equations are not hing more than
a reformulation of tautologies in the infinite-valued calculus, and since
the infinite-valued tautology problem is no more complicated than its
boolean counterpart, it becomes natural to apply to other mathemat-
ical areas logic-algorithmic not ions originating from the many-valued
calculus. For this purpose, one may use lattice-ordered abelian groups
with strong unit, via the r functor, as a bridge between MV-algebras
and other structures. Important examples are given by approximately
finite-dimensional (AF) O'-algebras, the algebras of operators currently
used for the mathematical description of infinite spin systems. As an-
other interesting example, disjunctive normal form reductions for for-
mulas in the infinite-valued calculus are essentiaHy the same as desin-
gularization algorithms for toric varieties, once the latter are described
by their associated fans-a fan being a complex of rational polyhedral
cones.
The relationships between MV-algebras, O'-algebras, and toric va-
rieties will be briefly discussed in Chapter 9; most of that chapter can
be safely skipped by readers only interested in the Lukasiewicz calcu-
lus and its algebras. Similarly, we have made no attempt to introduce
here "first-order" infinite-valued logic: as a matter of fact, the devel-
opment of the infinite-valued counterparts of such notions as "set",
INTRODUCTION 5
"equality", "structure" would result in a much less elementary text-

book of considerably larger size-Iet alone the problem of choosing the
right definitions. In a final section, we give appropriate bibliographical
references to the interested reader.
While it would be beyond the scope of the book to cover all the fasci-
nating and rapidly developing fields of research connected with infinite-
valued reasoning, for most of these topics the book is intended to pro-
vide sufficient background material. Thus, Chapter 8 is devoted to
the classification of equational classes of MV-algebras; these include all
classes associated to the finite-valued calculi of Lukasiewicz. The fruit-
ful interplay between MV-algebras and lattices is discussed at length
in Chapter 6. Extensive bibliographical references are given in the fi-
nal chapter on Further Readings, concerning such basic issues as the
"multiplication connective", as weIl as probability and proof theory in
the infinite-valued propositional calculus of Lukasiewicz.
***
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
(CNR), as weH as from "Fundaci6n Antorchas" (Buenos Aires), "Cen-

tre de Recerca Matematica, Institut d'Estudis Catalans" (Barcelona),
Funda<;äo de Amparo a Pesquisa do Estado de Säo Paulo (FAPESP,
Säo Paulo), Istituto Italiano per gli Studi Filosofici (NapIes), besides
our own universities, and COST Action 15. Their support allowed us
to meet several times in Argentina, Brazil and Italy.
R.C., LM.L.D.O, D.M.
Chapter 1
Basic nations
We introduce MV-algebras by means of a small number of simple equa-

tions, in an attempt to capture certain properties of the unit real inter-
val [0,1] equipped with truncated addition x tf) y = min(1, x + y) and
negation 1 - x. We show that every MV-algebra contains a natural
lattice-order. The chapter culminates with Chang's Subdirect Rep-
resentation Theorem, stating that if an equation holds in all totally
ordered MV-algebras, then the equation holds in all MV-algebras.
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:
MV1) x tf) (y tf) z) = (x tf) y) tf) z

MV2) x tf) y = y tf) x
MV3) x tf) 0= x
MV4) -,-,x = x
MVS) x tf) -.0 = -,0
MV6) -,( -,x tf) y) tf) y = -,( -,y tf) x) tf) x.
7
8 CHAPTER 1. BASIC NaTIONS
In particular, axioms MV1)-MV3) state that (A, EB, 0) is an abelian

monoid. Following tradition, we denote an MV-algebra (A, EB, -', 0) by
its universe A.
A singleton {O} is a trivial example of an MV-algebra. An MV-
algebra is said nontrivial Hf its universe has more than one element.
As a first example of a nontrivial MV-algebra, consider the real unit
interval [0,1] = {x E RIO ~ x ~ I}, and for all x,y E [0,1], let
x EB y =def min(l, x + y) and -,x =def 1 - x. It is easy to see that
[0,1] =def ([0,1], EB, -', 0) is an MV-algebra. As a second example,
if (A, V, 1\, -,0,1) is a boolean algebra, then (A, V, -,0) is an MV-
algebra, where V, - and Odenote, respectively, the join, the comple-
ment and the smallest element in A.
A subalgebra of an MV-algebra A is a subset S of A containing the
zero element of A, closed under the operations of A, and equipped with
the restriction to S of these operations.
The intersection of any nonempty family of subalgebras of A is a
subalgebra of A. Given a subset X of A the intersection of all sub alge-
bras of A containing X is a subalgebra of A, and is called the subalgebra
0/ A generated by X.
The rational numbers in [0,1], and, for each integer n ~ 2, the
n-element set
Ln =def {O, l/(n - 1), ... , (n - 2)/(n - 1), I},
yield examples of subalgebras of [0,1].
Given an MV-algebra A and a set X, the set A X of all functions
/: X -+ A becomes an MV-algebra if the operations EB and -, and the
°
element are defined pointwise. The continuous functions from [O,lJ
into [0, 1J form a subalgebra of the MV-algebra [0, l]l°,lJ.
On each MV-algebra A we define the constant 1 and the operations
8 and e as follows:
(1.1) 1 =def -,0,
(1.2) X8Y=def-,(-,xEB-,y),
(1.3) xe Y =def X 8 -'y.

An MV-algebra is nontrivial if and only if °i= 1. The following
identities are immediate consequences of MV 4):
1.1. MV-ALGEBRAS 9
MV7) -,1 = 0, and
MVS) x Ee y = -,(-,x 0 -,y).
Axioms MV5) and MV6) can now be written as:
MV5' ) x Ee1 = 1, and

MV6') (x9y)Eey= (y9x)Eex.
Setting y = -,0 in MV6) we obtain:
MV9) x Ee -,x = 1.
Note that in the MV-algebra [0,1] we have x0y = max(0,x+y-1)

and x 9 y = max(O, x - y).
Notation: Following common usage, we consider the -, operation more
binding than any other operation, and the 0 operation more binding
than Ee and 9.
Lemma 1.1.2 Let A be an MV-algebra and x, y E A. Then the lol-

lowing conditions are equivalent:
(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
is equivalent to MV9), antisymmetry follows from conditions (ii) and

(iii) , and transitivity follows from condition (iv).
An MV-algebra whose natural order is total is called an MV-chain.
Note that, by (iv), the natural order of the MV-chain [0,1] coincides
with the natural order of the real numbers.
Lemma 1.1.3 Let A be an MV-algebra. For each a E A, -,a is the

unique solution x of the simultaneous equations:
(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:
(i) x::; Y if and only if -,y ::; -,x;
(ii) If x::; y then for each z E A, x EB z ::; y EB z and x 0 z ::; y 0 z;
(iii) x0y::; z iJJx::; -'yEBz.
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
Proposition 1.1.5 On each MV-algebra A the natural order deter-

mines a lattice structure. Specijically, the join x V y and the meet x 1\ y
of the elements x and y are given by
(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
Proof: To prove (1.5), by MV6'), MV9) and Lemma 1. 1.4(ii) , x :::;

(x e y) E9 y and y:::; (x e y) E9 y. Suppose x :::; z and y:::; z. By (i)
and (iii) in Lemma 1.1.2, -,x E9 z = 1 and z = (z e y) E9 y. Then by
MV6') we can write
-,((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
Proposition 1.1.6 The following equations hold in every MV-algebra:
(i) x0(yVz)=(x0y)V(x0z),
(ii) x E9 (y!\ z) = (x E9 y) !\ (x E9 z).
Proof: By MV6') and Lemma 1. 1.4(ii) , x 0 y :::; x 0 (y V z) and

x 0 z :::; x 0 (y V z). Suppose x 0 y :::; t and x 0 z :::; t. Then by
Lemma 1.1.4(iii), y:::; -,x E9 t and z :::; -,x ElH, whence y V z :::; -,x tJj t.
One more application of Lemma 1.1.4 (iii) yields (y V z) 0 x :::; t, which
completes the proof of (i). It is now easy to see that (ii) is a consequence
of (i), using Lemma 1.1.4(i), together with MV4) and MV8). 0
Proposition 1.1. 7 Every MV-algebra satisfies the equation
(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)
= (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)
= (y e x) 8 (-,(y e x) E9 x) 8 (-,(-,x E9 (y e x» E9 -,y)
= y 8 -,x 8 (-,(y e x) E9 x) 8 ((x 0 -,(y e x» E9 -,y)
= -,x 8 (x E9 -,(y e x» 8 y 0 (-,y E9 (x 8 (-,y E9 x») =

-,x 8 (x E9 -,(y e x» 8 (x 0 (-,y E9 x» 8 (-,(x 8 (-,y E9 x» E9 y)
= 0, since by MV8) and MV9), -,x 8x = O. 0

Let A be an MV-algebra. For each x E A, we let Ox = 0, and für
each integer n ~ 0, (n + l)x = nx E9 x.
Lemma 1.1.8 Let x and y be elements of an MV-algebra A. If

x 1\ Y = 0 then for each integer n ~ 0, nx 1\ ny = O.
Proof: If x 1\ Y = 0 then by monotonicity (Lemma 1.1.4) and distribu-

tivity (Proposition 1.1.6), x = xE9(xl\y) = (xE9x) 1\ (xEBy) ~ 2xl\y,
whence 0 = x 1\ Y ~ 2x 1\ y. It follows that 0 = 2x 1\ 2y = 4x 1\ 4y =
8x 1\ 8y = ... . The desired conc1usion now follows from nx 1\ ny :s
2n x 1\ 2n y = O. 0
1.2 Homomorphisms and ideals

Let A and B be MV-algebras. A function h: A ---. B is a homomorphism
Hf it satisfies the follüwing conditions, for each x, y E A:
Hl) h(O) = 0,
H2) h(x E9 y) = h(x) E9 h(y),
H3) h(-,x) = -,h(x).

1.2. HOMOMORPHISMS AND IDEALS 13
Following current usage, if h is one-one we shall equivalently say

that h is an injective homomorphism, or an embedding. If the ho-
momorphism h: A - t B is onto B we say that h is surjective. By an
isomorphism we shall mean a surjective one-one homomorphism. We
write A C::! B Hf there is an isomorphism from A onto B.
The kernel of a homomorphism h: A - t B is the set
K er(h) =dej h-1(0) = {x E AI h(x) = O}.
Our next aim is to characterize kerneis of homomorphisms.
An ideal of an MV-algebra A is a subset 1 of A satisfying the fol-
lowing conditions:
11) 0 E 1,
12) If xE 1, y E A and y ~ x then y E 1,
13) If x E 1 and y E 1 then x ffi y EI.
The intersection of any family of ideals of A is still an ideal of A.
For every subset W ~ A, the intersection of all ideals 1 ;2 W is said to
be the ideal generated by W, and will be denoted (W).
The proof of the next lemma is immediate, and will be omitted.
Lemma 1.2.1 Let W be a subset of an MV-algebra A. If W = 0,
then (W) = {O}. If W =1= 0, then
(W) = {x E A I x ~ Wl EB··· EB Wk, fOT some Wl,"" Wk E W}.
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) For each x and y in A, either (x e y) E I or (y e x) E I.
An ideal I of an MV-algebra A is caHed maximal Hf it is proper and

no proper ideal of A strictly contains I, Le., for each ideal J =I- I, if
I ~ J then J = A.
The next proposition generalizes a weH known property of maximal
ideals in boolean algebras.
Proposition 1.2.2 Por any proper ideal J of an MV-algebra A the

following conditions are equivalent:
(i) J is a maximal ideal of A;
(ii) for each x E A, x rt. J iff -mx E J for some integer n 2:: 1.
Proof: (i) ----+ (ii): Suppose that J is a maximal ideal of A. If x rt. J,

then ({x}UJ) = A, and by (1.9), 1 = nxEBa for some integer n 2:: 1
and a E J. Stated otherwise, -,nx::; a E J, whence by 12), -,nx E J.
Conversely, if x E J, then nx E J for each integer n 2:: 1; since J is
proper, -mx rt. J.
(ii) ----+ (i): Let K =I- J be an ideal of A such that J ~ K. For
every x E K \ J we must have -,nx E J for some integer n 2:: 1. Hence
1 = nx EB -,nx E K, and K = A. 0
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.
Lemma 1.2.3 Let A, B be MV-algebras, and h: A ----+ B a homomor-

phism. Then the following properties hold:
(i) Por each ideal J of B, the set h-1(J) =def {x E AI h(x) E J} is

an ideal of A. Thus in particular, K er(h) E I(A);
(ii) h(x) ::; h(y) iff xe y E Ker(h);
(iii) h is injective iff Ker(h) = {O};
(iv) K er (h) =I- A iff B is nontrivial;

(v) K er(h) E P(A) iJ] B is nontrivial and the image h(A), as a

subalgebra of B, is an MV-chain. 0
Definition 1.2.4 The distance function d: A x A ----t A is defined by
(1.10) d(x, y) =def (x e y) EB (y e x).
In the MV-algebra [0,1], d(x, y) = Ix - yl. In every boolean algebra
the distanee funetion eoincides with the symmetrie differenee operation.
Proposition 1.2.5 In every MV-algebra A we have:
(i) d(x,y) = 0 iJ] x = y,
(ii) d(x, y) = d(y, x),
(iii) d(x, z) ~ d(x, y) EB d(y, z),
(iv) d(x, y) = d( -,x, -,y),
(v) d(x EB s, y EB t) ~ d(x, y) EB d(s, t).
Proof: Properties (i), (ii) and (iv) immediately follow by definition,
reealling the basie properties of the natural order on A (Lemmas 1.1.2
and 1.1.4). To prove (iii), note first that -,(x e z) EB (x e y) EB (y e z) =
(-,xV-'Y)EB(zVy) ~ -'yEBy= 1. Henee, (xez) ~ (xey)EB(yez).
In a similar way we obtain (z e x) ~ (y e x) EB (z e y), whenee (iii)
follows from the monotonieity of EB (Lemma 1.1.4(ii)). One similarly
proves (v) by observing that -,((x EB s) e (y EB t)) EB (x e y) EB (s e t) =
-,(x EB s) EB (x Vy) EB (t V s) ~ -,(x EB s) EB x EB s = 1. 0
As an immediate consequenee we have
Proposition 1.2.6 Let I be an ideal of an MV-algebra A. Then the
binary relation =1 on A defined by x =1 Y iJ] d(x, y) E I is a
eongruenee relation. (Stated otherwise, =1 is an equivalence relation
such that x =1 sand y =1 t imply -,x ==1 -,s and x EB y =1 S EB t.)
Moreover, I = {x E A I X =1 O}.
Conversely, if =
is a congruence on A, then {x E A I x == O} is an
= =
ideal, and x y iJ] d(x, y) O. Therefore, the correspondence I ~ =1
is a bijection from the set of ideals of A onto the set of congruences on
A. 0
Given x E A, the equivalence class of x with respect to = [ will

be denoted by xl1 and the quotient set AI = [ by A11. Since = [ is a
congruence, defining on the set AI1 the operations
(1.11) ,(xl1) =def ,xl1
and
(1.12) xI1$YI1=def (x$y)11,
the system (AI1, $",011) becomes an MV-algebra, called the quotient
algebra of A by the ideal 1. Moreover, the correspondence x t---t xl1
defines a homomorphism h[ from A onto the quotient algebra A11,
which is called the natural homomorphism from A onto A11. Note that
Ker(h[) = 1.
The next lemma is an easy consequence of Lemma 1.2.3(ii).
Lemma 1.2.7 1f A, Band C are MV-algebras, and f: A - Band
g:A - C are surjective homomorphisms, then Ker(f) ~ Ker(g) if
and only if there is a surjective homomorphism h: B - C such that
hof = g, i.e., h(f(x)) = g(x) for alt xE A. This homomorphism h is
an isomorphism if and only if Ker(f) = Ker(g). 0
Upon noting that Ker(h) = Ker(hKer(h») we immediately get
Theorem 1.2.8 Let A and B be MV-algebras. If h: A - B is a sur-
jective homomorphism, then there is an isomorphism f: AI K er(h) - B
such that f(xIKer(h)) = h(x) for alt xE A. 0
Proposition 1.2.9 1f A is an MV-chain, then alt proper ideals of A
are prime.
Proof: Let 1 be a proper ideal of A. Since h[: A - AI1 is a surjec-
tive homomorphism, then AI1 is also an MV-chain, and hence, by
Lemma 1.2.3(v), 1 must be a prime ideal. 0
Proposition 1.2.10 Let J be an ideal of an MV-algebra A. Then
the map 1 t---t hJ(I) determines an inclusion preserving one-one corre-
spondence between the ideals of A containing J and the ideals of the
quotient MV-algebra AI J. The inverse map also preserves inclusions,
and is obtained by taking the inverse image h;l(K) of each ideal K of
AIJ·
Proo/: Let I be an ideal of A such that J ~ I. Since h J maps A

onto AIJ and Ker(h J ) = J ~ I, by Lemma 1.2.3 (ii) and (MV6'),
we have hJ(I) E I(AI J) and, moreover, h J1 (h J (I)) ~ I. Since the
converse inclusion holds for all surjective mappings, one has that I =
h J1 (h J (I)). On the other hand, by Lemma 1.2.3 (i), h J1 (K) E I(A)
for each K E I(AI J). To complete the proof it is sufficient to note
that J = hJ1 ( {O}) ~ hJ1 (K) and hJ (h J1 (K)) = K. 0
Remark: If A is an MV-chain, then the set I(A) is totally ordered by

inclusion. Indeed, if I, J were ideals of A such that I ~ J and J ~ I,
then there would be elements a, b in A such that a E 1\ J and b E J \ I,
whence a i band b i a, which is impossible.
Theorem 1.2.11 The /ollowing properties hold in any MV-algebra A:
(i) Every proper ideal 0/ A that contains a prime ideal is prime;
(ii) For each prime ideal J 0/ A, the set {I E I(A) I J ~ I} is totally

ordered by inclusion.
Prao!" Let J be a prime ideal of A. By Lemma 1.2.3(v), AI J is an MV-

chain, and by Proposition 1.2.9 and the above remark, all proper ideals
of AI J are prime and are totally ordered by inclusion. This, together
with Proposition 1.2.10, implies (ii). To prove (i), note that if I is a
proper ideal of A such that J ~ I, then again by Proposition 1.2.10,
1= h J1 (h J (I)), whence I is the inverse image of a prime ideal of AI J.
o
Corollary 1.2.12 Every prime ideal J 0/ an MV-algebra A is con-

tained in a unique maximal ideal 0/ A.
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
The next proposition will play an important role in the proof of

Chang's Subdirect Representation Theorem.
Proposition 1.2.13 Let A be an MV-algebra, J an ideal 0/ A, and

a E A \ J. Then there is a prime ideal P 0/ A such that J ~ P and
a (j. P.
Proo/: A routine application of Zorn's Lemma shows that there is an

ideal I of A which is maximal with respect to the property that
J ~ land a (j. I. We shall show that I is a prime ideal. Let x and
y be elements of A, and suppose that both x 8 y (j. land y 8 x (j. I
(absurdum hypothesis). Then the ideal generated by land x8y must
contain the element a. By (1.9), a:::; sEIl p(x 8 y) for some sEI and
some integer p ~ 1. Similarly, there is an element tEl and an integer
q ~ 1 such that a:::; tEIl q(y 8 x). Let u = sEIlt and n = max(p, q).
Then u E I, a:::; u EIl n(x 8 y) and a:::; u EIl n(y 8 x). Hence by
(1.6) and (1.7), together with Proposition 1.1.6(ii) and Lemma 1.1.8,
we have a:::; (uEIln(x8Y)) 1\ (uEIln(Y8x)) = uEIl(n(x8Y) 1\ n(Y8x))
= u, whence a EI, a contradiction. 0
Corollary 1.2.14 Every proper ideal 0/ an MV-algebra is an intersec-
tion 0/ prime ideals. 0
From the above proposition and Corollary 1.2.12 we immediately

obtain:
Corollary 1.2.15 Every nontrivial MV-algebra has a maximal ideal.

o
In the next proposition we generalize so me properties of maximal

ideals of boolean algebras. Remarkably enough, (the analogues of)
these properties do not hold for other extensions of boolean algebras,
such as bounded distributive lattices or commutative rings with unit.
Proposition 1.2.16 Let A and B be MV-algebras, and M be a maxi-

mal ideal 0/ B. Then we have
(i) For any homomorphism h: A - B, the inverse image h- 1 (M) is a

maximal ideal 0/ A;
(ii) Por any subalgebra S 0/ B, Sn M is a maximal ideal 0/ S.
1.3. SUBDlRECT REPRESENTATION THEOREM 19
Proof" (i) By Lemma 1.2.3, h- 1 (M) is an ideal of A; since h(l) = 1 t/.

M, then h- 1 (M) must be a proper ideal. Suppose z t/. h- 1 (M). Since
h(z) t/. M, then by Proposition 1.2.2 there is an integer n ~ 1 such
that -,nh(z) E M. It follows that -,nz E h- 1 (M), whence, again by
Proposition 1.2.2, h- 1 (M) is a maximal ideal of A.
The proof of (ii) easily follows from (i), upon letting L: S -- B be
the natural embedding, given by L(X) = x for all x E S, and noting
that Mn B = L- 1 (M). 0
1.3 Subdirect representation theorem

Throughout this section I shall denote a nonempty set. The direct
product of a family {AdiEI of MV-algebras, denoted by TIiEI Ai, is the
MV-algebra obtained by endowing the set-theoretical cartesian product
ofthe family with the MV-operations defined pointwise. In other words,
TIiEI Ai is the set of all functions f: 1-- UiEI Ai such that f(i) E Ai
for all i E I, with the operations -, and EI7 defined by
(-,J)(i) =def -,f(i) and (f EI7 g)(i) =def f(i) EI7 g(i).
The zero element of TI iE1 Ai is the function i E I t---+ Oi E Ai. For each
j E I, the map 7rj: TIiEI Ai -- A j is defined by
7rj(f) =deff(j)·
Each 7rj is a homomorphism onto A j , called the lh projection func-
tion. In particular, for each MV-algebra A and nonempty set X, the
MV-algebra A X is the direct product of the family {AX}XEX, where
A x = A for all x EX.
Definition 1.3.1 An MV-algebra A is a subdirect product of a fam-
ily {AihEl of MV-algebras iff there exists a one-one homomorphism
h: A -- TIiEI Ai such that for each j EI, the composite map 7rj 0 his a
homomorphism onto A j •
If Ais a subdirect product of the family {AihEl, then A is isomor-
phie to the subalgebra h(A) of TIiEI Ai; moreover, the restriction to
h(A) of each projection is a surjective mapping.
The following result is a particular case of a theorem of Universal
Algebra, due to Birkhoff:
Theorem 1.3.2 An MV-algebra A is a subdirect product of a family

{AihEI of MV-algebras if and only if there is a family {JihEI of ideals
of A such that
(i) Ai ~ AI Ji for each i E I
and
(ii) n iEI Ji = {O}.

Prool Supposing first that A is a subdirect product of a family {Ai hEl
of MV-algebras, let h: A ---.. niEI Ai be a one-one homomorphism as
given by Definition 1.3.1; for each j E I, let Jj = K er(7rj 0 h). By
Theorem 1.2.8, A j ~ AI Jj . If x E niEI J i , then 7rj(h(x)) = 0 for all
j E I. This implies that h(x) = 0, and since h is injective, x = O.
Therefore niEI Ji = {O}, and conditions (i) and (ii) hold true.
Conversely, suppose {JihEI to be a family of ideals of A satisfying
conditions (i) and (ii). Let Ei be an isomorphism of AI J i onto Ai, as
given by condition (i). Let the function h: A ---.. niEI Ai be defined by
stipulating that, for each x E A, (h(x))(i) = Ei(xl J i ). It follows from
(ii) that Ker(h) = {O}, whence, by Lemma 1.2.3(iii), h is injective.
Since for each i E I the map a E A 1--+ al Ji E AI Ji is surjective, then
7ri 0 h maps A onto Ai' Thus, A is a subdirect product of the family
{AihEI, as required. 0
The following result is fundamental:
Theorem 1.3.3 (Chang's Subdirect Representation Theorem) Every

nontrivial MV-algebra is a subdirect product of MV-chains.
Proof: By Theorem 1.3.2 and Lemma 1.2.3(v), an MV-algebra A is a

subdirect product of a family of MV-chains if and only if there is a
family {~hEI of prime ideals of A such that niEI ~ = {O}. Nowapply
Corollary 1.2.14 to the ideal {O}. 0
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
all MV-algebras it is sufficient to check that the equation holds in all

MV-chains. To give a precise formulation to this result we shall now
develop the necessary syntactic machinery.
Definition 1.4.1 By astring (or, word) over a nonempty set S we

understand a finite list of elements of S. The latter are often called the
symbols of alphabet S.
For each natural number t ~ 1, let St =def {O", EB, Xl, ... , Xt, (, n.
An MV-term in the variables Xl, ... , Xt is astring over St arising from
a finite number of applications of the following rules:
°
(Tl) The elements and Xi, for i = 1, ... , t, considered as one-element
strings, are MV-terms.
(T2) If the string T is an MV-term, then so is 'T.
(T3) If the strings T and (J' are MV-terms, then so is (T EB o)
In other words, astring T over St is an MV-term if and only if there

is a formation (or, parsing) sequence for T, Le., a finite list of strings
over St, say Tl, ... , Tn , such that T n = T and for each i E {1, ... , n}, Ti
satisfies at least one of the following conditions:
(i) Ti = ° or Ti = Xj, for some 1 :::; j :::; t,

(ii) there is j < i such that Ti = 'Tj,
(iii) there are j < i and k < i such that Ti = (Tj EB Tk).
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
We shall henceforth write T(X}, ..• , X n ) to signify that T is an MV-

term in the variables Xl, ... , X n .
The following strings are examples of MV-terms in the variables
Xl, X2, X3:
Notation: Following tradition, for the sake of readability, in writing

terms we shall omit the outermost pair of brackets. Further , we shall
freely use the symbols 0, 8, V, /\ and 1 to write MV-terms in abbre-
viated form, in the light of (1.1)-(1.6). Thus, given MV-terms T and (T
in the variables Xl, ... , Xt, we shall freely call MV-terms (in the same
variables) also such abbreviations as T 0 (T, T 8 (T, T V (T, T /\ (T, and the
like.
Definition 1.4.3 Let A be an MV-algebra, T an MV-term in the vari-

ables Xl, ... ,Xt, and assurne al, ... , at are elements of A. Substitut-
ing an element ai E A for all occurrences of the variable Xi in T, for
i = 1, ... ,t, using the unique readability theorem, and interpreting the
symbols 0, EB and -, as the corresponding operations in A, we obtain
an element of A, denoted TA(al' . .. , at).
In more detail, proceeding by induction on the number of operation
symbols occurring in T, we define TA(al' .. . , at) as follows:
(i) xf = ai, for each i = 1, ... , t;
(ii) (-,(T)A = -,( (TA);
(iii) (0" EB p)A = (O"A EB pA).
By the above definition, given any MV-algebra A we can associate

each MV-term T in the variables Xl, ... ,Xn with a function TA: An -+ A.
Functions arising in this way are called term functions on A. The
dependence of TA on n is tacitly understood.
Definition 1.4.4 An MV-equation (for short, an equation) in the vari-

ables Xl, ... , Xt is a pair (T, (T) of MV-terms in the variables Xl,· .. ,Xt·
1.4. MV-EQUATIONS 23
Following tradition, we shall write r = a instead of (r, a). An MV-

algebra A satisfies the MV-equation r = a, in symbols,
A Fr = a,
Axioms (MV! )-(MV6) are examples of MV-equations in the vari-

ables x, y, z. By definition, these equations are satisfied by all MV-
algebras. In the previous sections many other equations have been
shown to hold for all MV-algebras. By contrast, the equation xEBxEBx =
x EB x is satisfied by the MV-algebra Ln if and only if n = 2 or n = 3.
Remark: By Proposition 1.2.5(i), an MV-equation r = a holds in an
MV-algebra A if and only if the equation (r e a) EB (a e r) = holds °
in A. Therefore we can safely assume that all MV-equations are of the
form p = 0, where p is an MV-term.
Lemma 1.4.5 Let A, B, Ai (for all i EI) be MV-algebras:
(i) If A Fr = athen S Fr = a for each subalgebra S of A;

(ii) If h: A - t B is a homomorphism, then for each MV-term r in the
variables Xl," ., X s and each s-tuple (al,"" as) of elements of A
we have rB(h(al),"" h(a s )) = h(TA(al"'" as)). In particular,
when h maps A onto B, from A F T = a it follows that B ~
T = a;
(iii) If Ai FT= a for each i E I, then I1EI Ai F T = a.

Proof" Conditions (i) and (ii) are immediate. As for condition (iii), let
JI, . .. , fs E A = niEI Ai' By hypothesis, for each i E I we can write
TA(fI"'" fs)(i) = TAi(JI(i), ... , fs(i)) =
a Ai (fl (i), ... , fs(i)) = aA(fI"'" fs)(i) ,
whence TA(JI, . .. , fs) = aA(JI, . .. ,fs). 0
Theorem 1.4.6 Let A be the subdirect product of a family {AihEl of

MV-algebras; let r = a be an MV-equation. Then A F T = a if and
only if Ai F T = a for each i EI.
Proof" Let h: A ~ [liEf Ai be a one-one homomorphism as given by

Definition 1.3.1. Suppose that A F r = (J'. Since for each i E I, 7ri 0 h
maps A onto Ai, it follows that Ai Fr = (J.
Conversely, suppose that Ai F r = (J' for all i EI. By the above
lemma, [liEf Ai F r = (J', and since h(A) is a sub algebra of [liEf Ai,
h(A) F r = (J'. Since h- 1 maps h(A) onto A, we conclude that A F
r = (J'. 0
Corollary 1.4.7 An MV-equation is satisfied by all MV-algebras if and

only if it is satisfied by all MV-chains.
Proof" Suppose that r = (J' is satisfied by all MV-chains, and let A

be an MV-algebra. If A = {O} then, trivially, rA(O, ... , 0) = 0
= (J'A(O, ... , 0), whence A F r = (J'. If A is nontrivial the desired
conclusion follows from Theorems 1.3.3 and 1.4.6. 0
Corollary 1.4.7 will be considerably strengthened in the next chap-
ter, by proving that an equation holds in all MV-algebras if and only
if it holds in the algebra [0, 1].
1.5 Boolean alge bras

We have already noted that boolean algebras are particular cases of
MV-algebras. In this section we shall characterize boolean algebras
among MV-algebras.
As shown in Section 1, the natural order makes every MV-algebra
A into a lattice with minimum element 0 and maximum 1. We shall
denote this lattice by
L(A).
Recall that the lattice operations of join and meet are definable via the
MV-operations by formulas (1.5) and (1.6).
A lattice is called distributive Hf the following distributive laws hold:
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
When A is an MV-chain and a, bE A, then a V b = max(a, b) and

a /\ b = min(a, b), whence clearly the distributive laws hold in A.
Using (1.5) and (1.6), the above two distributive laws can be equiva-
lently reformulated as MV-equations; since these equations are satisfied
byall MV-chains, by Corollary 1.4.7 we obtain
Proposition 1.5.1 For any MV-algebra A, L(A) is a distributive lat-

tice with smallest element 0 and greatest element 1. 0
Definition 1.5.2 An element x of a lattice L with 0 and 1 is said to

be complemented iff there is an element y E L (the complement of x)
such that x V y = 1 and x /\ y = O. When L is distributive each z E L
has at most one complement, denoted -z. We further let
B(L)
be the set of all complemented elements of the distributive lattice L.

Note that 0 and 1 are elements of B(L), because -0 = 1 and -1 = O.
As a matter of fact, B(L) is a sublattice of L which is also a boolean
algebra. For any MV algebra A we shall write B(A) as an abbreviation
of B(L(A)). Elements of B(A) are called the boolean elements of A.
Theorem 1.5.3 For every element x in an MV algebra A the following

conditions are equivalent:
(i) xE B(A);
(ii) x V -,x = 1;
(iii) x /\ -,x = 0;
(iv) x E9 x = x;
(v) x0x=x;
(vi) x E9 y = x V y, for all y E A;

(vii) x 0 y = x /\ y, for all Y E A.
Proof' The following equivalences are trivial: (ii){:}(iii), (iv){:} (v) ,

(vi){:}(vii). It is also trivial that (vi) => (iv). Further, the equivalent
conditions (ii) and (iii) state that ....,x is the complement of x. Thus, in
particular (iii)=> (i).
(i)=> (ii): By elementary manipulations, using Lemma 1.1.2 and
Proposition 1.1.6 we have ....,x = ....,x EB 0 = ....,x EB (x 1\ -x) = (....,x EB x) 1\
(....,x EB -x) = ....,x EB -x. Thus, -x ~ ....,x and 1 = x V -x ~ x V....,x ~ 1,
and we are done.
(iii)=>(vi): Using Proposition 1.2.5, together with the Subdirect
Representation Theorem 1.3.3 and the inequality xVy ~ xEBy, (which
also is an immediate consequence of Theorem 1.3.3) we have d(x EB
y,xVy) = (xEBY)0""'(xVy) = (xEBy)0(-,xl\-,y) ~ ((xEBY)0-'x)
I\((x EB y) 0 ....,y) = -,x 1\ Y 1\ ....,y 1\ x. Therefore, x 1\ -,x = 0 implies
d( x EB y, x V y) = 0, whence x EB y = x V y.
(iv)=>(ii): By hypothesis, 1 = ....,x EB x = ....,(x EB x) EB x = -,x V x. 0
Corollary 1.5.4 B(A) is a subalgebra of the MV algebra A. A subal-

gebra B of A is a boolean algebra iff B ~ B(A). 0
Corollary 1.5.5 An MV-algebra A is a boolean algebra if and only if

the operation EB is idempotent, i.e., the equation x EB x = x is satisfied
byA. 0
As the reader will recall, for every element z in an MV algebra A,

we denote by
(1.13) (z) =def {x E A I x ~ z EB ... EB z, (n times) far some n > O}
the ideal generated by z.
Corollary 1.5.6 For any MV algebra A and z E A, z is a boolean

element of A iff the set {x E A I x ~ z} is an MV ideal iff {x E A I
x::;z} = (z).
Proof' Immediate from Theorem 1.5.3(iv). 0

1.6. MV-GRAINS 27
1.6 MV-chains
In this section we collect several results on totally ordered MV-algebras,
to be used in the next chapter.
Lemma 1.6.1 The /ollowing properties hold in every MV-chain A:
(i) 1/ xtBy < 1 then x0y = 0;

(ii) 1/ x EB y = x EB z and x 0 y = x 0 z then y = z;
(iii) 1/ xEBy=xEBz<l then y=z;
(iv) 1/ x0y=x0z>0 then y=z;
(v) x EB y = x iff x = 1 or y = 0;
(vi) x EB Y = x iff ·x EB'y = .y;
(vii) 1/ x EB y = 1 and x EB z < 1 then (x 0 y) EB z = (x EB z) 0 y.
Proof: (i) By hypothesis, .x 'i y, whence y < .x. (ii) By hy-
pothesis, max(.x,y) = .xEB(y0x) = .xEB(z0x) = max(-,x,z).
Similarly, min( .x, y) = min( .x, z), whence y = z. Condition (iii) is
an immediate consequence of (i) and (ii). Condition (iv) follows from
(iii) by Lemma 1.1.4(i). Condition (v) follows from (iii). Prom (v) one
immediately obtains (vi). Finally, to prove (vii), since by assumption
.y :::; x, we get .y EB (x 0 y) EB z = (.y V x) EB z = x EB z < 1 and
.y EB (y 0 (x EB z)) = .y V (x EB z) = x EB z, whence (vii) follows from
(iii). 0
Remark: Prom Theorem 1.3.3 it follows that every MV-algebra satisfies
conditions (ii) and (vi).
Proposition 1.6.2 The /ollowing equations hold in every MV-algebra

A:
(1.14) xEByEB(x0y)=xEBy
(1.15) (x e y) EB ((x EB .y) 0 y) = x
(1.16) (x 0 y) EB ((x EB y) 0 z) = (x 0 z) EB ((x EB z) 0 y).
28 GHAPTER 1. BASIG NaTIONS
Proof: By Theorem 1.3.3 we can safely assume that A is a chain. If

x $ y = 1, then (1.14) follows by MV5'). If x $ y < 1, then (1.14)
follows from Lemma 1.6.1(i). To prove (1.15), note that if x:S y then
xe y = 0 and x = x A Y = (x $ -.y) 0 y; if, on the other hand, y < x
then (x e y) $ (x A y) = (x e y) $ y = x V y = x.
As aprerequisite for the proof of (1.16) we shall prove the following
equation:
(1.17) (x 0 y) $ ((x $ y) 0 z) = (x $ y) 0 ((x 0 y) $ z).
Indeed, if x$y = 1 both members of (1.17) coincide with (X0Y)$z.

If x $ y < 1 then by Lemma 1.6.1(i) both members coincide with
(x $ y) 0 z. Thus (1.17) holds for all MV-algebras.
Prom MV4) and MV8) we now obtain:
(1.18) -.((x 0 y) $ ((x $ y) 0 z)) = (.x 0 .y) $ ((.x $ .y) 0 .z).
To complete the proof of (1.16) we argue by cases as follows:

Gase 1: x $ y $ z < 1.
Then since A is a chain, by Lemma 1.6.1(i), both members of (1.16)
are equal to O.
Gase 2: .x $ .y $ .z < 1.
Same as Case 1, recalling (1.18).
There remains to consider
Gase 3: x $ y $ z = 1 and -.x $ -'y $ -.z = 1.
Subcase 3.1: x $ y = 1 and x $ z < 1, or x $ y < 1 and x $ z = 1.
Then by symmetry, it is sufficient to consider the case x$y = 1 and
x$z< 1. Then x0z = 0, and (1.16) becomes (x0Y)$z = (x$z)0y,
which follows from Lemma 1.6.1(vii).
Subcase3.2: x$y=x$z=1.
Then equation (1.16) becomes
(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
(-,y 0 z) EEl y = (x 0 z) EEl y. By symmetry, (1.19) holds under the

hypothesis x 0 z = O. We next observe that if one of the members
of (1.19) is equal to 1 then so is the other. Assume, for instance,
(x 0 y) EEl z = 1. Since -.x EEl-.y EEl-.z = 1 is equivalent to x 0 Y 0 z = 0,
it follows from Lemma 1.1.3 that z = -.(x 0 y) = -.x EEl -.y. Hence, by
Proposition 1.1.6, (x 0 z) EEl y = (x 0 (-,x EEl -.y)) EEl y = (x /\ -,y) EEl y =
(x EEl y) /\ (-,y EEl y) = 1. Thus, to complete the analysis of Subcase 3.2
we may restriet to the case when (x 0 y) EEl z < 1, (x 0 z) EEl y < 1,
x 0 y > 0, x 0 z > O. Under these hypotheses, by Lemma 1.6.1(vii)
we obtain:
x0 (zEEl (x0Y)) = (x0z) EEl(x0y) > 0,

x 0 (y EEl (x 0 z)) = (x 0 y) EEl (x 0 z),
thus establishing (1.19) by Lemma 1.6.1(iv).
Subcase 3.3: x EEl y < 1 and x EEl z < 1.
Then by Lemma 1.6.1(i), x0y = 0 and x0z = 0, i.e., -,xffi-,y = 1
and -,x$-.z = 1. Recalling (1.18) and arguing as in Subcase 3.2 (with
-,x, -'y, -,z instead of x, y, z) we conclude that (1.16) also holds in this
case. 0
1.7 Bibliographical remarks

MV-algebras were originally introduced by Chang in [36] to prove the
completeness theorem for the Lukasiewicz calculus. In the same paper
Chang proved the basic facts about the natural order, congruences and
boolean elements. Prime ideals were introduced by hirn in [38] to prove
Theorem 1.3.3. For historical information on MV-algebras see [40]. A
version of Proposition 1.2.2 is in [86]. Equation (1.16) was proved in
[51].
It is easy to see that the present definition of MV-algebras (which
is essentially due to Mangani [144]) is equivalent to Chang's original
definition. To this purpose one simply not es that for every MV-algebra
(A, $, -', 0) the operations V and /\ defined by (1.5) and (1.6) are
the join and the meet with respect to the natural order of A; thus in
particular, both operations are commutative and associative.
(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
Chang completeness theorem
In this chapter we shall prove Chang's completeness theorem stating

that if an equation holds in the unit real interval [0, 1], then the equation
holds in every MV-algebra. Thus, intuitively, the two element structure
{O, 1} stands to boolean algebras as the interval [0,1] stands to MV-
algebras. Our proof is elementary, and makes use of tools (such as "good
sequences") that shall also find applications in a subsequent chapter to
show the equivalence between MV-algebras and lattice-ordered abelian
groups with streng unit.
2.1 The functor r

A partially ordered abelian group is an abelian group (G, +, -,0) en-
dowed with a partial order relation::; that is compatible with addition;
in other words, ::; has the following translation invariance property, for
all x, y, t E G:
(2.1) If x::;y then t+x::;t+y.
The positive cone G+ of G is the set of all x E G such that x. When
the order relation is total, (Le., when G = G+ U -G+), G is said to be
°: ;
a totally ordered abelian group, or o-group for short. When the order of
G defines a lattice structure, G is called a lattice-ordered abelian group,
or f-group, for short. In any f-group we have
(2.2) t+(xVy)=(t+x)V(t+y)
31
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;
(2.5) x- =d,eJ 0 V -x;
(2.6) lxi =d,eJ x+ + x- = x V -x.
A strong (order) unit u of G is an archimedean element of G, Le., an

element 0 ~ u E G such that for each x E G there is an integer n ~ 0
with lxi ~ nu.
Definition 2.1.1 Let G be an f-group. For any element u E G, u > 0,

(not necessarily u being a strong unit of G) we let
[0, u] =d,eJ {x E G I 0 ~ x ~ u},

and for each x, y E [0, u],
x E9 y =d,eJ u A (x + y), and -,x =d,eJ u - x.
The structure ([0, u], E9, -,,0) is denoted r(G, u).
Proposition 2.1.2 r(G, u) is an MV-algebra.
Proof: We shalllimit ourselves to verifying that r(G, u) satisfies MV6).

For all x, y E [0, u] we have
-,( -,x E9 y) E9 y
= y E9 -,(y E9 -,x)
= u A (y + (u - (u A (y + u - x))))
2.1. THE FUNCTOR r 33
= u A (y + u + (-u V (-y - u + x)))

= u A «y + u - u) V (y +u - Y- u + x))
=uA(yVx)
= y V x = x Vy.
This shows that x and y are interchangeable. 0
Lemma 2.1.3 Let G be an f-group with strong unit u. Let A -

r(G,u).
(i) Forall a,bEA,a+b=(aEDb)+(a0b);
(ii) For all Xl, ... , X n E A, Xl ED ... ED X n = U A (Xl + ... + X n );
(iii) The natural order 01 the MV-algebra A coincides with the order
01 [0, u] inherited from G by restrietion.
Proof" (i) We easily obtain a + b - (a 0 b) = a + b - (0 V (a + b - u» =

(a+b) Au = aEDb. An easy induction on n proves (H). (iii) is proved by
direct verification, using 1.1.5, together with the above proof of 2.1.2.
o
Notation: Following common usage, we let R, Q, Z denote the additive
groups of reals, rationals, integers, with the natural order. In the par-
ticular case when G = R, r(R, 1) coincides with the MV-algebra [0,1].
We also have Q n [0,1] = r(Q, 1). Further, for each integer n 2: 2,
1 1 z
Ln = r(Z n _ 1,1), where Z n _ 1 = {n _ 1 I Z E Z}.
In particular, the boolean algebra {O, 1} = L 2 is the same as r(Z, 1).

When dealing with elements X E r(G,u), the notation nx may
ambiguously represent both X + ... + X (n times), and X ED ... ED X (n
times). To avoid any danger of confusion, we will adopt the following
notation:
(2.7) X ED··· ED X = u A (x
n.x =def "-----v---" + ... + x), = u A nx.
n times n ti~es
Let G be an f-group and 0< u E G. Let S = {x E GI for so me O:S: n E

Z, lxi :s: nu}. Then S is a subgroup and a sublattice of G containing u,
and r(G, u) = r(S, u). Therefore, when considering the MV-algebras
r(G, u) we can safely assurne that u is a strong unit of G.
Definition 2.1.4 Let G and H be f-groups. A function h: G ----t H is

said to be an f-group homomorphism iff h is both a group-homomor-
phism and a lattice-homomorphism; in other words, for each x, y E G,
h(x-y) = h(x)-h(y), h(xVy) = h(x)Vh(y) and h(x/\y) = h(x)/\h(y).
Suppose that 0 < u E G and 0 < v E H, and let h: G ----t H be an f-
group homomorphism such that h(u) = v. Then h is said to be a unital
f- homomorphism.
Letting r(h) =def hho,u] denote the restrietion of h to the unit

interval [0, u], then r(h) is a homomorphism from r(G, u) into r(H, v).
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
The above result shall be strengthened in a subsequent chapter,

where we shall prove that r is a natural equivalence (Le., a full, faithful,
dense functor) between A and MV.
2.2 Good sequences

A sequence a = (al, a2, .. .) of elements of an arbitrary MV-algebra A
is said to be good iff for each i = 1,2 ... ,
and there is an integer n such that ar = 0 for all r > n. Instead of

a = (al, ... , an, 0, 0, ... ) we shall often write, more concisely,
2.2. GOOD SEQUENCES 35
Thus, if om denotes an m-tuple of zeros, we have identical good se-

quences
Note that by pre-pending to the good sequence (ab" . ,an) an m-tuple

1m of consecutive ones, the resulting good sequence (1 m , al, ... ,an)
is different from (al,"" an). For each a E A, the good sequence
(a,O .. . ,0, ... ) will be denoted by (a).
The good sequences of a boolean algebra Aare the nonincreasing
sequences of elements of A having a finite number of nonzero terms.
For totally ordered MV-algebras we have the following characteri-
zation of good sequences:
Proposition 2.2.1 If A is an MV-chain then each good sequence of A
has the form
(2.9) (F, a) for some integer p ~ 0 and a E A.
Proof: Immediate from Lemma 1.6.1(v). 0
Lemma 2.2.2 Suppose that A ~ TIi Ai is the subdirect product of a

family {AihEl of MV-algebras. A sequence a = (al,"" an, .. ') of ele-
ments of A is a good sequence if and only if for each i E 1 the sequence
is a good sequence in Ai, and there is an integer no > 0 such that

whenever n > no then for all i E I, '7ri(an ) = O.
Prao!: It is sufficient to note that an EB an+l = an means that '7ri(a n EB

an+!) = '7ri(an ) for each i E I. 0
In the light of Theorem 1.3.3, the above Lemma 2.2.2 and Proposi-
tion 2.2.1 yield a very useful tool for dealing with good sequences. As
an example, consider the proof of the following result:
Lemma 2.2.3 Let A be an MV-algebra. If a = (al,"" an, . .. ) and

b = (bI,' .. , bn , . .. ) are good sequences of A, then so is the sequence c
= (Cl, ... , Cn , . .. ) given by Cn = an V bn for each n.
Proof: Since a and bare both good sequences, there is an integer no

such that en = 0 for all n > no. By Theorem 1.3.3, A is a subdi-
reet product of a family {CihEl of MV-chains. For each i E I the
sequences ~ = (7ri(al)"'" 7ri(an), ... ) and b i = (7ri(bd, ... , 7ri(bn), ... )
are good sequences of Ci' Hence, by Proposition 2.2.1, ~ = (l P , ai)
and b i = (l Q ,ßi), where ai and ßi are in Ci' Therefore, 7ri(en) = 1
if n :5 max{p, q} and 7ri(en) = 0 if n > max{p, q} + 1. For n =
max{p, q} + 1, we have 7ri(en) = ai if p > q, 7ri(Cn) = ßi if P < q
and 7ri(en) = max{ai,ßi} when p = q. Consequently, letting Ci =
(7ri(CI), ... , 7ri(en), ... ) it follows that Ci is a good sequence for each
i E I, whence we conclude that C is a good sequence of A. 0
Example: To have a better intuition of the meaning of good sequences,
for every real number a ~ 0 let laJ denote the greatest integer :5 a,
and let (a) =def a - laJ be the fractional part of a. There will be no
danger of confusion between this notation and the notation for principal
ideals. Then a can be written as
a = 1 + ... + 1 + (a) + 0 + 0 + ...
with laJ many consecutive l's. Considered as elements of the MV-
algebra [0,1], the above summands al, a2,' .. of a trivially satisfy the
identity ai EB ai+l = ai for every integer i ~ 1. For 0 :5 ß E R, let
similarly
ß = ßI + ... + ßm-l + (ß) + 0 + ... ,

whereßI = ... = ßm-l = 1 = al = ... = an-I, 0 = an+l = a n+2 = ... ,
and 0 = ßm+1 = ßm+2 = .... Let "I = a + ß· Then "I = "11 + "12 + ... ,
where "11 = ... = "In+m-2 = 1, "In+m-l = (a) EB (ß), "In+m = (a) 0 (ß),
and 0 = "In+m+l = "In+m+2 = .,. . In a more compact notation, for
each i = 1,2, ... , the summand "Ii is given by
(2.10) "Ii = aiEB(ai-10ßI)EB(ai-20ß2)EB .. .EB(a20ßi-2)EB(a10ßi-l)EBßi'
In the light of (2.10) and (2.8), we can now give the following
Definition 2.2.4 For any two good sequences a = (al"", an) and b
= (bI, ... , bm ) their sum C = a + b is defined by C = (Cl, C2,.· .),
where for all i = 1,2, ...
(2.11) Ci =def ai EB (ai-l 0 bd EB ... EB (al 0 bi-d EB bio
2.3. THE PART1ALLY ORDERED MON01D M A 37
Since ap = bq = 0 whenever p > n and q > m, then Cj identi-

cally vanishes for each j > m + n. The notation c = (Cl,"" Cn+m) =
(al,' .. ,an) + (bI, ... , bm ) is self-explanatory.
The following immediate consequence of (2.11) will be frequently
used to compute the sum of two good sequences in an MV-chain:
2.3 The partially ordered monoid MA

Since by equation (1.14), (a EB b, a 0 b) is a good sequence, applying
Theorem 1.3.3 and Lemma 2.2.2 together with (2.12), we immediately
get that the sum of two good sequences is a good sequence. We denote
by M A the set of good sequences of A equipped with addition.
Proposition 2.3.1 Let A be an MV-algebra. Then M A is an abelian

monoid with the following additional properties:
(i) (cancellation) For any good sequences a, b, c, if a + b = a +c

then b = c;
(ii) (zero-Iaw) 1f a + b = (0) then a = b = (0).
Proof: By (2.11), a+(O) = a, addition is commutative, and the zero-law
holds. To prove associativity, by Theorem 1.3.3 we can safely assume
A to be totally ordered. By Proposition 2.2.1 and equation (1.16) in
Proposition 1.6.2, letting a = (l P , a), b = (F, b), and c = (F, c), we
have the identities
(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,
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:
(i) There is a good sequence c such that b +c = a ,.
(ii) bi ~ ai for all i = 1, ... ,n.
Praof: (i):::} (ii) is immediate from (2.11). (ii):::} (i). Observe

that by the remark following Lemma 1.6.1, (-.bn , ... , -.b l ) is a good
sequence. Let us denote by c = a - b the good sequence obtained
by dropping the first n terms in (al,"" an) + (-.b n, ... , -.b l ). We shall
prove that c + b = a. Using Theorem 1.3.3 we can safely assume A
to be totally ordered, so that by (2.9), a = (1 P , a) and b = (1 q , b). To
avoid trivialities assume both a and b to be different from 0 and from 1.
Then q ~ p. Upon rewriting b = (1 Q, b, QP-Q) , from n = p + 1 we get
(-.bn , ... , -.bd = (F-Q, -.b, oQ), and hence c is obtained by dropping
the first p + 1 terms from (1 2p - Q, a EB -.b, a e b).
Gase 1: b ~ a.
Then a EB -.b = 1, c = (lp-q,a e b) and c + b = (1 P , (a e b) EB
b, (a e b) 0 b) = (1 P , b V a, 0) = (1 P , a) = a.
Gase 2: b> a.
Then p > q, aeb = 0, c = (1 P - q - l , aEB-.b) and c+b
(1 P - I , a EB -.b EB b, (a EB -.b) 0 b) = (lP, a!\ b) = (F, a) = a. 0
Definition 2.3.3 Given any two good sequences a and b of A we write:
(2.13) b ~ a iff band a satisfy the equivalent conditions of 2.3.2.

2.3. THE PARTIALLY ORDERED MONOID M A 39
Proposition 2.3.4 Let a and b be good sequences.

(i) If b $ athen there is a unique good sequence c such that b+c = a.
This c, denoted a - b, is given by
(ii) In particular, for each a E A we have
(2.15) (-,a) = (1) - (a).
(iii) The order is translation invariant, in the sense that b $ a implies

b + d $ a + d for every good sequence d.
Proof" By an easy adaptation of the proof of Proposition 2.3.2, together

with Proposition 2.3.1 (i). 0
Proposition 2.3.5 Let a = (al, . .. , an, ... ) and b = (bI, ... ,bn , ... )
be good sequences of an MV-algebra A.
(i) The sequence
is good, and is in fact the supremum of a and b with respect to the

order defined by (2.13).
(ii) Analogously, the good sequence
is the infimum of a and b.

(iii) For all a, b, c E A we have
(2.16) ((a) + (b)) /\ (1) = (a $ b).
Proof" By Lemma 2.2.3, together with Proposition 2.3.2(ii) and (2.11).

o
2.4 Chang's f-group GA

From the abelian monoid M A , enriched with the lattice-order of Propo-
sition 2.3.5, one can routinely obtain an f-group GA such that M A is
isomorphie, both as a monoid and as a lattice, to the positive cone GA +.
To this purpose, mimicking the construction of Z from N, let us agree
to say that a pair of good sequences (a, b) is equivalent to another pair
(a', b') iff a + b' = a' + b. Transitivity of this relation follows from
cancellation, Proposition 2.3.1(i).
Notation: The equivalence class of the pair (a, b) shall be denoted by
[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.
Definition 2.4.1 Let A be an MV-algebra, and a, b, c, d E M A . We

say that the equivalence class [c, d] dominates the equivalence class
[a, b], in symbols,
[a, b] :::5 [c, d],
2.4. CHANG'S i-GROUP GA 41
Hf [C, d]-[a, b] = [e, (0)] for some good sequence e E M A . Equivalently,

[a, b] ~ [c, d] Hf a + d ~ c + b, where ~ is the partial order of M A
given by Definition 2.3.3.
Proposition 2.4.2 Let A be an MV-algebra.

(i) The relation ~ is a translation invariant partial order, making GA
into an i-group. Specijically, for any two pairs of good sequences (a, b)
and (c, d) the supremum of their equivalence classes in GA is the equiv-
alence class of ((a + d) V (c + b), b + d), where V is the supremum in
M A given by Proposition 2.3.5. In symbols,
(2.17) [a, b] V[c, d] = [(a + d) V (c + b), b + d].
(ii) Similarly, the injimum [a, b] A[c, d] is given by
(2.18) [a, b] "[c, d] = [(a + d) /\ (c + b), b + d].

(iii) The map a E M A t-+ [a, (0)] is an isomorphism between the
monoid M A , equipped with the lattice-order of Proposition 2.3.5, and
the positive cone GA + =def {[c, d] E GA I c 2: d}, with the lattice-order
inherited by restriction of~.
Proof: (i) The proof that ~ is a translation invariant partial order on

GA is routine. In order to prove (2.17), first of all, from the inequality
a+d ~ (a+d)V(c+b) weobtain [a,b] ~ [(a+d)V(c+b),b+d],
and, symmetrically, [c,d] ~ [(a+d)V(c+b),b+d]. Thus, [(a+d)V
(c + b), b + d] is an upper bound of [a, b] and [c, d]. To show that this
is indeed the least upper bound, for any upper bound [p, q] we must
find an element z E M A such that
(2.19) p + d + b = z + q + ((a + d) V (b + c)).

By hypothesis, there are x, y E M A such that p + b = x + q + a and
p + d = y + q + c. Let z E M A be such that x + y = z + (x V y);
the existence of z is ensured by the inequality x + y 2: x V y, using
Propositions 2.3.2 and 2.2.3. One now establishes (2.19) using the
cancellation property of M A , as folIows:
2p+ b+d
= 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
2.5 Chang completeness theorem

An f-group term in the variables Xl, •.. ,Xt is astring of symbols over the
alphabet {Xl,"" X n , 0, -, +, V, A, (,)} which is obtained by the same
inductive procedure used in Chapter 1.4 to define MV-terms.
Let T be an f-group term in the variables Xl,' .. , Xt and G be an
f-group. Substituting an element ai E G for all occurrences of the
variable Xi in T, for i = 1, ... , t, and interpreting the symbols 0, -, +,
V and A as the corresponding operations in G, we obtain an element
of G, denoted TG(aI, ... , at).
To each MV-term T in the n variables Xl, ... , X n we associate an
f-group term f in the n + 1 variables (Xl,"" X n , y), according to the
following stipulations:
Xi =def Xi, for each i = 1, ... , n,
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).
Proof: By induction on the number of operation symbols in T. The

basis is trivial. For the induction step, by definition of r we have:
(-,(1 )A(gl' ... ,gn)
= -'((1A(gl"" ,gn))
The EB-case is similar. 0

In the light of Proposition 2.4.2(iii), we shall now identify M A with
the positive cone of GA :
Proposition 2.5.2 If A is an MV-chain, al,"" an are elements

of A, and r(xI, ... , x n) is an MV-term, then the one-term sequence
(rA(al"'" an)) E MA ~ G~ coincides with fGA((ad,···, (an), (1)).
Proof" By induction on the number of operation symbols in r. The

basis is trivial. For the induction step, if a = </>EB'l/J then using (2.16),
together with the definition of the mapping a""'" iJ, and omitting
unnecessary superscripts, we can write:
= min((l), (</>(aI, ... , an)) + ('l/J(al"'" an)))
= min((l), ~((al)'" ., (an), (1)) + {b((ad,· .. , (an), (1)))
= (y /\ (~+ {b))((al)"'" (an), (1))
= </>EB'l/J((ad,···, (an), (1)).
In the -,-case, one similarly uses (2.15). 0
Theorem 2.5.3 (Completeness Theorem) An equation holds in [0,1]

if and only if it holds in every MV-algebra.
Prool Suppose an equation fails in an MV-algebra A. By the remark

following Definition 1.4.4 we mayassume that the equation has the form
r(xI' ... ' x n ) = o. By Corollary 1.4.7, A may be assumed to be totally
ordered. There are elements al, ... ,an E A such that rA(al' . .. , an) >
o. Letting GA denote the Chang f-group of A, and again writing M A =
G!, by Proposition 2.5.2 we have 0 < fGA((ad,···, (an), (1)) ~ (1).
Let S = Z(l) + Z(ad + ... + Z(a n ) be the subgroup of GA generated
by the elements (1), (al), ... ,(an), with the induced total order.
Since the order in GA is translation invariant, it follows that GA
is torsion-free. Since S is a finitely generated subgroup of GA, by the
fundamental theorem on torsion-free abelian groups, we can identify S
with the free abelian group zr, for some integer r ;::: 1. Its elements (1),
(al), ... , (an) are respectively identified with vectors h o, h l , ... ,hn E
zr; the set of nonnegative elements of S then becomes a submonoid P
of zr such that
(2.20) P n -P = {O} and PU -P = Zr.
For any two vectors h, k E zr let us write
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
(2.23) hw(o) ~p hw(1) ~p ... ~p hw(t).
Our aim is to replace by another total order ~P' over zr in

~p
such a way that the above inequalities still hold with respect to ~P"
and (zr, $;p') is isomorphie to a subgroup of the additive group R

with the natural order. For each j = 1, ... , t, let the vector d j E P
be defined by
dj = hw(j) - hw(j-I) = (djll dj2 , ... , djr).
Embedding zr into R r , we define the positive and the negative span of
the d j 's as follows:
t
(2.24) P'" = {E Ajdj I 0 $; Aj ER}, N'" = -P"'.
j=I
Note that P'" is a closed and convex subset of Rr, and whenever h E P*
and 0 $; a E R, then ah E P*.
Claim 1: Whenever Al, ... , At are real numbers 2: 0 and E~=I Aidi = 0,
it follows that Ai = 0 for each i such that d i =I- O.
Otherwise (absurdum hypothesis ) let I = {i E {I, ... , t} I d i =I-
O}, and assurne i'" E I, Ai- > 0 and EiEI Aidi = 0 for suitable
Ai 2: O. Stated otherwise, the homogeneous system of linear equations
EiEI Aidik = 0, k = 1, ... ,r has nontrivial solutions Ai 2: O. Now, the
solutions of this system are obtained by fixing arbitrarily some values,
say Aal"'" Aal" with p strictly smaller than the number of elements
of I, and then, for each ß E 1\ { aI, ... , a p }, computing Aß by means of
the formulas Aß = Ef=I CßaiAai' Here, the coefficients Cßai arise as the
result of performing rational operations on the integers dij . Thus, by
continuity, the existence of positive real solutions of the system implies
the existence of rational positive solutions. Choosing all Ai rationals
and multiplying by their least common denominator, we finally obtain
integers {ni 2: 0 I i E I} such that EiEI nidi = 0, and ni- 2: 1. By
definition of P, we have 0 =I- d i - $;p EiEI nidi = 0, whence d i - = 0,
a contradiction.
Having thus settled our first claim, it easily follows that

(2.25) P* n N* = {O}
From equations (2.20) and (2.25) we obtain P* n P - P* n zr,
whence in partieular, for arbitrary i, j = 0, ... , t,
To conclude the proof we need the foIlowing weIl known result of

linear algebra, whose proof is included for the sake of completeness.
Here, as usual, h . k denotes the scalar product of vectors h, k E RT.
Glaim 2: Let el, ... ,em be vectors in RT such that for any sequence of
real numbers 0 ~ Ab ... , Am, if E~IAiei = 0, then Al = ... = Am = O.
Then there is a vector V E RT such that ei . V > 0 for each i = 1, ... , m.
The proof is by induction on m. The basis is trivial. For the induc-
tion step, let el, ... ,em+1 in RT have the property that for any real
numbers 0 ~ Ab ... , Am +l, ifE~tlAiei = 0, then Al = ... = Am +l = O.
A fortiori, for any 0 ~ 111, ... , 11m with E~l l1iei = 0, we must have
111 = ... = 11m = O. Then by induction hypothesis there is a vector
u E RT such that ei· u > 0 for each i = 1, ... , m. We now argue by
cases:
Gase 1: e m+1· u > O. Then the desired conclusion follows upon letting
v=u.
Gase 2: e m+l· u = O. Then one gets the desired result by letting
V = au + em+1, where a = 1 + max{ei~:~~t li = 1, ... , m}.
Gase 3: e m+l· u < O. Then for each i = 1, ... , mIet the vector gi be
defined by
gi = ei - em+l·
e m +l· U
By our assumptions ab out the vectors el, ... , e m+1, whenever Vi 2: 0
and E~1 Vigi = 0 then necessarily, VI = ... = Vm = O. By induction
hypothesis, there is a vector t E RT such that gi·t > 0 for i = 1, ... , m.
Let w be defined by
em+l . t
w= t - u.
e m+l· U
Then, by direct inspection, w satisfies the inequalities em+l . w = 0 and
ei . w > 0 for each i = 1, ... ,m. Proceeding now as in Case 2 we get
the desired conclusion, and the claim is settled.
It follows from Claims 1 and 2 that there is a vector
such that g. d j > 0 for all nonzero vectors d j , j = 1, ... , t. By

continuity, g can be assumed to be in general position, in the sense
that 1'1,"" I'r are linearly independent over Q. Let
Then from (2.24) it follows that
(2.27) P* ~ rr; and N* ~ -rr;.

Let us now focus attention on the totally ordered abelian group T'
= (zr, '5:.P' ). Although T' need not coincide with T = (zr, '5:.p), still
by (2.25)-(2.27), for all i,j = 0, ... , t such that h i =1= h j we have
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
Since the inequalities in (2.21) are preserved under isomorphism, letting

~o = O(ho), ~I = O(hd,···, ~n = O(hn ), ~t = O(h t ) we can write 0 '5:.
~I, ... ,~n '5:. ~o and 0 < ~t '5:. ~o. Assuming without loss of generality,
~o = 1, we have ~t = fU(~I"",~n,1) > O. By Proposition 2.5.1, in
the MV-algebra B = r(U, 1) we have TB(~I"'" ~n) =1= 0, whence, a
fortiori, the equation T = 0 fails in the MV-algebra [O,lJ. 0

The idea of associating a totally ordered abelian group to any MV-
algebra A is due to Chang, who in [36] and [38] gave the first purely
algebraic proof of the completeness of the Lukasiewicz axioms for the
infinite-valued calculus, using quantifier elimination for totally ordered
divisible abelian groups.
After the unpublished proof of Wajsberg, the literat ure contains
many other proofs of the completeness theorem: a proof of Rose and
Rosser, based on syntactic methods and linear inequalities [216], the
proof in [50] using the representation of free f-groups, the proof of Panti
[197] using techniques from algebraic geometry. For the one-variable
fragment the reader may also see [193]. Our present geometrie proof,
using elementary algebra and convexity theory in finite-dimensional
vector spaces, was first published in [51] (also see [52] for its counterpart
for lattice-ordered abelian groups). Claim 2 in the proof is a variant
of Farkas' lemma (see [250]). Good sequences and the r functor were
first introduced in [163].
Chapter 3
Free MV-algebras
Free algebras are universal objects: every n-generated MV-algebra Ais

a homomorphic image of the free MV-algebra Free n over n generators;
if an equation is satisfied by Free n then the equation is automatically
satisfied by all MV-algebras. As a consequence of the completeness the-
orem, Free n is easily described as an MV-algebra of piecewise linear
continuous [0, 1]-valued functions defined over the cube [0,1]n. Known
as McNaughton functions, they stand to MV-algebras as {O, 1}-valued
functions stand to boolean algebras. Many interesting classes of MV-
algebras can be described as algebras of [0, 1]-valued continuous func-
tions over some compact Hausdorff space. The various representation
theorems presented in this chapter all depend on our concrete repre-
sentation of free MV-algebras.
3.1 McNaughton functions

Let /'i, be an arbitrary, finite or infinite cardinal 2:: 1. (Readers not
familiar with ordinals and cardinals may think of /'i, as being a positive
integer, without any essential loss.)
Suppose we are given distinct propositional variables
for each ordinal a < /'i,. Then by definition, each MV-term r in the
51
52 CHAPTER 3. FREE MV-ALGEBRAS
variables {Xo}o<1I: is a finite string of symbols over the alphabet
{O, ..." ED, (, ), Xo}o<1I:

obtained by the same inductive procedure of Definition l.4.I.
For any MV-algebra A the elements of All: have the form
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
Proj~ =def {7ro la < K}
denote the set of projections of All:.

The following is a generalization of Definition 1.4.3:
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:
(i) X;; =def 7ro n
(ii) OA =def the constant function °over All:,

(iii) (...,p)A =def ...,(pA),
(iv) (p ED (T)A =def (pA ED (TA).
The dependence of TA on K is tacitly understood. We let
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
Lemma 3.1.2 For each MV-algebra A and cardinal/'i, 2: 1, Term~ is

the smallest subalgebra of AAl< containing each projection 7ra E Proj~.
o
Definition 3.1.3 An MV-algebra A with a distinguished subset Y of

elements generating A is said to be free over (the generating set) Y, and
is denoted by Freey, iff for every MV-algebra Band every function
f : Y --+ B, f can be extended to a homomorphism 1 : A --+ B.
Note that 1 is uniquely determined by f. A moment's refiection

shows that for any two sets Y and Y' of the same cardinality /'i, , if A
is free over Y and A' is free over Y' then A ~ A'. Thus there is no
danger of ambiguity in saying that A is the free MV-algebra over /'i,
many generators, and writing A = FreeK..
Proposition 3.1.4 For each cardinal/'i, 2: 1, Term~,l] is the free MV-

algebra over the generating set Projlo,l], in symbols, Term~,l] = FreeK..
Proof: Let B be an arbitrary MV-algebra, and f : Projlo,l] --+ B be a
function, with the intent of uniquely extending f to a homomorphism
1 : Term~,l] --+ B. Let b = (b o, bl , ... ,ba, . . . )a<K. E BK. be given by
ba =def f(7r a ). Let the function cp map each term T (in the variables
X a ) into the element TB(b) E B, where TB E Term~ is the term
function on BK. determined by T. By Definition 1.4.4 together with the
Completeness Theorem 2.5.3, for any two terms p and a we have
if [0,1]1= p = athen B 1= p = a and then pB(b) = (TB(b).

Stated otherwise, whenever p[O,l] = a[O,l] E Term~·l], then cp maps p and
a into the same element of B. Since cp(Xa ) = X:(b) = ba = f(7f a ),
then cp naturally determines an extension 1 of f. By induction one
easily sees that 1 is indeed a homomorphism of Term~,l] into B. To
prove uniqueness, let 9 : Term~,l] --+ B be a homomorphism extending
f. Since 1 and 9 coincide over a subalgebra of AAl< containing all
projections, by Lemma 3.1.2 they coincide over all of Term~·l]. 0
The proof of the following proposition is now an immediate conse-
quence of Lemma 1.2.3 and Theorem 1.2.8:
Proposition 3.1.5 Let K, ~ 1 be a cardinal. Suppose the MV-algebra

A is genera ted by :5 K, elements. Then there is an ideal J in the free
MV-algebra FreeK. such that A is isomorphie to the quotient algebra
FreeK./ J. 0
Dur next task is to give a more explicit description of the elements

of the algebras FreeK.'
Definition 3.1.6 Let n ~ 1 be an integer. Then a function
f : [0, l]n ~ [0, 1]

is called aMeNaughton junetion over [0, l]n iff it satisfies the following
conditions:
(i) f is continuous with respect to the natural topology of [0, l]n;
(ii) there are linear polynomials PI, .. ' ,Pk with integer coefficients,
(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)·
Definition 3.1.7 Let>. be an infinite cardinal. A function
g: [0, lJ'~ ~ [0,1]
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))'
Proposition 3.1.8 11 ajunction 1 belongs to afree MV-algebra FreeK.'

then 1 is a M cNaughton function.
Proof: Trivially, the projections are McNaughton functions, and so is

the function constantly taking the value Oover [0,1]K.. The set of Mc-
Naughton functions is closed under pointwise application of the EB and
.., operations. To see that McN aughton functions are closed under the
operation EB, if 1 and gare given by linear polynomials PI, ... ,Pm and
qll ... , qn, then 1 EB 9 is given by the linear polynomials Pi + qj, (for
3.1. MCNAUGHTON FUNCTIONS 55
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
(3.1) ,tt =def b V 0) /\ 1.
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.
Proof: By induction on m = Imol + ... + Imn-ll. If m = 0, then

°
gU coincides with either constant function or 1, whence it trivially
belongs to Free n. For the induction step, assume the lemma holds for
m - 1. It is no loss of generality to assume that max(lmol, ... , Imn-ll)
= Imol. Proceeding by cases, assume first mo > 0. Let h = 9 - Xo, so
that
(3.2) h = h(xo, ... , Xn-l) = b + (mo - l)xo + mlXl + ... + mn-lXn-l·

By induction hypothesis, both functions h t and (h+1)tt belong to Free n.
We shall prove that for each x = (xo, . .. , Xn-l) E [O,l]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,
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
3.2 The one-dimensional case

In this section we shall prove that, up to isomorphism, the free MV-
algebra Freel over one generator coincides with the MV-algebra of
one-variable McNaughton functions.
To this purpose, for each n = 0, 1, ... , the nth Farey partition
FareYn of the unit interval [0,1] is defined by
Fareyo =def {0,1},
1
FareYl =def {O, 2' 1},
112
FareY2 =def {O, 3' 2' 3' 1},
1 121 323
{O, 4' 3' 5' 2' 5' 3' 4' I},
112 132 3 1 4 3 5 2 5 3 4
FareY4 =def {0'5'4'7'3'8'5'7'2'7'5'8'3'7'4'5,1},
Thus, FareYn+1 is obtained by inserting between any two consecutive
elements alb and eid of FareYn their mediant (a + e)/(b + d), with
° = 0ll and 1 = 1/1. Dur present FareYn is a notationally simpler
variant of the traditional "Farey sequence" of order n + 1, where one
inserts only those mediants whose denominators do not exceed the value
n + 1. Farey observed, and Cauchy proved, the following elementary
properties of Farey partitions:
Proposition 3.2.1 Let n = 0,1,2, .... Then we have

(i) All jractions in FareYn are automatieally in irredueible form, and
the interval [alb, eid] determined by any two eonseeutive fraetions
alb< eid in FareYn has the unimodularity property eb - ad = 1.
Moreover, alb< (a + e)/(b + d) < eid.
3.2. THE ONE-DIMENSIONAL GASE 57
(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
often, which is impossible. 0

The 2n + 1 elements of Fareyn can be displayed in increasing order as
follows:
o < Q < ... < I < 8 = eid< c < ... < w < 1.
A (Schauder) hat of FareYn is a function of either form:
lid
w E
Thus each Schauder hat is a continuous piecewise linear function

h : [0, 1] -+ [0, 1), whose graph consists of the four segments joining
the points (0,0), (f,0), (8, lid), (c,O) and (1,0). As a consequence
of unimodularity, each linear piece of h has the form y = mx + q, for
suitable integers m and q. Of course, the graphs of the two extrem al
hats only consist of two segments. For each n = 0,1,2, ... , we shall
denote by
Schaudern =def (h l , ... ,hu )
the naturally ordered sequence of all Schauder hats of FareYn, where
u = 2n + 1. We naturally say that hj and h j +1 are contiguous. The
following is a picture of Schauderi, for i = 0,1,2:
Schaudero Schauderl Schauder2

For each hat hi of Schaudern the multiplicity J.-li of hi is the inverse

of the maximum value of h i . Thus by definition, J.-li coincides with the
denominator di of the rational point Ci/ di E [0, 1] at which hi attains
its maximum value. For any n, the Schauder hat hi precedes hj in
Schaudern iff i < j iff ci/di < cj/dj . Note that if li-jl ::::: 2 then hil\hj =
O. From the sequence (h 1 , ••. , hu ) one can unambiguously recover the
multiplicity of each hat, as wen as the number n = log2(u - 1). It is
easy to see that Ei J.-lihi = 1.
By anode of a continuous function / : [0, 1] ---+ [0, 1] we mean a
nondifferentiability point in the domain of /. The vertices and 1 are °
always included among the no des of /. If / arises from the identity
function via a finite number of applications of the operations ..." EB, 8,
1\ and V, then the no des of / are finitely many, and they are an rational
numbers. By Proposition 3.1.4 this is always the case when / E Freel.
Proposition 3.2.2 Let h 1 , ... , h u be the hats 0/ Schaudern, in their

natural order, and with their respective multiplicities J.-l1, ... , 11-1.1. Let
k1 , ... , k2u - 1 be the hats 0/ Schaudern+l' with their respective multi-
plicities 6, ... , 61.1-1. Then we have:
k2i = hi 1\ hi+1! with 6i = (l1-i + 11-i+1) , /or each i = 1,2, ... , u - 1;

k2i - 1 = hi - (h i 1\ (h i- 1 V hi+1)) = hi - (h i 1\ (h i - 1 + hi+l)) =
hi e (h i- 1 EB hi+d, with 6i-1 = J.-li, /or each i = 2,3, ... , u - 1.
°
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 +
di )] and [(CiH + Ci)/(di+l + di ), 1], and is linear over both intervals

[(Ci-l + ci)/(di- l + di ), Ci/di] and [Ci/di , (Ci+l + Ci)/(di+l + di )]. Since
(hi - l A hi+l) = 0, the function coincides with k 2i - 1 at the point cddi ,
where it attains its maximum value 1/di . It is now easy to see that the
identity k 2i - l = h i - (h i A (h i- l + h iH )) holds over all of [0,1]. One
similarly proves the remaining identities. 0
By direct inspection we have
Proposition 3.2.3 (i) Each one of the extremal hats h l and h u of

Schaudern is the sum of two hats of Schaudern+l; all the remaining hats
of Schaudern are obtainable as a sum of three hats of Schaudern+l.
(ii) Suppose a linear combination Alh l + ... + Auhu with real coefficients
Ai coincides with the constant function 0. Then each coefficient Ai must
vanish. For short, the hats of Schaudern form a linearly independent
set of functions. 0
Definition 3.2.4 Fix an integer n = 0,1, ... , and let h l , ... , hu be

the hats of Schaudern, with their respective multiplicities Ih, ... , f.Lu.
Then by a subsystem S of Schaudern (in symbols, and with a slight
abuse of notation, S ~ Schaudern) we mean a sequence of integers
Al, ... , Au, with °: :;Ai :::; f.Li for each i = 1, ... , u. The associated
function fs : [0, 1] ~ [0,1] is defined by fs = Ei Aihi. We further let
S-' ~ Schaudern =def (f.Ll - Al,· .. , f.Lu - Au), whence, recalling that
f.Llhl + ... + f.Luhu = 1, we have the identity fs~ = -,fs = 1 - fs.
Two subsystems S ~ Schaudern and R ~ Schauderm are said to be
equivalent iff fR = f s·
From Proposition 3.2.3 we get

Proposition 3.2.5 1f S ~ Schaudern then for each integer p 2:: n
there is precisely one subsystem S' ~ Schauderp equivalent to S. 0
Proposition 3.2.6 Given two subsystems R ~ Schauder m and S ~

Schaudern, there exists an integer q 2:: max(m, n) and a subsystem
T ~ Schauder q such that fT = fR ffi fs.
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
of q follows from Proposition 3.2.1 (ii), together with Proposition 3.2.2

ensuring that all the no des of IRffJ!s are rational numbers. Let 9 = IR+
Is. For any two consecutive fractions x and y in FareYq one cannot have
g(x) > 1 and g(y) < 1. Let R' = (..\1"'" ..\w) and S' = (111,"" IIw)
be the equivalent subsystems of S and of R in Schauderq, as given
by Proposition 3.2.5. Let kI , ... , k w be the hats in Schauderq, with
their multiplicities 6, ... , ~w' Thus, for any two consecutive indexes
i, j E {1, ... , w}, if ..\j + IIj < ~j then necessarily ..\i + lIi ~ ~i' Let
T ~ Schauderq be the subsystem ('f/I, ... , 'f/w), where 'f/i = min(~i,..\i +
IId. It follows that the function IT coincides with IR EB Is at each
point of FareYq, and both functions are linear on each interval between
consecutive points of FareYq. Thus, IT = IR EB Is as required. 0
Example: With the notation of Proposition 3.2.6, let m = n = and °
R = S = (0,1), so that Schauderm = (1 - x, x) and IR = Is = X. The
nodes of the function IR EB Is = min(2x, 1) are 0, 1/2 and 1, which all
occur in the Farey partition FareYq for q = 1. By definition,
Schauder 1 = (max(l - 2x, 0), min(x,l - x), max(O, 2x - 1)),
with multiplicities 1,2,1, respectively. As a particular case of Proposi-

tion 3.2.5, the subsystem R' ~ Schauderi given by the triplet (0,1,1)
is equivalent to R. Indeed, IR = IRI = x. Letting the subsystem T of
Schauderi be defined by T = (0,2,1), we have IR EB Is = Ir. For no
subsystem U of Schaudero we would be able to write IR EB Is = lu·
Theorem 3.2.7 Let I : [0,1] - [0,1] be an arbitrary lunction. Then

the lollowing are equivalent:
(i) I arises from the identity function x : [0, 1] - [0, 1] via a finite
number 01 applications 01 the operations --, and EB;
(ii) there is a subsystem T such that I = IT.
Proof" One direction immediately follows from Proposition 3.2.2, since

both operations 0 and A are definable in terms of --, and EB. For the
other direction, as shown by the above example, the identity function
x arises as IR for some subsystem R. Proceeding by induction on the
number of operation symbols --, and EB needed to obtain I, in case
I = --,g one simply not es that Is~ = --,(Is). In the remaining case,
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.
Proof: By Proposition 3.1.4, Freel is a subalgebra of the MV-algebra of

one-variable McNaughton functions. Conversely, given any such func-
tion /, let us display its nodes as follows:
o= ni < n2 < ... < nu-I< n u = 1.

As already remarked, every ni is a rational number, and / is linear over
each interval [ni, ni+1]' By Proposition 3.2.1(ii), there exists an integer
q such that all ni's are members of FareYq. Let
o= ml < m2 < ... < mv-l < m v = 1
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
3.3 Decomposing McNaughton functions

In this section, for later use, we discuss so me properties of McN aughton
functions of n variables.
Let / : [0, l]n - [0,1] be a McNaughton function. Recalling Defini-
tion 3.1.6, let PI, ... ,Pk be the linear eonstituents of /, i.e., the distinct
polynomials representing the linear pieces of /. For any two such poly-
nomials P and q, either P ~ q, or q ~ p, or there is a hyperplane H di-
viding Rn into two closed half-spaces H+ and H- such that p(x) ~ q(x)
3.3. DECOMPOSING MCNAUGHTON FUNCTIONS 63
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
(j - 1)-dimensional face of P. We now inductively define a sequence

S(O) , S(I) , ... ,s(n) where each SO) is a set of j-dimensional simplexes,
as folIows:
S(O) = ;:(0) , S(I) = ;:(1) ,
S(Hl) = {[bp, F] I PE ;:(Hl) , FE S(i) , F ~ P},
where [bp, F] is the simplex whose vertices are the vertices of F

together with b p . Letting S = s(n), we have proved:
Proposition 3.3.1 For any McNaughton function f : [O,l]n - [0,1]

with its linear constituents PI, ... ,Pk, there is a set S of compact n-
dimensional simplexes with rational vertices such that
(i) The union of all simplexes in S coincides with the cube [0, l]n ,.
(ii) Any two simplexes in S are either disjoint or they intersect in a

common face;
(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
3.4 Ideals in free MV-algebras

The aim of this section is to describe the maximal ideals in free MV-
algebras and to give conditions ensuring that an ideal of a free MV-
algebra is an intersection of maximal ideals.
Dur results essentially depend on the fact that free MV-algebras are
algebras of [0, l]-valued continuous functions on a compact Hausdorff
topological space. Therefore, we shall firstly recall the ideal theory of
such function algebras.
3.4. IDEALS IN FREE MV-ALGEBRAS 65
As the reader will remember, for every MV-algebra A, T(A) and

M(A) respectively denote the set of ideals and of maximal ideals of A.
For X an arbitrary nonempty set, let A be a subalgebra of the
MV-algebra [0, 1]X of all functions f: X --+ [0,1], with pointwise defined
operations. For each set S ~ X, let
(3.5) Js =def {f E AI f(x) = °for all x ES},
be the set of functions vanishing over S. Further, for any ideal J in A
let
(3.6) VJ =def n{f- 1 (0) 1/ E J}
be the intersection of the zerosets Z(J) =def f- 1 (0) of all functions
f E J. Trivially, Js E T(A); further, J0 = A and Jx = {O} = (0). For
each x E X we shall write J x instead of J{x}.
Lemma 3.4.1 Let X be a nonempty set and let A be a subalgebra of
the MV-algebra [0, 1]x. Thenfor each x E X, Jx E M(A).
Proof: Suppose first that A = [0,1]x. Jx is a proper ideal of A, because
the constant function 1 is not among its elements. If f E A \ Jx , then
f(x) > 0, and we can find an integer n ~ 1 such that nf(x) 2:: 1. It
follows that
->nf = 1 - (J ffi ... ffi f) E Jx ,
, "
'"
n times
whence, by Proposition 1.2.2, Jx E M(A). Now to complete the proof
it suffices to apply Proposition 1.2.16 (ii). 0
For any topological space X we shall denote by
Cont(X)
the subalgebra of [0,1]X given by the continuous [0,1]-valued func-
tions over X. It is understood that [0, 1] is equipped with its natural
topology.
Proposition 3.4.2 Let X be a compact Hausdor.fJ space, and A be
a subalgebra of the MV-algebra Cont(X). The map J ~ V J is an
inclusion reversing map from the set T(A) of ideals into the family of
closed subsets of X. Moreover, VJ =f 0 for each proper ideal J in A.
Proof" The continuity of each function I E A ensures that VJ is a closed

subset of X; trivially, the map J 1-+ VJ reverses inclusions. Let J be
a proper ideal in A. By way of contradiction, suppose VJ = 0. Then
by the assumed compactness of X there are finitely many functions
11, ... , Is E J such that the intersection of their zerosets is empty. Let
I = !I E9 ... E9 Is· Then I E J and the zeroset of I is empty. Since I
attains a minimum value > 0, there exists an integer m ~ 1 such that
m/(x) > 1 for all x E X. We conclude that I E9 ... E9 I (m times),
constantly takes the value 1. Thus 1 E J, and J = A, a contradiction.
o
A subalgebra A of [0, 1jX is said to be separating iff for any two
distinct points x and Y in X, there is I E A such that I (x) = 0 and
I(y) > o.
Theorem 3.4.3 Let X be a compact Hausdorff space and A be a sep-
arating subalgebra 01 Cont(X). Then we have
(i) The map x 1-+ Jx is a one-one correspondence between X and
M(A);
(ii) For each closed set S ~ X, VJs = S;
(iii) For each proper ideal J in A, JVJ is the intersection 01 alt maximal
ideals in A containing J.
Proof" (i) By Lemma 3.4.1 together with our assumption about A,

the map x 1-+ Jx is a one-one correspondence from X into M(A). To
see that this map is onto M(A), let J be a maximal ideal of A. By
Proposition 3.4.2, VJ is a nonempty closed subset of X. Since for each
y E VJ , J y 2 J, we conclude that VJ is a singleton, say VJ = {x} and
J= Jx '
(ii) Trivially, S ~ VJs ' To prove the converse inclusion, suppose
Z E X \ S. Since A is separating, for each y E S we can find I y E A
such that Iy(z) = 0 and ly(Y) = ay > O. By continuity, there is an open
neighborhood Uy of Y such that Iy(x) > by = ay/2 for each x E Uy.
By a standard compactness argument, there are finitely many functions
11, ... ,!k E A, together with real numbers numbers b1, ... ,bk > 0 such
that, upon defining I = !I E9 ... E9 Ik, we can write I(z) = 0 and
°
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
Remark: Let X be a compact Hausdorff space and A be a separating

subalgebra of Cont(X). It follows from (ii) in the above theorem that
each closed subset of X is an intersection of zerosets of functions in A.
Therefore, the complements of the zerosets of functions in A form an
open basis for the topology of X.
Corollary 3.4.4 Let X be a compact Hausdorff space, let A be a sep-

arating subalgebra of Cont(X), and J be an ideal in A. Then J is
an intersection of maximal ideals iff J = JVr Moreover, the map
U J--+ J X\U is an order isomorphism from the set of proper open subsets
of X and the set of ideals of A that are intersection of maximal ide-
als (both sets being ordered by inclusion). The inverse isomorphism is
given by J J--+ X \ VJ . 0
Example: Let f: [0, 1]-+ [0,1] be defined by f(x) = x sin(l/x), for

°< x ::; 1 and f(O) = 0; let further f+(x) = max(f(x) , 0) for each
x E [0,1]. Then (1+) is a proper ideal of Cont([O, 1]) and it is not
hard to verify that JV(f+) strictly contains (1+). Therefore, (1+) is
a principal proper ideal of Cont([O, 1]) which is not an intersection of
maximal ideals.
Let X be a compact Hausdorff space, and A a sub algebra of the

MV-algebra [O,I]x. For every nonempty subset Y of X, let
denote the subalgebra of [O,l]Y given by the restrietions to Y of the

functions in A. The map f J--+ pU) = fly defines a surjective ho-
momorphism p: A -+ Aly. In case Y happens to coincide with VJ for
some proper ideal J in A, it follows that Ker(p) = JVr Then from
Lemma 1.2.7 we immediately obtain
Proposition 3.4.5 Let X be a compact HausdorfJ space, and A be

a separating subalgebra 01 Cont(X). Por each J E I(A), the map
I I J ~ IlvJ is an isomorphism from AI J onto AlvJ il and only il J is
an intersection 01 maximal ideals 01 A. 0
In order to apply the above results to free MV-algebras, we prepare
Lemma 3.4.6 For each cardinal/'i, 2:: 1, Free", is a separating subalge-

bra 01 Cont([O, 1]"').
Proof" In the light of Proposition 3.1.8, we may visualize the elements

of Free", as McNaughton functions on the cube [0,1]"', the latter being
equipped with product topology. Therefore, Free", is a subalgebra of
Cont([O, 1]"'). In order to show that this sub algebra is separating, let
Y = (Yo, Yl, ... ) and z = (zo, Zl,' .. ) be two distinct points of [0,1]"'.
Assume without loss of generality Yo < Zoo Let r be a rational number
such that Yo < r < Zoo Let p = p(x) = ax + b be a linear polynomial
with integer coefficients such that a > 0 and r = -bla. Then by
Proposition 3.1.4 and Lemma 3.1.9, the McNaughton function I(x) =
p#(xo) belongs to the free MV-algebra Free"" and, moreover, I(y) = 0
and I(z) > o. 0
As an immediate consequence of the above lemma together with
Theorem 3.4.3, we record here the following result, that will find appli-
cation in subsequent sections.
Proposition 3.4.7 The map x ~ J x is a one-one correspondence

between points 01 [0, 1]'" and maximal ideals 01 Free",. The inverse
correspondence is given by JE M(Free",) ~ the only point oIVJ . 0
The following lemma is of independent interest:
Lemma 3.4.8 Let I, 9 E Free",. Then
(3.7) gE(J) ifJ Zg2ZI·

Proof: For the nontrivial direction, assuming Z 9 2 Z f, since both 9

and f only depend on finitely many variables, say Xl,"" X m we can
safely restrict attention to [0, l]n. By a simple adaptation of Propo-
sition 3.3.1 there exists a set S = {Tb"" Tu} of compact convex
n-dimensional simplexes, whose union coincides with [0, l]n, any two
Ti and T; being either disjoint or intersecting in a common face, and
with the additional property that both functions fand 9 are linear
over each Ti. Let ViO,"" Vin be the vertices of the simplex Ti. For
°
each 1 ~ i ~ u and ~ j ~ n there is an integer mij ~ 1 such that
g(Vij) ~ mijf(vij). As a matter of fact, if f(Vij) = °
then by hy-
pothesis g(Vij) = 0, and we can take mij = 1. If, on the other hand,
f(Vij) > 0, then the existence of mij follows from the archimedean
property of the real numbers. Let mi = max(miQ, ... , min). Since
each x E Ti is a convex combination of the vertices of Ti, and since
both 9 and f are linear over 7i, we get g(x) ~ md(x) for each xE 7i.
Let m = max(mb"" m u ). From [O,l]n = Uf=l7i , it follows that
g(x) ~ mf(x) for each x E [O,l]n. Since 9 ~ 1, we finally obtain
g(x) ~ min(l, mf(x)) = f(x) $ ... $ f(x) (m times). In conclusion,
gE (1). 0
In contrast with the example given after Corollary 3.4.4, we have

Theorem 3.4.9 Each proper principal ideal of FreeK, is an intersec-
tion of maximal ideals.
Proof: An immediate consequence of Corollary 3.4.4, Lemma 3.4.6,
together with Lemma 3.4.8. 0
In the next section we shall give an example of a nonprincipal ideal
of Freel which is an intersection of maximal ideals. The following
example shows that free MV-algebras also contain proper ideals which
are not intersections of maximal ideals.
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
3.5 Simple MV-algebras

An MV-algebra is called simple Hf it has exact1y two ideals. In other
words, an MV-algebra Ais simple if and only if Ais nontrivial and {O}
is its only proper ideal.
Theorem 3.5.1 For every MV-algebra A the following eonditions are

equivalent:
(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);
(iii) A is isomorphie to a subalgebra of [0,1].
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
Corollary 3.5.2 Every simple MV algebra A has at most the eardi-

nality of the eontinuum. 0
To obtain further information about simple MV-algebras, recall that

for each integer n = 2,3, ... , the n element (Lukasiewicz) chain Ln is
defined by
1 2 n- 2
(3.8) Ln =def {O, - - , - - , ... , - - , I}.
n-1 n-1 n-1
3.5. SIMPLE MV-ALGEBRAS 71
Proposition 3.5.3 Let A be a subalgebra of [0,1]. Let A+ = {x E A I

x > O}, and a = inf A+ be the infimum of A+. 1f a = then A is a °
°
dense subehain of [0, 1]. 1f a > then A = Ln for some n ;::: 2.
° °
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
K = {O, a, 2a, ... , (m - l)a, I} ~ A.
We shall prove that K = A. For otherwise, let x E A\K (absurdum

hypothesis ). If (m - l)a < x < 1 then -,x would be a nonzero element
of A strictly smaller than a, which is impossible. If for some j =
O,I, ... ,m - 2 we have ja < x < (j + l)a then, again, (j + l)a-
x = (j + l)a 0 -,x is a nonzero element of A which is strictly smaller
than a, another contradiction. Having thus proved that K = A, since
° < a < 2a < '" < (m - l)a < 1, it follows that a = 1 - (m - 1 )a,
whence a = l/m and A = L(m+l), as required. 0
Remark: Let m, n ;::: 2. Then from the above proof it follows that
L m ~ Ln iff for some k E {I, ... , n -I} we have the identity m~l = n~l
iff m - 1 is a divisor of n - 1.
Corollary 3.5.4 An MV-algebra A is finite and simple if and only if

A is isomorphie to an MV-algebra Ln for some integer n ;::: 2.
Proof: For the nontrivial direction, if A is finite and simple, by The-

orem 3.5.1 we can identify A with a finite sub algebra B of [0,1]. By
Proposition 3.5.3, letting n ;::: 2 be the number of elements of B, we
have that B = Ln. 0
As promised, we shall now show that there are maximal ideals in

Freel that are not principal.
Example: Consider the subalgebra BA of [0,1] generated by an irrational

number A E [0,1]. Sinee BA is simple and has one generator, we ean
write BA = Freed J for some maximal ideal J of Freel' Now, J eannot
be a principal ideal: for otherwise, (absurdum hypothesis ), reealling
Proposition 3.4.7 and writing J = Jx = (J), for some x E [0,1], and
1 E Freel, we get that the zeroset Z 1 eoincides with the singleton {x}.
On the otherhand, sinee all linear pieees of 1 have integer eoeffieients,
x must be a rational number, say x = alb for relatively prime integers
° ~ a ~ b with b > 0. It follows that Freed J is isomorphie to the
MV-algebra Lb+l of possible values of MeNaughton functions g at the
rational point alb. Sinee by Proposition 3.5.3, BA is infinite and L b+ 1 is
finite, these two algebras are not isomorphie. Thus J is nonprineipal.
3.6 Semisimple MV-algebras

In the light of Corollary 1.2.15, for any MV-algebra A, we eall radical
01 A the interseetion of all maximal ideals of A. The radieal of A will
be denoted by
Rad(A).
An MV-algebra A is said to be semisimple Hf A is nontrivial and

Rad(A) = {O}. In partieular, every simple MV-algebra is semisimple.
In what follows, we will eonsider only nontrivial MV-algebras.
As an immediate eonsequenee of Proposition 1.2.10, for each ideal J
of A, the quotient AI J is a simple MV-algebra if and only if J is max-
imal. Henee, by Theorem 3.5.1, AI J is isomorphie to a subalgebra of
[0, 1] if and only if J is a maximal ideal of A. The next proposition will
be promptly reeognized as an immediate eonsequenee of Theorem 1.3.2:
Proposition 3.6.1 An MV-algebra is semisimple il and only il it is a

subdirect product 01 subalgebras 01 [0,1]. 0
Corollary 3.6.2 Every free MV-algebra is semisimple.
Proof: Immediate from Proposition 3.1.4. 0

Our next aim is to eharaeterize the elements of the radical.
3.6. SEMISIMPLE MV-ALGEBRAS 73
Definition 3.6.3 An element a in an MV-algebra A is said to be in-

finitely small or infinitesimal iff a =I- 0 and na ~ ,a for each integer
n ;::: O. The set of all infinitesimals in A will be denoted by Infinit(A).
Remark: If ais infinitesimal, then the elements na, for n = 0,1,2, ... ,
form a strictly increasing sequence. Indeed, if for some n, na = (n+ 1 )a,
then one would have 1 = ,((n + l)a) EB na = ,a V na = ,a, i.e. a = 0,
a contradiction.
Proposition 3.6.4 For any MV-algebra A, Rad(A) = Infinit(A) U
{O}.
Proof: Suppose a rt Rad(A). Then there is a maximal ideal of A, say
M, such that a rt M. Hence (M U {al), the ideal generated by a and
M, must coincide with A, and by (1.9), there is an integer n 2: 0 and
an element z E M such that 1 = na EB z. If na ~ ,a, then one would
have a ~ -.na ~ z, whence a E M, a contradiction. Hence a is not
infinitesimal, whence Infinit(A) ~ Rad(A).
Conversely, assurne that 0 < a E A is not infinitesimal. Then
there is an integer m ;::: 0 such that ma 1:. -.a. Hence ma e -.a =
ma 0 a > 0, and by Proposition 1.2.13, there is a prime ideal P of A
such that ma e -.a rt P. By Equation (1.7) we have ,a e ma E P. By
Corollary 1.2.12 there is a maximal ideal M of A such that P ~ M.
Then -.(aEBma) = ,aema E M. Therefore, (m+1)a rt M, and hence,
art M. In conclusion, art Rad(A), and Rad(A) ~ Infinit(A) U {O}, as
required to complete the proof. 0
It follows from the above proposition that an MV-algebra A is
semisimple if and only if A has no infinitesimals. In particular, by the
remark following Definition 3.6.3, each finite MV-algebra Ais semisim-
pIe; moreover, since A has only a finite number of ideals, A is the
subdirect product of a finite family of finite subalgebras of [0, 1]. The
next proposition gives a sharper result:
Proposition 3.6.5 An MV-algebra A is finite if and only if A is iso-
morphie to a finite produet of finite ehains, in symbols,
(3.9) A ~ Ld1 X •.• X Ld,., for some integers 2 ::; d 1 ::; d2 ~ ••• ~ du.
This representation is unique, up to the ordering of faetors.
Proof: For the nontrivial direction, assume A is a finite MV-algebra.

As noted above, Ais semisimple. Let 11 , ••• ,Iv. be the list of an distinct
maximal ideals of A. Then A is a subdirect product of the MV-algebras
AlI1 , AII2 , ••• , AI Iv.. By Corollary 3.5.4, for each t = 1, ... , u the
quotient algebra AlIt can be identified with the chain L dt , for some
integer dt 2: 2. Let us now define the produet MV-algebra P by
P = II{L dt I t = 1, ... , u}.

For each element x E A and every ideal I t the element x I I t E L dt
ean be identified with a rational number cl (dt - 1), for some integer
o ~ c ~ dt -1. Let ß: x E A ~ (xIIll ... ,xIIv.) E P. Then from
the semisimplicity of A it follows that ß is injeetive. To see that ß
is surjeetive, it is sufficient to show that every element in P of the
form (0, ... , 1/(dt - 1), ... ,0) is in the range of ß. To this purpose, let
at be the minimum of an elements z E A such that z I I t = 1I (d t - 1).
The existence of at is ensured by the assumed finiteness of A. For every
maximal ideal Ir of A other than I t , there is an element w E A such that
at ~ wand w I Ir = O. For otherwise, Ir ~ I t , thus contradicting the
maximality of Ir. We have proved that ß(at) = (0, ... , 1/(dt -1), ... , 0),
and hence, A is isomorphie to the produet MV-algebra P. To prove
uniqueness of the deeomposition (3.9), suppose
A = L b1 X ••• X L bv
for suitable integers bi 2: 2. Then for each i = 1, ... , v, the kernel of
the projeetion function (Xl, ... , x r ) ~ Xi is a maximal ideal I i of A.
Moreover, AI I i ~ L bi and A ~ AlI1 x ... x AI Ir. 0
Lemma 3.6.6 Let A be a (nontrivial) MV-algebra, and J be an ideal

0/ A. Then the quotient algebra AI J is semisimple i/ and only i/ J is
an intersection 0/ maximal ideals 0/ A.
Proof: Let h J : A - 4 AI J be the natural projeetion.

Suppose that AI J is a semisimple MV-algebra. If {MdiEI denotes
the family of an maximal ideals of AI J, then
J = h"J1( {O}) = h"J1(n Mi) =

iEI
nh·:/(M
iEI
i ).
By Proposition 1.2.10, J is an intersection of maximal ideals of A.

Conversely, assume J to be an intersection of maximal ideals of A.
Then J is the intersection of all the maximal ideals of A containing J.
If {Mihel denotes this family, then, by Proposition 1.2.10, {hJ(Mi)heI
is the family of all maximal ideals of AI J and hJ(J) = Rad(AI J). 0
The following result strengthens Proposition 3.6.1:
Theorem 3.6.7 An MV-algebra A with K, many generators is semisim-

ple if and only if for some nonempty closed subset X ~ [0, 1]/t, A is
isomorphie to the MV-algebra of restrietions to X of all functions in
Free/t.
Proof: One direction is an immediate eonsequenee of Proposition 3.6.1.

The other direction follows at onee from Propositions 3.1.5 and 3.4.5
and Lemma 3.6.6. 0
Corollary 3.6.8 For any MV-algebra A the following eonditions are

equivalent:
(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
The following theorem direetly follows from Theorem 3.4.9 and

Lemma 3.6.6.
Theorem 3.6.9 Suppose A ~ Free/tl J, with J a prineipal ideal. Then

A is semisimple. 0

Our definitions of term and free algebras are particular eases of abstract
definitions in Universal Algebra (see, for instance, [25], [104] or [150]).
Lemma 3.1.9 is due to McNaughton [152]. The simplified proof
presented here is due to Rose and Rosser [216]. The proof of the one
variable McNaughton theorem, based on Farey series and Schauder

hats, is taken from [181].
Our FareYn is a variant, due to Stern and Brocot, of the traditional
Farey sequence of order n + 1, considered in most textbooks in number
theory.
For a proof of the fundamental theorem on polyhedra see, for in-
stance, [250] or [82, p.31]. The fact, mentioned at the beginning of
Section 3, that every polyhedral complex can be refined to a simpli-
cial complex without adding new vertices is also weIl known: see, for
instance, [82, Theorem III, 2.6]. Proposition 3.3.1 can be found in
McNaughton's paper [152].
Lemma 3.4.8 was first proved in [172, Proposition 2.4]. Theorem
3.4.9 is the algebraic counterpart of results of Hay, W6jcicki and Rose
that will be presented in the next chapter. The proof given here is
taken from [170]. The example after Theorem 3.4.9 is due to W6jcicki
[243].
Theorem 3.5.1, Proposition 3.6.1 and CoroIlary 3.5.4 are due to
Chang [36] (see also [13],[144]).
Definition 3.6.3 and Proposition 3.6.4 are due to Rodriguez [212]
(see also [132, Proposizione 7]). Proposition 3.6.5 appears in [144],
[233], [212].
Chapter 4
Lukasiewicz oo-valued
calculus
Since every MV-term T is astring of symbols over a finite alphabet,

one may naturally consider the following decision problem: does there
exist an effective procedure (for definiteness, a Turing machine) decid-
ing whether an arbitrary equation T = 1 holds in all MV-algebras ?
More generally, given two terms (J and T, does there exist an effective
procedure to decide whether the McNaughton function determined by
(J belongs to the principal ideal determined by T in the free MV-algebra
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
78 CHAPTER 4. LUKASIEWICZ oo-VALUED CALCULUS
word problem for finitely presented MV-algebras shall be reformulated

in terms of logical consequence. We shall give a positive solution to
the problem, after a detailed analysis of the subtleties of the notion of
consequence in the infinite-valued calculus.
4.1 Many-valued propositional calculi

In the early twenties, Jan Lukasiewicz introduced systems of logic in
which propositions admit as truth values real numbers between and °
1. As main propositional connectives he considered implication - and
negation -', as given by the following "truth tables", where x and y
denote arbitrary elements of the real unit interval [0,1]:
(4.1) x - Y =def min(1, 1- x + Y)

and
(4.2) -,x =def 1 - x.
Along with this infinite-valued propositional system, Lukasiewicz
also considered, for each natural number n ;::: 2, an n-valued system, in
which the truth values are the rational numbers
1 n- 2
0, --1' ... ' - - ' 1,
n- n-1
and the truth values for implication and negation are again given by
formulas (4.1) and (4.2). Thus in particular, for n = 2 the only pos-
°
sible truth values are and 1, and the above formulas give back the
truth tables for implication and negation in the classical propositional
°
calculus, provided 1 is interpreted as true and as false.
As the reader will recall, formula (4.2) gives the operation -, in the
MV-algebra [0,1]. On the other hand, formula (4.1) can be written in
terms of the MV-operations as follows:
(4.3) x - Y = -,x EB Y
whence,
(4.4) xEBy = -,x - y.

4.1. MANY-VALUED PROPOSITIONAL CALCULI 79
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
(Propositional) formulas are built, exactly as boolean propositions,

from a denumerable set of propositional variables, X, XI, XII, ... , for
short, Xo, Xl, X 2 , ••• , by means of the connectives of negation"" and
of implication -+.
In more detail, the set Form of formulas is given inductively as
follows:
Fl) Each propositional variable X k is a formula.

F2) If a is a formula, then ...,a is a formula.
F3) Ha and ß are formulas, then (a -+ ß) is a formula.
Notation: For each formula a, Var(a) will denote the set of all propo-
sitional variables occurring in a.
Definition 4.1.2 Let A be an MV-algebra. Then an A-valuation is

a function 11: Form -+ A satisfying the following properties, where a
and ß denote arbitrary formulas:
VI) 1I(...,a) =del ""1I(a),

V2) 1/(0 -+ ß) =def 1/(0) -+ 1/(ß)·

Since unique readability also holds for formulas, any A-valuation 1/
is uniquely determined by its values 1/(Xo) , 1/(X1 ), ....
An A-valuation 1/ is said to A-satisfy a formula 0 iff lI(a) = 1; 0 is
an A-tautology iff 0 is A-satisfied by an A-valuations. Formulas a and
ß are semantically A-equivalent iff 1/(0) = 1/(ß) for an A-valuations
1/. From condition (V2) we get that 0 and ß are A-equivalent iff both
formulas 0 -+ ß and ß -+ 0 are A-tautologies. Let e be a set of
formulas. We say that a formula 0 is a semantic A-consequence of
e iff each A-valuation 1/ that A-satisfies an the formulas in e also A-
satisfies o. Thus in particular, 0 is an A-tautology iff a is a semantic
A-consequence of the empty set.
Using the identities (4.3), (4.4), every formula 0 containing the

variables Xl, ... , X k can be straightforwardly transformed into an MV-
term Ta in the same variables. Conversely, upon replacing every occur-
rence of the constant 0 in a term T by, say, the formula -,(X ---+ X),
then T is transformed into a propositional formula 0'T' We shall tacitly
identify propositional formulas and MV-terms, using the maps 0 I - t Ta
and T f--+ 0'T' Also, using all abbreviations introduced in Section 1.4,
we shall freely write, e.g., 1 instead of -,-,(X ---+ X).
As a first application of this identification, an easy induction on the
number of connectives in 0 yields
Proposition 4.1.3 (i) Let A be an MV-algebra and 0 be a formula,

with Var(o) ~ {Xill ... , X ik }. Then for each A-valuation 1/ we have
(4.6) 1/(0) = oA(1/(Xi}), ... , 1/(Xik )),
where OA: A k ---+ A is the term junction defined in Section 1.4;
(ii) A formula 0 is an A-tautology if and only if the MV-equation 0 = 1
holds in A;
(iii) Formulas and ß are semantically A-equivalent iff the equation
0
o = ß holds in A iff OA = ßA. 0
For each integer n ~ 2, the n-valued Lukasiewicz propositional cal-

culus deals with propositional formulas equipped with the relation of
4.1. MANY- VALUED PROPOSITIONAL CALCULI 81
semantic Ln-equivalence. In the Lukasiewicz infinite-valued proposi-

tional calculus one considers propositional formulas equipped with the
relation of semantic [O,I]-equivalence.
By Proposition 4.1.3, the Completeness Theorem (Theorem 2.5.3)
has the following equivalent formulation:
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
Since OUf main concern here is the infinite-valued calculus, by a

valuation we shall henceforth mean a [0, 1]-valuation; by a tautology we
shall understand a [O,I]-tautology; also, semantic [O,I]-equivalence and
[O,I]-consequence, shall be simply referred to as semantic equivalence
and semantic consequence, respectively.
Notation: For each 8 ~ Form, 81= will denote the set of semantic con-
sequences of 8. In particular, 01= will denote the set of all tautologies.
The last two results above allow us to identify term functions a[O,I)
and semantic equivalence classes of propositional formulas. Recalling
now our analysis of free MV-algebras in Section 3.1, one may reasonably
expect that McNaughton functions play in the infinite-valued calculus
the same role played by boolean functions in the classical propositional
calculus. As an instance of this role, consider the proof of the following
result, which will be considerably strengthened in Corollary 4.5.3 and
in Theorem 4.6.10 below:
Proposition 4.1.5 There exists an effective procedure (for definite-

ness, a Turing machine) enumerating all formulas that are not tautolo-
gies.
Proof: Fixing an arbitrary lexicographic ordering of all strings over

alphabet E, let 'l/JI, 'l/J2, ... be a list of all n-variable formulas 'l/Jt =
'l/Jt(X I , ... , X n ) E Form. Let similarly xl, x 2 , ... list all n-tuples x =
Xl, ... , X n of rational numbers in the unit real interval [0,1]. Thus,
Xi = (xL . .. , x~) for suitable X} E [0,1] n Q. For each i = 1,2, ... let
vi be the valuation such that vi(Xj ) = X} for all j = 1, ... , n., and
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.
4.2 Wajsberg algebras

Stone's representation theorem for boolean algebras allows one to vi-
sualize the connectives of negation, disjunction and conjunction of the
classical propositional calculus as the set-theoretical operations of com-
plement, union and intersection. In the infinite-valued case, since EB and
0, together with negation -, can express V and A, the (additive) op-
erations EB and 0 are regarded as more fundamental than the (lattice )
operations V and A.
Historically, Chang introduced MV-algebras as generalized boolean
algebras; indeed, Chang's completeness theorem is a generalization of
the fact that if an equation holds in the boolean algebra {O, 1}, then the
equation holds in every boolean algebra. Already the proof of the one-
variable case of McNaughton's theorem given in Section 3 enhances the
role of the associative-commutative operation EB, giving normal form
reductions of all elements of Freel as sums of Schauder hat functions.
In a later chapter we shall establish a categorical equivalence between
MV-algebras and abelian f-groups with strong unit (the latter being
deeply related to the time-honoured theory of magnitudes). Altogether,
the operations 0 and EB have a distinctly arithmetic, rat her than a set-
theoretic significance.
On the other hand, the identities x ~ y = -,x EB y and x EB
y = ,x ~ y show that there is no essential difference of expressive
power between the additive connectives and the implication connective.
It turns out that for our analysis of proofs and consequence in this
chapter, it is more convenient to replace the 0 and EB operations by
4.2. WAJSBERG ALGEBRAS 83
the implieation eonneetive -. In this way one obtains the following

equivalent variant of MV-algebras:
Definition 4.2.1 A Wajsberg algebra (for short, a W-algebra) is a

system A = (A, -",1), where A is a nonempty set, and the binary
operation -, the unary operation ' and the distinguished element 1
satisfy the following equations:
W1) 1-x=x
W2) (x - y) - ((y - z) - (x - z)) = 1
W3) (x - y) - y = (y - x) - x
W4) (,x - ,y) - (y - x) = 1.
Lemma 4.2.2 Let A be an MV-algebra, and put x - y =def ,x EB y

and 1 =def ,0. Then (A, -, ,,1) is a W-algebra. 0
Our next task is to prove that eonversely, eaeh W-algebra beeomes

an MV-algebra, onee equipped with the operations -', x 8 Y =dej -'x -
y and 0 =def ,1.
Lemma 4.2.3 Let A = (A, -,1) be a system satisfying (Wl), (W2)

and (W3). Then the following properties hold fOT every x, y and z in
A:
W5) x - x = 1
W6) Ifx - y = y - x = 1, then x = y

W7) x -1 =1
W8) x - (y - x) = 1
W9) If x - y =Y ---4 Z = 1, then x - z = 1
W10) (x ---4 y) ---4 ((z ---4 x) ---4 (z - y)) = 1
W11) x - (y - z) = y - (x - z).
84 CHAPTER 4. LUKASIEWICZ 00- VALUED CALCULUS
Proof: W5) By (W2), (1 -+ 1) -+ ((1 -+ x) -+ (1 -+ x)) = 1, and then,

by (Wl), x -+ x = 1 -+ (x -+ x) = 1.
W6) If x -+ y = y -+ x = 1, then by (Wl) and (W3) we have x = 1 -+
x = (y -+ x) -+ x = (x -+ y) -+ y = 1 -+ Y = y.
W7) By (W3), (Wl) and (W5), (x -+ 1) -+ 1 = (1 -+ x) -+ x = x -+
x = 1. From this identity, (W2), (Wl) and (W5) we have 1 = (1 -+
x) -+ ((x -+ 1) -+ (1 -+ 1)) = x -+ ((x -+ 1) -+ 1) = x -+ 1.
W8) By (W2), (W7) and (WI), 1 = (y -+ 1) -+ ((1 -+ x) - (y -
x)) = 1 -+ (x -+ (y -+ x)) = x -+ (y -+ x).
W9) If x -+ y = y -+ z = 1, then by (W2), 1 = (x -+ y) -+ ((y - z) -+
(x -+ z)) = 1 -+ (1 -+ (x -+ z)), and applying (Wl) twice, we obtain
x-+z=1.
WlO) As a preliminary step, we shall prove the following weaker form
of (Wll):
Wll') If x -+ (y -+ z) = 1, then y - (x -+ z) = 1.
Suppose that x -+ (y - z) = 1. Replacing y by y -+ z in (W2) we
get 1 = (x -+ (y -+ z)) -+ (((y -+ z) - z) -+ (x -+ z)) = 1 - (((y -
z) - z) -+ (x -+ z)). Hence, by (Wl) and (W3), ((z -+ y) -+ y) -+
(x - z) = 1. On the other hand, by (W8), y -+ ((z - y) - y) = 1.
Applying (W9) to these identities, with x = y, y = (z - y) - y and
z = x -+ z, we obtain y -+ (x -+ z) = 1. Now (WIO) follows at onee
from (Wll') and (W2).
Wll) By (W3) and (W8), y -+ ((y -+ z) -+ z) = y -+ ((z -+ y) - y) =
1, and by WlO), ((y -+ z) -+ z) -+ ((x -+ (y -+ z)) -+ (x - z)) = 1.
Applying (W9) to the last two identities, we obtain y -+ ((x - (y -+
z)) -+ (x -+ z)) = 1. Hence, by (Wll'), (x -+ (y - z)) -+ (y -+
(x - z)) = 1. By interchanging x and y in this identity we also obtain
(y -+ (x -+ z)) -+ (x -+ (y -+ z)) = 1, and then we can apply (W6) to
obtain (Wll). 0
Lemma 4.2.4 The following equations hold in every W-algebra A:

(i) -d -+ x = 1;
4.2. WAJSBERG ALGEBRAS 85
(ii) -,x = x - -,1;

(m) -,-,x = x;
(iv) x - y = -'y - -,x.
Proof: (i) By (WlO) we have ((-,x - -,1) - x) - ((-,1 - (-,x -
-,1) - (-,1 - x» = 1. Since, by (W1) and (W4), (-,x - -,1) - x =
(-,x - -,1) - (1 - x) = 1 and, by (W8), -,1 - (-,x - -,1) = 1,
whence the desired conclusion follows from (W1).
(ii) By (W8), -,x - (-,-,1 - -,x) = 1. By (W4), (-,-,1 - -,x) -
(x - -,1) = 1. Then from (W9) we obtain -,x - (x - -,1) = 1, and
applying (W11) we have x - (-,x - -,1) = 1. As seen in the proof
of (i), we also have (-,x - -,1) - x = 1. Hence by (W6) we obtain
x = -,x - -,1, and taking into account (W3), together with (i) and
(W1) we get x - -,1 = (-,x - -,1) - -,1 = (-,1 - -,x) - -,x = 1 -
(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.
Proof' (A, EB, -', 0) satisfies MVl), because
x EB Y = -,x - y = -,y - -,-,x = Y EB x,

and also satisfies MV2), as shown by the following identities:
x EB (y EB z) = -,x - (-,y - z) = -,x - (-,z - y)

= -,z - (-,x - y) = z EB (x EB y) = (x EB y) EB z.
The remaining verifications, MV3)-MV6), are easy consequences of
the properties established in the previous lemmas. 0
In what follows, ---.. will denote the binary operation defined on

MV-algebras by formula (4.3). In the light of the above theorem, this
operation satisfies properties (W1) - (W11).
Lemma 1.1.4(iii) gives the following remarkable relation between the
operations ---.. and 0 on an MV-algebra A:
(4.7) For all x, y, Z E A, z 0 x::; y iff z::; x ---.. y.
Definition 4.2.6 An implicative filter of an MV-algebra Ais a subset

F of A satisfying the following conditions:
(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:
(i) F is an implicative filter;

(ii) F =I 0; if x E Fand x ::; y E Athen y E F; if x, Y E F then
x0y E F;
(iii) The set -,F =def {-,x I x E F} is an ideal of A.
Proof: (i) implies (ii): Let F be an implicative filter. By (F1), 1 E F.

Suppose x E Fand x ::; y. Since this last condition is equivalent to
x ---.. Y = 1, (F2) yields y E F. Note that y ---.. (x ---.. (x 0 y» = 1.
Therefore if x, y are both in F, then by (F2) we obtain that x 0 y E F.
(ii) implies (iii): This immediately follows by definition of the op-
eration 0 as given by equation (1.2).
(iii) implies (i): Suppose that -,F is an ideal. Since 0 E -,F, we
have 1 E F. Suppose now that x and x ---.. y are in F. This means that
-,x E -,F and xe y E -,F, and then -,x V -,y = ...,x EB (x e y) E ...,F.
Therefore, -,y E -,F, i.e., y E F. 0
4.3. PROVABILITY 87
The above lemma allows us to obtain properties of implicative filters

from properties of ideals. For instance, if h: A --+ A' is a homomorphism
then the set F(h) =del {x E AI h(x) = I} is an implicative filter;
moreover, h(x) = h(y) Hf (x --+ y) 1\ (y --+ x) E F(h).
Notation: Given an implicative filter F, by abuse of notation we shall
write AI F to denote the quotient algebra AI...,F.
The intersection of any nonempty family of implicative filters of an
MV-algebra A is an implicative filter, and since A itself is an implicative
filter, we can give the following
Definition 4.2.8 For any subset X of an MV-algebra A, the implica-

tive filter V(X) generated by X is the intersection of all implicative
filters containing X.
Proposition 4.2.9 Let A be an arbitrary MV-algebra.

(i) Let s, t E A and Y ~ A. Then t E V(Y U {s}) iff there is an
integer n;::: 1 such that sn --+ t E V(Y);
(ii) For any s, t E A we have t E V( {s}) iff there is an integer n ;::: 1
such that sn --+ t = 1;
(iii) For arbitrary Xl, ... , Xk E A we have
(4.8) V( {Xl,"" xd) = V(XI 0 ... 0 Xk).
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:
(Al) a --+ (ß --+ a);

(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.
Definition 4.3.2 A prooffrom a set e of formulas is a finite string of

formulas al," . ,an, with n ~ 1, such that, for each 1 :::; i :::; n:
(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).
A formula a is provable /rom e, in symbols, e f- a, iff there is a proof

al,' .. ,an from e, such that an = a. Equivalently, we say that a is a
syntactic consequence of e. The set of provable formulas from e shall
be denoted
As usual, we say that formula ß follows by modus ponens from formulas

a and a ~ ß. With this terminology, condition (iii) in the above
definition can be given by the following equivalent reformulation:
(iii / ) there are j, k E {I, ... , i-I} such that ai follows by modus
ponens from aj and ak.
In Definition 4.3.2 the case e = 0 is not excluded. In this case,

condition (ii) becomes vacuous.
Definition 4.3.3 By a proof we shall henceforth me an a proof from

the empty set; by a provable formula we shall mean a formula that is
provable from the empty set. We shall also write f- a in place of 0 f- a.
4.3. PROVABILITY 89
Given a finite sequence of strings of symbols from the alphabet E we

can effectively decide whether such string is a proof. Using some lexi-
cographic ordering, we then see that there exists an effective procedure
to list the set of provable formulas.
Our next aim is to show that the tautologies coincide with provable
formulas; this, together with our discussion in the first sections, will
give a positive solution to the decision problem for the infinite-valued
calculus of Lukasiewicz.
Notation: We shall use CiVß and CiAß as abbreviations of (Ci --+ ß) --+ ß
and ""((""Ci --+ ...,ß) --+ ...,ß), respectively. With this notation, (A3) can
be written as (Ci V ß) --+ (ß V Ci).
Proposition 4.3.4 For all formulas Ci, ß and, we have
(4.9) f- (Ci --+ (ß --+ ,)) --+ (ß --+ (Ci --+ ,))
(4.10) f- Ci --+ Ci
(4.11) f- (ß --+ ,) --+ ((Ci --+ ß) --+ (Ci --+ ,))
(4.12) f- ...,...,a --+ (ß ---t a)
(4.14) f- (a --+ ...,ß) --+ (ß --+ ""Ci)
(4.15) f- (a --+ ...,...,ß) --+ (...,ß --+ ""Ci)
(4.16) f- a --+ ...,...,a.
Proof: To prove (4.9) it suffices to check that the following string of

formulas is a proof. To help the reader, on the right-hand side of each
formula occurring in the string, we shall point out whether this formula
is an axiom, or else the formula follows by modus ponens from preceding
formulas.
al : ß --+ h V ß) = ß --+ (h --+ ß) --+ ß) (Al)
Ci2 : (ß --+ (, V ß)) --+ (((, V ß) --+ (ß V,)) --+ (ß --+ (ß V,))) (A2)
a3 : ((, V ß) -+ (ß V,)) -+ (ß -+ (ß V,)) (MP-al, a2)

a4 : b V ß) -+ (ß V ,) (A3)
as : ß -+ (ß V ,) (MP- a4, a3)
a6 : (ß -+ (ß V,)) -+ (((ß V ,) -+ (a -+ ,)) -+ (ß -+ (a -+ ,))) (A2)
a7 : ((ß V ,) -+ (a -+ ,)) -+ (ß -+ (a -+ ,)) (MP-as, a6)
a8 : (a -+ (ß -+ ,)) -+ (((ß -+ ,) -+ ,) -+ (a -+ ,))
= (a -+ (ß -+ ,)) -+ ((ß V ,) -+ (a -+ ,)) (A2)
ag : a8 -+ (a7 -+ ((a -+ (ß -+ ,)) -+ (ß -+ (a -+ ,)))) (A2)
alO : a7 -+ ((a -+ (ß -+ ,)) -+ (ß -+ (a -+ ,))) (MP-a8, ag)
an : (a -+ (ß -+ ,)) -+ (ß -+ (a -+ ,)) (MP- a7, alO)
This settles (4.9). To prove (4.10) note first that, taking , = a in

(4.9), we obtain
I- (a -+ (ß -+ a)) -+ (ß -+ (a -+ a)).
If al,"" an = (a -+ (ß -+ a)) -+ (ß -+ (a -+ a)) is a proof, then
is also a proof. Hence taking any axiom as ß, we obtain that the

sequence al,' .. , an> an+l, an+2, a n+3 = ß, a n+4 = a -+ a is a proof of
a -+ a.
To prove (4.11), as a particular case of (4.9) we have
I- {) = ((a -+ ß) -+ ((ß -+ ,) -+ (a -+ ,))) -+
((ß -+ ,) -+ (( a -+ ß) -+ (a -+ ,))).
If al, .... an = {) is a proof, then the string

4.3. PROVABILITY 91
O:n+2 = (ß --+ I) --+ ((0: --+ ß) --+ (0: --+ I))
is also a proof. This settles (4.11).

To prove (4.12) it is enough to check that the following string of
formulas is a proof:
(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) --+
(ß - -'0)) - ((0 - -.ß) - (ß - -'0))) (A2)

on+4 : ((-'-'0 - -.ß) - (ß - -'0)) -
((0 - -.ß) - (ß - -'0))
(A4)
a n +6 : (0 - -.ß) - (ß - -.a) (MP- 0n+5, On+4).
To obtain (4.15) it is sufficient to replace ß by -.ß in (4.14).
Finally, to prove (4.16), replacing 0 in (4.14) by -.a and ß by 0,
we get f- (-.a - -'0) - (a - -'-'0). Combining now this result with
(4.10), in which 0 is replaced by -'0, we get the desired conclusion. 0
From (Al) and modus ponens we immediately obtain

(4.17) If f- 0 then, for each ß, f- ß - o.
4.4 Lindenbaum algebra

Theorem 4.4.1 Let the binary relation = on Form be defined by
o = ß iff f- 0 - ß and f- ß - o. Then = is an equivalence relation,
called syntactic equivalence , satisfying the following conditions:
(4.18) If a = 'Y and ß = 0 then (0 - ß) = (-y - 0);
(4.19) If 0 =ß then =-.ß.-'0
Proof" It is obvious that 0 = ß implies ß = 0 and, by (4.10), 0 = o.

By (A2), {o - ß,ß - 'Y} f- 0 - 'Y and {'Y - ß,ß - o} f- 'Y - o.
Hence, = is transitive, and we have shown that = is an equivalence
relation on Form. By (A2), for any formulas 0, ß and 'Y we have
{'Y - o} f- (a - ß) - ('Y - ß)
and
{o - 'Y} f- ('Y - ß) - (0 - ß).
Therefore,
(4.20) If a ='Y then 0 - ß ='Y - ß·

4.4. LINDENBAUM ALGEBRA 93
Analogously, from (4.11) we obtain
(4.21) If ß =0 then 'Y -+ ß =: 'Y -+ o.

Now (4.18) follows from (4.20), (4.21) and the transitivity of ==.
By (4.13) and (4.16), for every formula ß, ß == "ß. Hence by
(4.21), for arbitrary formulas a and ß, a -+ ß == a -+ "ß. Therefore,
r (a -+ ß) -+ (a -+ "ß). From this result and (4.15) we obtain
r (a -+ ß) -+ (,ß -+ ,a). Since by (A4), r (,ß -+ ,a) -+ (a -+ ß),
we have (4.19). 0
Notation: The equivalence dass of formula a with respect to syntactic

equivalence shall be denoted by lai. In symbols,
lai =def {ß E Form I ß - a}.
Lemma 4.4.2 For each formula a, a E 0f- iJJ lai = 0f-.

Prool If lai = 0f- then a E 0f- because a E lai. Conversely, suppose
a E 0f-, Le., r a. Take ß E 0f-. By (4.17), ß == a. Hence 0f- ~ lai.
Now take ß E lai. Then r (a -+ ß) and, by modus ponens, r ß. Hence
lai ~ 0f-. 0
Theorem 4.4.3 The quotient set Form/=: becomes a Wajsberg alge-

bra, once equipped with the operations -+ and ' and the constant 1 as
given by the following stipulations:
(4.22) lai -+ IßI =def la -+ ßI

(4.23) ,lai =def I,al
(4.24) 1 =def 0f-.
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 =
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
By Theorem 4.2.5 we get
Corollary 4.4.4 The quotient set Form/ = becomes an MV-algebra

with the operations. and Ee and the constant 0 defined by
(4.25) ·Ial =def I·al

(4.26) lai EeIßI =def I·a --+ ßI
(4.27) 0 =def .0f- = {a E Form I there is ß E 0f- such that a =·ß}·
o
The MV-algebra
.c =def (Form/=, 0", Ee)
is called the Lindenbaum algebra 0/ Lukasiewicz infinite-valued propo-
sitional calculus.
4.5 All tautologies are provable

Lemma 4.5.1 Each /ormula provable from a set e 0/ /ormulas is also
a semantic consequence 0/ this set. In symbols,
Proof: Let v: Form --+ [0,1] be a valuation such that v(a) =

1 for all
a E e. By induction on n we shall prove that if al," . ,an is a proof
from e, then v( an) = 1.
If n = 1, then al is either an axiom or it belongs to e. In the
second case, v(ad = 1 by the hypothesis on v. In the first case we also
4.5. ALL TAUTOLOGIES ARE PROVABLE 95
have lI(al) = 1 because all axioms are tautologies, as shown by direct

inspection of (Al) - (A4).
Let n > 1 and suppose that, for each proof from 8, ßl,"" ßm,
with m < n, we have lI(ßm) = 1 (induction hypothesis). Let al,' .. , an
be a proof from 8. If an is not an axiom and does not belong to 8,
then there are i, j E {I, ... , n} such that aj coincides with the formula
(ai - an). Since both al,' .. ,ai and al, ... , aj are proofs from 8, by
induction hypothesis, lI(ai) = lI(aj) = 1. Therefore,
1 = lI(aj) = 1 - lI(an) = lI(a n).
o
Thus, in particular, all provable formulas are tautologies. The con-
verse follows from Chang's completeness theorem:
Theorem 4.5.2 Every tautology is provable. Thus, tautologies eoin-
eide with provable formulas, in symbols, 0F = 0'r-.
Proof" First of all , for each propositional variable Xi, the syntactic
equivalence class lXii is an element of the Lindenbaum algebra C. Let
a be a formula with Var( a) ~ {Xit , ... , X in }. By an easy induction on
the number of connectives in a, it follows that aC(IXi11, ... , IXin I) =
lai. Thus, in case a E Form is not provable, then by Lemma 4.4.2
and (4.24), lai =11, whence aC(IXitl,.··, IXinl) =11. Stated otherwise,
the Lindenbaum algebra C does not satisfy the equation a = 1. Hence
by Chang's completeness theorem 2.5.3, the MV-algebra [0,1] does not
satisfy this equation, Le., a is not a tautology. The rest follows from
Lemma 4.5.1. 0
Corollary 4.5.3 There is an effeetive proeedure (for definiteness, a
Turing maehine) deeiding whether an arbitrary formula is a tautology
in the infinite-valued ealeulus of Lukasiewiez.
Proof: By the above theorem, together with the remark following Def-
inition 4.3.3, tautologies can be effectively enumerated. The effective
enumerability of nontautologies was shown in Proposition 4.1.5. 0
This result shall be considerably strengthened in a later chapter,
where we shall give a more accurate estimate of the number of Thring
steps needed to decide whether a formula is a tautology.
Corollary 4.5.4 The relations of semantic and syntactic equivalence

coincide: thus, for arbitrary formulas a and ß, a[O,lj = ß[O,lj iff
a=ß·O
Following tradition, whenever two formulas are syntaetieally (= se-

mantically) equivalent, we shall heneeforth say that they are logically
equivalent.
Proposition 4.5.5 Up to isomorphism, the Lindenbaum algebra L co-

incides with the free MV-algebra over the generating set {IXoI, lXII, ... }
of logical equivalence classes of propositional variables.
Proof: Reealling the terminology introdueed in Seetion 3.1, let us de-

fine the map c.p: L - t Term([O, 1], w) by the stipulation c.p(lal) = a[O,lj.
Then Corollary 4.5.4 implies that c.p is an isomorphism of the Linden-
baum algebra L onto the term algebra Term([O, l],w). In particular,
the rest riet ion of c.p to the set of logical equivalenee classes of proposi-
tional variables yields a bijeetion from this set onto the set of projee-
tion functions {-lro, 'lr1, ... }. Henee, the desired eonclusion follows from
Proposition 3.1.4. 0
Lemma 4.5.6 Let A be an MV-algebra and v: Form -t A be an A-

valuation. Then the stipulation
defines a homomorphism h v : L - t A. Conversely, for each homomor-

phism h: L - t A, the stipulation vh(a) = h(lal) defines an A-valuation
VA: Form - t A. Moreover, the correspondence v ~ h ll is a one-one
mapping from the set of A-valuations onto the set of homomorphisms
into A of the Lindenbaum algebra L. The inverse mapping is given by
h ~ Vh.
Proof: Sinee ß E lai implies v(ß) = v(a), the stipulation hll(lal) =

v(a) defines a mapping h v : L - t A. The rest of the proof is obvious.
o
4.6. SYNTACTIC AND SEMANTIC CONSEQUENCE 97
4.6 Syntactic and semantic consequence

By Lemma 4.5.1, for any set e of formulas we have the inclusion er- ~
eF. By Theorem 4.5.2, 0F = 0r-. As we shall see in this seetion, in
general, eF =I- er-; we shall give neeessary and suffieient eonditions for
eF to eoincide with er-.
Definition 4.6.1 A theory of Lukasiewiez infinite-valued proposi-
tional ealeulus is a set e of formulas satisfying the following eonditions:
(Tl) All axioms belong to e;

(T2) If a E e and (a -+ ß) E e, then ß E e.
Proposition 4.6.2 The following conditions hold for each set e of
formulas:
(i) er- is the smallest theory containing e;

(ii) e is a theory iff e = er-;
(iii) 1f e is a theory and a E e, then lai ~ e.
Praof: (i) It follows at onee from Definition 4.6.1 that er- is a theory.
Suppose that :E is a theory and e ~ :E. If al, ... , an is a prooffrom e,
arguing as in the proof of Lemma 4.5.1, we see that an E :E. Therefore,
er- ~ :E. The proof of (ii) trivially follows from (i). To prove (iii), note
that if ß E lai then a -+ ß E er- and, by (T2), ß E e. 0
Let {eihEI be a nonempty family of theories. It is easy to see that
niEI ei is a theory; furt her , from the above proposition we immediately
get that (UiEI eit
is the smallest theory eontaining ei for all i E 1.
Thus, the set of theories of Lukasiewicz infinite-valued calculus, ordered
by inclusion, is a eomplete lattiee. We shall denote this lattice by Theo.
Comparing Definitions 4.2.6 and 4.6.1 we immediately get
Theorem 4.6.3 The correspondence
(4.28) e 1-+ lei =def {lai E J:, I a E e}

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.
Proposition 4.6.4 For arbitrary formulas a and ß, and for each e~

Form we have
ßE(8U{a}t
ij, and only ij, there is an integer n ~ 1 such that (an - ß) E 8f-.
Prool Since ß E (8U{a}t iff IßI E 1(8U{a}tl, the result follows

from Proposition 4.2.9 and 4.6.3. 0
Corollary 4.6.5 ß E {at iJJ there is an integer n > 1 such that

an _ ß is a tautology.
Prool By Proposition 4.2.9(ii). 0

The next theorem gives the desired necessary and sufficient condi-
tion for 8 F to coincide with 8f-.
Theorem 4.6.6 For each 8 ~ Form, 8 F = 8f- iJJ 18f-1 is an inter-

section of maximal implicative filters of the Lindenbaum algebra C.
Proof: For each valuation v: Form - [0,1] we have
where h v : C - [0,1] is the homomorphism defined in Lemma 4.5.6. We

next observe that whenever a does not belong to 8 F there exists a
valuation Va: Form - [0,1] such that va(ß) = 1 for all ß E 8 and
va(a) < 1. Therefore,
8F = n vc:- ({1}),
a~eF
1
4.6. SYNTACTIC AND SEMANTIC CONSEQUENCE 99
whence, by Theorem 4.6.3 and equation (4.30),
Consequently, if e~ = e~, then le~1 is an intersection of maximal

implicative filters of .c.
Conversely, suppose there is a family {MihEI of maximal im-
plicative filters of .c such that le~1 = niEI Mi. Since for each i E I
the quotient MV-algebra .clMi is simple, there is a homomorphism
hi:.c- [0,1] such that Mi = hi 1 ({1}). By Lemma 4.5.6, for each
i E I there is a valuation Vi: Form - [0,1] such that hi = h Vi • Hence,
again using Theorem 4.6.3 and (4.30), we conclude that
e~ = n
iEI
Vi- 1 ( {1 }).
Thus, whenever a <t. e~ there is j E I such that Vj (a) < 1. It follows

that a <t. e~. In conclusion, e~ ~ e~, and by Lemma 4.5.1, e~ = e~,
as required. 0
Every implicative filter F is an intersection of maximal implicative
filters if, and only if, -,F is an intersection of maximal ideals. Since by
Theorem 4.4.1, .c ~ Free w , the example given in Chapter 3 of an ideal
of Free w which is not an intersection of maximal ideals, immediately
yields an example of a set e ~ Form such that e~ ~ e~.
On the other hand, if e is a nonempty finite set of formulas, say
e = {al, ... ,ak}, from (4.8) together with Theorem 4.6.3, it follows
that
e~ = {al 0 ... 0 akt·
Then le~1 is a principal (Le., a one-generated) implicative filter of .c.

An application of Theorems 3.4.9 and 4.6.6 yields
Theorem 4.6.7 For each finite set e ~ Form, e~ = e~. 0
Definition 4.6.8 For each theory e, the MV-algebra .cIlei is called

the Lindenbaum algebra ofe, and is denoted by .c(e).
Note that .c ~ .c/10'·l Therefore,.c is the Lindenbaum algebra

of the theory 0r-. From Theorem 4.6.6, together with our criterion for
semisimplicity, (Theorem 3.4.9), for any set of formulas 8 we also have:
8r- = 81= Hf .c( 8) is semisimple.
Theorem 4.6.9 Up to isomorphism, every countable MV-algebra A is

the Lindenbaum algebra 01 same theory in the infinite-valued calculus
01 Lukasiewicz built /rom the variables X o, Xl, ....
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.
The logical counterpart of the ward problem lor finitely presented

MV-algebras, mentioned in the Introduction, has the following uni-
formly positive solution, yielding a generalization of Corollary 4.5.3:
Theorem 4.6.10 There is a Turing machine U such that, over any

input (8, 'I/J), where 8 is a finite set ollormulas and'I/J is a lormula, U
decides in a finite number 01 steps whether or not 'I/J is a syntactic (=
semantic) consequence 018.
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,
(ii) letting Je and JCP be the McNaughton functions corresponding to B

and 'P via Propositions 3.1.8 and 4.5.5, we have fe(x) = 1 and JCP(x) <
1.
By Theorem 4.6.7, the existence of such rational point x is a suffi-
cient condition for 'P not to be a consequence of B.
Claim. The existence of x E [0, l]m n Q is also a necessary condition
for 'P not to be a consequence of B.
As a matter or fact, assurne 'P is not a consequence of B, and let
Y E [O,I]m be an m-tuple of real numbers such that Je(Y) = 1 and
JCP(Y) < 1. It is sufficient to argue in case m = n. Let jcp be the
restriction of Jcp to the set
Z =def {z E [0, Ir I Je(z) = I}.
By Proposition 3.1.8 JCP is a continuous, piecewise linear function, each
piece having integer coefficients. Arguing as in Proposition 3.3.1, we
see that the minimum value of jcp is attained at some point Z E Z with
rational coordinates. Our claim is settled.
To complete the proof, from the above list of triplets (x, B, 'P) it is
now easy to construct the required effective enumeration of all pairs
(B, 'P) such that 'P is not a consequence of B. A routine argument now
yields the desired Turing machine U deciding whether 'P is a conse-
quence of B. 0

The paper [139] is a basic reference for Lukasiewicz infinite-valued cal-
culus. The equivalence between MV-algebras and Wajsberg algebras
was first proved by Rodriguez in [212] (see also Font, Rodrfguez and
Torrens [86], and Komori [130]).
The crucial distinction between syntactic and semantic consequence
relations in the infinite-valued ca1culus is due to W6jcicki [243] and [244,
Theorem 4.3.4]. The fact that the two notions coincide for finitely
axiomatizable theories was first proved by Hay in [113]. A related
result by Rose is in [215, Lemma 1]. A geometrie visualization of the
two notions, using the differential structure of McNaughton functions
was given in [170].
For the proof of the completeness theorem, following Chang [38]

we use examples of proofs already present in [216]. Historically, the
original list ofaxioms conjectured by Lukasiewicz to be complete for
the infinite-valued calculus also included the axiom
whose redundancy was proved simultaneously by Meredith [154] and

Chang [37].
Chapter 5
Ulam's game
The crucial problem of interpreting n truth values when n > 2 was

investigated, among others, by Lukasiewicz hirnself. As shown in this
chapter, a simple interpretation is given by Ulam game, the variant of
the game of Twenty Questions where n - 2 lies, or errors, are allowed
in the answers. The case n = 2 corresponds to the traditional game
without lies. The game is originally described by Ulam on page 281 of
his book [235] as follows:
Someone thinks of a number between one and one million (which is
just less than 220). Another person is allowed to ask up to twenty
questions, to each of which the first person is supposed to answer only
yes or no. Obviously the number can be guessed by asking first: Is
the number in the first half million? then again reduce the reservoir
of numbers in the next quest ion by one-half, and so on. Finally the
number is obtained in less than log2(lOOOOOO). Now suppose one were
allowed to lie once or twice, then how many quest ions would one need
to get the right answer?
5.1 Questions and answers

Let us inspect a round of Ulam game: initially, the two players agree to
fix a nonempty finite set S of numbers, called the search space, and an
integer m 2: O. Then the first player chooses a number x E S, and the
second player must find the unknown x, by asking the smallest possible
number of questions, to each of which the first player can only answer
103
104 CHAPTER 5. ULAM'S GAME
"yes" or "no", being allowed at most m errors/lies in his answers. By

definition, a question is a subset of S: thus for instance, the quest ion
is x an even number ?
is identified with the set of all even numbers in S.
We can safely assurne Pinocchio to be the first player, and identify
ourselves with the second player. Pinocchio's answers are propositions
of either form "yes, (x is even) " , or "no, (x is odd)". Dur state of
knowledge about x is uniquely determined by the conjunction of these
answers. The latter, in general, do not obey the rules of classicallogic,
for
(a) The conjunction 01 two equal answers to the same repeated question
need not be equivalent to a single answer. Thus, the classical idempo-
tence principle fails. To see this, let us assurne that Pinocchio can lie at
most once. Suppose we ask twice the following quest ion "is x even?".
If the answer is "yes" in both cases, then x must be even. However,
after the first answer we are not certain that x is even.
(b) The conjunction 01 two opposite answers to the same repeated ques-
tion need not lead to contradiction. To see this, again using the above
example, if Pinocchio answers "yes" to the first question, and "no" to
the second, then we cannot conclude that S is empty.
In the traditional error-free game, our knowledge on x is usually
represented by a function # : S ~ {O, I} giving, for every z E S, the
number a = #z of falsified answers, where all a ;::: 1 are collapsed to
1. Stated otherwise, for each z E S, #z = 0 Hf z does not falsify any
answer; #z = 1 iff z falsifies at least one answer.
In Ulam game with m lies, our knowledge is also given by a function
Cl : S ~ {O, 1, ... , m, m + I}, counting the number a = Cl(Z) of answers
falsified by z, where all a ;::: m + 1 are collapsed to m + 1: thus Cl(Z)
= 0,1, ... , m or m + 1, according as z falsifies 0,1, ... , m or ;::: m + 1
answers, respectively.
5.2 Dynamics of states of knowledge

Let us assurne that, after receiving a certain number of answers from
Pinocchio, our knowledge is represented by the function Cl : S ~
5.2. DYNAMICS OF STATES OF KNOWLEDGE 105
{O, ... , m + I}. Let D be an arbitrary subset of S, and D = S\D its

complementary subset. Suppose our quest ion "does x belong to D?",
now receives from Pinocchio the answer "yes". Then how should we
represent our new knowledge 0" ? Let the (m+2)-tuple (So, ... , Sm+l),
be defined by
Si =def {Z E S I o'(z) = i},
(i = 0, ... , m + 1). There is an obvious correspondence between 0' and
(So, . .. , Sm+l), generalizing the correspondence between subsets and
characteristic functions. The knowledge 0" = (Sb, ... ,S:n+l)' obtained
from 0' as a consequence of the positive answer to question D is given
by
(5.1) Sb = So n D,
and, for each i = 1, ... , m,
(5.2) S; = (Si n D) U (Si-l n D).
As a matter of fact, whenever y E S is a valid candidate for the unknown

x, the number o'(Y) of answers falsified by y will remain constant iff y
satisfies D iff y E D; on the other hand, o'(Y) must be increased by
one iff y falsifies D iff y rt. D. One can similarly deal with the negative
answer to question D; such answer has precisely the same effect as a
positive answer to the opposite quest ion D.
Formulas (5.1) and (5.2) describing the dynamics of Ulam game
acquire a particularly simple form if, instead of assigning to each y E S
the number o'(Y) of falsified answers, we assign the truth value r(y)
given by
(5.3) r(y) =def 1 - O'(y)/(m + 1).
Intuitively, the truth value r(y) measures, in units of m + 1, how

distant y is from the condition of falsifying too many answers. More
106 CHAPTER 5. ULAM'S GAME
precisely,
0, if y falsifies ;::: m +1 answers
1/ (m + 1), if y falsifies m answers
(5.4) r(y) =
m/(m + 1), if y falsifies one answer
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
Proposition 5.2.1 Let t ~ 1 be an integer. For every i = 1, ... ,t, let

D i ~ Sand b( i) E {yes, no}. Let r be the state 0/ knowledge arising
fw, b(l) b(2) b(t)
Jlom the sequence 0/ answers D 1 ,D2 , ... , D t . Then
r = Db(l)
1
0 ...
0 Db(t)
t,
where 0 is pointwise Lukasiewicz conjunction in the MV-algebra Lm +2 •

o
5.3 Operations on states of knowledge

Onee equipped with pointwise Lukasiewicz eonjunetion, the set KS,m
of states of knowledge in Ulam game over the seareh spaee S with m
lies/ errors beeomes an abelian monoid with neutral element 1 (the ini-
tial state). While Pinoeehio's answers may be false/erroneous, their
eonjunetion is all we know ab out the unknown number x. As the num-
ber of answers inereases, our state of knowledge beeomes sharper and
sharper. To aeeount for the natural order between states of knowledge,
let us agree to write
r ' ~ r"
(read r ' is sharper than r", or r" is coarser than r ' ) iff r ' (y) ~ r" (y)
for all y E S.
For every state r E KS,m there is a eoarsest state -,r that is incom-
patible with r, in the sense that r 0 -,r = O. Naturally, -,r = 1 - r.
We ean regard -, as a form of negation, without any fear of eonfusion
with (5.5) and (5.6). In partieular, our analysis in Seetion 5.1 (b) does
not apply to the -, operation.
Using the operations -, and 0 we ean express the natural order
between states of knowledge, by writing r 0 -'(7 = 0 instead of r ::; (7.
Sinee the operations 0, -', E9 and 1, -', 0 are mutually interdefinable,
by a slight abuse of terminology, let us refer to fes,m = (Ks,m, 1, -', 0)
as the MV-algebra of states 0/ knowledge in Ulam game over S with m
lies.
We say that an equation </>(X1! . .. , X n ) = 'IjJ(X1 , ... ,Xn ) is absolute
iff it is valid whenever the variables Xi are replaeed by arbitrary states
Chapter 6
Lattice-theoretical properties
In this chapter we study properties that are strongly related to the

lattice structure of MV-algebras. We start by considering relations
between the ideals of an MV-algebra A and the ideals of the lattice
L(A). A stonean ideal of a bounded distributive lattice L is an ideal
generated by complemented elements of L. We shall show that the
minimal prime lattice ideals of L(A), as weIl as the stonean ideals of
L(A), are always ideals of A.
An MV-algebra A is called hyperarchimedean iff all its ideals are
stonean ideals ofL(A). Hyperarchimedean MV-algebras have many fea-
tures in common with boolean algebras. For instance, an MV-algebra
is hyperarchimedean iff all its prime ideals are maximal. Further, ev-
ery hyperarchimedean MV-algebra A is semisimple, and so is every
homomorphic image of A. We shall describe the most general hyperar-
chimedean MV-algebra with one generator.
We shall also investigate the role of the complemented elements
of L(A) in the direct product decompositions of A, and we shall intro-
duce boolean products of MV-algebras as generalizations of finite direct
product decompositions.
An MV-algebra is said to be complete iff its underlying lattice is
closed under infinite sups and infs. The structure of complete MV-
algebras explicitly depends on the underlying lattice structure, and
every complete MV-algebra A is semisimple. We shall also prove that
every complete MV-algebra A has a decomposition A = Al X A 2 X A 3 ,
where Al is complete and atomic, A 2 is apower of [0,1], and A 3 is
111
112 CHAPTER 6. LATTICE-THEORETICAL PROPERTIES
complete and without linear factors.
6.1 Minimal prime ideals

As usual, by an ideal 0/ a lattice L with 0 and 1 we mean a subset I
of L satisfying the following conditions:
(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.
For every element z E L, the principal ideal ( z 1 is defined by
(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.
Lemma 6.1.1 Let J be an ideal 0/ an MV-algebra A. Then J is a

prime ideal 0/ A iff J is a prime ideal 0/ the underlying lattice L(A).
Proo/: Suppose that J is a prime ideal of the MV-algebra A and let

x, y E A be such that x A y E J. By hypothesis we can safely assurne
x9y E J. Hence (xAY)EB(x9Y) E J. Moreover, from (xAY)EB(x9Y) =
(xEB(x9Y))A(yEB(x9Y)) = (xEB(x9y))A(xVy) 2: xA(xVy) = x,
we get x E J. Thus, J is a prime ideal of L(A).
Conversely, assurne J to be a prime ideal of L(A), and let x, y E A.
Since by Proposition 1.1.7, (x 9 y) A (y 9 x) = 0 E J, then we must
either have x 9 y E J or y 9 x E J, and, since by assumption J is an
ideal of A, then J is a prime ideal in A. 0
6.1. MINIMAL PRIME IDEALS 113
Definition 6.1.2 An ideal K of an MV-algebra A (resp., of L(A)) is

said to be minimal prime iff it is a prime ideal of A (resp., of L (A) ),
and whenever I ~ K is a prime ideal of A, (resp., ofL(A)), then 1= K.
Theorem 6.1.3 Let A be an MV-algebra, and J be a proper ideal of

the underlying lattice L(A). For each z E A let
Jz = {x E A I zex ~ J} and K = K(J) = n{Jz IZ E A\J}.

Then K(J) is an ideal of A and K(J) ~ J. Ij, in addition, J is a
prime ideal of L(A), then K(J) is a prime ideal of A.
Proof: Since 1 E A\J, K is weH defined. We shall prove that K is

an ideal of the MV-algebra A. Thivially, 0 E K. If t ::; u E K, then
t E K. Indeed, for each z E A\J from z e u ~ J we get a fortiori,
z e t ~ J (because, z e t ;::: z e u). Thus t E K, as required. Assurne
now x, y E K, with the intent of proving x EIl y E K. Pick an element
z E A\J. Then from x E K we get z ex E A\J, and hence y E K
implies (z e x) e y ~ J. Thus,
z e (x EIl y) = z 0 (....,x 0 .y) = (z e x) e y ~ J,

whence x EIl y E K. We have thus shown that K is an ideal of A.
In order to prove that K ~ J, it is enough to observe that Z e Z =
OE J; thus, whenever Z ~ J, we must have z ~ K.
Claim: For each Z E A, if x 1\ y E J z , then either x E Jz or y E Jz.
As a matter of fact, we have the identities
(z 0 .x) V (z 0 .y) = z 0 (....,x V ....,y) = z e (x 1\ y) ~ J,

and since J is an ideal of L(A), we must either have z e x ~ J or
z e y ~ J, as required to settle our claim.
To conclude the proof, assurne J to be a prime ideal of L(A). As-
sume x ~ K and y ~ K. Then there exist elements s, t E A \ J such
that x ~ Js and y ~ Jt . By our assumption about J, s 1\ t ~ J. If
x 1\ Y E K (absurdum hypothesis) then x 1\ y E JSl\t and hence, by our
claim, we must either have x E JSl\t or y E JSl\t. From the inclusion
JSl\t ~ Js n Jt we obtain a contradiction in either case, thus showing
that x 1\ y ~ K. We conclude that K ~ J::/: A is a prime ideal of A. 0
Corollary 6.1.4 Let A be an MV-algebra and L(A) its underlying

lattice. Then the set of minimal prime ideals of A coincides with the
set of minimal prime ideals ofL(A).
Proof: This is an immediate consequence of Theorem 6.1.3, together
with Lemma 6.1.1. D
Let us recall that a dual ideal, or filter, of a lattice L is a subset F of
L satisfying the following conditions:
(Fl) 1 E F;
(F2) If xE Fand x ~ y, then y E F
(F3) If x E Fand y E F, then x A y E F.
Theorem 6.1.5 Let J be a minimal prime ideal of an MV-algebra A.
Then for every s E J there exists t E A \ J such that t A s = O.
Proof" By Corollary 6.1.4, J is a minimal prime ideal of L(A). Let
F = {y E A I there is x E A \ J such that sAx ~ y}. Then F is a filter
of L(A) and s E F. Moreover, for every x E A\J we have x E F, in
symbols,
(6.2) A\J ~ F.
Claim: If 0 fj. F then there exists a prime ideal P of L(A) such that
pnF=0.
As a matter of fact, by our standing assumption, the family of all
ideals of L(A) that are disjoint from F is nonempty. Applying Zorn's
lemma, let P be maximal in this family.. Then P will be a prime ideal of
L(A). For otherwise, letting a, b fj. P and a A bE P, by our assumption
about P there are p, q E P such that p V a E Fand q V bE F. Then
(p V a) A (q V b) E F. On the other hand, by distributivity,
(p V a) A (q V b) = (p A q) V (p A b) V (a A q) V (a A b) E P,
thus contradicting the disjointness of P and F. In conclusion, P is

prime and the claim is settled.
By (6.2) we now get P ~ J and s E J\P, thus contradicting the
fact that J is a minimal prime ideal of L(A). It follows that 0 E F,
whence there exists t E A \ J such that s A t = 0, as required. D
6.2. STONEAN IDEALS AND ARCHIMEDEAN ELEMENTS 115
6.2 Stonean ideals and archimedean ele-

ments
Definition 6.2.1 Let L be a lattice with 0 and 1. An ideal I of L
is said to be stonean Hf for every x E I there is a boolean element
Z E InB(L) such that x ~ z.
Stated otherwise, an ideal I of L is stonean iff it is generated by

the ideal I n B (L) in the boolean algebra B (L). For every x E L, the
ideal (x] is stonean iff x E B (L ). The following is a generalization of
Corollary 1.5.6.
Corollary 6.2.2 For any MV-algebra A, every stonean ideal of the

underlying lattice L(A) is automatically an ideal of A.
Proof: Let J be a stonean ideal of L(A). We have only to show

that J is closed under the EB operation. Let x, y E J. Then there are
s, tE JnB(A), such that x ~ sand y ~ t. Since by Theorem 1.5.3(iv),
x EB Y ~ s EB t = sV t E J n B(A), we obtain x EB y E J. 0
Definition 6.2.3 An element x of an MV-algebra A is said to be ar-

chimedean iff there is an integer n 2: 1 such that x EB ... EB x (n times)
is boolean.
Notation: Throughout this chapter, following the notation of (2.7) we

shall write
n.x as an abbreviation of x EB ... EB x (n times).
Corollary 6.2.4 Por every element x of an MV-algebra A, the follow-

ing conditions are equivalent:
(i) There is an integer n 2: 1 such that ...,x V n.x = 1;
(ii) There is an integer n 2: 1 such that n.x = (n + l).x;

(iii) x is archimedean;
(iv) (x) is a stonean ideal ofL(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 Hyperarchimedean algebras

Definition 6.3.1 An MV-algebra A is said to be hyperarchimedean
iff all its elements are archimedean.
Equivalently, by Corollary 6.2.4(ii), for every x E A there is an integer

n ~ 1 such that n.x = (n+1).x; one then immediately sees that boolean
algebras, finite MV-algebras and simple MV-algebras are examples of
hyperarchimedean MV-algebras.
If the MV-algebra A is a chain, then B(A) is the two-element set
{O, I}. In this case, a nonzero element x is archimedean iff 1 = n.x for
some integer n ~ 1. Then by Theorem 3.5.1, an MV-chain is hyperar-
chimedean iff it is isomorphie to a subalgebra of [0,1].
While the class of hyperarchimedean MV-algebras is closed und er
subalgebras, finite products and homomorphic images, this class is not
closed under infinite products. As a matter of fact, let
(6.3) A = II Ln·
n~2
Then each Ln is hyperarchimedean, but the product MV-algebra A is

not. To see this, let f : {2, 3, 4, ... } - t Un>2 Ln be defined by f(n) =
l/(n - 1). From 0< n.f(n + 2) = nf(n +-2) = n/(n + 1) < 1, we get
6.3. HYPERARCHIMEDEAN ALGEBRAS 117
n.f f/. B(A), for each n = 0,1,2, ... , whence f is a nonarchimedean

element of A, and A is not hyperarchimedean.
Theorem 6.3.2 For any MV-algebra A the following conditions are

equivalent:
(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;
(v) Every ideal of A is an intersection of the maximal ideals contain-

ing it;
(vi) For every ideal J of A, AI J is semisimple.
Proof: (i):=;. (ii): Let J be an ideal of A, and x E J. By Corollary

6.2.4(iii) there is an integer n > 0 such that n.x is a boolean element.
Since x ~ n.x E J, J is a stonean ideal of L(A).
(ii) :=;. (iii): Let P be a prime ideal in A and P ~ J for some ideal J
in A. Let Po = pnB(A), J o = JnB(A). Suppose x E J\P (absurdum
hypothesis ). Since by hypothesis J is stonean, there is z E Jo such that
x ~ z; it follows that z f/. Po. Since P is prime, it is easy to see that
Po is a maximal ideal in the boolean algebra B(A). Since Jo strictly
contains Po, we must have 1 E Jo, whence J 2 J o is not a proper ideal,
a contradiction showing that P is maximal.
(iii) <=? (iv): Thivial.
(iv) :=;. (i): Let 0 =1= s E A, and J.L = {x E A I x 1\ s = O}. Then J.L
is an ideal of A. As a matter of fact, skipping all trivialities, whenever
x 1\ s = 0 = y 1\ s, then (x EB y) 1\ s = 0; this is easily verified on all MV-
chains, and hence by the subdirect representation theorem, it holds in
general. Let K be the ideal generated by J.L and s. By (1.9) we have
K = {x E A Ix ~ n.s EB a, for some n = 1,2, ... and a E J.L}.
Assurne K to be a proper ideal (absurdum hypothesis). Then, by
Corollary1.2.14 there is a prime ideal J in A containing K. Since
by hypothesis, J is minimal, an application of Theorem 6.1.5 yields an

tt
element tE A\J such that s A t = O. Since J;2 K ;2 J1.., then t J1..,
a contradietion. We have shown that 1 E K. Thus, there is an integer
n ;::: 1 and an element a E J1.. such that n.sEBa = 1. From aAs = 0 and
->n.S ~ a it follows that -m.sAs = 0, whence 1 = n.sV-,s. By Corollary
6.2.4(i), s is an archimedean element of A, and A is hyperarchimedean.
(iii) {::} (v): This is a consequence of Corollary 1.2.14.
(v) {::} (vi): By Lemma 3.6.6. 0
By Corollary 3.6.8, since each hyperarchimedean MV-algebra A is
semisimple, there is a compact Hausdorff space X such that A is iso-
morphie to a separating subalgebra of Cont(X).
Proposition 6.3.3 Let X be a compact Hausdorff space. A separating

subalgebra A of Cont(X) is hyperarchimedean iff ZU) = f-l( {O}) is
an open subset of X for each f E A.
Proof" Suppose that A is hyperarchimedean, and for each x EX, let

Ox denote the ideal given by all functions in A that vanish over some
open neighbourhood of x. The fact that A separates points implies that
VOx = {x}, and then, from Theorem 3.4.3(iii) and Theorem 6.3.2(vi) it
follows that Ox = Jvox = J x' Therefore, whenever f E A vanishes in
some point x E X, then f also vanishes over an open neighbourhood of
x. This property clearly implies that ZU) is an open subset of X for
each f E A.
To prove the converse, we need the following
Claim: Let J be an ideal of A, and let f E A. If f vanishes over so me
open set U;2 VJ , then f E J.
As a matter of fact, since VJ n (X \ U) = 0, for each x E X \ U
there is a function gx E J such that gx(x) > O. Since X \ U is a
compact subset of X, arguing as in the proof of Theorem 3.4.3(ii), we
can find a function 9 E J such that g(x) = 1 for each x E X \ U. Since
f(x) = 0 ~ g(x), for all x E U, and f(x) ~ 1 = g(x), for all x E X \ U,
then f ~ g. We conclude that f E J, as required to settle our claim.
Suppose now that ZU) is open for each f E A, and let J be a
proper ideal of A. If f E JvJ , then VJ ~ ZU), and by the above
6.3. HYPERARCHIMEDEAN ALGEBRAS 119
claim, I E J. Therefore J = JvJ , and again by Theorem 3.4.3(iii) and

Theorem 6.3.2(vi) we can conclude that A is hyperarchimedean. 0
We shall now characterize hyperarchimedean MV-algebras with one
generator. In the light ofTheorem 3.6.7 and Proposition 3.1.5, we can
safely focus attention on MV-algebras of rest riet ions of McNaughton
functions, as follows:
Corollary 6.3.4 Let X =I 0 be a closed subset 01 the real interval [0, 1],
and let A be the subalgebra 01 Cont(X) given by the restrietions to X 01
the lunctions in Freel. Then A is hyperarchimedean iff every rational
point 01 X is isolated.
Proof" By Corollary 3.2.8, Freel coincides with the MV-algebra of

all one-variable McNaughton functions. Suppose that A is hyperar-
chimedean and let r be a rational point in X. We can easily define a
McNaughton function Ir: [0, 1] ~ [0,1] such that Z(fr) = {r}. Then
from Proposition 6.3.3, we obtain that {r} = Z (fr) n X is open in X,
i.e., r is isolated in X.
Conversely, suppose that all rational points in X are isolated, and
let I: [0, 1] ~ [0,1] be a McNaughton function. Clearly Z(flx) =
Z(f) n X. Let z E Z(f) n X. If z is rational, then {z} is an open
neighbourhood of z (in the topology of X) that is contained in Z(flx).
Suppose that z is irrational. Since I is aMeN aughton function, there
is an open interval U such that z E U ~ Z(f). Then U n X is an
open neighbourhood of z in X and is contained in Z(flx). This shows
that Z(flx) is open in X, whence by Proposition 6.3.3, A is hyperar-
chimedean. 0
Let X be a compact Hausdorff space. Following tradition, we say
that a set S ~ X is dopen iff S is simultaneously open and closed.
o and X are always clopen subsets of X. The clopen subsets of X
equipped with union, intersection and complement, form a boolean
algebra, denoted clop(X).
On the other hand, it is easy to see that
B(Cont(X)) = {I E Cont(X) I for all x E X, I(x) E {O, I}}.
Therefore the correspondence I 1----+ Z (f) defines an iso morph ism from
B(Cont(X)) onto clop(X). More generally, let U be a clopen sub set of
X and A be a separating subalgebra of Cont(X). Since U is compact,

for each x f/. U there is Ix E A such that Ix(x) = 1 and U ~ Z(fx).
Since X \ U is also compact, we can find I E A such that I(t) > ~ for
each tEX \ U and Z(f) = U. Hence 9 = 2.f E B(A) and Z(g) =
U. In condusion, we have proved that the correspondence 9 ~ Z(g)
defines an isomorphism 01 B(A) onto dop(X). In particular, for each
separating sub algebra A of Cont(X), we have B(A) ~B(Cont(X)).
Let X be a compact Hausdorff space and A be a separating subalge-
bra of Cont(X). If A is hyperarchimedean then, by Proposition 6.3.3,
the complements of the zero sets of functions in A are dopen and, by
the remark following Theorem 3.4.3, they form a basis of the topology
of X. Compact Hausdorff spaces having a basis of dopen sets are called
boolean spaces .
For every boolean algebra BIet
X(B)
be the set of maximal ideals of B, equipped with the topology having as

an open basis the sets ofthe form O"B(a) = {P E X(B) la f/. P}, where a
ranges over elements of B. Then X(B) is a boolean space. Furthermore,
Stone's celebrated theorem states that B ~ dop(X(B)), and X(B) is
uniquely determined, up to homeomorphism. Since B(Cont(X)) ~
clop(X), we conclude that a Hausdorff compact space X is a boolean
space iffB(Cont(X)) is a separating subalgebra oICont(X). We have
proved:
Corollary 6.3.5 Let X be a compact Hausdorff space. Then the MV-

algebra Cont(X) has a hyperarchimedean separating subalgebra A iff X
is a boolean space. In this case, X is homeomorphic to X(B(A)). 0
It is worthwhile to point out that the fact that X is a boolean space

need not imply that Cont(X) is a hyperarchimedean MV-algebra, as
the following example shows.
Example: Let X be Cantor's ternary subset of [0, 1]. It is well known

that X, with the topology inherited from [0, 1] is a boolean space having
no isolated points. Since X contains rational points, by Corollary 6.3.4,
6.4. DlRECT PRODUCTS 121
the separating subalgebra of Cont(X) formed by the restrietion of Mc-

Naughton functions to X is not hyperarchimedean, whence Cont(X) is
not hyperarchimedean.
6.4 Direct products

Let A be an MV-algebra. For each z E A let the functions -'z : A ~ A
and hz : A ~ A be defined by
(6.4) hz(x) = z A x and -'zX = z A -,x.
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].
Proof: By Corollary 1.5.6, (b] is an ideal of A, and hence it is closed

under the restrietion of the operation EB over (b]. We shall now show
that (b] is an MV-algebra. We first note that (b] is closed under the
operation -'b, and that for all x E (b], -'b-'bX = x. Further, by Theorem
1.5.3(vi), for every x E (b] we have x EB -'bO = x EB b = x V b = b. To
establish the identity -'b(X EB -'bY) EB x = -'b(-'bX EB y) EB y, we first note
that for any y E (b], b EB y = b V Y = b. It follows that
-'b(-'bX EB y) EB y = -'b((b A -,x) EB y) EB y
= -'b ( (b EB y) A (-,x EB y)) EB y
= (b A -,(b A (-,x EB y))) EB y = ((b A -,b) V (b A -,( -,x EB y))) EB y

= (b EB y) A (-, (-,x EB y) EB y) = b A (-, (-,x EB y) EB y).
By symmetry,
-'b(X EB -'bY) EB x = bA (-,(x EB -,y) EB x) = bA (-,( -,x EB y) EB y)
= -'b ( -'bX EB y) EB y,
as required. The remaining verifications to establish that (b] is an MV-

algebra, are all trivial. In order to prove that hb is a homomorphism,
for all x, y E A we have (x 1\ b) EB (y 1\ b) = ((x 1\ b) EB y) 1\ ((x 1\ b) EB b).

Since, by Theorem 1.5.3(vi), (x 1\ b) EB b = (x 1\ b) Vb = b, we get
(x 1\ b) EB (y 1\ b) = (x EB y) 1\ (b EB y) 1\ b = (x EB y) 1\ b.
We conclude that hb(x EB y) = hb(x) EB hb(y). The rest of the proof is

trivial. 0
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:
Proposition 6.4.3 For every MV-algebra A and

have
< b E B(A) we°
(i) The MV-algebras ( bJ and AI (-,b] are isomorphie;
(ii) (b] is a subalgebra of A iff b = 1 iJJ (b] = A;
(iii) B(( b]) = (b J n B(A). If in addition, (b J is a ehain, then b is an
atom of the boolean algebra B(A). 0
°
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
Lemma 6.4.4 Let {AihEl be a nonempty family of MV-algebras and

let P = IIiEl Ai. Then there is a set {8i I i E I} ~ B (P) satisfying the
following eonditions:
(i) ViEl 8i = 1;
(ii) 8i /\ 8j = 0, whenever i =1= j;
(iii) eaeh Ai is isomorphie to (8i ].
Proof: For each i E I, let 8i : 1---+ UiEl Ai be defined by
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:
Lemma 6.4.5 Let A be an MV-algebra. Let bl , ... ,bk, k ~ 2 be ele-

ments in B (A) sueh that
(i) bl V ... V bk = 1, and
(ii) bi /\ bj = 0 for i =1= j, i, j = 1, ... , k.
Then A ~ (bd x ... X (bk]'
Proof" From (i) we immediately get nf=l

(,bi] = {O}. Hence by The-
orem 1.3.2 and Proposition 6.4.1, the function h: A ---+ (bI] X ... x (bk]
given by h(a) = (a /\ bl , ... , a /\ bk), for each a E A, is an embedding.
Since by (ii), h(al V ... Vak) = (al,"" ak), then h is also surjective,
whence h is an isomorphism. 0
Definition 6.4.6 An MV-algebra A is called direetly indeeomposable

iff Ais nontrivial, and whenever we can write A Al X A 2 then either f"'oo.I
Al or A 2 is trivial.
As an immediate consequence of Lemmas 6.4.4 and 6.4.5 we have:
Theorem 6.4.7 An MV-algebra A is directly indecomposable if and

onlyifB(A)={O,l}.O
Prom Proposition 6.4.3(iii) and the above theorem we obtain that

for each atom b of B(A), the MV-algebra (b] ~ AI( -,b] is directly
indecomposable. This result can be generalized as follows: first of all,
let us say that an ideal J of an MV-algebra A is a maximal stonean
ideal iff J is a proper stonean ideal of A, and for each stonean ideal
I of A, I =I J and J ~ I imply I = A. In other words, J n B(A) is a
prime ideal of B (A).
Corollary 6.4.8 Let A be a nontrivial MV-algebra. For each maximal

stonean ideal J of A, the quotient algebra AI J is directly indecompos-
able.
Proof: Let x E A be such that xl JE B(AI J). Then x A -,x E J, and

there is z E J n B(A) such that x A -,x ~ z. Letting y = x V z, we get
y A -,y = (x A -,x A -,z) V (z A -,x A -,z) = 0. It follows that y E B(A),
and since J n B(A) is a prime (= maximal) ideal of B(A), then either
y E J or -,y E J. If y E J, then x E J, because x ~ y. Suppose y ~ J.
Then -,y E J, and -,x V z = -,y V z E J. Hence -,x E J. 0
6.5 Boolean products of MV-algebras

As shown by Lemmas 6.4.4 and 6.4.5, boolean elements have an im-
portant role in direct product decompositions of MV-algebras. We will
return to this topic in the next section, when considering complete
MV-algebras. Now we are going to consider a special kind of subdirect
product, where the index set is equipped with a boolean topology.
Definition 6.5.1 A weak boolean product of a family {AX}XEX of MV-

algebras (X =I 0) is a subdirect product A of the given family, in such a
way that X can be endowed with a boolean (i.e., totally disconnected,
compact Hausdorff) topology having the following two properties:
6.5. BOOLEAN PRODUCTS OF MV-ALGEBRAS 125
(i) For all J,g E A, the set [J = g] =def {x E X If(x) = g(x)} is

open in Xj
(ii) Whenever Z is a dopen (dosed-and-open) subset of X and J, gE

Athen Jlz U glx\z E A.
Replacing condition (i) by:
(i*) For all J, 9 E A, the set [J = g] is dopen,

we obtain the notion of a boolean product.
Recalling the properties of the distance function, since a = b is

equivalent to d(a, b) = 0, the above conditions (i) and (i*) can be
replaced, respectively, by:
(i') If J E A, then [J = 0] is open in X

and
(i*') If J E A, then [J = 0] is dopen in X.
In the following, when dealing with a weak boolean product A of

a family {AX}XEX, the dependence of A on the topology of X shall
be tacitly understood. As shown by the following example, boolean
products are a generalization of finite direct products.
Example: Let {AX}XEX, for X =f:. 0 be a family of nontrivial MV-

algebras, and let A = IIxEx A x . Then A is a weak boolean product of
the family {AX}XEX if, and only if, X is finite. In this latter case, A
is indeed a boolean productj further, the discrete topology on X is the
only possible topology making A into a boolean product of the family
{AX}XEX' As a matter offact, assuming X to be finite, then conditions
(i*') and (ii) are trivially satisfied upon equipping X with the discrete
topology. A moment's reflection shows that no other topology on X can
make A into a boolean product of the {AX}XEX' Conversely, assume X
to be infinite, with the intent of proving that A cannot be made into a
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 following theorem should be compared with Lemmas 6.4.4 and

6.4.5, together with the above example. Recall the definition of X( C)
for any boolean algebra C.
Theorem 6.5.2 Let A be a weak boolean produet of a family {A x }XEX,

X #0, of nontrivial MV-algebras. Let C be defined by
(6.5) C = {g E AI g(x) E {Ox, Ix} for eaeh x EX}.
Then C is a subalgebra of B(A) and we have:
(i) The eorrespondenee x ~ Qx = {g E CI g(x) = O} is a homeo-

morphism from X onto X( C);
(ii) For eaeh x E X, A x is isomorphie to AI(Qx);

(iii) C eoineides with B(A) iJJ all algebras A x are direetly indeeompos-
able.
Conversely, if A is a nontrivial MV-algebra and C is a subalgebra of

B(A), then A is isomorphie to a weak boolean produet of the family
{AI (Q) }QEX(C).
Proof: Suppose first that A is a boolean product of the family {A x } xEX,

and let C be as in (6.5). Then C is a subalgebra of B(A), and by
Definition 6.5.1, C ~ B(X); arguing as in Theorem 3.4.3 one easily
sees that the map x ~ Qx is a one-one correspondence of X onto X(C).
As a matter of fact, this correspondence is continuous, because for each
x E X and 9 E C we have Qx E ac(g) iff x E X \ [g = 0] = [.g = 0],
6.5. BOOLEAN PRODUCTS OF MV-ALGEBRAS 127
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],
whence condition (i) in Definition 6.5.1 holds. Finally, let Z be a dopen

subset of X. Then there is 9 E C such that Z = ac(g), and since for
all fand h in A, flz U hlx\z = (g 1\ 1) V (-,g 1\ h) E A, we condude
that (ii) in Definition 6.5.1 also holds. 0
Corollary 6.5.3 Eaeh nontrivial MV-algebra is isomorphie to a weak

boolean produet of direetly indeeomposable MV-algebras. 0
Let A be an MV-algebra. Since AI J is an MV-chain iff J is a prime

ideal, and MV-chains are directly indecomposable, we obtain:
Corollary 6.5.4 A nontrivial MV-algebra A is a weak boolean produet

of MV-ehains if and only if eaeh maximal stonean ideal of A is a prime
ideal of A. 0
Particular cases of MV-chains are given by simple MV-algebras, i.e.,

the subalgebras of [0, 1]. Since for any MV-algebra A, AI J is simple if
and only if J is maximal ideal, we obtain:
Corollary 6.5.5 An MV-algebra A is a weak boolean produet of simple

algebras i.tJ eaeh maximal stonean ideal of A is a maximal ideal of A.
o
Suppose that A is a weak boolean product of simple MV-algebras.

and let a E A and Q E X(B(A» be such that Q E [a =1= 0]. Then
a does not belong to the ideal (Q) generated by Q in A, and since by
Corollary 6.5.5 (Q) is a maximal ideal of A, there are e E (Q) and
n E N such that c EB na = 1, i.e., -,na ::; c. Hence there is b E Q such
that -,na ::; b, whence Q E [b = 0] ~ [a =1= 0]. Therefore [a = 0] is
dopen for each a E A. We have proved that weak boolean products of
simple MV-algebras are automatically boolean products.
On the other hand, if P is a prime ideal of an MV-algebra A, then
(P n B(A») is a maximal stonean ideal contained in P. Therefore, in
the light of Theorem 6.3.2 we conclude that each maximal stonean ideal
of A is a maximal ideal of A i.tJ A is hyperarehimedean.
From the above remarks and Corollary 6.5.5, we get the following
characterization of hyperarchimedean MV-algebras:
6.6. COMPLETENESS 129
Corollary 6.5.6 A nontrivial MV-algebra A is isomorphie to a boolean

produet of simple MV-algebras iJJ A is hyperarehimedean. 0
The example given at the end of Section 6.3 shows that in general,
Cont(X), for a boolean space X, is not a boolean product of subalge-
bras of [0,1].
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
n.x = n.x 1\ -,x = (n.x ED x) ex = (n + l).x e X ~ sex.

Therefore, s :::; sex:::; s, whence, 0 = se (s e x) = s 1\ x = x, as

required. 0
Lemma 6.6.3 Let A be a complete MV-algebra. Let {Xi li E I} ~ A.

Then
(6.7) 1\ Xi = -, V -'Xi
iEI iEI
and
(6.8) VXi = -, 1\ -'Xi •

iEI iEI
Proof" Prom the elementary properties of the natural order relation in

A.D
Lemma 6.6.4 Let A be a complete MV-algebra. Let {Xi li E I} ~ A.

Then for each X E A, the following generalized distributive laws hold:
(6.9) X 1\ VXi = V(x 1\ Xi)

iEI iEI
and
(6.10) X V 1\ Xi = 1\ (X V Xi).
iEI iEI
Proof" We first prove
(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)
X 1\ a = (-,a EB x) 8 a = V(( -,a EB x) 8 Xi)

iEI
6.6. COMPLETENESS 131
iEI iEI
Prom the trivial inequality ViEI(X 1\ Xi) :5 X 1\ a we finally obtain

(6.9). To complete the proof, it is sufficient to note that, by Lemma
6.6.3, the identities (6.9) and (6.10) are equivalent. 0
Corollary 6.6.5 Let A be a complete MV-algebra. Then
(i) B(A) is a complete boolean algebra. As a matter of fact, for every
set {bi li E I} ~ B(A) we have
(6.12) Vb E B(A)i
iEI
and
(6.13) 1\ bi E B(A);
iEI
(ii) For every b E B(A), letting (b land h b : A -+ (b 1 be as in

(6.1) and (6.4), it follows that (b 1 is a complete MV-algebra, and h b
preserves arbitrary infima and suprema. In more detail,
and
Proof" (i) Let X = ViEl bio By (6.7) and (6.9), we have

,x = 1\ ,bi and ,x 1\ x = V(,x 1\ bi ) ::; V(,bi 1\ bi ) = O.
iEI iEI iEI
Then by Theorem 1.5.3(iii), x is a boolean element of A; this settles

(6.12). The proof of (6.13) is similar.
(ii) Since (b 1 is a complete MV-algebra, (6.14) and (6.15) are an
immediate consequence of the definition of hb , in the light of (6.9) and
(6.10). 0
Using Lemma 6.6.4 an easy adaptation of the proof of Lemma 6.4.5
yields
Lemma 6.6.6 Let A be a eomplete MV-algebra. Let the set {bi I i E

I} ~ B(A) satisfy the following eonditions: ViEl bi = 1, and bi 1\ bj = 0
whenever i =J j. Then A is isomorphie to the direct produet MV-algebra
I1{(bd li E I}. 0
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.
Lemma 6.6.7 Let {Ai I i E I} be a family of MV-algebras. Then the

produet P = I1iEI Ai is a eomplete MV-algebra iff so is eaeh Ai·
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
6.7 Atoms and Pseudocomplements

Definition 6.7.1 By an atom of an MV-algebra A we mean an atom
of the underlying lattice L(A). We say that A is atomie Hf for each
o =J x E A there is an atom a E A with a ~ x. We say that A is
atomless iff no element of A is an atom.
Examples. For each n ;::: 2, the element l/(n - 1) is an atom of the

MV-chain Ln. By Proposition 3.5.3, all infinite subalgebras of [0, 1] are
atomless. By Theorem 3.5.1 and Proposition 3.5.3, up to isomorphism,
6.7. ATOMS AND PSEUDOCOMPLEMENTS 133
simple, complete, atomic MV-algebras coincide with the MV-chains Ln,

where n ~ 2. Further, [0,1] is the only simple, complete and atomless
MV-algebra.
In the rest of this section, we shall study algebras of the form ( b], when
bis an atom of the boolean algebra B(A).
Let A be a complete MV-algebra. For each z E A, recall that, by
definition, J/. = {x E A I x A z = O}. It is not hard to see that Jz 1.
is an ideal of A. (To this purpose, using the subdirect representation
theorem one simply notes that for all p, q, rE A if p A q = 0 then pAr
=pA(qEBr)).
Let zoO E A be defined by zoO = VJz 1. = V{x I x A z = O}.
Then z* is the pseudoeomplement of z, in the sense that for any x E A
(6.16) xAz=O iff x$z*.
Lemma 6.7.2 Let A be a eomplete MV-algebra. Then the lattiee L(A)

is pseudocomplemented, in the sense that eaeh z E A has its pseudo-
eomplement z". M oreover, for each z E A the pseudocomplement z" is
a boolean element of A.
Proof: By Lemma 6.6.4, L(A) is pseudocomplemented. By the above

discussion, together with the assumed completeness of A and Lemma
6.6.4(4), for every element z E A, the ideal Jz1. coincides with (zoO]. By
Corollary 1.5.6, z" E B(A). 0
With the above notation, a straight forward computation yields
(6.17) x $ xoOoO;
(6.18) If x $ y then xoOoO $ y .... ;
(6.19) (x A y)oOoO = x .... A y**;

(6.20) If z E B(L) then z .... = z.
Theorem 6.7.3 Let A be a eomplete MV-algebra and z an atom of

B(A). 1f there is an atom a of A sueh that a $ z, then the MV-algebra
(z] is isomorphie to the finite ehain Ln, for some n ~ 2. 1f no sueh
atom exists, then (z] is isomorphie to [0, 1].
Proo/: We first prove that is a prime ideal in L(A). Let us

(-.z]
assume x A y ::; -.z. By (6.17)-(6.20),
x** Ay** = (x Ay)**::; (-.z)** = -.z.
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
Lemma 6.7.4 For each atom a 0/ a complete MV-algebra A, a** is an

atom 0/ the boolean algebra B(A).
Proof: By Lemma 6.7.2, a** E B(A) and, by (6.18), a** # O. Let us

assume that z E B(A) and z ::; a**. By (6.19) and (6.20),
z = z A a** = z** A a** = (z A a)**,
and sinee a is an atom of A, we must either have z /\ a = 0 or z /\ a = a.

Therefore, z = 0 or z = a**. 0
6.8 Complete distributivity

As usual, a eomplete lattice L is said to be completely distributive iff
for every family {Ji I i E I} of nonempty sets, and for arbitrary Xij E L,
letting T = [LeI Ji we have
/\ V Xij = V /\ Xif(i) and V /\ Xij = /\ VXif(i) .

feTieI feTieI
We say that a eomplete MV-algebra A is completely distributive iff its

underlying lattice L(A) is eompletely distributive in the above sense.
6.8. COMPLETE DISTRIBUTIVITY 135
Theorem 6.8.1 For every MV-algebra A the following conditions are

equivalent:
(i) A is a direct product of totally ordered complete MV-algebras;
(ii) A is complete and completely distributive;
(iii) A is complete and the boolean algebra B(A) is atomic.
Proof' (i) => (ii): Trivial.

(ii) => (iii): By Corollary 6.6.5, the boolean algebra B(A) is eom-
plete and eompletely distributive. Then, by a classieal result of Tarski,
B(A) is isomorphie to apowerset boolean algebra. Thus in partieular,
B(A) is atomie.
(iii) => (i): The atoms of B(A) form a set {Zi I i E I} sueh that
Zi 1\ Zj =° whenever i =J j; further, ViEl Zi = 1. By Lemma 6.6.6, A
is isomorphie to the product MV-algebra ITiEI( zd. By Theorem 6.7.3,
eaeh (Zi] is a eomplete ehain. 0
Corollary 6.8.2 An MV-algebra is the direct product of copies 0/[0,1]

iff it is complete, completely distributive, and atomless. 0
Corollary 6.8.3 An MV-algebra is a direct product of finite chains iff

it is complete and atomic. 0
As a eorollary, we have a new proof ofthe result (Proposition 3.6.5)

that every finite MV-algebra is a direet product of finite ehains.
Definition 6.8.4 An MV-algebra A is said to have no linear factors

iff whenever A ean be written as a direct product of MV-algebras,
then no Ai is totally ordered. For any eomplete MV-algebra A let

{ai I i E I} be the set of atoms of A. Let {Zj I j E J} be the set of
atoms of B (A). Then we define
SA = Va** i , and ZA = V Zi'

iEI iEI
In the rest of this section, whenever in a product MV-algebra one

of the factors (xl coincides with the singleton (0], we shall tacitly un-
derstand that (xl must be deleted.
Theorem 6.8.5 I/Ais a complete MV-algebra then
I/ SA > 0 then (SA] is a direct product 0/ finite chains. I/ SA i- ZA then

(-'SA A ZA 1 is a direct product 0/ copies 0/ [0, 1], and the MV-algebra
(-'ZA 1 is complete and has no linear /actors.
Proof" Formula (6.21) immediately follows from Lemmas 6.6.6 and

6.7.4. If 0 i- Z E B( -'ZA], then Z is not an atom of B(A) and, by
Proposition 6.4.3, (z] is not totally ordered. By Lemma 6.4.4, (-'ZA]
has no linear factors. Assume SA =1= 0 and let x E (SA], x i- O. If ai i x
for each i E I we obtain
x = x A S A = x A Va;*
iEI iEI iEI
which is impossible. Therefore, ai ~ x, for some i E I, whence (SA 1

is a complete atomic MV-algebra. By Corollary 6.8.3, (SA] is a direct
product of finite chains. By Lemma 6.7.4, SA ~ ZA· Thus if SA i- ZA,
then -'SA A ZA i- O. If a were an atom of (-'SA A ZA ], then a would also
be an atom of A; then a ~ a** < SA, whence a = 0, which is impossible.
Therefore, (-'SA A ZA ] is an atomless complete MV-algebra. The set of
atoms of B((-'SA A ZA ]) = (-'SA A ZA] n B(A) coincides with the set
of those atoms Zi of B(A) such that Zi ~ -'SA. From the identities
-'SA A ZA = -'SA A VZi = V(-'SA A Zi)

iEI iEI
it follows that B(( -,sAAzA ]) is an atomic boolean algebra. By Theorem

6.8.1 and Corollary 6.8.2, if SA =1= ZA then (-,sAAzA] is a direct product
of copies of [0,1]. 0
6.9 Bibliographical re marks

In every MV-algebra A the operation -, satisfies the following condi-
tions:
(MI) -,0 = 1;
(M2) -,-,x = x;
(M3) -,(x V y) = -,x A -'y.
A distributive lattice with 0 and 1 equipped with an operation satisfying

(MI), (M2) and (M3), is called De Morgan algebra. Ais called a Kleene
algebra (see [12)) Hf it satisfies the additional condition:
By the subdirect representation theorem, for every MV-algebra A, its

underlying lattice (A, V, A, -', 0,1) is a Kleene algebra.
Theorem 6.1.3 was first proved in [146], and was used in [99] and
[100] to show that, in Zermelo-Fraenkel set theory without the axiom
of choice, the following statements are equivalent (one implication is
trivial):
• Every MV-algebra has a maximal ideal.
• Every boolean algebra has a maximal ideal.

It is known that this latter statement is strictly weaker than the axiom
of choice.
The result proved in the claim in Theorem 6.1.5 is a well known
elementary fact of the theory of distributive lattices, due to Birkhoff
and Stone. See, e.g., [103]. For a study of stonean ideals in lattices
see [45]. The proof of the claim in Proposition 6.3.3 can be found in
[163, Lemma 8.5]. Proposition 6.4.1 is due to Rodriguez, see [212]. For
direct product decompositions also see [123].
For boolean products of MV-algebras see [231], [232], [55] and [56J.
Corollary 6.5.6 is due to Torrens [231J (caution: our hyperarchimedean
MV-algebras are called archimedean in [231]).
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
MV-algebras and f-groups
As proved at the beginning of Chapter 2, r is a functor from the cat-

egory A of f-groups with a distinguished strong unit, to the category
MV of MV-algebras. In this chapter we shall prove that r is a natural
equivalence (i.e., a full, faithful and dense functor) between A and MV.
As a consequence, a genuine addition can be uniquely recovered from
the MV-algebraic structure. Several applications will be discussed.
7.1 Inverting the functor r

In this section we shall give an explicit construction of an adjoint func-
tor of r. Our starting point is the f-group GA with order unit UA
considered in Section 2 of Chapter 2. As the reader will recall, for
every MV-algebra A, GA is an f-group and there is an isomorphism be-
tween the ordered monoid GA + of its positive elements and the ordered
monoid M A of good sequences of A.
Let A and B be MV-algebras, and h: A ~ B a homomorphism. If
a = (al, a2, ... ) is a good sequence of A, then (h(ad, h(a2), ... ) is a
good sequence of B. Let h*: M A ~ MB be defined by
(7.1) h*(a) = (h(al), h(a2), .. .),
for all a E M A . By direct inspection, using (2.11) and Proposition
2.2.3, for all a, b E M A we have
(7.2) h*(a + b) = h*(a) + h*(b)j

139
140 CHAPTER 7. MV-ALGEBRAS AND l-GROUPS
(7.3) h*(a V b) = h*(a) V h*(b);

(7.4) h*(a/\ b) = h*(a) /\ h*(b).
Thus, h* ; M A -+ MB is both a monoid-homomorphism and a lattice-

homomorphism. Let us further define the map h# ; GA -+ G B by
(7.5) h#([a, b]) = [h*(a), h*(b)].

Let UA and UB be the strong units of GA and G B given by Proposition
2.4.4. Then by (7.2)-(7.4), the map h# is a unitall-group homomor-
phism of (GA, UA) into (G B, UB). Let us agree to write
(7.6) E(A) = (GA,UA) and E(h) = h#.
Proposition 7.1.1 E is a functor from MV into A. 0
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
Theorem 7.1.2 The composite functor rs is naturally equivalent to

the identity functor of MV. In other words, for all MV-algebras A, B
and homomorphism h ; A -+ B, we have a commutative diagram
h
A --+ B
'PA ! ! 'PB
r(S(h»
r(E(A)) --+ r(E(B))
in the sense that, for each a E A, 'PB(h(a)) = (f(E(h)))('PA(a)).

Proof: For each a E A, 'PB(h(a)) = [(h(a)), (0)] and 'PA(a) = [(a), (0)].
Further, by (7.1)-(7.6) E(h)([(a), (0)]) = [(h(a)), (0)], the latter being
an element of r(S(B)). Since r(E(h)) is the restrietion of S(h) to
r(E(B)), we can write r(E(h))('PA(a)) = [(h(a)), (0)] = 'PB(h(a)), as
required. 0
Our next aim is to prove that the composite functor sr is also
naturally equivalent to the identity functor of the category A. We first
prepare the "dual" of the maps 'PA.
7.1. INVERTING THE FUNCTOR r 141
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.
Proo/: For every integer k 2: 1, let ak be inductively defined by: al =

a 1\ u, and ak+1 = (a - al - ... - ak) 1\ u.
In order to prove that (al,"" ak,"') is a good sequence of A, we
first prove the identity
(7.7) a - al - ... - ak = (a - ku) V0 = (a - ku)+.
The proof is by induction on k. For k = 1, we have
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
= (a - mu)+ - ((a - mu)+ 1\ u)

= 0 V ((a - mu)+ - u)
= (a - (m + l)u)+,
which settles (7.7). Since for each k 2: 1, ak = (a - (k - l)u)+ 1\ u,
then necessarily each ak is in the unit interval [0, u] of G. Since u is
a strong unit of G, there is an integer n 2: 1 such that a :s; nu, and
hence, aj = 0 for all j > n.
We shall now prove that ak EB ak+l = ak. To simplify the notation,
let t = a - ku. Then by (7.7)
= ((t+ 1\ u) + ((t - u)+ 1\ u)) 1\ u

= (t+ + (t - u)+) 1\ u
= ((t V 0) + ((t - u) V 0)) 1\ u
142 CHAPTER 7. MV-ALGEBRAS AND f-GROUPS
= ((2t - u) /\ u) V (t+ 1\ u) = t+ 1\ u = ak.
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
and, proceeding inductively, for i 2:: 2,
= (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
Definition 7.1.4 An f.-ideal of an f.-group G is a subgroup J of G

satisfying the following condition:
(7.8) If xE J and lyl ~ lxi then y E J.

Letting for each a E G, al J = a + J be the coset determined by a,
and defining al J V bl J = (a V b)1 J, the quotient group GI J becomes
an f.-group. The map qJ: G ~ GI J sending each a E G to the coset
al J, is an f-homomorphism such that K er(qJ) = J. Moreover, if u is
a strong unit of G, then qJ(u) is a strong unit of GIJ. Conversely,
if G and H are f.-groups and f: G ~ H is an f.-homomorphism, then
Ker(f) = f- 1 ({O}) is an f.-ideal of G, and GIKer(f) is isomorphie to
the f-subgroup f(G) of H.
An f-ideal J of an f-group G is said to be prime Hf J is proper (Le.,
J =f G), and the quotient f.-group GI J is totally ordered.
Lemma 7.1.5 Let G be an f.-group with an order unit u. Then the

mapping g: G+ ~ MrCG,'I.I.) of Lemma 7.1.3 satisfies the following con-
ditions, for all a, b E G+:
(i) g(a + b) = g(a) + g(b);

(ii) g(a Vb) = g(a) V g(b);
(iii) g(a /\ b) = g(a) /\ g(b);

(iv) g(u) = (u).
Proof: The notion of a subdirect product of f.-groups is, mutatis mutan-

dis, the same as the corresponding not ion for MV-algebras (Definition
1.3.1). The following claim is a classieal result of Birkhoff :
Claim. Every f.-group G is a subdirect product of totally ordered
abelian groups.
Indeed, let 0 =f a E G. An adaptation of the proof of Proposition
1.2.13 shows that any f.-ideal of G whieh is maximal for not containing
a is automatically prime. It follows that the intersection of all prime
f.-ideals of Gis the zero ideal, and G can be embedded into the product
of all its prime quotients G / J. Thus, G is a subdirect product of totally

ordered f-groups, and our claim is proved.
Let X be the set of prime f-ideals of G. Our claim yields a family
{GX}XEX of totally ordered abelian groups and a one-one f-homomor-
phism
h:G ~ TIxEXGx
such that, for each x E X, the composite function 7rx h maps G onto
Gx (7r x being the canonical projection onto G x ). For every a E G,
letting ax = 7rx (h(a)), it follows that U x is a strong unit of G x. Given
arbitrary elements a, b E G+, let us write g(a) = (al,"" am ) and
g(b) = (bI,"" bn ). Then, for each x E X, both (alx,"" amx ) and
(blx,"" bnx ) are good sequences in r(G x , u x ). Moreover,
n
and bx = L:bix '
i=I
Since the MV-algebra r(G x , u x ) is totally ordered, by Proposition 2.2.1

every good sequence of A has the form (l P , a). Thus, there are elements
O'x, ßx E Gx with 0 ~ O'x < U x and 0 ~ ßx < u x , satisfying
the equations (aIX, ... ,amx ) = (u~"',O'x) and (blx, ... ,bnx ) = (u~"',ßx).
Further, by definition of g,
whence (a+b)x = ax+bx = (Px+qx)ux+O'x+ßx. Recalling (2.12), by

an application of Lemma 2.1.3(i), for all x EX we obtain
(u~"',O'x) + (ui"',ßx) = (u~",+q"',O'x$ßx,O'x 0ßx).

Then, by direct inspection, (a + b)x = (aIX"'" amx ) + (b lx , ... , bnx ).
We conclude that g(a + b) = g(a) + g(b), and (i) is proved.
To prove (ii), for any x E X, we either have ax ~ bx or bx ~ a x .
Suppose, without loss of generality, ax ~ bx . Then, from 0 ~ O'x ~ ßx <
u x, by (7.9) we get Px ~ qx, whence (u~"', O'x) ~ (u~"', ßx). Therefore,
letting r = max(m, n),
r
(a V b)x = :~:)aiX V bix ).
i=l
In other words, a V b = Ei=l (ai V bi ), whence, by Lemma 7.1.3, g(a V b)

= g(a) V g(b), which completes the proof of (ii). The proof of (iii) is
similar, and (iv) follows at once by definition of g. 0
Corollary 7.1.6 For every (G, u) E A let the map 'IjJ(G,u) : G -+
GrCG,u) be defined by
for all a E G. It follows that 'IjJ(G,u) is an f.-group isomorphism of G

onto GrcG,u), and 'IjJ(G,u)(u) = [(u), (0)]. 0
In the light of Corollary 7.1.6, using the maps '1jJ(G,u) (for all (G, u) E A),
we have the following
Theorem 7.1.7 The composite functor sr is naturally equivalent to

the identity functor of the category A. In other words, for any two f-
groups with strong unit (G, u) and (H, v) and uni tal f-homomorphism
f: (G, u) -+ (H, v), we have a commutative diagram
(G,u) (H,v)
'1jJ(G,u) ! ! 'IjJ(H,v)
S(r(G,u)) S(r(H, v))
in the sense that, for all a E G, 'IjJ(H,v)(j(a)) = S(r(j))('IjJ(G,u)(a)).
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
whence, recalling (7.1), g(j(a)+) = (h(al)' ... ' h(an)) = h*(g(a+)}.

Similarly, g(j(a)-) = h*(g(a-)), whence by (7.5), (7.6) and (7.10),
'IjJ(H,v) (j(a))
= [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
Corollary 7.1.8 The r functor defines a natural equivalence between

the category A 01 l-groups with strong unit, and the category MV 01
MV-algebras. 0
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
Therefore, r(f) = h. In order to establish the uniqueness of I, suppose

I(b) #- f'(b) for some bE G+, and unital i-homomorphism
1': (G, u) --+ (H, v).
Writing b = al + ... + ak for suitable ai E r(G, u), if r(f) = r(f')

we have I(b) = I(al) + ... + I(ak) = I'(ad + ... + f'(ak) = f'(b), a
contradiction.
(ii) Suppose that h maps r(G, u) onto r(H, v), and let b E r(H, v).
Then there exists an element a E r(G, u) such that h(a) = band
I(a) = h(a) = b. Since every element of H+ (resp., every element of
G+) is a sum of elements of the unit interval [0, v] (resp., a sum of
elements of [0, uD we have that I maps G+ onto H+, whence I maps
G onto H.
(iii) Finally, if I is not one-one then I maps some nonzero element
x E G into the zero element of H. Without loss of generality, ~ x,
°
whence = I(x /\ u) = h(x /\ u), and h is not one-one. 0
°
Let us agree to denote by T(G) the set of all i-ideals of G, ordered
by inclusion.
Theorem 7.2.2 Let G be an i-group with strong unit u, and A =

r (G, u). Then the correspondence
cp : J ~ cp( J) = {x E G I Ixl/\ u E J}
is an order-isomorphism /rom the set T(A) 01 ideals 01 the MV-algebra
A, ordered by inclusion, onto T( G). The inverse isomorphism 'IjJ is
given by H E T(G) ~ 'IjJ(H) = H n [0, u].
Proof" Trivially, 'IjJ(H) E T(A) whenever H E T(G). Conversely, to

prove that for each ideal J of A, cp(J) is an i-ideal of G, assume x, y E
cp(J). Then (Ixl/\ u) ffi (lyl/\ u) E J. Furthermore,
(Ixl/\ u) ffi (lyl/\ u)
= ((Ixl/\ u) + (lyl/\ u)) /\ u

= ((Ixl/\ u) + lyJ) /\ ((Ixl/\ u) + u) /\ u
148 CHAPTER 7. MV-ALGEBRAS AND i-GROUPS
= ((lxi /\ u) + Iyl) /\ u
= (lxi + Iyl) /\ u
~ Ix - yl /\u,
whenee Ix - yl /\ u E J, and henee, x - °

y E 4>(J). Sinee E 4>(J), then
4>(J) is a subgroup of G, and sinee 4>(J) satisfies (7.8), 4>(J) is an i-ideal
of G. Trivially, the funetion 4>: I(A) --+ I(G) preserves inc1usions. Now
it is not hard to see that 1/J(4)(J)) = J, for each ideal J of A. Henee to
eomplete the proof, there remains to be shown that for each i-ideal H of
G, 4>(1/J(H)) = H; stated otherwise, H = {x E G Ilxl /\ u EH n [0, un.
The inc1usion H S; {x E G Ilxl /\ u E H n [0, un is c1ear from (7.8). On
the other hand, if lxi /\ u E H n [0, u], then sinee u is a strong unit there
°
is an integer n ~ 1 such that lxi ~ nu. Thus, ~ lxi = lxi /\ nu ~
n(lxl /\u) E H, and x E H. 0
Corollary 7.2.3 The map J 1--+ J n [0, u] defines an isomorphism
between the partially ordered set 0/ prime i-ideals 0/ G and the partially
ordered set 0/ prime ideals 0/ r( G, u), both sets being equipped with the
inclusion ordering. 0
Theorem 7.2.4 Let G be an i-group with strong unit u. Then /or

every i-ideal J 0/ G, we have the isomorphism
r(GIJ,uIJ) ~r(G,u)/(Jn [O,u]).
Proo/: Let /: G -+ GI J be the natural i-homomorphism, and k = r(f).
Let A = r(G, u) and 1= Jn[O, u]. By Theorem 7.2.2, I is an ideal of A.
Sinee by Lemma 7.2.1(ii), k is a homomorphism of A onto r(GI J, ul J),
whose kernel eoincides with I, then AI I ~ r (GI J, u I J), as required.
o
Remark: It follows at onee that the eorrespondenee J 1--+ J n [0,1]
defines a 1-1 mapping from the set of maximal i-ideals of G onto the
set of maximal ideals of r(G, u).
We shall noweonstruet uneountably many non-isomorphie simple
MV-algebras: We first prepare the following well-known result; its proof
is inc1uded here to inerease readability.
7.2. APPLICATIONS 149
Proposition 7.2.5 Let G and H be f-subgroups of the additive group

R of real numbers with natural order. Assume 1 E G n H. Then
there is at most one f-isomorphism f ofG onto H such that f(l) = 1.
Whenever such f exists, then G necessarily coincides with H, and f is
the identity function on G.
Proof: By way of contradiction, assume G ~ H, and let a E G\H.

Then a =1= f (a) EH. In case a < f (a) there is a rational number
p/q such that a < p/q < f(a), Le., qa < p < qf(a). Therefore,
o < p - qa E G, and 0 > p - qf(a) = pf(l) - qf(a) = f(p - qa) E H,
whence f does not preserve the order, a contradiction. In a similar
way, if f(a) < a, a negative element of G would be mapped by f into
a positive element of H. We have proved that G ~ H. Symmetrically,
H ~ G, and we are done. 0
Recalling now Lemma 7.2.1 we obtain
Corollary 7.2.6 Two subalgebras A and B of [0,1] are isomorphie iff

A = B; the identity function is the only automorphism of A. 0
Example: As in the example following Theorem 3.4.9, for any irrational

°
number E [0,1] let the MV-algebra Sa be defined by
Sa = {m+na I m,n E Z,O ~ m+na ~ I}.
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
Corollary 7.2.7 There are uncountably many nonisomorphie simple

subalgebras of [0, 1] with one generator. 0
7.3 The radical

Let G be an f-group. Given elements a, b in G+, we say that a is
injinitely smaller than b, in symbols, a « b, iff na :5 b for each integer
n~O.
Lemma 7.3.1 For any f-group G with strong unit u, we have:

(7.11) Rad(r(G, u» = {x E G+ I x« u}.
Proo/: By Proposition 3.6.4, for every MV-algebra A we have Rad(A)
= Injinit(A) U {O}. Let H = {x E G+ Ix «u}. If x E H then a
fortiori, x E [0, u]. For each integer 0 :5 n we have (n+ l)lxl :5 u; using
the notational convention of (2.7), we can write
n.lxl = u 1\ nix I :5 u 1\ (u -lxI) = -,x,
whence H ~ Injinit(r(G,u».
Conversely, let x E r(G, u) be infinitesimal. By induction on n we
shall prove that nx :5 u. The case n = 0 is trivial. For the induction
step, suppose nx :5 u. Then nx = u 1\ nx = n.x :5 u - x, whence
(n + l)x :5 u. In conclusion, x « u, as required to complete the proof.
o
Recall from Theorem 7.2.2 the definition of the map 4>.
Lemma 7.3.2 Let J be an ideal 0/ the MV-algebra r(G, u) such that
J ~ Rad(r(G, u». Then 4>(J) = {x E G Ilxl E J}.
Proof: In the light of Proposition 3.6.4, we must prove that whenever
Ixll\u is infinitesimal then lxi :5 u. As a matter offact, by Lemma 7.3.1,
if Ixll\u is infinitesimal, then Ixll\u« u. Since u is a strong unit of G,
there is 0:5 mEZ such that lxi :5 mu. Therefore, lxi :5 m(lxll\u) :5 u.
o
Let A be an MV-algebra and J an ideal of A contained in Rad(A).
With reference to Theorem 2.4.5, since 'PA: A -+ r(G A, UA) is an iso-
morphism, it follows that 'PA(J) is an ideal of r(G A, UA) contained in
Rad(r(GA, UA». As an f-ideal of GA, 4>('PA(J» is an f-group (more
precisely, an f-subgroup of GA)' Since by Lemma 7.3.2, 4>('PA(J» ~
{x E GA Ilxl « UA} ~ [0, UA], we conc1ude that 'PA(J) = (4)('PA(J)))+.
As a consequence we have
7.4. PERFECT MV-ALGEBRAS 151
Lemma 1.3.3 Let A be an MV-algebra and J an ideal 0/ A such that

J ~ Rad(A). Then we have
(i) Por all x, Y E J, <PA(X) ffi <PA(Y) = <PA(X) + <PA(Y);

(ii) J is nilpotent, i.e., J2 = {x 0 Y I x, Y E J} = {O}.
Proo/: Suppose x, Y E J. Since <PA(X) and <PA(Y) both belong to

(4)(<PA(J)))+, then <PA(X) + <PA(Y) E (4)(<PA(J)))+. Therefore, <PA(X) +
<PA(Y) < UA, whence (i) folIows. Moreover, <PA(X) 0 <PA(Y) = (<pA(X) +
<PA(Y) - UA) V 0 = O. Since <PA is an isomorphism, also (ii) is settled.
o
Theorem 1.3.4 Let A be an MV-algebra. Por each ideal J ~ Rad(A)

there is an l-group G(J), unique up to isomorphism, such that (J, ffi, 0)
is isomorphie to the ordered monoid G( J)+ 0/ positive elements 0/
G(J).
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.
7.4 Perfeet MV-algebras

Let A be an MV-algebra. As an intersection of (maximal) ideals of
A, Rad(A) is itself an ideal of A. Let the f-group G(Rad(A» be as
in Theorem 7.3.4. If A is semisimple, then G(Rad(A» = {O}. At
the other extreme, one can naturally investigate MV-algebras having
as many infinitesimals as possible. This leads to the following
Definition 7.4.1 An MV-algebra A is called perfeet iff A is nontriv-

ial and A = Rad(A) U -,Rad(A), where -,Rad(A) = {x E A I-,x E
Rad(A)}. For each perfect MV-algebra A, let
~(A) =def G(Rad(A)),
and for each homomorphism h from A into a perfect MV-algebra B, let
~(h) =def 8(h)I{XEGA Ilxl«UA}
be the restriction of S(h) to the set {x E GA Ilxl « UA}.
Since f-group homomorphisms preserve the « relation, it follows

that ~(h) is an f-group homomorphism from ~(A) into ~(B). It
is easy to check that ~ is a functor from the the category of perfect
MV-algebras (considered as a full subcategory of the category of MV-
algebras) into the category of f-groups. Conversely, we shall give a
method for constructing perfect MV-algebras from f-groups.
To this purpose, let H be a totally ordered abelian group and G be
an f-group. The lexicographic product Ho G of Hand G is the f-group
whose elements are the pairs (x, a) such that x E Hand a E G, with
the group operations defined pointwise and the order relation given by
(x, a) < (y, b) iff either x < y or x = y and a < b. Note that the
infimum of two elements in H 0 G is obtained as follows:
(x,a) if x< y
(7.12) (x, a) 1\ (y, b) = { (y, b) if y <x
(x,al\b) if x = y
For each t'-group G, we write A( G) = ZoG, where, as usual, Z denotes

the additive group of integers with natural order. Further, for any
f-group Hand f-group homomorphism h: G ~ H, let us define the
function A(h): A(G) ~ A(H) by
(7.13) A(h)((m, a)) = (m, h(a)) for each pair (m, a) E ZoG.
Then A(h) is an f-group homomorphism. As a matter of fact, Ais

a functor from the category of lattice-ordered abelian groups into itself.
Let G be an f-group. Since for each m > 0 and each a E G, the

pair (m, a) is a strong unit of ZoG, we can write
~(G) = r(A(G), (1,0)).

The zero element of ~(G) is (0,0) and the operations E9 and -, are
defined as folIows:
(O,a+b), if i+j=O
(i,a)E9(j,b)= { (l,(a+b)/\O), if i+j=l
(1,0), if i +j = 2
and
-,(i, a) = (1 - i, -a).
By definition of lexicographic order it follows that
~(G) = {(O, a) la E G+} U {(1, b) I bE G-}.

The MV-algebra C = ~(Z) is a first example of a nonsemisimple
MV-algebra. More generally, since Rad(~(G)) = {(O, a) la E G}, and
for each a E G, (1, -a) = (1,0) - (0, a) = -,(0, a), the algebras ~(G)
are examples of perfeet MV-algebras.
Lemma 7.4.2 Por each perfeet MV-algebra A, the stipulation

if xE Rad(A)
if x E -,Rad(A)
defines an isomorphism ClA from A onto ~(ß(A)).
Proof' Since <PA(Rad(A)) = (ß(A))+, ClA is a one-one mapping from

A into ~(ß(A)). To see that ClA maps A onto ~(ß(A)), take (i, z) E
~(ß(A)). If i = 0, then z = <PA(X), for some x E Rad(A), and
(0, z) = ClA(X), If i = 1, then z = -<PA(Y), with Y = -,-,y E Rad(A).
Hence (1, z) = ClA(-'Y)' We shall now show that ClA is a homomor-
phism. Trivially, ClA(O) = (0,0), and a simple computation shows that
ClA(-'X) = (1,0) - ClA(X) = -'ClA(X), There remains to be proved
(7.14) ClA(X E9 y) = ClA(X) E9 ClA(Y)'
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
If h is an f-group homomorphism from G into an f-group H, it fol-

lows that A(h): A(G) -+ A(H) is an f-group homomorphism such that
A(h)((l,O)) = (1,0). Therefore the map E(h) defined by E(h) =def
r(A(h)): E(G) -+ E(H) is a homomorphism. It is now easy to see
that E is a functor from the category of f-groups to the category of
MV-algebras.
For each f-group G, we have the identity Rad(E(G)) = {O} x G+. It
follows that
</>((E(G))) = {(m, x) E A(G) II(m, x)1 « (1, O)} = {O} x G,
whence the mapping a 1---+ ßG(a) = (0, a) is an f-group isomorphism
from G onto ß(E(G)).
The next lemma shows that the composite functor ßE is naturally
equivalent to the identity functor of the category of f-groups.
Lemma 7.4.3 For all f-groups G, Hand for any f-group homomor-
phism h: G -+ H, we have a commutative diagram
G ~ H
ßG 1 1 ßH
ß(E(G)) ~~» ß(E(H))
in the sense that, for each a E G, ßH(h(a)) = .6.(~(h))(ßA(a)).
Proof: It is sufficient to observe that ßH(h(a)) = (0, h(a)), and
.6.(~(h) )(ßa( a))
= .6.(~(h))(O, a) = (0, h(a)).
o
Conversely, the following lemma shows that the composite functor
is naturally equivalent to the identity functor of the category of
~.6.
MV-algebras:
Lemma 7.4.4 For all MV-algebras A, Band for any homomorphism

h: A -+ B, we have a commutative diagram
in the sense that, for each a E A, aB(h(a)) = ~(.6.(h))(aA(a)).
Proof: If a E Rad(A) then, trivially, h(a) E Rad(B) and aB(h(a)) =

(0, 'PB(h(a))). On the other hand,
~(.6.(h))(aA(a)) = ~(.6.(h)(O, 'PA(a))) = (0, r(B(h))('PA(a))),

and the result follows from Theorem 7.1.2. 0
From Lemmas 7.4.3 and 7.4.4 we immediately get:
Theorem 7.4.5 The.6. functor establishes a natural equivalence from

the category of perfeet MV-algebras, considered as a full subcategory of
the category of MV-algebras, and the category of f-groups. 0

The natural equivalence between MV-algebras and f-groups with strong
unit was first established in [163], building on previous work by Chang
[38] for the totally ordered case (see also [131]).
The proof of Theorem 7.2.4 can be found in [56]. The results estab-
lished in the remark following Lemma 7.3.3 are due to Belluce [14].
As a generalization of the MV-algebra algebra C originally intro-
duced by Chang in [36], perfeet MV-algebras were introduced in [20].
The natural equivalence between the categories of perfeet MV-algebras
and f-groups was established by Di Nola and Lettieri [65] (see also [60]).
Chapter 8
Varieties of MV-algebras
A dass C of MV-algebras is said to be a variety (or, an equational class) ,

iff there is a set E of MV-equations such that for every MV-algebra A,
A E C iff A satisfies all equations in E. For instance, when E = 0, we
obtain the variety MVofMV-algebras. When E = {x = y}, we obtain
the variety of trivial MV-algebras. The main aim of this chapter is to
describe all varieties of MV-algebras.
8.1 Basic definitions

A variety C of MV-algebras is called proper iff it is nontrivial and is dif-
ferent from MV. By Corollary 1.5.5, the dass of boolean algebras is the
proper equational dass ofMV-algebras determined by E = {x = xEBx}.
By Lemma 1.4.5 every variety C is dosed under subalgebras, homomor-
phic images and direct products. Using this observation we obtain that
the following three dasses of MV algebras are not varieties: semisimple,
hyperarchimedean, and complete MV algebras. Indeed, since by Corol-
lary 3.6.2 all free MV-algebras are semisimple, from Proposition 3.1.5
it follows that the dass of semisimple MV-algebras is not dosed un-
der quotients. As noted after Definition 6.3.1, the dass of hyperar-
chimedean MV algebras is not dosed under direct products. Finally,
the dass of complete MV algebras is not dosed under subalgebras, be-
cause Q n [0, 1] is an incomplete subalgebra of the complete MV algebra
[0,1].
157
158 CHAPTER 8. VARIETIES OF MV-ALGEBRAS
Let K. be a set of MV-algebras and E be the set of all MV-equations

which are satisfied by all MV-algebras in K.. The variety determined
by E is said to be generated by K., and is denoted by C(K.). When
the set K. eonsists of one MV-algebra A, we shall write C(A) instead
of C({A }), and we will eall C(A) the variety genera ted by A. With
this notation, the Completeness Theorem 2.5.3 amounts to the identity
MV = C([O, 1]). More generally we have
Proposition 8.1.1 1/ A is an infinite subalgebra 0/ the MV-algebra

[0,1], then C(A) = MV.
Proo/: Sinee the operations ....,x = 1 - x and x ES y = min(l, x + y)

are eontinuous over [0,1], then for each MV-term r(xl, ... , x n ), the
°
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.
Proposition 8.1.2 1/2 :5 no < nl < ... is an infinite sequence 0/

natural numbers, then C( {L ni li = 0, 1, ... }) = MV.
Proof: For each a E [0, 1] there is a number k such that 1 :5 k :5 n - 1

and la- n~11 < n':l· The desired result now follows from the eontinuity
of the term funetions r[O,I): [0, l]n -+ [0,1]. 0
As a eonsequenee of Theorems 1.3.3 and 1.4.6, every variety 0/ MV-
algebras is generated by the collection 0/ its MV-chains.
If A is a proper MV-chain, AI Rad(A) is a simple MV-algebra, be-
eause Rad(A) is the unique maximal ideal of A. By Theorem 3.5.1,
AI Rad(A) is isomorphie to a subalgebra of the MV-algebra [0,1].
Definition 8.1.3 Let A be a proper MV-chain. If for some n ;::: 2,

AI Rad(A) 9:! Ln, then we say that A is of rank n. If AI Rad(A) is
isomorphie to an infinite subalgebra of [0,1], we say that A is of infinite
rank.
8.1. BASIC DEFINITIONS 159
As an example, for each integer n;::: 2 the MV-algebra Ln is a

simple MV-chain of rank n.
For every totally ordered abelian group G, the algebra :E (G) =
r(A(G), (1,0)) introduced in Section 7.4 is a nonsimple MV-chain of
rank 2. More generally, for each bEG and each integer n ;::: 2, con-
sider the MV-algebra r(A(G), (n - 1, b)). By definition oflexicographic
product, r(A(G), (n - 1, b)) coincides with the union AUBUC, where
A is the set of all pairs of the form (0, x), for arbitrary b ~ x E G,
B is the set of all pairs (n - 1, y), for arbitrary b ;::: y E G, and
C = U~":i2( {i} x G). The pair (0,0) is the zero element, and the opera-
tions ..., and E9 given by
(i + j, x + y) if i + j < n - 1 or if
(8.1) (i,x)E9(j,y) = { i+j=n-landx+y<b
(n - 1, b) otherwise.
(8.2) ...,(i, x) = (n-l-i,b-x)
It follows that the 0 operation of r(A(G), (n - 1, b)) is given by
(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
r(7l"l): r(A(G), (n - 1, b)) - r(Z, n - 1)
is a surjective homomorphism. Since
Ker(r(7l"l)) = {(O, t) ItE G+} = Rad(r(A(G), (n - 1, b))),
we have
r(A(G), (n - 1, b))j Rad(r(A(G), (n - 1, b))) ~ r(Z, n - 1) f'oj Ln.

For each integer n ~ 2, we shall write, in abbreviated form,
Kn = r(A(Z), (n - 1,0))
and
Rn = r(A(Z), (n - 1,1)).
Proposition 8.1.4 1/ C is a proper variety 0/ MV-algebras then there

is an integer m ~ 2 such that rank(A) :$ m /or alt MV-chains A E C.
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.
8.2 Varieties from simple algebras

Following the notation in Section 1 of Chapter 2 (see Equation (2.7)),
for x E [0,1], we will write n.x instead of x EB ... EB x (n times), and
write nx instead of x + ... + x (n times).
Definition 8.2.1 Let n ~ 2 be an integer. By an n-bounded MV-

algebra we shall mean an algebra satisfying the equation
(Eno ) (n - 1).x = n.x
The variety of n-bounded MV-algebras will be denoted by Uno
Theorem 8.2.2 Let A be an MV-algebra and n ~ 2 an integer. Then

A E Un i/ and only i/ A is a subdirect product 0/ algebras L k , with
2:$ k :$ n.
8.3. MV-CHAINS OF FINITE RANK 161
Proof: Let A E Uno By the remark following Definition 3.6.3 we have

Rad(A) = (0). Hence A is semisimple, Le., A is a subdirect product
of subalgebras of [0, 1]. By Theorem 1.4.6, all these subalgebras satisfy
the equation (n - l).x = n.x. Let S be a sub algebra of [0,1] such
that, for each k = 2,3, ... ,n, S =f Lk. Then by Proposition 3.5.3 there
°
is t E S such that < t < l/(n - 1). It follows that (n - l).t =
min(l, (n -l)t) < min(l, nt) = n.t. Therefore A is a subdirect product
of algebras L k , for suitable integers 2 ~ k ~ n.
Conversely, let A be a subdirect product of algebras L k , with 2 ~
°
k ~ n. For all x E L k , we either have x = or x 2: l/(k - 1). From
k -1 ~ n -1 it follows that min (1, (n -l)x) = 1 = min (1, nx). Thus,
for each 2 ~ k ~ n the MV-chain L k belongs to Uni by Theorem 1.4.6,
A E Uno 0
Recalling now Theorem 1.4.6 we have
Corollary 8.2.3 For each integer n 2: 2, Un = C( {L2 , L3 , ••• , Ln}). 0
Corollary 8.2.4 An MV-algebra A belongs to the variety C(L 2 ) iJJ A

is a boolean algebra, i.e., iJJ A satisfies the equation xEBx = X. An MV-
algebra belongs to C(L3 ) iJJ it satisfies the equation x EB x = x EB x EB X.
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
8.3 MV-chains of finite rank

As we have seen above, for every totally ordered group G, algebras
of the form r(A(G), (n - 1, b)) yield examples of MV-chains of rank
n. We shall see in this section that these are in fact the most general
examples of nonsimple MV-chains of rank n. Firstly we introduce some
notations.
Let A be a totally ordered MV-algebra. By Theorem 7.3.4 we can
associate with Rad(A) the f-group G(A), which is an f-subgroup of the
.e-group GA. Hence, by the remark following Theorem 2.4.5, G(Rad(A))

is a totally ordered group, denoted p(A). When A is a perfect MV-
chain, then p(A) = a(A).
Theorem 8.3.1 Let A be a nonsimple MV-chain 01 rank n. Then there

is 0 < b E Rad(A) such that A ~ r(A(p(A)), (n - 1, c,oA(b))).
Proof: Let I: A ~ Ln be the composition ofthe natural homomorphism
from A onto AI Rad(A) and the isomorphism from AI Rad(A) onto Ln.
By Lemma 7.2.1, there is a unique .e-group homomorphism g: GA ~ Z
such that g(UA) = n - 1 and r(g) = I. Since 1 is surjective, by the
same Lemma we know that 9 is surjective. Since 1 is not injective
(because A is not simple), 9 is not injective, that is, K er(g) = {x E
GA Ig(x) = O} =f {O}. By Theorem 7.2.4, we have isomorphisms
r(G A, UA) Ic,oA (Rad(A))
~ r(Z, (n - 1))
~ r(GAI Ker(g), uAI Ker(g))

~ r(G A, uA)I'!jJ(Ker(g)).
Since every (nontrivial) MV-chain has exactly one maximal ideal, we
obtain p(A) = 4>(c,oA(Rad(A))) = 4>('!jJ(Ker(g))) = Ker(g). Therefore,
to prove the theorem it is sufficient to prove the following
Claim. There is an .e-group isomorphism h: GA ~ A(Ker(g)) such that
h(UA) = (n -l,v), for some 0 < v E Ker(g).
To prove the claim, we first show that there is 0 < z E GA such that
g( z) = 1 and (n-l)z < UA. Indeed, since 9 is surjective, there is y E GA
such that g(y) = 1. Note that y > 0, for otherwise g(y) ::; g(O) = 0,
which is impossible. If (n - l)Y < UA, then we can put z = y. If
(n-1)y = UA, then we can put z = y-t, where tE Ker(g), t > 0 (note
that t < y, for otherwise g(y) ::; g(t) = 0, a contradiction). Finally, if
(n - l)y > UA, then (n - l)y = U + 8, for some 0 < 8 E Ker(g), and
we can put z = Y - 28. To complete the proof of the claim it is enough
to let h be defined by h(x) = (g(x), x - g(x)z), for each x E GA, and
v = UA - (n - l)z. 0
8.3. MV-GRAINS OF FINITE RANK 163
Lemma 8.3.2 Por each integer n 2:: 2 and each nonsimple MV-chain
A of order n, Rn E C(A).
Proof: By Lemma 8.3.1 we can assurne A ~ r(A(G), b) for some to-

tally ordered group G and 0 < bEG. The mapping h: A(Z) -+ A(G)
defined by h((j, m)) = (j, mb), for each (j, m) E Z x Z, is an inject-
ive l-group homomorphism and h((n - 1,1)) = (n - 1, b). Therefore
r: Rn -+ r(A(G), (n - 1, b)) is an injective homomorphism, whence it
follows that Rn E C(A). 0
For all integers k and n, n 2:: 2, let the map h: A(Z) -+ A(Z) be
defined by h((j, m)) = (j, (n - l)m - kj). Then h is an injective l-
group homomorphism with h((n - 1, k)) = (n - 1,0). It follows that_
r(h): r(A(Z), (n - 1, k)) -+ Kn
is an embedding, whence
(8.4) for all integers k and n, n 2:: 2, r(A(Z), (n - 1, k)) E C(K n ).

This result is considerably strengthened by the following proposi-
tion:
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 .
Proof: We need the following c1assical result from the theory of l-

groups:
Por any lattice-ordered abelian group G there exists a nonempty set I,
an l-subgroup H of the l-group ZI and an l-group homomorphism from
H onto G.
We sketch here a proof for the reader's convenience. Since G is a
subdirect product of totally ordered abelian groups (see the proof of
the Claim in Lemma 7.1.5), one can easily adapt the proof of the Com-
pleteness Theorem (Theorem 2.5.3) to show that an l-group equation
holds in all l-groups Hf it holds in Z. For each nonempty set I let H
be the l-subgroup of the l-group ZI generated by the set P of pro-
jection functions '7ri: ZI -+ Z, for all i E I. Arguing as in the proofs
of Lemma 3.1.2 and Proposition 3.1.4, one immediately shows that H

has the following property: Each function from the set P to an f-group
G can be uniquely extended to an f-group homomorphism from H to
G. The desired conclusion now follows by letting I be an arbitrary
set whose cardinality is greater or equal than that of G (compare with
Proposition 3.1.5).
Returning to the proof of the proposition, suppose that G is an f-
group and 0 < bEG. Let H an f-subgroup of ZI and h: H ~ G be
a surjeetive f-group homomorphism. Then there is c E H sueh that
h( c) = b; replacing, if neeessary, c by c V 0, we ean safely assume c > O.
It follows that A(h): A(H) ~ A(G) is a surjeetive f-group homomor-
phism and A(h)((n - 1, c)) = (n - 1, b). Henee
r(h): r(A(H), (n - 1, c)) ~ r(A(G), (n - 1, b))
is a surjeetive homomorphism. Sinee His an f-subgroup of ZI, eaeh ele-

ment of xE H is a function x: I ~ Z. The funetion f: A(H) ~ A(Z)I
defined by f((m, x))(i) = (m, x(i)), for i E I, x E Hand mEZ,
is an injeetive f-group homomorphism. Let J = {i E I I c(i) i= O}.
Sinee c > 0, J is nonempty; moreover, i E J iff c(i) > O. There-
fore r(A(H), (n - 1, c)) is isomorphie to a sub algebra of the produet
algebra
IIiEJr(A(Z), (n - 1, c(i))).
By (8.4), for eaeh i E J, we get r(A(Z), (n - 1, c(i))) E C(K n ). There-
fore,
r(A(Z), (n - 1, c)) E C(Kn ).
Sinee the MV-algebra r(A(G), (n - 1, b)) is a homomorphic image of
r(A(Z), (n - 1, c)), we finally get
r(A(G), (n - 1, b)) E C(K n ),
as required to eomplete the proof. 0
Lemma 8.3.4 For each integer n 2:: 2, K n E C(H n ).
Proof" Suppose that an equation does not hold in K n . By the remark

following Definition 1.4.4 we can assume that the equation has the form
8.3. MV-GRAINS OF FINITE RANK 165
Xk) = 0, for some MV-term 7r in the variables Xl! ... , Xk.

7r(XI, ••• ,
There are elements Cl, ... ,Ck E Kn such that
(8.5) 7r Kn (CI' ... ' Ck) ~ (0,1).
For each integer m ~ 1, let fm: A(Z) -+ A(Z) be the f-group homomor-
phism defined by fm((v, w)) = (v, mw) for all pairs (v, w) E Z x Z. For
every MV-term a in the variables Xl! ... ,Xk let g(a) denote the total
number of symbols..., and EB occurring in a. By induction on g(a),
for each k-tuple (bl! ... , bk ) E Kn ~ Rn we shall prove the following
inequalities:
(8.6) fm(aKn(b l , ... , bk )) - (0, g(a))
(8.7) ~ aHn (fm(b l ), ... , fm(b k ))
(8.8) ~ fm(aKn(b l , . .. , bk )) + (0, g(a)).

If g(a) = 0, then a = Xi, for some i E {1, 2, ... , k} or a = 0. In this
latter case, the result is obvious. In the former case, if bi = (v, w) then
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,
= (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)).
From these inequalities one immediately sees that u satisfies (8.6).

In Case 2 we have g(u) = g(J.L) + g(lI) + 1. By induction hypothesis,
uHn(fm(b)) = (n - 1,1) A (J.lHn(fm(b)) + lI Hn Um(b)))
::; (n - 1,1) A Um (J.lK n(b)) + fm (lIK n(b)) + (0, g(J.l)) + (0, g(lI)))
::; (n - 1, g(J.l) + g(lI) + 1)
A(fm(J.lKn(b)) + fm (lIK n(b)) + (0, g(J.l) + g(lI) + 1))

= ((n - 1,0) A (fm (J.lK n(b)) + fm (lIK n(b))) + (0, g(J.l) + g(lI) + 1)
= fm(uKn(b)) + (0, g(u))
and
uHn(fm(b) )
~ (n - 1,1) A (fm(J.lKn(b)) + fm (lIK n(b)) - (0, g(J.l)) - (0, g(lI)))
~ (n - 1,1 - g(J.l) - g(u))
A(fm(J.lKn(b)) + fm (lIK n(b)) - (0, g(J.l)) - (0, g(lI)))
= ((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
2: ((n - 1,0) A (fm(J.lKn(b)) + fm (l/K n(b))) - (0, g(J.l) + g(a) + 1)

= /m(aKn(b)) - (O,g(a)).
Rence the inequalities (8.6) also hold in this case. From (8.5) and (8.6)
we obtain
(0, m - g(7r)) ~ fm(7r Kn (b 1 , ••• , bk )) - (0, g(7r))
~ 7r Hn (fm(b1 ), .•. , /m(bk )).
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
Theorem 8.3.5 For each integer n 2: 2 and each nonsimple MV-chain

A 0/ rank n, C(A) = C(K n ). 0
8.4 Komori 's classification

The two formulas (8.1) and (8.3) show that, for each (j, a) E K n =
r(A(Z), (n - 1,0)) and integer k 2: 1,
(kj, ka) if kj < n - 1 or kj = n - 1 and
(8.9) k.(j, a) = { a<0
(n - 1,0) otherwise
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.
Lemma 8.4.1 Let n 2: 2 be an integer and A an MV-chain. Then A

has rank ~ n i/ and only if A satisfies the equation
Proof: Let (j, a) E Km. Whenever 0 ~ j ~ m - 2, from (8.10) it

follows that (j, a)m-l E Rad(Km ), (j, a)m = (0,0), and (m -1, b)m-l =
(m -1, b)m = (m -1,0). Hence the equation (Fno ) holds in the algebra
Km for m = 2, ... , n; further, by Theorem 8.3.5, (Fno ) holds in all
MV-chains of rank m ~ n. Since AI Rad(A) E C(A), to complete the
proof it is sufficient to show that for each m > n, the MV-algebra L m
does not satisfy equation (Fno ). To this purpose, we must prove that
there is t E L m such that n.tn - 1 = 1 and 2.tn < 1. Note that for
each t E L m , n.tn - 1 = 1 iff n((n - l)t - (n - 2)) ~ 1 and 2.tn < 1 iff
2(nt - (n -1)) < 1. Therefore, for each m > n we must find an integer
k such that
n -1 k 2n -1
-n- < - m-l- < 2n .
In case n < m ~ 2n we can take k = m - 2. In case 2n < m, there are

integers q and r such that q ~ 1, 0 < r < 2n and m = 2nq + r. Thus,
letting k = m - (q + 2) we obtain the desired conclusion. 0
Theorem 8.4.2 For each nontrivial MV-algebra A and integer n ~ 2,

the following conditions are equivalent:
(i) A satisjies the equation
(ii) A E C( {K 2 , K 3 ,· •. , K n });
(iii) AI Rad(A) E C( {L 2 , L 3 , ... ,Ln})'
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
Corollary 8.4.3 For each nontrivial MV -algebra A we have:

8.4. KOMORI'S CLASSIFICATION 169
(i) A E C(K 2 ) iff A satisfies the equation
(ii) A E C(K 3 ) iff A satisfies the equation
Proof: (i) is an immediate consequence of Proposition 8.4.2, and (ii)

follows from the same proposition and the fact that K 2 is a subalgebra
of K3 • 0
Remark: It follows from (i) in the above corollary that alt per/ect MV-
algebras are in C(K 2 ).
From Proposition 8.1.4 we know that each proper variety of MV-
algebras is generated by a set of MV-chains of bounded rank. From
Theorems 8.2.2 and 8.4.2 we conclude that, conversely, each set of MV-
chains of bounded rank generates a proper variety. By Theorem 8.3.5,
we then obtain
Theorem 8.4.4 A class C 0/ MV-algebras is a proper variety iff there

are two finite sets land J 0/ integers ~ 2 such that Iu J is nonempty
and C = C({LihEI, {KjhEJ).O
Notation: For each integer n ~ 2 we let
Div(n)
be the set of all divisors d ~ 1 of n.
Lemma 8.4.5 Let m and n be integers such that n ~ m ~ 3. For each

integer p such that 2 ::; p < m - 1, the MV-algebra Km satisfies the
equations:
(Enp ) (p.xp-1)n = n.xp
and
(Fnp ) n.xp = (n + l).xp

if and only if p does not divide m - 1.
Proof: Let y = (j, a) be an arbitrary element of Km. Let us assurne

p ~ Div(m-1). If yP-l = 0, then yP = 0, whenee (p.yp-l)n = n.yP and
n.yP = (n + l).yp. If, on the other hand, yP-l > 0, then there are two
subeases to eonsider:
If yP = 0, from (8.10), together with our standing hypothesis that p
does not divide m - 1, it follows that jp < (m - 1)(P - 1). Therefore,
p(p - l)j - p(p - 2)(m - 1) ~ m - 2. Henee from (8.9), we have
p.yP-l = (p(p _ l)j - p(p - 2)(m - l),p(P - l)a)

~ (m - 2,p(p - l)a),
and again by (8.10), (p.yp-l)n = 0 = n.yp.

If yP > 0 then jp > (m - l)(p - 1), whenee p.yp-l = 1 = yP and
n.yP = n.(pj - (p - l)(m - l),pa) = (m - 1,0) = (n + l).yp.

Therefore both equations (Enp ) and (Fnp ) hold in Km.
Conversely, let us assurne that n 2: p 2: 2. If y = 7' then yP = 0
and yp-l = ~. Therefore L p +1 does not satisfy equation (En(p_l)),
and sinee Lp +1 is isomorphie to the subalgebra {(O, 0), (1,0), ... , (p, O)}
~ Kp+1, we also obtain that equation (En(p_l)) does not hold in Kp+1'
On the other hand, from z = (p, -1) E Kp+1 and zP = (p, -p) we get
n.zP < (n + l).zp, whenee equation (Fnp ) does not hold in Kp +1 . To
eomplete the proof it is sufficient to note that if p E Div( m - 1) then
K p +1 is a sub algebra of Km. 0
Theorem 8.4.6 Let land J be sets of integers 2: 2 such that IU J is

nonempty; suppose that whenever r, sEI u J and r < s then r - 1 ~
Div( s -1). Let n be the greatest element of I U J. Let A be a nontrivial
MV-algebra. Then
A E C({LihEI, {KjhEJ)
iff A satisfies the equations
(Fno ) (n.x n- 1)2 = 2.xn ,

8.5. VARIETIES GENERATED BY A FINITE CHAIN 171
(Enp ) (p.xp-1)n = n.xp,

for each integer p such that 2 ~ p < k and p ~ UrEIUJ Div(r - 1),
as well as the equations
(Fnq ) k.x q = (k + 1).xq,
for each integer q 2: 2 such that
q E U(Div(r -1) \ U Div(s -1)).

rEI SEJ
Proof: Let C be the dass of all MV-algebras satisfying equations (Eno ),

(Enp ), for each integer p such that 2 ~ p < k and p ~ UrEIUJ Div(r-l),
and (Fnq ), for each integer q 2: 2 such that q E UrEI(Div(r - 1) \
UsEJ Div(s -1)). It follows from Lemmas 8.4.1 and 8.4.5, that Km E C
if and only m -1 divides s -1 for some s E J. Let r E land y E L r +1 .
Since yq E Lr +1, by Theorem 8.2.2 we have (r - 1).yq = r.yq, whence
a fortiori, n.yq = (n + 1).yq. Since for each m, L m is isomorphie to a
subalgebra of Km, by Lemma 8.4.5 we can now condude that L m is a
member of C if and only if k - 1 divides r - 1 for some r E I U J. Hence
by Theorem 8.4.4, C = C({LihEI, {KjhEJ). 0
8.5 Varieties generated by a finite chain

Corollary 8.2.4 gives a simple equational characterization of the vari-
eties of MV-algebras generated by L 2 and L 3 • The following theorem,
which is an immediate consequence of Theorems 1.4.6 and 8.2.2 to-
gether with Lemma 8.4.5, gives a set of defining equations for each
variety generated by a finite simple MV-algebra, in all remaining cases.
These equations are simpler than those obtainable by letting I = {n}
and J = 0 in Theorem 8.4.6.
Theorem 8.5.1 For every integer n 2: 4 and every MV-algebra A, the

following conditions are equivalent:
(i) A satisfies the equations
(Eno ) (n - l).x = n.x
and
(Enp ) (p.xp-1)n = n.xp.
/or every integer p = 2, ... , n - 2 that does not divide n - 1;
(ii) A is a produet 0/ subalgebras 0/ Ln;

(iii) A E C(Ln ). 0
Definition 8.5.2 Algebras in the variety C(L n ) are called n-valued

MV-algebras, or, for short, MVn-algebras .
Corollary 8.5.3 Let A be an MV-algebra.

(i) 1/ A is a finite MVn -algebra, then A is isomorphie to the direet
produet 0/ a /amily 0/ subalgebras 0/ Ln;
(ii) 1/ A is finite then there is an integer n ~ 2 sueh that A E C(L n );

(iii) 1/ A is isomorphie to the direet produet 0/ a finite /amily 0/ sub-
algebras 0/ Ln then the members 0/ this /amily eoineide with the
images 0/ A under all possible homomorphisms 0/ A into Ln .
Proo/: (i) By Proposition 3.6.5, Ais isomorphie to the direct product of

a finite family of finite chains. Since any such chain L is a homomorphic
image of A, L E C(L n ). By Theorem 8.5.1(ii), L must be a subalgebra
ofLn ·
(ii) By Proposition 3.6.5, we can write
Let s be the least common multiple of nl - 1, ... , n r - 1, and n = s + 1.

By the remark following Proposition 3.5.3 each of L n1 , .•• ,Lnr is a
subalgebra of Ln. Thus, A E C(L n ).
(iii) Let J be an ideal of A. Then a fortiori J is an ideal of the
underlying lattice L(A). By our finiteness assumption, there is a E A
such that J = (a 1 = {x E A I x ~ a}. By Corollary 1.5.6, a E
B(A). It is easy to see that J is a maximal ideal of A iff there is
8.6. THE CARDINALITY OF F REEff. 173
an atom b of B(A) sueh that J = (-.b]. Let h : A - SI X ... X Sr

an arbitrary isomorphism, where SI," ., Sr are all subalgebras of Ln.
Let PI,'" ,Pr be the compositions of the canonical projections of L~
onto Ln with h. Trivially, Pi is a homomorphism of A into Ln. Furt her ,
for each i = 1, ... , r, it is easy to see that pi(A) = Si' Thus, for each
i there is an atom bi in B(A) such that Ker(pd = (-.bi ]. It follows
that (-.b l ] n ... n (-.br ] = (0], i.e., b1 V ... V br = 1. In conclusion,
b1 , ... ,br exhaust all possible atoms of B (A), and PI, ... ,Pr are all
possible homomorphisms of A into Ln. 0
An MVn-algebra A, with a distinguished subset Y of elements, is

said to be a free MVn-algebra over the (generating set) Y , and is
denoted by Freey, iff for every MVn-algebra Band every function
f: Y - B, f ean be uniquely extended to a homomorphism of A into
B.
As in the ease of free MV-algebras (cf. Section 3.1), there is no
danger of ambiguity in calling Free~ the free MVn-algebra over fi, many
generators.
Since for MV-terms p, q, p Ln = qL n iff pA = qA for each MVn-
algebra A, arguing as in the proof of Proposition 3.1.4 we obtain:
Proposition 8.5.4 For each natural number n ~ 2 and cardinal fi, ~

1, Free~ = Term(L n , fi,). 0
8.6 The cardinality of Free~

When fi, is a finite cardinal, say fi, = r ~ 1, L~: is finite, and has in
fact nn elements. Therefore by the above Proposition 8.5.4 together
T
with Corollary 8.5.3, Free~ is isomorphie to a direct product of a finite

number of subalgebras of Ln. In this seetion, for each rand n we shall
determine how many times each subalgebra of Ln appears in this direct
product.
Let Y = {Yb"" Yr} be a generating set of the free MV n-algebra
Free~. There are n r functions from Y into Ln, denoted by el,' .. , enr.
Let Si be the smallest subalgebra of Ln eontaining all ei (Yl), ... , ei (Yr),
i = 1,2, ... , n r • Letting hi = hei be the unique homo mo rphi sm of
Free~ into Ln extending ei, by Proposition 8.5.4, h l , .. . ,hnr is the list

of all homomorphisms of Free~ into Ln; further, hi(Free~) = Si, for
each i = 1, ... , n r . Thus, by Corollary 8.5.3(iii) we have
Free~ = SI X ..• X Snr.

For each i = 1, ... ,nr and j = 1, ... , r, we define
aij
ei(Yj) = (n - 1)'
Recall that Div(n), (n 2: 2) denotes the set of all divisors d 2: 1 of
n. Given integers nl," ., nk 2: 1, we shall denote by gcd(nl" .. , nk)
their greatest common divisor. By Proposition 3.5.3 together with the
remark following it, the set of subalgebras of Ln coincides with the
set of algebras Ld+l, where d E Div(n - 1). Obviously, a subalgebra
Si coincides with L 2 if and only if, for all j = 1, ... , reither aij = 0
or aij = n - 1. Further , for each d < n - 1 we have Si = Ld+l iff
gcd(ail,"" air, n - 1) = (n - 1)jd. Since el,"" enr lists all functions
of {Yl'" ., Yr} into Ln, for any given d E Div(n -1) there is an i such
that aij = (n - 1)jd for all j = 1, ... , r. Such i also has the property
that Si = L d+l . Therefore, all subalgebras of Ln are included among
the Si 's, and we have
F reern = TI deDiv(n-l) LQ(n,r,d)

d+l ,
where a(n, r, 1) = 2 , and for each d > 1, a(n, r, d) is the number of r-
r
tuples (ail,' .. , air) E {O, 1, ... ,n -1 Y such that gcd(ail, ... ,air, n -1)
= (n -1)jd.
Let us fix n 2: 3, r 2: 1 and 1 =I d E Div(n-1). In order to compute
the number a(n, r, d), we shall first introduce some notations. For each
a and b in Div(n - 1), by writing
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
MaxDiv(c) = {a E Div(c) la -< cl,

8.6. THE CARDINALITY OF FREE:' 175
Min(c, n - 1) = {a E Div(n - 1) I c -< a},

and
H(c) = {i E {I, ... , n r } I cE Div(ail) n ... n Div(airn.
Let us note that MaxDiv(c) = 0 Hf c = 1; moreover, Min(c, n-I) = 0
iff c = n - 1. Further, i E H(c) iff there are integers 0 :5 b1 , ... ,br such
that bjc = aij :5 n - I,for all j = 1, ... , T. It follows that ~H(c) =
(n~l + IY. We then have
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.
Theorem 8.6.1 For any two integers n 2: 2 and r 2: 1 we have
Free~ = rr
dEDiv(n-l)
La:(n,r,d)
d+l ,
where
a(n, r, d) - (_I)~X (gcd(X) + Ir. 0

X~MaxDiv(d)
From the above theorem, we get in particular Free~ = Lr, which is

a weIl known characterization of the free boolean algebra with r free
generators. Furt her , whenever n - 1 is prime we get
A simple counting argument shows that ijFreer = 4 = number of func-

tions from L 2 into L 2 •
As another example, while the number of functions from L 3 into L 3
is 27, a simple computation shows that ijFreef = 12.
In the rest of this section we shall prove some interesting and less
immediate consequences of Theorem 8.6.l.
Let 'Jr be an MV-term in the variables Xl, ... ,Xn and suppose that
S is a subalgebra of an MV-algebra A. From the definitions given in
Section 1.4, it follows that for all al, . .. , ar E S
'JrA(al' ... ' ar ) = 'JrS(al' ... ' ar ) E S.

Therefore, 'JrA(sr) ~ S. For short, we say that term functions preserve
subalgebras.
Since the subalgebras of Ln are precisely the algebras L d+ l , for d E
Div( n-l), we can say that a function I: L~ -+ Ln preserves subalgebras
if and only if 1 satisfies the following condition:
(*) For each r-tuple (n~l'···' n':l) E L~, il I(n~l'···' n':l) = n~l'
then b is a multiple 01 gcd(al' ... , ar , n - 1).
Conversely, the next result shows that if a function I: L~ -+ Ln pre-
serves subalgebras, then 1 is a term function. More precisely we have
Corollary 8.6.2 Given any arbitrary integers n 2: 2 and r 2: 1, let

f: L~ ---jo Ln· Then fE Term(L n , r) iff f preserves subalgebras.
Proof: Let 2 :5 d 1 < d2 < ... < ds = n - 1 display the divisors of n - 1

other than 1. Let (bil , ... , biT)' 1 :5 i :5 n T display all r-tuples in L~,
where the indexes i are so chosen that
• precedence is given to the 2T r-tuples that generate the subalgebra

L 2 of Ln,
• followed by the a(n, r, d2 ) many r-tuples that generate the subal-
gebra L d2 +1!
• and so on.
Let D be the subalgebra of L~ given by those functions satisfying

condition (*). Let E be the product algebra
rv T
11 l$i$s L o(n,T,d;)
d; = erm (Ln, r ) .
It follows that the correspondence
is an isomorphism of D onto E. 0

The study of varieties of algebras is one of the main concerns of univer-
sal algebra. See, for instance, [104] or [150]. Proposition 8.1.1 originates
from the early logico-algebraic Polish School (see [139]). Proposition
8.1.2 is due to Tarski [139, Theorem 20].
The notion of rank was introduced by Komori [130], who also proved
the main results of Section 3. To prove Theorem 8.3.1 Komori intro-
duced a special dass of ordered abelian groups. The proof given here,
based on the properties of the variety of f-groups is borrowed from [50].
After Komori's dassification of all varieties of MV-algebras, con-
siderable work has been done, by Gispert, Torrens, Panti and others,
on their axiomatization, and on the classification and axiomatization

of classes generated by MV-chains (see, e.g., [94], [95] [67], [68], [199]).
The present equational characterization of all proper varieties of MV-
algebras is due to Di Nola and Lettieri. MVn-algebras were introduced
by Grigolia [105], [106], who also gave their equational characterization.
See also [212]. The variety generated by K 2 was considered in [7] and
[65].
The description of the free MVn-algebra given in Theorem 8.6.1 was
announced by Ant6nio Monteiro in the VI Congress of Mathematicians
of Latin Expression, held in Bucharest during September 1969.
The case n = 3 of equation (8.11) was established by A. Monteiro
in his lectures delivered at the Universidad Nacional deI Sur, Bahia
Blanca, in 1963. The case n = 4 was first obtained in [44].
Corollary 8.6.2. was first proved by McNaughton in [152]. Our proof
is obtained by means of direct techniques from universal algebra.
A brute force method to establish the nonisomorphism of finite al-
gebras is to count their elements. For instance, as proved in [44), the
5-valued Moisil-Lukasiewicz algebra over one free generator has 192 el-
ements, while, by (8.11), the free MV5-algebra over one generator has
300 elements.
Chapter 9
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
180 CHAPTER 9. ADVANCED TOPICS
9.1 McNaughton's theorem

For each (closed) n-dimensional simplex S S; [0, l]n with rational ver-
tices we let
(I) VA, VI,. •. , V n be the list of vertices of S;
(11) 1 ~ di = least common denominator of the coordinates of Vi;

(111) Vij be the uniquely determined integers satisfying
with gCd(ViO, ... , Vi(n-I) , di ) = 1, where, as in the previous chap-

ters, gcd denotes greatest common divisor;
(IV) vfom E zn+! be the homogeneous coordinates of the vertices of S,
(9.2) v?om = (ViO, . .. , Vi(n-I) , di ), i = 0,1, ... , n;
(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).
Following tradition, the set
(9.3) Ps = {z E Rn+! I Z = Aov~om+ .. .+Anv~om, °~ Ai < 1, Ai E R}
shall be called the half-open parallelepiped of S.
Definition 9.1.1 Adopt the above notation and terminology. Then S

is said to be a unimodular simplex Hf det(Ms) = 1. By a unimod-
ular triangulation of [0, l]n we mean a set W of closed n-dimensional
unimodular simplexes with rational vertices, such that the union of all
simplexes in W coincides with [O,I}n, and any two simplexes in Ware
either disjoint or intersect in a common face.
The following result strengthens Proposition 3.3.1:

9.1. MCNAUGHTON'S THEOREM 181
Theorem 9.1.2 Let 1 : [O,l]n -+ [0,1] be a McNaughton function,

with its linear constituents PI! ... ,Pk. Then there is a unimodular tri-
angulation U 01 [0, l]n such that lor each simplex S EU, 1 coincides
with some Pj over S.
Proof: Let S be as given by Proposition 3.3.1. Adopting the above

notation, let T be an n-dimensional simplex in S such that det(M T )
is maximum. Let VO, ••• , V n be the positively oriented list of vertices
of T. If det(MT) = 1 we are done. Otherwise, det(M T ) > 1 and
(MT )-1 is not an integral matrix.
Let eo, ... ,en be the standard basis vectors of Rn+!: thus, eo =
(1,0,0, ... ,0), el = (0,1,0, ... ,0), ... , e n = (0,0,0, ... ,1). We can
safely assume that at least one of the coordinates of eo in the basis given
by the vectors v~om, ... , v~om, is not an integer. Among all vectors that
are obtainable by adding to eo linear combinations with integral coef-
ficients of the vectors vfom, there is precisely one, denoted wERn+ 1 ,
(read: w = eo modulo PT), lying in the half-open parallelepiped PT'
We have shown
°=J. w = (wo, .. ·, W(n-l), wn ) E zn+l n PT.
Let VT be the inhomogeneous correspondent of w, in symbols, VT =

(wold, . .. , W(n-l)/d), where d = Wn = last co ordinate of w. Trivially,
VT is a point of the n-simplex T, and is not one of its vertices. For
any (n - l)-dimensional face F of T not containing VT, the point VT
together with the vertices of F determines a new n-dimensional simplex
[VT, F] ~ T. Let M = M[VT,F) be the associated matrix. Since the
nonzero integral vector w belongs to the half-open parallelepiped PT,
by (9.3), the determinant of M is an integer satisfying the inequalities
(9.4) 1 ~ det(M) < det(M T ).
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
(ii) any two simplexes in SI are either disjoint or they intersect in

a common face.
The set of vertices of (simplexes in) SI is the same as in S, with
the only addition of VT. By (9.4), the number of simplexes R in SI
such that det(M R ) = det(M T ) is decreased. In this way, proceeding
by induction, we obtain a sequence of sets of n-dimensional simplexes
This sequence must terminate after a finite number Z of steps, yielding a

set Sz = U of unimodular n-dimensional simplexes having the required
properties. 0
Definition 9.1.3 Given a unimodular triangulation W of [0, l]n, and a

vertex v of some simplex in W, let d be the least common denominator
of the coordinates of v. Then the Schauder hat h v of W at v is the
continuous function h v : [0, l]n ~ [0, 1] determined by the following
conditions:
(i) hv(v) = lid;
°
(ii) hv(u) = for each vertex u E W different from v;
(iii) hv is linear over each simplex U E W.
We denote by 1iw the set of Schauder hats of W.
The star of v, in the triangulation W-in symbols, star(v)-is the
set of all n-dimensional simplexes of W having v among their vertices.
Proposition 9.1.4 Let W be a unimodular triangulation of [0, l]n, v

a venex in W, and h v be its corresponding Schauder hat. Then we have
(i) For any simplex S E star(v), let 9 be the linear polynomial coin-
ciding with h v over S; then there are integers ao, . .. ,a(n-l), b such that
for every x = (xo, ... ,X(n-l») E S,
(9.5) hv(xo, ... , X(n-l») = aoxo + ... + a(n-l)X(n-l) + b = g(x);

(ii) h v E Free n .
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
matrix (M S)-I; the latter is an integral matrix, by the unimodularity

ofW.
(ii) Let 91, ..• ,9q be the distinct linear polynomials given by (i).
Our aim is to show that h v coincides, over U star (v), with a (V A)-
combination of the 9i 'so To this purpose, for any permutation U of the
set {I, ... ,q}, we define the closed convex polyhedron Pu by
(9.6) Po' = {x E [0, ljn I 90'(1) (x) ~ 9u(2)(X) ~ ••• ~ 9u(q) (x)}.
Let {} be the set of permutations 7r such that PTr is n-dimensional. Then
(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
(9.8) hv = 9Tr(i.".) over PTr n Ustar(v).

Again keeping 7r fixed, let us define the function 9Tr : Rn - R by
(9.9) 9 7r = 97r(1) A ... A 9Tr(i.".).

Claim. 9 Tr ~ hv over each simplex W E star(v).
It is sufficient to prove the claim for all y in a suitably small open n-
dimensional ball B centered at v. By continuity, we can safely assume
y to lie in the interior of Pp for so me permutation p E [1. By (9.8)
there is an index i E {I, ... , q} such that hv = 97r(i) over Pp n B. If
i ~ i Tr , the claim trivially follows by definition of 9 7r . If i > i 7r , there is
a point x =1= y in the interior of PTr n B such that 9Tr(i) (x) < 9Tr(i.".)(X) =
hv(x). We may assume x to be so close to v that the line segment [x, y]
joining x and y lies in the interior of B n(U star(v)). Let X, Y E Rn+l
be defined by X = (x, hv(x)) and Y = (y, hv(Y)). Let [X, Y] denote
the line segment joining X and Y. Let 'fJ be the restriction of hv to
[x, y]. Then'fJ is a continuous function, consisting of finitely many linear
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
9.2 Nonsingular fans and normal forms

This section requires familiarity with the so called vocabulary from toric
varieties to fans. Writing in homogeneous coordinates the vertices of
each simplicial complex arising in the disjunctive normal form of a
McNaughton function, one obtains a sequence leading to a nonsingular
refinement of the fan (a fan being a complex of simplicial cones, as
defined below) corresponding to the linear domains of the proposition.
This refinement process in turn amounts to aresolution of singularities
for toric variety corresponding to the fan.
In more detail, let So, SI, ... , Sz = U be a sequence of simplicial
complexes over [0, l]n, as in the proof of Theorem 9.1.2. Each simplex
T E Si has vertices Vo, ... , V n . The homogeneous counterpart of T
is a simplicial cone, u T = (vo hom , ... , v n hom ), Le., the positive span
in Rn+l ofvectors vohom, ... ,vnhom. As a vector in zn+l, each Vjhom is
primitive, Le., minimal along its ray. The v/om are called the primitive
generating vectors of u T . From each Si we obtain a simplicial fan .6.i ,
i.e., a complex of simplicial cones. The fact that Sz is a unimodular tri-
angulation is equivalent to saying that the primitive generating vectors
of each (n + l)-dimensional cone in .6. z form a unimodular matrix-for
short, .6. z is a nonsingular fan.
As is weIl known, every (nonsingular) fan.6. is canonically associated
with a (nonsingular) toric variety XtJ., in such a way that the sequence
.6.0 , ... ,.6.z corresponds to a desingularization X z of the toric variety
X o. Thus, desingularizing a toric variety amounts to subdividing a
simplicial complex into a unimodular triangulation, precisely as is done
to compute DNF reductions of McNaughton functions. For the sake of
definiteness, recalling Definition 9.1.3, let us give the following
Definition 9.2.1 A Schauder set in [0, l]n is a set of the form H = H u

for some (necessarily unique) unimodular triangulation of [O,l]n. We
say that H' is a one-step star refinement of H iff it is obtained from
H as follows:
(a) Pick a set S = {h 1 , ••• , h q } ~ Hand let h s = h 1 1\ ... 1\ h q ;
(b) For each j = 1, ... , q replace hj by hj e h s ;
°
(c) If =I hs put hs in H'.
186 CHAPTER 9. ADVANCED TOPIes
Trivially, H = H' if hs = 0 or if S is a singleton. When S has two

elements and hs =J. 0 we say that the refinement is binary. Writing
H = H u , and assuming 0 =J. h s , it follows that H' = HU', where U'
is the unimodular triangulation obtained by starring U at the mediant
point determined by the face S ofU. We say that H* is a star refinement
of H iff it is obtained from H via a path H = Ho, H 1 , ... ,Ht = H*,
where each Hi is a one-step star refinement of H i - 1 •
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.
Proof: This is a consequence of the De Concini-Procesi theorem on

elimination of points of indeterminacy in toric varieties. 0
In the particular case when n = 1, using the Hirzebruch-Jung con-
tinued fraction algorithm, one can compute, for every fan ß in the
cartesian plane, the coarsest nonsingular subdivision of ß. In MV-
algebraic terms we have:
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
• (ii) Whenever a Schauder set.c is a DNF reduction o/Q, then.c

is also a DNF reduction 0/ H.
o
9.3 Complexity of the tautology problem

In this section we require some familiarity with the theory of NP-
completeness. We shall consider the following problem:
INSTANCE: A formula 4>.

QUESTION: Is 4> a tautology in the infinite-valued calculus
of Lukasiewicz ?
We shall prove that the problem is co-NP-complete, Leo, the com-

plementary problem of deciding whether a formula is not a tautology,
is NP-complete. Cook's theorem states the co-NP-completeness of the
tautology problem for the classical propositional calculuso
We shall denote by 14>1 the number of occurrences of symbols in 4>.
We shall also use the notation lxi for the absolute value of areal number
x. This will never cause any confusion. Recalling Propositions 301.8 and
4.5.5, for each formula 4> = 4>(X l , . .• , X n ) its associated McNaughton
function shall be denoted by f", = f",(Xl"" ,xn). For all x,y E [O,l]n,
the one-sided derivative of f", at x along direction d = y - x is defined
by
/.' ( . d) - l' f",(x + Ed) - f",(x)
'" x, - 1m
dO
0
Proposition 9.3.1 With the above notation, the directional derivative

f~(x; d) is well defined, and we have the inequality
(9.11) If~(x; d)1 ::; Ildll·I4>I,

where Ildll denotes euclidean norm in Rn.
Proof: The existence of f~(x; d) follows by definition of McNaughton

function. Inequality (9.11) is proved by induction on the number m of
connectives occurring in </>. The basis m = 0 is trivial. For the induc-

tion step, if </> = -,'ljJ for some formula 'ljJ, then the desired conclusion
immediately follows from the identity f I/> = 1 - f",· Finally, if </> = 'ljJ EEl X
then, assuming without loss of generality that both 'ljJ and X have the
same variables, the desired conclusion immediately follows by definition
of truncated addition, upon noting that fl/> = f", EEl fx' 0
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 ).
Proof: By Proposition 3.3.1 there is a finite number of distinct poly-

nomials PI, .. . ,Pm with integer coefficients, and a finite number of n-
dimensional simplexes VI!"" V s , whose union is [0, l]n, and such that
over each V j the function fl/> coincides with some Pi(j) ( i(j) = 1, ... , m).
Suppose without loss of generality, x E VI and fl/>(X) > O. Then we
can safely assurne x to be a vertex of VI' Therefore, the coordinates
of x are all rational, say x = (aI/b, ... , an/b), for suitable integers ai
and b with 0 ::; ai ::; b. Moreover, by the above corollary, x is the
solution of a system of n linear equations in n unknowns, and each row
has its coefficients ::; 21</>1. Since, trivially, n ::; 1</>1 then by Hadamard's
inequality we conclude that the determinant /:). of this system satisfies
the inequality
Since b::; 1/:).1, the desired conclusion immediately follows. 0

9.3. COMPLEXITY OF THE TAUTOLOGY PROBLEM 189
Theorem 9.3.4 The tautology problem for the infinite-valued calculus

is in the class co-NP.
Proof" A nondeterministie proeedure quiekly deciding if a formula
is not a tautology is as follows: Firstly, applying Proposition 9.3.3 to

the function 1 - fljJ, guess a rational point
x = (aI/b, ... , an/b) E [O,l]n

°
such that fljJ(x) < 1 and < b < 2(411/>1 2 ). Seeondly, for the purpose of
eheeking that fl/>(x) < 1, write eaeh eoordinate ai/b as a pair of binary
integers; let [ai] and [bI denote the number of bits of ai and b. Note
that [ai] ~ [b] ~ 414>1 2 for all i = 1, ... , n; also note that, onee x is
written down as a sequenee of pairs of binary numbers, its length [x]
will satisfy the inequalities
Sinee the operations of negation and truneated addition do not inerease

denominators, for some polynomial q : N ~ N the value fljJ(x) ean be
eomputed by a deterministic Thring maehine within a number of steps
~ q(I4>I). 0
In order to prove that the tautology problem in the infinite-valued

ealculus is eo-NP-hard, we prepare:
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
fnAX) = {(Xl V -,xd 0 ... 0 (Xl V -'XI)} 0 ...

, v '
t times
t times
Further, for eaeh integer i 2: 1 the formulas 4>i, 'l/Ji,t, and Pn,t are
defined by
(i) 4>i = Xi V..,Xi ;

(ii) 'ljJi.t = 4>i 0 ... 04>i (t times);
(iii) Pn.t = 'ljJl.t 0 ... 0 'ljJn.t.
As an immediate consequence of the definition we have fn.t = !Pn,t'

A tedious but straightforward inspection yields the following
Lemma 9.3.6 Fix an enumeration of the vertices of the cu be [0, l]n.

Let Vj be the jth vertex (j = 1, ... , 2n). Let
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.
Proof: For the nontrivial direction, by definition of implication, we

must prove
fn.t $ fl/l'
With reference to Lemma 9.3.6(iv), the inequality holds over the set
[0, l]n\ Uj Tj. So let us assume that for some j = 1, ... , 2n, fn.t(x) >
fl/l(x) with x E Tj (absurdum hypothesis). By continuity we can safely
assume x to be in the interior of Tj, whence in particular, x i= Vj' Let
9.4. MV-ALGEBRAS AND AF C*-ALGEBRAS 191
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],
f~,t(Y; u) :5 -14>1 :5 f~(Y; u).

Since by assumption, 4> is a boolean tautology, ftf> = 1 on {O,1}n. By
Lemma 9.3.6,
fn,t(Vj) = ftf>(vj) = 1.
Now, fn,t is linear on the interval [Vj, x]; on the other hand, ftf> is
(continuous and) piecewise-linear on [Vj, x]. Thus ftf> ~ fn,t on [Vj, x],
a contradiction. 0
Theorem 9.3.8 The tautology problem in Lukasiewicz infinite-valued

calculus is co-NP-complete.
Proof: We have just given a polynomial-time reduction of the boolean

tautology problem into the tautology problem for the infinite-valued
calculus. The desired conclusion is now an immediate consequence of
Theorem 9.3.4 and the above Lemma 9.3.7, in the light of Cook's NP-
completeness theorem for the boolean satisfiability problem. 0
9.4 MV-algebras and AF C*-algebras

This section requires some familiarity with AF C" -algebras. Every
C"-algebra A considered in this paper shall have a unit element 1A.
By a projection p in A we mean a self-adjoint idempotent p = p. = p2.
Up to isomorphism, the most general possible finite-dimensional C"-
algebra :F is a finite direct sum Md(l) + M d(2) + ... +Md(t) where
Md(i) denotes the C"-algebra of all d(i) x d(i) complex matrices, for
suitable 1 :5 d( i).
192 CHAPTER 9. ADVANCED TOPfCS
An approximately finite-dimensional (for short, AF C*-algebra) is

the norm closure of the union of a sequence F 1 ~ F 2 ~ ..• of finite-
dimensional C*-algebras, all with the same unit, where each F i is a
*-subalgebra of F i +1'
For every AF C*-algebra A, two projections p, q E A are said to
be equivalent Hf there exists an element v E A such that vv* = p and
v*v = q. We denote by [P] the equivalence class of p, and by L(A)
the set of equivalence classes of projections of A. The Murray-von
Neumann order over L(A) is defined by
[P] ::; [q] iff p is equivalent to a projection r such that rq = r.

Elliott's partial addition, denoted +, is the partial operation on L(A)
given by adding two projections whenever they are orthogonal. Then
+ is associative, commutative, monotone, and satisfies the following
residuation property:
(*) 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].
Theorem 9.4.1 For every AF C*-algebra A we have:

(i) There is at most one extension 0/ Elliott 's partial addition to an
associative, commutative, monotone operation $ over the whole L(A)
having the above residuation property (*). Such extension $ exists iff
L(A) is a lattice;
(ii) Let K(A) = (L(A), [0],..." $). Then the map A f-t K(A) is a
one-one correspondence between isomorphism classes 0/ AF C*-algebras
whose L(A) is lattice-ordered, and isomorphism classes 0/ countable
MV-algebras;
(iii) In particular, up to isomorphism, the map A f-t K(A) is a one-
one correspondence between commutative AF C*-algebras and count-
able Boolean algebras. The inverse correspondence is given by the map
X t-+ C(X), the latter denoting the C*-algebra 0/ complex-valued con-
tinuous junctions over an arbitrary separable totally disconnected com-
pact Hausdorff space X. 0
9.5. DI NOLA '8 REPRE8ENTATION THEOREM 193
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).
9.5 Di Nola's representation theorem

This section requires familiarity with model theory.
Theorem 9.5.1 Up to isomorphism, every MV-algebra Bis an algebra

0/ [0, l]*-valued junctions over some set, where [0,1]* is an ultrapower
0/ [0,1], only depending on the cardinality 0/ B.
Proof" Let P( B) be the set of prime ideals of B. In the light of
the Subdirect Representation Theorem 1.3.3, let us embed B into the
MV-algebra II{B/I I I E P(B)}. For each prime ideal I of B, in
the light of Theorem 7.1.7, let G(I) be the totally ordered abelian
group with strong unit u(I) uniquely determined by the stipulation
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

McNaughton's representation theorem 9.1.5 was first proved in [152], a
few years before the advent of MV-algebras. The present proof, using
unimodular triangulations and Schauder hats, was first given in [181].
Schauder hats over n-dimensional unimodular triangulations were in-
troduced in [170]. Generalized Schauder hats are a standard tool in the
theory of Banach spaces.
Ewald's book [82] gives a full account of the vocabulary between
fans and toric varieties. Also see [195], where toric desingularizations
are discussed, along with the various results of Danilov, De Concini-
Procesi and Hirzebruch-Jung mentioned in this chapter. See Morelli
[162] for a solution of the strong form of Oda's conjecture.
The relationship between toric desingularizations and the infinite-
valued calculus of Lukasiewicz was first investigated in [183]. Direct
applications to three-dimensional toric varieties are given in [5].
Theorem 9.3.8 was first proved in [167]. Any textbook on computa-
tional complexity theory contains a proof of Cook's NP-completeness
theorem. A standard reference is [88].
See [77] and [101] for Elliott's theory, and for the role of Grothen-
dieck's functor K o in connecting AF C"-algebras and partially ordered
abelian groups. AF C"-algebras are a standard tool to give a math-
ematical description of spin systems in quantum statistical mechanics

(see the references in [178]). The mutual relations between MV-algebras
and AF C*-algebras were first investigated in [163]. Theorem 9.4.1 was
proved in [190] building on [163]. Gödel incompleteness phenomena for
AF C*-algebras are discussed in [163]. The fact that many weH known
examples of AF C*-algebras are coded by polynomial-time theories in
the infinite-valued calculus is noted in [166] and [177].
Theorem 9.5.1 is due to Di Nola, [63], [64]. The model theoretic
machinery needed for its proof can be found in [42] and [158].
Chapter 10
Further Readings
10.1 More than two truth values

Lukasiewicz introduced many-valued logics in 1920. The history of
studies of Lukasiewicz's original philosophical ideas and motivations is
fairly long, and is definitely outside the scope of this book. Interested
readers are referred to Wolenski's monograph [246], where Lukasiewicz's
motivations are analyzed and his work on many-valuedness is presented
in a wide perspective. In her essay [73] the author discusses many-
valuedness in the framework on nonclassicallogics. In their essay [204],
Priest and Routley study Lukasiewicz logic from the viewpoint of para-
consistency. Paztig [200} discusses the relations between Lukasiewicz's
ideas on many-valuedness and ideas in Chapter 9 of Aristotle's De In-
terpretatione. The short paper by Rosser [218] surveys the early stages
of many-valued logic, and offers succinct historical and bibliographical
remarks to an intended audience of physicists. The books [149], [30]
and [227] contain English translations of papers by Lukasiewicz and
Wajsberg.
The problem of interpreting nonboolean truth values was considered
by many people, including Lukasiewicz. Prior to the interpretation via
Ulam game, agame theoretic interpretation of infinite-valued logic was
given by Giles in [93], in the context of subjective probability. While
no probability is involved in Ulam game, it is quite possible that an
analysis of the Questioner's ''willingness to bet" in agame of Twenty
197
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
logic. Hähnle's monograph [108] is mainly devoted to automatic deduc-

tion in n-valued logics. Comprehensive bibliographies can be found in
[157], [206] and [74] (also see the selected bibliography in [141].) The
handbook chapters by Urquhart [237] and Panti [198] offer a survey of
several many-valued systems. See [52] for a compact technical survey
of MV-algebras and their neighbours.
10.2 Current Research Topics

We conclude with a list of active areas of current research in infinite-
valued propositionallogic and MV-algebras. The relevant literature is
rapidly growing, and several papers are under review at the time of
writing this section.
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-
tributes over the Lukasiewicz disjunction x EB y whenever x 0 y = 0

(Le., whenever x EB y = x + y). This distributivity law is a basic pre-
requisite to analyze and develop the expressive power needed for
• the approximation of a (continuous, real-valued, control) function
by means of a disjunction of pure tensors,
• the definition, in every a-complete MV-algebra, of such notions
as "independent events", "conditional", and "product of two ob-
servables", for furt her applications in MV-algebraic probability
theory (see below).
10.2.2 States, Observables, Probability, Partitions

Introduced in [182] and [184], states are the MV-algebraic generaliza-
tion of jinitely additive probability measures on boolean algebras. Their
AF C* -algebraic counterparts are known as "tracial states". On the
other hand, countably infinitary operations are needed for the develop-
ment of MV-algebraic measure theory. Accordingly, a-complete MV-
algebras and a-additive states are systematicaHy used in the book by
Riecan and Neubrunn [211]. As shown by Riecan and his School, many
important results of classical probability theory based on a-complete
boolean algebras and a-fields of sets have interesting MV-algebraic gen-
eralizations. One more example can be found in [187]. '
While the theory of a-additive MV-algebraic states is fairly weH
understood, random variables (alias, observables) still lack a definitive
systematization in the context of MV-algebras. A number of technical
problems, also involving product and infinite distributive laws are posed
by the theory of continuous functions of several (joint) MV-algebraic
observables. (See [208], [209] and [210} for interesting positive results).
A useful tool for understanding such observables is given by the MV-
algebraic generalization of the not ion of boolean partition [184], [185]
and [188]. An MV-partition in A is a multiset of linearly independent
elements of A whose sum equals 1. This definition makes perfect sense,
by referring to the underlying Z-module structure of the unique lattice-
ordered abelian group C with unit 1 given by r( C, 1) = A. The joint
refinability of any two MV-algebraic partitions on an MV-algebra A
depends on the "ultrasimplicial property" of its associated f-group C,
10.2. CURRENT RESEARCH TOPICS 201
in the sense that every finite set in G+ is contained in the monoid

generated by some basis B S; G+, Le., a set B of positive elements
that are independent in the Z-module G. After some partial results of
Elliott, Panti, Handelman and others (see [171], [186], [192], [191] and
references therein), recently Marra [145] has proved that every abelian
f-group is ultrasimplicial.
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.
10.2.4 Further constructions

MV-algebraic coproducts (= MV-algebraic free products) are considered
in [172} and by Di Nola, Lettieri in [66]. In [168] it is proved that the
amalgamation property holds in the variety of MV-algebras. This result
is generalized in [67].
For automorphisms of free MV-algebras see [71]. A moment's re-

flection shows that the automorphism group of the free MV-algebra
with one generator consists of precisely two elements: identity and ro-
tation of McN aughton functions around the vertical li ne through the
point (1/2, 0). By contrast, already the automorphism group of the free
MV-algebra with two generators is highly nontrivial.
Order-theoretic and topological properties 0/ the set 0/ ideals of MV-
algebras, as weH as various kinds of representations, including sheaf
representations, are studied in [221, [57], [17], [16], [72}, [146], [84J,
[147], [148], [205], [116], [117], [118], [120], [70J. The MV-algebraic
reformulation of a long-standing open problem in f-group theory asks
for a purely topological characterization of the possible spaces of prime
ideals of MV-algebras with the natural (huH-kernel) topology. The
order-theoretic characterization was given in [56].
The relations between MV-algebras and various structures, notably
distributive lattices, ordered monoids and f-groups, are investigated in
[213], [214J, [31], [32], [13], [173], [63], [35], [96], [196], [115], [19], [180],
[78J and [59]. Special classes of MV-algebras are considered in [7], [21],
[20], [54], [99], [100], [114], [131], [133], [134], [180], [85], [75] [18].
A classification of the universal classes generated by certain totally
ordered MV-algebras is given in [95]. For subvarieties, quasi-varieties
and related topics, see [215], [230], [248], [27], [83], [87], [94J. In
this book, finite-valued Lukasiewicz logics have been considered only
via their Lindenbaum algebras, namely Grigolia's MV n-algebras, when
dealing with the classification problem for subvarieties of MV-algebras.
For more information on the algebras and proof theory of such systelJls
see, for instance, [160], [161], [47], [48], [157J, [8], [10], [11], (122), [173],
[226J.
As already noted, infinite-valued "first-order" not ions have not been
considered at all. A rapidly growing literature is concerned with the
important problem of giving infinite-valued generalizations of the clas-
sical notions of point, set, cartesian product, union, subset, relation,
function, equality, model, quantifier. For various constructs and at-
tempts in this direction see, e.g., [41], [222], [90], [48}, [110], [112J and
references therein.
Bibliography
[lJ ACKERMANN, R. (1967) Introduction to Many-valued Logics.

Dover, New York.
[2J AGUZZOLI, S. (1998) The complexity of McNaughton functions of one

variable. Advances in Applied Mathematics, 21, p. 58-77.
[3J AGUZZOLI, S. (1998) A note on the representation of McNaughton func-

tions by basic literals. Soft Computing, 2, p. 111-115.
[4J AGUZZOLI, S., CIABATTONI, A. (2000) Finiteness in infinite-valued

Lukasiewicz logic. Journal of Logic, Language and Information. Special
issue on Logics of Uncertainty (Mundici,D., Ed.) To appear.
[5J AGUZZOLI, S., MUNDICI, D. (1994) An algorithmie desingularization of

three-dimensional torie varieties. Tohoku Mathematical Journal, 46, p.
557-572.
[6J AIGNER, M. (1996) Searching with Lies. Journal of Combinatorial The-

ory, Series A, 74, p. 43-56.
[7J AMBROSIO, R., LETTIERI, A. (1993) A classification of bipartite MV-

algebras. Mathematica Japonica, 38, p. 111-117.
[8J ANSHAKOV, O.M., RYCHKOV, S.V. (1995) On finite-valued proposi-

tional caleuli. Notre Dame Journal of Formal Logic, 36, p. 606-629.
[9J BAAZ, M., FERMÜLLER, C.G., SALZER, G. (199?) Automated dedue-

tion for many-valued logics. In: Handbook of Automated Reasoning,
(Robinson, A., Voronkov A., Eds.) Elsevier, Amsterdam. To appear.
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.
[12] BALBES, R., DWINGER, P. (1974) Distributive Lattices. Columbia:

University of Missouri.
[13] BELLUCE, L.P. (1986) Semisimple algebras of infinite valued logie.

Canadian Journal of Mathematics, 38, p. 1356-1379.
[14] BELLUCE, L.P. (1992) Semisimple and eomplete MV-algebras. Algebra

Universalis, 29, p. 1-9.
[15] BELLUCE, L.P. (1995) a-complete MV-algebras. In: Non-classical

Logics and their Applications to Fuzzy Subsets, (Höhle, U., Kle-
ment, E.P., Eds.) Kluwer Academic Publishers, Dordrecht, p. 7-21.
[16] BELLUCE, L.P., DI NOLA, A. (1996) Yosida type representation for

perfeet MV-algebras. Mathematical Logic Quarterly, 42, p. 551-563.
[17] BELLUCE, L.P., SESSA, S. (1997) Orthogonal deeompositions of MV-

spaces. Mathware and Soft Computing, 4, p. 5-22.
[18] BELLUCE, L.P., DI NOLA, A., GEORGESCU, G. (1999) Perfeet MV-

algebras and i-rings. Journal of Applied Nonclassical Logics, Special issue
on many-valued logic, (Carnielli, W.A., Ed.,) 9. To appear.
[19] BELLUCE, L.P., DI NOLA, A., LETTIERI, A. (1993) Subalgebras, di-

reet produets and associated lattiees of MV-algebras. Glasgow Math. Jour-
nal, 34, p. 301-307.
[20] BELLUCE, L.P., DI NOLA, A., LETTIERI, A. (1993) Loeal MV-

algebras. Rendiconti Circolo Matematico di Palermo, 42, p. 347-361.
[21] BELLUCE, L.P., DI NOLA, A., SESSA, S. (1991) Triangular norms,

MV-algebras and bold fuzzy set theory. Mathematica Japonica, 36, p.
481-487.
BIBLIOGRAPHY 205
[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.
[24J BIGARD, A., KEIMEL, K., WOLFENSTEIN, S. (1971) Groupes et An-

neaux Reticules. Springer-Verlag, Berlin, Lecture Notes in Mathemat-
ics, v. 608.
[25J BIRKHOFF, G. (1967) Lattice Theory. 3rd ed., American Mathemat-

ical Society, Providence.
[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)
[28J BOICESCU, V., FILIPOIU, A., GEORGESCU, G., RUDEANU, S. (1991)

Lukasiewicz-Moisil Aigebras. North-Holland, Amsterdam. (Annals of
Discrete Mathematics, v. 49)
[29J BOLC, L., BOROWIK, P. (1992) Many-valued Logics 1: Theoreti-

cal Foundations. Springer-Verlag, Berlin.
[30J BORKOWSKI, L. (Ed.) (1970) Selected Works of J. Lukasiewicz.

North-Holland, Amsterdam.
[31J BOSBACH, B. (1981) Concerning brieks. Acta Mathematica Academiae

Scientiarum Hungaricae, 38, p. 89-104.
[32J BOSBACH, B. (1982) Concerning cone algebras. Algebra Universalis,

15, p. 58-66.
[33J BUFF, H.W. (1985) Decidable and undecidable MV-algebras. Algebra

Universalis, 21, p. 234-249.
206 BIBLIOGRAPHY
[34] BUTNARIU, D., KLEMENT, E.P. (1995) Triangular norm-based

measures and games with fuzzy coalitions. Kluwer, Dordrecht.
[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)
[36] CHANG, C.C. (1958) Aigebraic analysis of many-valued logics. Trans-

actions 0/ the American Mathematical Society, 88, p. 467-490.
[37] CHANG, C.C. (1958a) Proof of an axiom of Lukasiewicz. Transactions

0/ the American Mathematical Society, 87, p. 55-56.
[38] CHANG, C.C. (1959) A new proof of the completeness of the Lukasie-
wicz axioms. Transactions 0/ the American Mathematical Society, 93, p.
74-90.
[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.
[41] CHANG, C.C., KEISLER, H.J. (1966) Continuous Model Theory.

Princeton University. (Annals of Mathematics Studies)
[42] CHANG, C.C., KEISLER, H.J. (1973) Model Theory. North-Holland,

Amsterdam.
[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.
[44] CIGNOLI, R. (1970) Moisil Aigebras. Bahia Blanca: Universidad Na-

cional deI Sur. Notas de L6gica Matematica, v. 27.
[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
[46] CIGNOLI, R (1980) Some algebraic aspects of many-valued logics. In:

Third Brazilian Conference on Mathematical Logic, Recife, 1979.
(Arruda,A. 1., da Costa, N. C. A., Sette, A. M., Eds.,) Sociedade Brasileira
de L6gica, Säo Paulo, p. 49-69.
[47] CIGNOLI, R (1982) Proper n-valued Lukasiewicz algebras as S-algebras

of Lukasiewicz n-valued propositional calculi. Studia Logica, 41, p. 3-16.
[48] CIGNOLI, R (1984) An algebraic approach to elementary theories based

on n-valued Lukasiewicz logics. Zeitschrift für math. Logik und Grundla-
gen der Mathematik, 30, p. 87-96.
[49] CIGNOLI, R (1991) Complete and atomic algebras ofthe infinite-valued

Lukasiewicz logic. Studia Logica, 50, p. 375-384.
[50] CIGNOLI, R (1993) Free lattice-ordered abelian groups and varieties of

MV-algebras. In: IX Latin American Symposium on Mathematical
Logic, Bahia Blanca, p. 113-118. (Notas de L6gica Matematica, v. 38, part
I)
[51] CIGNOLI, R, MUNDlCI, D. (1997) An elementary proof of Chang's

completeness theorem for the infinite-valued calculus of Lukasiewicz. Stu-
dia Logica, 58, p. 79-97.
[52] CIGNOLI, R, MUNDlCI, D. (1998) An invitation to Chang's MV-

algebras. In: Advances in Algebra and Model-Theory, (M.Droste
and R.GÖbel, Eds.,) Gordon and Breach Publishing Group, Readings UK,
p.171-197.
[53] CIGNOLI, R, MUNDlCI, D. (1998a) An elementary presentation of the

equivalence between MV-algebras and f-groups with strong unit. Studia
Logica, special issue on Many-valued logics, (Mundici,D., Ed.,) 61, p.
49-64.
[54] CIGNOLI, R, TORRENS, A. (1995) Retractive MV-algebras. Mathware

and Soft Computing, 2, p. 157-165.
[55] CIGNOLI, R., TORRENS, A. (1996) Boolean products of MV-algebras:

hypernormal MV-algebras. Journal 0/ Math. Analysis and Applications,
199, p. 637-653.
208 BIBLIOGRAPHY
[56] CIGNOLI, R., TORRENS, A. (1996a) The poset of prime f-ideals of an

abelian f-group with strong unit. Journal of Algebra, 184, p. 604-612.
[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.
[58] CIGNOLI, R. D'OTTAVIANO I.M.L., MUNDICI, D. (1995) Alge-

bras of Lukasiewicz Logics, (in Portuguese). Second Edition. Editions
CLE, State University of Campinas, Campinas, S.P., Brazil. (Cole(fiio
CLE/UNICAMP, v. 12)
[59] CIGNOLI, R., ELLIOTT, G.A., MUNDICI, D. (1993) Reconstructing

C*-algebras from their Murray von Neumann orders. Advances in Math-
ematics, 101, p. 166-179.
[60] CORNISH, W.H. (1980) Lattice-ordered groups and BCK-algebras.

Mathematica Japonica, 25, p. 471-476.
[61] CORNISH, W.H. (1982) On Iseki's BCK-algebras. M. Dekker, New

York, p. 101-122. (Lecture Notes in Pure and Applied Mathematics, v.
74)
[62] CZYZOWICZ, J., MUNDICI, D., PELC, A. (1989) Ulam's searching game
with lies. Journal of Combinatorial Theory, Series A, 52, p. 62-76.
[63] DI NOLA, A. (1991) Representation and reticulation by quotients of

MV-algebras. Ricerche di Matematica, (Napies), 40, p. 291-297.
[64] DI NOLA, A. (1993) MV-algebras in the treatment of uncertainty. In:

Fuzzy Logic, Proceedings of the International Congress IFSA, Bruxelles
1991. (Löwen P., Roubens E., Eds.,) Kluwer, Dordrecht, p. 123-131.
[65] DI NOLA, A., LETTIERI, A. (1994) Perfeet MV-algebras are categori-

cally equivalent to abelian f-groups. Studia Logica, 53, p. 417-432.
[66] DI NOLA, A., LETTIERI, A. (1996) Coproduct MV-algebras, nonstan-

dard reals, and Riesz spaces. Journal of Algebra, 185, p. 605-620.
[67] DI NOLA, A., LETTIERI, A. (199?) One chain generated varieties of

MV-algebras. Journal of Algebra. To appear.
BIBLIOGRAPHY 209
[68J DI NOLA, A., LETTIERI, A. (199?a) Equational characterization of all

varieties of MV-algebras. Journal 0/ Algebra. To appear.
[69J DI NOLA, A., SESSA, S. (1995) On MV-algebras of continuous func-

tions. In: Non-classical Logics and their Applications to Fuzzy
Subsets, (Höhle, U., Klement, E. P., Eds.,) Kluwer Academic Publish-
ers, Dordrecht, p. 23-32.
[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.
[71J DI NOLA, A., GRIGOLIA, R., PANTI, G. (1998) Finitely generated

free MV-algebras, and their automorphism groups. Studia Logica, special
issue on Many-valued logics, (Mundici,D., Ed.,) 61, p. 65-78.
[72J DI NOLA, A., LIGUORI, F., SESSA, S. (1993) Using maximal ideals in
the classification of MV-algebras. Portugalia Mathematica, 50, p. 87-102.
[73J D'OTTAVIANO, LM.L. (1992) A Logica Classica e 0 Surgimento das

Logicas näo-classicas. In: Seculo XIX: 0 Nascimento da Ciencia
contemporänea. (Evora, F.R.R., Ed.,) University of Campinas, Camp-
inas, S.P., Brazil, p. 65-94. (Colet;iio CLE/UNICAMP, v. 11)
[74J DUNN, J. M., EpSTEIN, G., (Eds.) (1977) Modern Uses ofmultiple
valued Logics. D. Reidel, Dordrecht.
[75J DVURECENSKIJ, A., RIECAN, B. (199?) Weakly divisible MV-algebras

and product. Journal 0/ Math. Analysis and Applications. To appear.
[76J DWINGER, P. (1977) A survey of the theory of Post algebras. In:

DUNN, J.M., EpSTEIN, E., (Eds.) Modern Uses of multiple valued
Logics. D. Reidel, Dordrecht, p. 53-75.
[77J ELLIOTT, G.A. (1976) On the classification of inductive limits of se-

quences of semisimple finite-dimensional algebras. Journal 0/ Algebra, 38,
p.29-44.
[78J ELLIOTT, G.A., MUNDICI, D. (1993) A characterization of lattice-

ordered abelian groups. Mathematische Zeitschrift, 213, p. 179-185.
210 BIBLIOGRAPHY
[79] EpSTEIN, G. (1960) The lattice theory of Post algebras. Tmnsactions

of the American Mathematical Society, 95, p. 300-317.
[80] ESCALADA IMAz, G., MANYIA SERRES, F. (1994) The satisfiability

problem for multiple-valued Horn formulae. In: Proceedings of the In-
ternational Symposium on Multiple-Valued Logics, ISMVL '94.
Boston, MA, USA, IEEE Press, Los Alamitos, p. 250-256.
[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.
[82] EWALD, G. (1996) Combinatorial convexity and algebraic geom-

etry. Springer-Verlag, Berlin, Heidelberg, New York (Gmduate Texts in
M athematics, v. 168)
[83] FERREIRIM, I.M.A. (1992) On varieties and quasivarieties of hoops

and their reducts. University of Illinois, Chicago. (Ph D Thesis)
[84] FILIPOIU, A., GEORGESCU, G. (1995) Compact and Pierce represen-

tations of MV-algebras. Romanian Journal Pure and Applied Math., 40,
p.599-618.
[85] FILIPOIU, A., GEORGESCU, G., LETTIERI, A. (1997) Maximal MV-

algebras. Mathware and Soft Computing, 4, p. 53-62.
[86] FONT, J.M., RODRIGUEZ, A.J., TORRENS, A. (1984) Wajsberg alge-

bras. Stochastica, 8, p. 5-31.
[87] GAlTAN, H. (1991) Quasivarieties of Wajsberg algebras. Journal of

Non-classical Logic, 8, p. 79-101.
[88] GAREY, M.R., JOHNSON, D.S. (1979) Computers and Intractabil-

ity: a guide to the theory of NP-completeness. W.H. Freeman and
Company, San Francisco.
[89] GEORGESCU, G., LEUSTEAN, I. (1998) Convergence in perfect MV-

algebras. Journal of Math. Analysis and Applications, 228, p. 96-111.
[90] GEORGESCU, G., IORGULESCU, A., LEUSTEAN, I. (1998) Monadic and

closure MV-algebras. Multi-valued Logic, 3, p. 235-257.
BIBLIOGRAPHY 211
[91] GEORGESCU, G., LIGUORI, F., MARTINI, G. (1997) Convergence in

MV-algebras. Mathware and Soft Computing, 4, p. 41-52.
[92] GERLA B. (1999) A probabilistic approach to Ulam games. Theoretical

Computer Science. To appear.
[93] GILES, R. (1977) A non-classicallogic for physics. In: WOJCICKI, R.,

MALINOWSKI, G., (Eds.) Selected Papers on Lukasiewicz Senten-
tial Calculi. Polish Academy of Sciences, Wroclaw, p. 13-51.
[94] GISPERT, J., TORRENs, A. (1998) Quasivarieties generated by sim-

ple MV-algebras. Studia Logica, special issue on Many-valued logics,
(Mundici,D., Ed.,) 61, p. 79-99.
[95] GISPERT, J., MUNDICI, D. TORRENS, A. (1999) Ultraproducts of Z

with an application to many-valued logics. Journal of Algebra. To appear.
[96] GIUNTINI, R. (1996) Quantum MV algebras. Studia Logica, 56, p. 393-

417.
[97] GIVANT, S.R., McKENZIE, R.N. (Eds.) (1986) Collected Papers of

Alfred Tarski, 4 volumes. Birkhäuser, Basel.
[98] GLASS, A.M.W., HOLLAND, W.C., (Eds.) (1989) Lattice Ordered

Groups. Kluwer, Dordrecht.
[99] GLUSCHANKOF, D. (1989) Objetos inyectivos en algebra de la 16gica.

Universidad de Buenos Aires. (Ph.D. Thesis)
[100] GLUSCHANKOF, D. (1992) Prime deductive systems and injective ob-

jects in the algebras of Lukasiewicz infinite-valued calculi. Algebra Uni-
versalis, 29, p. 354-377.
[101] GOODEARL, K. (1982) Notes on Real and Complex C*-algebras.

Birkhäuser, Boston. (Shiva Mathematics Series, v. 5)
[102] GOTTWALD, S. (1989) Mehrwertige Logik. Akademie-Verlag, Ber-

lin. Expanded version in English, in preparation.
[1031 GRÄTZER, G. (1971) Lattice Theory, First Concepts and Dis-

tributive Lattices. Freeman and Company, San Francisco.
212 BIBLIOGRAPHY
[104J GRÄTZER, G. (1979) Universal Algebra. 2nd ed. Springer-Verlag,

New York.
[105J GRIGOLIA, R.S. (1973) An algebraic analysis of n-valued systems of

Lukasiewicz-Tarski. Proceedings 01 the University 01 Tbilisi, A 6-7, p.
121-132. (in Russian)
[106J GRIGOLIA, R.S. (1977) Algebraic analysis of Lukasiewicz-Tarski's n-

valued logical systems. In: WOJCICKI, R., MALINOWSKI, G., (Eds.)
Selected Papers on Lukasiewicz Sentential Calculi. Ossolineum,
Wroclaw, p. 81-92.
[107J GUZICKI, W. (1990) Ulam's searching game with two lies. Journal 01
Combinatorial Theory, Series A, 54, p. 1-19.
[108J HÄHNLE, R.. (1993) Automated Deduction in multiple-valued

Logics. Clarendon Press, Oxford.
[109J HÄHNLE, R. (1996) Exploiting data dependencies in many-valued log-

ics. Journal 01 Applied Non-Classical Logics, 6, p. 49-69.
[110J HÄHNLE, R. (1998) Commodious axiomatization of quantifiers in

multiple-valued logic. Studia Logica, special issue on Many-valued logics,
(Mundici,D., Ed.,) 61, p. 101-121.
[111J HÄHNLE, R., ESCALADA-IMAZ, G. (1997) Deduction in many-valued

logics: a Survey. Mathware and Soft Computing, 4, p. 69-97.
[112J HAJEK, P. (1998) Metamathematics of fuzzy logic. Kluwer, Dor-

drecht. (Trends in Logic, Studia LogicaLibrary)
[113J HAY, L.S. (1963) Axiomatization of the infinite-valued predicate cal-

culus. The Journal 01 Symbolic Logic, 28, p. 77-86.
[114J HÖHLE, V. (1995) Enriched MV-algebras. Mathware and Soft Com-

puting, 2, p. 167-180.
[115J HÖHLE, V. (1995a) Commutative, residuated, f-monoids. In: Non-

classical Logics and their Applications to Fuzzy Subsets, (Höhle,
V., Klement, E. P., Eds.,) Kluwer Academic Publishers, Dordrecht, p.
53-106.
BIBLIOGRAPHY 213
[116] Hoo, C.S. (1987) Bounded commutative BCK-algebras satisfying

D.C.C. Mathematica Japonica, 32, p. 217-225.
[117] HOo, C.S. (1989) MV-algebras, ideals and semisimplicity. Mathemat-

ica Japonica, 34, p. 563-583.
[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.
[121] ISEKI, K., TANAKA, S. (1978) An introduction to the theory ofBCK-

[122] ITURRIOZ, L. (1977) An axiom system for three-valued Lukasiewicz

propositional calculus. Notre Dame Journal oi Formal Logic, 18, p. 616-
620.
[123] J AKUBfK, J. (1994) Direct product decomposition of MV-algebras.

Czechoslovak Mathematical Journal, 44, p. 725-739.
[124] JAKUBfK, J. (1995) On complete MV-algebras. Czechoslovak Mathe-

matical Journal, 45, p. 473-480.
[125] JAKUBfK, J. (1995) Sequential convergences on MV-algebras. Czech-

oslovak Mathematical Journal, 45, p. 709-726.
[126] JAKUBfK, J. (1998) On archimedean MV-algebras. Czechoslovak

Mathematical Journal, 48, p. 575-582.
[127J JAKUBfK, J. (1998) Complete generators and maximal completions of

MV-algebras. Czechoslovak Mathematical Journal, 48, p. 597-608.
[128J KAWANO, Y. (1994) On the structure of complete MV-algebras. Jour-

nal oi Algebra, 163, p. 773-776.
[129] KOMORI, Y. (1978) The separation theorem of the ~o-valued Luka-

siewicz proposition al logic. Reports oi the Faculty oi Sciences, Shizuoka
University, 12, p. 1-5.
214 BIBLIOGRAPHY
[130] KOMORI, Y. (1981) Super Lukasiewicz propositional logics. Nagoya

Mathematical Journal, 84, p. 119-133.
[131] LACAVA, F. (1979) Alcune proprieta delle L-algebre e delle L-algebre

esistenzialmente chiuse. Bollettino Unione Matematica Italiana, A(5), 16,
p.360-366.
[132] LA CAVA , F. (1979a) Sulla struttura delle L-algebre. Atti Accademia

Nazionale dei Lincei: Rendiconti Classe di Scienze Jisiche, matematiche
e naturali, 67, p. 275-281.
[133] LACAVA, F. (1980) Sulla classe delle L-algebre esistenzialmente chiu-

se. Atti Accademia Nazionale dei Lincei: Rendiconti Classe di Scienze
Jisiche, matematiche e naturali, 68, p. 319-322.
[134] LACAVA, F. (1989) Sulle L-algebre iniettive. Bollettino Unione Mate-

matica Italiana, 3-A, 7, p. 319-324.
[135] LACAVA, F., SAELI, D. (1976) Proprieta e model-completamento di

alcune varieta di algebre di Lukasiewicz. Atti Accademia Nazionale dei
Lincei: Rendiconti Classe di Scienze Jisiche, matematiche e naturali, 60,
p.359-367.
[136J LACAVA, F., SAELI, D. (1977) Sul model-completamento della teoria

delle L-catene. Bollettino Unione Matematica Italiana, 14-A, 5, p. 107-
110.
[137J LEHMKE, S. (1995) Resolution in many-valued logics. University of

Dortmund, Germany. (Dissertation)
[138] LUKASIEWICZ, J. (1920) 0 logice trojwarküSciowej [On three-valued

logic]. Ruch FilozoJiczny, 6, p. 170-171. Translated into English by Woj-
tasiewicz, 0., in: BORKOWSKI 1970 and, by Hiz, H., in: MCCALL 1967.
[139] LUKASIEWICZ, J., TARSKI, A. (1930) Untersuchungen über den Aus-

sagenkalkül. Comptes Rendus des seances de la Societe des Sciences et
des Lettres de Varsovie, Classe III, 23, p. 30-50. Reprinted in: GI-
VA NT & McKENZIE 1986; English translation in: TARSKI 1956 and in:
BORKOWSKI 1970.
BIBLIOGRAPHY 215
[140] MAC WILLIAMS, J.F., SLOANE, N.J.A. (1986) The Theory ofEr-
ror-correcting Codes. 5th ed. North-Holland, Amsterdam. (Mathem-
atical Library)
[141] MALINOWSKI, G. (1977) Bibliography ofLukasiewicz logics. In: WOJ-

CICKI, R., MALINOWSKI, G., (Eds.) Selected Papers on Lukasiewicz
Sentential Calculi. Polish Academy of Sciences, Wroclaw, p. 189-199.
[142] MALINOWSKI, G. (1977a) S-algebras for n-valued sentential calculi of

Lukasiewicz. The degrees of maximality of some Lukasiewicz logics. In:
WOJCICKI, R., MALINOWSKI, G. (Eds.) Selected Papers on Luka-
siewicz Sentential Calculi. Polish Academy of Sciences, Wroclaw, p.
149-159.
[143] MALINOWSKI, G. (1993) Many-valued logics. Oxford University

Press, Clarendon Press, Oxford.
[144] MANGANI, P. (1973) Su certe algebre connesse con logiche a phI valori.
Bollettino Unione Matematica Italiana (4), 8, p. 68-78.
[145] MARRA, V. (199?) Every abelian f-group is ultrasimplicial. Journal

of Algebra. To appear.
[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.
[148] MARTfNEZ, N.G., PRIESTLEY, H.A.P. (1995) Uniqueness of MV-

algebra implication and De Morgan negation. Mathware and Soft Com-
puting, 2, p. 229-245.
[149J MCCALL, S. (Ed.) (1967) Polish Logic 1920-1939. Cl aren don Press,
Oxford.
[150J McKENZIE, R.N., McNuLTY, G.F., TAYLOR, W. F. (1987) Alge-

bras, Lattices, Varieties. Wadsworth & Brooks: Cole Publ., Monterrey,
v. 1.
[151] McLANE, S. (1971) Categories for the Working Mathematician.
Springer-Verlag, Berlin.
216 BIBLIOGRAPHY
[152J MCNAUGHTON, R. (1951) A theorem about infinite-valued sentential

logie. The Journal of Symbolic Logic, 16, p. 1-13.
[153J MENU, J., PAVELKA, J. (1976) A note on tensor produets of the unit
interval. Commentarii Math. Univ. Carolinae, 17, p. 71-83.
[154J MEREDITH, C.A. (1958) The dependenee of an axiom of Lukasiewiez.

Transactions of the American Mathematical Society, 87, p. 54.
[155J MOISIL, G. (1940) Recherches sur les logiques nonehrysippiennes. An-

nales Scientifiques de l' Universite de Jassy, 26, p. 431-436. Reprinted in:
MOISIL, G. 1972, p. 195-232.
[156J MOISIL, G. (1941) Notes sur les logiques nonehrysippiennes. Annales

Scientifiques de l'Universite de Jassy, 27, p. 86-98. Reprinted in: MOISIL,
G. (1972), p. 233-243.
[157J MOISIL, G. (1972) Essays sur les Logiques NonchrysippiEmnes.

Bueharest: Aeademie de la Republique Socialiste de Roumanie.
[158J MONK, D. (1976) Mathematical Logic. Springer, Berlin, New

York.
[159J MONTAGNA F. (2000) An algebraic approach to proposition al fuzzy

logic. Journal of Logic, Language and Information. Special issue on Logies
of Uneertainty (Mundici,D., Ed.) To appear'
[160J MONTEIRO, A. (1964) Sur la Definition des Aigebres de Luka-

siewicz trivalentes. Universidad Nacional deI Sur, Bahia Blanca. Notas
de L6gica Matematica, v. 21.
[161J MONTEIRO, A. (1967) Construetion des Algebres de Lukasiewicz triva-

lentes dans les Algebres de Boole Monadiques-I. Mathematica Japonica,
12, p. 1-23. (Reprinted in: Notas de L6gica Matematica, Bahia Blanca,
v. 11, 1974)
[162J MORELLI, R. (1996) The birational geometry of torie varieties. Journal

Algebmic Geometry, 5, p. 751 - 782. Errata: Homepage at the University
of Utah (1997), p. 767-770.
[163} MUNDICI, D. (1986) Interpretation of AF C*-algebras in Lukasiewicz

sentential ealeulus. Journal of Punctional Analysis, 65, p. 15-63.
BIBLIOGRAPHY 217
[164] MUNDICI, D. (1986a) MV-algebras are categorically equivalent to

bounded comrnutative BCK-algebras. Mathematica Japonica, 31, p. 889-
894.
[165] MUNDICI, D. (1986b) Mapping abelian l-groups with strong unit one-
one into MV-algebras. Journal 0/ Algebra, 98, p. 76-81.
[166] MUNDICI, D. (1987) The Turing complexity of AF C*-algebras with

lattice-ordered Ko. Lecture Notes in Computer Science, 270, p. 256-264.
[167] MUNDICI, D. (1987a) Satisfiability in many-valued sentential logic is

NP-complete. Theoretical Computer Science, 52, p. 145-153.
[168] MUNDICI, D. (1987b) Bounded comrnutative BCK-algebras have the

amalgamation property. Mathematica Japonica, 32, p. 279-282.
[169} MUNDICI, D. (1987c) Every abelian l-group with two positive genera-
tors is ultrasimplicial. Journal 0/ Algebra, 105, p. 236-241.
[170] MUNDICI, D. (1988) The derivative of truth in Lukasiewicz senten-

tial calculus. Contemporary Mathematics, American Mathematical Soci-
ety, 69, p. 209-227.
[171] MUNDICI, D. (1988a) Farey stellar sub divisions, ultrasimplicial groups

and K o of AF C*-algebras. Advances in Mathematics, 68, p. 23-39.
[172] MUNDICI, D. (1988b) Free products in the category of abelian f-groups

with strong unit. Journal 0/ Algebra, 113, p. 89-109.
[173] MUNDICI, D. (1989) The C*-algebras of three-valued logics. In: Logie

Colloquium 1988, Padova. (Ferro, R., Bonotto, C., Valentini, S" Za-
nardo, A., Eds.), North-Holland, Amsterdam, p. 61-77. (Studies in Logic
and the Foundations 0/ Mathematics)
[174] MUNDICI, D. (1991) Complexity of adaptive error-correcting codes.

Lecture Notes in Computer Science, 533, p. 300-307.
[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
[176J MUNDICI, D. (1992a) Normal forms in infinite-valued logic: the case

of one variable. Lecture Notes in Computer Science, 626, p. 272-277.
[l77J MUNDICI, D. (1992b) Thring complexity of the Behncke-Leptin C*-

algebras with a two-point dual. Annals 0/ Mathematics and Artificial In-
telligence, 26, p. 287-294.
[178J MUNDICI, D. (1993) Logic of infinite quantum systems. International

Journal 0/ Theoretical Physics, 32, p. 1941-1955.
[179J MUNDICI, D. (1993a) Ulam's game, Lukasiewicz logic and AF C*-

algebras. Fundamenta In/ormaticae, 18, p. 151-161.
[180J MUNDICI, D. (1993b) K a , relative dimension, and the C*-algebras of

Post logic. In: IX Latin American Symposium on Mathematical
Logic, Bahia Blanca, Argentina. Notas de L6gica Matematica, v. 38, part
I, p. 83-92.
[181J MUNDICI, D. (1994) A constructive proof of McNaughton's Theorem

in infinite-valued logics. The Journal 0/ Symbolic Logic, 59, p. 596-602.
[182J MUNDICI, D. (1995) Averaging the truth value in Lukasiewicz senten-

tiallogic. Studia Logica, Special issue in honor of Helena Rasiowa. 55, p.
113-127.
[183J MUNDICI, D. (1996) Lukasiewicz normal forms and toric desingular-

izations. In: Proceedings of Logic Colloquium 1993, Keele, England.
(Hodges, W., et al., Eds.,) Oxford University Press, p. 401-423.
[184J MUNDICI, D. (1996a) Uncertainty measures in MV-algebras and states

of AF C*-algebras. Notas de la Sociedad de Matematica de Chile, Special
issue in memoriam Rolando Chuaqui, 15, p. 42-54.
[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.
[186} MUNDICI, D. (1999) Classes of ultrasimpliciallattice-ordered abelian

groups. Journal 0/ Algebra, 213, p. 596-603.
[187J MUNDICI, D. (1999a) Tensor products and the Loomis-Sikorski theo-

rem for MV-algebras. Advances in Applied Mathematics, 22, p. 227-248.
BIBLIOGRAPHY 219
[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.
[189] MUNDICI, D., OLIVETTI, N. (1998) Resolution and model building in

the infinite-valued calculus of Lukasiewiez. Theoretical Computer Science,
200, p. 335-366.
[190] MUNDICI, D., PANTI, G. (1993) Extending addition in Elliott's local

semigroup. Journal 01 F'unctional Analysis, 171, p. 461-472.
[191] MUNDICI, D., PANTI, G. (1999) A constructive proof that every 3-

generated f-group is ultrasimplicial. In: Logic, Algebra and Com-
puter Science Polish Academy of Sciences, Warszawa, p. 169-178. (Ba-
nach Center Publications, vol. 46)
[192} MUNDICI, D., PANTI, G. (1999a) Twenty questions with many-valued

answers, In: Models, Aigebras and Proofs, Selected papers of the Xth
Latin American Symposium on Mathematieal Logic, held in Bogota, in
memoriam Rolando Chuaqui, (Caicedo, X., and Montenegro, C.H., Eds.,)
Marcel Dekker, Inc., New York, Basel, p. 295-313.
[193] MUNDICI, D., PASQUETTO, M. (1995) A proof of the completeness

of the infinite-valued calculus of Lukasiewicz with one variable. In: Non-
classical Logics and their Applications to Fuzzy Subsets, (Höhle,
V., Klement, E.P., Eds.), Kluwer Academic Publishers, Dordrecht, p. 107-
123.
[194] MUNDICI, D., TROMBETTA, A. (1997) Optimal comparison strategies

in Vlam's searching game with two errors. Theoretical Computer Science,
182, p. 217-232.
[195] ODA, T. (1988) Convex Bodies and Aigebraic Geometry.

Springer-Verlag, Berlin, New York. (Eryebnisse der Math. Grenzgeb. (3),
v. 15)
[196] ÜNO, H. (199?) Logics without contraction rule and residuated

lattiees, 1. Manuscript.
[197] PANTI, G. (1995) A geometrie proof of the completeness of the

Lukasiewiez calculus. Journal 01 Symbolic Logic, 60, p. 563-578.
220 BIBLIOGRAPHY
[198] PANTI, G. (1998) Multi-valued Logics. In: Quantified Representa-

tion ofUncertainty and Imprecision, vol. 1, (Smets, P., Ed.), Kluwer,
Dordrecht, p. 25-74.
[199] PANTI, G. (1999) Varieties of MV-algebras. Journal of Applied Non-

classical Logics, Special issue on many-valued logics. (Carnielli, W.A.,
Ed.), 9, p. 141-157.
[200] PATZIG, G. (1973) Aristotle, Lukasiewicz and the origins of many-

valued logic. In: Logic, Methodology and Philosophy of Science,
IV, (Suppes, P., et al., Eds.), North-Holland, Amsterdam, p. 921-929.
(Studies in Logic and the Foundations of Mathematics, vol. 74)
[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.
[202] PELC, A. (1987) Solution of Ulam's problem on searching with a lie.

Journal of Combinatorial Theory, Series A, 44, p. 129-140.
[203] POST, E.L. (1921) Introduction to a general theory of elementary

propositions. American Journal of Mathematics, 43, p. 163-185.
[204] PRIEST, G., ROUTLEY, R. (1989) First Historical Introduction: apre-

liminary History of paraconsistent and dialectic Approaches. In: (Priest,
G., Routley, R., Norman, J., Eds.,) Paraconsistent Logic: Essays on
the Inconsistent. Philosophia-Verlag, Munieh, p. 3-75.
[205] RAY, S., SESSA, S. (199?) Representation of an MV-algebra by its

triangular norms. International Journal of Approximate Reasoning. To
appear.
[206] RESCHER, N. (1969) Many-valued Logic. McGraw Hill, New York.
[207] RIECAN, B. (1999) On the product MV-algebras. Tatra Mountain

Math. Publications, 16. To appear.
[208] RIECAN, B. (199?) On the joint distribution of observables. Soft Com-

puting. To appear.
[209] RIECAN, B. (199?a) On the LP space of observables on product MV-

algebras. Manuscript.
BIBLIOGRAPHY 221
[21OJ RIECAN, B. (199?b) On the probability theory on MV-algebras. Soft

Computing. To appear.
[211J RIECAN, B., NEUBRUNN, T. (1997) Integral, Measure, and Or-

dering. Kluwer, Dordrecht.
[212J RODRfGUEZ, A.J. (1980) Un Estudio algebraico de los Calculos pro-

posicionales de Lukasiewicz. Universidad de Barcelona. (Ph.D. Thesis)
[213J ROMANOWSKA, A., TRACZYK, T. (1980) On commutative BCK-al-

gebras. Mathematica Japonica, 24, p. 567-583.
[214J ROMANOWSKA, A., TRACZYK, T. (1982) Commutative BCK-alge-

bras: subdirectly irreducible algebras and varieties. Mathematica Japoni-
ca, 27, p. 35-48.
[215J ROSE, A. (1953) The degree of completeness of the ~o-valued Luka-

siewicz propositional calculus. The Journal 0/ the London Mathematical
Society, 28, p. 176-184.
[216J ROSE, A., ROSSER, J.B. (1958) Fragments ofmany-valued statement

calculi. Transactions 0/ the American Mathematical Society, 87, p. 1-53.
[217J ROSENBLOOM, P. C. (1942) Post algebras, I: postulates and general

theory. American Journal 0/ Mathematics, 64, p. 167-188.
[218J ROSSER, J .B. (1941) On the Many-valued Logics. American Journal

0/ Physics, 9, p. 207-212.
[219J ROSSER, J.B. (1960) Axiomatization of infinite-valued logics. Logique
et Analyse, 11-12, p. 137-153.
[220J ROSSER, J.B., TURQUETTE, A. (1952) Many-valued Logics.

North-Holland, Amsterdam.
[221J SAELI, D. (1975) Problemi di decisione per algebre connesse a logiche

a piu valori. Atti Accademia Nazionale dei Lincei: Rendiconti Classe di
Scienze fisiche, matematiche e naturali, 59, p. 219-223.
[222J SCHWARTZ, D. (1980) Polyadic MV-algebras. Zeitschrift für math.

Logik und Grundlagen der Mathematik, 26, p. 561-564.
222 BIBLIOGRAPHY
[223] SCOTT, D. (1974) Completeness and axiomatizability in many-valued

logic. In: Tarski Symposium, American Mathematical Society, Provi-
dence, p. 411-435. (Proceedings 0/ Symposia in Pure Mathematics, v. 25)
[224] SESSA, S., TURUNEN, E. (1998) On complete and strongly stonian

MV-algebras. Scientiae Mathematicae, 1, p. 23-26.
[225] SPENCER, J. (1992) Ulam's searching game with a fixed number of

lies. Theoretical Computer Science, 95, p. 307-321.
[226] STACHNIAK, Z. (1998) On finitely-valued inference systems. Studia

Logica, special issue on Many-valued logics, (Mundici,D., Ed.,) 61, p.
149-169.
[227] SURMA, S. (1977) Logical Works of Wajsberg. Ossolineum, Wro-

claw.
[228] TARSKI, A. (1956) Logic, Semantics, Metamathematics. Claren-

don Press, Oxford. Reprinted (1983), Hackett, Indianapolis.
[229] TOKARZ, M. (1977) A method ofaxiomatization of Lukasiewicz logics.

In: WOJCICKI, R., MALINOWSKI, G. (Eds.) Selected Papers on Lu':
kasiewicz Sentential Calculi. Polish Academy of Sciences, Wroclaw,
p. 113-117.
[230] TOKARZ, M. (1977a) Degrees of completeness of Lukasiewicz logics.

In: WOJCICKI, R., MALINOWSKI, G. (Eds.) Selected Papers on Lu-
kasiewicz Sentential Calculi. Polish Academy of Sciences, Wroclaw,
p. 127-134.
[231] TORRENS, A. (1987) W-algebras which are Boolean products of mem-

bers of SR[l] and CW-algebras. Studia Logica, 46, p. 263-272.
[232] TORRENS, A. (1989) Boolean products of CW-algebras and pseudo-

complementation. Reports on Mathematical Logic, 23, p. 31-38.
[233] TRACZYK, T. (1979) On the variety of bounded commutative BCK-

[234] TRACZYK, T. (1983) Free bounded commutative BCK-algebra with

one free generator. Demonstratio Mathematica, 16, p. 1049-1056.
BIBLIOGRAPHY 223
[235J ULAM, S. (1976) Adventures of a Mathematician. Scribner's,

New York.
[236J URQUHART, A. (1973) An interpretation of many-valued logic.

Zeitschr. J. math. Logik und Grundl. der Math., 19, p. 111-114.
[237J URQUHART, A. (1986) Many-valued logic. In: Handbook of Philo-

sophical Logic, III, (Gabbay, D., Ed.,) Reidel, Dordrecht, p. 71-116.
Updated version entitled "Basic many-valued Logic". To appear.
[238J VASYUKOV, V.L. (1993) The completeness of the factor semantics for
Lukasiewicz's infinite-valued logics. Studia Logica, 52, p. 143-167.
[239J VOJTAS, P., PAULlK, L.(1996) Soundness and completeness of non-

classical extended SLD-resolution. In: Extensions of Logic Program-
ming, (Dyckhoff R, et al., Eds.), Lecture Notes in Arlificial Intelligence,
Springer, Berlin, v. 1050, p. 289-301.
[240J WAGNER, H. (1999) A new resolution calculus for the infinite-valued

propositionallogic of Lukasiewicz. To appear.
[241J WAJSBERG, M. (1931) Aksjomatyzacja trowartosciowego rachunku

zdan [Axiomatization ofthe three-valued proposition al calculusJ. Comptes
Rendus des Seances de la Societe des Sciences et des Lettres de Varso-
vie, Classe 3, 24, p. 126-148. English translation by R Gruchman and S.
McCall, in: MCCALL 1967, and in: SURMA 1977.
[242J WAJSBERG, M. (1935) Beiträge zum Metaaussagenkalkül I. Mona-

tshefte für Mathematik und Physik, 42, p. 221-242. English translation in:
SURMA 1977.
[243J WOJCICKI, R. (1973) On matrix representations of consequence op-

erations of Lukasiewicz sentential calculi. Zeitschrift für math. Logik und
Grundlagen der Mathematik, 19, p. 239-247. Reprinted in: WOJCICKI,
R, MALINOWSKI, G. (Eds.) (1977) Selected Papers on Lukasiewicz
Sentential Calculi, p. 101-111.
[244J WOJCICKI, R (1988) Theory of Logical Calculi: Basic Theory

of Consequence Operations. Kluwer, Dordrecht. (Synthese Libmry, v.
199)
224 BIBLIOGRAPHY
[245] WOJCICKI, R., MALINOWSKI, G. (Eds.) (1977) Selected Papers on

Lukasiewicz Sentential Calculi. Ossolineum, Wroclaw.
[246] WOLENSKI, J. (1989) Logic and Philosophy in the Lvov-Warsaw

School. Kluwer, Dordrecht. (Synthese Library, v. 198)
[247] WOLF, R.G. (1977) A survey of many-valued logics (1966-1974). In:

DUNN, J. M., EpSTEIN, G. (Eds.) Modern Uses of multiple valued
Logics. D. Reidel, Dordrecht, p. 167-323.
[248] WOZNIAKOWSKA, B. (1978) Algebraic proof of the separation theorem

for the infinite-valued logic ofLukasiewicz. Reports on Mathematical Logic,
10, p. 129-137.
[249} YUTANI, H. (1977) On a system ofaxioms of commutative BCK-

algebras. Mathematics Seminar Notes of the Kobe University, 5, p. 255-
256.
[250} ZIEGLER, G.M. (1995) Lectures on Polytopes. Springer-Verlag,

Berlin, Heidelberg, New York. (Graduate Texts in Mathematics, v. 152)
Index
01=, the tautologies, 81 Adaptive search, 109

A/F, the quotient of A by F, 87 Adjoint functor, 139
a « b, a is infinitely smaller than b, 150 AF C"'-algebra, 192
Ho G, lexicographic product, 152 commutative, 193
(W), the ideal generated by W, 13 continuous trace, 193
n.x, the n-fold truncated addition of x, 33 finite-dimensional, 193
lai, the equivalence class of formula a, 93 liminary, 193
r functor, 34 with comparability of projections, 193
el-, the provable formulas from e, 88 Aguzzoli, S., 194, 201
el=, the semantic consequences of e, 81 Algebra
AF C*-, 192
C(Je), the variety generated by Je, 158 BCK-, 29
Cont(X), 65 C*-, 191
Div(n), the divisors of n, 169 CN-, 30
Form, the set of formulas, 79 De Morgan, 137
Free" , 53 Kleene, 137
H n , Komori chain of the second type, 160 Lindenbaum, 94
I(A), the ideals of A, 14 1-, 30
Injinit(A), the infinitesimals of A, 73 MV-, 7
Js, 65 Moisil-Lukasiewicz, 198
Ker(h), the kernel of h, 13 Post, 198
K n , Komori chain of the first type, 160 So, 30
.c(S), the Lindenbaum algebra of S, 99 Wajsberg, 83
Ln, the n element Lukasiewicz chain, 70 Alphabet, 21
M(A), the maximal ideals of A, 14 of Lukasiewicz calculus, 79
"P(A), the prime ideals of A, 14 Amalgamation property, 202
Q, the additive o-group of rationals, 33 Ambrosio, R., 202
R, the additive o-group of reals, 33 Analytic calculus, 201
Rad(A), the radical of A, 72 Answer in Ulam game, 104
u-additive state, 200 Archimedean element, 32
u-complete MV-algebra, 129 of an MV-algebra, 115
u-field of sets, 200 Aristotle's, De Interpretatione, 197
Theo, the lattice of theories, 97 Atom
U n , the variety of n-bounded MV-algebras, in a poset, 122
160 in an MV-algebra, 132
Var(a), the variables in a, 79 Atomic MV-algebra, 132
VJ, 65 Atomless MV-algebra, 132
Z, the additive o-group of integers, 33 Automated deduction, 201
Automorphisrns of free MV-algebras, 202
Abelian monoid, 8 Axiom, 87
Absolute
equation, 107 Baaz, M., 201, 202
value, 32 Basis, 57
Ackermann, R., 198 BCK-algebra, 29
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
Generated minimal prime in a lattice, 113

ideal, 13 nilpotent, 151
subalgebra, 8 prime, 13
Generating set of a subalgebra, 8 prime in a lattice, 112
Georgescu, G., 138, 199, 202 principal in a lattice, 112
Giles, R., 197 principal, 13
Gispert, J., 178,202 proper, 13
Giuntini, R., 202 proper in a lattice, 112
Gluschankof, D., 137, 202 stonean in a lattice. 115
Gödel incomplete Ideals
AF C"'-algebra, 193 poset of prime-, 202
the6ry, 193 spectral space of prime-, 202
Godo, L., 199 Idempotence law, 26
Good sequences, 34 Implication connective, 79
equivaIent, 40 Implicative filter, 86
natural order of-, 38 Incompatible
sum of-, 36 pair of states, 107
Gottwald, S., 198 state of knowledge, 106
Grothendieck's group Ko, 193, 194 Infinite-vaIued calculus of Lukasiewicz, 77
Grigolia, R., 178, 202 Infinitely small, 73
Group Infinitely smaller, 150
abelian, 31 Infinitesimal, 73
divisible, totally ordered, 194 Inhomogeneous
enveloping, 40 coordinates, 57
lattice-ordered, 31 correspondent, 181
partially ordered, 31 Initial state of knowledge, 106
torsion-free abelian, 45 Injective homomorphism, 13
totally ordered, abelian, 31 Iorgulescu, A., 202
ultrasimplicial, 201 Iseki, K., 29
Isomorphism, 13
Hähnle, R., 199,201 Iturrioz, L., 202
Hajek, P., 198, 199, 202
Half-open parallelepiped, 180 Jakublk, J., 138
Hay, L., 76, 101 Joint embedding property, 194
Hirzebruch-Jung algorithm, 186
Höhle, U., 202 Keisler, H. J., 202
Homogeneous coordinates, 180 Kernel of a homomorphism, 13
Homomorphism, 12 Kleene algebra, 137
injective, 13 Klement, E. P., 30, 199
kernel of-, 13 Komori, Y., 30,101,167,177
natural, 16
of i-groups, 34 Lacava, F., 30, 138,202
surjective, 13 Lattice
unital, 34 complete, 129
Hoo, C. S., 30, 138, 202 completely distributive, 134
Horn formula, 201 distributive, 24
Hyperarchimedean MV-algebra, 116 ideal of, 112
Hyperplane, 62 Lattice-ordered group, 31
Lehmke, S., 201
Ideal Lettieri, A., 156, 178, 201, 202
dual in a lattice, 114 Leustean, I., 199, 202
generated by a subset, 13 Lexicographic product, 152
in a lattice, 112 i-group, 31
in an MV-algebra, 13 homomorphism, 34
maximal, 14 term, 43
maximal stonean, 124 of an MV-algebra, 42
228 INDEX
Lindenbaum algebra, 94 idempotent, 26

of a theory, 99 maximal stonean ideal in-, 124
Linear constituent, 62 natural order, 10
Logical equivalence, 96 n-bounded, 160
Lukasiewicz, J., 29,78, 102, 103, 178, 197 nontrivial, 8
axioms, 49 one-generated hyperarchimedean, 119
calculus, 77 one-generated free, 62
chain, 122 partition, 200
connectives, 78 perfect, 152
finite-valued calculi, 78 probability measure, 200
product, 19
Malinowski, G., 198 quotient, 16
Mangani, P., 29 radical, 72
Manyili Serres, F., 201 semisimple, 72
Marra, V., 201 separating, 66
Martfnez, N., 202 simple, 70
Maximal state, 200
ideal of an MV-algebra, 14 subalgebra, 8
stonean ideal, 124 subdirect product, 19
Maximum satisfiability problem, 109 subterm, 21
McNaughton, R., 62,75, 178, 194 tensor product, 199
function, 54 term, 21
functions, of one variable, 62 totally ordered, 10
representation theorem, 184 valuation, 79
Mediant, 56 variety, 157
Menu, J., 30 without linear factors, 135
Meredith, C. A., 102 MVn -algebra, 172
Minimal prime ideal of a lattice, 113
Model completion of an MV-algebra, 138 Natural
Modus ponens, 88 homomorphism, 16
Moisil, G., 178, 198,202 order between states of know ledge, 107
Moisil-Lukasiewicz algebra, 198 order in an MV-algebra, 10
Monoid, zero-law, 37 order of good sequences, 38
Montagna, F., 199 n-bounded MV-algebra, 160
Monteiro, A., 178, 202 Negation
Morelli's proof of Oda's conjecture, 186 connective, 79
Multiplicity of a Schauder hat, 59 in Ulam game, 107
Murray-von Neumann order of projections, Negative
192 answer, 106
MV-algebra, 7 part, 32
u-complete, 129 Neubrunn, T., 200
archimedean element, 115 Nilpotent ideal, 151
atom, 132 Node of a function, 59
atomic, 132 Non-isomorphic simple MV-algebras, 148
atomless, 132 Nonsingular fan, 185
chain, 10
completely distributive, 134 Oda conjecture, strong form, 186
complete, 129 o-group, 31
completion, 138 Olivetti, N., 201
coproduct, 202 One-generated hyperarchimedean MV-alge-
directly indecomposable, 123 bra, 119
equation, 22 One-step star refinement, 185
free, 53 Ono, H., 202
free product, 202 Order
homomorphism, 12 natural, 31
hyperarchimedean, 116 translation invariant, 31
INDEX 229
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
Symmetrie difference, 15 desingularization, 185

Synta.ctic variety, 185
consequence, 88 vocabulary, 185
equivalence, 92 Torrens, A., 30, 101, 137, 178, 202
Totally ordered abelian group, 31
Thnaka, S, 29 Tra.czyk, T., 30, 202
Tarski, A., 101, 135, 138, 177 Translation invariance, 31
Tautology, 80 Triangulation, 63
Tensor product of MV-algebras, 199 star in a-, 182
Term unimodular, 180
function, 22 Truth-value, 105
of an l-group, 43 Turing
Theorem decidability of absolute equations, 108
r is fuH, faithful and dense, 146 decidability of word problem, 95
r preserves quotients, 148 enumeration of nontautological formu-
MV = C([O, 1», 158 las, 81
Chang's completeness, 44 enumeration of provable formulas, 89
Chang's subdirect representation, 20 Turquette, A., 198
Cook, 187 Turunen, E., 138
Oanilov, 186 Two-valued calculus, 78
Oe Concini-Procesi, 186
Oi Nola, 193 Ulam, S., 103, 197
Frayne, 194 Ulam game of Twenty Questions, 103
Komori 's dassification of MV-varieties, absolute equation, 107
169 answer, 104
McNaughton, 184 incompatible pair of states, 107
McNaughton-, for Freel, 62 incompatible state, 106
Morelli, 186 initial state, 106
Stone representation, 82 natural order, 107
completeness in Ulam game, 108 negation, 107
completeness of the infinite-valued cal- negative answer, 106
culus, 81 positive answer, 106
co-NP-completeness of tautology prob- question, 104
lem, 191 search spa.ce, 103
decidability of the word problem, 95 sharper state, 107
deduction, 98 state of knowledge, 106
finitely axiomatizable theories are de- truth-value, 105
cidable, 100 Ultrapower of [0, 1], 194
categorical equivalence, 146 Ultrasimplicial group, 201
on complete MV-algebras, 134 Underlying lattice of an MV-algebra, 10
on finite-valued MV-algebras, 168 Uni modular
on Free~, 175 simplex, 180
on hyperarchimedean MV-algebras, 117 triangulation, 180
on perfect MV-algebras, 155 Unique readability
on principal ideals, 69 of formulas, 80
on rank n nonsimple MV-chains, 167 of l-group terms, 43
on semisimple MV-algebras, 75 of MV-terms, 21
on simple MV-algebras, 70 Unit interva1 in an l-group, 34
on unimodular refinement, 180 Unitall-homomophism, 34
principal MV-quotients are semisimp- Universal dass of an MV-chain, 202
le, 75 Urquhart, A., 199
spectral representation, 66
Theory, 97 Valuation, 79
T-norm, 199 in [0,1], 81
Tokarz, M., 202 Variety, 157
Toric generated by, 158
INDEX 231
Vertices of a polyhedron, 63
Vojtas, Po, 201
Wagner, Ho, 201

Wajsberg, Mo, 49
algebra, 83
Weak boolean product, 124
W6jcicki, R., 76, 101, 198
Wolenski, Jo, 197
Word, 21
Word problem
decidability of, 95
for finitely presented algebras, 100
Wozniakowska, Bo, 202
Yutani, Ho, 29
Zero law in a monoid, 37

Zeroset, 65
TRENDS IN LOGIe
1. G. Schurz: The Is-Ought Problem. An Investigation in Philosophical Logic. 1997

ISBN 0-7923-4410-3
2. E. Ejerhed and S. Lindsträm (eds.): Logic, Action and Cognition. Essays in Philo-
sophical Logic. 1997 ISBN 0-7923-4560-6
3. H. Wansing: Displaying Modal Logic. 1998 ISBN 0-7923-5205-X
4. P. Hajek: Metamathematics ojFuzzy Logic. 1998 ISBN 0-7923-5238-6
5. H.J. Ohlbach and U. Reyle (eds.): Logic, Language and Reasoning. Essays in Honour
ofDov Gabbay. 1999 ISBN 0-7923-5687-X
6. K. Do§en: Cut Elimination in Categories. 1999 ISBN 0-7923-5720-5
7. R.L.O. Cignoli, I.M.L. D'Ottaviano and D: Mundici (eds.): Algebraic Foundations
ojmany-valued Reasoning. 2000 ISBN 0-7923-6009-5
KLUWER ACADEMIC PUBLISHERS - DORDRECHr / BOSTON / LONDON

Cignoli, Mundici-Algebraic Foundations of Many-Valued Reasoning PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Cignoli, Mundici-Algebraic Foundations of Many-Valued Reasoning PDF

Uploaded by

Copyright:

Available Formats

ALGEBRAIC FOUNDATIONS OF MANY-VALUED REASONING

SCOPE OF THE SERIES

ITALA M.L. D'OTfAVIANO

Springer-Science+Business Media, B.V.

ISBN 978-90-481-5336-7 ISBN 978-94-015-9480-6 (eBook)

Printed on acid-free paper

All Rights Reserved

1 Basic not ions 7

2 Chang completeness theorem 31

4 Lukasiewicz oo-valued calculus 77

5 Ulam's game 103

6 Lattice-theoretical properties 111

7 MV-algebras and f-groups 139

8 Varieties of MV-algebras 157

8.4 Komori's c1assification . . . . . . . . · 167

9 Advanced topics 179

10 Further Readings 197

and transform valid MV-equations into tautologies by writing p ~ q

give an MV-algebraic definition of "partition of unit", thus generalizing

for aH x E G there is n = 0, 1,2, ... such that x::; nu.

This archimedean-like property is not only beyond the expressive power

"equality", "structure" would result in a much less elementary text-

(CNR), as weH as from "Fundaci6n Antorchas" (Buenos Aires), "Cen-

We introduce MV-algebras by means of a small number of simple equa-

MV1) x tf) (y tf) z) = (x tf) y) tf) z

MVS) x tf) -.0 = -,0

MV6) -,( -,x tf) y) tf) y = -,( -,y tf) x) tf) x.

In particular, axioms MV1)-MV3) state that (A, EB, 0) is an abelian

(1.3) xe Y =def X 8 -'y.

MV7) -,1 = 0, and

MVS) x Ee y = -,(-,x 0 -,y).

Axioms MV5) and MV6) can now be written as:

MV5' ) x Ee1 = 1, and

Setting y = -,0 in MV6) we obtain:

Note that in the MV-algebra [0,1] we have x0y = max(0,x+y-1)

Lemma 1.1.2 Let A be an MV-algebra and x, y E A. Then the lol-

is equivalent to MV9), antisymmetry follows from conditions (ii) and

Lemma 1.1.3 Let A be an MV-algebra. For each a E A, -,a is the

(i) x::; Y if and only if -,y ::; -,x;

(ii) If x::; y then for each z E A, x EB z ::; y EB z and x 0 z ::; y 0 z;

(iii) x0y::; z iJJx::; -'yEBz.

Proposition 1.1.5 On each MV-algebra A the natural order deter-

Proof: To prove (1.5), by MV6'), MV9) and Lemma 1. 1.4(ii) , x :::;

Proposition 1.1.6 The following equations hold in every MV-algebra:

Proof: By MV6') and Lemma 1. 1.4(ii) , x 0 y :::; x 0 (y V z) and

Proposition 1.1. 7 Every MV-algebra satisfies the equation

= (y e x) 8 (-,(y e x) E9 x) 8 (-,(-,x E9 (y e x» E9 -,y)

= y 8 -,x 8 (-,(y e x) E9 x) 8 ((x 0 -,(y e x» E9 -,y)

= -,x 8 (x E9 -,(y e x» 8 y 0 (-,y E9 (x 8 (-,y E9 x») =

= 0, since by MV8) and MV9), -,x 8x = O. 0

Lemma 1.1.8 Let x and y be elements of an MV-algebra A. If

Proof: If x 1\ Y = 0 then by monotonicity (Lemma 1.1.4) and distribu-

1.2 Homomorphisms and ideals

H3) h(-,x) = -,h(x).

Following current usage, if h is one-one we shall equivalently say

14) For each x and y in A, either (x e y) E I or (y e x) E I.

An ideal I of an MV-algebra A is caHed maximal Hf it is proper and

Proposition 1.2.2 Por any proper ideal J of an MV-algebra A the

(i) J is a maximal ideal of A;

Proof: (i) ----+ (ii): Suppose that J is a maximal ideal of A. If x rt. J,

Lemma 1.2.3 Let A, B be MV-algebras, and h: A ----+ B a homomor-

(i) Por each ideal J of B, the set h-1(J) =def {x E AI h(x) E J} is

(ii) h(x) ::; h(y) iff xe y E Ker(h);