Sergei P. Odintsov Constructive Negations and Paraconsistency

Constructive Negations and Paraconsistency
TRENDS IN LOGIC
Studia Logica Library
VOLUME 26
Managing Editor
Ryszard Wjcicki, Institute of Philosophy and Sociology,
Polish Academy of Sciences, Warsaw, Poland
Editors
Vincent F. Hendricks, Department of Philosophy and Science Studies,
Roskilde University,
Denmark
Daniele Mundici, Department of Mathematics Ulisse Dini,
University of Florence, Italy
Ewa Orowska, National Institute of Telecommunications,
Warsaw, Poland
Krister Segerberg, Department of Philosophy, Uppsala University,
Sweden
Heinrich Wansing, Institute of Philosophy, Dresden University of Technology,
Germany
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
relations to other disciplines. These include artificial intelligence, informatics,
cognitive science, philosophy of science, and the philosophy of language.
However, this list is not exhaustive, moreover, the range of applications, com-
parisons and sources of inspiration is open and evolves over time.
Volume Editor
Heinrich Wansing
The titles published in this series are listed at the end of this volume.
Sergei P. Odintsov
Constructive Negations
and Paraconsistency
123
Sergei P. Odintsov
Russian Academy of Sciences
Siberian Branch
Sobolev Institute of Mathematics
Koptyug Ave. 4
Novosibirsk
Russia
ISBN 978-1-4020-6866-9 e-ISBN 978-1-4020-6867-6
Library of Congress Control Number: 2007940855
2008 Springer Science+Business Media B.V.

No part of this work may be reproduced, stored in a retrieval system, or transmitted in any form or by
any means, electronic, mechanical, photocopying, microfilming, recording or otherwise, without written
permission from the Publisher, with the exception of any material supplied specifically for the purpose
of being entered and executed on a computer system, for exclusive use by the purchaser of the work.
Printed on acid-free paper.
9 8 7 6 5 4 3 2 1
springer.com
Contents
1 Introduction 1
I Reductio ad Absurdum 13
2 Minimal Logic. Preliminary Remarks 15

2.1 Denition of Basic Logics . . . . . . . . . . . . . . . . . . . . 15
2.2 Algebraic Semantics . . . . . . . . . . . . . . . . . . . . . . . 21
2.3 Kripke Semantics . . . . . . . . . . . . . . . . . . . . . . . . . 28
3 Logic of Classical Refutability 31

3.1 Maximality Property of Le . . . . . . . . . . . . . . . . . . . 32
3.2 Isomorphs of Le . . . . . . . . . . . . . . . . . . . . . . . . . 35
4 The Class of Extensions of Minimal Logic 41

4.1 Extensions of Le . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.1.1 Intuitionistic and Negative Counterparts
for Extensions of Le . . . . . . . . . . . . . . . . . . . 45
4.2 Intuitionistic and Negative Counterparts for Extensions
of Minimal Logic . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.2.1 Negative Counterparts as Logics of Contradictions . . 52
4.3 Three Dimensions of Par . . . . . . . . . . . . . . . . . . . . . 53
5 Adequate Algebraic Semantics for Extensions

of Minimal Logic 57
5.1 Glivenkos Logic . . . . . . . . . . . . . . . . . . . . . . . . . 57
5.2 Representation of j-Algebras . . . . . . . . . . . . . . . . . . 59
5.3 Segerbergs Logics and their Semantics . . . . . . . . . . . . . 62
5.4 Kripke Semantics for Paraconsistent Extensions of Lj . . . . 78
v
vi Contents
6 Negatively Equivalent Logics 81

6.1 Denitions and Simple Properties . . . . . . . . . . . . . . . . 81
6.2 Logics Negatively Equivalent to Intermediate Ones . . . . . . 84
6.3 Abstract Classes of Negative Equivalence . . . . . . . . . . . 88
6.4 The Structure of Jhn+ up to Negative Equivalence . . . . . . 91
7 Absurdity as Unary Operator 101
7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
7.2 Le and L
ukasiewiczs Modal Logic . . . . . . . . . . . . . . . 104
7.3 Paradox of Minimal Logic and Generalized Absurdity . . . . 108
7.4 A- and C -Presentations . . . . . . . . . . . . . . . . . . . . . 113
7.4.1 Denitions and First Results . . . . . . . . . . . . . . 113
7.4.2 Logic CLuN . . . . . . . . . . . . . . . . . . . . . . . 119
7.4.3 Settes Logic P1 . . . . . . . . . . . . . . . . . . . . . 123
II Strong Negation 129

8 Semantical Study of Paraconsistent Nelsons Logic 131
8.1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
8.2 Fidels Semantics . . . . . . . . . . . . . . . . . . . . . . . . . 135
8.3 Twist-structures . . . . . . . . . . . . . . . . . . . . . . . . . 138
8.3.1 Embedding of N3 into N4 . . . . . . . . . . . . . . . 142
8.4 N4-Lattices . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
8.5 The Variety of N4-Lattices . . . . . . . . . . . . . . . . . . . 147
8.6 The Logic N4 and N4 -Lattices . . . . . . . . . . . . . . . 155
9 N4 -Lattices 159
9.1 Structure of N4 -Lattices . . . . . . . . . . . . . . . . . . . . 161
9.2 Homomorphisms and Subdirectly Irreducible N4 -Lattices . 167
10 The Class of N4 -Extensions 177
10.1 EN4 and Int+ . . . . . . . . . . . . . . . . . . . . . . . . . . 177
10.2 The Lattice Structure of EN4 . . . . . . . . . . . . . . . . . 185
10.3 Explosive and Normal Counterparts . . . . . . . . . . . . . . 195
10.4 The Structure of EN4C and EN4 C . . . . . . . . . . . . . . 201
10.5 Some Transfer Theorems for the Class of N4 -Extensions . . 211
11 Conclusion 223
Bibliography 227
Index 237
Chapter 1
Introduction
The title of this book mentions the concepts of paraconsistency and construc-
tive logic. However, the presented material belongs to the eld of paracon-
sistency, not to constructive logic. At the level of metatheory, the classical
methods are used. We will consider two concepts of negation: the nega-
tion as reduction to absurdity and the strong negation. Both concepts were
developed in the setting of constrictive logic, which explains our choice of
the title of the book. The paraconsistent logics are those, which admit in-
consistent but non-trivial theories, i.e., the logics which allow one to make
inferences in a non-trivial fashion from an inconsistent set of hypotheses.
Logics in which all inconsistent theories are trivial are called explosive. The
indicated property of paraconsistent logics yields the possibility to apply
them in dierent situations, where we encounter phenomena relevant (to
some extent) to the logical notion of inconsistency. Examples of these situ-
ations are (see [86]): information in a computer data base; various scientic
theories; constitutions and other legal documents; descriptions of ctional
(and other non-existent) objects; descriptions of counterfactual situations;
etc. The mentioned survey by G. Priest [86] may also be recommended for
a rst acquaintance with paraconsistent logic. The study of the paraconsis-
tency phenomenon may be based on dierent philosophical presuppositions
(see, e.g., [87]). At this point, we emphasize only one fundamental aspect of
investigations in the eld of paraconsistency. It was noted by D. Nelson in
[65, p. 209]: In both the intuitionistic and the classical logic all contradic-
tions are equivalent. This makes it impossible to consider such entities at all
in mathematics. It is not clear to me that such a radical position regarding
contradiction is necessary. Rejecting the principle a contradiction implies
everything(ex contradictione quodlibet) the paraconsistent logic allows one
1
2 1 Introduction
to study the phenomenon of contradiction itself. Namely this formal logical

aspect of paraconsistency will be at the centre of attention in this book.
We now turn to constructive logic. Constructive logic is the logic of con-
structive mathematics, logic oriented on dealing with the universe of con-
structive mathematical objects. The common feature of dierent variants
of constructive mathematics is the rejection of the concept of actual inn-
ity and admitting only the existence of objects constructed on the base of
the concept of potential innity. In any case, passing to constructive logic
from the classical one changes the sense of logical connectives. For example,
Markov [60] denes the constructive disjunction as follows: The construc-
tive understanding of the existence of a mathematical object corresponds to
the constructive understanding of the disjunction of sentences of the form
P or Q. Such a sentence is considered as accepted if at least one of the
sentences P , Q was accepted as true. Of course, this understanding of dis-
junction does not allow one to accept the law of excluded middle and leads
to the rejection of classical logic. In the setting of constructive logic, there
are two basic approaches to the concept of negation and they are considered
in our investigation.
Since the Brouwer works, the negation of statement P , P , is under-
stood as an abbreviation of the statement assumption P leads to a con-
tradiction. Note that this concept agrees well with paraconsistency. The
above understanding of negation does not assume the principle contra-
diction implies everything (ex contradictione quodlibet) responsible for the
trivialization of inconsistent theories. The rst formalization of intuitionistic
logic suggested by A.N. Kolmogorov [44] in 1925 was paraconsistent. In this
work, A.N. Kolmogorov reasonably noted that ex contradictione quodlibet
(in the form p (p q)) has appeared only in the formal presentation
of classical logic and does not occur in practical mathematical reasoning.
However, A. Heyting was sure that using ex contradictione quodlibet is ad-
missible in intuitionistic reasoning and he added the axiom p (p q)
to his variant Li of intuitionistic logic [35]. Note that adding ex contradic-
tione quodlibet creates some problems with interpretation of Li as calculus
of problems [45]. One cannot consider the implication P Q as the prob-
lem of reducing the problem Q to the problem P . In Li, the implication
P Q means that the problem Q can be reduced to the problem P or the
problem P is meaningless. This diculty was known to A. Heyting, but he
did not considered this as a serious problem. According to A. Heyting [36,
p. 106], . . . it (ex contradictione quodlibet S.O.) adds to the precision of
the denition of implication and I shall interpret implication in this wider
sense.
1 Introduction 3
Only in 1937 I. Johansson [41] questioned the using of ex contradictione

quodlibet in constructive reasoning and suggested the system, which we de-
note by Lj. Axiomatics for Lj can be obtained by deleting ex contradictione
quodlibet from the standard list of axioms for intuitionistic logic, more ex-
actly, Li = Lj + {p (p q)}. In [41], Johansson proved that many im-
portant properties of negation provable in the Heyting logic Li can be proved
also in the system Lj. Since that the logic has the name Johanssons logic
or minimal logic(see the title of Johanssons article). Note that, in fact,
Johansson came back to the Kolmogorovs variant of intuitionistic logic.
More exactly, the implication-negation fragment of Lj coincides with the
propositional fragment of the system from [44]. Kolmogorov considered the
rst-order logic, but in the language with only two propositional connectives,
implication and negation.
Unfortunately, the logic Lj was for a long time on the borderline of
studies in the eld of paraconsistency, which was traditionally motivated by
the following paraconsistent paradox of Lj. Although Lj is not explosive,
admits non-trivial inconsistent theories, we can prove in Lj for any formulas
and that
, Lj .
This means that the negation makes no sense in inconsistent Lj-theories,

because all negated formulas are provable in them. In this way, inconsis-
tent Lj-theories are positive. It should be noted that studies in the eld
of paraconsistency were directed during a long period to searching for the
most natural system of paraconsistent logic, which is maximally close to
classical logic (cf. [39, p. 147]). The above paradox obviously shows that Lj
cannot play the role of such logic. However, recently more attention has been
paid to the study of paraconsistent analogs of well-known logical systems. In
this respect, Johanssons logic Lj is worthy of attention as a paraconsistent
analog of intuitionistic logic Li.
Turning to the second main approach to negation in constructive logic,
the concept of strong negation. Note that the strong negation is namely a
proper constructive negation.
As happens with most fundamental logical concepts, the concept of
strong negation was developed independently by many authors and with dif-
ferent motivations. Constructive logic with strong negation was suggested
for the rst time by D. Nelson in 1949 [64]. The truth of a negation of
statement in intuitionistic and minimal logic can be stated only indirectly,
via reducing a negated sentence to an absurdity. As a consequence of this,
the negation in these logics has the following feature, unsatisable from the
4 1 Introduction
constructive point of view. When the negation of a conjunction ( )

is provable, it does not follow in the general case that either , or is
provable. In the mentioned work, D. Nelson suggested a new constructive
interpretation of the negation connective based on the idea that the falseness
of atomic formulas can be seen directly, which leads to parallel constructive
procedures reducing the truth and falseness of complex statements to the
truth and falseness of their components. As a result, D. Nelson obtained a
logical system possessing the property:
if ( ), then or ,
where denotes the negation connective and the derivability in Nelsons
system. Now, the above property is traditionally considered as a character-
istic property of constructive negation, and Nelson-type negations are called
strong. One year later constructive logic with strong negation was consid-
ered by A.A. Markov [59]. The propositional variant of Nelsons logic was
studied by N.N. Vorobiev [114, 115, 116]. Independently, Gentzen-style cal-
culus equivalent to Nelsons system was developed by F. von Kutschera [49].
A system closely related to strong negation systems also arose in the work
by J.P. Cleave [18], who constructed the predicate calculus adequate for
the algebra of inexact sets by S. Korner [46]. The paraconsistent variant of
Nelsons system was studied independently by R. Routley (later R. Sylvan)
in the propositional case in [96], by Lopez-Escobar in [51] and by Nelson
himself [1], both in the rst-order case. It should be noted that the term
strong negation is connected not with the idea of direct falsication, but
with comparing strong and intuitionistic negations in the explosive variant
of Nelsons logic [64]. In this logic, one can dene an intuitionistic negation
via a strong negation as follows := , and prove the implication
showing that the negation is stronger than the intuitionis-
tic one. In the paraconsistent version of Nelsons logic, one cannot dene
the intuitionistic negation and the above comparison loses its meaning, but
traditionally the name strong negation is used also in this case.
We now say a few words about denotation of logics under consideration.
There is no generally accepted convention. Nelson used the denotation N
and N for his system of strong negation and for its paraconsistent vari-
ant (see [64, 1]), respectively. In Dunns systematization [26], these systems
receive the denotation N and BN1 , respectively. We will follow another
tradition (see, e.g., [120]) and denote explosive Nelsons logic by N3 and
paraconsistent Nelsons logic by N4. This choice is motivated by the Kripke-
style semantics for these logics. Kripke semantics for N3 was developed by
R. Thomason [107] and R. Routley [96]. As in the case of intuitionistic logics,
1 Introduction 5
N3-frames are partial orderings. But since verication and falsication are
treated in N3 independently, N3-models have two valuations, v + for veri-
cation and v for falsication, with the additional restriction that v + (p)
v (p) = , i.e., no atomic statement can be true and false in the same world
simultaneously. Omitting the latter restriction we obtain a semantics for N4.
It is not hard to check (see [93]) that from the pair (v + , v ) one can pass
to one many-valued valuation, which is three-valued (true, false, neither) in
case of N3 and four-valued (true, false, neither, both) in case of N4.
Of course, the logic N4 is more attractive for applications, because it
allows one to work with inconsistent information. A view of N4 as a logic
convenient for information representation and processing is reected in a
series of books (see [40, 117, 118]). Also, N4 has proved useful for solving
some well-known philosophical logic paradoxes [119, 121]. At the same time,
the attention paid to this logic is incomparable with that for N3. In par-
ticular, semantic investigations of N4 were restricted mainly to the case of
Kripke-style semantics. There was no specic information about the class
of N4-extension, except for the information about its proper subclass, the
class of N3-extensions. It should be noted that the latter was thoroughly
studied (see [33, 47, 99, 100, 101]).
Thus, we have two explosive logics Li and N3, and their paraconsistent
analogs Lj and N4. It will be shown that Li can be faithfully embedded
into Lj, whereas N3 is faithfully embedded into N4. In this way, refusing
the explosion axiom does not lead to a decrease in the expressive power of
a logic. Here arises the question: which new expressive possibilities have the
logics Lj and N4 as compared to the explosive logics Li and respectively
N3, and how regularly this family of new possibilities is structured? In this
book we give answers to these question by studying the lattices of extensions
of the logics Lj and N4.
Studying the lattices of extensions of dierent logics such as, e.g., the
intuitionistic logic Li (see, e.g., [16]), the normal modal logic K4 [30, 31],
etc., plays an extremely important part in the development of modern non-
classical logic. In the rst part of the book (Chapters 27) we concentrate on
the study of the class of extensions of Johanssons logic. This was the rst
attempt to systematically study the lattice of extensions for a paraconsistent
logic. We will see that there is one important feature, which distinguishes
the class of Lj-extensions from the classes of extensions of the explosive log-
ics Li and K4. The class Jhn of non-trivial extensions of minimal logic has a
non-trivial and interesting global structure (it is three-dimensional in some
sense), which allows one to reduce its description (to some extent) to the
well-studied classes of intermediate and positive logics.
6 1 Introduction
More exactly, the class Jhn is the disjunctive union of three classes: the
class Int of intermediate logics, which are explosive; the class Neg of negative
logics, i.e., logics with degenerate negation containing the scheme p; and
the class Par of properly paraconsistent extensions of Lj containing logics
which do not belong to the rst two classes. So we have
Jhn = Int Neg Par.
Note that negative logics are denition equivalent to positive ones.

For any L Par, one can dene its intuitionistic counterpart Lint (neg-
ative counterpart Lneg ) as the least logic from the class Int (respectively,
from the class Neg) containing L. There are strong translations (i.e., trans-
lations preserving the consequence relation) of logics Lint and Lneg to the
original paraconsistent L. The logic Lint may also be obtained by adding
ex contradictione quodlibet to L. In this way, the above-mentioned transla-
tion of Lint shows that the usual explosive reasoning can be modelled in a
paraconsistent logic. On the other hand, as was noted above, the important
advantage of paraconsistent logics is that they allow one to distinguish con-
tradictions: dierent contradictions are not equivalent in them. In case of
Lj-extensions, the structure of contradictions in the paraconsistent logic L
can be presented as a formal system, and namely the logic Lneg plays this
part. The strong translation of Lneg in L can be done via the contradiction
operator C() := . Due to this fact, the logic Lneg can really be
treated as the logic of contradictions of the logic L.
We conclude our study of the class Jhn with an eort to describe the
structure of Jhn up to the negative equivalence. Two logics L1 , L2 Jhn are
said to be negatively equivalent if they have the same negative consequence
relation, i.e., X L1 i X L2 for an arbitrary set of formulas X and
any formula . The negative equivalence of logics from Lj is equivalent to
the fact that they have the same family of inconsistent sets of formulas. From
the constructive point of view, these facts mean that negatively equivalent
logics have essentially the same concepts of negation and of contradiction.
Concluding the rst part of the book, we suggest a way to overcome the
above mentioned paradox of minimal logic. It can be done via introducing
the unary operator of absurdity A() instead of the constant and dening
the negation as the reduction to this generalized absurdity:
:= A().
The idea of such a denition arose from comparing the contradiction oper-
ator in the logic Le of classical refutability [22] with the necessity operator
1 Introduction 7
in L
ukasiewiczs modal logic L [52, 53]. For the rst time, a similar inter-
connection between Le and L was noted by Porte [84, 85]. We prove that
one of the modal paradoxes of L exactly corresponds to the fact that the
absurdity operator is constant, i.e., is like in Le. Moreover, it turns out that
negation in several well-known paraconsistent logics can be dened in this
way. For example, in the logic CLuN of Batens [5, 6] and in Settes maxi-
mal paraconsistent logic P 1 [102, 88], the negation can be presented as the
reduction to a unary absurdity operator.
In the second part of the book we study the lattice of extensions of
paraconsistent Nelsons logic. This investigation was motivated not only by
the interest in Nelsons logic as an alternative formalization of intuitionistic
logic, but also by the desire to prove whether is it possible to apply to this
new object the approach developed in the rst part of our work? The answer
to this question is positive, although we discovered essential dierences in
the structures of lattices of extensions of minimal logic and paraconsistent
Nelsons logic.
In connection with the paraconsistent Nelson logic there also arises a
question: in which language should this logic be considered? The explosive
N3 is usually considered in the language
, , , , with symbols for
two negations, strong and intuitionistic . As was noted above, the in-
tuitionistic negation is superuous in this case, because it can be dened
via the strong one. If we pass to the paraconsistent N4, the interpretation
of is not clear and it looks natural to consider the language with only
the negation symbol . This variant of the paraconsistent Nelson logic will
be denoted N4. However, it turns out that the presence of intuitionistic
negation is natural and desirable. The conservative extension of N4 in the
language
, , , , obtained by spreading N4-axioms to the new lan-
guage and adding axioms p and p for the new constant is
denoted N4 . The intuitionistic negation is dened in N4 in the usual
way, := .
To study the class EN4 (EN4 ) of extensions of Nelsons logic N4 (N4 )
we need adequate algebraic semantics. This means that we have to describe
the variety of algebras determining N4 (N4 ) such that there is a dual iso-
morphism between the lattice of subvarieties of this variety and the lattice
of N4(N4 )-extensions. For explosive N3, the algebraic semantics is pro-
vided by N -lattices, which are well studied [90, 28, 29, 33, 99, 100, 110]. The
N4-lattices introduced in [72] provide this kind of semantics for N4. The
algebraic semantics for N4 is provided by N4 -lattices, a natural modi-
cation of N4-lattices. An interesting peculiarity of N4(and N4 )-lattices is
that they have a non-trivial lter of distinguished values.
8 1 Introduction
The advantage of the language with intuitionistic negation becomes

obvious, when we start the investigation of the class of N4 -extensions.
Its structure diers essentially from that of Jhn. First of all, unlike Jhn
containing the subclass Neg of contradictory logics, N4 does not admit
contradictory extensions. Despite its paraconsistency the logic N4 admits
only local contradictions, adding any contradiction as a scheme to N4 re-
sults in a trivial logic. However, the class EN4 decomposes into subclasses
of explosive logics, normal logics, and logics of general form. This decompo-
sition reects the local structure of contradictions inside N4 -models and
is very similar to the decomposition of Jhn into subclasses of intermediate,
negative and properly paraconsistent logics.
Note that the negative equivalence relation, which played an important
role in the study of extensions of minimal logic, degenerates if we pass to
N4(N4 )-extensions. Two extensions of N4 (N4 ) are negatively equiva-
lent if and only if they are equal.
We shall now describe more precisely the structure of the book. Chapter 2
contains denitions of the most important logics from the class Jhn and
necessary information concerning algebraic and Kripke-style semantics for
Lj-extensions. Chapter 3 is devoted to the logic of classical refutability, the
maximal paraconsistent extension of Lj playing the key role in the studying
the class of Lj-extensions.
In Chapter 4, we investigate the logic Le = Lj + { ( p)} and
prove that the class of its extensions coincides with the class of all possi-
ble intersections of intermediate and negative logics. Moreover, any logic L
extending Le has a unique presentation as an intersection of intermediate
logic L1 and negative logic L2 . The logic L1 (resp., L2 ) will be taken as intu-
itionistic (resp., negative) counterpart of L. The notions of intuitionistic and
negative counterparts allow a generalization to the class of all Lj-extensions
and it turns out that the class Par of properly paraconsistent Lj-extensions
decomposes into a disjoint union of classes Spec(L1 , L2 ) consisting of all
logics having L1 and L2 as its intuitionistic and negative counterparts, re-
spectively. Each of the classes Spec(L1 , L2 ) forms an interval in the lattice
Par with the upper point L1 L2 . In this way, studying the structure of Jhn
reduces to the investigation of intervals of the form Spec(L1 , L2 ).
The next chapter will be devoted to constructing an adequate algebraic
semantics, in fact, a suitable presentation of j-algebras, which is convenient
to determine the location of dierent logics inside the intervals Spec(L1 , L2 ).
The eectiveness of the obtained presentation will be demonstrated via its
application to numerous extensions of Lj considered by K. Segerberg [98].
We also provide several facts concerning Kripke semantics for Lj-extensions.
1 Introduction 9
In Chapter 6, we introduce the negative equivalence of logics (see above),

which we denote as neg , and by modifying the technique of Jankovs for-
mulas prove that the quotient lattice Spec(L1 , L2 )/ neg is isomorphic to
the interval Spec(Lk, L2 ). We also prove that every interval Spec(L1 , L2 )
contains innitely many classes of negative equivalence and that there is a
continuum of negative equivalence classes in Jhn.
The last chapter of the rst part of the book, Chapter 7, will be devoted
to studying absurdity as a unary operator.
Chapter 8 starts the second part of the book, devoted to strong nega-
tion. In the rst section, we dene two variants of paraconsistent Nelsons
logic. The logic N4 is determined in the language
, , , , where is
a symbol for strong negation, whereas the logic N4 is a logic in the lan-
guage
, , , , with an additional constant . Moreover, N4 is a
conservative extension of N4 as well as of intuitionistic logic. The explosive
logic N3 is obtained by adding to N4 the explosion axiom p (p q).
Notice that by putting := (p0 p0 ) one can prove in N3 the additional
axioms of N4 .
In the second section, the logic N4 is characterized via Fidel structures
[29]. This is direct generalization of M. Fudels result for N3 obtained in
[29]. Fidel structures are implicative lattices augmented with a family of
unary predicates.
In the third section, we describe a semantics for N4 with the help of
twist-structures over implicative lattices (see [28, 110]). The completeness re-
sult will follow from the equivalence of Fidel structures and twist-structures,
also established in this section. A twist-structure is an algebraic structure
dened over the Cartesian square of an implicative lattice, the operations
of this structure agrees with the operations of the underlying implicative
lattice on the rst component and are twisted on the second component.
Further, in Section 4 of this chapter, we prove that the class of algebras
isomorphic to twist-structures admits a lattice theoretical denition. We
distinguish the class of N4-lattices, prove that any twist-structure is an
N4-lattice and that any N4-lattice A is isomorphic to a twist-structure over
A , the implicative lattice dened as quotient of A wrt to a congruence of
a special form. These results imply that N4 is characterized by N4-lattices.
In the next section, it is proved that N4-lattices form a variety VN4 such
that the lattice EN4 of N4-extensions is dually isomorphic to the lattice of
subvarieties of VN4 .
In the last section of Chapter 8, we transfer all these results to the logic
N4 and the lattice of its extensions EN4 . In this case, the twist-structures
are dened over Heyting algebras and for any N4 -lattice A, the quotient
10 1 Introduction
A is also a Heyting algebra. We call A the basic Heyting algebra of an
N4 -lattice A.
In Chapter 9, we develop the origins of the algebraic theory of N4 -
lattices necessary to study the lattice of extensions of the logic N4 . In par-
ticular, N4 -lattices are represented in the form of Heyting algebras with
distinguished lter and ideal. We dene a pair of adjoint functors between
categories of N4 -lattices and of Heyting algebras. We prove that if a homo-
morphism of basic algebras can be lifted to N4 -lattices, it can be done in a
unique way. It is shown that congruences on an N4 -lattice are in one-to-one
correspondence with implicative lters and that the lattices of congruences
of an N4 -lattices and of its basic algebra are isomorphic. As a consequence,
we describe subdirectly irreducible N4 -lattices as lattices with subdirectly
irreducible basic algebra. Finally, in terms of the above-mentioned represen-
tation, we formulate an embeddability criterion and describe the quotients
of N4 -lattices.
In the last chapter, we study the structure of the lattice of N4 -exten-
sions and show that it is similar to the structure of the class of Lj-extensions.
Although the distinctions of the structures of these two classes of logics are
also essential. The rst of these distinctions is that N4 has no contradictory
extensions, whereas minimal logic has the subclass of inconsistent extensions
isomorphic to the class of extensions of positive logic.
We investigate the interrelations between a logic L extending N4 and
its intuitionistic fragment. In the lattice EN4 , we distinguish the subclasses
Exp of explosive logics, Nor of normal logics, and Gen of logics of general form,
which play the roles similar to that of classes Int, Neg, and Par in the lattice
of extensions of minimal logic. The interrelations between classes Exp, Nor
and Gen are investigated with the help of notions of explosive and normal
counterparts for logics in Gen.
Finally, we give some rst applications of the developed theory of the
lattice of N4 -extensions. First, we completely describe the lattice of exten-
sions of the logic N4 C obtained by adding the Dummett linearity axiom
to N4 . We prove that all extensions of N4 C are nitely axiomatized and
decidable and that given a formula, one can eectively determine which of
the N4 C-extensions is axiomatized by this formula. Second, we describe
tabular, pretabular logics and logics with Graigs interpolation property in
the lattice of N4 -extensions.
Regarding the authorship of the presented results, this book contains
mainly the investigations of the author, previously published in a series of
articles [6681]. Chapter 2 and Section 8.1 have a preliminary character and
here we do not carefully trace the authorship of the presented results. Except
1 Introduction 11
for Chapter 2 and Section 8.1, we give explicit references to all results quoted
from other authors.
Acknowledgments. I am deeply indebted to Professors L.L. Maksimova
and K.F. Samokhvalov for our fruitful discussions, which inspired, in fact,
the beginning of this investigation. The investigations presented in the rst
part of the book were carried out during my stay in Torun, at the Logic
Department of Nicholas Copernicus University. I am very grateful to Prof.
Jerzy Perzanowski, the head of this department, for the invitation, hospital-
ity and helpful criticism. I want to acknowledge my deep indebtedness to the
Alexander von Humboldt Foundation for granting the research fellowship at
Dresden University of Technology and the return fellowship. The investi-
gations presented in the second part of the book were carried out during
this period. Finally, I am especially grateful to Prof. Heinrich Wansing, my
academic host in Dresden, for the very fruitful collaboration.
Chapter 2
Minimal Logic. Preliminary

Remarks
2.1 Denition of Basic Logics

A propositional language L is a nite set of logical connectives of dierent
arities, L = {f1n1 , . . . , fknk }. A propositional constant is a connective of
arity 0. Given a set of propositional variables, we dene formulas of the
language L via the standard inductive denition.
In the rst part of the book we will consider logics and deductive systems
formulated in the following propositional languages: the language of positive
logic L+ := {2 , 2 , 2 }, the language L := L+ {0 } extending L+ with
the constant for absurdity, and the language L := L+ {1 } with
the symbol for negation. Extensions of minimal logic admit equivalent
formulations in the languages L and L .
If is a formula in some propositional language and p1 , . . . , pn are propo-
sitional variables, the denotation (p1 , . . . , pn ) means that all propositional
variables of are from the list p1 , . . . , pn .
By a logic we mean a set of formulas closed under the rules of substitution
and modus ponens:
(p1 , . . . , pn )
and .
(1 , . . . , n )
If (1 , . . . , n ) is obtained from (p1 , . . . , pn ) by the substitution rule, we

say that it is a particular case or a substitution instance of . A deductive
system is a collection of axioms and inference rules. A theorem of a deductive
system is a formula provable in this system. We will usually dene logics as
15
16 2 Minimal Logic. Preliminary Remarks
sets of theorems of Hilbert style deductive systems with only the inference
rules of substitution and modus ponens. Therefore, to dene a logic it is
enough to list its axioms.
For a logic L and a set of formulas X, L + X denotes the least logic
containing L and all formulas of X. The symbol + also denotes the operation
of taking the least upper bound in the lattice of logics.
With any logic L, we associate in a standard way an inference relation L .
For a set of formulas X and a formula , the relation X L means that
can be obtained from elements of X and tautologies of L in a nite number
of steps by using the rule of modus ponens. A set X is said to be non-trivial
wrt L if X L for some .
Let Li be a logic in a propositional language Li , i = 1, 2, and L1 L2 .
We say that L2 is a conservative extension of L1 if L1 L2 and for any
formula in the language L1 ,
L1 L2 .
In this case we say also that L1 is an L1 -fragment of L2 . In what follows by

a positive fragment we mean an L+ -fragment.
Denote by F the trivial logic, i.e., the set of all formulas in the language
L , {+, , }.
We now dene several important logics. In the choice of denotation we
follow the book [93] by W. Rautenberg.
Positive logic Lp is the least logic in the language L+ containing the
following axioms:
1. p (q p)
2. (p (q r)) ((p q) (p r))
3. (p q) p
4. (p q) q
5. (p q) ((p r) (p (q r)))
6. p (p q)
7. q (p q)
8. (p r) ((q r) ((p q) r))

2.1 Denition of Basic Logics 17
Positive logic satises Deduction Theorem:

X {} Lp X Lp .
To prove this theorem we need axioms 1 and 2 of positive logic and the fact
that modus ponens is the only inference rule. All logics considered in the
book satisfy these conditions, therefore, Deduction Theorem remains true
for all logics considered below.
Classical positive logic Lk+ also is a logic in the language L+ and can
be axiomatized modulo Lp by either of the following two axioms:
P. ((p q) p) p (Peirce law)
E. p (p q) (extended law of excluded middle)
The version Lj of minimal logic (or Johanssons logic) in the language
L can be dened as a logic axiomatized by the axioms 18 above. The
equivalent version of minimal logic Lj in the language L with the negation
symbol can be axiomatized by the axioms 18 and the following axiom:
A. (p q) ((p q) p) (reductio ad absurdum)
To make precise the statement on the equivalence of two versions of
minimal logic, we dene the translations from the language L to L and
from L to L as follows.
For any F , let () be a formula in the language L obtained from
by replacing each subformula of the form by the subformula .
For any F , we denote by () a formula in the language L obtained
from by replacing every occurrence of by the subformula (p p),
where p is some xed propositional variable. For a set of formulas X F ,
denote by (X) the set {() | X}. Respectively, for X F , put
(X) := {() | X}.
Proposition 2.1.1 The following statements hold.
1. For an arbitrary set of formulas X F and for any formula F ,
X Lj if and only if (X) Lj ().
Moreover, Lj () for any formula .
2. For an arbitrary set of formulas X F and for any formula F ,
X Lj if and only if (X) Lj ().
Moreover, Lj () for any formula .

Thus, the translations dened above preserve all deductive properties and
the subsequent application of two translations results in a formula equivalent
to the original one. Due to these facts we pass freely in the following from the
language L to the language L and vice versa. We will omit the superscripts
in the denotation of minimal logic and will not explicitly indicate with which
version of minimal logic or of its extension we are dealing at the time. And
we write F for either F or F .
Intuitionistic logic Li and minimal negative logic Ln can be axiomatized
modulo minimal logic in the language L as follows:
Li = Lj + { p}, Ln = Lj + {};
and in the language L as follows:
Li = Lj + {p (p q)}, Ln = Lj + {p}.
Classical logic Lk, logic of classical refutability Le, and maximal negative
logic Lmn can be axiomatized modulo intuitionistic logic Li, minimal logic
Lj, and negative logic Ln respectively, via either the Peirce law P or the
extended law of excluded middle E.
Lk = Li+{p(p q)}, Le = Lj+{p(p q)}, Lmn = Ln+{p(p q)}.
The positive fragments of Lk, Le, and Lmn coincide with classical pos-
itive logic, whereas the positive fragments of Li, Lj, and Ln coincide with
positive logic.
Lk+ = Lk F + = Le F + = Lmn F + ,
Lp = Li F + = Lj F + = Ln F + .
All logics introduced above except for positive and classical positive log-
ics are extensions of minimal logic. The class of all non-trivial extensions of
the logic Lj we denote by Jhn, the class of all extensions by Jhn+ . Clearly,
the class of logics Jhn+ forms a lattice, where L1 + L2 is the least upper
bound of logics L1 and L2 , and the intersection L1 L2 is the greatest lower
bound. For an arbitrary logic L, the lattice of its extensions with lattice
operations + and we denote as EL. Notice that EL is a complete lattice.
For any family {Li | i I} of logics
in EL, the intersection iI Li is a logic
and it extends L. Obviously, iI Li is the greatest logic contained in all
logics Li . For this reason, in EL, there also exists the sum of logics iI Li ,
i.e., the least logic containing all logics Li , i I.
Recall several important formulas provable in Lp and Lj.
2.1 Denition of Basic Logics 19
Proposition 2.1.2 The following formulas are provable in Lp:
1. p p (the identity law).
2. (p q) (q p), (p q) (q p) (the commutativity of and ).
3. (p (q r)) ((p q) r), (p (q r)) ((p q) r)

(the associativity of and ).
4. (p (q r)) ((p q) (p r)), (p (q r)) ((p q) (p r))

(the distributivity laws).
5. (p (q r)) (q (p r)) (the permutation law).
6. (p (p q)) (p q) (the contraction law).
7. (p (q r)) ((p q) r) (import and export of the premiss).
8. ((p q) r) ((p r) (q r)).
9. ((p q) r) ((p r) (q r)) r.

2
Proposition 2.1.3 The following formulas are provable in Lj:
1. (p p),
2. (p q) (q p),
3. (p q) (q p),
4. (p q) (p q),
5. (p q) (p q),
6. p p,
7. (p p),
8. (p q) (p q),
9. (p q) (p q),
10. (p q) (p q)
For the proof of this and the previous proposition the reader may consult
one or another traditional textbook in classical logic and observe that the
standard proofs of formulas listed in these propositions use only axioms of
Lj or Lp respectively. It is also not hard to prove all these formulas directly
or with the help of Deduction Theorem.
2
The next proposition gives some information on the results of adjoining
dierent new axioms to Lj.
Proposition 2.1.4 [98]
1. The equality Lk = Lj + {} holds, where is one of the following

formulas:
(a) p p,
(b) (p q) (q p),
(c) (p q) (q p),
(d) (p q) (p q),
(e) (p q) (p q),
(f ) (p q) (p q).
2. Lk = Lj + {p p} = Lj + {(p q) (p q)}.
3. Lj + {p p} = Lj + {p p} = Lj + {(p q) (p q)}.
4. Li = Lj + {(p q) (p q)}.
The next proposition shows how to construct axioms of an intersection

of logics. In fact, we repeat the proof of Miuras result [63] for intersec-
tions of superintuitionistic logics (see also [16, p. 111]). We call the formula
(p1 , . . . , pm ) (pn+1 , . . . , pn+m ) the repeatedless disjunction of the formu-
las (p1 , . . . , pn ) and (p1 , . . . , pm ) and denote it by .
Proposition 2.1.5 Let L {Lp, Lj}, L1 = L + {i | i I}, and L2 :=

L + {j | j J}. Then L1 L2 = L + {i j | i I, j J}.
Proof. Suppose L1 L2 . By Deduction

Theorem and the
properties of
(see Proposition 2.1.2) we have iI i L and jJ j
L, where I I, J J, I and J are nite, and every i and j are
2.2 Algebraic Semantics 21
substitution instances of i and j respectively. Using axiom 8 of positive

logic and the distributivity laws, we then obtain

(i j ) L,
iI ,jJ
from which L + {i j | i I, j J}.

Conversely, assume that L + {i j | i I, j J}. Then is
derivable in L from some nite set of substitution instances i j of axioms
of this logic. Using axioms 6 and 7 of positive logic we can also derive from
the set of i as well as from the set of j . Consequently, L1 L2 .
2
Proposition 2.1.6 The lattices ELp and ELj are distributive. Moreover,
the intersection distributes with the innite sum in these lattices.
Proof. Let L {Lp, Lj}. We prove only that
L iI Li = iI (L Li ),
L, Li EL. Assume L = L + and Li = L + i for i I.

L iI Li = (L + ) (L + iI i )
= L+ { | , iI i }
= L + iI { | , i }
= iI (L + { | , i })
= iI (L + ) (L + i )
2.2 Algebraic Semantics

In this section, we give necessary denitions and facts concerning the alge-
braic semantics for propositional logics. Detailed information can be found
in [92, 93].
Let L = {f1n1 , . . . , fknk } be a propositional language. An algebra A =
A, f1A , . . . , fkA of the language L is a set, where the connectives of L are

interpreted as operations of respective arities, fiA : Ani A. The set A is
the universe of A and is denoted |A|. We write a A instead of a |A|.
If A1 , . . . , An are algebras of the same language, then the direct product
A1 . . . An is dened as an algebra whose universe is the direct product
of universes |A1 | . . . |An | and the operations are componentwise. Note

that i : |A1 | . . . |An | |Ai |, the projection onto the ith coordinate,
determines an epimorphism of A1 . . . An onto Ai .
By A B we denote that the algebra A is embeddable into B, i.e., that
there exists a monomorphism h : A B.
If K is a class of algebras, we denote by I(K) the class of algebras isomor-
phic to algebras from K, by H(K) the class of homomorphic images of alge-
bras from K, by S(K) the class of algebras embeddable into algebras from K,
and, nally, Up(K) denotes the class of algebras isomorphic to ultraproducts
[14] of algebras from K.
A matrix is, as usual, a pair M =
A, DA , where A is an algebra
and D A |A| is the set of distinguished elements of this matrix. In cases
where D A = {1} is one-element, we write
A, 1 instead of
A, {1} and
identify, in fact, a matrix with an algebra in a language with an additional
constant 1. A valuation in an algebra A is dened as a mapping from the
set of propositional variables into |A|. A valuation extends to the set of
all formulas in a homomorphic way. A formula is said to be true on a
matrix M =
A, DA , M |= , if for any A-valuation v, v() DA . An
identity = is true on an algebra A, A |= = , if v() = v() for
any A-valuation v. The set L(M) := { | M |= } is the logic of a matrix
M and the set Eq(A) := { = | A |= = } is the equational theory
an algebra A. For a class Kof matrices (algebras), we dene L(K) :=
of
{L(M) | M K} (Eq(K) := {Eq(A) | A K}).
The direct product of matrices M1 =
A1 , DA1 , . . . , Mn =
An , DAn
is dened as
M1 . . . Mn =
A1 . . . An , DA1 . . . DAn .
Since the operations on the direct product are componentwise, we have
L(M1 . . . Mn ) = LM1 . . . LMn .
In this part of the book we deal mainly with matrices having one distin-
guished element. Let A be an algebra of the language L+ {1} (L {1},
L {1}). Note that A |= is equivalent to = 1 Eq(A). Elements of
LA are called identities of A in this case.
An algebra A is a model for a logic L if L LA. If L = LA, we say
that A is a characteristic model for L. It is clear that the class of models
for some logic L forms a variety. Write A |= L if A is a model of L. Denote
M od(L) := {A | A |= L}.
Proposition 2.2.1 [93] Every Lj-extension has a characteristic model.
The reader is expected to be familiar with lattices and with distributive

lattices.
If A =
A, , is a lattice, then the lattice ordering A is dened by
the condition a A b a b = a. If a, b A and a A b, we denote
by [a, b] an interval wrt the lattice ordering with end points a and b, i.e.,
[a, b] := {c A | a A c A b}. In what follows we omit the lower index in
the denotation of the ordering if it does not lead to a confusion.
For a b and c [a, b], an element d is said to be a complement of c in
the interval [a, b] if c d = b and c d = a. Recall that if the lattice A is
distributive and c A has a complement in [a, b] A, then this complement
is unique.
An algebra A =
A, , , , 1 is called an implicative lattice if
A, , , 1
is a lattice with the greatest element 1 and such that for any a, b A there
exists a supremum {x | a x b} equal to the value of the implication
(or relative pseudo-complement) operation a b. Here denotes the lattice
ordering of A. All implicative lattices form a variety and the logic of this
variety is Lp [92].
Proposition 2.2.2 [92] Let A be an implicative lattice and a, b A. Then

the following holds.
1) a b = 1 i a b;
2) a = b i a b = 1 and b a = 1;
3) b a b;
4) a (a b) = a b.
2
By a j-algebra we mean an algebra A =

A, , , , , 1 of the language
L {1} such that
A, , , , 1 is an implicative lattice and the constant
is interpreted as an arbitrary element of the universe A. Minimal logic
Lj corresponds to the variety of j-algebras [92], which we denote as Vj .
Equivalently, we can dene j-algebras in the language L {1} as implicative
lattices with the negation operation satisfying the property a b = b
a. These equivalent denitions are related as follows: a = a , = 1.
A Heyting algebra is a j-algebra with the least element . Intuitionistic
logic Li is the logic of the variety Vi of Heyting algebras [92].
A negative algebra is a j-algebra with = 1. Obviously, negative algebras

are distinguished in the variety of j-algebras via the identity . Therefore,
minimal negative logic Ln is the logic of the variety Vn of negative j-algebras.
An arbitrary variety V of implicative lattices, j-algebras, Heyting alge-
bras, or negative algebras determines a logic LV extending Lp, Lj, Li, or
respectively Ln. More exactly, let Sub(V ) denote the lattice of subvarieties
of an arbitrary variety V . For a logic L Jhn+ , dene a variety of j-algebras
V (L) := {A | A Vj , A |= L},
and for a variety V Sub(Vj ), dene a logic
L(V ) := { | A |= for all A V }.
It is clear that for any L Jhn+ and V Sub(Vj ) we have
V (L) Sub(Vj ) and L(V ) Jhn+ .
Moreover, the following statement holds.
Theorem 2.2.3 The mappings
V : Jhn+ Sub(Vj ) and L : Sub(Vj ) Jhn+
are mutually inverse dual lattice isomorphisms. The restrictions V Sub(Vi )

and L ELi are mutually inverse dual isomorphisms of the lattices ELi and
Sub(Vi ), whereas V Sub(Vn ) and L ELn are mutually inverse dual isomor-
phisms of the lattices ELn and Sub(Vn ).
2
By a dual isomorphism of lattices A1 and A2 we mean an isomorphism

from the lattice A1 onto the lattice (A2 )op with the inverse ordering.
For a j-algebra A and a Heyting algebra B we denote by AB the direct
sum of these algebras. It is a j-algebra in which the unit element of A is
identied with the zero of B, and for any a A and b B, we have a b.
Recall that a non-empty subset F of an implicative lattice (a j-algebra)
A is a lter on A if it satises the following conditions: 1) for any x, y F ,
xy F ; 2) for any x F and y A, if x y, then y F . Denote by F(A)
the set of all lters on A. If X A, then
X denotes a lter generated by
the set X, i.e., the least lter on A containing X. It is clear that

X = {a A | b1 . . . bn a for some b1 , . . . , bn X}.
Instead of
{a} we write
a.
For a Heyting algebra A, denote by Fd (A) its lter of dense elements

and by R(A) the Boolean algebra of its regular elements. Recall that
Fd (A) = {a A | a = 1} = {a a | a A},
R(A) = {a A | a a = 1} = {a A | a = a},
and R(A) = A/Fd (A).
Let A be an implicative lattice (a j-algebra). For any congruence on
A, we dene a lter F := {a A | a1A }. For any lter F on A, dene a
congruence F := {(a, b) | a b, b a F }. It is clear that = F and
F = FF . We have thus dened a one-to-one correspondence between the
set of congruences and the set of lters on the implicative lattice (j-algebra)
A. Notice that for an identity congruence IdA , FIdA = {1A }.
We dene a subdirectly irreducible algebra A as an algebra, which has
minimal non-identity congruence (comp. [14]). Taking into account the above
correspondence between lters and congruences on implicative lattices and
j-algebras, we arrive at the following statement.
Proposition 2.2.4 An implicative lattice (a j-algebra) is subdirectly irre-

ducible if and only if

{1A } = {F | F is a lter on A, F = {1A }}.
An element A of an implicative lattice (a j-algebra) A is called an

opremum, if A = 1 and for any a A, the inequality a = 1 implies a A .
Proposition 2.2.5 An implicative lattice (a j-algebra) A is subdirectly ir-

reducible i it has an opremum.
2
For Heyting algebras, a similar result was stated by C.M. McKay [62].
It can be transferred to implicative lattices and j-algebras in a trivial way.
Due to the well-known Birkho theorem [14], any algebra A is isomorphic
to a subdirect product of subdirectly irreducible algebras (being homomor-
phic images of A). This immediately implies that every variety is completely
determined by its subdirectly irreducible algebras. Let M odf si (L) denote the
set of nitely generated subdirectly irreducible models of a logic L. In view
of the correspondence between logics and varieties, we have the following
Proposition 2.2.6 Let L1 and L2 be logics extending Lp (Lj). We have

L1 = L2 if and only if M odf si (L1 ) = M odf si (L2 ).
2
We call a Peirce algebra an implicative lattice satisfying the identity P

(or, equivalently, E).
Let 2P =
{0, 1}, , , , 1 be a two-element Peirce algebra.
Proposition 2.2.7 L2P = Lk+ .
Proof. It is clear that Lk+ L2P . To prove the inverse inclusion we show
that there is only one subdirectly irreducible Peirce algebra, 2P .
Let A be a Peirce algebra with more than two elements. We show that
for any 1 = a A there exists a lter Fa = {1} on A with a Fa .
Take 1 = a A. There is also a b A with 1 = b = a. If b a, then
a
b. Assuming b a, we consider an element a b. Since a = b, we
have a a b and a 1 b. By denition of relative pseudo-complement
we conclude a a b and a b = 1, i.e., a
a b and
a b =
{1}.
We have thus constructed a collection {Fa | a A} of lters on A such
that
Fa = {1},
aA
and Fa = {1} for all a A. By Proposition 2.2.4 this means that A is not
subdirectly irreducible.
2
Let 2 =
{0, 1}, , , , 0, 1 be a two-element Heyting algebra, which
is a characteristic model for classical logic, L2 = Lk.
By 2 =
{0, 1}, , , , 1, 1 we denote a two-element negative algebra.
Proposition 2.2.8 L2 = Lmn.
Proof. Obviously, and ((p q) p) p are identities of 2 , and so

L2 Lmn. The inverse inclusion can be stated similar to Proposition 2.2.7.
2
Proposition 2.2.9 [93] The logic Lj has exactly two maximal non-trivial
extensions, Lk and Lmn. Every non-trivial Lj-extension is contained in one
of them.
Proof. Consider an arbitrary non-trivial extension L of Lj and its charac-

teristic model A, L = LA. Obviously, A is not one-element. If A = 1A ,
then for every a A, a = 1, the set {a, A } is the universe of a subalgebra

isomorphic to 2 . Consequently, LA L2 . If A = 1A , then the subalgebra
with the universe {A , 1A } is isomorphic to 2, whence LA Lk.
2
Recall several facts from the universal algebra. A variety V is called
congruence distributive if for any algebra A V, the lattice Con(A) of
congruences of algebra A is distributive. A variety V is called congruence
permutable if for any algebra A V, the congruences are permutable wrt
composition. In this case the join of two congruences coincide with their
composition 1 1 = 1 2 for any 1 , 2 Con(A). An arithmetic variety
is a variety, which is congruence permutable and congruence distributive.
According to Pixleys theorem (see [14]) a variety V is arithmetic if and
only if there exists a term m(x, y, z) such that the identities
m(x, y, x) = m(x, y, y) = m(y, y, x) = x
hold in V.
Proposition 2.2.10 The variety of j-algebras is arithmetic.
Proof. In case of j-algebras, as well as in case of Heyting algebras (see [14]),

we can use the term
m(x, y, z) := ((x y) z) ((z y) x) (x z)
to establish that the varieties of j-algebras and Heyting algebras are arith-
metic. The verication is straightforward.
2
Let us consider an -generated free j-algebra A and its congruence
lattice Con(A ), which is distributive by the last proposition. Moreover,
congruences of Con(A ) are permutable wrt the composition. Elements of
A can be identied with classes of equivalence of formulas wrt Lj,
|A | = {[] | F},
where [] := { | Lj}. With any L Jhn+ we associate the

congruence
L := {([0 ], [1 ]) | 0 1 L}.
Clearly, the mapping L L is one-to-one and preserves the ordering.
Consequently, to prove that it is a lattice embedding it is enough to check
that for any L0 , L1 Jhn+ , the congruences L0 L1 and L0 L1 also have
the form L for a suitable L. Observe that L is closed under substitution,

i.e., if [0 ]L [1 ], then [0 (1 , . . . , n )]L [1 (1 , . . . , n )] for any 1 , . . . , n .
It can easily be seen that if Con(A ) is closed under substitution, then
L = { | []1}, where 1 is the class of Lj-tautologies, is a logic from Jhn+
and = L .
In this way, it is enough to check that L0 L1 and L0 L1 are closed
under substitution. We consider only the non-trivial case of L0 L1 . Since
A is congruence permutable, L0 L1 = L0 L1 . So, [0 ]L0 L1 [1 ] if
and only if there is a formula such that [0 ]L0 [] and []L1 [1 ]. This
immediately implies that L0 L1 is closed under substitution.
So, the set of congruences of the form L forms a lattice. It is easy to
see that L L is an order isomorphism of the lattice Jhn+ and the lattice
of congruences of the form L . If two lattices are isomorphic as orders, they
are isomorphic as lattices too. We have thus proved in an algebraic way the
distributivity of Jhn+ .
Corollary 2.2.11 The lattice Jhn+ is distributive.
2.3 Kripke Semantics

In conclusion of this chapter we say a few words on the Kripke-style seman-
tics for minimal logic and its extensions.
A j-frame is a triple W =
W, , Q, where W is a set of possible worlds,
is an accessibility relation such that
W, is an ordinary Kripke frame
for intuitionistic logic, i.e., a partially ordered set, and Q W is a cone
(upward closed set) with respect to , which we call the cone of abnormal
worlds. Worlds lying out of Q are called normal. A valuation v of a j-frame
W is a mapping from the set of propositional variables to the set of cones
of the ordering
W, . A model =
W, v is a pair consisting of a j-frame
and its valuation. Say in this case that is a model on W.
The forcing relation between models and formulas is dened in exactly
the same way as for ordinary Kripke frames. The only exception is the case of
constant . More precisely, we dene the relation |=x , where =
W, v
is a model, W =
W, , Q, x W , and is a formula, by induction on the
structure of formulas as follows. For a propositional variable p, put
|=x p x v(p).
2.3 Kripke Semantics 29
And further,
|=x |=x and |=x ;

|=x |=x or |=x ;
|=x y W (x y ( |=y |=y )).
We did not consider yet the case of constant , and we put
|=x x Q.
In particular, for a negated formula considered as an abbreviation for

, we have
|=x y W (x y ( |=y y Q)).
Read |=x as a formula is true at a world (or at a point) x in a

model . A formula is true on a model =
W, v, |= , if |=x
holds for all x W . A formula is true on a j-frame W, W |= , if it
is true on a model
W, v for an arbitrary valuation v of the j-frame W.
A formula is valid on the class K of Kripke j-frames if W |= for any
j-frame W K.
Let W =
W, , Q be a j-frame and let K W be a cone wrt .
We dene a j-frame W K in the following way: W K :=
K, K , QK , where
K := K 2 , QK := Q K. If =
W, v is a model on W, then K :=
W K , v K , where v K (p) := v(p) K for all propositional variables p.

If [x] := {y W | x y} is a cone generated by x, we write W x and
x instead of W [x] and [x] respectively.
Lemma 2.3.1 Let W =

W, , Q be an arbitrary j-frame, a model on
W, and K W a cone. For any x K and an arbitrary formula , we
have
|=x K |=x .
In particular,
W |= = W K |= .
We say that a j-frame W is a model for a logic L Jhn, W |= L, if

W |= for all L. For a class of j-frames K we put
LK := { | W K (W |= )}.
A logic L Jhn is characterized by a class of j-frames K if L = LK.

We call a j-frame W =
W, , Q normal if Q = , i.e., if all worlds
of this frame are normal. It is clear that normal j-frames can be identied
with ordinary Kripke frames for intuitionistic logic. We call a j-frame W =
W, , Q abnormal if Q = W , i.e., if all worlds are abnormal. Finally, a

j-frame W =
W, , Q will be called identical if the accessibility relation
coincides with the identity relation on W , = IdW .
Proposition 2.3.2 [98]
1. Minimal logic Lj is characterized by the class of all j-frames.
2. Intuitionistic logic Li is characterized by the class of all normal
j-frames.
3. Minimal negative logic Ln is characterized by the class of all abnormal
j-frames.
4. Logic of classical refutability Le is characterized by the class of all
identical j-frames.
5. Classical logic Lk is characterized by the class of all identical normal
j-frames.
6. Maximal negative logic Lmn is characterized by the class of all iden-
tical abnormal j-frames.
Further, we dene several special classes of j-frames. Let W =
W, , Q
be a j-frame. We say that W is separated if
x, y W ((x Q y Q) x y).
And we say that W is closed if
x, y W ((x Q y Q) (x y)).
Denote by Sep the class of all separated j-frames and by Cl the class of all
closed j-frames.
Proposition 2.3.3 [98] The logic Lj+{(p )( p)} is characterized
by the class Sep, and the logic Lj + { ( p)} is characterized by Cl.
A j-frame W =
W, , Q is called dense if
x W (x Q y xz y(z Q)).
The class of all dense j-frames is denoted by Den.
Proposition 2.3.4 [124] The logic Lj + {( p)} is characterized by
the class Den.
Chapter 3
Logic of Classical
Refutability1
We start the investigation of the class of Lj-extensions with the logic of

classical refutability Le = Lj + {((p q) p) p}. This important logic
arose in the work of dierent authors and with dierent motivations. It was
introduced for the rst time in the P. Bernays review [10] of H.B. Currys
articles [20, 21]. P. Bernays observed that one can obtain a new logical
system, namely Le, by extending axiom schemes of classical positive logic
Lk+ to the language with additional constant in the same way as one can
obtain minimal logic Lj by extending axiom schemes of positive logic to the
language L . Two years later, the system equivalent to Le was introduced
in S. Kangers work [42]. Kangers reason for dening Le is that . . . it
constitutes a weakened classical calculus in the same sense as the minimal
calculus is a weakened intuitionistic calculus. Further, this logic was studied
by S. Kripke [48], who stated, in particular, the equation Le = Lk Lmn.
The name logic of classical refutability was suggested in the H.B. Curry
monograph [22]. In [22, Ch.6, Sec.A], one can nd the discussion of this
name. In [98], K. Segerberg studied the Kripke-style semantics for numerous
extensions of minimal logic, and among them for Le [98, p.46]. J. Porte
[84, 85] investigated interrelations between logic of classical refutability and
L
ukasiewiczs modal logic [52, 53]. His results will play an essential part
in Chapter 7, where we will study the generalized version of negation as
reduction to absurdity. In [85], it was stated, in particular, that Le is a
four-valued logic. The same four-element matrix for Le will be introduced
in Section 3.1 in a dierent way, as a simplest characteristic model for Le.
1
Parts of this chapter were originally published in [67, 68].
31
32 3 Logic of Classical Refutability
Another time, the logic Le arose under the name of Carnots logic CAR in
the work [19] by N.C.A. da Costa and J.-Y. Beziau to explicate some ideas
of Lasare Carnot. The equality CAR = Le was stated by I. Urbas [108].
The author studies in [66] also led to the system Le, and it arises here
in a rather unexpected way, from the investigation of the constructivity
concept suggested by K.F. Samokhvalov [97]. In [66], it was proved that
Le coincides with the logic of all exactly constructive systems in the sense
of K.F. Samokhvalov. We also established in [66] the maximality property
for Le. Adjoining to Le a new classical tautology gives classical logic, and
adjoining a new maximal negative tautology results in maximal negative
logic. In this respect, Le is similar to the logic P 1 suggested by A. Sette
[102], the rst example of maximal paraconsistent logic. The maximality
property of Le will be presented in Section 3.1. In the conclusion to this
section we show that the class Jhn of all nontrivial extensions of minimal
logic is divided into three intervals: the interval of well-known intermediate
logics lying between the intuitionistic and the classical logics; the interval
of negative logics lying between minimal and maximal negative logics and
the interval of properly paraconsistent logics, which all lie between minimal
logic and Le.
The results of Section 3.2 were inspired by A. Karpenko article [43],
where isomorphs of classical logic into three-valued Bochvars logic B3 were
considered. It turns out that in Le one can naturally dene one isomorph of
classical logic and two dierent isomorphs of maximal negative logic. Start-
ing from these isomorphs, we dene translations of Lmn and Lk into Le.
In the next chapter, these translations will be generalized to translations
of arbitrary negative and intermediate logics into properly paraconsistent
extensions of minimal logic, which allow one to dene the notions of intu-
itionistic and negative counterparts of a paraconsistent Lj-extension.
3.1 Maximality Property of Le

According to Proposition 2.1.5 the intersection of logics Lmn = Lj + {, p
(p q)} and Lk = Lj + { p, p (p q)} is axiomatized as follows:
Lk Lmn = Lj + {p (p q), ( p)}.
The second formula is a substitution instance of the rst one and we have
thus proved the following statement, announced for the rst time by S. Kripke
in 1959 [48].
3.1 Maximality Property of Le 33
Proposition 3.1.1 Le = Lk Lmn.
This representation of Le allows one to make the following observations.
Lemma 3.1.2 1. The following formulas are in Le:
p p, (p p), (p q) (p q), (p q) (p q).
2. Le does not contain p p and p (p q).
Proof. 1. These formulas are classical tautologies, which can be easily de-
duced in maximal negative logic using the scheme .
2. Assuming Le p p we have Lmn p p. But Lmn p,
hence Lmn p. The substitution rule implies that any formula is provable
in Lmn, a contradiction.
If Lmn p (p q), then Lmn p q. Substituting p for p in
the latter formula we again have Lmn q.
2
Now we consider models for Le.
We call A =
A, , , , , 1 a Peirce-Johansson algebra (pj-algebra)
if
A, , , , 1 is a Peirce algebra and the constant is interpreted as
an arbitrary element of the universe A. These algebras provide a semantics
for the logic Le. Recall that Peirce algebras provide algebraic semantics for
Lk+ and that Le can be considered as an expansion of Lk+ to the language
L . These facts immediately imply the following
Proposition 3.1.3 A j-algebra A =

A, , , , , 1 is a model for Le if
and only if A is a pj-algebra.
2
List some simple properties of pj-algebras.
Proposition 3.1.4 Let A =

A, , , , , 1 be a pj-algebra.
1. The interval [, 1]A forms a subalgebra of A, which is a Boolean

algebra.
2. For any a A, if a , then a = 1.
1 and [, 1]A = A, then A contains an element incomparable

3. If =
with .
Proof. All statements of the proposition can easily be deduced from the
denition of the relative pseudo-complement. Consider, for example, the
last statement.
3. There exists an element a under by assumption. Then a is
incomparable with in view of the equality ( a) = 1.
2
Proposition 3.1.5 Let A be a pj-algebra. Then either A is a model for

Lmn, or A is a model for Lk, or LA Le, i.e., A is a characteristic model
for Le.
Proof. Let A be a pj-algebra. If = 1 = 1, then a = 1 for any a A

by Item 2 of the previous proposition. This means that A is a model for
Lmn.
Assume = 1, then [, 1]A is a non-trivial Boolean algebra. If [, 1]A =
A, then A is a model for classical logic.
Finally, assume that = 1 and [, 1]A = A. In this case, A contains a
subalgebra isomorphic to 2, whence, LA L2 = Lk. Consider the lter

and the corresponding quotient algebra. Since a 0 for any a A, the
algebra A/
satises the identity p = 1. Therefore, L(A/
) Ln.
Moreover, ((p q) p) p is an identity of A/
as a quotient of A.
Consequently, L(A/
) = Lmn, and we have LA Lk Lmn = Le.
2
Consider the lattice 4 =
{0, 1, 1, a}, , where 1 a 1, 1 0
1, and the elements a and 0 are incomparable.
It is a Peirce algebra. Interpreting as 0 turns it into a pj-algebra. By
the previous proposition we obtain.
Corollary 3.1.6 4 is a characteristic model for Le.
Proof. Indeed, 4 is neither a Boolean algebra, nor a negative algebra.

2
Remark. The j-algebra 4 is the simplest among characteristic models
for Le. Propositions 3.1.4 and 3.1.5 easily imply that any characteristic
model for Le must contain at least four elements. Indeed, the unity element
diers from , there is a third element under , and there is a fourth element
incomparable with .
Now we are ready to prove the maximality property for Le.
3.2 Isomorphs of Le 35
Theorem 3.1.7 Let Le. There are three possible cases:
1. Le + {} is trivial;
2. Le + {} = Lmn;
3. Le + {} = Lk.
Proof. Assume that Le+ {} is non-trivial and A is its characteristic model.

The inclusion LA Le fails, since is not in Le. By Proposition 3.1.5 we
have either LA = Lk or LA = Lmn.
2
We now make an important observation on the structure of the class Jhn
of all nontrivial extensions of Lj. Let Int := {L | L Jhn, p L} be the
class of all intermediate logics; let Neg := {L | L Jhn, p L} be the class
of all negative logics, i.e., Neg consists of logics with a degenerate negation.
Finally, let Par := Jhn \ (Int Neg) be the class of all properly paraconsistent
logics. Obviously, the class Jhn is a disjoint union of the classes Int, Neg, and
Par. It is well known that L Int if and only if Li L Lk. It turns out
that the other two classes also form intervals in the lattice Jhn+ .
Proposition 3.1.8 Let L Jhn+ . Then the following equivalences hold:
1. L Neg if and only if Ln L Lmn;
2. L Par if and only if Lj L Le.
Proof. 1. If L Neg, then Ln is contained in L by denition. At the same

time, adding the axiom p p to L leads to a trivialization, therefore, L
cannot be extended to Lk, consequently, L Lmn by Proposition 2.2.9.
2. Let L Par, and let A be a characteristic model for L. Since L Neg,
the inequality = 1 holds in A, hence {, 1} is a nontrivial Boolean algebra
and a subalgebra of A. Consequently, L Lk. Further, L Int, therefore,
the quotient A/
is nontrivial and has the greatest element , hence it
contains a two-element subalgebra isomorphic to 2 . The latter means that
L Lmn. Thus, L Le = Lk Lmn.
2
3.2 Isomorphs of Le
The term isomorph was used in the rst monograph on multi-valued logics
[94], and now it looks a bit archaic. Let L1 and L2 be logics and let L2 be
given via its logical matrix. Due to N. Rescher [94], an isomorph of the logic
L1 in the logic L2 is a denition of a matrix for L1 in the matrix for L2 with

the help of term operations.
We can dene a translation of L1 into L2 whenever some isomorph of
L1 in L2 is given. The relations between interdenability of logical matri-
ces and mutual translations of logics was studied in detail in the works by
P. Wojtylak [122, 123]. However, it will be quite enough for our goal to use
the old notion of the isomorph.
The interest of the author in isomorphs, which can be dened inside
logic of classical refutability, is connected with the talk of A. Karpenko at
the First World Congress on Paraconsistency (see [43]), in which he consid-
ered isomorphs of classical logic in three-valued Bochvars logic B3 given via
so-called internal and external connectives. He also tried to argue that
dierent isomorphs of classical logic in the given many-valued logic deter-
mine the paraconsistent structure of this logic. The logic Le as well as all
other extensions of minimal logic takes a borderline position among paracon-
sistent logics, and this fact naturally gives rise to the question of isomorphs,
which can be dened inside the logic Le. It turns out that there is only one
natural isomorph of classical logic in Le. At the same time, there are two
isomorphs of maximal negative logic. As we will see in the next chapter,
the translations of Lk and Lmn dened by these isomorphs can be used to
dene translations of intermediate and negative logics into arbitrary para-
consistent extensions of Lj. In this sense, the studying of isomorphs of the
logic Le plays the key role in the investigation of the class of Lj-extensions.
Consider the 4-element matrix 4 =
{1, 0, a, 1}, , , , , {1} for Le
and dene the mapping (x) := x.
x (x)
1 1
a 1
0 0
1 0
The operations , , , and are dened as follows:

x y := (x y) = (x) (y), {, , },
x := (x),
These operations determine an isomorph of classical logic in Le, which we
denote Lk . The fact that Lk really is an isomorph of Lk can easily be
veried by considering the truth tables for the above operations.
1 a 0 1 1 a 0 1
1 1 1 0 0 1 1 1 1 1
a 1 1 0 0 a 1 1 1 1
0 0 0 0 0 0 1 1 0 0
1 0 0 0 0 1 1 1 0 0
1 a 0 1
1 1 1 0 0 1 0
a 1 1 0 0 a 0
0 1 1 1 1 0 1
1 1 1 1 1 1 1
As we can see, rows and columns corresponding to elements 1 and a

are identical. The same holds for elements 0 and 1. Thus, identifying these
pairs of elements we obtain two-valued truth tables for operations of classical
logic. It is not hard to check with the help of the above truth tables that the
mapping denes an epimorphism from 4 onto the two-element Boolean
algebra 2 with the universe {1, 0}. It is also clear that 2 is a subalgebra
of 4 .
Lemma 3.2.1 The mapping : 4 2 , where 2 is a subalgebra of 4 with

the universe {1, 0}, is an epimorphism.
2
We now consider the mapping (x) := x x, which acts on the set

of truth-values as follows.
x (x)
1 1
a a
0 1
1 a
As above, dene the operations , , , :
x y := (x y) = (x) (y), {, , },
x := (x).
The truth tables of these operations look as follows.

1 a 0 1 1 a 0 1
1 1 a 1 a 1 1 1 1 1
a a a a a a 1 a 1 a
0 1 a 1 a 0 1 1 1 1
1 a a a a 1 1 a 1 a
1 a 0 1
1 1 a 1 a 1 1
a 1 1 1 1 a 1
0 1 a 1 a 0 1
1 1 1 1 1 1 1
As we can see, the pairs of elements 1 and 0, a and 1 are indiscernible

with respect to the introduced operations. Identifying these elements, we
obtain truth tables of the two-element negative algebra 2 with the universe
{1, a}. The algebra 2 is a characteristic model for maximal negative logic
Lmn. This fact allows one to conclude that the introduced operations dene
an isomorph of Lmn into Le, which we denote Lmn . Moreover, one can
check that the mapping preserves the operations of 4 .
Lemma 3.2.2 The mapping : 4 2 , where 2 is a two-element negative
algebra with the universe {1, a}, is an epimorphism.
2
Note that 2 is not a subalgebra of 4 , though it is an implicative sub-
lattice of 4 .
Finally, we dene the mapping (x) := x (its action on the truth
values of Le is in the table below)
x (x)
1 0
a 1
0 0
1 1
and the operations

x y := (x y) = (x) (y),
x y := (x y) = (x) (y),
x y := (x y) = ( (x) (y)),
x := (x).
Consider the truth tables of these operations.
1 a 0 1 1 a 0 1
1 0 1 0 1 1 0 0 0 0
a 1 1 1 1 a 0 1 0 1
0 0 1 0 1 0 0 0 0 0
1 1 1 1 1 1 0 1 0 1
1 a 0 1
1 0 1 0 1 1 0
a 0 0 0 0 a 0
0 0 1 0 1 0 0
1 0 0 0 0 1 0
Again, we see that the pairs of elements 1 and 0, a and 1 are indis-
cernible with respect to the introduced operations and that their identi-
cation yields the truth tables of the two-element negative algebra 2 :=
{0, 1}, , , , , where 0 plays the part of a unit element and the
negation is identically equal to 0, the conjunction and disjunction oper-
ations are induced by the respective operations of the algebra 4 , whereas
the implication is dened as x y := (x y) . Since L2 = Lmn,
we conclude that the operations , , , and dene an isomorph of
Lmn into Lk with a new distinguished value 0, i.e., the set of tautologies
of the matrix
4 =
{1, 0, a, 1}, , , , , {0}
with the only distinguished value 0 coincides with Lmn. This isomorph is
denoted as Lmn .
Again we note the following fact
Lemma 3.2.3 The mapping : 4 2 , where 2 is a two-element nega-

tive algebra with the universe {0, 1} and unit element 0, is an epimorphism.
2
The isomorphs dened above lead to the following translations of classi-

cal and maximal negative logics into Le.
Proposition 3.2.4 For any formula , the following equivalences hold:
1. Lk Le ;
2. Lmn Le ;
3. Lmn Le ( ) Le .
Proof. 1. Assume Le . In this case Lk since Lk extends Le.

In Lk, any formula is equivalent to its double negation, whence Lk .
Let us prove the inverse implication. Suppose that a formula =
(p1 , . . . , pn ) is such that Lk , but Le . In this case, there is
a 4 -valuation v such that v() = 1. Due to Lemma 3.2.1 the double
negation preserves the operations of 4 , and so we have
v((p1 , . . . , pn )) = v((p1 , . . . , pn )).
Let v1 be a 2 -valuation with the property v1 (p1 ) := v(p1 ), . . . , v1 (pn ) :=

v(pn ). In view of the last equality, we have
v1 ((p1 , . . . , pn )) = v((p1 , . . . , pn )) = 1.
The latter inequality means that is not provable in Lk.

2. We can prove this item in the same way as was done above, using
Lemma 3.2.2 instead of Lemma 3.2.1. We can also reduce this item to the
next one. Indeed, Le , whence
Le Le ( ) Le .
3. Let Le . Then Lmn since Lmn extends Le. In

view of Lmn we immediately obtain Lmn .
To prove the inverse implication we consider a formula such that Le
and a 4 -valuation v such that
v( ) = v( ( )) = 1.
The latter means that v( ) = 1. Let = (p1 , . . . , pn ). Due to epi-

morphism properties of (x) = x (see Lemma 3.2.3) we obtain
v((p1 , . . . , pn ) ) = v((p1 , . . . , pn )).
Consider a 2 -valuation v1 such that v1 (p1 ) := v(p1 ), . . . , v1 (pn ) :=

v(pn ). Then v1 () = v( ) = 1, which refutes the provability
Lmn .
2
Chapter 4
The Class of Extensions

of Minimal Logic1
In this chapter, we assign to every properly paraconsistent extension L of

minimal logic an intermediate logic Lint and negative logic Lneg called intu-
itionistic and negative counterparts of L, respectively. It will be proved that
the negative counterpart Lneg explicates the structure of contradictions of
paraconsistent logic L. We show that both counterparts Lint and Lneg are
faithfully embedded into the original logic L. Finally, we investigate a ques-
tion: to what extent is a logic L Par determined by its counterparts? As a
rst step, we study paraconsistent extensions of the logic
Le := Li Ln = Lj + { ( p)}.
The class of extensions of this logic has a nice property that every logic
L ELe Par is uniquely determined by its intuitionistic and negative
counterparts.
4.1 Extensions of Le

In this section, we state that properly paraconsistent extensions of Le are
exactly intersections of two logics, one of which is intermediate and the
other is negative. Prior to this, we study the algebraic semantics for logics
extending Le .
1
Parts of this chapter were originally published in [70] (Nicholas Copernicus University
Press, Poland) and in [76] (Elsevier, UK). Reprinted here by permission of the publishers.
41
42 4 The Class of Extensions of Minimal Logic
For an implicative lattice A =

A, , , , 1 and a A, we put
Aa := {b A | b a} and Aa := {b A | b a}.
The set Aa is obviously closed under the operations of A and we can dene
an implicative sublattice Aa of A, with the universe Aa . Except for the case
a = 1, the set Aa forms a sublattice but not an implicative sublattice of A,
because Aa is not closed under the implication (a a = 1). However, the
operation x a y := (x y) a turns Aa into an implicative lattice with
unit element a. Denote this implicative lattice by Aa .
If A is a j-algebra and a = , Aa can be treated as a Heyting algebra
and Aa as a negative one. In the following we call Heyting algebra A an
upper algebra of A. Negative algebra A is a lower algebra of A.
Recall one well-known fact from the theory of distributive lattices. Let A
be a distributive lattice, a an arbitrary element of A, and let sublattices Aa
and Aa be dened as above. The mappings (x) := x a and (x) := x a
are epimorphisms of A onto Aa and Aa respectively. The mapping (x) :=
(x a, x a) gives an embedding of A into the direct product of lattices Aa
and Aa .
These facts do not generally hold for implicative lattices. As before, the
mapping is an epimorphism of implicative lattices. But : A Aa and
: A Aa Aa are an epimorphism and an embedding of implicative
lattices only if some additional condition is imposed on A. More precisely,
the following assertions take place.
Proposition 4.1.1 For an implicative lattice A and a A, the mapping

: A Aa , (x) = x a, is an epimorphism of implicative lattices.
Proof. It follows from the denition of implication in Aa and the identity

(x y) z = ((x z) (y z)) z satised in all implicative lattices. The
latter fact follows from Item 9 of Proposition 2.1.2.
2
Proposition 4.1.2 Let A be an implicative lattice and a A. The following

three conditions are equivalent.
1. For all x, y A, we have
(x a) (y a) (x y) a.
2. The mapping : A Aa given by the rule (x) = x a is an epimor-

phism of implicative lattices.
4.1 Extensions of Le 43
3. The mapping : A Aa Aa given by the rule (x) = (x a, x a)

is an isomorphism of A onto a direct product of implicative lattices
Aa Aa .
Proof. 1 2. Check that preserves the implication, i.e., that the equality
(x y) a = (x a) (y a) holds. We have
Lp ((p q) r) ((p r) (q r))
by Item 8 of Proposition 2.1.2, whence the inequality (x y)a (xa)
(y a) is valid in any implicative lattice. The inverse inequality holds by
assumption.
2 3. It follows easily by assumption that is a homomorphism of A
into Aa Aa . We prove the injectivity of . Take an element b A, it is a
complement of a in the interval [b a, b a]. Assuming (c) = (b) for some
c A yields that c is a complement of a in the same interval [b a, b a].
We have b = c, since complements are unique in distributive lattices. Thus,
it remains to prove that maps A onto Aa Aa .
For x Aa and y Aa , we set z := (a y) x. The direct computation
shows that z a = y and z a = ((a y) a) x. Further, (a y) a =
(a a) (y a) = a a = 1 in view of the assumption that is a
homomorphism, whence z a = x. We have thereby proved (z) = (x, y).
3 1. Obviously, 3 implies 2. Therefore, the desired inequality follows
from the fact that preserves the implication.
2
Let us consider the following formulas
P. ((p q) p) p
E. p (p q)
D. ((p r) (q r)) ((p q) r)
We have
Lk+ = Lp + {P} = Lp + {E} = Lp + {D},
where Lk+ is the positive fragment of classical logic. It is well-known that
Lk+ is axiomatized relative to positive logic by the Peirce law P or by the
extended law of excluded middle E. It can be veried directly that D is true
on the 2-element Peirce algebra 2P . On the other hand, substituting p for
r in D, we immediately obtain Lp + D E. We have thus obtained that
D axiomatizes Lk+ modulo Lp. Combining this fact and Proposition 4.1.2
yields a characterization of Peirce algebras in terms of mappings and .
Proposition 4.1.3 Let A be an implicative lattice. The following condi-

tions are equivalent.
1. A is a Peirce algebra.
2. For any a A, the mapping a (x) = x a denes an epimorphism of

A onto Aa .
3. For any a A, the mapping a (x) = (x a, x a) denes an isomor-

phism of A and Aa Aa .
2
We now turn to the subsystem Le of Le, which can be axiomatized relative
to Lj by each of the following substitution instances of E and D:
E . ( p).
D . ((p ) (q )) ((p q) ).
The equality Lj + {E } = Lj + {D } can be checked as follows. On the
one hand, E follows from the instance of D obtained by replacing p for .
On the other hand, D is equivalent in Lj to (p (q )) ((p q) ),
and the latter formula follows in Lj from ( q). Indeed, implies ,
and formulas q and p (q ) imply p q.
Note a curious fact that the instance of the Peirce law
P = ((p ) p) p = (p p) p,
which is known as the Clavius law, is not equivalent to the above formulas
relative to Lj. Indeed, Lj P (p p) as a particular case of the
equivalence P E, and the logics Lj + (p p) and Le are incomparable
in the lattice of Lj-extensions. To prove the latter assertion, consider the
3-element linearly ordered Heyting algebra 3 and 3-element j-algebra 3
with the universe {1, , 1}, 1 1. It can be checked directly that
3 2 |= p p, 3 2 |= E , 3 |= E , 3 |= p p.
Consider the algebraic semantics for Le .
Proposition 4.1.4 Let A be a j-algebra. A is a model for Le if and only

if one of the following equivalent conditions holds.
1. The mapping (x) = x denes an epimorphism of the j-algebra A

onto the Heyting algebra A .
2. The mapping (a) = (a , a ) determines an isomorphism of

j-algebras A and A A .
3. For any a, b A with a b, has a complement in the interval
[a, b].
Proof. The inclusion Le LA is equivalent to the fact that D is an identity
of A, which is equivalent in its own right to Item 1 of Proposition 4.1.2 for
a = . In this way, Proposition 4.1.2 implies that each of the conditions 1, 2
characterizes models for Le . Proving Proposition 4.1.2 we established, in
fact, that 2 implies 3. Now we check the inverse implication, which completes
the proof.
Condition 3 means exactly that an embedding of distributive lattices
: A A A is onto, i.e., that is an isomorphism of distributive
lattices A and A A . The implication is dened in terms of the ordering
preserved by , consequently, also preserves the implication.
2
Corollary 4.1.5 Let L Jhn. Then Le L Le if and only if L =
L1 L2 , where L1 Int and L2 Neg.
Proof. Let L be an intersection of intermediate and negative logics L1 and
L2 . Then Li L1 and Ln L2 , whence Le = Li Ln L. It is clear that
the L is neither intermediate nor negative, therefore, L Par and L Le.
Conversely, let Le L Le and let A be a characteristic model for
L. By the above proposition A is presented as a direct product of Heyting
algebra A and negative algebra A , hence, L = LA = LA LA . It
remains to note that LA is an intermediate logic and LA is a negative
one.
2
4.1.1 Intuitionistic and Negative Counterparts

for Extensions of Le
First we state one more property of models for Le .
Letting A be a j-algebra, Le LA, put CA () := {a A | a = 1}
and decompose A into a direct product A A . Then CA () = {(1, b) |
b A }. Indeed, for a = (x, y) A A , we have
1 = a (1, 1) = (x, y) (0, 1) = (x, 1) x = 1.
It follows immediately that the set CA () is closed under , , and . We
will consider CA () as a negative algebra with operations induced from A
and 1 = .
Proposition 4.1.6 Let a j-algebra A be a model for Le . Then CA ()

=
A and the mapping (x) = x denes an epimorphism of the j-algebra
A onto the negative algebra CA ().
Proof. Again, we need a presentation of A as a direct product of Heyting and

negative algebras. The isomorphism CA () = A follows from the above
equality CA () = {(1, b) | b A }.
Check that is an epimorphism. For a = (b, c) A A , we have
(a) = (b, c) = (0, 1) (b, c) = (0 b, 1 c) = (1, c). Consequently, for
{, , } and for any (a, b), (c, d) A A ,
((a, b) (c, d)) = ((a c, b d)) = (1, b d) = (1, b) (1, d) = (a, b) (c, d).
It remains to note that () = 1 and (a) = a for any a CA ().
2
We are now ready to dene translations of intermediate and negative
logics into properly paraconsistent extensions of Le , which are similar to
translations of classical logic and maximal negative logic into Le dened at
the end of the previous chapter.
Theorem 4.1.7 Let L extend Le , L Par, and let A be a characteristic

model for L. Set L1 = LA and L2 = LA . Then for an arbitrary formula
, the following equivalences hold.
1. L1 L .
2. L2 L .
Proof. 1) Assume L1 and for an A-valuation v, compute the value

v( ). By Proposition 4.1.2, : A A is an epimorphism, from which
we have v( ) = v(). Here v denotes an A -valuation obtained as
a composition of v and . By denition L1 = LA , whence v() = 1. We
have thus proved that v( ) = 1 for any A-valuation v, i.e., L .
Conversely, let L . Every A -valuation v can be treated as an
A-valuation with the property v = v. As above, we have v() = v() =
v( ) = 1, which immediately implies L1 .
2) This proof is similar to the previous one with replaced by .
2

As follows from the theorem, the logics L1 := LA and L2 := LA
do not depend on the choice of a characteristic model A for the logic L
extending Le . Indeed,
L1 = { | L }, L2 = { | L }.
It is clear that L1 Int and L2 Neg. We call the logics L1 and L2 dened as
above intuitionistic and negative counterparts of L Le and denote them
Lint and Lneg respectively. Since A = A A , we have L = Lint Lneg .
Let, on the contrary, L = L1 L2 , where L1 Int and L2 Neg.
For a suitable Heyting algebra B and for some negative algebra C, we have
L1 = LB and L2 = LC. The direct product A = B C is a characteristic
model for L since L(B C) = LB LC = L1 L2 = L. Moreover, B = A
and C = A , consequently, L1 = Lint and L2 = Lneg . In this way, we arrive
at the following statement.
Proposition 4.1.8 The mapping L (Lint , Lneg ) denes a lattice iso-

morphism between [Le , Le] and the direct product Int Neg. The inverse
mapping is given by the rule (L1 , L2 ) L1 L2 .
Proof. In fact, it was stated above that the mapping under consideration is
a bijection. According to Theorem 4.1.7 it preserves an ordering. It remains
to note that an order isomorphism of two lattices is a lattice isomorphism
too.
2

We can now describe the class of models for L Le as follows.
Proposition 4.1.9 Let L Le . A j-algebra A is a model for L if and only

if A
= A A , A |= Lint , and A |= Lneg .
Proof. Let A |= L. According to Proposition 4.1.4 the condition L Le

implies A = A A . Denote L := LA. Then A |= Lint and A |= Lneg
by Theorem 4.1.7. In view of the previous proposition Lint Lint and
Lneg Lneg , whence A |= Lint and A |= Lneg .
Conversely, let A = A A , A |= Lint , and A |= Lneg . Then the
direct product A is a model for the intersection Lint Lneg . But L Le ,
hence, L = Lint Lneg by Corollary 4.1.5.
2

Thus, the class of properly paraconsistent extensions of Le is completely
described in terms of intermediate and negative logics. It should be em-
phasized that the mapping dened in Proposition 4.1.8 has an essentially
eective character. Theorem 4.1.7 allows one to eectively reconstruct intu-
itionistic and negative counterparts from the given paraconsistent L, whereas
the L itself is simply an intersection of its counterparts, i.e., a formula is
proved in L if and only if it is proved in both Lint and Lneg .
However, the interval [Le , Le] constitutes a relatively small part of the
class Par of all properly paraconsistent extensions of Lj. There are many
interesting logics, which do not belong to this interval. One of them is the
Glivenko logic treated in the beginning of the next chapter.
4.2 Intuitionistic and Negative Counterparts

for Extensions of Minimal Logic
As the rst stage in studying the whole class Par we dene intuitionistic and
negative counterparts for an arbitrary extension of minimal logic.
For extensions of Le , our denitions will be equivalent to those of the
previous section.
We dene the following translation
In() = , In(p) = p , In( ) = In() In(),
where p is a propositional variable, and arbitrary formulas, and

{, , }. In other words, if = (p0 , . . . , pn ), then
In() = (p0 , . . . , pn ).
For L Jhn+ , dene
Lint := { | L In()}, Lneg := { | L }.
It can easily be seen that Lint and Lneg are logics. Moreover, Li Lint
since (p ) Lj, and Ln Lneg in view of Lj. We call Lint
and Lneg intuitionistic and negative counterparts of the logic L respectively.
Notice that this denition of negative counterpart is exactly the same
as the denition of negative counterparts for Le -extensions given in the
previous section. As for Lint , using formula D we can easily prove in Le
the equivalence ( ) In() for any formula . Therefore, if L extends
Le , Lint coincides with the intuitionistic counterpart dened in the previous
section.
List some simple properties of the notions introduced above.
Proposition 4.2.1 1. For any L Lj, we have Lint Int {F}, Lneg
Neg {F}, and L Lint Lneg . The last inclusion is not proper if
and only if L extends Le .
2. L Int if and only if L = F, L = Lint , and Lneg = F.
3. L Neg if and only if L = F, L = Lneg , and Lint = F.

4.2 Intuitionistic and Negative Counterparts for Extensions of Minimal Logic 49
4. If Lj L1 L2 , then L1int L2int and L1neg L2neg .
5. If L L1 Int, then Lint L1 .
6. If L L1 Neg, then Lneg L1 .
Proof. We only prove the last two assertions.

5. If L In(), then also L1 In(). Since L1 is intermediate, we have
L1 In() , and so L1 , which implies the desired inclusion.
6. Again, from L1 we conclude L1 since belongs to any
negative logic.
2
We have thus proved, in particular, that Lint is the least intermedi-
ate logic containing L, and Lneg is the least negative logic with the same
property. It can easily be seen that the mappings ()int : Jhn+ Int and
()neg : Jhn+ Neg can be dened as follows. For any L Jhn+ , put
Lint := L + { p} = L + Li and Lneg := L + {} = L + Ln.
Proposition 4.2.2 The mappings ()int and ()neg are lattice epimor-
phisms.
Proof. This fact easily follows from the distributivity of Jhn+ (Proposi-
tion 2.1.6).
2
Further, we prove that upper and lower algebras associated with a given
j-algebra are semantic analogs of intuitionistic and negative counterparts.
Proposition 4.2.3 For any j-algebra A and formula , the following equiv-
alences hold.
1. A |= In() A |= .
2. A |= A |= .
Proof. 1. Assume A |= and prove A |= In(). For an A-valuation v, dene

an A -valuation v by the rule v (p) := v(p) . Then it follows easily that
v(In()) = v (), which immediately implies the desired conclusion.
Conversely, let A |= In(). For any A -valuation v, we have v = v , in
particular, v(In()) = v(), which completes the proof.
2. We use the mapping (x) = x , which is an epimorphism of A onto
A by Proposition 4.1.1. Note also that is equivalent to ( )
in Lj.
Assuming A |= we take an A-valuation v and consider the com-

position v, which is an A -valuation. By epimorphism properties of
we have v( ) = v(). But v() = by assumption, which yields
v( ( )) = 1. Thus, A |= ( ) by an arbitrary choice of v.
Now, we let A |= ( ). Clearly, v = v for any A -valuation v.
By assumption v() v( ) = v() = v(). The greatest element of
A is , whence v() = . In this way, A |= .
2
Corollary 4.2.4 Let L Jhn+ .
1. If A is a model for L, then
A |= Lint and A |= Lneg .
2. If A is a characteristic model for L, then
LA = Lint and LA = Lneg .
Consider classes of logics with given intuitionistic and negative counter-

parts. For L1 Int and L2 Neg, we dene
Spec(L1 , L2 ) := {L Lj | Lint = L1 , Lneg = L2 }.
It is clear that for any pair of intermediate and negative logics, (L1 , L2 ), the
set Spec(L1 , L2 ) is non-empty. It contains at least the intersection L1 L2 .
Moreover, in view of Item 1 of Proposition 4.2.1 L1 L2 is the greatest
element of Spec(L1 , L2 ). It turns out this set also contains the least element
and forms an interval in the lattice of Lj-extensions. Let
L1 L2 := Lj + {In(), | L1 , L2 },
where L1 Int and L2 Neg.
Proposition 4.2.5 Let L1 Int and L2 Neg. Then
Spec(L1 , L2 ) = [L1 L2 , L1 L2 ].
Proof. Let L := L1 L2 . It follows from denition that L1 Lint and L2

Lneg . On the other hand, for any L Spec(L1 , L2 ), we have L L. Indeed,
L contains all axioms of L . As noted above, L1 L2 is the greatest element
4.2 Intuitionistic and Negative Counterparts for Extensions of Minimal Logic 51
of Spec(L1 , L2 ), whence, by Item 4 of Proposition 4.2.1 the logic L and all

logics intermediate between L and L1 L2 also belongs to Spec(L1 , L2 ).
2
The logic L1 L2 , the least element of the interval Spec(L1 , L2 ), we call
a free combination of logics L1 and L2 . This name is justied by the next
proposition saying that models for L1 L2 are all j-algebras such that their
upper and lower algebras are models for L1 and L2 , respectively.
Proposition 4.2.6 Let L1 Int and L2 Neg. For any j-algebra A, we

have A |= L1 L2 if and only if A |= L1 and A |= L2 .
Proof. This statement easily follows from the denition of free combination
and Corollary 4.2.4.
2
The next proposition allows one to write axioms for L1 L2 relative to
Lj given an axiomatics of L1 relative to Li and of L2 relative to Ln.
Proposition 4.2.7 Let L1 Int, L1 = Li + {i | i I} and L2 Neg,

L2 = Ln + {j | j J}. Then
L1 L2 = Lj + {In(i ), j | i I, j J}.
Proof. Denote the right-hand side of the last equality by D. The inclusion
D L1 L2 is trivial. To state the inverse inclusion we show that L1 Dint
and L2 Dneg .
We argue for L2 Dneg . Note that Ln = Ljneg , i.e., Ln if and only
if Lj . Assume L2 , then Ln (j 1 . . . j n ) for suitable
particular cases j 1 , . . . , j n of axioms j1 , . . . , jn , j1 , . . . , jn J. Whence
Lj ((j 1 . . . j n ) ).
The last formula implies in Lj
(( j 1 ) . . . ( j n )) ( ),
from which we infer D. Consequently, L2 Dneg .

The remaining inclusion follows in the same way from the equality Li =
Ljint .
2
As we can see from Proposition 4.2.5, the class of properly paraconsistent
Lj-extensions decomposes into a union of disjoint intervals

Par = {Spec(L1 , L2 ) | L1 Int, L2 Neg}.
Fr
X XX
XXX
XXX
XX
'
Lmn rX
XX$ ' X
rLk $
XXX
XXX
XX Le
' X
PPr $
PP
Li Lmn r PPrLk Ln
A
A
Neg A Int
A
A H rLk Lmn
A HH
A H
&
Ln Pr Li% Lmn r A H
HrLk Ln& rLi %
QPP S A

Q PP
Q PPS A
Q PSPP A
Q
Q S PPA r
Q
Q S Le
Q S
Q S
Q Par %
& QSr
Lj
Figure 4.1
It is interesting that the upper points of these intervals also form an interval
in Jhn+ , [Le , Le]. Figure 4.1 illustrates the structure of the class Jhn+ .
In this way, the investigation of the class of Lj-extensions is reduced to
the problem: what is the structure of the interval Spec(L1 , L2 ) for the given
intermediate logic L1 and negative logic L2 ? This problem will be treated
in the subsequent sections but rst we make an observation on the nature
of the negative counterpart Lneg of a paraconsistent logic L.
4.2.1 Negative Counterparts as Logics of Contradictions

We dene a contradiction operator C() := and extend this operator
to sets of formulas as follows. Put C() := {} and C(X) := {C() |
X} for X = . The contradiction operator is trivial in all intermediate
logics. If the law ex contadictione quodlibet holds, we have C() for
any formula . Rejecting ex contadictione quodlibet we obtain the possibility
to distinguish contradictions constructed with the help of dierent formulas.
In particular, for L Par we have L C() if and only if Lneg .
Moreover, it turns out that relative to deducibility properties, the behavior
4.3 Three Dimensions of Par 53
of formulas in the negative counterpart Lneg is completely similar to the

behavior of contradictions constructed with the help of these formulas in
the original logic L. More precisely, the following fact takes place.
Proposition 4.2.8 Let L Par. For an arbitrary set of formulas X and

for any formula , the following equivalence holds:
X Lneg C(X) L C().
The proof of this proposition is an easy exercise on the deducibility in min-

imal logic.
2
We have thus proved that the contradiction operator denes a strong
translation of the negative counterpart Lneg in a paraconsistent logic L
Par. This fact allows one to consider the negative counterpart Lneg as a logic
of contradictions associated with a given paraconsistent logic L.
4.3 Three Dimensions of Par

We can see now that the class Par has a three-dimensional structure. The
position of a logic L in this class is determined by its intuitionistic counter-
part Lint , which represents reasoning in L under the additional assumption
of inconsistency, or of impossibility of contradictions, and by its structure
of contradictions explicated in the negative counterpart Lneg . When an ex-
plosive pattern of reasoning and a structure of contradictions are xed, we
have a variety of possibilities for combining them presented by the interval
of logics Spec(Lint , Lneg ). The place of L in this interval can be considered
as its third coordinate in Par, the sense of which is not quite clear yet. It
becomes clearer in the next chapter. Now we turn to the question of a scale
for this third coordinate. Unlike rst and second coordinates having abso-
lute scales, Int and Neg respectively, the intervals Spec(I, N ) are mutually
disjoint for dierent pairs of logic I Int and N Neg. However, one can
nd natural interrelations between these scales, i.e., between the intervals
of the form Spec(I, N ) for various I Int and N Neg.
Consider two pairs of logics P1 = (I1 , N1 ) and P2 = (I2 , N2 ), where
I1 , I2 Int, N1 , N2 Neg. P1 P2 means that I1 I2 and N1 N2 . We
write also Spec(P1 ) for Spec(I1 , N1 ).
Let P1 = (I1 , N1 ) and P2 = (I2 , N2 ) be such that P1 P2 . Mappings
rP2 ,P1 : Spec(P2 ) Par and eP1 ,P2 : Spec(P1 ) Par are dened as follows
rP2 ,P1 (L) := L (I1 N1 ), eP1 ,P2 (L) := L + (I2 N2 ).
Proposition 4.3.1 Let pairs of logics P1 and P2 be such that P1 P2 . The

following facts hold.
1. For any L Spec(P2 ), we have eP1 ,P2 rP2 ,P1 (L) = L.

2. For any L Spec(P1 ), we have
rP2 ,P1 eP1 ,P2 (L) = L + rP2 ,P1 (I2 N2 )
3. eP1 ,P2 is a lattice epimorphism from Spec(P1 ) onto Spec(P2 ).

4. rP2 ,P1 is a lattice monomorphism from Spec(P2 ) into Spec(P1 ) and
rP2 ,P1 (P2 ) = [rP2 ,P1 (I2 N2 ), I1 N1 ].
5. For any P3 such that P2 P3 , we have
eP1 ,P2 eP2 ,P3 = eP1 ,P3 , rP3 ,P2 rP2 ,P1 = rP3 ,P1 .
Proof. 1. We calculate
eP1 ,P2 rP2 ,P1 (L) = (L (I1 N1 )) + (I2 N2 )
= (L + (I2 N2 )) ((I1 N1 ) + (I2 N2 )).

By Proposition 4.2.5 I2 N2 is the least point of Spec(P2 ), therefore, we have
L + (I2 N2 ) = L. Further, we need one lemma.
Lemma 4.3.2 For any L Spec(I, N ), I N = L + Le .
Proof. (Le )int equals to Li, the least logic in Int, and (Le )neg = Ln, which
is the least logic in Neg. Now, it follows from Proposition 4.2.2 that L + Le
has the same counterparts as L. By Corollary 4.1.5 L + Le coincides with
the greatest point of Spec(I, N ).
2
Using this lemma and the obvious relation I1 N1 I2 N2 we obtain
(I1 N1 ) + (I2 N2 ) = ((I1 N1 ) + Le ) + (I2 N2 ) = I2 N2 + Le = I2 N2 .
And nally, eP1 ,P2 rP2 ,P1 (L) = L (I2 N2 ) = L.

2. The direct computation shows
rP2 ,P1 eP1 ,P2 (L) = (L + (I2 N2 )) (I1 N1 )
= (L (I1 N1 )) + ((I2 N2 ) (I1 N1 )) = L + rP2 ,P1 (I2 N2 ).

4.3 Three Dimensions of Par 55
3. It follows from the distributivity of Jhn+ that eP1 ,P2 is a lattice ho-
momorphism. Let L Spec(P1 ) and L := eP1 ,P2 (L) = L + (I2 N2 ). By
Proposition 4.2.2 (L )int = Lint + (I2 N2 )int = I1 + I2 = I2 . In the same
way, (L )neg = N2 . Consequently, L Spec(P2 ). The fact that eP1 ,P2 is an
epimorphism follows from Item 1.
4. As above, we use Proposition 4.2.2 to check that rP2 ,P1 maps Spec(P2 )
into Spec(P1 ). This is a homomorphism due to the distributivity of Jhn+ . If
rP2 ,P1 (L1 ) = rP2 ,P1 (L2 ), then applying the formula of Item 1 we obtain L1 =
L2 . Thus, rP2 ,P1 is a monomorphism. The equality rP2 ,P1 (P2 ) = [rP2 ,P1 (I2
N2 ), I1 N1 ] follows from Item 2.
5. This item follows from the obvious relations I2 N2 I3 N3 and
I1 N1 I2 N2 .
2
The above proposition shows, in particular, that any interval Spec(I, N )
is isomorphic to an upper subinterval of Spec(Li, Ln). In this way, the latter
interval can be considered as a scale for the third dimension of the class Par.
Extending intuitionistic and negative counterparts, we restrict simultane-
ously the part of the scale that can be used to construct a logic with given
counterparts. It is also worth noticing the following consequence of the last
proposition.
Corollary 4.3.3 Let P1 = (I1 , N1 ) and P2 = (I2 , N2 ) be pairs of logics such

that P1 P2 . For any logics L1 , L2 Spec(P2 ), L1 = L2 , there is a formula
I1 N1 such that (L1 \ L2 ) (L2 \ L1 ).
Proof. Let L1 , L2 Spec(P2 ). If L1 = L2 , but these logics are not distin-

guished by a formula I1 N1 , then rP2 ,P1 (L1 ) = rP2 ,P1 (L2 ). By Item 4
of the previous proposition rP2 ,P1 is a monomorphism, whence L1 = L2 , a
contradiction.
2
In particular, any two logics from Spec(I, N ) can be distinguished via a
formula from Le = LiLn. Moreover, any logic from the interval Spec(I, N )
can be axiomatized by formulas from Le modulo the least logic of the in-
terval I N . Indeed, for any L Spec(I, N ) we have by Item 1 of Propo-
sition 4.3.1 L = (L Le ) + (I N ). In this way, any possible combination
of intuitionistic and negative logics can be determined by corollaries of the
formula ( p).
For further investigations of the class Par, we need semantic considera-
tions.
Chapter 5
Adequate Algebraic
Semantics for Extensions
of Minimal Logic1
The goal of this chapter is to nd a representation of j-algebras, convenient

for working with logics lying inside the intervals Spec(L1 , L2 ). We have to
understand the structure of an arbitrary j-algebra A with given upper alge-
bra A and lower algebra A . The semantic characterization of Glivenkos
logic considered in Section 5.1 prompts the solution to this problem. The
desired representation is described in Section 5.2. In Section 5.3 with the
help of the obtained representation we characterize the Segerberg logics and
demonstrate its eectiveness in this way. Finally, in Section 5.4 we consider
the Kripke semantics and dene for j-frames analogs of upper and lower
algebras associated with a j-algebra.
5.1 Glivenkos Logic

Consider the following substitution instance of the Peirce law:
P . (( p) ) = ( p).
We call the logic Lg := Lj + {P } Glivenkos logic. It was mentioned in

[98, p. 46] that Glivenkos logic is the weakest one in which is derivable
whenever is derivable in classical logic. Unfortunately, this work contains
neither the proof of this assertion, nor any further reference. In this section,
1
Parts of this chapter were originally published in [69].
57
58 5 Adequate Algebraic Semantics for Extensions of Minimal Logic
we present a natural algebraic proof of this statement. We also show that

Lg is a proper subsystem of Le .
Proposition 5.1.1 1. Let A be a j-algebra. Then A is a model for Lg

if and only if ( a) Fd (A ) for any a A.
2. Let A be a model for Lg and := Fd (A ). Then the mapping (a) =
(a )/ denes an epimorphism of A onto A /.
Proof. 1. This item immediately follows from the denition of Glivenkos

logic and the fact that (a ) = a = a = a for any
j-algebra A and a A. The last equality implies, in particular,
Fd (A ) = {a A | a = 1}.
2. In fact, we need only check that preserves the implication, i.e.,
(a b) / = (a ) (b )/.
We have ((a b) ) ((a ) (b )) = 1 , since the correspond-
ing formula is provable in Lj (see Item 8 of Proposition 2.1.2). Further, it
can be veried directly that
Lj ( q) (((p ) (q )) ((p q) )).
In view of Lj (p q) (p q) we obtain
Lj ( q) (((p ) (q )) ((p q) )),
i.e.,
Lg (((p ) (q )) ((p q) )).
By assumption A is a model for Lg, consequently,
((a ) (b )) ((a b) ) ,
which completes the proof.
2
Theorem 5.1.2 (Generalized Glivenkos Theorem.) For every logic L

Jhn, the following conditions are equivalent.
1. For any ,
Lk L .
2. L Lg and L Neg.
5.2 Representation of j-Algebras 59
Proof. 1 2. This implication is trivial.

2 1. Since L Neg, the logic L is contained in Lk. If L , then
Lk , and so Lk .
Assume Lk . Let A |= L and v be an A-valuation. Consider an
A /Fd (A )-valuation v such that v (p) = v(p )/Fd (A ). Using the as-
sumption Lg L and Proposition 5.1.1 we obtain v () = v()/Fd (A ).
On the other hand, A /Fd (A ) is a Boolean algebra and Lk , therefore,
v () = 1. In this way, v( ) Fd (A ), i.e., v(( )) = v() = 1.
Since A and v are arbitrary, we obtain L .
2
Let us prove that Glivenkos logic does not belong to the class of Le -
extensions. To this end it will be enough to show that Glivenkos logic has
models dierent from direct products of Heyting and negative algebras.
Proposition 5.1.3 Let A be a model for Le , and B a Heyting algebra.

Then A B is a model for Lg.
Proof. It follows from two facts. For all a A B, we have ( a) B.

All elements of B are dense in (A B) .
Corollary 5.1.4 The inclusion Lg Le is proper.
Proof. Indeed, according to Proposition 4.1.4 the algebra A B is not a

model for Le if B is a non-trivial Heyting algebra. But this is a model of
Glivenkos logic by the previous proposition.
2
5.2 Representation of j -Algebras

In this section we give a convenient representation of j-algebras, which allows
one to describe classes of models for logics lying inside intervals of the form
[L1 L2 , L1 L2 ], where L1 Int and L2 Neg. We know that an intersec-
tion L1 L2 of intermediate and negative logics is characterized by the class
of all direct products of the form AB, where A is a Heyting algebra being a
model for the logic L1 and B is a negative algebra modelling L2 . Indeed, due
to Corollary 4.1.5 the intersection L1 L2 extends Le and each model A for
L1 L2 is isomorphic to the direct product A A by Proposition 4.1.4.
It remains to note that by Proposition 4.1.8 we have L1 = (L1 L2 )int
and L2 = (L1 L2 )neg . Thus, Proposition 4.1.9 implies A |= L1 and
A |= L2 .
At the same time, a free combination of logics, L1 L2 , is characterized

by the class of all j-algebras A such that the upper algebra A is a model
for L1 and the lower algebra A models L2 (see Proposition 4.2.6). At this
point the following question arises. If a Heyting algebra B and a negative
algebra C are given, what is the dierence between an arbitrary j-algebra A
with the condition A = B and A = C and the direct product of algebras
B C? Proposition 5.1.1 allows us to assume that elements of the form
( a), where a A , play a special part in the structure of A.
Proposition 5.2.1 Let A be a j-algebra and a mapping fA : A A is

given by the rule fA (x) = ( x). Then the following two conditions
are met.
1. The mapping fA : A A is a semilattice homomorphism preserv-

ing the meet and the greatest element, fA () = 1;
2. The embedding : A A A , where (x) = (x , x ), has

the following image
(A) = {(x, y) | x fA (y), x A , y A }.
Proof. 1. For brevity, we omit the lower index in denotation fA . We have

f () = ( ) = 1. Further,
f (y1 ) f (y2 ) = ( ( y1 )) ( ( y2 )) =
= (( y1 ) ( y2 )) = ( (y1 y2 )) = f (y1 y2 ).
We have thus veried that f is a semilattice homomorphism preserving the
meet and the unit element.
2. If a A, then (a , a ) (A) and the inequality a
( (a )) holds. The latter can be checked, for example, by proving
in Lj the formula p ( p ). Thus, the inclusion
(A) {(x, y) | x f (y), x A , y A }
is proved. Now, let x, y A, x , y , and x ( y). We show

that there is an element a A such that x = a and y = a . Put
a := x ( y), then
a = ( x) ( ( y)) = x ( ( y)) = x,
5.2 Representation of j-Algebras 61
moreover, a = x ( y) = ( y) = y. The inverse inclusion

is also checked.
2
As we can see from the above proposition, every j-algebra A determines a
triple (A , A , fA ) consisting of a Heyting algebra, a negative algebra and a
semilattice homomorphism. Now, let us take a triple (B, C, f : C B), where
B is an arbitrary Heyting algebra, C a negative algebra and f a semilattice
homomorphism from C to B preserving the meet and the greatest element.
Starting from this triple we try to construct a j-algebra A, the upper and
lower algebras of which are isomorphic to B and C respectively, and the
mapping fA is induced in a natural way by the homomorphism f .
Dene a lattice B f C as a sublattice of the direct product B C with
the universe
|B f C| := {(x, y) | x B, y C, x f (y)}.
This is really a sublattice of B C, because f preserves the meet and, hence,
the ordering, which easily implies the relation f (y1 ) f (y2 ) f (y1 y2 ).
From the latter immediately follows that the set |B f C| is closed under
componentwise lattice operations on the direct product of lattices. As we can
see from the proposition below, this lattice can be considered as a j-algebra.
Proposition 5.2.2 Let B, C, f , and A := B f C be as above. The lattice A

has a natural structure of j-algebra, where the relative pseudo-complement
operation is given by the rule
(x1 , y1 ) (x2 , y2 ) = ((x1 x2 ) f (y1 y2 ), y1 y2 ),
1A = (1B , C ), and A = (B , C ). Moreover, B = A , C
= A , and these
isomorphisms are given by the rules x (x, C ), x B, and y (B , y),
y C, respectively. Finally, for all y C, we have
(f (y), C ) = A (A (B , y)) = fA ((B , y)).
Proof. First, we check that the relative pseudo-complement is well dened.

Let b1 , b2 B, c1 , c2 C, b1 f (c1 ), and b2 f (c2 ). The element (b1 , c1 )
(b2 , c2 ), if it is dened, is the greatest among all elements (x, y) such that
x f (y) and (b1 , c1 ) (x, y) (b2 , c2 ). This is equivalent to relations x
(b1 b2 ) f (y) and y c1 c2 . Taking into account that f preserves the
ordering, we immediately obtain that the element
((b1 b2 ) f (c1 c2 ), c1 c2 )
is the desired relative pseudo-complement.
All other relations, except the last, are trivial. Check the last relation
using the obtained formula for relative pseudo-complement. We have
A (A (B , y)) = (B , C ) ((B , C ) (B , y)) =
= (B , C ) (1B f (C y), C y) = (B , C ) (f (y), y) = (f (y), C ).

2
As we can see from the above considerations, to dene a class of j-
algebras characterizing some extension L of minimal logic, we must choose a
class of Heyting algebras and a class of negative algebras isomorphic to upper
and lower algebras respectively, associated with models for L. In this way,
we x intuitionistic and negative counterparts of the logic L. Moreover, to
determine the place of L inside the interval [Lint Lneg , Lint Lneg ], we have
to distinguish in one or another way the class of admissible homomorphisms
from negative algebras into Heyting ones. If no restrictions are imposed on
the class of homomorphisms, we obtain a free combination of intermediate
and negative logics characterized by the selected classes of Heyting and
negative algebras (see Proposition 4.2.6). If we admit only homomorphisms
identically equal to the unit element, we obtain the intersection Lint Lneg .
Indeed, a j-algebra A f B coincides with the direct product A B if and
only if f (y) = 1 for all y B.
5.3 Segerbergs Logics and their Semantics

It is interesting to consider logics dierent from intersections and free com-
binations of intermediate and negative logics, i.e., logics lying inside in-
tervals of the form Spec(L1 , L2 ). In this section, using the representation
for j-algebras obtained above, we describe an algebraic semantics for logics
studied previously by K. Segerberg [98], who characterized these logics in
terms of Kripke semantics. Except for Lj K. Segerberg [98] considered logics
obtained by adding to Lj one or several axioms from the list below.
I. p
K. p p
X. p p
L. (p q) (q p)
E. p (p q)
5.3 Segerbergs Logics and their Semantics 63
L . p ( p)(= (p ) ( p))
E . ( p)
Q.
LN . (p q ) (q p )
LQ
1 . (p q) (q p)
LQ
2 . ( (p q)) ( (q p))
EQ
1 . p (p q)
EQ
2 . ( p) ( (p q))
P . ( p)(= (( p) ) )
We may combine these axioms, which gives rise to a large number of

new logics. Some of these logics have traditional denotation, for example
Li = Lj + {I}, and others have not. Due to this fact, we need some notational
conventions.
If some logic is obtained from the logic already having a denotation,
say L, by adding some axiom denoted by a capital letter, say X, then the
denotation of this new logic will be obtained by joining the corresponding
small letter to the existing denotation, Lx := L+{X}. Of course, in this way
one logic may obtain dierent denotations. According to this convention we
have, for example, Lji = Li, Lje = Le, Ljq = Ln, Lix = Ljix = Lk, and
nally, Ljp = Lg.
We shall say a few words on how the above list of axioms arises. The
Kripke semantics for extensions of Lj was described in Chapter 2. Recall
that any j-frame is divided into two parts consisting of normal worlds and
of abnormal worlds. The rst axiom I distinguish the class of j-frames in
which all worlds are normal. The next two axioms are the well-known law
of excluded middle X and weak law of excluded middle K. These axioms
impose some restrictions on the accessibility relation only in the normal part
of a j-frame [98]. It must be identical in case of X and directed in case of K.
The Dummett linearity axiom L and the extended law of excluded middle
E dene properties of accessibility relation in the whole frame. A j-frame
satisfying L is linear, whereas in a j-frame satisfying E the accessibility
relation is identical. The next two axioms, L and E , are particular cases of L
and E, respectively. They do not impose any restrictions on either the normal
(rLk
(
r ( ( ((((
((((Le
( ( ( r(((
( Lje x
r (((
( (rLil
(
Ljx
r( ( ((((
(((( Lje l
r(((
N
Lje l

((r

( ( (((( Ljl
(

r(

( ((((
(
(((( LjlN l (((
((rLik
r((( ( ( (
LjlN (((
( ( (r (((
(((( Lje k
( ( (((r( ((rLi
r((( Ljkl
( ( ( (((
(((
Ljk
( ( (r (((
(((( Lje
(( ( ((r(
r ((
( Ljl
Lj
Figure 5.1
or abnormal part of a j-frame, but they dene the way in which the cone of
abnormal worlds is situated in the whole frame (see Proposition 2.3.3).
The interrelations between logics obtained by joining to Lj one or several
axioms reviewed up to this point are presented in Figure 5.1.
Note that this diagram (as well as the diagram presented in Figure 5.2
below) respects only the ordering, but not the lattice structure of Jhn+ . All
logics presented at the diagram are distinct, and a logic L1 is contained in a
logic L2 if and only if there is a path leading from L1 to L2 , which at every
point is either rising or horizontal and directed to the right.
To explain the explicit irregularities of the above diagram K. Segerberg
put some new axioms into consideration, which are not as natural at rst
glance as the axioms considered up to this point. But as long as we can-
not account for the irregularities in the above diagram, we cannot claim to
understand the situation fully [98, p. 41].
As we can see from the above the axiom X can be considered as a
relativization of the axiom E to the normal part of a j-frame. Indeed, the
axiom E imposes the condition to be identical on the accessibility relation,
whereas X imposes essentially the same condition to be identical but on
the accessibility relation restricted to the normal part of a j-frame. The next
six axioms in the list are the axiom Q distinguishing the class of abnormal
r Lnl
r Lmn
r Ln
Le
u r u uLk

r r
Ljx u u u r uLil

Ljl

u r

u u r r uLik

LjlN u
r r
u
r r Leu r r uLi

Ljk u r r Ljl u r r

u r r

Lj LjlQ
2 LjeQ
2
Figure 5.2
j-frames and relativizations of axioms E and L to the normal or to the

abnormal part of a j-frame. The axiom LN is a restriction of L to normal
worlds. The axioms LQ Q
1 and L2 are variants of relativization of L to the
abnormal part of a j-frame. Relativizing E to abnormal words K. Segerberg
also suggests two variants, EQ1 and E2 .
Q
The last axiom in the list, P , is similar to E and L because it restricts

only the way of combination of normal and abnormal parts of a j-frame (see
Proposition 2.3.4). This axiom, as well as axioms LQ Q
1 and E1 lie out of the
main line of considerations in [98].
If we exclude from the above list the axioms P , LQ Q
1 , and E1 , the logics
that can be constructed via adjoining to Lj the other axioms from the list
form the beautiful diagram presented in Figure 5.2. The logics of Figure 5.1
are depicted in this diagram by bigger circles. The way, in which these
logics are situated in Figure 5.2, explains the irregularities of the previous
diagram. Only a few points on the diagram are endowed with the names
of corresponding logics. The other logics are obtained via a combination of
axioms of explicitly designated logics and one can easily reconstruct which
logic corresponds to one or another point on the diagram. For example,
the non-designated logics lying on the horizontal line ended with Lik are the
following: Ljke , Ljke lQ Q
2 , Ljke e2 (from left to right). We also note the
equalities Ljl = Ljl lN lQ Q
2 and Le = Lje xe2 . As we will see the equality
Ljl = Ljl lN lQ Q Q
1 does not hold. So using axiom L1 instead of L2 results in a
diagram of logics, which is not as regular as that of Figure 5.2. This explains
the choice of K. Segerberg between variants of relativization of the axiom L
to abnormal worlds.
In this diagram there are only four intermediate logics, namely, the logics
lying on the vertical line from Li to Lk. The three negative logics on the
diagram are those lying on the horizontal line from Ln to Lmn. All other
logics on the diagram belong to the class Par. They form a three-dimensional
gure, the dimensions of which, as we can see later, correspond to the three
parameters, which determine the position of a paraconsistent logic in the
class Par.
To better explain this correspondence we turn to the algebraic semantics
of Segerbergs logics.
Recall that a Stone algebra is a Heyting algebra satisfying the identity K.
Let A be a Heyting (negative) algebra. We call A a Heyting (negative)
l-algebra if A |= (p q) (q p).
Proposition 5.3.1 Let A be an arbitrary j-algebra. The following equiva-

lences hold.
1. A |= Ljk if and only if A is a Stone algebra.
2. A |= Ljx if and only if A is a Boolean algebra.
3. A |= Ljl if and only if fA (A ) R(A ).
4. A |= Ljl if and only if A and A are l-algebras, fA (A ) R(A ),

and for all y1 , y2 A , we have fA (y1 y2 ) fA (y2 y1 ) = 1.
5. A |= Lg if and only if fA (A ) Fd (A ).
6. A |= LjlN if and only if A is a Heyting l-algebra.

7. A |= LjlQ
1 if and only if A is a negative l-algebra.
8. A |= LjlQ
2 if and only if A is a negative l-algebra and for all y1 , y2
A , we have fA (y1 y2 ) fA (y2 y1 ) = 1.
9. A |= LjeQ
1 if and only if A is a negative Peirce algebra.
10. A |= LjeQ 2 if and only if A is a negative Peirce algebra and for all
y1 , y2 A , we have fA (y1 ) fA (y1 y2 ) = 1.
Proof. 1. Let A |= Ljk. We represent A in the form A fA A , take an

arbitrary element (x, y) A, and compute
((x, y) (, )) (((x, y) (, )) (, )) = (x , y )
((x , y ) (, )) = (x, ) ((x, ) (, )) =

= (x, ) (x, ) = (x x, ) = (1, ).
The latter identity is satised if and only if the identity x x = 1 is
true on A , i.e., if and only if A is a Stone algebra.
2. This item can also be proved via a direct computation.
3. Let (x, y) A fA A . The direct computation shows
((x, y) (, )) ((, ) (x, y)) = ((x ) f (y), ).
Here after, we omit the lower index in the denotation fA if it does not lead
to confusion. As we can see, L is an identity of A if and only if for all
x A , y A , x f (y), the equality (x ) f (y) = 1A holds. In
particular, we have (f (y) )f (y) = f (y)f (y) = 1 , i.e., each element
of the form f (y) is regular. The inverse implication immediately follows from
the above and the fact that the implication is descending with respect to the
rst argument. Indeed, if for some y A we have (f (y) )f (y) = 1A ,
then for all x A , x f (y), we also have (x ) f (y) = 1A .
4. Assume that A |= Ljl. In this case, the upper algebra A is a Heyting
l-algebra as a subalgebra of A. The inclusion fA (A ) R(A ) holds by
Item 3, because L is a substitution instance of L. Further, recall that the
implication of A is dened via the implication of A as x y =
(x y) . Calculate
(x y) (y x) = ((x y) ) ((y x) ) =
((x y) (y x)) = 1 = .
Thus, A is a negative l-algebra. To check the last of the conditions listed

in this item take arbitrary elements y1 , y2 A and represent them in the
form (, y1 ) and (, y2 ). We have
(1, ) = ((, y1 ) (, y2 )) ((, y2 ) (, y1 )) =
(f (y1 y2 ) f (y2 y1 ), (y1 y2 ) (y2 y1 )),

in particular, f (y1 y2 ) f (y2 y1 ) = 1.
Prove the inverse implication. Let A and A be l-algebras, and let
fA (A ) R(A ) and for all y1 , y2 A , we have fA (y1 y2 ) fA (y2
y1 ) = 1. Take (x1 , y1 ), (x2 , y2 ) A and using the formula for implication
calculate
((x1 , y2 ) (x2 , y2 )) ((x2 , y2 ) (x1 , y1 )) = (h, (y1 y2 ) (y2 y1 )).
The second component of the last pair equals = 1A , because A is

a negative l-algebra, whereas the rst component has the following form:
h = ((x1 x2 ) (x2 x1 )) (f (y1 y2 ) (x2 x1 ))
((x1 x2 ) f (y2 y1 )) (f (y1 y2 ) f (y2 y1 )).

From our assumptions we immediately infer that rst and last conjunctive
terms of the last expression are equal to the unit element. In this way, we
obtain that the satisability of L on A is equivalent to the condition:
for all (x1 , y1 ), (x2 , y2 ) A, (x1 x2 ) f (y2 y1 ) = 1A .
Taking into account the facts that the implication is descending in the
rst argument and ascending in the second and that x f (y) for all (x, y)
A, we obtain the chain of inequalities (x1 x2 ) f (y2 y1 ) (x1
) f ( y1 ) (f (y1 ) ) f (y1 ) = f (y1 ) f (y1 ) = 1. The latter
equality holds due to the condition that every element of the form f (y) is
regular.
Items 510 can be checked via direct computation.
2
Corollary 5.3.2 1. Ljk = Lik Ln.
2. Ljx = Lk Ln.
3. For all L1 Int and L2 Neg, we have (L1 L2 )p , (L1 L2 )l
Spec(L1 , L2 ) and the following equality holds
(L1 L2 )p + (L1 L2 )l = L1 L2 .
In particular, Le = Lg + Ljl .
4. For every L1 Int and L2 Neg such that L1 = Lk, the logics
(L1 L2 )p and (L1 L2 )l are dierent from the endpoints of the
interval Spec(L1 , L2 ). At the same time, if L1 = Lk, we have (Lk
L2 )p = Lk L2 and (Lk L2 )l = Lk L2 .
5. LjlN = Lil Ln.
6. LjlQ
1 = Li Lnl.
7. LjlQ Q Q
2 Spec(Li, Lnl), Ljl2 = Li Lnl, Ljl2 = Li Lnl.
8. LjeQ
1 = Li Lmn.
9. LjeQ Q Q
2 Spec(Li, Lmn), Lje2 = Li Lmn, Lje2 = Li Lmn.
10. The logic Ljl is a proper extension of (Lil Lnl)l = Ljl lN lQ

1.
Proof. Items 1,2,5,6,8 easily follow from Propositions 4.2.3 and 4.2.7 and
suitable items of the last proposition.
3. By Item 3 of Proposition 5.3.1 all elements of the form ( a) are
regular in models of the logic (L1 L2 )l . On the other hand, in models of the
logic (L1 L2 )p all elements of this form are dense, as follows from Item 5
of Proposition 5.3.1. Thus, in models of the least upper bound of logics (L1
L2 )p and (L1 L2 )l elements of the form ( a) are regular and dense
simultaneously, i.e., they are all equal to the unit element. Consequently,
models of the considered least upper bound are exactly direct products of
the form B C, where B |= L1 and C |= L2 , whence we immediately obtain
the desired equality by Proposition 4.1.4 and Corollary 4.1.5.
4. The assertion of this item is true due to the fact that for any Heyting
algebra A the following three conditions are equivalent: A is a Boolean
algebra; the unit element is the only dense element of A; all elements of A
are regular.
7. By Item 8 of Proposition 5.3.1 the logic LjlQ 2 belongs to Spec(Li, Lnl).
Consider a model A for the free combination Li Lnl structured as follows.
An upper algebra A is arbitrary; a lower algebra A is a 4-element negative
Peirce algebra with universe {, a, b, 0}, where 0 a , 0 b , and

elements a and b are incomparable; f () = 1, f (x) = for x = . Calculate
f (a b) f (b a) = f (b) f (a) = = ,
which proves that LjlQ 2 diers from Li Lnl.

We now point out a model for LjlQ 2 dierent from the direct product
of Heyting and negative algebras. This will prove that LjlQ 2 is not equal to
the intersection of logics Li and Lnl. Let B and C be Heyting and negative
l-algebras respectively, which are isomorphic as implicative lattices, and let
f : C B be an arbitrary lattice isomorphism. It is not hard to check that
B f C is the desired model of LjlQ 2.
Q
9. The fact that Lje2 belongs to the interval Spec(Li, Lmn) follows
from Item 10 of Proposition 5.3.1. Examples of j-algebras showing that
LjeQ2 diers from the endpoints of the indicated interval can be constructed
in a way similar to that of Item 7.
10. This item can also be proved in a way similar to Item 7. As a coun-
terexample showing that the indicated extension is proper we may take the
j-algebra A from Item 7 with the additional restriction that A is a Heyting
l-algebra.
2
Now we have enough information about j-algebras modelling Segerbergs
axioms and we can come back to the analysis of Figure 5.2. We denote
by N eg the line passing trough the logics Ln and Lmn and by Int the
line passing through the logics Li and Lk. Recall that logics lying on the
line Int (N eg) form the intersection of the class D of logics presented in
Figure 5.2 with the class Int (respectively, with the class Neg), D Int = Int
and D Neg = N eg. These lines play a part of the coordinates for the three-
dimensional part of Figure 5.2, which we denote by P ar, P ar = D Par. For
any logic L P ar we can naturally dene its projections I(L) and N (L) to
the lines Int and N eg respectively. For example,
I(Lj) = Li, N (Lj) = Ln, I(Ljl) = Lil, N (Ljl) = Lnl,
I(Ljx) = Lk, N (Ljx) = Ln.
Using Proposition 5.3.1 and Corollary 5.3.2 we can easily check that for all
logics L P ar the equalities
I(L) = Lint and N (L) = Lneg

take place. Thus, for any line L on the diagram, which is parallel to the
line (Lj, Le ), the logics of this line have xed intuitionistic and negative
counterparts, say L1 and L2 . And so we have L = D Spec(L1 , L2 ).
We stated in this way that the three dimensions of the part P ar of
Figure 5.2 exactly correspond to the three parameters determining a posi-
tion of a logic in the class Par. One coordinate of a logic L is its intuitionis-
tic counterpart Lint Int, the second coordinate is its negative counterpart
Lneg Neg, and the third coordinate corresponds to a position of L in-
side the interval Spec(Lint , Lneg ), which is determined in turn by the class
of admissible semilattice homomorphisms from models of Lneg to models
of Lint .
At this point we note one obvious defect of Figure 5.2. Let us consider the
planes in the part P ar of the gure parallel to the plane with points Lj, Ljk,
and LjlQ 1 . There are three such planes. We denote by Pj the plane containing
the point Lj, by Pl the plane containing the point Ljl, and, nally, by Pe
the plane containing the point Le. If we follow the geometrical analogues
sketched above, we would expect that all logics belonging to one of the planes
Pj, Pl, Pe will dene the same class of admissible homomorphisms. But this
holds only for the plane Pe. For any logic L Pe we have ( p) L,
and so L = Lint Lneg is the greatest point of the interval Spec(Lint , Lneg ),
which is determined by the class of homomorphisms identically equal to the
unit element.
Let us consider the plane Pj. Elements of this plane are the least points
in the sets P ar Spec(L1 , L2 ), where L1 {Li, Lik, Lil} and L2 {Ln,
Lln, Lmn}. As we know from Proposition 4.2.5 the least point of Spec(L1 , L2 )
is the free combination L1 L2 of logics L1 and L2 . Moreover, for free combi-
nations all semilattice homomorphisms from models of negative counterpart
to models of intuitionistic counterpart are admissible. However, only three
points of Pj, namely, the logics Lj, Ljk, and LjlN are free combinations of
their intuitionistic and negative counterparts (see Items 1 and 5 of Corol-
lary 5.3.2). Logics LjlQ Q
2 and Lje2 are proper extensions of free combinations
LiLnl and LiLmn respectively, as it follows from Items 7 and 9 of Corol-
lary 5.3.2. Regarding the remaining four logics in Pj, we can easily modify
the proofs of Items 7 and 9 of Corollary 5.3.2 to show that the restrictions,
which axioms LQ Q
2 and E2 impose on the class of admissible semilattice ho-
momorphisms remain non-trivial, even if the intuitionistic counterpart of a
logic satises axioms K or LN (see also Propositions 5.3.4 and 5.3.5 below).
In case of the plane Pl we have a similar situation. Only the logics in
the leftmost vertical line have the class of admissible semilattice homomor-
phisms with a range contained in the set of regular elements of an upper
algebra (see Item 3 of Proposition 5.3.1). The other logics are characterized
by narrower classes of admissible homomorphisms (see Propositions 5.3.4
and 5.3.5).
The indicated defect can easily be overcome if we replace the axioms
LQ Q Q Q
2 and E2 by L1 and E1 respectively. As follows from Items 7 and 9 of
Proposition 5.3.1, these axioms do not impose any restrictions on the class
of admissible homomorphisms and restrict only the class of lower algebras.
These axioms can thus be considered as an adequate relativization of the
axioms L and E to the negative counterpart of a logic. After the above-
mentioned replacement and deleting axiom L, we obtain a diagram of logics
having exactly the same conguration as that of Figure 5.2.
As we have seen above, the axioms LQ Q
2 and E2 impose restrictions on the
classes of lower algebras of their models and simultaneously on the classes
of admissible homomorphisms from the lower algebras of their models to
the upper ones. We can separate these restrictions. As follows from Propo-
sition 5.3.1 axioms LQ Q
1 and E1 restrict the classes of lower algebras in the
same way as axioms LQ Q
2 and E2 respectively, and have no inuence on the
classes of admissible homomorphisms. On the other hand, as follows from
the next proposition, the axioms
F1 . ( (p q)) ( (q p))(= LQ
2)
F2 . ( p) ( (p q))(= EQ
2)
will restrict the classes of admissible homomorphisms in the same way as

was done by axioms LQ Q
2 and E2 respectively, and will not change the classes
of lower algebras.
Proposition 5.3.3 Let A be an arbitrary j-algebra. The following equiva-

lences hold.
1. A |= Ljf1 if and only if we have fA (y1 y2 ) fA (y2 y1 ) = 1 for

all y1 , y2 A .
2. A |= Ljf2 if and only if we have fA (y1 ) fA (y1 y2 ) = 1 for all
y1 , y2 A .
This statement can be proved via a direct computation. It is clear that

LjlQ
2 = LjlQ Q Q N Q
1 f1 , Lje2 = Lje1 f2 , and Ljl = Ljl l l1 f1 .
Let us consider the class D1 consisting of logics which can be obtained
by adjoining to Lj some subset of the following set of axioms
{I, Q, K, X, L, E, L , E , P , LN , LQ Q
1 , E1 , F1 , F2 }.
r L e = L1 L2
C@
C @
C @
C @
C @
C @
C @
r L gf2 CX rL l f2 @
r A X XrXX @

\
r
L g L gf\1 A L l f1 XXX
@rL l
@ \ A
@ \ Ar
@ \ L f2
@ \
@ \ r
@ L f1
@
@
@
@r
L
Figure 5.3
In this way, we take into account all properties involved in Segerbergs ax-
ioms. Obviously, D D1 . At the same time, D satises the condition that
for any L1 Int D and L2 Neg D the intersection Spec(L1 , L2 ) D is
linearly ordered. In case of D1 , this condition fails as we can see from the
propositions below.
Proposition 5.3.4 Let L1 {Li, Lik, Lil}, L2 {Ln, Lnl}, and let L :=
L1 L2 . The set of logics Spec(L1 , L2 ) D1 forms an upper semilattice,
shown on Figure 5.3.
In the course of proving this and subsequent propositions, we will con-

struct various j-algebras to check the interrelations between dierent logics.
The following Heyting and negative algebras will play the part of breaks in
our constructions: 2 and 2 are two-element Heyting and negative algebras;
3H and 3N are three-element Heyting and negative algebras, the elements of
which are linearly ordered; nally, 4H and 4N are four-element Heyting and
negative algebras respectively, whose implicative lattices are Peirce algebras.
For any Heyting algebra B, negative algebra C, and for any j-algebra
constructed from them B f C, we will identify an element b of B (c of C)
with the corresponding element (b, ) of the upper algebra (B f C) ((, c)
of the lower algebra (B f C) ).
r r r1 r
1 @ @
@ @
r r r r ar @rb cr @rd
1 a c @ @
@ @
r r1 r r1 @r @r 1
2 2 3H 3N 4H 4N
Figure 5.4
Proof (of Proposition 5.3.4.) First of all, we note that due to our as-
sumption L1 = Lk. This fact together with Items 3 and 4 of Corollary 5.3.2,
implies that logics L g and L l are dierent from the endpoints of the inter-
val Spec(L1 , L2 ) and the least upper bound of these logics coincides with the
greatest point of the interval, L g + L l = L e , which means, in particular,
that L g and L l are incomparable.
Let us consider the logics L f1 and L f2 . Take an arbitrary model A
for L f2 . Due to Proposition 5.3.3 we have fA (y1 ) fA (y1 y2 ) = 1 for
all y1 , y2 A . Since y1 y2 y1 , we have fA (y1 ) fA (y2 y1 ),
and also fA (y2 y1 ) fA (y1 y2 ) = 1 for all y1 , y2 A . In view of
Proposition 5.3.3 the latter means that A is a model for L f1 , and we have
the inclusion L f1 L f2 .
Let us consider a j-algebra A1 := 3H f1 3N , where f1 : 3N 3H is
a uniquely dened implicative lattice isomorphism (see Figure 5.5, in which
the structures of algebras constructed in this and the next proposition are
presented). For any y1 , y2 (A1 ) we have
f1 (y1 y2 ) f1 (y2 y1 ) = (f1 (y1 ) f1 (y2 )) (f1 (y2 ) f1 (y1 )) = 1
since 3H |= (p q) (q p). Thus, A1 |= L f1 . Now we take the elements
1, c 3N . It is clear that f1 (1) = and that f1 (c) = a (see Figure 5.4).
We have
f1 (c) f1 (c 1) = f1 (c) f1 (1) = a = a = 1.
This means that A1 is not a model for L f2 , and so the inclusion L f1 L f2
is proper.
Consider j-algebras A2 := 2 f2 4N , where f2 () = 1 and f2 (x) = for
x < , and A3 := 4H f3 4N , where f3 is an implicative lattice isomorphism
between 4N and 4H . As in Items 7 and 9 of Corollary 5.3.2 we can show
that A2 is a model of L , but is not a model of L f1 , respectively, that A3
is a model of L f2 , but is not a model of L e . We have thus proved that
L f1 and L f2 are dierent from the endpoints of the interval Spec(L1 , L2 ).
r r r1
1 1 @
@
ra r ar @rb
@ @ @ @
@ @ @ @
r @r cr @r d r @r @r
@ @ @ @
@ @ @ @
@r @r @r @r
c 1 c@ d
@
r1 A2 @r1
A1 A3
r1 r1 r1 r1
ra ra r ra
@ @ @
@ @ @
r r @r r @r rc r @r
@ @ @ @
@ @ @ @
r @r @ @r r r @r
@ c d 1 1 @ c
@ @
@r A5 A6 @r
1 r 1
1
A4 A7
r

r1
A8
Figure 5.5
Now we check that each of the logics L f1 and L f2 is incomparable with

either of the logics L g or L l . The j-algebras A1 and A3 are models of L f1
and L f2 respectively, but theirs are not models of L g, which implies that
L g is not contained in either of the logics L f1 or L f2 . Dene j-algebra
A4 as 3H f4 4N , where f4 () = 1 and f4 (x) = a for x < . A4 is a model
for L g, since the element a is dense in 3H , but it is not a model for L f1 ,
in which case it is not also a model for L f2 . Indeed, for c, d 4N , we have
f4 (c d) f4 (d c) = f4 (d) f4 (c) = a a = a. We have thus proved
that L f1 and L f2 are incomparable with L g.
The algebra A2 provides a counterexample, demonstrating that either of

the logics L f1 or L f2 is not contained in L l . To state that the inverse in-
clusions also fail we consider a j-algebra A5 := 3H f5 2 , where f5 (1) = a.
This is not a model for L l since a is not regular. At the same time, the
direct calculation shows that A5 |= L f2 . In this way, L l is not contained
in L f2 , moreover, it is not contained in L f1 .
The above facts on incomparability of logics imply, in particular, that
L gfi is a proper extension of L fi and of L g, i = 1, 2, and that L l fi is a
proper extension of L fi and of L l , i = 1, 2. So, it remains to verify that
the inclusions L gf1 L gf2 L e and L l f1 L l f2 L e are proper.
A j-algebra A6 := 3N 2 ( = 2 f6 3N , where f6 (x) = for x < )
shows that L l f1 L l f2 is a proper inclusion. It is a model for L lf1 since

for any y1 , y2 3N , either y1 y2 = or y2 y1 = . On the other

hand, f6 (c) f6 (c 1) = = . Note that A3 is a model for L l f2
dierent from the direct product of 4H and 4N . This proves that L e is a
proper extension of L l f2 .
Finally, consider the algebras A7 := 3H f7 3N , where f7 (x) = a for
x < , and A5 dened above. The rst of these algebras is a counterexample
showing that the inclusion L gf1 L gf2 is proper. The second algebra can
be used to check that L e is a proper extension of L gf2 .
2
Proposition 5.3.5 Let L1 {Li, Lik, Lil} and L := L1 Lmn. The set of
logics Spec(L1 , Lmn) D1 forms an upper semilattice shown on the semi-
lattice diagram in Figure 5.6.
rL e = L1 Lmn
@B
B@
B @
B @
B @
L gf1 = L gf2r L l f1@
BrP = L l f2
L P PP@

r L P@
Pr

L g@ L L l
@ L
@ Lr
@ L f1 = L f2
@
@
@
@r
L
Figure 5.6
Proof. As in the previous proposition, we have the assumption L1 = Lk,

which implies that logics L g and L l are dierent from the endpoints of
the interval Spec(L1 , L2 ), are incomparable and their upper bound coincides
with the greatest point of the interval.
We argue to prove the equality L f1 = L f2 . The inclusion L f1 L f2
was stated above. Let us check the inverse inclusion. Take an arbitrary model
A of L f1 , which means that fA (x y) fA (y x) = 1 for all x, y A .
By assumption A satises the Peirce law, and so for any x, y A , we
have x = (x y) x. On the other hand, in any j-algebra we have the
identity x y = x (x y). In this way, for any x, y A , we have
fA (x) fA (x y) = fA ((x y) x) fA (x (x y)) = 1,
which proves the desired equality.

The lower algebras of j-algebras A2 , A3 , A4 , and A5 dened in Propo-
sition 5.3.4 are models for Lmn and so these algebras can be used in the
following reasoning. In particular, j-algebras A2 and A3 can be used to check
that the logic L f1 lies inside the interval Spec(L1 , Lmn).
With the help of A4 and A8 := 2 2 we can show that the logics
L f1 and L g are incomparable. A4 is a model for L g, but not for L f1 .
Conversely, A8 is a model for L f1 , but not for L g.
In a similar way, one can use algebras A2 and A5 to check that logics
L f1 and L l are incomparable.

We are left to check that the following inclusions are proper: L f1 g

L e and L f1 l L e . The suitable counterexamples are provided by alge-
bras A5 and A3 , respectively.
2
We have not yet considered the case when the intuitionistic counterpart
coincides with the classical logic. It turns out that only in this case sets of
the form Spec(L1 , L2 ) D1 are linearly ordered with respect to inclusion.
Proposition 5.3.6 Let L2 {Ln, Lnl, Lmn}, and let L := Lk L2 . The

sets of logics Spec(Lk, L2 ) D1 have the structure presented in Figure 5.7.
Proof. First, consider the case L2 {Ln, Lnl}. Algebras A6 , A8 , and A2

can be used to verify that the inclusions L L f1 , L f1 L f2 , and
respectively L f2 L e are proper.
In case L2 = Lmn we may again use A8 and A2 to check the corre-
sponding relations between logics.
2
rL e = L g
rL f rL e = L g
2
rL f rL f = L f
1 1 2
rL = L l r
L = L l
L2 {Ln, Lnl} L2 = Lmn
Figure 5.7
5.4 Kripke Semantics for Paraconsistent

Extensions of Lj2
In this section we dene analogs of upper and lower algebras associated with
a given j-algebra for j-frames.
For an arbitrary j-frame W =
W, , Q we dene the following frames
W (+) :=
W \ Q, (W \ Q)2 , ,
W () :=
Q, Q2 , Q.
It is obvious that W (+) is a model for intuitionistic logic and W () is a
model for minimal negative logic.
Remark. For any j-frame W and any formula , the translation In()
is true on j-frame W () ,
W () |= In().
This fact can be checked via an easy induction on the structure of formulas.
Lemma 5.4.1 Let W be an arbitrary j-frame, v a valuation of W (+) , and

let v be a valuation of W such that for any propositional variable p we have
2
The content of this section was originally published in [76] (Elsevier, UK). Reprinted
here by permission of the publisher.
5.4 Kripke Semantics for Paraconsistent Extensions of Lj 79
v(p) = v (p) (W \Q). Then for any formula and for an arbitrary element
x W \ Q the following equivalence holds

W, v |=x In()
W (+) , v |=x .
Proof. Let :=
W, v and (+) :=
W (+) , v.
We argue by induction on the structure of formulas. The case of constant
is trivial. For an arbitrary propositional variable p and x W \ Q we have
|=x p if and only if either x v (p) or x Q. The second alternative
is impossible by assumption. Thus we have x v (p) and x W \ Q, i.e.,
x v(p). The latter is equivalent to (+) |=x p.
Now, we assume that for formulas and and for all x W \ Q the
equivalences
|=x In() (+) |=x and |=x In() (+) |=x
hold. Prove that the desired equivalence takes place for the implication
.
Let |=x In( )(= In() In()) for some x W \ Q. This
means that for all y W , the relations x y and |=y In() imply
|=y In(). In view of the assumed equivalences, we have
y W \ Q(x y ((+) |=y (+) |=y )),
and so (+) |=x .

Conversely, let (+) |=x for some x W \ Q. By assumption for
all y W \ Q such that x y, if |=y In(), then |=y In(). If y Q,
then |=y In() and |=y In(). Thus, for all y W such that x y,
we have
|=y |=y ,
which means that |=x In( ).
The cases of disjunction and conjunction are trivial.
2
The next proposition demonstrates that frames W (+) and W () can be
considered as analog of upper and lower algebras.
Proposition 5.4.2 For a j-frame W and a formula , the following equiv-

alences hold
W |= In() W (+) |= ,
W |= W () |= .
Proof. The rst equivalence immediately follows from the previous Lemma.
If W |= , then for any valuation v of W, is true in all abnormal
worlds of the model
W, v, which means by Lemma 2.3.1 that
W () , v Q |=
. Any valuation v of W () can be considered as a valuation of W, in which
case v = v Q . Thus, for all valuations v of W () , we have
W () , v |= , i.e.,
W () |= .
Conversely, the assumption W () |= implies that for any valuation v
of W,
W () , v Q |= . In view of Lemma 2.3.1, the latter means that for all
valuations v of W, is true at any abnormal world of
W, v, which implies,
in turn,
W, v |= .
2
The following fact immediately follows from the denition of intuition-
istic and negative counterparts and from the last proposition.
Corollary 5.4.3 Let L Jhn and W |= L. Then
W (+) |= Lint and W () |= Lneg .
2
For a class of j-frames K, we dene
K(+) := {W (+) | W K}, K() := {W () | W K}.
Proposition 5.4.4 Let K be a class of j-frames and let L = LK. Then

Lint = LK(+) and Lneg = LK() .
Proof. The inclusion Lint LK(+) follows from Corollary 5.4.3. We argue
for the inverse inclusion. Take a Lint , in which case In() L. Conse-
quently, there exist a frame W K, its valuation v, and an element x W
such that
W, v |=x In(). As was remarked above, a formula of the form
In() is true in any model at any abnormal element, therefore, x Q.
Whence, by Lemma 5.4.1 we have W (+) |= .
Now we turn to the second equality. Again, we have to prove only
the inclusion LK() Lneg since the inverse inclusion follows from Corol-
lary 5.4.3. Let Lneg , i.e., L . Consider a j-frame W K such
that W |= . From the last relation we obtain by Proposition 5.4.2
W () |= , i.e., LK() .
Chapter 6
Negatively Equivalent
Logics1
In the following, by negative formulas we mean formulas of the form . The

well-known Glivenko theorem implies, in particular, that in intuitionistic and
in classical logic the same negative formulas are provable. This means that
intuitionistic and classical logic, as well as all intermediate logic have, in a
sense, the same negation. Generalizing this relation between logics we dene
negatively equivalent logics as logics where the same negative formulas are
inferable from the same sets of hypotheses. From the constructive point of
view we need negation to refute formulas on the basis of one or another set of
hypotheses, therefore, negatively equivalent logics have essentially the same
negation. Unlike the class of intermediate logics, the relation of negative
equivalence is non-trivial on the class Jhn+ and in this chapter we obtain
several interesting results on the structure of negative equivalence classes.
Simultaneously, we prove the results on cardinality of intervals of the form
Spec(L1 , L2 ).
6.1 Denitions and Simple Properties

Let L1 and L2 be logics in Jhn+ . We say that L1 is negatively lesser than
L2 , and write L1 neg L2 , if for any set of formulas X and formula , the
following implication holds:
X L1 = X L2 .
1
Parts of this chapter were originally published in [75] (Springer, Netherlands).
Reprinted here by permission of the publisher.
81
82 6 Negatively Equivalent Logics
In other words, one logic is negatively lesser that the other if passing from
one to the other preserves the negative consequence relation, i.e., the conse-
quence relation of the form X , in which the conclusion is negative.
As we can see from the proposition below, the condition of preserving
the negative consequence relation can be replaced by that of preserving the
class of inconsistent sets of formulas. However, the equivalence proved in this
proposition is typical for the class Jhn+ , because in this class the negation
is dened via the constant absurdity, whereas the absurdity can be
dened as a negation of tautology.
Proposition 6.1.1 For any L1 , L2 Jhn+ , the following conditions are

equivalent.
1. L1 neg L2 .
2. For an arbitrary set of formulas X, if X L1 , then X L2 .
Proof. 1) 2) If X L1 , then X L1 for any formula . By the

assumption that L1 neg L2 , we have X L2 for any . Take an L2 -
tautology , then X L2 , , whence X L2 .
2) 1) Let X L1 . Then X {} L1 . In this case, we have
X {} L2 by assumption, consequently, X L2 by Deduction
Theorem, i.e., X L2 .
2
The denition of neg can also be re-worded as follows.
Proposition 6.1.2 For any L1 , L2 Jhn+ , the relation L1 neg L2 holds

if and only if for any formula , the following implication is true:
L1 = L2 .
Proof. Let L1 neg L2 . If L1 , then {} L1 . By the last proposition

we have {} L2 , from which we immediately obtain L2 .
Now we assume that the right-hand side of the desired equivalence holds.
Let X L1 . This means that for some formulas 1 , . . . , n X, L1
(1 . . .n ) . According to our assumption L2 (1 . . .n ). In
view of Lj p p we obtain L2 (1 . . .n ), which immediately
implies that X L2 . Again, Proposition 6.1.1 allows one to conclude that
L1 is negatively lesser than L2 .
2
If one of the logics is nitely axiomatizable relative to the other, the last
statement can be simplied as follows.
6.1 Denitions and Simple Properties 83
Proposition 6.1.3 Let L1 , L2 Jhn+ and L2 = L1 + {1 , . . . , n }. Then

L2 neg L1 if and only if 1 , . . . , n L1 .
Proof. We consider only the non-trivial implication. Let 1 , . . . , n

L1 . Take an arbitrary set of formulas X with X L2 , then X {1 , . . . , k }
is inconsistent in L1 , or equivalently,
X L1 (1 . . . k ),
where 1 , . . . , k are substitution instances of formulas from the list 1 , . . .,

n . We have L1 1 , . . . , k by assumption. Consider an arbitrary
model A |= L1 and an A-valuation v. The elements v(1 ), . . . , v(k ) are
dense in A . Consequently, the element v(1 . . . k ) is also dense, in
particular, v((1 . . . k )) = A .
Let formulas 1 , . . . , m X be such that
L1 (1 . . . m ) (1 . . . k ).
In view of the above considerations, for any model A |= L1 and any A-

valuation v, we have v(1 . . . m ) A . Consequently, we have L1
(1 . . . m ) , which means that X is inconsistent in L1 .
2
Dene the relation neg as an intersection of neg and its inverse relation:
neg :=neg (neg )1 .
One can easily prove
Lemma 6.1.4 1. The relation neg is a preordering.

2. The relation neg is an equivalence.
2
In view of this lemma, logics L1 , L2 Jhn+ with L1 neg L2 will be

called negatively equivalent. From Propositions 6.1.1 and 6.1.2 we immedi-
ately obtain
Corollary 6.1.5 For any L1 , L2 Jhn+ , the following conditions are equiv-
alent.
1. L1 neg L2 .
2. An arbitrary set of formulas X is inconsistent in L1 if and only if it
is inconsistent in L2 .
3. For any formula , the following implications hold:

(L1 = L2 ) and (L2 = L1 ).
Remark. Notice that the inclusion L1 L2 implies L1 neg L2 . There-

fore, the logics L1 and L2 from Proposition 6.1.3 will be negatively equivalent.
Remark. Any two negative logics are negatively equivalent due to the
following fact. In an arbitrary negative logic any set of formulas in inconsis-
tent since the absurdity belongs to the set of logical tautologies. Any two
intermediate logics are also negatively equivalent, which easily follows from
Proposition 6.1.3 and Glivenkos theorem. It is well known that negation in
intuitionistic logic is not constructive, from the deducibility Li ( )
does not follow, in a general case, that either of the formulas or is
provable in Li. We have just noted that negation in an arbitrary intermedi-
ate logic is close in some sense (negatively equivalent) to classical negation.
This fact can be considered as a generalization of Glivenkos theorem and
also emphasizes the non-constructive character of negation in intermediate
logics.
6.2 Logics Negatively Equivalent

to Intermediate Ones
In this section we consider the question: which logics from the class Jhn are
negatively equivalent to intermediate ones? More exactly, let some logics
L1 Int and L2 Neg be xed. Which logics L Par having L1 and
L2 as intuitionistic and negative counterparts respectively, are negatively
equivalent to intermediate logic L1 and so to an arbitrary intermediate logic?
In other words, to which extent can one weaken the law ex contradictione
quodlibet while preserving the negative equivalence?
Proposition 6.2.1 Let L1 Int, L2 Neg, and L Spec(L1 , L2 ). The
equivalence L neg L1 holds if and only if (L1 L2 )p L.
Proof. Recall that L1 = L + { p}. According to Proposition 6.1.3
L neg L1 whenever L ( p). By denition (L1 L2 )p = L1
L2 + {( p)}.
2

Call G(L1 , L2 ) := (L1 L2 )p the relativized Glivenkos logic wrt L1
and L2 .
From the last fact we easily infer the following strengthening of Gener-
alized Glivenkos Theorem (Theorem 5.1.2).
6.2 Logics Negatively Equivalent to Intermediate Ones 85
Corollary 6.2.2 Glivenkos logic Lg is the least logic in Jhn, which is neg-
atively equivalent to Lk.
2
As we can see from Item 4 of Corollary 5.3.2, the interval [LkL2 , LkL2 ]
contains a unique paraconsistent logic negatively equivalent to Lk, namely
Lk L2 . Note that this logic is axiomatized modulo the least logic Lk L2
of the interval Spec(Lk, L2 ) via the axiom ( p) having essentially a
non-constructive character. At the same time, if L1 = Lk, there is a proper
subinterval [G(L1 , L2 ), L1 L2 ] consisting of logics negatively equivalent to
intermediate logics.
It turns out that the disjunction property can be transferred from an
intuitionistic counterpart to the relativized Glivenkos logic. This fact was
established by M. Stukacheva [104]. Recall that a logic L has the disjunction
property if L implies L or L.
Let L Jhn. By induction on the length of formula we dene an
expression |L (Kleenes slash, see [13]) as follows (further on, instead of
|L and L we write L ):
|L := L , where is an atomic formula;

|L := |L and |L ;
|L := L or L ;
|L := (L |L ).
Proposition 6.2.3 [104] Let L1 Int, L2 Neg, and L1 has the disjunc-
tion property. If L1 L2 , then |G(L1 ,L2 ) .
Proof. Let G(L1 ,L2 ) . By induction on the length of proof, we show that
|G(L1 ,L2 ) . In the proof we omit the lower index G(L1 , L2 ).
Prove that this statement holds for axioms of G(L1 , L2 ).
a) The case of Lj-axioms can be easily veried;
b) For L1 L2 -axioms of the form , where L2 , the conclusion

is obvious since G(L1 ,L2 ) ;
c) By induction of the structure of In(), L1 , prove that In()

implies |In().
The basis is obvious. Indeed, since L1 is non-trivial, we have L1 p,
i.e., G(L1 ,L2 ) In(p);
Let In() and In() be such that In() |In() and In()
|In().
If In( ), then In() and In(). By induction hypothesis
we have |In() and |In(), which means by denition |In() In().
Recall that In() In() = In( ).
If In() In(), then In() or In(), since L1 satises the
disjunction property. By induction hypothesis In() or In(),
i.e., |In() In().
Assume In() In() and In(), then In() and |In() by
induction hypothesis.
d) It remains to prove | ( p). By denition we have
| ( p) (( ( p) and | ( p) ) | ).
Since ( p) and , we have ( p) , which means

that the right-hand side implication is true.
Finally, let is obtained by modus ponens from G(L1 , L2 ) and

G(L1 , L2 ). We have by induction hypothesis | and | .
Consequently, implies | , hence, | .
2
Proposition 6.2.4 Let L1 Int, L2 Neg, and L1 has the disjunction

property. Then G(L1 , L2 ) has the disjunction property.
Proof. Let G(L1 , L2 ) . According to Proposition 6.2.3 we have

|G(L1 ,L2 ) , consequently, G(L1 ,L2 ) or G(L1 ,L2 ) .
2
This fact shows that we can resign the law ex contradictione quodlibet
preserving not only the class of inconsistent sets of formulas, but also con-
structive properties of intuitionistic logic. But the disjunction property does
not hold in all relativized Glivenkos logics. In particular, if L1 = Lk, we
have G(Lk, L2 ) = Lk L2 ( p) (Item 4 of Corollary 5.3.2).
Moreover, if L1 does not possess the disjunction property, then G(L1 , L2 )
also does not. Indeed, let be a corresponding counterexample, i.e.,
L1 , but neither nor are provable in L1 . In this case, the for-
mula In( )(= In() In()) will refute the disjunction property for
G(L1 , L2 ): G(L1 , L2 ) In() In(), but neither In() not In() are
provable in G(L1 , L2 ).
6.2 Logics Negatively Equivalent to Intermediate Ones 87
However, one can point out an interesting weak analog of the disjunc-
tion property, which holds in all relativized Glivenkos logics G(L1 , L2 ) with
L1 = Lk. We try to nd a property that holds in all relativized Glivenkos
logics, independently of constructive properties of intuitionistic counter-
parts. Therefore, it should be a property that is trivially satised in all
intermediate logics, but becomes non-trivial in paraconsistent extensions of
Lj. The property of a logic to be closed under the rule

can serve as an example of such property. It can be considered as a weak ana-
log of the disjunction property, because as well as in case of the disjunction
property we conclude from a deducibility of disjunction to a deducibility of
disjunction term.
Proposition 6.2.5 Let Lk = L1 Int, L2 Neg, and let be an arbitrary

formula. If G(L1 , L2 ) , then G(L1 , L2 ) .
Proof. Let = (p1 , . . . , pn ). Assume that G(L1 , L2 ) , but is not

provable in G(L1 , L2 ). This implies, in particular, that is not provable in
L2 . Indeed, if L2 , then G(L1 , L2 ) and one can easily infer
G(L1 , L2 ) . Thus, there exists a negative algebra B being a model for
L2 , B |= L2 , and elements b1 , . . . , bn B such that (b1 , . . . , bn ) = .
By assumption Lk = L1 , hence, there exists a Heyting algebra A with
A |= L1 and a non-trivial lter of dense elements, Fd (A) = {1}. Take an
element a Fd (A), a = 1, and consider a j-algebra A f B, where a semi-
lattice homomorphism f is dened as follows: f () = 1 and f (x) = a for
x = . In this case, for any pair (x, y) A f B, we have x a when y = .
Moreover, f = {a, 1} Fd (A), which means that A f B is a model for
G(L1 , L2 ) (see Proposition 5.3.1).
Compute the value of on the elements (0, b1 ), . . . , (0, bn ) A f B.
Taking into account that the mapping (x, y) y denes an epimorphism
of j-algebras A f B B we have the equality ((0, b1 ), . . . , (0, bn )) =
(x, (b1 , . . . , bn )), where x a in view of (b1 , . . . , bn ) = . Thus, we have
((0, b1 ), . . . , (0, bn )) (, ) = (x, ) = (1, ),
which contradicts our assumption that G(L1 , L2 ) .

2
Remark. It is interesting that in the class of extensions of minimal logic
the inference rule
is equivalent to disjunctive syllogism in the following
sense. Let L Jhn and let Ld be a deductive system with the set of axioms
L and the only deductive rule modus ponens. Adding to Ld either of the
rules
,
or

results with the deductive system having exactly the same consequence
relation.
6.3 Abstract Classes of Negative Equivalence

For an arbitrary logic L Jhn+ , we dene
(L) := { | L}.
We now observe that the set (L) is itself a logic, possibly a trivial one,
and point out some simple properties of the operator : Jhn+ Jhn+ .
Proposition 6.3.1 For an arbitrary L Jhn+ , the following facts take

place.
1. L (L).
2. If L1 Jhn+ and L L1 , then (L) (L1 ).
3. (L) Jhn+ .
4. (L) = (L).
5. (L) = F if and only if L Neg {F}.
Proof. 1. This is true because Lj.

2. This item trivially follows from the denition.
3. Let formulas and belong to (L). Consider a model A |=
L and take an arbitrary A-valuation v. By denition of (L) we have
, ( ) L, which means that the values of formulas
and ( ) are dense, v( ), v(( ) ) Fd (A ). Calculate
v( ) v(( ) ) = v(( ( )) )
= v(( ) ) v( ) (A).
Thus, A |= for an arbitrary model A for L, i.e., L , whence

(L). In this way, the set (L) is closed under modus ponens. The
6.3 Abstract Classes of Negative Equivalence 89
fact that it is closed under the substitution rule follows directly from the
denition. We have thus proved that (L) is a logic, the fact that it extends
Lj follows from Item 1.
4. First, we note that the object (L) is well dened in view of the
previous item. The inclusion (L) (L) follows from Item 1.
Take a formula (L), in this case (L) and L.
The last formula is equivalent in Lj to , and so (L), which proves
the inverse inclusion.
5. If L Neg {F}, then (L) = F, because an arbitrary negative
formula belongs to L in this case. Assume L Jhn \ Neg. Then L Lk and
by Item 2 (L) (Lk) = Lk. The last equality is due to the fact that a
formula and its double negation are equivalent in Lk.
2
The operator is closely related to the negative equivalence relation, as
we can see from the following
Proposition 6.3.2 1. For any L Jhn+ , we have L neg (L).
2. For any L1 , L2 Jhn+ , the following equivalence holds
L1 neg L2 (L1 ) = (L2 ).
Proof. 1. It follows from Item 3 of Corollary 6.1.5.

2. Let L1 neg L2 . By denition (L1 ) if and only if L1 . In
virtue of the negative equivalence of L1 and L2 , the last fact is equivalent to
L2 , which is equivalent, in turn, to (L2 ). We have thus proved
that (L1 ) = (L2 ).
To prove the inverse implication assume (L1 ) = (L2 ). If L1 , then
also (L1 ), whence, by assumption (L2 ), and so L2 . In
the same way, L2 implies L1 . Applying Item 3 of Corollary 6.1.5
we conclude that L1 and L2 are negatively equivalent.
2
+
For a logic L Jhn , we denote by [L]neg its abstract class with respect
to negative equivalence,
[L]neg := {L1 Jhn+ | L1 neg L}.
It turns out that each of such abstract classes forms an interval in the
lattice Jhn+ , moreover the greatest point of the interval [L]neg can be cal-
culated by .
Proposition 6.3.3 For any L Jhn \ Neg,
[L]neg = [L , (L)],
where L Lg.
Proof. First we state that the set [L]neg is convex. Letting L1 , L2 [L]neg we
check that the interval [L1 , L2 ] is contained in [L]neg . Take an arbitrary L
[L1 , L2 ], we then have L1 neg L neg L2 . Taking into account L1 neg L2
we immediately obtain L [L]neg .
The logic (L) is the greatest point of [L]neg . Indeed, if L neg L and
L , then L, i.e., (L), and we have the inclusion L (L).
To state that [L]neg has the least point it is enough to observe that
the intersection of an arbitrary family of logics from the class [L]neg again
belongs to this class. One can give a more explicit presentation of the least
logic from [L]neg . Put
L := Lj + { | L}.
Due to Proposition 6.1.2 every logic negatively equivalent to L must contain

L . On the other hand, the logic L itself belongs to [L]neg . Indeed, the
relation L neg L follows from an obvious inclusion L L, the inverse
relation L neg L follows from Proposition 6.1.2.
According to Corollary 6.2.2, the logic Lg is the least logic in [Lk]neg ,
and so it has a presentation
Lg = Lj + { | Lk}.
Using this fact and the inclusion L Lk we immediately obtain L Lg.

2
Logics of the form (L) admit another interesting characterization in-
dependent of the operator and the notion of negative equivalence.
We dene -logics as xed-points of the operator , i.e., we say that a
logic L Jhn is a -logic if (L) = L. In view of Item 4 of Proposition 6.3.1,
any logic of the form (L) is a -logic. The -logics have a description, in
which again arises the rule .
Proposition 6.3.4 A logic L Jhn is a -logic if and only if Lint = Lk

and L is closed under the rule

.

6.4 The Structure of Jhn+ up to Negative Equivalence 91
Proof. Recall that Lj (p p). This means that for every L Jhn,
the formula p p belongs to (L). It was proved in Item 2 of Corollary
5.3.2 that Lj + {p p} = Lk Ln, and so any logic of the form (L)
contains Lk Ln. An inclusion of logics implies the inclusion of respective
counterparts (see Proposition 4.2.2), therefore, Lk ((L))int . We have
thus proved that intuitionistic counterparts of -logics are classical.
We now observe that every logic (L) is closed under the rule .
This fact easily follows from the idempotentness of . If (L), then
(L) = (L). According to the lemma below, the double negation
is equivalent to in Lk Ln, and so in any -logic, which completes
the proof of the direct implication.
Lemma 6.3.5 p (p ) Lk Ln.
Proof. By denition of free combination we have
(p ) (p ) = In(p p) Lk Ln.
It remains to note that (p ) p Lj.

2
Prove the inverse implication. The condition Lint = Lk implies the in-
clusion Lk Ln L, and we apply Lemma 6.3.5 to conclude that L is closed
under the rule
. If (L), then by denition L, and applying
the above rule we obtain L.
2
6.4 The Structure of Jhn+ up to Negative

Equivalence
In this section, we give a characterization of the partial ordering

Jhn+ / neg , neg ,
where neg :=neg / neg . To obtain the main results we apply the technique
of Jankovs formulas suggested by V. A. Jankov [37, 38] and modied by
H. Ono [83] and A. Wronski [125, 126]. Usually, this technique is used for
constructing uncountable families of logics. We are interested rst of all for
Jankovs formulas themselves. In our considerations, they will have the form
of negative formulas, which allows one to prove that dierent logics are not
negatively equivalent. We recall basic elements of Jankovs method adopting
it for j-algebras.
A relation X |=A , where X is a set of formulas, a formula, and A a

j-algebra, means that for any A-valuation v, if v() = 1 for all X, then
v() = 1. If K is a class of j-algebras, then X |=K means that X |=A
for all A K. Finally, write X |=L instead of X |=M od(L) .
Let A =
A, , , , , 1 be a not more than countable and subdirectly
irreducible j-algebra. For each element a A, a = , we attach a unique
propositional variable pa . Further, for any a A, we attach a unique atomic
formula Za as follows

pa , if a =
Za :=
, if a = .
A diagram D(A) of A is the following set of formulas

D(A) := {Zab (Za Zb ) | a, b A}
{Zab (Za Zb ) | a, b A}
{Zab (Za Zb ) | a, b A}.
Let A be a nite subdirectly irreducible j-algebra. Then D(A) is a nite

set of formulas and we can dene a Jankov formula of A by

J(A) := ( D(A)) Z
A ,

where ( D(A)) is the conjunction of all formulas in D(A), and A is the
opremum of A. It is easy to see that J(A) LA. Moreover, the following
statement holds.
Lemma 6.4.1 Let A be a nite and subdirectly irreducible j-algebra. For

each j-algebra B, the following two conditions are equivalent.
1. J(A) LB.
2. A is embeddable into a quotient algebra of B.

Proof. 1 2. Assume B | = J(A). Let v be a B-valuation such that v( D(A))
v(Z
A ). Put a0 := v( D(A)) and consider the quotient B/
a0 . Dene
a mapping
h : A B/
a0 by the rule h(a) := v(Za )/
a0 . It follows from
v( D(A))
a0 that h is a homomorphism. Since a0 v(Z
A ), we have
h( A ) = 1, i.e., A Ker(h). This means that h is an embedding.
2 1. Let F be a lter on B and h : A B/F be an embedding.
Consider a B-valuation v such that v(Za )/F = h(a) for all a A. Homo-
morphism properties of h imply that for all D(A) we have v() F ,

and so v( D(A)) F . At the same time, h( A ) =1/F since h is an em-
bedding, which implies v(Z
A ) F . In this way, v( D(A)) v(Z
A ), i.e.,
B |= J(A).
2
In case A is not nite, we cannot, of course, dene a Jankov formula
of A. However, one can prove
Lemma 6.4.2 Let A be a countable and subdirectly irreducible j-algebra.

For each j-algebra B, the following conditions are equivalent.
1. D(A) |=B Z
A .
2. A is embeddable into B.
Proof. 1 2. Let v be a B-valuation such that v() = 1 for all D(A) and
v(Z
A ) = 1B . Consider a mapping h : A B given by the rule h(a) = v(Za ).
It follows easily from our assumption and the denition of D(A) that h is
a homomorphism. If h is not a monomorphism, then Ker(h) = {1A } and
A Ker(h), i.e., h( A ) = 1B . The latter conicts with the assumption that
v(Z
A ) = 1B .
2 1. Assume h embeds A into B. Consider a B-valuation such that
v(pa ) = h(a) for a = A . Naturally, v() = h(A ) = B . It is clear that
v() = 1 for all D(A) and v(Z
A ) = 1B , i.e., D(A) |=B Z
A .
2
A sequence {Li }i< of logics from Jhn is said to be strongly independent
if Li j
=i Lj for each i < , where j
=i Lj is the least logic containing all
Lj , j = i.
The following two facts are natural generalizations of Proposition 1.2
and Lemma 1.4 of [106] to the class of extensions of minimal logic.
Proposition 6.4.3 Let {Li }i< be a strongly independent sequence of logics

from Jhn. For any subsets I and J of , iI Li = iJ Li if and only if
I = J.
Proof. Prove the non-trivial implication. Assume that I = J and there is k

I \ J. Obviously, iJ Li i
=k Li . Since {Li }i< is strongly independent,
we have Lk i
=k Li , moreover, Lk iJ Li . Therefore, iI Li = iJ Li .
2
Proposition 6.4.4 Let {Ai }i< be a sequence of subdirectly irreducible

j-algebras satisfying the following conditions:
(1) each Ai is nite;
(2) for every i, j < , i = j implies that Ai cannot be embedded into any
quotient algebra of Aj .
If a logic L is contained in every LAi (i < ), the sequence {Li }i< of logics
dened by Li = L + {J(Ai )} (i < ) is strongly independent.
Proof. It suces to show that J(Ai ) j
=i Li for all i. According to
Lemma 6.4.1 we have J(Ai ) LAj for j = i. Consequently, L {J(Aj ) |
j = i} LAi , and we conclude j
=i Lj LAi . In view of J(Ai ) LAi we
have J(Ai ) j
=i Li .
2
We now turn to studying the partial ordering
Jhn+ / neg , neg . First
we prove
Proposition 6.4.5 Let L Neg. For any L1 , L2 Spec(Lk, L), we have
the equivalence
L1 neg L2 L1 = L2 .
Proof. We need only prove that dierent logics from the class Spec(Lk, L)
are not negatively equivalent. Let logics L1 , L2 Spec(Lk, L) be such that
there is a formula L1 \ L2 . Since L2 , according to Proposition 2.2.6
it must be refuted at some subdirectly irreducible model of L2 .
Let A be a subdirectly irreducible j-algebra such that A |= L2 and
A |= . First of all we note that A is not negative. If A is negative, then A
is a model of (L2 )neg = L2 + {} = L. The logics L1 and L2 have a common
negative counterpart L and L1 L. In this way, we conclude that A |= L1 ,
which conicts with our assumption that A refutes .
We now consider the upper algebra A , which is non-trivial due to the
last remark. The opremum of A equals to the opremum of A . Indeed,
A = 1A , and so A A . Therefore, Heyting algebra A is a subdirectly
irreducible model of (L1 )int = Lk, whence, it is a Boolean algebra. It is well
known that only two-element Boolean algebra is subdirectly irreducible, and
so A = A . We have thus proved the following
Lemma 6.4.6 Let L Neg and L1 Spec(Lk, L). Every subdirectly ir-
reducible model A of L1 is either negative, in which case A |= L, or has
the form A
= B 2, where B is a negative algebra modelling L and 2 is a
two-element Boolean algebra.
Assume additionally that A is nite. Since A is not a model of L1 , it

is not embeddable into a quotient algebra of any j-algebra B modelling L1 .
According to Lemma 6.4.1, this means that for any model B of L1 , B |= J(A),
and so J(A) L1 . It is clear that J(A) L2 since J(A) LA.
As noted above, A = A , whence,

J(A) = ( D(A)) = ( D(A)).
This formula proves that logics L1 and L2 are not negatively equivalent.
It remains to consider the case when a j-algebra A refuting L1
is innite. We use Lemma 6.4.2 to conclude that for any j-algebra B with
B |= L1 , we have D(A) |=B , i.e., that D(A) |=L1 .
We may consider j-algebras as models of the rst-order predicate calcu-
lus of the language
, , , , 1 with identity = and classical negation
. Propositional variables will be treated as individual ones, propositional
formulas as terms, and valuations as denotations of individual variables. To
avoid confusion we will denote the rst order satisability relation by |=1 .
In this case, for any j-algebra B, B-valuation v, and formula , we have the
following equivalences:
v() = 1 B |=1 = 1[v],
B |= B |=1 ( = 1) ,
where () denotes the universal closure of a rst-order formula .
The relation D(A) |=L1 is equivalent to the statement that there is no
j-algebra B and B-valuation v such that B |= L1 , B = 1B , and v() = 1 for
all D(A). In view of the above equivalences, the latter means exactly
that the set of formulas
:= {( = 1) | L1 } { = 1 | D(A)} { ( = 1)}
is not satisable. Due to the Malcev local theorem for rst-order logic, there
exists a nite non-satisable subset X .
Further, we argue to state that ( = 1) X. Consider the quotient
algebra A/
. The set of formulas { = 1 | D(A)} is satisable in A, it
will be also satisable in the quotient algebra A/
. The algebra A/
is
negative, and so it is a model for L = (L2 )neg . Since L1 L, the sentences
from the set {( = 1) | L1 } will be true in A/
. We have thus
proved that the set of formulas \ {( = 1)} is satisable, from which we
conclude that ( = 1) X.
Denote by X the following subset of D(A),
X := { | D(A), = 1 X}
Coming back to identities of j-algebras, we can see that
X |=B
for any B with B |= L1 , i.e., X |=

L1 .
We claim that the formula ( X ) = ( X ), where X is the
conjunction of all formulas from X , is true in all subdirectly irreducible
models B of L1 , and so belongs to L1 .
Take an arbitrary subdirectly irreducible model B of L1 . If Bis negative,
then any negated formula is true on B, in particular, B |= ( X ) .
Assume that B is not negative and v is a B-valuation. Relation X |=L1
means that the set of formulas X is not satisable in any non-negative
model of L1 , whence v( X ) = 1. Taking into account that B is subdirectly
irreducible and B = B by Lemma 6.4.6 we obtain v( X ) B , con-
sequently, v(( X ) ) = 1. We have thus proved that ( X ) L1 .
Recall that X is a subset of D(A), the diagram of the non-negative model
A for L2 , therefore, it is satisable
in A and ( X ) does not belong to L2 .
In this way, the formula ( X ) proves that L1 and L2 are not negatively
equivalent.
2
Proposition 6.4.7 For any logics L Jhn \ Neg and L1 Neg such that
L1 Lneg , there exists a logic L Spec(Lk, L1 ) such that L neg L.
Proof. Consider the logic (L). By Proposition 6.3.2 (L) neg L. Due to
Proposition 6.3.4 ((L))int = Lk. At the same time, by Proposition 6.3.1
L (L) and consequently, Lneg (L)neg . Combining the last two facts
yields (L) Spec(Lk, L2 ) for some L2 Lneg L1 , in particular, Lk
L2 (L).
Dene a logic L by setting
L := Lk L1 + { | L}(= Lk L1 + { | (L)}).
It is clear that Lk L1 (L), whence, L (L), and so L neg L.

Further, let L. Then also L, and L by denition. We
now apply Proposition 6.1.2 to conclude L neg L . We have thus proved
that L and L are negatively equivalent and it remains to check that L
belongs to the desired interval Spec(Lk, L1 ).
Obviously, L1 Lneg . If this inclusion is proper, there is a negative

algebra A, which is a model of L1 but not of Lneg . It is clear that A models
Lk L1 . All negated formulas are true in A, therefore, A |= L . Moreover,
A is negative, whence, A |= Lneg , a contradiction.
2
+
Now the ordering
Jhn / neg , neg can be described as follows.
Proposition 6.4.8 1. The following isomorphism takes place

Jhn+ / neg , neg
=
Spec(Lk, Ln) {F}, .
2. For L1 Int and L2 Neg, we have

Spec(L1 , L2 )/ neg , neg
=
Spec(Lk, L2 ), .
Proof. 1. Applying Proposition 6.4.7 for the case L1 = Ln, we conclude that
for any L Jhn \ Neg, there is a logic in Spec(Lk, Ln) negatively equivalent
to L. Such logic is unique due to Proposition 6.4.5 and we denote it by h(L).
For every L Neg {F}, put h(L) = F. We have thus dened a mapping
h : Jhn+ Spec(Lk, Ln) {F}.
Obviously, h(L) = L for all L Spec(Lk, Ln), therefore, h is a map-

ping onto. Moreover, it satises the property that L neg h(L) for every
L Jhn+ .
Take arbitrary L1 , L2 Jhn \ Neg such that L1 neg L2 . In view of the
envisaged property of h we have h(L1 ) neg h(L2 ). If h(L1 ), then
h(L1 ) and by Proposition 6.1.2 h(L2 ). Due to the same proposition
we have h(L1 ) neg h(L1 ) h(L2 ). The inverse relation h(L1 ) h(L2 ) neg
h(L1 ) follows from the set-theoretical inclusion. Applying Proposition 6.4.5
to h(L1 ) neg h(L1 ) h(L2 ) we infer h(L1 ) h(L2 ). Note also that for
any L1 Jhn and L2 Neg {F}, L1 neg L2 and h(L1 ) h(L2 ) =
F. We have thus proved that h is an epimorphism of
Jhn+ , neg onto
Spec(Lk, Ln) {F}, . Passing to the quotient mapping h/ neg of this

epimorphism with respect to the negative equivalence we obtain the desired
isomorphism of
Jhn+ / neg , neg and
Spec(Lk, Ln) {F}, .
2. Arguing as in the previous item we construct a mapping
h : Spec(L1 , L2 ) Spec(Lk, L2 )
such that L neg h(L) for any L Spec(L1 , L2 ). We need only check
that this is a mapping onto. Take an arbitrary L Spec(Lk, L2 ) and put
r r r
...
1 1 1
r r r
@ @
r r @r r r @r
B1 2 @ @ @
@r r @r @
B2 2 @
@r
B3 2
Figure 6.1
L := L1 L2 + { | L}. As in Proposition 6.4.7 we prove that

L neg L. Now we conclude the proof as in the previous item.
2
In conclusion, we make some remarks on the cardinality of intervals
Spec(L1 , L2 ). It turns out that any interval of this form is innite, more-
over, innite up to negative equivalence, and in many cases such an inter-
val has the power of the continuum. We start with the simplest interval
Spec(Lk, Lmn).
Proposition 6.4.9 The interval Spec(Lk, Lmn) has the following struc-
ture:
Lk Lmn . . . Ln . . . L1 L0 ,
where L0 = Le = Lk Lmn; for n > 0, Ln = L(Bn 2) = Lk Lmn +
{J(Bn+1 2)}, where Bn is a negative Peirce algebra with n atoms (see
Figure 6.1); nally, Lk Lmn = L(B 2), where B is an arbitrary innite
negative Peirce algebra.
Proof. All logics from the interval Spec(Lk, Lmn) have the same negative
models, namely, the models of Lmn. Therefore, every logic L in Spec(Lk,
Lmn) is determined by the class of its non-negative nitely generated sub-
directly irreducible models and we denote this class by M od+
f si (L). Due to
Lemmas 6.4.6 and 6.4.10 every such model model has the form B 2, where
B is a nitely generated model of Lmn, i.e., a nitely generated negative
Peirce algebra.
Lemma 6.4.10 The algebra B 2 is nitely generated if and only if B is

nitely generated.
Proof. Let B := B 2 and := B2 . Note that |B | = |B| {1B } and

= 1B = B .
If X generates B , then X \ { } generates B. Indeed, if b B is a value

of some term dened on generators, then replacing by c c, where c
X \ { }, does not change the value of this term.
If X generates B, then X { } generates B . It is a consequence of the
following obvious relations. For any a, b B, we have a B b = a B b,
a B b = a B b, if a B b = 1B , then a B b = a B b.
2
The only dierence between Boolean and negative Peirce algebras is
in the interpretation of . In Boolean algebras, is interpreted as the
least element, whereas in negative Peirce algebras, it is interpreted as the
greatest element of an algebra. In this way, all nitely generated negative
Peirce algebras are nite. And we have a countable chain of dierent (up
to isomorphism) nitely generated negative Peirce algebras {Bn | n },
where Bn is a negative Peirce algebra with n-atoms, i.e., with n minimal
elements in the set Bn \ {f } for f denoting the least element of Bn . Clearly,
Bn is isomorphically embedded into Bm if and only if n m. Therefore, the
class M od+
f si (L) for L Spec(Lk, Lmn) is of the form
M = {Bn 2 | n < },
where 1 . It is not hard to check that each of these sets can be
realized as a set of non-negative nitely generated subdirectly irreducible
models of a suitable logic L from the interval Spec(Lk, Lmn). Indeed, the
class M1 = {B0 2} = {2} corresponds to the logic Le = Lk Lmn, which
has a unique non-negative subdirectly irreducible model, 2. For 1 < < ,
consider a logic
L := Lk Lmn + {J(B 2)}.
According to Lemma 6.4.1 the class M od+ f si (L ) coincides with the class of
algebras of the form Bn 2 such that the algebra B 2 is not embeddable
into a quotient algebra of Bn 2. Every proper quotient algebra of Bn 2 is
negative and non-negative algebra can not be embedded into it. The algebra
Bm 2 is embeddable into Bn 2 if and only if Bm is embeddable into Bn .
Combining all these facts we obtain
M od+
f si (L ) = M .
From the last equality and the fact that Lmn = L(Bn ) for any n we have
L = L(B 2).
The class Spec(Lk, Lmn) has the least element L = Lk Lmn and,
obviously, M od+f si (Lk Lmn) = M . If we take an arbitrary innite nega-
tive Peirce algebra B, any algebra of the form Bn 2 will be embeddable
into B 2. Therefore, L = L(B 2).
Classes of models M , 1 , form an ascending chain of type

+ 1 with respect to inclusion, which means that the corresponding logics
L , 1 , form a descending chain of type ( + 1) .
2
Further, we obtain a sucient condition guaranteeing that Spec(L1 , L2 )
is of the power of the continuum.
Proposition 6.4.11 Let L1 Int and L2 Neg. Assume that there exists
a family {Bi | i < } of nite negative algebras such that Bi |= L2 for all
i < , and Bi is not embeddable into Bj for i = j. Then
|Spec(L1 , L2 )| = 2 .
Proof. In view of Proposition 6.4.8, it is enough to consider the case L1 = Lk.

Consider the sequence {Ai }i< of j-algebras dened by Ai = Bi 2, i < .
It is a sequence of subdirectly irreducible j-algebras modelling Lk L2 . By
assumption if i = j, Bi is not embeddable into Bj , in which case also Ai is
not embeddable into Aj . Further, note that every proper quotient algebra
of Aj is negative. This means that the non-negative algebra Ai cannot be
embedded into a quotient algebra of Aj for i = j.
Dene a sequence of logics {Li }i< by Li = Lk L2 + {J(Ai )}. By
Proposition 6.4.4 this sequence is strongly independent and we obtain a
continuum of dierent logics of the form iI Li , I . Each of these logics
belongs to the interval Spec(Lk, L2 ), because all J(Ai ) are negative formulas
and belong to L2 .
2
Corollary 6.4.12 |Jhn/ neg | = |Spec(Lk, Ln)| = 2 .
Proof. It is not hard to check that the family {Bi | i } (see Figure 6.2)
of negative algebras satises the condition of the previous proposition.
2
r r r
r @r r @r r @r
r @r @r
@r @r
@r
@r @r
r @r @r
@r @r
@r @r
@r @r
r @r @r
...
B0 @r @r @r
B1 @r
B2
Figure 6.2
Chapter 7
Absurdity as Unary
Operator1
7.1 Introduction
This chapter nishes the rst part of the book devoted to the concept of
negation as reduction to absurdity. As was mentioned in Chapter 1, minimal
logic lies on the border line of paraconsistency. We have in Lj for arbitrary
formulas and ,
{, } .
This means that although inconsistent Lj-theories may be non-trivial, they
are trivial with respect to negation. Any negated formula is provable in any
inconsistent Lj-theory.
Due to this paraconsistent paradox of minimal logic it looks natural
to nish the investigation of the class of Lj-extensions with an attempt
to overcome this paradox. We try to do it by merging the class of Lj-
extensions into a more general class of paraconsistent logics and pointing
out some special property distinguishing extensions of minimal logic in the
latter class. We also suggest that the negation in logics from the above-
mentioned class should preserve the most essential property of intuitionistic
negation, namely, that the negation must be dened as reduction to absur-
dity. In intuitionistic logic, the negation is characterized by three important
features: 1) we assert if supposing leads to absurdity; 2) absurdity may
be explicated as a propositional constant ; 3) absurdity implies everything,
1
Parts of this chapter were originally published in [73] (Nicholas Copernicus University
Press, Poland) and in [81] (Rodopi, Netherlands). Reprinted here by permission of the
publishers.
101
102 7 Absurdity as Unary Operator
p, or p (p q). Note that Item 3 implies, in fact, Item 2. If any

contradiction implies everything, all contradictions are equivalent, and so
we may use a propositional constant do denote an arbitrary contradictory
or absurd statement. In minimal logic Lj, we omit Item 3, which allows
one to distinguish non-equivalent contradictions. In an extension L of Lj,
the contradiction is equivalent to if and only if is provable in
Lneg = L + {}, the negative counterpart of L (see Chapter 4), moreover, it
was stated that Lneg may be considered as the logic of contradictions of L.
At the same time, Item 2 is preserved in Lj, which results, in particular, in
the above-mentioned paraconsistent paradox. It looks natural to make the
next step and to resign not only Item 3, but also its consequence, Item 2.
This gives rise to a question: how should the absurdity be explicated in a
logical system in this case?
We suggest an answer, which is based on the interrelations between logic
of classical refutability Le and L ukasiewiczs modal logic L
. In Chapter 3 it
was proved that the simplest matrix for Le has four elements, and so Le can
be considered as a four-valued logic as well as L ukasiewiczs modal logic.
In Section 3.2, it was proved that in Le, one can dene one isomorph of
classical logic and two dierent isomorphs of maximal negative logic. These
isomorphs, on the one hand, correspond to translations of Lk and Lmn
into Le playing the key role in studying the structure of the class Jhn (see
Section 4.1). On the other hand, the isomorphs were induced by mappings,
which can be identied in a natural way with the modalities of L , which eas-
ily implies the fact that Le is denition equivalent to the positive fragment
of L and that Le extended by the classical negation is denition equiva-
lent to L . Similar interrelations between Le and L were stated previously
by J. Porte [84, 85]. For our goals the most essential is the fact that the
necessity operator of L may be identied with the contradiction operator
C() := of logic of classical refutability. This leads to the idea of
considering a contradiction operator C as a primary logical connective and
dening a negation as a reduction to this operator, := C(). In
this way, we merge the class Jhn into the class of so-called C-logics and show
that logics from Jhn will be distinguished by some property of C, which be-
comes paradoxical if C is considered as a modal operator. So, more natural
modal properties of the contradiction operator result in a more satisable
consequence relation from the point of view of paraconsistent logic. Further,
it turns out that the same negation may be dened via dierent operators.
There is a sense to distinguish, for example, between the contradiction op-
erator C that should satisfy the axiom C() and the more general
absurdity operator, for which a similar axiom does not obey.
7.1 Introduction 103
At the end of this chapter, we show that in such well-known paracon-

sistent logics as the logic CLuN by D. Batens [6, 7] and Settes maximal
paraconsistent logic P 1 [88, 102] the negation can be dened via a suitable
contradiction or absurdity operator. It is interesting that the negation of
CLuN may be dened via an absurdity operator about which nothing is
postulated. In this way, CLuN may be considered as a counterpart of Le,
which has the same positive fragment and where the negation is dened via
the constant , about which nothing is postulated.
Let L1 and L2 be propositional languages, L1 and L2 logics in languages
L1 and L2 respectively, and let : L1 L2 and : L2 L1 be translations
from the language L1 into L2 and vice versa, i.e., mappings from the set of
L1 -formulas to the set L2 -formulas and vice versa. We say that L1 is faithfully
embedded into L2 via if for any L1 -formula , we have the equivalence
L1 () L2 .
Further, the logic L1 is said to be denition equivalent to L2 via translations

and if L1 is faithfully embedded into L2 via , L2 is faithfully embedded
into L1 via and moreover, for any L1 -formula and L2 -formula we have
L1 () and L2 ().
A translation from L1 into L2 preserves propositional variables if

(p) = p for any propositional variable p and it preserves an n-ary con-
nective if L1 L2 and
((1 , . . . , n )) = (1 , . . . n )
for all L1 -formulas 1 , . . . , n .

Remark. If L1 is faithfully embedded into L2 via , preserves the
implication connective and logics L1 and L2 satisfy the Deduction Theorem,
then it is not hard to prove that for any set of L1 -formulas X and L1 -formula
, the equivalence holds:
X L1 (X) L2 (),
where (X) = {() | X}. In other words, is a strong translation.

Taking into account this remark and the above denitions, Proposi-
tion 2.1.1 can be re-worded as follows.
Proposition 7.1.1 Logics Lj and Lj are denition equivalent via trans-

lations and .
2
7.2 Le and L
ukasiewiczs Modal Logic
Recall that logic of classical refutability Le is characterized by the class
of Peirce-Johansson (or pj-)algebras. A pj-algebra is an implicative lattice
satisfying the Peirce law and with interpreted as an arbitrary element of its
universe. This characterization easily implies that any characteristic model
of Le must contain at least four elements: unit element 1, 0(= = 1),
some element 1 under 0, because Le is paraconsistent and nally, element
a = 0 1 incomparable with 0. It turns out that the four-element lattice
4 with universe {1, 0, a, 1} and interpreted as 0 is really a characteristic
model for Le (see Corollary 3.1.6).
L
ukasiewiczs modal logic L was introduced in [52] (see also [53]). The
intention of J. L
ukasiewicz was to construct a system of modal logic, which
is an extension of classical propositional logic with two interdenable modal
operators of necessity L and of possibility M , Lp M p and M p
Lp. These operators, in turn, should satisfy the conditions
Lp p, p M p,
and
p Lp, M p p.
He used the matrix approach to dene his system. Since the ordinary two-
element matrix 2 for classical logic does not allow one to dene modal
operators satisfying the above-mentioned conditions, L ukasiewicz took the
four-element matrix 2 2, which also denes classical logic.
Note that if we identify the elements of 2 2 and the elements of Le-
model 4 in the following way (1, 1) 1, (1, 0) a, (0, 1) 0, and (0, 0)
1, the binary operations of classical logic and of Le will coincide on 4 ,
only negations will act dierently on the set of truth values {1, a, 0, 1}.
q1 q1
@
6 @
I
6
@ @
@ @
a
q -
@q0 aq -q0
R
@
@ @
@ @
@ @
@?
q @q
1 1
It these diagrams the classical negation is on the left and the negation
of Le on the right.
7.2 Le and L
ukasiewiczs Modal Logic 105
Modalities L and M and dual modalities W and V are dened on 2 2

via the following truth-tables:
x Lx Mx Wx Vx
1 0 1 a 1
a 1 a a 1
0 0 1 1 0
1 1 a 1 0
It is easy to check that the modalities L and M are interdenable via clas-
sical negation and satisfy L
ukasiewiczs conditions. We dene L
ukasiewiczs
modal logic L in the language L+ + {L, } as the set of tautologies of the
following algebra
4L
:=
{1, a, 0, 1}, , , , , L, 1.
Denote by L + its positive fragment in the language L+ + {L}. Note that

the second modality M is denable through L in positive fragment L + as
follows
M p L(p p) p.
The pairs of modalities L, M and W , V are interdenable, e.g., as follows
W p p M ((p p)), Lp p V ((p p)).
If we use modalities W and V instead of L and M , respectively, we obtain

the logic with the same set of tautologies. Taking into account the above
identication of truth-values of logics L and Le we can easily see that the
actions of modalities L, M , and V on the set of truth-values coincide, re-
spectively with the mappings , , and , which are denable in the logic
of classical refutability and which were used in Section 3.2 to construct iso-
morphs of classical () and maximal negative logic ( , ) in Le. On the other
hand, the modality W cannot be dened in Le, because algebra 4 has a
subalgebra with the universe {1, 0}, i.e., the set {1, 0} is closed under all
operators denable in Le.
Thus, we can see that there is a close interrelation between L and Le.
J. Porte [84, 85] was the rst to pay attention to the above-mentioned in-
terrelation but he had L ukasiewiczs modal logic as a starting point for his
considerations. More exactly, in [84], it was proved that the modalities of
L
can be dened in a system obtained from classical logic by the addition
of a constant, in Portes denotation, about which nothing is postulated.
And conversely, that the modalities of L can be used to dene the constant
. In [85], J. Porte observed that his constant is close to the negation of
logic of classical refutability and stated that Le is denition equivalent to
the positive part of L.
Extend the language L of Le via a new negation symbol , L, :=
L {}, and dene the logic Le as a logic in the language L, with
axiom schemes of Le and, additionally, axiom schemes of classical negation
for :
1 . (p q) ((p q) p)
2 . p (p q)
3 . p p
Dene a translation : L, L+ +{L, } from the language of Le to

the language of L
ukasiewiczs modal logic and an inverse translation from
to the language of Le in such a way that they preserve
the language of L
propositional variables and all classical connectives and moreover,
() = L(p p), (L) = ,
where p is a propositional variable and is an arbitrary formula in the

language L+ + {L, }.
Dening an algebra 4 as an expansion of 4 via the classical negation,
i.e.,
4 :=
{1, a, 0, 1}, , , , , , 1,
we state the following facts.
Proposition 7.2.1 4 is a characteristic model for Le .
Proof. Obviously, Le has only two subdirectly irreducible models, 2 and

2 , which are expansions of Boolean algebra 2 and negative algebra 2 via
an additional Boolean negation . Therefore, Le = L2 L2 . Taking
into account 4
= 2 2 we immediately arrive at L4 = Le .
2
Proposition 7.2.2 Le is a conservative extension of Le.

7.2 Le and L
ukasiewiczs Modal Logic 107
Proof. If is a formula in the language L and Le , then 4 |= , and

also 4 |= , because the connective does not occur in . Since L4 = Le,
we have Le, which completes the proof.
2
We now give in a slightly modied version of Portes results [84, 85].
Proposition 7.2.3 The logics Le and L are denition equivalent via

and , the logic Le is denition equivalent to the positive fragment of L
via
the same translations.
The proof is by direct verication, because all logics involved in this state-
ment are four-valued.
2
The only dierence in the above translations from Portes translations is
the item (L) = . The original version was (L) = , but,
obviously, Lj ( ) ( ).
Proposition 7.2.3 means, rst of all, that there is a close connection
between L ukasiewiczs modalities and the paraconsistent negation of logic
of classical refutability. From this fact Porte [85, pp. 8788.] inferred a rather
categorical conclusion that the modalities of the L -system are very far from
what everybody calls possibility and necessity and/or that the weak
negation of CR (the logic of classical refutability) is very far from what
everybody calls negation. As noted above, L ukasiewicz constructed his
system so as to satisfy the minimal list of requirements for modal operators,
which gives rise to a long history of critics of L
ukasiewiczs modal logic L
, but
special discussion of this logic lies outside the scope of the present research.
The resent work by Font and Hajek [32] may be recommended to become
acquainted with the topic. Regarding the critics of the negation in the logic
of classical refutability based on its similarity to modal operators, in recent
years such similarity was not considered as something negative and was
intensively studied. For example, K. Dosen in a series of works [23, 24, 25]
treated the negation namely as a modal operator. In fact, the denition
equivalence of L ukasiewiczs modal logic and logic of classical refutability
means that the list of properties of modal operators given in the beginning
of this section is not broad enough to distinguish the necessity operator and
the operator of contradiction. The interrelation between Le and L stated
in Proposition 7.2.3 will be used in the next section to suggest the way to
generalize the notion of negation, which allows one to overcome the paradox
of minimal logic.
7.3 Paradox of Minimal Logic and Generalized

Absurdity
Recall that by the paradox of minimal logic we mean the following property
of Lj: from any contradiction one can infer in Lj an arbitrary negative
formula, i.e., for any and , we have
, Lj .
Due to this property, the negation makes no sense in inconsistent Lj-theories,

because all negative formulas are provable in them. This property is condi-
tioned, on the one hand, by the axiom for implication
p (q p)
(which is sometimes called the positive implication paradox) and, on the

other hand, by the unrestricted law reductio ad absurdum,
(p q) ((p q) p),
saying that if a formula implies a contradiction, one can negate this formula
without any restriction on the nature of this contradiction. Indeed, let T be
some inconsistent Lj-theory and , T . Then for an arbitrary formula
, we can infer in T the implications and using the positive
paradox, and we then infer applying reductio ad absurdum.
Of course, one can try to overcome the above mentioned paradox via
rejecting the positive implication paradox and passing in this way into the
eld of relevant logic. But we choose another way, leave intuitionistic impli-
cation unchanged and consider possible ways of restricting reductio ad ab-
surdum. It is worth noting that this idea has been exploited many times in
investigations in the eld of paraconsistency. For example, such well-known
paraconsistent logics as Settes P 1 [88, 102] and Da Costas C1 [19] have
reductio ad absurdum restricted to complex formulas, in case of P 1 , and to
formulas behaving consistently, in case of C1 .
Our approach is based on the correspondence between the necessity op-
erator L of L ukasiewiczs modal logic and the contradiction operator C,
C() := , in the logic of classical refutability, which was stated in the
previous section. In Le as well as in an arbitrary extension of minimal logic,
the negation can be dened via the constant absurdity, := , but
it can also be dened via the contradiction operator C(), := C().
It is clear that
Lj ( C()).
7.3 Paradox of Minimal Logic and Generalized Absurdity 109
This leads to the idea of dening a negation via the contradiction operator
C considered as a primary logical connective, := C(). It will
be shown below that the unrestricted reductio ad absurdum for negation
dened as above exactly corresponds to some paradoxical properties of C()
considered as a modal operator.
The results of Section 7.2 allow us to identify Le with a subsystem of L

and after such identication, we have
C() L() and L

L ( L()).
Notice that modal operators L and M have the following properties:
L( ) L() and L
L M ( ) M () .
Each of these properties can be inferred from the other modulo classical logic
and the relation dening M through L, and both of these properties have a
paradoxical nature. Indeed, if we accept the conjunction of two conditions
as necessary, it means from the intuitive point of view more than stating
that one of these conditions is necessary and the second just takes place. In
a similar way, assuming that it is possible that one of the two conditions
takes place should be weaker than the alternative of one of the conditions
and the possibility of the other. It is interesting that the numerous authors
who criticized L ukasiewiczs modalities did not pay any attention to these
paradoxes.
For any formula , we have
(p p) and (p p),
from which we infer, using the above paradoxical properties,
L() L(p p) and L

L M () M ((p p)).
The rst of these equivalences corresponds to the relation C()

(p p) for the contradiction operator in Le. Thus, the paradoxical
properties envisaged above allow us to dene the negation via the constant
absurdity.
Now we consider the language LC =
, , , C, i.e., L+ extended
by the contradiction operator C, and dene a C-logic as a logic in this
language containing axioms of Lp and the formula C(p) p. We say that
C is extensional in a C-logic L if L is closed under the rule

.
C() C()
Lemma 7.3.1 Let L be a C-logic and
C(p q) C(p) q, C(p q) p C(q) L,
then C is extensional in L.
Proof. Let L satisfy the conditions of the lemma. Assume L. Using

the rst equivalence and the axiom C(p) p we have
C( ) C() C() C().
In a similar way, we infer from the equivalence C(p q) p C(q) that

C( ) C(), and, nally, C() C().
2
Proposition 7.3.2 Let L be a C-logic and := C(). Then
C(p) p p L, (7.1)
moreover,
(p q) ((p q) p) L
i
C(p q) C(p) q, C(p q) p C(q) L.
Proof. From the well-known property p(p q) pq of the intuitionistic

implication and the axiom C(p) p, we immediately obtain the equivalence
C(p) p p, which justies the name contradiction operator for C.
Assume that (p q) ((p q) p) L and show that in this
case the negation may be dened via a constant, , where
:= C(p0 p0 ) for some xed variable p0 . Substituting the denition
of negation in reductio ad absurdum and exporting the second premise we
obtain
((p q) (p (q C(q)))) (p C(p))
Taking into account Lp p (q r) (p q) (p q) we arrive at
(p (q (q C(q))) (p C(p)).
Due to (7.1), the latter is equivalent to
(p C(q)) (p C(p)). (7.2)

7.3 Paradox of Minimal Logic and Generalized Absurdity 111
Substituting the tautology p0 p0 for q, we obtain
L (p C(p0 p0 )) p.
To state the inverse implication, we substitute p0 p0 for p and p for q

in (7.2)
((p0 p0 ) C(p)) ((p0 p0 ) C(p0 p0 )),
from which we have C(p) C(p0 p0 ). Taking into account that the
intuitionistic implication is increasing in the second argument, we arrive at
p (p C(p0 p0 ))
and, nally, at
p p .
Further, from this fact and the equivalence C(p) p p we infer as in
Lj that L C(p) p . This equivalence implies in a trivial fashion the
extensionality of C in L, moreover,
C(p q) (p q) (p ) q C(p) q.
The second desired equivalence easily follows from the one just proved by
extensionality.
We now assume that the equivalences C(pq) C(p)q and C(pq)
p C(q) hold in L. By Lemma 7.3.1 the operator C will be extensional in
L. Dening := C(p0 p0 ) and using the extensionality of C we obtain
C(p) C((p0 p0 ) p) C(p0 p0 ) p p
and, further,
p p C(p) p (p ) p .
The equivalence p p easily implies the unrestricted version of

reductio ad absurdum.
2
As noted at the beginning of this section, the paradox of minimal logic
is conditioned by the positive implication paradox and the unrestricted
version of reductio ad absurdum. In the denition of C-logics we leave the
intuitionistic implication unchanged, therefore, a C-logic meets the paradox
of minimal logic if and only if the negation dened via its operator C satises
the unrestricted version of reductio ad absurdum. In this way, we infer from
the last proposition
Corollary 7.3.3 A C-logic L satisfy for all formulas and the condition
, L
i
C(p q) C(p) q, C(p q) p C(q) L.
As we can see from this corollary, deleting the paradoxical property of

C allows one to avoid the paradox of minimal logic. In other words, if the
operator C of contradiction has more natural properties from the viewpoint
of modal logic, then the negation dened via C generates more a satisable
inference relation from the viewpoint of paraconsistent logic.
It was noted above that in minimal logic the negation may be dened in
two equivalent ways: via the constant and via the non-constant operator
C() of contradiction. This simple observation leads naturally to the distinc-
tion between the contradiction operator and what we will call the absurdity
operator, which we denote A(). In case of Lj, the absurdity operator is con-
stant (A() ). There are dierent intuitions behind these operators. By
C() we mean a contradiction expressed in terms of , i.e., a simultaneous
stating of and its negation , from which follows that C() should imply
. By A() we mean such a statement that reducing to it, i.e., proving
the implication A(), is enough to negate . This understanding of
A() does not assume that A() implies . The formulas A() and may
be incomparable, as takes place in case of Lj, where the formulas and
are incomparable in the general case. In fact, the contradiction operator C
can be considered as a special case of an absurdity operator satisfying an
additional assumption that C() for any . Moreover, if a negation
of some system can be dened in terms of an absurdity operator, such a
negation itself can be taken as an absurdity operator, which will produce
the same negation. To present these considerations in a precise form, we
dene an A-language as the positive language L+ with an additional unary
operator A, LA := L+ {A}, and an A-logic as a logic in the language LA
containing axioms of positive logic Lp.
Proposition 7.3.4 Let L be an A-logic. Dene the contradiction operator

C as C() := A() and the negation as := A(). The
following formulas are provable in L:
7.4 A- and C -Presentations 113
p p p and p p C(p);
C(p) p p and C(p) p;
C(p) A(p) and A(p) p.
Proof. All the formulas listed in the proposition can easily be inferred from
the given denitions and axioms of Lp. For example, the equivalence p
p p is an abbreviation for
p A(p) p (p A(p))
and the latter is just a particular case of the contraction law.

2
This proposition shows that C dened in terms of the absurdity operator
A really can be considered as a contradiction operator corresponding to the
negation dened via A and that there is the whole interval of operators,
from C to dening the same negation.
7.4 A- and C -Presentations

7.4.1 Denitions and First Results
Now we want to dene more exactly what a negation in one or another logic
can be dened via an absurdity operator means. We restrict ourselves to the
class of -logics, which are sets of formulas in the language L containing
axioms of positive logic and closed under the rules of substitution and modus
ponens. Further, we need several translations between languages L , LC , and
LA .
Translations : L LC , : LC L , : L LA , and : LC LA
preserve propositional variables and all positive connectives and act on other
connectives as follows:
() := C(); (C()) := ;
() := A(); (C()) := A().

Let L be a -logic. An A-logic L such that denes a faithful embedding
of L into L and is called an A-presentation of L. A C-presentation of L is
a C-logic L such that faithfully embeds L into L .
By an exact A-(C-)presentation of a -logic L we mean its A-(C-) pre-
sentation, which is the least among all A-(C-)presentations of L.
A strong C-presentation of L is a C-logic L denition equivalent to L

via and .
If L is a C-logic, then its A-presentation is an A-logic L such that
faithfully embeds L into L , and its exact A-presentation is the least
A-presentation.
If L is a C-logic, then its -presentation is a -logic L such that faith-
fully embeds L into L , and its exact -presentation is the least
-presentation.
First we state some simple properties of the notions introduced.
Proposition 7.4.1 1. Let L be an A-(C-)logic. There exists a unique

-logic L such that L is an A-(C-)presentation of L .
2. Let L be an A-logic. There exists a unique C-logic L such that L is

an A-presentation of L .
Proof. 1. It is enough to consider the case of A-logics. We dene by induction

the subset of -formulas in the set of all LA -formulas:
1. a propositional variable is a -formula;
2. if and are -formulas, then , , and are -formulas;

3. if is a -formula, then A() is a -formula.
For a -formula , dene a formula as a result of replacing all sub-

formulas of the form A() by subformulas . Note that = ( ).
Consider the intersection S of the logic L with the set of all -formulas. It is
clear that S is closed under the substitution of -formulas and the restricted
version of modus ponens: if , S and does not have the form
A(), then S. Therefore, the set
L := { | S} (7.3)
is a -logic. Check that L is an A-presentation of L .

If L , then = for a suitable formula S L. Consequently,
= ( ) = L. Conversely, if L , then S and () = ,
whence, L .
The uniqueness of a -logic L , for which L is an A-presentation, easily
follows from the above considerations. Indeed, every such logic L must have
the form (7.3).
2. Dene a set of C-formulas by replacing Item 3 in the denition of
-formulas by
3 . if is a C-formula, then A() is a C-formula.
The proof of this item is similar to the previous one. The desired C-logic
has the form
L := {C | L and is a C-formula},
where C is obtained from by replacing every subformula A() by

C().
2
In view of this statement Proposition 7.3.2 can be re-worded as
Corollary 7.4.2 Let L be a C-logic. Then
C(p q) C(p) q, C(p q) p C(q) L
i L is a C-presentation for some extension of Lj.

2
Proposition 7.4.3 1. If L is a -logic and L has an A-(C-)presentation,

then L has also an exact A-(C-)presentation.
2. If L is a C-logic having an A-presentation, it also has an exact

A-presentation.
Proof. This statement follows from the next simple lemma.

Lemma 7.4.4 Let L0 and L1 be propositional languages and : L0 L1
a translation. Further, let L be a logic in the language L0 and {Li | i I}
a family of logics in the language L1 such that faithfully embeds L into Li
for all i I. Then faithfully embeds L into iI Li .
In this way, if the set of A-(C-)presentations of some logic is non-empty, the
intersection of all such presentations gives an exact A-(C-)presentation of
that logic.
2
To prove further results we impose the restriction of extensionality on
the considered logics. Dene the extensionality of negation and absurdity
operators in the same way as for the operator C above. Let L be a -(A-)
logic. We say that the negation (respectively, the absurdity operator A)
is extensional in L if the logic L is closed under the rule

(respectively ).
A() A()
If L is a -(C-, A-)logic and the operator (C, A) is extensional in L, we

say that L is an extensional -(C-, A-) logic.
Remark. Note that in all logics under consideration the positive con-
nectives , , and are extensional in a natural sense, which easily follows
from the fact that -(C-, and A-) logics was dened so as to satisfy all
axioms of the positive logic Lp.
Proposition 7.4.5 If a -logic L has an A-presentation, then we have
L p p p.
Conversely, if the equivalence p p p can be proved in a -logic L

and additionally, L is an extensional -logic, then L has an A-presentation.
Proof. Let a -logic L has an A-presentation L . By the contraction law for

, we have
L (p A(p)) (p (p A(p))).
The last equivalence graphically equals
(p) (p p) = (p p p).
By assumption is a faithful embedding of L into L , consequently,
L p p p.
We now assume that L is an extensional -logic and L p p p.

For a formula in the language L , dene a formula A as a result of
replacing every subformula of the form by A(). Put
LA := {A | L}.
It is clear that LA is an A-logic and the extensionality of in L immediately

implies that A is extensional in LA . We check that is a faithful embedding
of L into LA . First of all, we note that for any formula ,
L i LA A .
Therefore, it is enough to prove that LA A () for any . We use

the induction on the structure of formulas.
This statement trivially holds for propositional variables. The case of
positive connectives easily follows by extensionality. It remains to consider
the case = . By the induction hypothesis we have A () in LA .
This fact and the extensionality of A imply LA A( A ) A(()). On

the other hand, we have A() (() A(())) since the equivalence
p (p p) holds in L by assumption. The last two facts together imply
LA A( A ) (() A(())),
i.e., LA ()A ().

2
Proposition 7.4.6 1. If an extensional -logic L has an A-presentation,

it has also a C-presentation.
2. If an extensional -logic L has a C-presentation, this C-presentation

is unique and it is also an exact and strong C-presentation.
Proof. 1. The set C of formulas in the language L we dene as the least set
of formulas satisfying the following conditions:
1) propositional variables belong to C;
2) if C and C, then the formulas , , and are in

C;
3) if C, then C.
Let LC := L C. The set LC is closed under the rules of modus ponens

and substitution of formulas from C, moreover, it contains all formulas of
the form ( ) for in C. Therefore, the set
L := {C | LC },
where C is obtained from by substituting a subformula C() for any

subformula of the form , forms a C-logic.
Check that L is the desired C-presentation of L. First we note that the
equivalence
L i L C
holds for any C. The logic L has an A-presentation by assumption,
whence, by Proposition 7.4.5 we have in L the equivalences
p (p p) (p (p p)).
Therefore, in L any formula is equivalent to a formula obtained by

replacing every subformula of by ( ). Note that belongs
to C, moreover, () = C . In this way, we arrive at the following chain of

equivalences
L L L C L ,
which completes the proof of this item.

2. It follows easily from the considerations of the previous item that
any C-presentation of an extensional -logic L coincides with the set {C |
L C}. Therefore, the logic L may have at most one C-presentation.
The exact C-presentation equals the intersection of all C-presentations, it
therefore coincides with the unique C-presentation of L.
It remains to prove that a C-presentation L of L is strong, i.e., that
is a faithful embedding of L into L and that the translations and are
mutually inverse. It is clear that for any formula in the language LC , the
formula () belongs to C and, conversely, any formula in C has the form
(). Therefore, the above presentation of L immediately implies that is
a faithful embedding.
Let be a formula in the language L . The equivalence ()
follows from the fact () graphically equals the formula dened in
the previous item. For a formula in the language LC , the equivalence
() easily follows from the extensionality of L and the equivalence
C(p) (p (p C(p))) which holds in any C-logic.
2
Proposition 7.4.7 Let L be an extensional -logic, L1 an extensional

A-presentation of L, and L2 a C-presentation of L. Then L1 is an
A-presentation of L2 .
Proof. 1. By the previous proposition L2 is a strong C-presentation of L.

Therefore, we have for any formula ,
L2 L L1 .
Moreover, C() = ( ) = ( A()). The latter

formula is equivalent in L1 to A(). From this fact and the exten-
sionality of L1 , we obtain by induction
L1
for any .
2
The above results concerning the existence of A- and C-presentations,

uniqueness of C-presentations, etc., used essentially the extensionality prop-
erty. Most of the well-known systems of paraconsistent logic are not exten-
sional and need, therefore, special treatment. In the next sections, we take
two basic systems of paraconsistent logic and consider the problem of pre-
senting the negation via absurdity and contradiction operators for these
systems.
7.4.2 Logic CLuN

All logics considered in this and the following subsections have the same pos-
itive fragment, namely, Lk+ , and admit semantic characterization in terms
of {0, 1}-interpretations non-truth-functional for operators of negation, ab-
surdity or contradiction. The latter means that the value of an interpretation
on a formula of the form (A(), C()) is not determined uniquely by
the value ().
For the language L , where # {, A, C}, we dene an interpretation
as a mapping from the set of L -formulas to the set {0, 1} satisfying for
arbitrary and the following properties:
1 ) ( ) = 1 i () = 1 and () = 1;
2 ) ( ) = 1 i () = 1 or () = 1;
3 . ( ) = 1 i () = 0 or () = 1.
The logic CLuN was introduced in [6] (under the name P I). The name
CLuN we can understand as classical logic admitting gluts (see also [7]). It
is a fairly weak paraconsistent logics, which is in a sense basic for the class of
logics considered by D. Batens in [6]. The logic CLuN can be axiomatized
modulo classical positive logic via the law of excluded middle, i.e., we have
CLuN = Lk+ + {p p},
where Lk+ is an expansion of Lk+ to the language L .

We denote by U the class of interpretations satisfying for any the
property:
U . If () = 0, then () = 1.
Proposition 7.4.8 [6, p.205] For any formula , CLuN i () = 1

for all U.
We start to consider A- and C-presentations of CLuN with the following

observation.
Proposition 7.4.9 The logic CLuN has no strong C-presentation.
Proof. If CLuN has a strong C-presentation, the equivalence holds

in CLuN for any . Take a formula p and compute
(p) = (p (p p)) ((p (p p)) (p (p p)))
(p p) (p (p p)).
Let be an interpretation from U such that (p) = 0, (p) = (p) = 1,
and ((p (p p))) = 0. One can nd out such interpretation in U , be-
cause (p) = 0 implies (p (p p)) = 1. As we can see, ((p)) = 0,
and so (p (p)) = 0. By the previous proposition we have
CLuN p (p).
2
However, one can nd exact A- and C-presentations of CLuN.
Dene an A-logic CLuNA as an expansion of Lk+ to the language LA ,
and let U A be the set of all interpretations of the language LA . We have the
following completeness theorem.
Proposition 7.4.10 For any , CLuNA if and only if () = 1 for

all U A .
Proof. The soundness can be checked directly and to prove the completeness
we use an easy modication of Henkins construction. As usual, dene a
prime CLuNA -theory as a set of formulas T , which contains CLuNA , is
closed under modus ponens and has the disjunction property, i.e.,
T implies T or T .
In a standard way, one can prove the extension lemma.
Lemma 7.4.11 Let a set of formulas S and a formula be such that

S CLuNA . There is then a prime CLuNA -theory T such that S T
and T .
2
Let CLuNA 0 and T be a prime CLuNA -theory such that 0 T .

Dene an interpretation T on propositional variables and formulas of the
form A() as follows:
T (p) = 1 p T, T (A()) = 1 A() T.
By denition of interpretation, the mapping T extends uniquely to the set

of all formulas. For any prime CLuNA -theory, the following equivalences
hold:
T T or T ;
T T and T ;
T T or T .
The rst equivalence follows from the disjunction property, the second is
obvious, check the last one. If T and T , then T by modus
ponens. Conversely, if T , then T in view of ( )
CLuNA . If T , then T by the positive axiom ( ).
From the denition of T and the above equivalences we obtain
T () = 1 T,
consequently, T (0 ) = 0.
2
Proposition 7.4.12 CLuNA is an exact A-presentation of CLuN.
Proof. First we check that for any , if CLuN , then CLuNA . The
logics CLuN and CLuNA have the same positive fragment, therefore, it is
enough to check that the formula (p p) = p (p A(p)) is provable in
CLuNA , which is obviously true.
Conversely, assume that for some 0 we have CLuN 0 . In this case,
(0 ) = 0 for suitable U. Consider an interpretation of the language
LA satisfying the following conditions: (p) = (p) for any propositional
variable p; (A()) = ( ) for any in the language L . Obviously,
we can construct such an interpretation and it belongs to the class U A , which
contains all interpretations of the language LA .
Prove by induction on the structure of formulas that () = () for
all . This is obviously true for propositional variables. The case of positive
connectives is trivial, so it remains to consider the case of negation. Assume

that () = () and calculate
(()) = ( A()) = () (A()) =
( ) = ( ( )) = ().
The last equation is due to the fact that CLuN p p (p p).
According to the proven equality (0 )) = (0 ) = 0, i.e., the formula
0 is not provable in CLuNA . We have thus proved that CLuNA is an
A-presentation of CLuN. The exactness of this presentation immediately
follows from the fact that CLuNA is the least A-logic with classical positive
fragment. Obviously, any A-presentation of CLuN must have Lk+ as its
positive fragment.
2
Similarly to the case of A-presentation we can take the least C-logic with
classical positive fragment as a C-presentation of CLuN. Put
CLuNC := Lk+C + {C(p) p},
where Lk+C is an expansion of Lk+ to the language LC . We denote by U C

the class of all interpretations of the language LC , satisfying for any the
condition:
U C . If () = 0, then (C()) = 0.
Proposition 7.4.13 For any , CLuNC if and only if () = 1 for

all U C .
Proof is analogous to Proposition 7.4.10.

2
Proposition 7.4.14 CLuNC is an exact C-presentation of CLuN.
Proof of this statement can be obtained, in fact, by replacing A by C and

by in the proof of Proposition 7.4.12.
2
C
Due to Proposition 7.4.9 the logic CLuN cannot be a strong C-
presentation, but we can prove that the inverse translation also embeds
faithfully CLuNC into CLuN.
Proposition 7.4.15 CLuN is an exact -presentation of CLuNC .

Proof. The fact that CLuNC implies CLuN for any can be
checked directly.
Assume that for some formula 0 we have CLuNC 0 and consider an
interpretation U C with (0 ) = 0. In U, one can nd an interpretation
of the language L such that (p) = (p) for all propositional variables
p and () = ( C) for any . Using induction on the structure
of formulas, we check that = for any . Clearly, it is enough to
consider the case of C-operation. Let = for some . Then
(C) = ( ) = () = ( C) = (C).
The desired equality is proved, in particular, we have (0 ) = 0. Conse-

quently, CLuN is a -presentation of CLuNC .
Further, any -presentation of CLuNC has positive fragment Lk+ and
contains the formula p p, because this formula is equivalent modulo Lk+
to p (p (p p)) = (p (p Cp)). Therefore, CLuN is the least
C-presentation of CLuNC and so it is exact.
2
7.4.3 Settes Logic P1

We now turn to the rst example of maximal paraconsistent logic, namely,
to the system P 1 suggested by A. Sette in [102]. This is a sublogic of classical
logic maximal in a sense that adjoining to P 1 any new classical tautology
yields classical logic. The negation of P 1 is not extensional, however, as we
will see, P 1 has even a strong C-presentation. Originally P 1 was dened as a
three-valued logic with the set of truth-values {T 0 , T 1 , F }, where T 0 and F
are, in fact, the classical truth and falsehood, and T 1 is in a sense a second
truth, whose negation is also true. Both truths T 0 and T 1 are distinguished
values. The matrix for P 1 is
{T 0 , T 1 , F }, , , {T 0 , T 1 }, where operations
and are dened via the following truth tables:
T0 T1 F
T0 T0 T0 F T0 F
T1 T0 T0 F T1 T0
F T0 T0 T0 F T0
The connectives and are introduced via the denitions:
:= ( ) ( ),
:= ((( ) ) (( ) )) ( ).
A. Sette [102] gives the following axiomatics for P 1 :

1. p (q p)
2. (p (q r)) ((p q) (p r))
3. (p q) ((p q) p))
4. (p q) (p q)
The only inference rules are, as usual, substitution and modus ponens. It is
clear that P 1 is a sublogic of Lk, moreover, the following theorem holds.
Theorem 7.4.16 [102] For any Lk, if is not provable in P 1 , then

P 1 + {} = Lk.
2
In [2], it was stated that P 1 is equivalent to the system 2 introduced in

[50]. Axiomatics of 2 includes traditional axioms of classical positive logic
and the following axiom schema for negation:
1 . ( ) (( ) ), where is molecular, i.e., dierent

from a propositional variable.
From 2 we can easily pass to an axiomatization of P 1 convenient for

our goals.
Proposition 7.4.17 The logic P 1 is axiomatizable modulo the classical pos-

itive logic via the following axiom schemes:
1S ) ;
2S ) ( ) (( ) ), where is molecular.
Proof. On the one hand, we can use truth-tables to check that schemes
1S and 2S hold in P 1 . On the other, the schema 1 easily follows from
( ) (( ) ), a substitution case of 2S , and 1S .
2
Using the maximal property we can easily obtain a semantic characteri-
zation of P 1 in terms of {0, 1}-interpretations. Denote by P the class of all
interpretations of the language L satisfying the properties:
1P . If = 0, then = 1.
2P . If is molecular and = 1, then = 0.
Proposition 7.4.18 For any formula , P 1 if and only if = 1 for

all P.
Proof. Put L := { | = 1 for all P}. We have P 1 L Lk.

Indeed, the former inclusion can be veried directly, the latter follows from
the fact that all classical interpretations belong to P. Due to Theorem 7.4.16
one of these inclusions is not proper. In P, there is an interpretation such
that p = p = 1 and p = 0, which means that p p L , i.e.,
L = Lk. Therefore, L = P 1 .
2
Having convenient axiomatics and semantics for P 1 , we may turn to its
presentations.
Lemma 7.4.19 If L is an A-presentation of P 1 , then L contains the fol-

lowing schemes:
1A . A( A) ;
2A . ( A) , where is molecular.
Proof. Consider a -translation of 2S ,
( ) (( ( A)) ( A).
Modulo Lk+ it is equivalent to
( ( A)) ( A)
( ( ( A))) A ( ( A)) A
( A) ( A).
In particular, we have
L ( A) (( A) A( A)).
At the same time, the formula
(( A) A( A)) , (7.4)
a -translation of , is provable in L. From the last two formulas

we immediately have
L ( A) ,
where is molecular.
Further, the positive axiom

A( A) (( A) A( A))
together with (7.4) yields the scheme 1A .
2
Denote by P A the A-logic axiomatized modulo Lk+ by schemes 1A and
2A and by P A the class of interpretations of the language LA satisfying
the condition
1A . If is molecular and = 1, then (A) = 0.
Lemma 7.4.20 If P A , then = 1 for all P A .
The proof is by direct verication.
2
Proposition 7.4.21 The logic P A is an exact A-presentation of P 1 .
Proof. Let L := { | P A }. As it was stated in the proof of Lemma
7.4.19, the -translation of 2S is equivalent in P A to a particular case of 2A .
The formula ( ) has the following quasi-proof in P A :
1) A( A) , an axiom;
2) (( A) A( A)) (( A) ), by monotonicity of
in the second argument from 1);
3) (( A) ) , the Peirce law;
4) (( A) A( A)) (= ( )), by transitivity
from 2) and 3).
We have thus proved the inclusion P 1 L .

Let P A 0 and is an arbitrary classical interpretation. Dene an
interpretation of LA by the conditions p = p for any p and (A) =
|1 |. It is not hard to see that P A and = for any . By
Lemma 7.4.20, 0 = 1. And we have in this way L Lk. This inclusion is
proper, because the formula (p p) is not in P A . The last fact can be
easily checked using Lemma 7.4.20. Only one of the inclusions P 1 L Lk
is proper, whence P 1 = L .
We have thus proved that P A is an A-presentation of P 1 , the exactness
immediately follows from Lemma 7.4.19.
2
We now turn to C-presentations of P . 1
Lemma 7.4.22 If L is a C-presentation of P 1 , then it contains the schema:
1C ) C , where is molecular.
Proof. By Lemma 7.4.19 L ( C) for molecular , which is

equivalent to 1C in view of the axiom Cp p of C-logics.
2
C +
Let P be a C-logic axiomatized modulo Lk by the axiom scheme
1C . We denote by P C the class of all interpretations of LC satisfying the
conditions:
1C . If = 0, then C = 0.
2C . If is molecular and = 1, then C = 0.
Lemma 7.4.23 If P C , then = 1 for all P C .

2
Proposition 7.4.24 The logic P C is an exact and strong C-presentation

of P 1 .
Proof. The fact that P C is an exact presentation can be proved similarly to

Proposition 7.4.21. We can directly check that P 1 is a -presentation of P C .
It remains to prove that any formula is equivalent to in P 1 and that
is equivalent to in P C . This can be easily stated using induction on the
structure of formulas and the facts that P 1 ( ) ( ) for
molecular and and that P C ( ) (C() C()) for molecular
and .
2
1
Remark. Note that P is not an exact -presentation of P . The least
C
-presentation of P C can be dened via the class of interpretations satis-

fying the condition = 1 = 0 for molecular . And this logic has
no C-presentation.
Chapter 8
Semantical Study
of Paraconsistent
Nelsons Logic1
This chapter starts the second part of the book devoted to paraconsistent
Nelsons logic N4 and to the class of its extensions.
The natural rst step in the investigation of the class of N4-extensions
is to provide an adequate algebraic semantics for the logic N4, i.e., char-
acterizing N4 via a variety of algebraic systems V such that there exists
a natural dual isomorphism between the lattice of N4-extensions and the
lattice of subvarieties of V.
We consider two variants of Nelsons paraconsistent logic. The logic N4 is
determined in the language
, , , and its positive fragment coincides
with positive logic Lp. The logic, which we denote N4 , is a conservative
extension of N4 in the language with an additional symbol for the constant
absurdity and
, , , -fragment of N4 coincides with intuitionistic
logic Li. In the course of semantic investigations, we work mainly with the
logic N4 and we then show how the obtained results can be adapted to
N4 .
For the logic N3, an explosive variant of Nelsons logic, the algebraic
semantics was introduced in [90], where N3 was characterized via the variety
of N -lattices. Later D. Vakarelov [110] and independently M.M. Fidel [28]
found a very convenient representation of N -lattices in the form of so-called
twist-structures. Finally, M.M. Fidel [29] suggested a characterization of N3
1
Parts of this chapter were originally published in [72].
131
132 8 Semantical Study of Paraconsistent Nelsons Logic
via Heyting algebras with a special family of unary predicates distinguishing

classes of counterexamples for elements of a Heyting algebra. Structures of
this kind are called Fidel-structures, or F-structures.
In our semantic studies of N4, we go in the opposite direction com-
pared to N3 and start with Fidel-structures (Section 8.2). A semantics of
this kind for N4 can be obtained as an immediate generalization of the
semantics from [29] and the completeness theorem has an easy and short
proof. Then (Section 8.3) we introduce twist-structures for N4 and show that
Fidel-structures and twist-structures are mutually denable, which leads to
a characterization of N4 in terms of twist-structures. In Section 8.4, general-
izing the notion of N -lattices we dene N4-lattices and state that all twist-
structures are N4-lattices and conversely, that any N4-lattice is isomorphic
to a twist-structure. With any N4-lattice A, we associate the set DA of its
distinguished elements and prove the completeness result with respect to
the obtained class of logical matrices. It is essential that the set D A is not
one-element in the general case. This fact has the following explanation. In
N4, as well as in N3, the usual equivalence := ( ) ( )
does not have the congruence property with respect to negation. It is possible
that the negations of equivalent formulas are not equivalent. Only the strong
equivalence := ( ) ( ) is a congruence wrt negation.
Fortunately, due to the axiom schema ( ), all negations of tau-
tologies are equivalent in N3. Namely this fact allows one to characterize
N3 by the class of N -lattices, where the unit element (the greatest element
wrt the lattice ordering) is the only distinguished element. Another situa-
tion is in case of N4. For example, (p p) and (p (q p)) are not
equivalent in this logic. The rst formula is equivalent to p p, the second
to (p p) q, and these two contradictions are not equivalent in the para-
consistent logic N4. This complication can be overcome due to the fact that
sets of distinguished elements in N4-matrices are denable. In Section 8.5,
we prove that N4-lattices form a variety and establish a dual lattice isomor-
phism between the lattice of N4-extensions and the lattice of subvarieties
of the variety of N4-lattices. To obtain an axiomatization of the variety of
N4-lattices the results of A. Pynko [89] are essentially used. The desired
axiomatization is obtained as appropriate weakening of an axiomatization
of the class of implicative De Morgan lattices introduced in [89].
In the last section of this chapter we show how to dene twist-structures
for the logic N4 , and then dene N4 -lattices and establish the dual iso-
morphism between the lattice of N4 -extensions and the lattice of subvari-
eties of the variety of N4 -lattices.
8.1 Preliminaries 133
8.1 Preliminaries
In this part of the book, we deal with the propositional languages L :=
{, , , }, where is a symbol for strong negation, and L := L {}
with an additional symbol for the constant absurdity. By F or (F or ) we
denote the trivial logic in the language L (L ), i.e., the set of formulas of
this language. The connectives of equivalence and of strong equivalence
are dened as follows:
:= ( ) ( ), := ( ) ( ).
As above, logics will be dened via Hilbert-style deductive systems with only
the rules of substitution and modus ponens. In this way, to dene a logic it
is enough to give its axioms. Paraconsistent Nelsons logic N4 is a logic in
the language L characterized by the following list of axioms:
A1) p (q p);
A2) (p (q r)) ((p q) (p r));
A3) (p q) p;
A4) (p q) q;
A5) (p q) ((p r) (p (q r)));
A6) p (p q);
A7) q (p q);
A8) (p r) ((q r) ((p q) r));
A9) p p;
A10) (p q) ( p q);
A11) (p q) ( p q);
A12) (p q) (p q).
To obtain explosive Nelsons logic N3 we add to the list of N4-axioms,

the explosion axiom for strong negation:
A13) p (p q).
The logic N4 is a logic in the language L determined by the axioms

A1A13 and the two additional axioms for constant :
A14) p and A15) p .
Due to axiom A14 the intuitionistic negation can be dened in N4 as
:= . If we put := (p0 p0 ), one can prove
N3 p, p .
The rst formula is equivalent to a particular case of axiom A13 modulo

positive logic, the second follows from the equivalence (p0 p0 ) ob-
tained in turn from A9. In particular, the intuitionistic negation is denable
in N3. This is why we do not consider two dierent logics N3 and N3 .
We say that a formula of the language L or L is a negation normal
form (nnf) if it contains the symbol only before atomic formulas. The
following translation () sends every formula to a negation normal form,
where p P rop and # {, , }:
p = p p = p
= # = #
( ) = ( ) =
( ) = =
Proposition 8.1.1 For any formula in the language L (L ),
N4(N4 ) .
The proof easily follows from strong negation axioms A9A12.

2
Proposition 8.1.2 The logic N4 is a conservative extension of N4.
Proof. Let be a formula in the language L and 0 , . . . , n = is a proof

N4 . Let P be the set of all variables occurring in this proof. Put
of in
:= pP p p. Replacing in the above proof any occurrence of by
pi pi , pi P , and then any occurrence of by we obtain a quasi-proof
in N4. Indeed, every axiom of the form is obviously replaced by
a provable formula. That every axiom is replaced by a provable
formula follows from Proposition 8.1.1 and the fact that in positive logic
Lp, every formula follows from the conjunction of its variables.
2
8.2 Fidels Semantics 135
An important peculiarity of Nelsons logics N4 and N4 is that the

provable equivalence is not a congruence relation. However, axioms A1-A8
constitute the axiomatization of positive logic Lp (see Chapter 2). This
means that the provable equivalence is a congruence relation with respect
to positive connectives. More exactly, for any formulas 0 , 1 and a positive
formula (p) with a propositional parameter, the provability of 0 1 in
N4 (or N4 ) implies the provability of (0 ) (1 ) in that logic. The
provable strong equivalence will be a congruence relation in N4 as well
as in N4 .
Proposition 8.1.3 The logics N4 and N4 are closed under the weak re-
placement rule
0 1 0 1
.
(0 ) (1 )
Proof. It follows easily from Proposition 8.1.1.

2
8.2 Fidels Semantics

First, we present a semantics for N4 in terms of F-structures (Fidel-
structures), which is an immediate generalization of the semantics for N3
developed by M.M. Fidel in [29].
Definition 8.2.1 An F-structure is a tuple
A =
A, , , , 1, {Na }aA ,
where
A, , , , 1 is an implicative lattice and {Na }aA is a family of
sets satisfying the following properties:
1) for any a A, = Na A;
2) for any a, b A, a Na , and b Nb , the following relations hold
a b Nab , a b Nab , a Na ;
3) for any a, b A and b Nb , we have a b Nab .
We say that an F-structure A is saturated if the equality Na = A holds

for all a A. For F-structures we use the denotation
A, {Na }a|A| , where
A is an implicative lattice and {Na }a|A| a family of subsets of |A|.
With an arbitrary set of formulas non-trivial with respect to N4 we

associate an equivalence relation on the set of formulas in a standard
way. For any formulas and , we set
N4 .
The coset of an element with respect to is denoted by [] and the

family of all cosets by L . In case of N4, the relation is a congruence
with respect to all positive connectives, which allows one to dene on the
set L operations , and as follows:
[] [] := [ ] , where {, , }.
Due to the fact that N4 contains all axioms of positive logic, the struc-
ture
L , , , , 1, where 1 denotes the coset of formulas deducible from
, 1 = [p p] , is an implicative lattice. We extend it to an F-structure
putting
N[] := {[ ] | [] } for any [] L .
Later on, we omit the lower index when it does not lead to confusion.
Let
L :=
L , , , , 1, {N[] }[]L .
Lemma 8.2.2 For every set of formulas non-trivial wrt N4, the structure
L is an F-structure.
Proof. It is enough to check Items 2, 3 of Denition 8.2.1, which can easily

be done with the help of the axioms for strong negation.
2
We say that an F-structure
A, {Na }a|A| is a substructure of an
F-structure
B, {Nb }b|B| if (1) A is an implicative sublattice of B and
(2) for any a |A|, Na Na .
Lemma 8.2.3 Any F-structure is a substructure of a saturated F-structure.
Proof. Any structure

A, {Na }a|A| can be embedded into a saturated struc-
ture
A, {Na }a|A| , where A is the same implicative lattice and Na = |A|
for all a A.
2
8.2 Fidels Semantics 137
Definition 8.2.4 Let

A, {Na }a|A| be an F-structure. The mapping v from
F or into |A| is said to be a valuation into this structure if the following con-
ditions hold for arbitrary formulas and :
1) v( p) Nv(p) for a propositional variable p;
2) v( ) = v() v() for {, , };
3) v( ( )) = v( ) v( ) and v( ( )) = v( ) v( );
4) v( ( )) = v() v( );
5) v( ) = v().
Remark. It can easily be seen from this denition that every valuation
into an F-structure is uniquely determined by its restriction to the set of
propositional variables and their negations.
The following fact can be checked by induction on the structure of for-
mulas.
Lemma 8.2.5 For any F-structure

A, {Na }a|A| , any valuation v and for-
mula , we have v( ) Nv() .
2
Now we dene a semantic consequence relation |=F between sets of for-

mulas and formulas. Let be a set of formulas and be a formula. For an
F-structure A, the relation |=AF holds if and only if for any valuation v
into A the condition v() = 1 for all implies v() = 1. The relation
|=F means that |=A F for any F-structure A.
Lemma 8.2.6 Let B be a substructure of an F-structure A. Then |=A

F
implies |=B
F .
Proof. Indeed, if v is a valuation into B, it is also a valuation into A.

2
It is not hard to check that the introduced consequence relation is closed
under substitution.
Lemma 8.2.7 Let (p1 , . . . , pn ) be a set of formulas, (p1 , . . . , pn ), 1 , . . .,

n be formulas, and A an F-structure. If |=A F , then
(1 , . . . , n ) |=A
F (1 , . . . , n ).
2
Theorem 8.2.8 Let be a set of formulas non-trivial with respect to N4

and a formula. The following conditions are equivalent:
1) N4 ;
2) |=F ;
3) |=A
F for any saturated F-structure A;
4) |=L
F .
Proof. 1) 2) The soundness of consequence relation with respect to the

introduced semantics can be proved in a standard way via induction on the
length of proof.
Implications 2) 3) and 3) 4) are obvious.
4) 1) Assume |=L
F and consider a mapping v from the set of
propositional variables and their negations to L such that for a variable p,
we have v (p) = [p] and v ( p) = [ p]. This mapping extends in a unique
way to a valuation v into L . Using Items 25 of Denition 8.2.4 one can
easily check that v() = [] for any formula .
If , then clearly N4 , i.e., [] = 1. In view of the above
property of v, we have v() = 1 for all . This implies by assumption
that v() = [] = 1, i.e., N4 .
2
If is a positive formula, the validity of this formula on an F-structure
A, {Na }a|A| is equivalent to its validity on the implicative lattice A. This

fact and Theorem 8.2.8 imply
Corollary 8.2.9 N4 is a conservative extension of positive logic.
8.3 Twist-structures
Following M.M. Fidel [28] and D. Vakarelov [110] we introduce the notion of
twist-structures. Because we need a semantics for the logic N4, where the
intuitionistic negation is not denable, as underlying algebraic systems for
twist-structures we take implicative lattices unlike the works [28, 110], where
twist-structures were dened over Heyting algebras. The apt term twist-
structure was introduced in [47]. We dene an algebraic structure on the
direct square of the universe of an underlying algebra in such a way that
operations are not componentwise, but they are twisted in some sense.
8.3 Twist-structures 139
Definition 8.3.1 Let A =

A, , , , 1 be an implicative lattice.
1. The full twist-structure over A is the algebra
A =
A A, , , ,
with the twist-operations dened for (a, b), (c, d) A A as follows:
(a, b) (c, d) := (a c, b d), (a, b) (c, d) := (a c, b d)
(a, b) (c, d) := (a c, a d), (a, b) := (b, a).
2. A twist-structure over A is an arbitrary subalgebra B of the full twist-

structure A such that 1 (B) = A (in which case also 2 (B) = A).
3. The class of all twist-structures over A is denoted S (A).
It should be noted that our terminology diers from that of [47], where
the term twist-structure is used for twist-structures over Heyting algebras
(see Section 8.6) and with the unverse of the form {(a, b) | a b = 0}, where
0 is the least element of the underlying algebra.
A valuation into a twist-structure B is dened in the usual way as a
homomorphism of the algebra of formulas into B. The semantic consequence
relation |= over twist-structures is dened as follows. Let be a set of
formulas, a formula, and B a twist-structure. The relation |=B holds
if and only if for any valuation v in B the condition that 1 v() = 1 for all
implies 1 v() = 1. The relation |= means that |=B for all
twist-structures B.
One can easily pass from F-structures to twist-structures and back. Let
A =
A, {Na }a|A| be an F-structure. We associate with A a twist-
structure A over A with the following universe
|A | = {(a, b) |A| |A| | b Na }.
Using the denitions of twist-operations and of F-structures one can check

that for any F-structure A the set |A | is closed under all twist-operations,
and so the structure A is well dened.
On the other hand, given a twist-structure B S (A) we dene an
F-structure B F =
A, {Na }a|A| , where
Na = {b A | (a, b) B}
for any a |A|. The correctness of this denition can be veried directly.
Further, let A be an F-structure and v a valuation into A. We dene a

valuation v into A for a propositional variable p as
v (p) := (v(p), v( p)).
By induction on the structure of formulas we obtain then the following

statement.
Lemma 8.3.2 For any formula , v () = (v(), v( )).

2
Now, let B be a twist-structure and v a valuation. We dene a valuation

vF into B F as a unique valuation satisfying the conditions
vF (p) = 1 v(p) and vF ( p) = 2 v(p)
for all propositional variables p. Again, by induction on the structure of

formulas one can prove the following lemma.
Lemma 8.3.3 For any formula , vF () = 1 v().

2
Proposition 8.3.4 1. Let A =

A, {Na }a|A| be an F-structure and v
a valuation into A . Then
(A )F = A and (v )F = v.
2. Let B be a twist-structure and v a valuation into B. Then
(B F ) = B and (vF ) = v.
Proof. 1. Let (A )F =

A, {Na }a|A| . For any pair of elements a, b |A|,
we have the equivalences
b Na (a, b) A b Na .
This proves the equality (A )F = A .

Take an arbitrary formula and calculate
(v )F () = 1 v () = 1 ((v(), v( ))) = v().
The second equality in this chain is due to Lemma 8.3.2.

2. Let B be a twist-structure over A, B F =

A, {Na }a|A| and let a, b
|A|. The rst equality of this item follows from the equivalences:
(a, b) (B F ) b Na (a, b) B.
For a formula , we calculate applying Lemma 8.3.3
(vF ) () = (vF (), vF ( )) =
= (1 v(), 1 v( )) = (1 v(), 2 v()) = v().

2
Proposition 8.3.5 Let be a set of formulas and a formula.
1. For any F-structure A,
|=A A
F |= .
2. For any twist-structure A,

F
|=A A
|=F .
Proof. 1. In view of the previous proposition, any valuation v into A has the
form vF for a suitable A -valuation v and any valuation v into A can be

for some A-valuation v . These observations easily imply
represented as v
the desired equivalence.
2. Again, in view of the previous proposition (AF ) = A, which allows
one to reduce this item to the previous one.
2
For a set of formulas non-trivial with respect to N4, we dene a Lin-
denbaum twist-structure L := (L ) , i.e.,
=
{([] , [ ] ) | [] }, , , , .
L
The next completeness result easily follows from the completeness theo-
rem for Fidel structures, Proposition 8.3.5 and an obvious equivalence:
(an F-structure B is saturated) (the twist-structure B is full).

Theorem 8.3.6 Let be a set of formulas, non-trivial with respect to N4,

and a formula. The following conditions are equivalent:
1) N4 ;
2) |= ;
3) |=B
for any full twist-structure B;
L
4) |= .
Remark. It was already mentioned that the provable equivalence is
a congruence in N4, therefore, on the set
L
N4
:= {[] | F or},
where [] := { | N4 }, one can dene a structure of Lindenbaum
algebra in a standard way. Denote this algebra by L
N4 . Note that the map-
ping [] ([], [ ]) determines an isomorphism of Lindenbaum algebra
LN4 and Lindenbaum twist-structure LN4 .

8.3.1 Embedding of N3 into N42

Having the semantics for N4 in terms of twist-structures we are ready to
prove that explosive N3 is faithfully embedded into paraconsistent N4. First
we describe twist-structures, which are models of N3.
Lemma 8.3.7 Let A be an implicative lattice and B S (A). We have
B |= N3 if and only if A has the least element 0 and for any (a, b) B,
a b = 0.
Proof. As follows from the denition of twist-operations, the truth of axiom
A13 on B is equivalent to the fact that for any (a, b), (c, d) B, a b c.
By Denition 8.3.1 c is an arbitrary element of A, whence for any (a, b) B,
a b is the least element of A.
2
Definition 8.3.8 Let := (p0 p0 ), where p0 is a xed propositional
variable. We dene a transformation () of formulas as follows.
1. For a propositional variable p,
(p) := p , ( p) := ( p (p )) .
2
This section was originally published in [71].
2. ( # ) := # , where and are nnf and # {, , }.
3. () () for any formula not a nnf.
Theorem 8.3.9 For any formula ,
1. N3 if and only if N4.
2. N3 .
Proof. 1. First, we prove the direct implication. For any substitution instance
of one or another N4-axiom, is provable in N4 due to Proposition 8.1.1.
The transformation () preserves all positive connectives for nnf, therefore,
() is also provable in N4 as a substitution instance of . Let ,
N3 and , ( ) N4. Note that ( ) = ( ) = (

) = , which immediately implies that = () N4. We have
thus proved that the set of formulas, for which the considered implication
holds, is closed under modus ponens. In this way, it remains to check that
formulas of the form
( ( )) = ( ) ( )
are provable in N4. The latter formula is equivalent in N4 to (( ) )

, therefore, the desired result follows from the next lemma.
Lemma 8.3.10 For any formula , the following holds:
a) N4 ;
b) N4 ( ) .
Proof. a) This item is true by denition for propositional variables and their
negations. Any formula can be obtained from formulas of the form p and
( p) with the help of positive connectives, which allows one to complete
the proof by an easy induction on the structure of formulas.
b) We use again an induction on the structure of formulas. For a propo-
sitional variable p, we have
p ( p) (p p (p )) ( p p ) .
In case of the conjunction of formulas, we apply the induction hypotheses

and Item a) to obtain the following chain of equivalences:
( ) ( ( )) ( ( ))
( ) (( ) ( ) )
(( ( ) ) ) ( ( ( ) )) ( ) ( ) .
The case of disjunction can be considered similarly. For the implication, we
have
( ) ( ( )) ( ) ( ( ))
( ) ( ) ( ) ( )
( ) .
Thus, it remains to consider the case of negation, which is trivial in view of
the graphical equality ( ) ( ) = ( ) .
2
We now turn to the inverse implication. Assume N3 . Then there
is a twist-structure B, B |= N3, and its valuation v such that 1 v = 1.
By Lemma 8.3.7, B S (A), where A is an implicative lattice with the
least element 0. Since p0 p0 N4, we have 1 v(p0 p0 ) = 1. In view
of Lemma 8.3.7 2 v(p0 p0 ) = 0. In this way, v() = (0, 1). For any
propositional variable p, v(p) = (a, b), we have
v(p ) = v(p) v() = v(p) (0, 1) = v(p),
v(( p) ) = ( v(p) (v(p) (0, 1))) (0, 1) =

= (b, a) ((a, b) (0, 1)) = (b (a 0), a) = (b, a) = v( p).
Here we have taken into account the fact that a b = 0 is equivalent to
b a 0. Thus, if is a propositional variable or its negation, we have
v( ) = v(). Due to the fact that () preserve all logical connectives for
nnf, the relation v( ) = v() holds for any nnf . So for any formula ,
by Item 3 of Denition 8.3.8 we have v( ) = v(). By Proposition 8.1.1
we have N4 , whence 1 v( ) = 1 v() for any . In particular,
1 v( ) = 1, i.e., N4.
2. In the proof of the previous item, it was stated that for any formula
, for any twist-structure B, B |= N3, and any B-valuation v, 1 v( ) =
1 v(), which implies N3 .
2
8.4 N4-Lattices 145
We have proved that as well as in case of intuitionistic and minimal

logics deleting the explosion axiom from the logic N3 does not decrease
the expressive power of the logic. For this reason, we conclude this section
with the following problem: to nd a natural example of an explosive logic
such that it is not faithfully embeddable into its paraconsistent analog. An
example should be natural in a sense that the logic must be described in the
literature and not constructed especially to solve this problem.
8.4 N4-Lattices
In this section, we give an algebraic denition of twist-structures. Namely,
we describe the class of algebraic systems isomorphic to twist-structures.
The denition below is closely related to that of N -lattices (see [92, 33]).
Recall that in case of semantic investigations for N3, the notion of N -lattices
was rst introduced [90] and only later was it proved that N -lattices can be
represented as twist-structures [28, 110].
Definition 8.4.1 An algebra A =

A, , , , is said to be an N4-
lattice if the following hold.
1. The reduct
A, , , is a De Morgan algebra, i.e.,
A, , is a dis-
tributive lattice (non-bounded in general case) and the following iden-
tities hold: (p q) = p q and p = p.
2. The relation $, where a $ b denotes (a b) (a b) = a b, is a

preordering on A.
3. The relation , where a b if and only if a $ b and b $ a, is a

congruence relation with respect to , , and the quotient-algebra
A :=
A, , , / is an implicative lattice.
4. For any a, b A, (a b) a b.
5. For any a, b A, a b if and only if a $ b and b $ a, where is

a lattice ordering on A.
First we check that the twist-structures belong to the just dened class
of lattices.
Proposition 8.4.2 Every twist-structure B S (A) is an N4-lattice, more-

over, the following facts are true for all (a, b), (c, d) |B|:
a) (a, b) $ (c, d) if and only if a c;
b) (a, b) (c, d) if and only if a = c;
c) (a, b) (c, d) if and only if a c and d b;
d) The mapping [(a, b)] a determines an isomorphism of implicative

lattices B and A.
Proof. Check statements a)d).

a) We calculate (a, b) (c, d) = (a c, a d) and (a c, a d)
(a c, a d) = (1, (a c) a d) = (1, a c d). Thus, the inequality
(a, b) $ (c, d) is equivalent to the conditions a c = 1 and a d = a c d.
The rst condition means a c and implies the second.
Item b) immediately follows from a) and Item c) follows from the de-
nition of twist-operations.
d) In view of b) we have [(a, b)] = {(a, c) | (a, c) |B|}. Moreover,
twist-operations are componentwise with respect to the rst component and
1 (B) = A. These facts easily imply that [(a, b)] a determines the desired
isomorphism.
We now turn to the denition of N4-lattices. The fact that
|B|, , ,
is a De Morgan algebra easily follows from the denition of twist-operations.
Items 2 and 3 of the denition follows from b) and d), respectively. Item 4
can be veried directly, and 5 follows from a) and c).
2
Proposition 8.4.3 Every N4-lattice A is isomorphic to a twist-structure

over A .
Proof. Consider a mapping h from A to (A ) dened by the rule
h(a) := ([a] , [ a] ).
We claim that h is an isomorphic embedding. Indeed, for any a, b A, we

have
h(a b) = ([a b] , [ (a b)] ) = ([a] [b] , [ a] [ b] ) = h(a) h(b).

8.5 The Variety of N4-Lattices 147
We have used De Morgans law and the congruence properties of . That h

preserves can be checked similarly. Consider the case of implication:
h(a b) = ([a] [b] , [a] [ b] ) = h(a) h(b).
We have again used the congruence properties of and Item 4 of Denition

8.4.1. Finally, by identity p = p we have
h( a) = ([ a] , [ a] ) = ([ a] , [a] ) = h(a).
Now assume h(a) = h(b), which means that a b and a b. By

Item 5 of Denition 8.4.1 we immediately obtain a = b, which completes the
proof.
2
A
For an N4-lattice A, dene a set D := {a |A| | a a = a}. Let
N denote the class of all matrices of the form
A, DA , where A is an N4-
lattice, and let |=N be a consequence relation dened in a natural way by
the class N .
Lemma 8.4.4 Let A S (B). Then
D A = {(1, b) | (1, b) A}.
Proof. Calculate (a, b) (a, b) = (1, ab). Obviously, the equality (1, ab) =
(a, b) is equivalent to a = 1.
2
From this lemma and Theorem 8.3.6 we immediately obtain that N4 is
characterized by the class of matrices N .
Theorem 8.4.5 Let be a set of formulas and a formula. The following

equivalence holds:
N4 |=N .
8.5 The Variety of N4-Lattices

We have thus obtained a characterization of the logic N4 via the class
N of matrices. We now show that the underlying algebras of these matri-
ces, namely, N4-lattices, form a variety. The fact that sets of distinguished
elements of matrices in N are denable allows one to state the dual iso-
morphism between the lattice EN4 of logics extending N4 and the lattice
of subvarieties of the variety of N4-lattices. In this section, we will use
essentially the results of [89], where the class of implicative De Morgan lat-
tices was introduced and it was stated that this class provides an algebraic
semantics for Belnaps four-valued logic in the language
, , , . Our
axiomatization of the class of N4-lattices is obtained by weakening the ax-
iomatization of the class of implicative De Morgan lattices.
Definition 8.5.1 Let VN4 be a variety of algebras in the language L satis-

fying the identities of De Morgan algebras and the following identities:
1N . (p p) q = q
2N . (p q) r = p (q r)
3N . p (q r) = (p q) (p r)
4N . (p q) r = (p r) (q r)
5N . q p q
6N . (p q) (q r) $ p r
7N . p p (p q) p
8N . q q p ( q (p q))
9N . q q (p q) q
10N . p (p q) q ( q p)
11N . p (q q) ( q p)
Here, we consider p q as an abbreviation of the identity p q = q and

p $ q as an abbreviation for (p q) (p q) = p q.
Remark. Since algebras from VN4 are De Morgan algebras, for any
A VN4 and a, b A, holds
a b b a. (8.1)
Indeed, a b = (a b) = b.
As in the case of N4-lattices for an arbitrary algebra A VN4 , we dene
D A := {a |A| | a a = a}. First we state some properties of sets DA in
algebras of VN4 .
Lemma 8.5.2 Let A VN4 . For any a, b A, the following holds:
a) if a D A , then a b = b;
b) D A = {a a | a |A|};
c) if a DA and a $ b, then b DA ;
d) if a b, then a $ b;
e) if a D A and a b, then b DA ;
f ) if a, b D A , then a b DA .
g) a $ a b if and only if a $ b.
Proof. a) We have a b = (a a) b = b by identity 1N .

b) If a D A , then by denition a = a a. On the other hand, by 1N
we have (a a) (a a) = a a, whence a a DA .
c) By assumption (a b) (a b) = a b. On the other hand,
a D A implies by a) a b = b, whence b b = b.
d) By 2N we have (a b) (a b) = ((a b) a) b. The latter is
equal to a b, because a b by assumption and b a b by identity 5N .
e) It follows immediately from c) and d).
f) Applying identity 2N and Item a) we calculate (a b) (a b) =
a (b (a b)) = b (a b) = a b.
g) By denition a $ a b means a (a b) DA , which is equivalent
to (a a) (a b) D A . By Items e) and f) the latter is equivalent to
a a D A and a b DA . By b) a a DA , which completes the
proof.
2
Theorem 8.5.3 For any algebra A =

A, , , , , the following holds:
A VN4 A is an N4-lattice.
Proof. If A is an N4-lattice, then by Proposition 8.4.3 A can be isomorphi-

cally embedded into B for a suitable implicative lattice B. Checking that
the dening identities of VN4 are satised in A is reduced in this way to a
routine calculation with the help of twist-operations and the application of
well-known identities of implicative lattices. In case of identity 6N , we will
need also Proposition 8.4.2.a).
We now take an algebra A VN4 and check successively Items 15 of

Denition 8.4.1.
1. A is a De Morgan algebra by denition.
2. Let us check that the relation $ on A is reexive and transitive. Note
that for a, b A, a $ b if and only if a b DA . By Lemma 8.5.2 the
relation a $ a holds for any a A.
Take elements a, b, c A such that a $ b and b $ c. We then have
(a b) (b c) D A by Lemma 8.5.2.f). This fact and identity 6N imply
then a $ c. In this way, $ is transitive.
3. Check rst the congruence properties of the equivalence . Let a, b, c,
d A, a b and c d.
We have by Lemma 8.5.2.d) b (b d) D A , moreover, a b DA by
assumption. By Lemma 8.5.2.f) and identity 6N we obtain a (bd) DA .
In a similar way, from c d D A we infer c (b d) DA . Further,
applying Lemma 8.5.2.f) and identity 4N yields
(a c) (b d) = (a (b d)) (c (b d)) DA ,
i.e., a c $ b d. We omit the verication of the inverse inequality.

From the relations a b, c d D A one can conclude by 6N and
Lemma 8.5.2.f) that the implications (a c) b and (a c) d are in DA .
Now we need identity 3N to obtain (a c) (b d) DA .
Finally, consider the case of implication. We have (b a) (a c) $
(b c) by identity 6N , whence, (b a) ((a c) (b c)) DA by
2N . By assumption, b a DA , from which we infer by Lemma 8.5.2.c)
(a c) $ (b c). Similarly, the inequality (b c) $ (b d) can be
obtained from (b c) (c d) $ (b d) and the assumption c d DA .
By transitivity of $ we have (a c) $ (b d). Analogously, one can check
the inverse inequality.
We now show that A :=
A, , , / is an implicative lattice. A
is a distributive lattice as a quotient of a distributive lattice. Therefore, it
remains to prove that is a relative pseudo-complement operation on A ,
i.e., that for any a, b, c A, the following condition is met:
[c] [a b] if and only if [a] [c] [b] .
We can replace it by the condition:
c c (a b) if and only if a c b a c,
which is equivalent by Lemma 8.5.2.d) to:

c $ c (a b) if and only if a c $ a c b,
and by Lemma 8.5.2.g) to:
c $ (a b) if and only if a c $ b.
The latter immediately follows from the equality c (a b) = (a c) b,
which is true by 2N and the commutativity of .
4. In view of Items b) and e) of Lemma 8.5.2 and 2N , identities 7N 9N
can be rewritten as (p q) $ p, (p q) $ q, and p q $ (p q)
respectively. These facts together mean that the identity (p q) p
q holds on A .
5. Let a, b A. Assume a b, then the inequality b a follows
by (8.1). By Lemma 8.5.2.d) we have a $ b and b $ a.
We now assume that a $ b and b $ a and try to prove a b. The
idea of this proof is taken from the work [89] by A. Pynko, where it was
realized in another setting. But as we will see, it works also for VN4 . First
we state some new identities in essentially the same way as was done in [89]
for the class of implicative De Morgan lattices.
Lemma 8.5.4 The following identities hold in VN4 :
a) p p = p p;
b) (p p) q q;
c) q q (p q) (p q).
Proof. a) Take an algebra A VN4 and any a, b A. From 1N and 8N we

obtain
b b (a a) ( b ((a a) b)) = b b.
Similarly, one can state the inverse inequality b b b b.

b) By Lemma 8.5.2 we have (a a) (b b) D A . At the same time,
by 5N and a) for any c DA , holds c c c = c c = c. Hence,
(a a) (b b) (a a) (b b). Finally, (a a) b b.
c) We calculate
(a b) (a b) = (a b) ((a a) b) = [by 2N ]
= (a b) (a (a b)) = [by 2N ] (a (a b)) (a b) = [by 2N ]

= a ((a b) (a b)) = [by a)]

= a ( (a b) (a b)) = [by 5N ]
= a (( b (a b)) (a b)) = [by 2N ]
= (a b) (a ( b (a b))) [by 5N ]
a ( b (a b)) [by 8N ] b b.
2
Due to Item c) of the previous lemma conditions a $ b and b $ a
imply b b a b and a a b a. These inequalities
together with 10N and (8.1) yield
a (b b) a (a b) b ( b a) b ( a a).
The inequality a a b a, identity 11N and (8.1) allow us to

obtain
a (b b) ( a a).
From this fact and item b) of Lemma 8.5.4, we infer a b b. Substituting
in 11N b for p and a for q yields b a a, and so ( a
a) b by (8.1). Comparing this with the chain of inequalities above and
taking into account a b b we obtain a b.
2
This theorem allows one to state one more completeness result for N4.
Theorem 8.5.5 For any formula , the following equivalence holds:

N4 = Eq(VN4 ).
Proof. Due to Theorem 8.4.5 the fact N4 means that for any N4-lattice
A and valuation v, v() D A . By denition of DA the latter means exactly
that for any A VN4 , the identity = holds in A.
2
The latter characterization of the logic N4 is the most attractive, because
it can be extended to any extension of N4, moreover, we can establish a
dual isomorphism between the lattice EN4 and the lattice Sub(VN4 ) of all
subvarieties of VN4 .
For any L EN4, dene a variety
V ar(L) := {A | = Eq(A) for all L}.
Clearly, V ar(L) Sub(VN4 ). Moreover, the logic L is characterized by the

variety V ar(L).
Proposition 8.5.6 For any logic L EN4 and a formula , we have the
equivalence:
L = Eq(V ar(L)).
Proof. The direct implication is by denition of V ar(L). To state the in-
verse implication consider the canonical twist-structure L
L for the logic L
(see Section 8.3). Prove that L
L V ar(L). By denition
L | = {([], [ ]) | []},
|L
where [] is a coset of wrt the equivalence L . Taking into account that
[] = [] for [] we obtain
L | = {([], [ ]) | F or}.
|L
Let = (p1 , . . . , pn ) and an L L -valuation v be such that v(pi ) = ([i ],
[ i ]), i = 1, . . . , n. Using induction on the structure of formulas, we state
that
v() = ([(1 , . . . , n )], [ (1 , . . . , n )]).
In this way, if L, then for any L
L -valuation v we have 1 v() = 1, i.e.,
L
v() D . Consequently, = Eq(L
L
L ) and LL V ar(L).

Take a formula L and LL -valuation v such that v(p) := ([p], [ p]).

Then
v() = ([], [ ]).

By assumption L , consequently, [] = 1. This means that v() DLL ,
i.e., = Eq(L
L ).
2
Corollary 8.5.7 1. If L1 , L2 EN4, then
L1 L2 V ar(L2 ) V ar(L1 ).
2. The mapping V ar : EN4 Sub(VN4 ) is one-to-one.

Proof. 1. The direct implication immediately follows from the denition.
Assume that Eq(V ar(L1 )) Eq(V ar(L2 )) and take L1 . Then
= Eq(V ar(L2 )). By the previous proposition we have L2 .
2. This item follows from the previous one.
2
Now for any V Sub(VN4 ), dene a set of formulas
L(V ) := { | = Eq(V )}.
Proposition 8.5.8 1. For any V Sub(VN4 ), the set L(V ) is a logic

and L(V ) EN4.
2. The mappings V ar : EN4 Sub(VN4 ) and L : Sub(VN4 ) EN4 are
mutually inverse.
Proof. 1. It is clear that N4 L(V ) and that the set L(V ) is closed under
substitution. Assume , L(V ) and take an arbitrary N4-lattice
A V and a valuation v into A. By assumption v( ) = v() and
v(( ) ( )) = v( ), i.e., v(), v() v() DA . Hence,
due to Lemma 8.5.2 we have v() D A . The latter means that =
Eq(V ). Thus, L(V ) is closed under modus ponens and L(V ) EN4.
2. It follows from Proposition 8.5.6 that L(V ar(L)) = L for any L
EN4. Let us compare the varieties V and V ar(L(V )). It follows from the
denition that V ar(L(V )) V . To state the inverse inclusion we need the
following lemma.
Lemma 8.5.9 For any N4-lattice A and formulas and , we have =

Eq(A) if and only if
( ) ( ) = Eq(A).
Proof. Without loss of generality we may assume that A is a twist-structure

over a suitable implicative lattice B. It is enough to check the equivalence:
= Eq(A) if and only if v( ) DA for any A-valuation v. Due to
Lemma 8.4.4 the condition v( ) D A is equivalent to 1 v( ) = 1.
Let v be an arbitrary A-valuation. Assume v() = (a, b) and v() =
(c, d) for some a, b, c, d B. The direct computation shows that
1 v( ) = (a c) (b d).
It is obvious that (a c) (b d) = 1 if and only if a = c and b = d, i.e.,
v() = v().
2
For brevity we denote the identity = as N .
We now turn to the inclusion V V ar(L(V )). Let = Eq(V ).
By the previous lemma we have ( )N Eq(V ), then
L(V ) and ( )N Eq(V ar(L(V ))). Again by Lemma 8.5.9 =
Eq(V ar(L(V ))).
2
Theorem 8.5.10 The mapping V ar is a dual lattice isomorphism between

EN4 and Sub(VN4 ).
8.6 The Logic N4 and N4 -Lattices 155
Proof. By Item 2 of Proposition 8.5.8 the mapping V ar is a bijection. There-

fore, the fact that V ar is a dual lattice isomorphism follows from Corol-
lary 8.5.7.
8.6 The Logic N4 and N4 -Lattices

First, we consider when twist-structures over implicative lattices are
bounded.
Lemma 8.6.1 Let A be an implicative lattice and B S (A).
1. If A has the least element 0, then (0, 1), (1, 0) B and B is a bounded
lattice with zero element (0, 1) and unit element (1, 0).
2. If B is bounded, then A has the least element 0, zero element of B is

(0, 1) and unit element is (1, 0).
Proof. 1. By denition of twist-structures there are a, b A such that

(0, a), (1, b) B, then (0, 1) = (0, a) (b, 1) B and (1, 0) = (0, 1) B.
That (0, 1) is the least and (1, 0) the greatest element of B follows from the
denition of twist-operations.
2. Assume that (a, b) is the least element of B and c A. There is d A
such that (c, d) B. Then (a, b) (c, d), whence a c. Thus, a is the least
element of A.
2
This allows us to dene twist-structures over Heyting algebras as follows.
Definition 8.6.2 Let A =

A, , , , 0, 1 be a Heyting algebra.
1. A full twist-structure over A is an algebra
A =
A A, , , , ,
such that
A A, , , , is the full twist-structure over the im-
plicative lattice
A, , , , 1 and = (0, 1).
2. A twist-structure over A is an arbitrary subalgebra B of the full twist-

structure A such that 1 (B) = A .
3. The class of all twist-structures over A is denoted S (A).

Definition 8.6.3 An algebra A =

A, , , , , is an N4 -lattice if
A, , , , is an N4-lattice and is the least element of A. In this

case A :=
A, , , , / is a Heyting algebra, which we call a basic
Heyting algebra of A.
We can easily prove the equivalence of the last two denitions.
Proposition 8.6.4 Every twist-structure B S (A) over a Heyting al-

gebra A is an N4 -lattice. Every N4 -lattice A is isomorphic to a twist-
structure over A and an isomorphism is given by the rule
a ([a] , [ a] ).
2
The mapping i : A (A ) , i (a) = ([a] , [ a] ), is called a

canonical embedding of an N4 -lattice into the full twist-structure over the
basic Heyting algebra.
The validity of formulas in twist-structures over Heyting algebras and in
N4 -lattices is dened in the same way as for twist-structures over implica-
tive lattices and for N4-lattices. Note that although an N4 -lattice A has
the greatest element, its set of distinguished values is D A = {a a | a A}
and it is not one-element in the general case.
To prove that N4 is complete with respect to the class of twist-struc-
tures over Heyting algebras we have only to note that LN4 is a Heyting
algebra due to axiom A14. Consequently, the Lindenbaum twist-structure
L
N4 is a twist-structure over a Heyting algebra.
Theorem 8.6.5 For any formula , N4 if and only if A |= for any

N4 -lattice A.
2
Taking into account that the action of twist-operations on the rst com-
ponent agreed with the operations of the basic Heyting algebra, we easily
obtain
Corollary 8.6.6 N4 is a conservative extension of intuitionistic logic.
2
Since N4 -lattices are exactly bounded N4-lattices, they also form a va-
riety determined by the identities of de Morgan algebras, the set of identities
1N 11N and the new identity:
12N . p
8.6 The Logic N4 and N4 -Lattices 157
The mappings V ar : EN4 Sub(VN4 ) and L : Sub(VN4 ) EN4

are dened exactly as for the logic N4 and the variety of N4-lattices.
Repeating the reasoning from the previous section we prove
Theorem 8.6.7 The mappings V ar and L are mutually inverse dual lattice
isomorphisms between EN4 and Sub(VN4 ).
2
Chapter 9
N4-Lattices1
In the previous chapter, we considered two variants of Nelsons paraconsis-

tent logic, N4 and N4 , which were dened in dierent languages. N4 is a
conservative extension of N4 in the language with the additional constant
allowing us to dene in this logic the intuitionistic negation. The addition of
this constant results in an easy modication of the semantics. Models of N4
are isomorphic to twist-structures over implicative lattices and due to the
fact that implicative lattices do not necessarily have the least element and
the lattices modelling N4 do not have the greatest element either. In fact,
no constant can be naturally dened in N4. Extending the language with
we obtain the class of models isomorphic to twist-structures over Heyting
algebras, i.e., implicative lattices with the least element 0. A twist-structure
over bounded lattice is also bounded, it contains necessarily elements (0, 1)
and (1, 0), which are the least and the greatest elements. Thus, the seman-
tics for N4 is given by the class of bounded lattices. It turns out that
the introduction of has essential consequences for the class of extensions
EN4 . As we will see in Chapter 10, adding the constant enriches the
class of N4 -extensions as compared to EN4, and, which is more important,
provides it with a regular structure close to some extent to the structure of
the class of extensions of minimal logic studied in the rst part of the book.
This is why we will work mainly with the logic N4 and with N4 -lattices.
However, all results in this chapter remains true if we replace N4 -lattice
with N4-lattice and Heyting algebra with implicative lattice.
Our investigation of N4 -lattices is based essentially on the represen-
tation of N -lattices obtained by A. Sendlewski [100]. N -lattices are exactly
1
Parts of this chapter were originally published in [74] (Springer, Netherlands).
159
160 9 N4 -Lattices
N4 -lattices modelling N3. It is well known (see, [28, 110, 47]) that
N -lattices can be represented as twist-structures over Heyting algebras sat-
isfying the condition ab = 0 for any (a,b). We call such twist-structures ex-
plosive. D. Vakarelov [110] suggested the following intuitive interpretation of
an explosive twist-structure: the underlying Heyting algebra A represents a
set of assertions of an intuitionistic theory, i.e., it is a Lindenbaum algebra of
this theory. Let a and b be assertions. We say that b is a counter-example of a
if ab is contradictory. For intuitionistic theory, the latter means exactly that
ab = 0. Thus, an explosive twist-structure is a set of statements of an intu-
itionistic theory together with their counter-examples. The twist-operations
determine how to construct a counter-example of a complex statement from
counter-examples of components. For instance, a counter-example of a con-
junction of two assertions is a disjunction of counter-examples of these as-
sertions, (a, b) (c, d) = (a b, c d). Any statement can be considered
as a counter-example of its counter-example, therefore, (a, b) = (b, a).
Of course, any assertion has a counter-example, therefore, for any twist-
structure B over A, 1 (B) = A.
To represent arbitrary N4 -lattices we need arbitrary, not only explosive
twist-structures over Heyting algebras. Therefore, the condition a b = 0
should be omitted and the Vakarelov interpretation should be modied. One
of the possible solutions to this problem is as follows: we consider the notion
of a counter-example as a primary notion, independent of the notion of
contradiction in the underlying intuitionistic theory. As above, we consider
a twist-structure B over A as some set of pairs (a, b) consisting of an assertion
a A and its counter-example b. Of course, this set should be closed with
respect to twist-operations. The contradictory assertion is now dened as an
assertion of the form a b for (a, b) B, i.e., as a conjunction of an assertion
and its counter-example.
A. Sendlewski [100] proved that an explosive twist-structure B over a
Heyting algebra A is completely determined by a lter on A consisting of
all elements of the form a b, where (a, b) B. On the other hand, for
any lter over A containing the lter of dense elements, there exists an
explosive twist-structure B over A such that = {a b | (a, b) B}. In
what follows, we call a lter of completions of the twist-structure B.
In this chapter, we generalize Sendlewskis characterization to arbitrary
twist-structures over Heyting algebras. An explosive twist-structure is char-
acterized by conditions on the join and meet of an assertion and its counter-
example, i.e., on the join and meet of the rst and second components of its
elements. The join belongs to some xed lter of the basic Heyting algebra
and the meet is contradictory. The second condition means the same for
9.1 Structure of N4 -Lattices 161
any explosive twist-structure, namely, to be the least element of the basic

Heyting algebra. But in case of arbitrary twist-structures, this condition
may be varied. This leads to the hypothesis that a twist-structure B over a
Heyting algebra A is determined by this algebra, by the set of possible joins
of assertions and their counter-examples, which plausibly forms a lter on A
as well as in Sendlewskis characterization, and by the set of contradictions,
i.e., elements of A of the form a b for (a, b) B. It turns out that this set
forms an ideal on the basic Heyting algebra, which we call an ideal of con-
tradictions. One may assume that there is also some interrelation between
the ideal of contradictions and the lter of completions but as we will see
below, it is not the case.
We prove the main result of the chapter on the representation of twist-
structures in Section 9.1. Moreover, we formulate and prove the embeddabil-
ity criterium for N4 -lattices in terms of the above representation.
In Section 9.2, we deal with homomorphisms of N4 -lattices. In particu-
lar, we dene special lters of the rst kind (sk) on N4 -lattices and prove
that sk are exactly kernels of homomorphisms of N4 -latices. Moreover, it
is proved that sk on an N4 -lattice A are in a one-to-one correspondence
with lters on its basic Heyting algebra, which implies the isomorphism of
congruence lattices Con(A) = Con(A ). The latter fact allows one to ob-
tain a characterization of subdirectly irreducible N4 -lattices. Namely, an
N4 -lattice is subdirectly irreducible if and only if its basic Heyting al-
gebra is subdirectly irreducible. A similar result on subdirectly irreducible
N -lattices was proved in [99] and independently in [17], the proof was based
on the Priestley duality.
The results of this and the previous chapters constitute the necessary
semantic basis for the investigation of the lattice of logics EN4 .
9.1 Structure of N4 -Lattices

In this section, we obtain a characterization of arbitrary twist-structures
similar to that of A. Sendlewski [100].
First, we note that the condition a b = 0 can be replaced by the
condition a b , where is an arbitrary ideal of the basic Heyting
algebra of a twist-structure.
Recall that an ideal on a Heyting algebra (implicative lattice) A is a
non-empty subset X A such that: 1) if a, b X, then a b X; 2) if
a X and b a, then b X. The set of all ideals on A is denoted by I(A).
162 9 N4 -Lattices
Proposition 9.1.1 Let A be a Heyting algebra, be a lter on A such that

Fd (A), and let be an ideal on A. Then there exists a twist-structure
T w(A, , ) S (A) with the universe
|T w(A, , )| = {(a, b) | a, b |A|, a b , a b }.
Proof. Check that the set T := |T w(A, , )| is closed under twist op-
erations. Let (a, b), (c, d) |A| |A| be such that a b, c d and
a b, c d .
For the disjunction (a, b) (c, d) = (a c, b d), we have (a c)
(b d) = (a c b) (a c d). Both conjunction terms in the latter
expression belong to , and so (a c) (b d) . In a similar way, we
have (a c) (b d) = (a b d) (c b d) . Thus, (a, b) (c, d) T .
The case of conjunction can be considered similarly. Let us consider the
implication (a, b) (c, d) = (a c, a d). We have (a c) (a d) =
((a c) a) ((a c) d). The element (a c) a belongs to ,
because it is dense. At the same time, c d (a c) d . In this way,
(a c) (a d) . Further, (a c) (a d) = a c d. This element
belongs to by the condition c d . We have proved (a, b) (c, d) T .
It is obvious that (a, b) = (b, a) T .
2
It turns out that any twist-structure can be represented in the form
described in the last proposition.
Proposition 9.1.2 Let A be a Heyting algebra and B S (A). We dene
I(B) := {a a | a B}, (B) := 1 (I(B)), (B) := 2 (I(B)).
Then Fd (A) (B), (B) is a lter on A, and (B) is an ideal on A.

Moreover,
B = T w(A, (B), (B)).
Proof. First, we obtain a more convenient denition for I(B) and a repre-
sentation of B via I(B).
Lemma 9.1.3 Let A be a Heyting algebra and B S (A). Then
|B| = {(a, b) | a, b A, (a b, a b) I(B)},
I(B) = {(a b, a b) | (a, b) B} = {(a, b) | (a, b) B, a b}.

Proof. Let us check the rst equation. If (a, b) B, then (a b, a b) =

(a, b) (b, a) = (a, b) (a, b) I(B).
To check the inverse inclusion take an element (a, b) |A| |A| such
that (a b, a b) I(B) and prove that (a, b) B. By Denition 8.6.2 there
is an element c A such that (a, c) B. We have I(B) B by denition,
therefore, (a b, a b) = (a b, a b) B. Further,
(a, c) (a b, a b) = (a (a b), a (a b)) = (a b, a) B.
And nally, we have
(a b, a b) (a b, a) = ((a (a b)) (b (a b)), a) =
= ((a b) b, a) = (b, a).

Thus, the elements (b, a) and (a, b) = (b, a) are in B.
The representation for I(B) immediately follows from the facts that for
any (a, b) B, (a, b) (a, b) = (a b, a b) , and (a, b) (a, b) = (a, b)
whenever a b.
2
This allows us to conclude that a twist-structure B is completely deter-
mined by the set I(B).
Corollary 9.1.4 Let B1 and B2 be twist-structures over a Heyting algebra

A. Assume I(B1 ) = I(B2 ). Then B1 = B2 .
2
We now turn to the proof of the proposition and show that (B) is an
ideal of A. We need one more lemma.
Lemma 9.1.5 (B) = {a A | (1, a) B} = 2 (DB ).
Proof. If (1, a) B, then (1, a) (1, a) = (1, a) (a, 1) = (1, a), i.e.,
(1, a) I(B) and a (B). Conversely, let a (B). This means that
there is a d = (b, c) B such that 2 (d d) = a, i.e., a = b c. Further,
(b, c) (b, c) = (1, b c) = (1, a) B.
2
Let a, b (B). In view of the previous lemma (1, a), (1, b) B, whence
(1, a) (1, b) = (1, a b) B, and so a b (B).
Now assume that a (B) and b A, b a. Let c A be such that
(c, b) B. We have (1, a) (c, b) = (1, a b) = (1, b) B, which means that
b (B).
164 9 N4 -Lattices
We have thus proved that (B) is an ideal. Let us check that (B) is a
lter on A and Fd (A) (B).
Let a, b (B) and elements c, d (B) are such that (a, c), (b, d)
I(B). In view of Lemma 9.1.3 we have c a, d b. The fact that (B)
is an ideal and Lemma 9.1.5 imply (1, c d) I(B), whence the element
(a, c) (c d, 1) = (a (c d), a) is in B. We have also (a (c
d), a) (a, c) = (a c d, a c) = (c d, a) B. From this fact and
Lemma 9.1.3 we have (a, c d) I(B). Similarly, (b, c d) I(B). Further,
(a, cd)(b, cd) = (ab, cd) I(B), from which we conclude ab (B).
Assume a (B), b A, and a b. In this case, there exist elements
c, d A such that a c, (a, c) I(B), and (b, d) B. We have then
(ab, cd) = (b, cd) B. Moreover, cd a b, whence (b, cd) I(B),
i.e., b (B).
Thus, (B) is a lter. Let a, b A be such that (a, b) B. Then
(a, b) ((a, b) (0, 1)) = (a (a 0), a b) B.
We have a b a a (a 0), consequently, a (a 0) (B) and we

conclude that Fd (A) (B).
It remains to prove the equality B = T w(A, (B), (B)). As it follows
directly from denitions (a, b) T w(A, (B), (B)) if and only if (a b,
a b) I(B). In view of Lemma 9.1.3, the latter fact means that (a, b) B.
2

This proposition can be generalized to arbitrary N4 -lattices as follows.
Corollary 9.1.6 Let A =

A, , , , , be an N4 -lattice. We dene
l (A) := e ({a a | a A}), l (A) := e ({a a | a A}),
where e :
A, , , , A is a canonical epimorphism. Then l (A) is
a lter on A , Fd (A ) l (A), and l (A) is an ideal on A . Moreover,
we have i (A) = T w(A , l (A), l (A)), where i : A (A ) is a
canonical embedding, i.e.,
A
= T w(A , l (A), l (A)).
Proof. Let us consider a twist-structure A := i (A). Clearly, A S (A )

and it has the form A = T w(A , (A ), (A )) by the previous proposition.
In this way, the conclusion of the corollary will follow from the equations
l (A) = (A ) and l (A) = (A ), which we prove now.
By Proposition 8.4.3 each element b A is of the form b = (e (a),

e ( a)) for some a A. Thus,
b b = (e (a), e ( a)) (e ( a), e (a)) = (e (a a), e (a a)).
We have in this way the equalities:
(A ) = 1 (I(A )) = {e (a a) | a A} = l (A),
(A ) = 2 (I(A )) = {e (a a) | a A} = l (A).
2
Remark. If B S (A) and h : B A is an isomorphism given by
the rule h([(a, b)] ) = a (see Proposition 8.4.2), then
h(l (B)) = (B) and h(l (B)) = (B).
For this reason we omit later on the lower index l in the denotation
l (A) and l (A).
As mentioned, all results in this chapter also hold true for N4-lattices. To
formulate the above results for twist-structures over implicative lattices, we
have to dene a lter of dense elements for an arbitrary implicative lattice.
Let A be an implicative lattice. We dene
Fd (A) :=
{a (a b) | a, b A}
and call Fd (A) a lter of dense elements of A. If A has the least element, the
lter Fd (A) will coincide with the lter of dense elements of A considered as
a Heyting algebra. This explains why we use the same name and denotation.
It turns out that any element of Fd (A) has the form a (a b).
Proposition 9.1.7 For any implicative lattice A,
Fd (A) = {a (a b) | a, b A}
Proof. Let a F an , b1 , . . . ,bn A
d (A). Then one can nd elements a1 , . . . ,
such that a ni=1 ai (ai bi ). For the element d = ni=1 ai ni=1 bi ,
consider the implicative lattice Ad , a sublattice of A with the universe {a
A | a d}. It is clear that a Fd (Ad ) Fd (A). On the other hand,
Ad has the least element and the lter Fd (Ad ) coincides with the lter of
dense elements of the Heyting algebra Ad . Therefore, for some c Ad ,
a = c c = c (c d). We have thus obtained the desired representation
for an arbitrary element of Fd (A).
2
Analogs of Propositions 9.1.1 and 9.1.2 can be proved for twist-structures
over implicative lattices in essentially the same way.
166 9 N4 -Lattices
Proposition 9.1.8 Let A be an implicative lattice, be a lter on A such

that Fd (A), and let be an ideal on A. Then there exists a twist-
structure T w(A, , ) S (A) with the universe
|T w(A, , )| = {(a, b) | a, b |A|, a b , a b }.
Proposition 9.1.9 Let A be an implicative lattice and B S (A). We

dene
I(B) := {a a | a B}, (B) := 1 (I(B)), (B) := 2 (I(B)).
Then Fd (A) (B), (B) is a lter on A, and (B) is an ideal on A.

Moreover,
B = T w(A, (B), (B)).
An analog of Corollary 9.1.6 also holds. In the following we will formulate

results only for twist-structures over Heyting algebras and for N4 -lattices.
At the end of this section, we make some remarks on subalgebras of
twist-structures.
Proposition 9.1.10 1. Let B S (A). Then B = A if and only if

(B) = (B) = A.
2. Let A0 and A1 be Heyting algebras, i F(Ai ), Fd (Ai ) i , and

i I(Ai ), i = 0, 1. Moreover, let A0 A1 ,0 1 , and 0 1 .
Then
T w(A0 , 0 , 0 ) T w(A1 , 1 , 1 ).
3. Let B S (A) and C B. Then C = T w(A1 , , ), where A1 A,

F(A1 ) and (B), I(A1 ) and (B).
4. Let B S (A) and A1 A. There exists a subalgebra C of B such

that C S (A1 ).
5. Let B S (A) and A1 A. The greatest subalgebra C of B satisfying

the condition C S (A1 ) has the following form
C = T w(A1 , (B) A1 , (B) A1 ).

9.2 Homomorphisms and Subdirectly Irreducible N4 -Lattices 167
Proof. 1. Let B = A , i.e., |B| = |A| |A|. Then
I(B) = {(a, b) | a b, a, b A}.
This fact and the denition of (B) and (B) as projections of I(B) onto
the rst and the second coordinates, respectively, imply (B) = (B) = A.
If (B) = (B) = A, then
|B| = {(a, b) | a b (B), a b (B)} = {(a, b) | a, b A}.
2. This item follows immediately from the denition of twist-structures

T w(Ai , i , i ), i = 0, 1.
3. The operations , , and are componentwise with respect to the
rst coordinate, therefore, A1 := 1 (C) = 2 (C) is a subalgebra of A. In this
case C S (A1 ). Obviously, I(C) = {a a | a C} I(B), from which
we have (C) = 1 (I(C)) 1 (I(B)) = (B). Similarly, (C) (B).
4. Note that Fd (A1 ) Fd (A), therefore, Fd (A1 ) (B) A1 . Moreover,
0 (B) A1 = . Since (B) A1 is obviously a lter and (B) A1 an
ideal on C, we can consider a twist-structure C = T w(A1 , (B) A1 , (B)
A1 ). That C B follows from Item 2.
5. As was noted above, C is a subalgebra of B. By Item 2 it is the greatest
subalgebra with the condition C S (A1 ).
2
9.2 Homomorphisms and Subdirectly

Irreducible N4 -Lattices
In Denition 8.6.2 for any Heyting algebra A, there was dened the N4 -
lattice A , the full twist-structure over A. At the same time, for any N4 -
lattice A, there was dened the Heyting algebra A . We extend these map-
pings to homomorphisms of Heyting algebras and of N4 -lattices. For any
Heyting algebra homomorphism h : A B, the homomorphism of N4 -
lattices h : A B is dened by the rule: h (a, b) := (h(a), h(b)). That
this is really a homomorphism follows immediately from the denition of
twist-structures.
Now let A and B be N4 -lattices. For a homomorphism h : A B, we
dene a mapping h : A B by the rule h ([a] ) := [h(a)] . The facts
that h is a homomorphism and that the relation is denable via basic
operations of an N4 -lattice guarantee that the above denition is correct.
Due to the congruence properties of , h is a homomorphism of Heyting
168 9 N4 -Lattices
algebras. It turns out that we have dened a functor from the category
H of Heyting algebras and their homomorphisms to the category N 4 of
N4 -lattices and their homomorphisms and the adjoint functor from N 4
to H.
Theorem 9.2.1 2 The mapping () is a covariant functor from the category
H to the category N 4 . The mapping () is a covariant functor from the
category N 4 to the category H. The functor () is left adjoint to () .
Proof. One can directly verify that () and () are covariant functors. We
prove only that () is left adjoint to () . To this end, it will be enough to

show (see e.g. [12]) that the family of canonical embeddings i B, B N4 ,
determines a natural transformation of an identical functor 1N 4 to the
composition (() ) such that for any B N 4 , the pair (B , i B ) is a
reection of B along () . The latter means that for any Heyting algebra
A and homomorphism f : B A there exists a unique homomorphism
g : B A such that f = g i B.
First, take a homomorphism f : B C of N4 -lattice and check the
C f = (f ) iC . For any b B, we have
equality i
C f (b) = ([f (b)]C , [ f (b)]C ) = ([f (b)]C , [f ( b)]C ) =

i
= (f ([b]B ), f ([ b]B )) = (f ) ([b]B , [ b]B ) = (f ) i
C (b).

We have thus established that the family i B , B N 4 , determines the
desired natural transformation.
Further, for a Heyting algebra A and a homomorphism f : B A we
dene a mapping g : B A by the rule g([a]B ) := 1 f (a) for all a B.
This denition is correct due to the following observation. If a, b B and
a B b, then f (a) A f (b), from which we obtain 1 f (a) = 1 f (b) by
Proposition 8.4.2. That g is a Heyting algebra homomorphism follows from
the congruence properties of B and the denition of positive operations on
A . For b B, we calculate
g i
B (b) = (g([b]B ), g([ b]B )) = (1 f (b), 1 f ( b)) =
= (1 f (b), 1 ( f (b))) = (1 f (b), 2 f (b)) = f (b).

2
This is the unique place where we use the special notions from the category theory.
For this reason, we do not give all the necessary definitions and ask the reader to consult,
e.g., [12]. Note that categorical properties of N3-lattices were studied by A. Sendlewski
[100] and R. Cignoli [17].
On the other hand, for a homomorphism h : B A, the equality f =

h i
B immediately implies h([a]B ) := 1 f (a) for a B. Thus, g is a
unique homomorphism with the desired property.
2
For Heyting algebras and N -lattices, interrelations of this kind were
proved by V. Goranko [33] and A. Sendlewski [99, 100].
We note also that every homomorphism h of N4 -lattices is uniquely
determined by the homomorphism h of basic Heyting algebras.
Proposition 9.2.2 Let A and B be N4 -lattices, and let hi : A B,

i = 1, 2, be homomorphisms such that h1 = h2 . Then h1 = h2 .
Proof. Let gi := (hi ) , i = 1, 2. Since h1 = h2 , we have g1 = g2 . For any
a, b A,
gi ([a]A , [b]A ) = (hi ([a]A ), hi ([a]A )) = ([hi (a)]B , [hi (b)]B ).
Further, for a A, we have

1 1
hi (a) = (i i i i i
B ) ([h (a)]B , [ h (a)]B ) = (iB ) ([h (a)]B , [h ( a)]B ).
In view of the above calculation, we obtain

1
hi (a) = (i
B) gi ([a]A , [ a]A ).
In this way, the equality h1 = h2 follows from g1 = g2 .

2
We now introduce the notion of a special lter of the rst kind on an
N4 -lattice in exactly the same way as a special lter of the rst kind on
an N -lattice was dened by H. Rasiowa [92]. Let A be an N4 -lattice. A
non-empty subset = A is called a special lter of the rst kind (sk)
on A if: 1) a and b imply a b ; 2) a and a $ b imply
b . It is obvious that the set of all sk on A forms a lattice, which we
denote F 1 (A). For a homomorphism h : A B of N4 -lattices we dene
a kernel Ker(h) := h1 (DB ). As we can see from the following proposition,
sk are in one-to-one correspondence with congruences on an N4 -lattice
and the least sk on A coincides with D A .
Proposition 9.2.3 1. For any N4 -lattice A, DA is an sk.
2. Let h : A B be an epimorphism of N4 -lattices. Then Ker(h) is

an sk. For any a, b A, h(a) = h(b) if and only if a b Ker(h).
170 9 N4 -Lattices
3. Let A be an N4 -lattice and an sk on A. Then DA . The

relation , a b i a b , is a congruence relation on A and
= Ker(h), where h : A A/ is a canonical epimorphism.
Proof. 1. This item follows from the lemma below.
Lemma 9.2.4 Let A be an N4 -lattice. Then the following holds:
1. If a, b D A , then a b DA .
2. If a A and b D A , then a $ b;
3. For any a, b A, a = b if and only if a b DA .
4. For any a, b A, a (a b) $ b.
Proof. Due to Proposition 8.4.3 we can identify A with a twist-structure over

A , in which case DA = {(1, a) | a A } and (a, b) $ (c, d) is equivalent
to a b. The latter two facts easily imply all assertions of the lemma.
2
B
2. By the previous item D is an sk, moreover, h preserves the relation
$ as denable via basic operations. These facts imply that Ker(h) is an
sk. By Lemma 9.2.4 the equality h(a) = h(b) is equivalent to h(a b) =
h(a) h(b) D B , i.e., a b Ker(h).
3. The relation DA follows from Item 2 of Lemma 9.2.4. Recall that
the strong equivalence has congruence properties in N4 . This means,
in particular, that the formula (p q) ((r t) ((p r) (q t))) is
provable in N4 . By Theorem 8.4.5 for any a, b, c, d A,
(a b) ((c d) ((a c) (b d))) DA .
If we assume a b, c d , we may conclude by properties of sk that

(a c) (b d) . Other congruence properties can be checked similarly.
Let [a] be a coset of a with respect to . Assume that a Ker(h),
i.e., [a] D A/ . This is equivalent to [a a] = [a], i.e., a a a or
(a a) a . Since a b implies a $ b, the relation (a a) a
implies (a a) a . By a a D A and Item 4 of Lemma 9.2.4
we obtain a .
Conversely, let a . By identity 5N (see Denition 8.5.1), we have
(a a) a . It follows from a a DA and 5N that a (a
a) . At the same time,
a (a a) a (a a) a a.
Since a , we have a a and so, a (a a) . Finally,
(a a) a ( a a) a DA .
Whence (a a) a . In this way, (a a) a and

a Ker(h). We have thus proved the equality = Ker(h).
2
Corollary 9.2.5 Let A be an N4 -lattice and = A. The set is

an sk on A if and only if: 1) a, b implies a b ; 2) a, a b
implies b .
Proof. The direct implication follows from Item 4 of Lemma 9.2.4. The in-
verse implication follows from the fact that DA is the least sk on A. Indeed,
if a and a $ b, then a b D A .
2

In this way, sk are exactly implicative lters on N4 -lattices.
Corollary 9.2.6 For any N4 -lattice A, F 1 (A) is isomorphic to Con(A)

and mutually inverse isomorphisms are dened by the rules: for
1
F (A) and {a A | a(a a)} for Con(A).
2
Corollary 9.2.7 Let h : A B be a homomorphism of N4 -lattices. Then

h is a monomorphism if and only if Ker(h) = D A .
Proof. Assume h : A B is a monomorphism and a DA , i.e., a a = a,

then h(a) h(a) = h(a). The equality Ker(h) = D A now follows from the
fact that D A is the least sk on A.
If Ker(h) = DA and h(a) = h(b) for some a, b A, then a b DA by
Item 2 of Proposition 9.2.3 and we conclude that a = b by Item 3 of Lemma
9.2.4.
2
As we can see from Proposition 9.2.2, homomorphisms of N4 -lattices
are completely determined by homomorphisms of basic Heyting algebras.
This leads to a suggestion that there is a close connection between sk on
an N4 -lattice and lters on the basic Heyting algebra. This is really so and
we consider rst the case of twist-structures.
172 9 N4 -Lattices
Proposition 9.2.8 Let B S (A).
1. For any F(A), := 11 () = {(a, b) |A| |A| | a } is

an sk on B.
2. For any F 1 (B), := 1 () is a lter on A. Moreover, =

{(a, b) |A| |A| | a } = ( ) .
3. The lattices F(A) and F 1 (B) are isomorphic and the mappings
, F 1 (B) determine mutually inverse
, F(A) and
isomorphisms.
Proof. Items 1 and 2 can be proved via direct verication, whereas Item 3
is an immediate consequence of them.
2
Recall that due to Proposition 8.4.3 and Denition 8.6.3 any N4 -lattice
A is isomorphic to a twist-structure over A and the canonical epimorphism
e : A A corresponds to the projection 1 of the twist-structure onto
the rst coordinate. In this way, we generalize the previous statement to all
N4 -lattices.
Proposition 9.2.9 Let A be an N4 -lattice.
1. For any F(A ), := (e )1 () is an sk on A.
2. For any F 1 (A), := e () is a lter on A . Moreover, =

( ) .
3. The lattices F(A ) and F 1 (A) are isomorphic and the mappings
, F(A ) and , F 1 (A) determine mutually
inverse isomorphisms.
2
Taking into account the dualities of congruences and lters on Heyting

algebras and of congruences and sk on N4 -lattices, we obtain the following
Corollary 9.2.10 For any N4 -lattice A, Con(A)

= Con(A ).
2
The property of an algebra to be subdirectly irreducible is determined by

the structure of its congruence lattice. Combining this fact with the previous
corollary, we obtain
Corollary 9.2.11 An N4 -lattice A is subdirectly irreducible if and only if

the Heyting algebra A is subdirectly irreducible.
2
The above description of the congruence lattice of an N4 -lattice allows

one to prove that the lattice of N4 -extensions is distributive.
Proposition 9.2.12 VN4 is an arithmetic variety.
Proof. The congruence distributivity of the variety of Heyting algebras and

Corollary 9.2.10 allows one to conclude that the variety of N4 -lattices is
congruence distributive. Prove that VN4 is congruence permutable.
Let A be a Heyting algebra and B S (A). Consider some congruences
1 , 2 Con(B) determined by sk F1 and F2 , where F1 , F2 F(A), i.e.,
((a, b), (c, d)) i i (a, b) (c, d) Fi , i = 1, 2.
The denition of lters Fi and calculation with the help of twist-operations
allows one to conclude that (a, b) (c, d) Fi is equivalent to (a
c) (b d) Fi . In this way,
((a, b), (c, d)) i i (a, c), (b, d) Fi , i = 1, 2,
where Fi is a congruence on A determined by Fi . This equivalence easily

implies that
((a, b), (c, d)) 1 2 i (a, c), (b, d) F1 F2 .
The permutability of congruences on A imply that 1 and 2 are now per-

mutable.
2
Proposition 9.2.13 The lattice EN4 is distributive.
Proof. This reasoning is similar to Corollary 2.2.11 but contains several

peculiarities.
Consider a free -generated N4 -lattice A and its congruence lattice
Con(A ), which is distributive according to the previous proposition. More-
over, the congruences of Con(A ) are permutable wrt composition. In this
case, the elements of A can be identied with cosets of formulas wrt the
strong equivalence in N4 ,
|A | = {[] | F},
174 9 N4 -Lattices
where [] := { | N4 }. Each logic L EN4 corresponds to

the congruence
L := {([0 ] , [1 ] ) | 0 1 L}.
Clearly, the mapping L L is one-to-one and preserves the ordering. Check

that this is a lattice embedding. First, we verify that for any L0 , L1 EN4 ,
the congruences L0 L1 and L0 L1 are of the form L for a suitable
logic L. Note that L is closed under substitution, i.e., [0 ] L [1 ] implies
[0 (1 , . . . , n )] L [1 ((1 , . . . , n ))] for any formulas 1 , . . . , n .
Let Con(A ) be closed under substitution. Put L := { | [] [
] }. Check that L is a logic from EN4 . Obviously, the set L is closed
under substitution. If N4 , it is not hard to prove that ( )
N4 . Therefore, N4 L . It remains to check that L is closed under
modus ponens. According to Corollary 9.2.6 F = {[] | [] [ ] } is
an sk on A . If , L , then [] , [ ] = [] [] F .
From Corollary 9.2.5 we obtain [] F , i.e., L .
Check the equality = L . It can easily be seen that this equality is
equivalent to the following relation
([] , [] ) ([ ] , [( ) ( )] )
for any and , which is equivalent in turn to
[ ] F [ ] [( ) ( )] F .
The latter is a particular case of the relation
aF a (a a) F,
which was established in the proof of Item 3 of Proposition 9.2.3 for any sk
F on any N4 -lattice.
Thus, we have proved that the congruences of the form L are exactly
the congruences closed under substitution.
Check that L0 L1 and L0 L1 are closed under substitution. Con-
sider the non-trivial case of L0 L1 . Since the lattice A is congruence
permutable, L0 L1 = L0 L1 . Therefore [0 ] L0 L1 [1 ] i there
is a formula such that [0 ] L0 [] and [] L1 [1 ] . This equivalence
immediately implies that L0 L1 is closed under substitution.
We have proved that the set of congruences L forms a lattice. It is clear
that the mapping L L is an order isomorphism of EN4 and the lattice
of congruences of the form L . If two lattices are isomorphic as orderings,

they are isomorphic as lattices too.
2
We now prove the criterion when there exists a homomorphism from one
N4 -lattice into the other, which generalizes the embeddability criterion
from the previous section. But rst, we point out some facts on interrelations
of homomorphisms h and h .
Proposition 9.2.14 Let h : A B be a homomorphism of N4 -lattices.
1. Ker(h ) = (Ker(h)) and Ker(h) = (Ker(h )) .
2. h is a monomorphism if and only if h is a monomorphism.
Proof. 1. By denition (Ker(h)) = {[a]A | h(a) DB }. The condition

h(a) D B is equivalent to h ([a]A ) = [h(a)]B = 1B . From this equiv-
alence we immediately obtain Ker(h ) = (Ker(h)) . The second equality
immediately follows from the rst one and Proposition 9.2.9.
2. If h is a monomorphism, then Ker(h) = D A by Corollary 9.2.7. Tak-
ing into account that (D A ) = {1A } we infer from the previous item
Ker(h ) = {1A }, i.e., h is a monomorphism. The inverse implication is
similar.
2
Proposition 9.2.15 Let A and B be N4 -lattices. There exists a homo-

morphism (monomorphism) h : A B if and only if there exists a homo-
morphism (monomorphism) g : A B such that g((A)) (B) and
g((A)) (B).
Proof. If h : A B is a homomorphism, then h can be chosen as g. Indeed,

for [a a]A (A), we have
h ([a a]A ) = [h(a a)]B = [h(a) h(a)]B (B).
Analogously, h ((A)) (B).

If h is additionally a monomorphism, then h is a monomorphism by
Item 2 of the previous proposition.
Assume that there exists a homomorphism g : A B such that
g((A)) (B) and g((A)) (B). Let us consider g and check g
A (A) iB (B).
i
A (A). By Corollary 9.1.6 we have a b (A) and

Let (a, b) i
a b (A). Applying to these facts our assumption we obtain
176 9 N4 -Lattices
g(a) g(b) = g(a b) (B) and g(a) g(b) = g(a b) (B). Again, by
Corollary 9.1.6 we have g ((a, b)) = (g(a), g(b)) i
B (B).
Thus, the desired homomorphism from A to B can be dened as h :=
(i )1 g i .
B A
If g is a monomorphism, then g and h are also monomorphisms.
2
A homomorphic image of a twist-structure has the following presen-
tation.
Proposition 9.2.16 Let B be a Heyting algebra, F a lter on B. Let A

S (B) and A = T w(B, , ). Then
A/F
= T w(B/F, /F, /F ).
Proof. Dene a mapping h : A/F (B/F ) as follows. For any (a, b) A,
h((a, b)/F ) := (a/F, b/F ). Clearly, h is a homomorphism. Check that this
is a monomorphism. The equality (a/F, b/F ) = (c/F, d/F ) is equivalent to
(a c)(b d) F . At the same time, (a, b)/F = (c, d)/F is equivalent
by Proposition 9.2.3 to (a, b) (c, d) F , which is equivalent in turn to
(a c) (b d) F . Thus, h is a monomorphism and it remains to check
that
h(A/F ) = T w(B/F, /F, /F ),
Let A := h(A/F ) = {(a/F, b/F ) | (a, b) A}. Then
(A ) = 1 (I(A )) = 1 ({(a/F, b/F ) (a/F, b/F ) | (a, b) A}) =
1 ({(a b/F, a b/F ) | (a, b) A}) = {a b/F | (a, b) A} = /F.

In a similar way, (A ) = 2 (I(A )) = /F .
2
Chapter 10
The Class
of N4-Extensions1
In this chapter, we study the structure of the lattice EN4 and discover
a denite similarity to the structure of the lattice Jhn+ studied in the rst
part of the book. It should be noted that dierences in the structure of these
two classes of logics are also essential. Moreover, we give rst applications of
the developed theory: the class of extensions of the logic N4 C obtained by
adding Dummetts linearity axiom (p q) (q p) to N4 is completely
described; two classical results by L.L. Maksimova, namely, the description
of pretabular logics and the description of logics with Craig interpolation
property, are transferred from the class of superintuitionistic logics to the
class of N4 -extensions.
10.1 EN4 and Int+

We start the investigation of the class EN4 with the question on interre-
lations between a logic in EN4 and its intuitionistic fragment. We dene a
mapping from EN4 into the class Int+ of extensions of intuitionistic logic
Li so that (L) is a
, , , -fragment of L. Later on, we call formulas
of the language
, , , intuitionistic formulas and denote the set of all
intuitionistic formulas by F. As in the rst part of the book, Int denotes the
lattice of intermediate logics, i.e., Int+ = Int {F}. The restriction of to
the class EN3 was investigated by M. Kracht [47] and A. Sendlewski [101].
1
Parts of this chapter were originally published in [77, 78, 79] (Springer, Netherlands).
177
178 10 The Class of N4 -Extensions
First, we point out that (L) is determined by basic Heyting algebras of

L-models. For a class K of N4 -lattices, put
K := {A | A K}.
Proposition 10.1.1 For any L EN4 and a class K of N4 -lattices, if

L = LK, then (L) = LK .
Proof. Due to Proposition 8.4.3 it is enough to consider twist-structures
isomorphic to elements of K. Let B IK and B S (A) for some Heyting
algebra A. Assume that A( = B ) is not a model for (L). Let (L) and
A-valuation v are such that v() = 1. For any propositional variable p, there
is an element bp |A| with (v(p), bp ) |B|. Let v be a B-valuation such that
v (p) = (v(p), bp ) for any p. The twist-operations are componentwise wrt the
rst component, therefore, 1 v () = v() for any . Thus, 1 v () = 1 and
B |= , which conicts with the assumption B |= (L) L.
We have thus proved the inclusion (L) LK . To check the inverse
inclusion take an intuitionistic formula (L). Let B IK, B S (A)
and a B-valuation v be such that 1 v() = 1. Then 1 v is an A-valuation
proving that A |= .
2
The following fact was stated in the last proof and it makes sense to
distinguish it as a separate lemma.
Lemma 10.1.2 Let A be an N4 -lattice and an intuitionistic formula.
Then
A |= A |= .
2
We now study the inverse image 1 (L) for any L Int+ , i.e., the class
of all conservative extensions of L in the class EN4 . We show that 1 (L)
forms an interval in the lattice EN4 and consider the mappings sending L
to the end points of the interval 1 (L). Let for any L Int+ ,
(L) := N4 + L,
i.e., (L) is obtained by extending language and adding the strong negation
axioms to L, and
(L) = (L) + { p (p q), (p p)}.
Logics in EN4 having the form (L) we call special or s-logics, and logics
of the form (L) are called normal explosive or ne-logics.
10.1 EN4 and Int+ 179
Prior to proving that (L) and (L) are the end points of the inverse
image of L wrt , we describe models of (L) and (L) and classes of models
generating such logics.
Proposition 10.1.3 Let L Int+ and A be an N4 -lattice.
1. A |= (L) i A |= L
2. A |= (L) i A |= L, (A) = Fd (A ) and (A) = {0}.
Proof. 1. The direct implication follows by Lemma 10.1.2. The inverse im-
plication follows from the same lemma and the fact that the axioms of N4
hold in any N4 -lattice.
2. If A |= (L), then A is a model of N3 and (A) = {0}. The direct
computation shows that for a twist-structure A S (B), the condition
A |= (p p) is equivalent to (a b) = 1 for any (a, b) A, i.e.,
(A) Fd (B). The inclusion Fd (B) (A) holds for any twist-structure,
which proves the direct implication.
We now assume that the right hand conditions of the equivalence are
satised. Due to Lemma 10.1.2 the rst condition implies A |= L. We have
A |= p (p q) by (A) = {0}. Finally, in view of the above consider-
ations, (A) = Fd (A ) implies A |= (p p).
2
For a Heyting algebra A, dene A := T w(A, Fd (A), {0}). This is the

least twist-structure over A. For any B S (A), A B. Namely, such

lattices are up to isomorphism models of logics of the form (L). If an
N4 -lattice A is isomorphic to a lattice of the form B , we call it a normal
N3-lattice. This is really an N3-lattice, since in this case, (A) = {0}. In
the following, N4 -lattices isomorphic to full twist-structures will be called
special N4 -lattices.
Further, we dene translations of (L) and (L) into L. Extend the
set P rop of propositional variables adding a new propositional variable p
for any p P rop. This extended set of variables will be denoted P rop. To
any formula in the language L we assign intuitionistic formulas and
over P rop dened as follows. If is a nnf, then is the result of
replacement of every occurrence of p in by p. If is not in nnf, then
:= () . We dene formulas
and as follows:

:=
(p p) (p p),
pvar() pvar()
:=

.
Proposition 10.1.4 Let A be a Heyting algebra and F or . The fol-

lowing equivalences hold.
1. A |= if and only if A |= .
2. A
|= if and only if A |= .
Proof. In the course of the proof we assume that is a nnf.

1. For any A -valuation v, we dene v : P rop |A| by the rule
v (p) := 1 v(p) and v (p) := 2 v(p). For any v : P rop |A|, we dene
A -valuation v : P rop |A| |A| by the rule v (p) := (v(p), v(p)).
Note that (v ) = v for any A -valuation v, and (v ) = v for any A-
valuation v. We have thus dened a bijection between A - and A-valuations.
Taking into account the fact that the action of intuitionistic operations on
a twist-structure agrees with their action on the rst component and the
equality 1 v( p) = 2 v(p) we obtain
1 v() = v ( ) and 1 v () = v( )
for any nnf . These relations and the above-stated bijection between A -
and A-valuations imply the desired equivalence.
2. Let A |= and v be an A
-valuation. By denition of A we have

1 v(p) 2 v(p) = 0 and 1 v(p) 2 v(p) Fd (A) for any p var(),

= 1, whence v ( ) = 1. By the equivalence from the
which implies v ()
previous item v() = 1.
Assume now A
|= and v( ) = 1 for some A-valuation v. Then

v( ). Put F =
v()
v()
and consider the quotient algebra A/F and
the quotient valuation v := v/F . We have v ()

= 1 and v ( ) = 1. The

= 1 implies that 1 (v ) (p) 2 (v ) (p) = 0 and 1 (v ) (p)
equality v ()

2 (v ) (p) Fd (A) for any p var(), i.e., v is an (A/F )

-valuation.

Arguing as in the previous item from v ( ) = 1, we infer (A/F ) |= . It
remains to note that (A/F )
= (A) /F by Proposition 9.2.16.

2
Corollary 10.1.5 Let L Int+ and F or . The following equivalences

hold.
1. (L) if and only if L.
2. (L) if and only if L.

10.1 EN4 and Int+ 181
Proof. 1. Let (L) and A |= L. Then A |= (L) by Proposition 10.1.3,

and A |= by Item 1 of the last proposition. Conversely, let L and
A |= (L). Then A |= L by Proposition 10.1.3, and (A ) |= (L). Since
A (A ) , we have A |= .
2. This item follows similarly from Item 2 of Proposition 10.1.3.
2
Corollary 10.1.6 For any L Int+ , the logics (L) and (L) are conser-
vative extensions of L.
Proof. Intuitionistic formulas are -free, therefore, = for any such

formula. In this way, the fact that (L) is a conservative extension of L
follows from Item 1 of Corollary 10.1.5.
For any intuitionistic ,
(L). Thus, the equivalence is
provable in (L) for any intuitionistic . Now, Item 2 of Corollary 10.1.5
implies that (L) is conservative over L.
2
Let K be a class of Heyting algebras. Put
K := {A | A K} and K := {A

| A K}.
Proposition 10.1.3 states, in fact, that
(L) = L(M od(L)) and (L) = L(M od(L))

.
The proposition below generalizes this result.
Proposition 10.1.7 Let L Int+ and L = LK. Then
(L) = LK and (L) = LK .
Proof. The inclusions (L) LK and (L) LK follow from Proposi-
tion 10.1.3.
Let (L). Then L and there is A K such that A |= . It
follows by Item 1 of Proposition 10.1.4 that A |= , i.e., LK .
Similarly, we use Item 2 of Proposition 10.1.4 to prove the second
equality.
2
Taking into account Proposition 10.1.1, the last proposition can be re-
worded as follows.
Corollary 10.1.8 Let L EN4 .
1. L is a special logic if and only if L is determined by some family of

special N4 -lattices.
2. [33] L is an ne-logic if and only if L is determined by some family of

normal N3-lattices.
2
Proposition 10.1.9 For any L Int+ , 1 (L) = [(L), (L)].
Proof. If L1 1 (L) and an N4 -lattice A is such that A |= L1 , then

A |= L by Lemma 10.1.2 and A |= (L) by Item 1 of Proposition 10.1.3.
Thus, (L) L1 .
embeds into A and belongs to M od(L1 ).
For any A M od(L1 ), (A )
Let K = {A | A M od(L1 )}. Then K M od(L1 ). By Proposition
10.1.1, L = LK, and (L) = LK by Proposition 10.1.7. We have thus
proved L1 (L).
That (L) and (L) belong to 1 (L) was stated in Corollary 10.1.6.
2
Remark. The analog of the above proposition does not hold for the lattice
of N4-extensions. If p (L) is the positive fragment of L EN4, one can
still prove that N4 + L is the least element of the inverse image ( p )1 (L),
where L is some extension of positive logic. But in the general case, the set
( p )1 (L) has no greatest element. This follows from the fact that the family
of intermediate logics with the same positive fragment does not necessarily
have a greatest element.
Following M. Kracht [47] we obtain another useful characterization of
special and normal explosive logics.
Proposition 10.1.10 Let L EN4 .
1. L is a special logic if and only if all rules of the form / are ad-
missible in L.
2. L is an ne-logic if and only if L EN3 and (p p) L.
Proof. 1. Let L be a special logic. According to Corollary 10.1.8 L is de-

termined by some family of special N4 -lattices. The admissibility of rules
/ now follows from Proposition 10.1.4 and Lemma 10.1.2.
Conversely, assume that all rules of the form / are admissible in L.
10.1 EN4 and Int+ 183
Lemma 10.1.11 All rules of the form / are admissible in any L

EN4 .
Proof. Let A be an N4 -lattice. If A |= , then A |= by Lemma

10.1.2, and (A ) |= by Proposition 10.1.4. Since A (A ) , we con-
clude A |= .
2
Taking into account this lemma and Corollary 10.1.5, we obtain for any
, L if and only if L if and only if ((L)). Thus, L = ((L))
and L is a special logic.
2. If L is an ne-logic, then p (p q) and (p p) belong to L
by denition.
Let L EN3, (p p) L, and A |= L. In this case, A is an
N3-lattice, i.e., (A) = {0}. The validity of (p p) is equivalent to
(A) = Fd (A). We have thus proved that any model of L is a normal
N3-lattice. By Corollary 10.1.8 L is an ne-logic.
2
Denote N3 := N3 + {(p p)}.
Proposition 10.1.12 1. : EN4 Int+ is a lattice epimorphism

commuting with innite meets and joins.
2. : Int+ EN4 is a lattice monomorphism commuting with innite

meets and joins.
3. is a lattice isomorphism of Int+ and EN3 .
Proof. 1. It follows immediately from denition that commutes with innite

meets. Prove that it commutes with innite joins. The inclusion iI (Li )
(iI Li ), where Li EN4 , i I, is obvious. Check the inverse inclusion.
Take a model A of iI (Li ). By Proposition 10.1.3 we have A
|= (Li )
for i I. Since Li (Li ) by Proposition 10.1.9, we obtain A |= Li for
all i I, and so A |= iI L i . According to Lemma 10.1.2 A |= (iI Li ),
which proves the inverse inclusion.
2. That is one-to-one follows fromCorollary 10.1.6.Check that com-
mutes with innite meets. Let L := iI Li . Since ( iI (Li )) is obvi-
ously
equal to L , the equality (L ) = iI (Li ) follows from the fact that
iI (Li ) is a special logic. Each of (Li ) is closed under all rules of the
form / by Proposition 10.1.10. Obviously, the intersection iI (Li ) is
also closed under such rules and it is also a special logic.
3. embeds Int into EN3 by Corollary 10.1.6. That is onto follows

from Item 2 of Proposition 10.1.10.
2
We have thus presented the class EN4 as a union of disjoint intervals
of the form 1 (L):

EN4 = [(L), (L)].
LInt+
Now, we establish interrelations between these intervals. It turns out that if

L1 L2 , then 1 (L2 ) is embedded into 1 (L1 ) as upper subinterval, at
the same time, 1 (L2 ) is isomorphic to a homomorphic image of 1 (L1 ).
Let L1 , L2 Int+ and L1 L2 . Mappings rL2 ,L1 : 1 (L2 ) EN4 and
eL1 ,L2 : 1 (L1 ) EN4 are dened as follows:
rL2 ,L1 (L) = L (L1 ) and eL1 ,L2 (L) = L + (L2 )
Proposition 10.1.13 Let L1 , L2 Int+ and L1 L2 . The following facts

hold.
1. For any L 1 (L2 ), we have eL1 ,L2 rL2 ,L1 (L) = L.
2. For any L 1 (L1 ), we have
rL2 ,L1 eL1 ,L2 (L) = L + rL2 ,L1 ((L2 )).
3. eL1 ,L2 is a lattice epimorphism from 1 (L1 ) onto 1 (L2 ).
4. rL2 ,L1 is a lattice monomorphism from 1 (L2 ) into 1 (L1 ) and
rL2 ,L1 ( 1 (L2 )) = [rL2 ,L1 ((L2 )), (L1 )].
5. For any L3 Int+ such that L2 L3 , we have
eL1 ,L2 eL2 ,L3 = eL1 ,L3 and rL3 ,L2 rL2 ,L1 = rL3 ,L1 .
Proof. 1. Let L 1 (L2 ). We calculate
eL1 ,L2 rL2 ,L1 (L) = (L + (L2 )) ( (L1 ) + (L2 )).
We have L+ (L2 ) = L since (L2 ) is the least element of 1 (L2 ). Consider

L := (L1 )+(L2 ). Due to homomorphism properties of we have (L ) =
L2 . Moreover, L EN3 since (L1 ) L . By Item 3 of Proposition 10.1.12
10.2 The Lattice Structure of EN4 185
we conclude L = (L2 ). Finally, L (L2 ) = L since (L2 ) is the greatest

element of 1 (L2 ).
2. Again for L 1 (L1 ), we have
rL2 ,L1 eL1 ,L2 (L) = (L (L1 )) + ((L2 ) (L1 )),
where the rst term is equal to L since (L1 ) is the greatest element of
1 (L1 ) and the second term is exactly rL2 ,L1 ((L2 )).
3. That eL1 ,L2 is a lattice homomorphism follows from the distributivity
of EN4 . Let L 1 (L1 ). Since is a homomorphism, we have
eL1 ,L2 (L) = (L) + (L2 ) = L1 + L2 = L2 .
Thus, the range of eL1 ,L2 is contained in 1 (L2 ). Item 1 implies that eL1 ,L2
is onto.
4. As above, we use the distributivity of EN4 and the homomorphism
properties of to prove that rL2 ,L1 is a lattice homomorphism from 1 (L2 )
into 1 (L1 ). If rL2 ,L1 (L) = rL2 ,L1 (L ), then applying the formula of Item 1
we obtain L = L . In this way, rL2 ,L1 is a monomorphism. The equality
rL2 ,L1 ( 1 (L2 )) = [rL2 ,L1 ((L2 )), (L1 )] follows from Item 2.
5. This follows immediately from denitions.
2
10.2 The Lattice Structure of EN4

It was proved in the previous section that the lattice EN4 decomposes
into a union of disjoint intervals 1 (L), where L Int+ and that 1 (L1 )
is isomorphic to an upper subinterval of 1 (L2 ), whenever L2 L1 . Since
Lk is the greatest point of Int, an isomorphic copy of 1 (Lk) is an upper
part of any interval of the form 1 (L) for L = F. For this reason, we start
the study of EN4 with a description of the interval 1 (Lk).
Consider subdirectly irreducible models of logics in 1 (Lk). According
to Corollary 9.2.11, any such model is isomorphic to an element of S (A),
where A is a subdirectly irreducible model of Lk. Any subdirectly irreducible
model of Lk is isomorphic to the two-element Boolean algebra 2. Therefore,
we have exactly four subdirectly irreducible models of logics in 1 (Lk):
2 = T w(2, {0, 1}, {0, 1}), 2

= T w(2, {1}, {0}),
R = T w(2, {1}, {0, 1}), 2L = T w(2, {0, 1}, {0}).

2
These lattices dene logics closely related to well-known nite valued

logics. Consider matrices of the form M (A) =
A, DA corresponding to the
above N4 -lattices. The lattice 2 is two-element. Its elements (0, 1) and
(1, 0) can be identied with the classical truth-values f and t respectively.
The strong negation coincides with the classical one and DA = {t} in this
case. Thus, the matrix corresponding 2 denes, in fact, the classical logic
with an additional negation , such that p p. We denote Lk := L2 .

The matrix M (2 ) has additional truth-values & := (1, 1) and :=

(0, 0). It has two distinguished values, D2 = {t, &}, and the following
lattice structure.
tq
@
@
q@ @q &
@
@q
f
In fact, M (2 ) can be considered as a four-valued Belnaps matrix ([8, 9])

enriched with the weak implication and the constant . We denote B 4 :=
L2 .
The logic of 2R diers from the maximal extension RM3 of the relevant
logic RM [3, 4] only by an additional constant . Indeed, RM3 can be
dened via the matrix
M =
{f, t, &}, , , , , {t, &},
where f & t, and are the usual lattice operations, is an order

reversing involution and the implication is dened by the rule:

b, if a D M
a b := (10.1)
t, if a DM .

2R = {t, &}. The lattice structures of
R | = {f, t, &} and D
We have |2
M and 2R are identical, the strong negation is also an order-reversing in-
volution, nally, the direct computation shows that coincides with the
R . Thus, L2R is RM3 enriched with the constant . We
implication of 2
denote RM3 := L2 R.

It was pointed out in [110] that the logic of 2

L is denition equivalent
ukasiewicz. We have |2
3 of L
to the three-valued logic L L | = {f, , t}. Let
us set
a b := (a b) ( b a)
and calculate the truth tables for and on {f, , t}:
f t
f t t t f t
t t
t f t t f
Thus,
{f, , t}, , , {t} is the well-known matrix for L
3 . All operations
of 2
L can be dened through and as follows:
a := a a, a b := (a b) b,
a b := ( a b), a b := a (a b).
For this reason, we denote L2
L as L 3.
Note that N4 -lattices 2 , 2
2R , and 2L have no non-trivial homo-
,
morphic images and are embedded one into another as follows
2
@
I
@
@
2
R 2
L
I
@
@
@
2

Thus, the interval 1 (Lk) contains exactly ve logics determined by

the following sets of subdirectly irreducible N4 -lattices:
{2 }, {2
R , 2L }, {2R }, {2L }, {2 }.

We have thus proved the following

Proposition 10.2.1 The interval 1 (Lk) has the following structure:

Lk
q
@
@
q @q
RM3 @ L
3
@
@q
RM3 L
3
q
B
4
In particular, (Lk) = B
4 and (Lk) = Lk.
2
Taking into account that L (L) for any L EN4 and that
(L
1 ) (L2 ) whenever L1 L2 , we conclude
Corollary 10.2.2 1. Any non-trivial extension of N4 is contained in

the logic Lk.
2. The logic N4 has no contradictory non-trivial extension.

2
We have thus pointed out the rst essential dierence between the struc-
ture of EN4 and the structure of Jhn+ . Minimal logic has the subclass of
contradictory extensions isomorphic to the class of extensions of positive
logic, whereas in the case of N4 , no contradiction can be added to N4
as a scheme. Further, unlike Jhn, the class EN4 \ {F or } of non-trivial
N4 -extensions forms a lattice with the unit element Lk, and our next step

is to describe coatoms of the lattice EN4 \ {F or }.
We know two examples of coatoms. It follows from Proposition 10.2.1
that RM3 and L 3 are coatoms in EN4 \ {F or }. A further example of
coatoms provides the twist-structure 3
, where 3 is a three-element linearly
ordered Heyting algebra, |3| = {0, 1, 2}, 0 1 2. Since Fd (3) = {1, 2},
the lattice 3
has four elements (0, 1), (1, 0), (0, 2), and (2, 0). The lattice
structure and the action of strong negation on 3 are presented in the
diagram below.
(2,q 0)
6
@
@
@
(0, 1) q -@q(1, 0)
@
@
@
@q?
(0, 2)
We can see that 3 is isomorphic to 2

as a lattice, but not as an
implicative lattice, the negation also acts on 3

in another way.
Proposition 10.2.3 The lattice EN4 \ {F or } has exactly three coatoms:

RM3 , L
. Each element of EN4 \ {F or } dierent from Lk is
3 , and L3
contained in one of them.
Proof. Let L EN4 \{F or } and L = Lk. Assume L is contained neither in

RM3 nor in L 3 . By Proposition 10.2.1 (L) = Lk in this case. It follows that
L has a model A = T w(B, , ), where B is not a Boolean algebra, in which
case = {1}. Take an element a in B such that a = 0, 1 and a , and
consider a twist-structure A0 := T w({0, a, 1}, {a, 1}, {0}). Clearly, A0 A
and A0 . Thus, 3 |= L and L L3 .
= 3

L3 is incomparable with RM3 and L 3 . On the one hand, the intuition-

istic fragment of L3 is dierent from classical logic. On the other hand,
p (p q) L3
\ RM3 and (p p) L3 \ L
.
3
2
Recall that an element x of a lattice A is called a splitting element if
there exists an y such that for every element z A, exactly one of the
following conditions holds: z x or y z. If x is a splitting element, the
corresponding y is called the splitting of A by x and is denoted by A/x. We
write A/{x, y} for A/x A/y.
Proposition 10.2.4 The logics RM3 , L 3 , and L3

are splitting elements

in the lattice EN4 and the following holds:
EN4 /RM3 = N3, EN4 /L3 = N4 + (p p), EN4 /L3

= B4
and
EN4 /{RM3 , L
3 } = N3 .
Proof. If L EN3, then (A) = {0} for any A |= L. Therefore, 2R is not
a model of L and L RM3 . If L EN3, there is A = T w(B, , ) such
that A |= L and = {0}. Choose an a such that a = 0 and consider
the quotient A0 := A/
a . By Proposition 9.2.16
A0
= A1 := T w(B/
a, /
a, /
a).
Since a , /
a = B/
a. Consequently, A1 contains a subalge-
bra T w({0/
a, 1/
a}, {1/
a}, {0/
a, 1/
a}), which is isomorphic to 2
R.
Thus, L RM3 and we have proved the equality EN4 /RM3 = N3.
If (p p) L, then every model of L is up to isomorphism of the
form T w(B, Fd (B), ). Obviously, 2
L does not satisfy this condition since
(2L ) = 2. Therefore, L |= 2L , and L is not contained in L
3.
If (p p) L, then L has a model A = T w(B, , ) such that
= Fd (B). Passing to the quotient A/(Fd (B)) we obtain a model A0 of
L such that A0 = T w(B0 , 0 , 0 ), B0 is a Boolean algebra and 0 = {1}.
Let a 0 and a = 1. Let a be a Boolean complement of a in B0 . By
Proposition 9.2.16 we have
A0 /
a
= A2 := T w(
a,
a, 1 ).
Since a = 1, the twist-structure A2 contains a subalgebra T w({1, a}, {1, a},

{a}) that is isomorphic to 2 L . Therefore, L |= 2L . We have thus proved

EN4 /L3 = N4 + (p p).
To prove EN4 /L3
= B4 , we notice that L3 = (L3) according
to Proposition 10.1.7. Thus, if (L) L3, then L (L) (L3). It is

well known that L3 is the opremum of Int. Consequently, if (L) L3, then
(L) = Lk and L extends (Lk). It remains to notice that (Lk) = B 4 by
Proposition 10.2.1.
The last equality immediately follows from the rst and the second.
2
We denote N4N := N4 + {(p p)} and distinguish in EN4 the
following subclasses:
Exp := {L EN4 | p (p q) L},
Nor := {L EN4 | (p p) L},
Gen := EN4 \ (Exp Nor).

Let L EN4 . We say that L is explosive if L Exp, we call L normal if

L Nor. Finally, if L Gen, we say that L is a logic of general form. In
the proposition below, we collect a series of simple facts on the introduced
classes.
Proposition 10.2.5 1. Exp Nor = EN3 and L3

is an opremum in

(Exp Nor) \ {F or }.
2. Exp = EN3.
3. Exp \ Nor = [N3, L

3]
4. L Exp i L RM3 i for any A M od(L), (A) = {0}.
5. Nor = EN4N .
6. Nor \ Exp = [N4N , RM3 ]
3 i for any A M od(L), (A) = Fd (A ).

7. L Nor i L L
8. Gen = [N4 , RM3 L

3 ].
2
All statements in this proposition easily follow from Propositions 10.2.3

and 10.2.4. Figure 10.1 presents the structure of EN4 .
Remark. Let us note the similarity of classes Exp, Nor, Gen and respec-
tively, the classes Int, Neg, Par studied in the rst part of the book. Both
classes Exp and Int consist of explosive logics. The similarity of classes Nor
and Neg is not so obvious. The class Neg consists of contradictory logics,
which represent the structures of contradictions in all Lj-extensions in the
following sense. If L1 Neg is a negative counterpart of an Lj-extension L,
then L1 is embedded into L by the contradiction operator (see Section 4.2.1).
According to Corollary 10.2.2 the logic N4 has no contradictory extension
and the structures of contradictions cannot be explicated in the same way as
for Lj-extensions. However, they can be represented semantically as ideals of
contradictions (A). Every N4 -lattice is determined by its basic algebra
A , the lter of completions (A) and the ideal of contradictions (A).
By Proposition 10.1.1 basic algebras of L-models for L EN4 determine
the intuitionistic fragment of L. According to the last proposition, the ideal
of contradictions is trivial in models of explosive logics and so, a logic from
Exp is characterized only by its intuitionistic fragment and by lters of com-
pletions of its models. At the same time, the lter of completions has a xed
rF or

rLk
X XXX
XXX
XX
XX X

'rL3$ XXX L
RM3 r Xr 3
' $
XXX ' $
XX
XXX
XXX
XX
Exp Nor r $
XXX
'
RM3 L
3
Nor \ Exp Exp \ Nor
& Xr %
X XXX
N3 XXX
XX
XXX
& r %
XXXr
& %
HH Gen
N4N HH N3
HH
HH
HH
H
HH
HH
& Hr %
N4
Figure 10.1
value, the least possible one, in models of normal logics. Thus, a normal logic
is characterized by its intuitionistic fragment and by ideals of contradictions
of its models. In this sense, the normal logics represent the structures of
contradictions in all N4 -extensions and are similar to negative logics in
the class of Lj-extensions. It will be shown in the next section that the class
Gen is connected to the classes Exp and Nor via the system of counterparts
in a similar way to the connection of the class Par with the classes Int and
Neg.
Our plans for the rest of this section are as follows. First, we consider the
restrictions of the operator to classes Exp and Nor and point out the perfect
analogy with the situation described in Section 10.1. For any L Int+ , the
inverse image of L with respect to the corresponding restriction of forms
an interval in the class Exp (Nor), and the end points of such interval can
be translated into L. Further, we study interrelations of the class Gen and
classes Exp and Nor.
Denote
3 := Exp and n := Nor .
The mappings 3 : Int+ Exp and n : Int+ Nor are dened as follows.
For every L Int+ ,
3 (L) := (L) + { p (p q)} and n (L) := (L) + {(p p)}.
Clearly, (L) = 3 (L) + n (L).

Logics in Exp having the form 3 (L) we call special explosive or se-logics,
and logics in Nor of the form n (L) are called special normal or sn-logics.
Models of se- and sn-logics are described in the following.
Proposition 10.2.6 Let L Int+ and A be an N4 -lattice.
1. A |= 3 (L) i A |= L and (A) = {0}.
2. A |= n (L) i A |= L and (A) = Fd (A ).
Proof. It can be proved similarly to Proposition 10.1.3.

2
In the following, we omit the proofs if they can be obtained similarly to
the results of the previous section.
For a Heyting algebra A, we dene
3 := T w(A, A, {0}) and An := T w(A, Fd (A), A).

A
If an N3-lattice A is isomorphic to a lattice of the form B3 , we call it a

special N3-lattice. If an N4 -lattice A is isomorphic to a lattice of the form
Bn , we call it a special normal N4 -lattice.
Translations of 3 (L) and n (L) into L can be dened in the following
way. The extended set P rop of propositional variables and formula were
dened in the previous section. Put

3 := (p p) ,
pvar()

n := (p p) .
pvar()
Proposition 10.2.7 Let A be a Heyting algebra and a formula in the

language L . The following equivalences hold.
1. A 3
3 |= if and only if A |= .
n |= if and only if A |= .
2. A n
Corollary 10.2.8 Let L Int+ and F or . The following equivalences

hold.
1. 3 (L) if and only if 3 L.
2. n (L) if and only if n L.

2
Let K be a class of Heyting algebras. Put
K3 := {A
3 | A K} and Kn := {An | A K}.

Proposition 10.2.9 Let L Int+ and L = LK. Then
3 (L) = LK3 and n (L) = LKn .
Corollary 10.2.10 Let L EN4 .
1. L is a special explosive logic if and only if L is determined by some

family of special N3-lattices.
2. L is a special normal logic if and only if L is determined by some

family of special normal N4 -lattices.
2
Proposition 10.2.11 For any L Int+ ,
( 3 )1 (L) = [ 3 (L), (L)] and ( n )1 (L) = [ n (L), (L)].
By analogy with Proposition 10.1.10, special explosive and special nor-

mal logics can be characterized via the admissibility of inference rules.
10.3 Explosive and Normal Counterparts 195
Proposition 10.2.12 Let L EN4 .
1. L is a special explosive logic if and only if L Exp and all rules of the
form /3 are admissible in L.
2. L is a special normal logic if and only if L Nor and all rules of the
form /n are admissible in L.
This characterization allows one to prove an analog of Proposition 10.1.12.

The statements on 3 and 3 were established earlier in [47].
Proposition 10.2.13 1. 3 : Exp Int+ and n : Nor Int+ are

lattice epimorphisms commuting with innite meets and joins.
2. 3 : Int+ Exp and n : Int+ Nor are lattice monomorphisms
commuting with innite meets and joins.
2
10.3 Explosive and Normal Counterparts

The decomposition of EN4 into classes Exp, Nor, and Gen is very similar to
the decomposition of the class Jhn of non-trivial extensions of minimal logic
into subclasses of intermediate, negative and properly paraconsistent logics.
Our next step is to dene explosive and normal counterparts for logics in
EN4 , in exactly the same way as we dened intuitionistic and negative
counterparts for extensions of minimal logic.
The mappings ()exp : EN4 Exp, ()nor : EN4 Nor, and ()ne :
EN4 Exp Nor are dened by the rules:
Lexp := L + N3, Lnor := L + N4N , Lne := L + N3 ,
where L EN4 . Obviously, Lne := Lexp + Lnor . Call the logic Lexp an
explosive counterpart of L, Lnor a normal counterpart of L, and Lne a nor-
mal explosive counterpart of L. Thus, by denition the explosive (normal)
counterpart of a logic L EN4 is the least explosive (normal) logic con-
taining L.
Notice that N4 has the following counterparts:
N4 N
exp = N3, N4nor = N4 , N4ne = N3 .
Some simple properties of counterparts are collected in the proposition

below.
Proposition 10.3.1 1. ()exp , ()not , and ()ne are lattice epimor-

phisms.
2. L Exp i L = Lexp i Lnor = Lne .
3. L Nor i L = Lnor i Lexp = Lne .
4. L EN3 i Lexp = Lnor .
5. (L) = (Lexp ) = (Lnor ) = (Lne ) for every L EN4 .
6. Lne = (L) for every L EN4 .
Proof. Item 1 follows from the distributivity of the lattice EN4 . Items 2
and 3 hold by the denition of counterparts. Item 4 follows from the relation
N3 = N3 + N4N . Homomorphism properties of imply Item 5.
We prove the last item. By Item 5 (Lne ) = (L) and Lne EN3 by
denition. Thus, the equality Lne = (L) follows from the fact that is
a lattice isomorphism of Int+ and EN3 (see Proposition 10.1.12).
2
The semantic characterization of logics N3, N4N , and N3 allows one
to characterize models of counterparts as follows.
Proposition 10.3.2 Let L EN4 and A be an N4 -lattice.
1. A |= Lexp if and only if A |= L and (A) = {0}.
2. A |= Lnor if and only if A |= L and (A) = Fd (A).
3. A |= Lne if and only if A |= L, (A) = {0}, and (A) = Fd (A).
2
We now study how the counterparts of L can be dened in L. For any
formula , we put

e := (p p), n := (p p),
pvar() pvar()
and
exp := e , nor := n , ne := (e n ) .
Let A be a twist-structure, A = T w(B, , ). We associate with A the
following substructures:
Aexp = T w(B, , {0}), Anor = T w(B, Fd (B), ),
Ane = T w(B, Fd (B), {0}).
According to Proposition 10.3.2 if A is a model of L, then Aexp , Anor ,

and Ane are models of explosive, normal and normal explosive counterparts
of L, respectively. The validity of formulas on Aexp, Anor , and Ane can be
simulated in A as follows.
Proposition 10.3.3 Let A S (B) and be a formula.
1. Aexp |= if and only if A |= exp .
2. Anor |= if and only if A |= nor .
3. Ane |= if and only if A |= ne .
Proof. 1. Let A |= exp and v be an Aexp-valuation. Note that v is also an

A-valuation,
since Aexp A. For any (a, b) Aexp , a b = 0. Consequently,
1 v( pvar() (p p)) = 1. And since A |= exp , we have 1 v() = 1.
We proved Aexp |= .
Assume now that Aexp |= but A |= exp . Let v be an A-valuation
such that 1 v(exp ) = 1. Put a := 1 v( pvar() (p p)) and F :=
a .
Then a 1 v(). Consider the quotient structure A/F and the quotient
valuation v/F . By Proposition 9.2.16
A/F
= A1 := T w(B/
a, /
a, /
a).
Applying
the same proposition we obtain (A1 )exp = Aexp /F . Clearly, 1 v/
F ( pvar() (p p)) = 1, whence v/F can be considered as a valuation
in (A1 )exp . Since a 1 v(), we have 1 v/F () = 1. Thus, Aexp /F |= ,
which conicts with the assumption Aexp |= .
The proofs of Items 2 and 3 are similar and are based on the following
facts. For any A-valuation v, v is an Anor -valuation i 1 v((p p)) = 1
for all p, and v is an Ane -valuation i for all p,
1 v((p p)) = 1 and 1 v((p p)) = 1.
Proposition 10.3.4 Let L EN4 and be a formula.
1. Lexp if and only if exp L.
2. Lnor if and only if nor L.
3. Lne if and only if ne L.

Proof. We prove only the rst item. The proofs of others are similar and are
based on the respective items of the last proposition.
Assume Lexp . Let A be a twist-structure such that L |= A. Then
Aexp |= Lexp . Indeed, Aexp |= L as a substructure of A and (Aexp ) = 0. By
Item 1 of the previous proposition we have A |= exp . Finally, since every
model of L can be represented as a twist-structure, we conclude exp L.
Conversely, let exp L. Consider a twist-structure A |= Lexp . By Propo-
sition 10.3.2 (A) = 0, i.e., A = Aexp . By Item 1 of the last proposition
A |= if and only if A |= exp , from which we conclude Lexp .
2
Let K be a class of twist-structures and {exp, nor, ne}. We put
K := {A | A K}.
Proposition 10.3.5 Let K be a class of twist-structures, L = LK, and

{exp, nor, ne}. Then L = LK .
Proof. Consider the case of explosive counterparts. Obviously, Lexp LKexp .

If Lexp , then exp L by the previous proposition. Consequently, there
is A K such that A |= exp . By Proposition 10.3.3 Aexp |= .
2
At the end of this section, we consider classes of logics having given
explosive and normal logics as explosive and normal counterparts. According
to Proposition 10.3.1 explosive and normal counterparts of a logic have the
same intuitionistic fragment. Therefore, for given L1 Exp and L2 Nor,
there is a logic L with Lexp = L1 and Lnor = L2 if and only if (L1 ) = (L2 ).
For L1 Exp and L2 Nor with (L1 ) = (L2 ), we dene the family of
logics
Spec(L1 , L2 ) := {L EN4 | Lexp = L1 and Lnor = L2 }
and the logic
L1 L2 := N4 + {exp | L1 } {nor | L2 }.
First of all, we note that all logics of Spec(L1 , L2 ) have the same in-
tuitionistic fragment and that if one of the logics L1 or L2 belongs to the
intersection Exp Nor, the class Spec(L1 , L2 ) is one-element.
Proposition 10.3.6 Let L1 Exp and L2 Nor be such that L = (L1 ) =

(L2 ).
1. Spec(L1 , L2 ) 1 (L).
2. If L2 Exp, then Spec(L1 , L2 ) = {L1 }.
3. If L1 Nor, then Spec(L1 , L2 ) = {L2 }.
Proof. 1. If L Spec(L1 , L2 ), then Lexp = L1 and
L = (Lexp ) = (L ) + (N3) = (L ) + Li = (L ).
2. If L2 ExpNor = EN3 , then L2 = (L). Since (L) is the greatest

element of 1 (L), we obtain L1 L2 .
Let L Spec(L1 , L2 ). We then have L + N4N = L2 L1 N3. We
claim that N3 L in this case. Assume that L is not explosive. There
then is a twist-structure A |= L with (A) = {0}. By Proposition 10.3.2
Anor |= L2 = Lnor and (Anor ) = (A) = {0}. The latter is impossible,
since L2 is explosive. From N3 L and L1 = L + N3 we obtain L = L1 .
3. From L1 Nor we conclude L2 L1 and nish the proof similar to
the previous item.
2
In the case when neither of L1 or L2 is normal explosive, the class
Spec(L1 , L2 ) forms an interval in the lattice EN4 containing at least two
points.
Proposition 10.3.7 Let L1 Exp and L2 Nor be such that (L1 ) =

(L2 ). Then Spec(L1 , L2 ) = [L1 L2 , L1 L2 ]. If L1
Nor and L2 Exp,
then L1 L2 = L1 L2 .
Proof. It follows by denition that L Lexp Lnor for all L EN4 ,

i.e., L L1 L2 for all L Spec(L1 , L2 ). Let us check that L1 L2
Spec(L1 , L2 ). We calculate
(L1 L2 )exp = (L1 + N3) (L2 + N3) = L1 (L1 ) = L1 .
That (L1 L2 )nor = L2 , can be checked similarly.

We have thus proved that L1 L2 is the greatest point of Spec(L1 , L2 ).
Let us consider the logic L1 L2 . If L Spec(L1 , L2 ), then
{exp | L1 } {nor | L2 } L
by Proposition 10.3.4, i.e., L1 L2 L. It remains to verify that L1 L2

belongs to Spec(L1 , L2 ).
Let L := L1 L2 . The inclusions L1 Lexp and L2 Lnor follow from
the denition of L1 L2 and Proposition 10.3.4. Let us prove the inverse
inclusions.
By denition
Lexp = N3 + {exp | L1 } {nor | L2 } = L1 + {ne | L2 }.
The last equality is due to the fact for any ,
N3 exp and N3 nor ne ,
which follows from (p p) N3. By Proposition 10.3.4 a formula of

the form ne belongs to Lexp if and only if (Lexp )ne . The equality
(L1 ) = (L2 ) implies L2 (L2 )ne = (L1 )ne , whence, L2 implies
ne L1 . Thus, Lexp = L1 . That Lnor = L2 , can be proved similarly.
Assume L1 Nor and L2 Exp. Prove that in this case
( p (p q)) (r r) (L1 L2 ) \ (L1 L2 ).
Since p (p q) L1 and (r r) L2 , the disjunction of these

formulas belongs to the intersection L1 L2 . To prove ( p (p q))
(r r) L1 L2 note that on the one hand, L1 L2 B 4 =L 3 RM3 ,
and on the other hand,
( p (p q)) (r r) B
4 .
2
As a consequence, we obtain a semantic characterization of logics of the
form L1 L2 .
Corollary 10.3.8 Let L1 Exp and L2 Nor be such that (L1 ) = (L2 ).
For every N4 -lattice A holds the equivalence
A |= L1 L2 Aexp |= L1 and Anor |= L2 .
Proof. The direct implication follows from the fact that L1 L2 Spec(L1 ,
L2 ) stated above. The inverse implication is by the denition of L1 L2 .
2
10.4 The Structure of EN4C and EN4 C 201
10.4 The Structure of EN4C and EN4 C
In this section, we completely describe the extensions of the logics N4 C :=

N4 + {C} and N4C := N4 + {C}, which are obtained by adding to N4
and respectively to N4 Dummetts linearity axiom C = (p q) (q p).
This result is interesting for the following reasons.
First, comparison of the structures of EN4 C and EN4C vividly demo-
nstrates how the regular structure of the lattice N4 is collapsed if we delete
the intuitionistic negation from the language. In particular, one cannot dene
normal logics in the class EN4.
Second, it is worth comparing the structures of the lattices EN4 C,
EN3C and ELC, where N3C := N3 + {C} and LC is Dummetts logic
obtained by adding the linearity axiom to intuitionistic logic, i.e. LC = Lil
in the denotation of the rst part of the book. Dummetts logic is the rst
example of superintuitionistic pretabular logic, whose structure of exten-
sions was completely described [27]. Recall that a logic is pretabular if any
of its extensions is tabular, i.e., is determined by a nite algebra, but the
logic is not tabular itself. In fact, the article [27] by M.J. Dunn and R. Meyer
was published earlier then the notion of pretabular logic was introduced and
generated an interest in pretabular logics as logics whose class of extensions
admits an exhaustive description. M. Kracht [47] described the lattices of
N3C-extensions and proved that although this logic is not pretabular, it
preserves the most essential properties of pretabular logics. Namely, all ex-
tensions of N3C are nitely axiomatizable and decidable. Moreover, given a
formula , one can determine which of the N3C-extensions is axiomatized
by . It turns out that the class of N4 C-extensions also satises all these
properties. Moreover, comparing lattices EN4 C and EN3C demonstrates
the complexity of the lattice of N4 -extensions.
Finally, notice that the description of the lattice N3C by M. Kracht [47]
was motivated by M. Gelfonds question on the structure of the upper part of
the lattice EN3. This is connected with the so-called answer set semantics for
logic programs. The logic of here-and-there with strong negation lying in
the upper part of EN3 plays an important role in the study of the answer set
semantics. For this reason, it is important to have same information on the
surroundings of this logic in the lattice EN3. In the study of paraconsistent
answer sets [82] there arises a paraconsistent version of the logic here-and-
there with strong negation, which belongs to the upper part of EN4 . This
naturally leads to the question on the description of the lattice EN4 C.
q Lch2 = Lk
q Lch3
q Lch4
pp
p
q LC
Figure 10.2
Recall that the class ELC of extensions of Dummetts logic has the fol-
lowing structure. Let chn be a linearly ordered n-element Heyting algebra.
Algebras chn are all up to isomorphism nitely generated subdirectly irre-
ducible models of LC. Obviously, for any n , chn chn+1 . Therefore,
every proper LC-extension has the form Lchn for some n , and ELC is
isomorphic to a linear order of type ( + 1) (see Figure 10.2).
Given a logic L, consider the class M odf si (L) of its nitely generated
subdirectly irreducible models. Dene on M odf si (L) the following preorder-
ing:
A B LA LB.
The following theorem by Jonsson is well known. In a congruence dis-

tributive variety, all subdirectly irreducible elements of the subvariety gener-
ated by the class K of algebras belong to HSUp(K). If an algebra A is nite,
then HSUp({A}) = HS({A}). The varieties of N4-lattices and N4 -lattices
are congruence distributive, therefore, for any nite N4-(N4 -)lattices A
and B, holds the equivalence
LA LB B HS({A}).
As was noted above, all nitely generated subdirectly irreducible models of

Dummetts logic LC are nite. According to Corollary 9.2.11, an N4(N4 )-
lattice A is subdirectly irreducible if and only if A is subdirectly irreducible.
The fact that if A is nitely generated, then A is nitely generated, follows
from the lemma below.
Lemma 10.4.1 If an N4(N4 )-lattice A is nitely generated, then A is

nitely generated.
Proof. If A is generated by the set {a1 , . . . , an }, then its reduct

|A|, , ,
(
|A|, , , , ) is generated by the set {a1 , . . . , an , a1 , . . . , an }. This
follows from the fact that every formula is equivalent in N4(N4 ) to an nnf.
By denition A =
|A|, , , / (
|A|, , , , / ). Consequently,
A is generated by the set
{[a1 ] , . . . , [an ] , [ a1 ] , . . . , [ an ] }.
2
Thus, every nitely generated subdirectly irreducible model of N4C
(N4 C) is isomorphic to a twist-structure over chn for suitable n ,
therefore, all nitely generated subdirectly irreducible models of N4C and
N4 C are nite. In this section, we consider only extensions of the logics
N4C and N4 C, therefore, in the following for A, B M odf si (L), holds
the equivalence:
A B B HS({A}).
Every extension of a logic L is determined by the class of its nitely
generated subdirectly irreducible models. This class forms a cone in the pre-
ordering (M odf si (L), ). Of course, not every cone of (M odf si (L), ) can be
represented in the form M odf si (L ) for a suitable L-extension L in the gen-
eral case. However, we prove that for every cone U of (M odf si (N4 C), ),
there exists L EN4 C such that M odf si (L) = U . Cones of the pre-
ordering (M odf si (N4 C), ) are closed under isomorphism. Note that the
isomorphism relation, coincides with the equivalence determined by the pre-
ordering . Factoring out the preordering (M odf si (N4 C), ) wrt the iso-
morphism relation, we obtain the partial ordering of isomorphism types of
nitely generated subdirectly irreducible N4 C-models, which we denote
(M odf si (N4 C), ). In this way, the description of the lattice EN4 C is
reduced to the description of the ordering (M odf si (N4 C), ).
Let |chn | = {0, 1, . . . , n 1}, 0 1 . . . n 1. Ideals of chn are
exactly non-empty initial segments of chn . The lter Fd (chn ) equals {1, . . . ,
n 1}. Therefore, if A = T w(chn , , ), then = chn or = Fd (chn ), and
= {0, . . . , m} for some m n 1. We have thus proved |S (chn )| = 2n.
Denote
chn (k, +) := T w(chn , chn , {0, . . . , k 1}),
chn (k, ) := T w(chn , Fd (chn ), {0, . . . , k 1}),

where 1 k n.
chn (1, ) q
@
@
chn (2, ) q @q
p @ chn (1, +)
p p @
@q
chn (n 2, ) q p chn (2, +)
@ p p
@
chn (n 1, ) q @q
@ chn (n 2, +)
@
chn (n, ) q @q
@ chn (n 1, +)
@
@q
chn (n, +)
Figure 10.3
Let A, B S (chn ). As it follows from Propositions 9.2.8 and 9.2.16

every proper homomorphic image of B is isomorphic to a twist-structure
over a proper homomorphic image of chn , i.e., over chm , m < n. In this way,
B A if and only if A B. Thus, according to Proposition 9.1.10 B A if
and only if |(A)| |(B)| and |(A)| |(B)|. The structure of S (chn )
ordered by is depicted in Figure 10.3.
Proposition 9.2.2 states that a homomorphic image h(A) of an N4 -
lattice A is uniquely determined by the homomorphic image h (A ) of
the basic Heyting algebra. Since every lter on chn has the form Fm :=
{m, . . . , n 1} and chn /Fm
= chm+1 , Proposition 9.2.16 allows one to con-
clude that quotients of twist-structures over chn have the following form:
chn (k, )/(Fm )

= chm+1 (min{k, m + 1}, ), m n 1, {+, }.
Consider the question of embedding twist-structures of the form chn (k, ).
If h : chn1 (k1 , 1 ) chn2 (k2 , 2 ), then according to Proposition 9.2.15 h
embeds chn1 into chn2 and the inclusions
h ((chn1 (k1 , 1 ))) (chn2 (k2 , 2 )), h ((chn1 (k1 , 1 ))) (chn2 (k2 , 2 ))
hold. Consequently, n1 n2 and k1 k2 . From h (0) = 0 and rst of the

above inclusions we conclude that 1 = + implies 2 = +.
We set 1 2 if 1 = 2 , or 1 = and 2 = +.
It is easy to see that the conditions 1 2 , n1 n2 , and k1 k2
guarantee that there exists an embedding g : chn1 chn2 such that
g({0, . . . , k1 1}) {0, . . . , k2 1}. Again, by Proposition 9.2.15 we conclude

that there exists an embedding
h : chn1 (k1 , 1 ) chn2 (k2 , 2 )
such that h = g.

We have thus proved the following statement.
Lemma 10.4.2 Let n1 , n2 , k1 , k2 , k1 n1 , k2 n2 , and 1 , 2

{+, }.
1. chn1 (k1 , 1 ) chn2 (k2 , 2 ) if and only if n1 n2 , k1 k2 , and 1 2 .
2. If chn1 (k1 , 1 ) H(chn2 (k2 , 2 )), then chn1 (k1 , 1 ) chn2 (k2 , 2 ).
2

Let T := n S (chn ). Combining Lemma 10.4.2 with the structure of
(S (chn ), ) presented in Figure 10.3, we obtain that T is ordered by as
depicted in Figure 10.4.
Since every nitely generated subdirectly irreducible model of N4 C is
isomorphic to some element of T, we arrive at the following statement.
Proposition 10.4.3 The preordering (M odf si (N4 C), ) has the struc-
ture depicted in Figure 10.4.
qch2 (1, )

@
S (ch2 ) q q @ qch2 (1, +)
@ @

q@qq @qch3 (1, +)

@ @

S (ch3 ) q q@ q @qch4 (1, +)
q
@ @
@
q@qq@q @q

@ @ q
S (ch4 ) q q @q @

q
q
@ @ q

q @q @ q
@
q
q @q q
q q
q
Figure 10.4
Describe now the cones of (T, ).

Let A T. If A = chn (k, ), we say that A is of type (n, k, ) and write
tp(A) = (n, k, ). By tpi (A), i = 1, 2, 3, we denote the i-th coordinate of
tp(A).
For a cone U of (T, ) and i = 1, 2, we put
tp
i (U ) := max{tpi (A) | A U and tp3 (A) = },
tp+
i (U ) := max{tpi (A) | A U and tp3 (A) = +}.
Clearly, tp+
i (U ) tpi (U ) for i = 1, 2.
Further, we distinguish the following cones:
T := {A T | tp3 (A) = },
Tk := {A T | tp2 (A) k},
T
k := {A T | tp3 (A) = and tp2 (A) k}.
Proposition 10.4.4 Every cone of (T, ) can be represented as U V ,
where U is a nite cone and V is one of the following cones:
, T, T , Tk , T
k , T Tk , Tk Tk ,
where k < k .
Proof. Let U be a cone in (T, ). If tp
1 (U ) < and tp2 (U ) < , it can
easily be seen from Figure 10.4 that U is nite or empty.
If tp+ +
1 (U ) = and tp2 (U ) = , then U = T.
Let tp
1 (U ) = and tp2 (U ) = . Then T
U . If additionally
+ +
tp1 (U ) < and tp2 (U ) < , then U = T U , where U is nite.

If tp+ + +
1 (U ) = and tp2 (U ) < , there exists m tp2 (U ) such that
chn (m, +) U for all n. Let
k := max{m | chn (m, +) U for all n}.
Then U = T Tk U , where U is nite.
Let tp
1 (U ) = and tp2 (U ) < . Put
k := max{m | chn (m, ) U for all n}.

Then T
k U and arguing as above, we obtain that either U = T
k U , or
U = T
k Tk

U , k < k, or U = Tk U , where U is some nite cone.
2
Theorem 10.4.5 The lattice EN4 C is isomorphic to the lattice of cones
of the partial ordering presented in Figure 10.4. All elements of EN4 C are
nitely axiomatized and decidable.
Proof. In view of Proposition 10.4.3, the lattice EN4 C is embedded into

the lattice of cones of the partial order presented in Figure 10.4. We have to
prove that this embedding is onto, i.e., that for any cone of (T, ), there is
L EN4 C such that T(L) := M odf si (L) T = U . Since T contains all up
to isomorphism nitely generated subdirectly irreducible models of N4 C,
the equality T(L) = U implies LU = L.
First, we note that the empty cone corresponds to the trivial logic F or .
According to Corollary 10.2.2 every non-trivial extension of N4 is consis-
tent, therefore, every contradiction axiomatizes F or over N4 C. Consider
p0 p0 as a standard axiom of the trivial logic.
Prove the equality T = T(N4 C + {(p p)}). By denition the
cone T contains exactly algebras from T satisfying (A) = Fd (A). This
condition is equivalent to the validity of (p p) on A, which implies
the desired equality.
Consider the formulas

Dn := (pk pk ) (pm pm ),
1k<mn+1

En := pk pm .
1k<mn+1
Let A T. Recall that by denition (A) = 2 {a a | a A}. We have

2 (a a) = 1 ( (a a)) = 1 (a a). Thus, (A) = 1 ({a
a | a A}). In this way, A |= Dn is equivalent to the fact that v(En ) = 1
for any A -valuation such that v(pk ) (A) for 1 k n + 1. Since
A is linearly ordered, we immediately obtain that A |= Dn if and only if
(A) contains at most n elements. In other words, A |= Dn if and only if
tp2 (A) n. We have thus proved Tn = T(N4 C + {Dn }).
Since T
n = T Tn , this cone is distinguished in T by the set of axioms
{(p p), Dn }.
For A T, the condition tp1 (A) n is equivalent to A = chm , m n.
The class of models with this property is distinguished in M odf si (LC) by
the formula En .
By Lemma 10.1.2 the validity of an intuitionistic formula on a twist-
structure over a Heyting algebra B is equivalent to B |= . In particular,
A |= En if and only if A |= En . The cone of algebras satisfying the con-
dition tp1 (A) n corresponds to the logic N4 C + {En }. Note that by
Lemma 10.4.2 this cone is generated by chn (n, +), i.e., by the full twist-
structure over chn .
The cone generated by chn (k, +) consists of algebras chm (s, ) satisfying
the conditions m n and s k. Therefore, it is distinguished in T by the
formulas {En , Dk }. Algebras over chn (k, ) satisfy additionally the condition
= . Therefore, the cone generated by chn (k, ) is distinguished in T by
the formulas {En , Dk , (p p)}.
Note that all logics corresponding to the cones up to this moment are
nitely axiomatizable.
In view of Proposition 10.4.4 every cone of (T, ) can be represented as
a nite union of cones of the forms considered above. The logic of a union
of cones is the intersection of logics of cones. If L1 = L + {1 , . . . , n } and
L2 = L + {1 , . . . , n }, moreover, the formulas 1 , . . . , n do not contain
propositional variables occurring 1 , . . . , n and vice versa, then
L1 L2 := L + {i j | 1 i n, 1 j m}.
In the case when L, L1 and L2 are extensions of positive or minimal logic,

this result was stated in Proposition 2.1.5. Its proof is based on the facts
that modus ponens is the only inference rule and that L, L1 and L2 contain
axioms of positive logics. Therefore, the result remains true for extensions of
N4 . In this way, Proposition 10.4.4 implies that every extension of N4 C
is nitely axiomatizable.
We know axiomatics for any logic in EN4 C. Given a logic L EN4 C
with a nite list of axioms A and some formula , we enumerate all formulas
deducible from A and from A {}. At a nite step we deduce from A or
we deduce from A {} an axiom of some other logic not contained in L.
Thus, L is decidable.
2

Note that every logic L EN4 C such that T M odf si (L) is not
tabular. Consequently, N4 C is not pretabular. Moreover, it has innitely
many non-tabular extensions.
It follows from the proof of Theorem 10.4.5 that we can construct such
an enumeration L0 , L1 , . . . of all logics from EN4 C that given n one can
eectively construct the list of axioms of Ln . Fix some enumeration of this
kind.
Proposition 10.4.6 There is an algorithm nding for any formula a

logic Ln EN4 C such that Ln = N4 C + {}.
Proof. Let 0 , 1 , . . . be an eective enumeration of all formulas inferable

from N4 C + {}. The desired algorithm works as follows.
Step n. For all m n, check whether all axioms of Lm are contained in

the list of formulas 0 , . . . , n . If we have a positive answer for some m0 n,
i.e., we have established the inclusion Lm0 N4 C + {}, we then check
whether Lm0 . In case of the second positive answer, we have proved
Lm0 = N4 C + {} and the algorithm terminates. Otherwise, we pass on
to the next step.
Since the list L0 , L1 , . . . contains all logics from EN4 C, and the list
0 , 1 , . . . all formulas from N4 C + {}, at some nite step we meet a
number n such that Ln = N4 C + {}.
2
We have thus proved that given a formula, one can determine which of
the extensions of N4 C is axiomatized by this formula.
In conclusion to this section, we consider briey how the situation changes
if we pass on to the logic N4C in the restricted language L =
, , ,
and to the explosive logic N3C.
Consider the logic LpC := Lp + {(p q) (q p)}. Finitely gen-
erated subdirectly irreducible models of LpC are nite linearly ordered
implicative lattices, the reducts of Heyting algebras chn to the language
, , . As above, we have Fd (chn ) = {1, . . . , n 1} (recall that for im-

plicative lattices Fd (A) =
{a (a b) | a, b |A|}). The lattice ELpC
as well as ELC is a linear order of type ( + 1) . We can identify twist-
structures from M odf si (N4C) with elements of T. As above chn (k, ) de-
notes a twist-structure A over n-element linearly ordered implicative lattice
with k-element ideal (A) and (A) = A if = +, (A) = Fd (A) if = .
However, the ordering diers in this case (see Figure 10.5).
The structures chn (k, ) have the same homomorphic images but the
embedding relation changes. In the language
, , , , the condition
h(0) = 0 should not be satised for a homomorphism h : A B of N4-
lattices. Therefore, the additional embeddings chn (k, +) chn+1 (k + 1, )
are possible. For this reason, the ordering (T, ) has the structure de-
picted in Figure 10.5. This ordering has fewer cones than the ordering pre-
sented in Figure 10.4. For example, in denotation of Proposition 10.4.4 we
have Tk T k+1 . This shows that N4-lattices A satisfying the condition
(A) = Fd (A) do not form a variety and we cannot dene analogs of nor-
mal logics as was done in the class of N4 -extensions. Similarly to Theo-
rem 10.4.5, we can prove the following statement.
Theorem 10.4.7 The lattice EN4C is isomorphic to the lattice of cones of

the partial order presented in Figure 10.5. All elements of EN4C are nitely
axiomatizable and decidable.
qch2 (1, )

@
S (ch2 ) q q!@ ! qch2 (1, +)
@!@

!q @qch (1, +)
@
q!!q 3
!
! @ ! !!
@

q! q!
S (ch3 ) !
!qq @qch (1, +)
@ 4
@! @ !!@
! !!
!qq!
!q @q
q!
! @ @
!
! @ ! @
! !
! ! q

S (ch4 ) !
q
!
q
@
! q
@ q

@! ! @ q
! q
!
q @q @
q
@
q q
q @ q
q q
q
Figure 10.5
ch1 (1, ) q
@
@
ch2 (1, ) q @q
@ ch1 (1, +)
@
ch3 (1, ) q @q
@ ch2 (1, +)
@
ch4 (1, ) q @q
p @ ch3 (1, +)
p p @
@q
p ch4 (1, +)
p p
Figure 10.6
Let us consider the logic N3C. Only twist-structures of the form chn (1, )
are models of N3C. Therefore, the isomorphism types of algebras from
M odf si (N3C) are ordered wrt as shown in Figure 10.6.
According to Theorem 10.4.5, the lattice EN3C is isomorphic to the

lattice of cones of the above ordering. It is easy to see that EN3C has the
structure depicted in Figure 10.7. This diagram was obtained for the rst
time by M. Kracht [47].
10.5 Some Transfer Theorems for the Class of N4 -Extensions 211
q
Lch1 (1, +) q
@
Lch2 (1, +) q @qLch1 (1, +)
@
Lch3 (1, ) q @q
@ @
Lch4 (1, ) q @q @qLch2 (1, +)
p @ @
p p @q @q
p @ @
N3Cne q p p @ q @qLch3 (1, +)
@ p @
@q p p @q
@ p @
@q p p @qLch4 (1, +)
@ p
@q p p
@
@q
p p
p
q
N3C
Figure 10.7
10.5 Some Transfer Theorems for the Class

of N4 -Extensions
In this section we transfer several results well known for the class of interme-
diate logics to the class of N4 -extensions. We obtain the characterization
of tabular and pretabular logics and logics with the Craig interpolation
property in the class EN4 . The corresponding results for Int we mention
without proofs.
First, we describe tabular and pretabular logics in EN4 , which gener-
alizes the result by A. Sendlewski [99] for EN3.
Let E be a lattice of logics and L E. We say that L is of nite codi-
mension [93] if there is no innite descending chain of logics containing L.
It is known that
Theorem 10.5.1 [93] Let L Int. L is tabular if and only if L is of nite

codimension in Int.
2
It turns out that an N4 -extension is tabular if and only if its intuition-
istic fragment is tabular.
Proposition 10.5.2 Let L EN4 . L is tabular if and only if (L) is

tabular.
Proof. Let L = LA for some nite N4 -lattice A. Then (L) = LA by
Proposition 10.1.1. Thus, if L is tabular, (L) is tabular too.
Assume (L) = LB for some nite Heyting algebra B. Since Heyting
algebras are congruence distributive, all nitely generated subdirectly ir-
reducible (L)-models belong to HS(B). Consequently, (L) has only a -
nite number of nitely generated subdirectly irreducible models {B1 , . . . , Bn }
and all these models are nite. For every nitely generated subdirectly ir-
reducible model A of L, A is a subdirectly irreducible model of (L) by
Corollary 9.2.11, and A is nitely generated by Lemma 10.4.1. Conse-
quently, every element of M odf si (L) is isomorphic to some A S (Bi ),
i = 1, . . . , n. Thus, M odf si (L) contains only a nite number of algebras up
to isomorphism, which means that L is tabular.
2
This fact and Theorem 10.5.1 yield the description of tabular logics in
EN4 .
Theorem 10.5.3 Let L EN4 . L is tabular if and only if L is of nite
codimension in EN4 .
Proof. If L is tabular, up to isomorphism it has only a nite number of
nitely generated subdirectly irreducible models {A1 , . . . , An }. This follows
from the fact that the variety of N4 -lattices is congruence distributive and
the above-mentioned result by Jonsson. Any extension of L is determined
by some subset of {A1 , . . . , An }. Therefore, L has nitely many extensions,
in particular, it has a nite codimension.
Let L have a nite codimension in EN4 . By Proposition 10.1.12 the
mapping embeds the lattice E(L) into EL. Therefore, (L) also has a
nite codimension. Now the tabularity of (L) follows from Theorem 10.5.1
and the tabularity of L from Proposition 10.5.2.
2
In [55], L.L. Maksimova proved that there are exactly three pretabular
intermediate logics.
Theorem 10.5.4 [55] The class Int contains exactly three pretabular logics:
LC := Li + {(p q) (q p)},
LJ := Li + {p (p (q q))},
LH := Li + {p p, p (p (q (q (r r))))}.
2
Proposition 10.5.5 Let L EN4 . L is pretabular if and only if L = Lne

and (L) is pretabular.
Proof. Let L be pretabular. Then (L) is pretabular by Proposition 10.5.2.
Indeed, every proper extension of (L) is of the form (L ) for some proper
extension L of L. By assumption L is tabular, whence (L ) is tabular. At
the same time, (L) is not tabular since L is not tabular.
Assume L = Lne . We have (Lne ) = (L) and L Lne , consequently,
Lne is a non-tabular proper extension of L, which conicts with the assump-
tion that L is pretabular.
Assume L = Lne and (L) is pretabular. Then L EN3 . By Proposi-
tion 10.1.12, every proper L-extension is of the form (L ) for some proper
(L)-extension L . By assumption L is tabular, consequently, (L ) is tab-
ular by Proposition 10.5.2. Thus, L is pretabular.
2
By Item 6 of Proposition 10.3.1 if L = Lne , then L = (L). This fact
and the last two assertions imply.
Theorem 10.5.6 The class EN4 contains exactly three pretabular logics:
(LC), (LJ), and (LH).
2
We say that a logic L possesses the Craig interpolation property (CIP )
if L implies that there exists a formula such that L and
L, and has occurrences of common variables of and only.
The algebraic counterpart of this property is as follows. Let K be a class
of algebras. We say that K has an amalgamation property if for any algebras
A0 , A1 , A2 K and monomorphisms i1 : A0 A1 and i2 : A0 A2 , there
exist an algebra A K and monomorphisms 1 : A1 A and 2 : A2 A
such that 1 i1 = 2 i2 , i.e., the diagram
A1
@
i1 @ 1
@@
R
A0 A
@
i@
2 2
@
R
@
A2
commutes. The triple (A, 1 , 2 ) is called an amalgam of A1 and A2 over A0 .

It was proved in [56] that an intermediate logic possesses CIP if and

only if the corresponding variety of Heyting algebras has the amalgamation
property. The next statement can be obtained by a natural modication of
the proof from [56].
Proposition 10.5.7 Let L EN4 . L possesses CIP if and only if the

variety V ar(L) has the amalgamation property.
Proof. Assume that L possesses CIP and prove the amalgamation property
for V ar(L). Let A0 , A1 , A2 V ar(L) be such that A0 is a common subalge-
bra of A1 and A2 . Assign to each element a Ai , i = 0, 1, 2, a propositional
variable pia . Assume additionally that p0a = p1a = p2a for a A0 and all other
variables are dierent. Denote by Fi the set of formulas with variables pia .
Put F := F0 F1 F2 . Let v : F A1 A2 be such that v(pia ) = a. For
i = 1, 2, denote
Ti := { Fi | v() DAi }.
It is easy to see that Fi L Ti and that the sets Ti are closed under modus
ponens. Put
T := { F | T1 T2 L }.
Lemma 10.5.8 Let {i, j} = {1, 2}, Fi , and Fj . Then
T F0 (v() $ v() and v() $ v()).
Proof. If v() $ v() and v() $ v(), then v( ) DAi and v( )

D Aj . Therefore, , T , whence T .
Conversely, let T . There are nite subsets i Ti and j Tj
such that i , j L . By Deduction Theorem

L i ( j ( )),
which is equivalent to

L i ( j ).

According
to CIP , there is a F0 such that L i and L
( j ). We obtain i L and j L . Since i Ti ,
we have Ti . Consequently, v( ) DAi , i.e., v() $ v().
Similarly, v() $ v().
2
Let Fj be such that T L . Take some LFi , then T L .

By the previous lemma there is a formula F0 such that v() $ v() and
v() $ v(). It follows from LFi that v() DAi . Taking into account
the rst inequality we obtain v() D Ai , or, equivalently, v( ) = v().
Consequently, we have also v() D Aj . Hence, v() DAj , i.e., Tj . In
this way, we have proved T Fj = Tj , j = 1, 2.
Dene on F the relation =:= {(, ) | T }. Since the prov-
able strong equivalence has in N4 the congruence properties, = is a
congruence on F.
Let , Fi . The equality T Fi = Ti implies

= if and only if v() = v() (*)
Put A := F/ =. Clearly, A is an N4 -lattice. Let mappings i : Ai A,

i = 1, 2, be given by the rules
= , a Ai .
i (a) := [pia ]
The mapping i is a homomorphism since = is a congruence, and is one-to-

one in view of (). Since p0a = p1a = p2a , we have 1 (a) = 2 (a) for a A0 .
Thus, (A, 1 , 2 ) is an amalgam of A1 and A2 over A0 .
Prove the inverse implication. Assume that V ar(L) has the amalgama-
tion property. We need one more lemma from [56].
Lemma 10.5.9 Let A0 is a common subalgebra in Heyting algebras A1 and

A2 . If a A1 and b A2 are such that there is no c A0 with a A1 c
and c A2 b, then there are lters F1 on A1 and F2 on A2 such that a F1 ,
b F2 , and A0 F1 = A0 F2 .
Proof. We write i for Ai , i = 1, 2. Consider the sets
:= {z A0 | a 1 z} and := {z A0 | z 2 b}.
By assumption = . Consider an ideal on A2 generated by b, i.e.,

the set I = {x A2 | x 2 b}. Then F2 := A2 \ I is a lter on A2 such that
b F2 . Put
F0 := F2 A0 and I0 := I A0 .
It is clear that F0 = A0 \ I0 , F0 , and I0 . Further, consider a lter
on A1 generated by F0 {a}, i.e.,
F1 = {x A1 | a z 1 x for some z F0 }.
We claim that F1 I0 = . Indeed, if x F1 I0 , then a z 1 x for z F0 .

The latter means that a 1 z x. Since x I0 A0 and z A0 , we
have z x F0 . Taking into account z F0 we obtain x F0 , which
conicts with F0 I0 = .
From F0 F1 and F1 I0 = we immediately infer that F1 A0 = F0 ,
i.e., A0 F1 = A0 F2 .
2
We adopt this statement for N4 -lattices.
Lemma 10.5.10 Let an N4 -lattice A0 is a common subalgebra in N4 -

lattices A1 and A2 . If a A1 and b A2 are such that there is no c A0
with a $A1 c and c $A2 b, then there are sk 1 on A1 and 2 on A2 such
that a 1 , b 2 , and A0 1 = A0 2 .
Proof. Without loss of generality we may assume that Ai S (Bi ), i =

0, 1, 2, a = (a1 , a2 ) and b = (b1 , b2 ). As we can see from Proposition 8.4.2
the algebras B0 , B1 , B2 and the elements a1 B1 and b1 B2 satisfy the
conditions of the previous lemma. Let the conclusion of this lemma hold for
lters F1 on B1 and F2 on B2 . Then sk 1 := F1 and 2 := F2 satisfy
all the desired conditions. We check only the equality A0 1 = A0 2 .
If (a, b) A0 1 , then a B0 F1 = B0 F2 , whence (a, b) F2 . The
inverse inclusion can be checked similarly.
2
Let x, y, z be disjoint tuples of variables and let for (x, y) and (x, z)
there is no (x) such that (x, y) (x) L and (x) (x, z) L.
The relation L is equivalent to $ Eq(V ar(L)). Therefore,
there is no (x) such that V ar(L) |= (x, y) $ (x) $ (x, z). Let Ax be a
free algebra in V ar(L) with free generators x and Ax,y,z a free algebra with
generators x, y, z. Consider Ax as a subalgebra of Ax,y,z .
According to Lemma 10.5.10 there are sk 1 and 2 on Ax,y,z such that
[(x, y)] 1 , [(x, z)] 2 , and Ax 1 = Ax 2 . Here [] denotes the
element of the free algebra corresponding to the formula .
Put A1 := Ax,y,z /1 and A2 := Ax,y,z /2 . Then [(x, y)]/1 DA1 and
[(x, z)]/2 DA2 . Recall that for any u, v Ax , the equality u/1 = v/1
is equivalent to u v 1 , which is equivalent in turn to u v 2 in
view of Ax 1 = Ax 2 . Thus,
u/1 = v/1 if and only if u/2 = v/2 .
Therefore, there is a natural embedding i2 (u/1 ) := u/2 of N4 -lattice
A0 := {u/1 | u Ax } A1 into N4 -lattice A2 .
By the amalgamation property of V ar(L) there are A V ar(L) and

monomorphisms i : Ai A, i = 1, 2, such that 1 A0 = 2 i2 . Put
h([x]) := 1 ([x]/1 ), h([y]) := 1 ([y]/1 ), h([z]) := 2 ([z]/2 ),
where h([x]) := h([p1 ]), . . . , h([pn ]) for x = p1 , . . . , pn , etc. Extend this map-
ping to a homomorphism h : Ax,y,z A. We obtain
h([(x, y)]) = 1 ([(x, y)]/1 ) DA , h([(x, z)]) = 2 ([(x, z)]/2 ) DA .
Therefore, V ar(L) |= (x, y) $ (x, z), i.e., (x, y) (x, z) L.

2
The description of intermediate logics with CIP looks as follows:
Theorem 10.5.11 [56] In the class Int, there exists exactly 7 logics with
CIP :
L1 := Li,
L2 := Li + {p p},
L3 := Li + {p (p (q q))},
L4 := L3 + {(p q) (q p) (p q)},
L5 := L3 + {p p},
L6 := Li + {(p q) (q p)},
L7 := Lk.
We use the facts presented above to nd all logics in EN4 possessing

CIP . First of all, we notice that the intuitionistic fragment of a logic in
EN4 inherits CIP .
Proposition 10.5.12 If L EN4 possesses CIP , then (L) possesses

CIP .
Proof. According to Proposition 10.5.7 the variety V ar(L) has the amalga-
mation property. We show that the class of (L)-models has the amalgama-
tion property too. Let Heyting algebras A0 , A1 , and A2 model (L), and
i1 : A0 A1 and i2 : A0 A2 are embeddings.
Consider twist-structures (A0 )
, (A1 ) , and (A2 ) . According to Pro-
position 10.1.3 these twist-structures are models of (L). Logics L and
(L) have the same intuitionistic fragment, therefore, L (L) by
Proposition 10.1.9. Thus, (A0 )
, (A1 ) , and (A2 ) model L. Moreover,
i
1 : (A0 ) (A1 ) and i2 : (A0 ) (A2 ) are embeddings according
to Proposition 9.2.14. By assumption one can nd an amalgam (A, 1 , 2 )
of (A1 )
and (A2 ) over (A0 ) . In particular, 1 i1 = 2 i2 . Using functor
properties of () we obtain
(1 ) i1 = (1 i
1 ) = (2 i2 ) = (2 ) i2 .
Again by Proposition 9.2.14 (1 ) : A1 A and (2 ) : A2 A are

embeddings. Thus, (A , (1 ) , (2 ) ) is an amalgam of A1 and A2 over A0 .
2
The implication in the above proposition can be reversed for some special
logics from the interval 1 (L).
Proposition 10.5.13 Let L Int. If L possesses CIP , then the logics (L),
3 (L), n (L), and (L) possess CIP .
Proof. Recall that for any logic L 1 (L) for any N4 -lattice B modelling
L , B is isomorphic to a twist-structure over a Heyting algebra A such that
A |= L.
Let A0 , A1 , A2 |= L. Consider twist-structures Bi := T w(Ai , i , i ),
i = 0, 1, 2, and their embeddings i1 : B0 B1 and i2 : B0 B2 . Then (i1 )
embeds A0 into A1 , and (i2 ) embeds A0 into A2 . Since L possesses CIP ,
there is an amalgam (A, 1 , 2 ) of A1 and A2 over A0 . Proposition 9.2.15
allows us to lift up monomorphisms 1 and 2 so that we obtain the following
diagram.
B1
@
i1 @ 1
@
R
@
B0 A
@
i@
2
2
@@
R
B2
The functor properties of () guarantee that this diagram commutes.

The full twist-structure A is a model of (L) by Proposition 10.1.3 and we
have thus proved that (L) possesses CIP .
If Bi |= 3 (L), then i = {0}, i = 0, 1, 2, by Proposition 10.2.6. In this
3
i embeds Bi into Aexp . Again by Proposition 10.2.6 Aexp |= (L),
case,
which proves that 3 (L) possesses CIP .

In a similar way, applying Propositions 10.2.6 and 10.1.3, we prove that

n (L) and (L) possess CIP .
2
It turns out that all logic from 1 (L) that can possess CIP were listed
in the last proposition.
Proposition 10.5.14 If L EN4 and L possesses CIP , then L is of the

form (L ), 3 (L ), n (L ), or (L ) for some L Int.
Proof. Let L possesses CIP . If L = Lne , then L = (L). Let L = Lne .

Then one of the formulas (p p) or (p p) does not belong to L.
Assume (p p) L and (p p) L and prove that L = 3 (L)
in this case.
Consider the Lindenbaum twist-structure of L, L
L . Recall that
L | = {([]L , [ ]L ) | []L },
|L
where []L is a coset of the equivalence
L L.
L is a twist-structure over the Heyting algebra LL of all cosets []L .

Thus, L
Note that (p p) L implies the equality (L L ) = {0}. Put
F := {[]L | (p0 p0 ) L}.
Obviously, F is a lter on LL and F is an sk on L L . Consider a quotient

algebra L0 := L L /F .
Let ([0 ]L , [0 ]L ), ([1 ]L , [1 ]L ) L

L . By Proposition 9.2.3 the equality
([0 ]L , [0 ]L )/F = ([1 ]L , [1 ]L )/F is equivalent to
([0 ]L , [0 ]L ) ([1 ]L , [1 ]L ) F .
By denition of F the latter is equivalent to
[0 0 ]L , [1 1 ]L F.
Thus, an element of L0 can be identied with a pair of elements of the

quotient LL /F = LF ,
[([]L , []L )]F = ([]F , []F ).
Consider the element a := ([p0 ]F , [ p0 ]F ) L0 . Then
a a = ([p0 p0 ]F , [p0 p0 ]F ),
which means that [p0 p0 ]F (L0 ). By denition [(p0 p0 )]L F ,

i.e., [(p0 p0 )]F = [p0 p0 ]F = 1. Consequently, [p0 p0 ]F = 0.
We have thus proved that (L0 ) = LF , i.e., L0 is a special N3-lattice.
Let us check the equality L = LL0 . We have L LL0 , since L0 is a
quotient of L L . Let L0 |= (q0 , . . . , qn ) and p0 does not occur in . Then
[(q0 , . . . , qn )]L F , i.e.,
(p0 p0 ) (q0 , . . . , qn ) L.
By CIP we have (p0 p0 ) L and (q0 , . . . , qn ) L, where

is some formula without propositional variables, i.e., constructed from
and logical connectives. Denote & := . Then & N4 . By an easy
induction on the structure of formulas one can prove that N4
or & N4 . In this way, we obtain (p0 p0 ) L or
& (q0 , . . . , qn ) L. In view of our assumption (p0 p0 ) L,
consequently, (q0 , . . . , qn ) L.
We have thus proved that L is determined by the special N3-lattice L0 .
By Corollary 10.2.10 L is a special explosive logic, i.e., a logic of the form
3 (L ).
Now, we assume (p p) L and (p p) L and prove that
L = n (L). In this case, the Lindenbaum twist-structure LL is normal. Let
us consider the lter
F1 := {[]L | (p0 p0 ) L}
and the quotient L1 := L

L /F1 . The element b = ([p0 ]F1 , [ p0 ]F1 ) is such
that
b b = ([p0 p0 ]F1 , [p0 p0 ]F1 ).
Since [p0 p0 ]L F1 , we obtain [p0 p0 ]F1 = 1 and (L1 ) = L1 .
Thus, L1 is a special normal N4 -lattice. Arguing as above, we check that
L = LL1 . Thus, L is determined by a special normal N4 -lattice. By Corol-
lary 10.2.10 L is a special normal logic, i.e., a logic of the form n (L ).
It remains to consider the case when (p p) L and (p p)
L. We have ((p p) (q q)) L. Indeed, this is equivalent to
(p p) (q q) L. If (p p) (q q) belongs to L, by
CIP we have
(p p) L or & (q q) L.
Both alternatives conict with our assumption. Let us consider the lter
F2 := {[]L | (p0 p0 ) (p1 p1 ) L}

on LL . Arguing as above, we can prove that the quotient L2 := L

L /F2 is
a characteristic model for L and that (L2 ) = (L2 ) = LF2 , i.e., L2 is a
special N4 -lattice. Since L is determined by a special N4 -lattice, it is a
special logic, i.e., L = (L).
2
3
The logics (L), (L), (L), and (L) are dierent for any L in Int.
n
This fact and the above propositions imply
Theorem 10.5.15 In EN4 , there are exactly 28 non-trivial logics possess-

ing CIP : (Li ), 3 (Li ), n (Li ), and (Li ), i = 1, . . . , 7, where L1 , . . . , L7
are logics from Theorem 10.5.11.
2
Chapter 11
Conclusion
Discussing the question Why is paraconsistency worthy? J.-Y. Beziau [11]

emphasized that paraconsistent logic is an important contribution to the
theory of negation and to modern logic in general. The distinction between
triviality and inconsistency made in paraconsistent logic is similar to the
distinction between implication and inference relation and allows one to
elucidate new features of traditional logical notions. Suppose that the in-
vestigations presented in this book also contribute to the general theory of
logical systems. For two explosive logics with the same positive fragment and
with essentially dierent kinds of negation, we investigated how the lattice
of extensions of a logic changes when the explosion axiom is deleted, i.e., if
we pass from a logic to its paraconsistent analog. It turns out that in both
cases the lattices of extensions extend in a rather regular manner. In the
class of extensions of a paraconsistent logic, one can distinguish the sub-
class of explosive logics, i.e., the class of extensions of the original explosive
logic; the subclass of logics, which can be used to represent the structures
of contradictions in all extensions of the considered paraconsistent logic (see
Remark after Proposition 10.2.5). Finally, all other logics can be obtained
via a combination of logics from the above two subclasses. The manner of
combination can be explicated via a suitable representation theory for alge-
bras modelling the logics under consideration. Thus, admitting non-trivial
inconsistent theories leads to quite natural and interesting mathematical
structures arising. Moreover, the above results allow one to consider para-
consistent logic not as an alternative to traditional explosive logic, but as
a more general setting for considering explosive systems. It would be nat-
ural to continue the study of lattices of extensions of paraconsistent logics
for some other concepts of negation. According to the authors opinion the
223
224 11 Conclusion
most natural candidate for such investigation is the subminimal negation

by D. Vakarelov [111, 112]. The subminimal logic has the same positive
fragment as intuitionistic logic. Its negation is a natural generalization of
minimal negation and can be dened via two constants and & such that
& holds by the formula
p (p ) &.
In this case, the constants are denable through negation as := (p p)

and & := (p p). To obtain minimal logic from the subminimal one we
have to add the axiom &. As other candidates for such investigations one
could consider the negation as impossibility by K. Dosen [24, 25], dierent
kinds of negations considered by M. Dunn in [26] and nally, the regular
and co-regular negations by D. Vakarelov.
At the same time, one can continue the study of the classes of extensions
of minimal logic and of Nelsons logic N4 using the coordinate system de-
picted in this work. It should be mentioned that there are interesting results
on minimal logic extensions that are not reected in this book. L.L. Maksi-
mova [57, 58] studied interpolation and denability properties in extensions
of positive and minimal logic. She constructed several series of logics with
Craig interpolation property in the class of minimal logic extensions. How-
ever, it is still unknown whether the total number of logics with Craig in-
terpolation property in this class of logics is nite. M.V. Stukacheva studied
the disjunction property in the class of minimal logic extensions [103, 104]
and transferred the technique of canonical formulas by M. Zakharyaschev
to this class of logics [105].
And the very nal remark concerns the technique of twist-structures. In
the second part of the book we used twist-structures to solve algebraiza-
tion problems for Nelsons logics N4 and N4 . The main diculty here
was connected with the fact that Nelsons logics are not closed under the
replacement rule. But they are closed under weak forms of replacement rules:
,
and ,
+ () +
() () ()
where + is a -free formula. Namely, these rules allow one to dene the
semantics for Nelsons logics in terms of twist-structures. The rst of these
rules and the fact that -free fragment of N4 (N4 ) is equal to positive (in-
tuitionistic) logic imply that implicative lattices (Heyting algebras) should
be taken as basic algebras of twist-structures for N4 (N4 ). The number
of premises in the second rule corresponds to the number of components in
11 Conclusion 225
elements of twist-structures. Finally, the strong negation axioms determine

the twist-operations. A more general situation, where a technique of this
kind could be applied would be as follows. Assume that we have a logic L
in a language L, L0 is a sublanguage of L and L0 is an L0 -fragment of L.
Further, let the logic L0 be extensional, the class of algebras K0 provides an
algebraic semantics for L0 , and the logic L is closed under the rule:

, where 0 is an L0 -formula.
0 ()
0 ()
Finally, assume that there are formulas 1 (p), . . . , n (p) such that L is
closed under the following form of replacement rule:
1 () 1 (), . . . , n () n ()
,
() ()
where is an arbitrary L-formula. Note that in case of N4 and N4 we

have n = 2, 1 (p) = p and 2 (p) = p.
In this case, one can try to construct a semantics for the logic L in
terms twist-structures over algebras from K0 with the number of components
equalling n. Naturally, to dene twist-operations some reduction axioms for
connectives from L \ L0 are needed. Possibly, some additional conditions
should be imposed on the form of formulas 1 , . . . , n . It is interesting
to look for natural examples of logics, which are closed under rules of the
form described above and to dene semantics for such logics in terms of
twist-structures. It may also be interesting to develop a general theory of
twist-structures.
Bibliography
[1] A. Almukdad and D. Nelson. Constructible falsity and inexact predi-

cates. Journal of Symbolic Logic, 49, No. 1, 1984, 231233.
[2] E.H. Alves, A.M. Sette. On the equivalence between some systems of
non-classical logic. Bulletin of the Section of Logic, 25, No. 2, 1996,
6872.
[3] A. Avron. On an implicative connective of RM . Notre Dame Journal

of Formal Logic, 27, No. 2, 1986, 201209.
[4] A. Avron. Natural 3-valued logics: characterization and proof theory.

Journal of Symbolic Logic, 56, No. 1, 1991, 276294.
[5] D. Batens. A completeness proof method for extensions of the implica-

tional fragment of the propositional calculus. Notre Dame Journal of
Formal Logic, 21, No. 3, 1980, 509517.
[6] D. Batens. Paraconsistent extensional propositional logics. Logique et

Analyse, Nouv. Ser., 23, 1980, 195234.
[7] D. Batens, K. De Clercq, N. Kurtonina. Embedding and interpolation

for some paralogics. The propositional case. Reports on Mathematical
Logic, 33, 1999, 2944.
[8] N.D. Belnap. A useful four-valued logic. In: G. Epstein and M.J. Dunn
(eds.). Modern Uses of Multiple-Valued Logic, Oriel Press, 1977, 737.
[9] N.D. Belnap. How computers should think. In: G. Ryle (ed.). Contem-
porary Aspects of Philosophy, Oriel Press, 1977, 3056.
[10] P. Bernays. Review of [20] and [21]. Journal of Symbolic Logic, 18,
No. 1, 1953, 266268.
227
228 Bibliography
[11] J.-Y. Beziau. What is paraconsistent logic? in: D. Batens et al. (eds.)
Frontiers of Paraconsistent Logic, Research Studies Press, Baldock,
2000, 95111.
[12] F. Borceux. Handbook of categorical algebra 1. Basic category theory.

Cambridge University Press, Cambridge, 1994.
[13] M. Bozic and K. Dosen. Models for Normal Intuitionistic Modal Logics.
Studia Logica, 43, 1984, 217245.
[14] S. Burris, H.P. Sankappanavar. A course in universal algebra. Gradu-

ate Texts in Math., 78, New York, Springer, 1981.
[15] A. Chagrov, M. Zakharyaschev. The undecidability of disjunction prop-

erty of propositional logics and other related problems. Journal of
Symbolic Logic, 58, No. 3, 1999, 9671003.
[16] A. Chagrov, M. Zakharyaschev. Modal logic. Clarendon Press, Oxford,

1997.
[17] R. Cignoli. The class of Kleene algebras satisfying interpolation prop-

erty and Nelson algebras. Algebra Universalis, 23, No. 3, 1986, 262
292.
[18] J.P. Cleave. The notion of logical consequence in the logic of inexact
predicates. Zeitschrift fur Mathematische Logic und Grundlagen der
Mathematik, 20, No. 3, 1974, 307324.
[19] N.C.A. da Costa, J.-Y. Beziau. Carnots logic. Bulletin of the Section
of Logic, 23, No. 2, 1994, 98105.
[20] H.B. Curry. On the denition of negation by a xed proposition in

the inferential calculus. Journal of Symbolic Logic, 17, No. 1, 1952,
98104.
[21] H.B. Curry. The system LD. Journal of Symbolic Logic, 17, No. 1,
1952, 3542.
[22] H.B. Curry, Foundations of mathematical logic, McGraw-Hill Book

Company, New York, 1963.
[23] K. Dosen. Negative modal operators in intuitionistic logic. Publications

of the Institute of Mathematics (Beograd), Nouv. Ser. 35(49), 1984,
1520.
Bibliography 229
[24] K. Dosen. Negation as a modal operator. Reports on Mathematical

Logic, 20, 1986, 1528.
[25] K. Dosen. Negation in the light of modal logic. in D. Gabbay, H. Wans-

ing (eds.) What is Negation?, Kluwer, Dordrecht, 1999, 7786.
[26] J.M. Dunn. Partiality and its Dual. Studia Logica, 66, No. 1, 1980,
540.
[27] J.M. Dunn, R.K. Meyer. Algebraic completeness results for Dum-
metts LC and its extensions. Zeitschrift fur Mathematische Logic und
Grundlagen der Mathematik, 17, No. 2, 1971, 225230.
[28] M.M. Fidel. An algebraic study of a propositional system of Nelson.

Mathematical Logic, Proc. of the First Brasilian Conf. on Math. Logic,
Campinas 1977, Lect. Notes Pure Appl. Math. 39, 1978, 99117.
[29] M.M. Fidel. An algebraic study of logic with constructive negation.

Proc. of the Third Brazilian Conf. on Math. Logic, Recife 1979, 1980,
119129.
[30] K. Fine. Logics containing K4. I., Journal of Symbolic Logic, 39, No. 1,
1974, 3142.
[31] K. Fine. Logics containing K4. II., Journal of Symbolic Logic, 50,
No. 3, 1985, 619651.
[32] J.M. Font, P. Hajek. On Lukasiewiczs four-valued modal logic. Studia

Logica, 70, No. 2, 2002, 157182.
[33] V. Goranko. The Craig interpolation theorem for propositional logics

with strong negation. Studia Logica, 44, No. 3, 1985, 291317.
[34] J. van Heijenoort. From Frege to Godel. A Source Book in Mathemat-

ical Logic, 18791931. Harvard University Press, Cambridge, 1967.
[35] A. Heyting. Die formalen Regeln der intuitionistischen Logik.

Sitzungsber. preuss. Akad. Wiss. Berlin, 1930, 4256.
[36] A. Heyting. Intuitionism. An introduction. North-Holland, Amster-

dam, 1971.
[37] V.A. Jankov. The relationship between deducibility in the intuitionistic

propositional calculus and nite implicational structures. Soviet Math.
Dokl., 4, 1963, 12031204.
230 Bibliography
[38] V.A. Jankov. Constructing a sequence of strongly independent superin-

tuitionistic propositional calculi. Soviet Math. Dokl., 9, 1968, 806807.
[39] S. Jaskowski. Propositional calculus for contradictory deductive sys-

tems. Studia Logica, 24, 1969, 143157.
[40] J. Jaspars. Calculi for constructive communication. ILLC Dissertation

Series 1994-4, ILLC, 1994.
[41] I. Johansson. Der Minimalkalkul, ein reduzierter intuitionistischer

Formalismus. Compositio Mathematika, 4, 1937, 119136.
[42] S. Kanger. A note on partial postulate sets for propositional logic.

Theoria, 21, No. 1, 1955, 99104.
[43] A.S. Karpenko. Two three-valued isomorphs of classical propositional

logic and their combinations. First World Congress on Paraconsis-
tency, Abstracts, Ghent, 1997, 9294.
[44] A.N. Kolmogorov. On the principle of excluded middle. Matematich-

eskij sbornik, 32, No. 4, 1925, 646667 [in Russian]. English transla-
tion [34], 414437.
[45] A. Kolmogoro. Zur Deutung der intuitionistischen Logik. Mathema-

tische Zeitschrift, 35, 1932, 5865.
[46] S. Korner. Experience and Theory. Kegan Paul, London, 1966.
[47] M. Kracht. On extensions of intermediate logics by strong negation.

Journal of Philosophical Logic, 27, No. 1, 1998, 4973.
[48] S.A. Kripke. Distinguished constituents (abstract). J. Symb. Logic, 24,

1959, 323.
[49] F. von Kutschera. Ein verallgemeinerter Widerlegungsbegri fur

Gentzenkalkule. Archiv fur Mathematische Logik und Grundlagen-
forschung, 12, No. 2, 1969, 104118.
[50] A. Loparic, N.C.A. da Costa. Paraconsistency, paracompleteness and

induction. Logique et Analyse, Nouv. Ser., 113, 1986, 7380.
[51] E.G.K. Lopez-Escobar. Refutability and Elementary Number Theory.

Indagationes Mathematicae, 34, 1972, 362374.
Bibliography 231
[52] J. L
ukasiewicz. A system of modal logic. The Journal of Computing
Systems, 1, 1953, 111149. Reprinted in [54], 352390.
[53] J. L
ukasiewicz. Aristotles syllogistic from the standpoint of modern
formal logic (2nd enlarged edition). Clarendon Press, Oxford, 1957.
[54] J. L
ukasiewicz. Selected works, edited by L. Borkowski. Stud. in Logic
and the Found. of Math. North-Holland, Amsterdam, 1970.
[55] L.L. Maksimova. Pretabular superintuitionist logic. Algebra and Logic,

11, No. 5, 1972, 308314.
[56] L.L. Maksimova. Craigs theorem in superintuitionistic logics and

amalgamable varieties of pseudo-Boolean algebras. Algebra and Logic,
16, No. 6, 1977, 427455.
[57] L.L. Maksimova. Implicit Denability and Positive Logics. Algebra

and Logic, 42, No. 1, 2003, 3753.
[58] L.L. Maksimova. Interpolation and Denability in Extensions of the

Minimal Logic. Algebra and Logic, 44, No. 6, 2005, 407421.
[59] A.A. Markov. Constructive logic. Usp. Mat. Nauk, 5, No. 3, 1950,
187188 (in Russian).
[60] A.A. Markov. On constructive mathematics. Proceedings of Steklov

Mathematical Institute, 67, 1962, 814 (in Russian).
[61] E. Mendelson. Introduction to mathematical logic, 2nd edition.

D. Van Nostrand Company, New York, 1979.
[62] C.M. McKay. On nite logics. Indag. Math., 29, No. 3, 1967, 363365.
[63] S. Miura. A remark on the intersection of two logics. Nagoja Math. J.,
26, No. 2, 1966, 167171.
[64] D. Nelson. Constructible falsity. Journal of Symbolic Logic, 14, No. 1,

1949, 1626.
[65] D. Nelson. Negation and separation of concepts. In: Constructivity in

Mathematics, North-Holland, Amsterdam, 1959, 208225.
[66] S.P. Odintsov. Propositional relatively constructive systems. Vychisli-

telnye sistemy, 158, Novosibirsk, 1997, 110126 (in Russian).
232 Bibliography
[67] S.P. Odintsov. Isomorphs of the logic of classical refutability and their
generalizations. Proceedings of the seminar of logical center, Inst. of
Philosophy of RAS, Moscow, 1998, 4861.
[68] S.P. Odintsov. Maximal paraconsistent extension of Johansson logic.

Logique et Analyse, Nouv. Ser, 161/163, 1998, 107120.
[69] S.P. Odintsov. Representation of j-algebras and Segerbergs logics. Log.

Anal., Nouv. Ser., 165/166, 1999, 81106.
[70] S.P. Odintsov. Logic of classical refutability and class of extensions of

minimal logic. Logic and Logical Philosophy, 9, 2002, 91107.
[71] S.P. Odintsov. On the embedding of Nelsons logics. Bull. Sect. Log.,
Univ. Lodz, Dep. Log., 31, No. 4, 2002, 241250.
[72] S.P. Odintsov. Algebraic semantics for paraconsistent Nelsons Logic.

Journal of Logic and Computation, 13, No. 4, 2003, 453468.
[73] S.P. Odintsov. Reductio ad Absurdum and L ukasiewiczs Modalities.

Logic and Logical Philosophy, 11, 2003, 149166.
[74] S.P. Odintsov. On representation of N4-lattices. Studia Logica, 76,

No. 3, 2004, 385405.
[75] S.P. Odintsov. Negative Equivalence of Extensions of Minimal Logic.

Studia Logica, 78, No. 3, 2004, 417442.
[76] S.P. Odintsov. On the Structure of Paraconsistent Extensions of

Johanssons Logic. Journal of Applied Logic, 3, No. 1, 2005, 4365.
[77] S.P. Odintsov. The Class of Extensions of Nelson Paraconsistent

Logic. Studia Logica, 80, No. 2-3, 2005, 291320.
[78] S.P. Odintsov. Transfer Theorems for Extensions of the Paraconsistent

Nelson Logic. Algebra and Logic, 45, No. 4, 2006, 232247.
[79] S.P. Odintsov. On Extensions of Nelsons Logic Satisfying Dummetts

Axiom. Siberian Mathematical Journal, 48, No. 1, 2007, 112125.
[80] S.P. Odintsov. The Lattice of Extensions of the Minimal Logic.

Siberian Advances in Mathematics, 17, No. 2, 2007, 112143.
Bibliography 233
[81] S.P. Odintsov. Absurdity as Unary Operator. in: J. Malinowski,

A. Pietruszczak (eds.) Essays in Logic and Ontology, Poznan Stud-
ies in the Philosophy of Sciences and the Humanities 91, Rodopi, NY,
2006, 225242.
[82] S.P. Odintsov, D. Pearce. Routley semantics for answer sets. in:
G.Ch. Baral et al. (eds.) Logic Programming and Nonmonotonic
Reasoning, 8th International Conference, LPNMR 2005, Diamante,
Italy, September 5-8, 2005, Proceedings (LNCS 3662), Springer, 2005,
343355.
[83] H. Ono. Kripke models and intermediate logics. Publ. of Research Inst.
for Math. Sci., Kyoto Univ., 6, 1970, 461476.
[84] J. Porte. The -system and the L-system of modal logic. Notre Dame
J. Formal Logic, 20, No. 4, 1970, 915920.
[85] J. Porte. Lukasiewiczs L-modal system and classical refutability.

Logique et Analyse, Now. Ser., 27, 8792 (1984).
[86] G. Priest. Paraconsistent logic. in: D. Gabbay, H. Guenter (eds.) Hand-

book of Philosophical Logic, Vol. 6, 2nd edition, Dordrecht: Kluwer,
2002, 287393.
[87] G. Priest, R. Routley, J. Norman (eds.) Paraconsistent logic. Essays

on the inconsistent. Munchen, Philosophia Verlag, 1989.
[88] A.P. Pynko. Algebraic study of Settes maximal paraconsistent logic.

Studia Logica, 54, No. 1, 1995, 89128.
[89] A.P. Pynko. Functional completeness and axiomatizability within

Belnaps four-valued logic and its expansions. Journal of Applied
Non-classical Logics, 9, No. 1, 1999, 61105.
[90] H. Rasiowa. N -lattices and constructive logic with strong negation.

Fundamenta Mathematicae, 46, No. 1, 1968, 6180.
[91] H. Rasiowa. Algebraische Charakterisierung der intuitionistischen

Logik mit starker Negation. in: A. Heyting (ed.) Constructivity in
Mathematics, Amsterdam, North-Holland, 1959, 234240.
[92] H. Rasiowa. An algebraic approach to non-classical logics. Amsterdam,

North-Holland, 1974.
234 Bibliography
[93] W. Rautenberg. Klassische und nichtclassische Aussagenlogik. Braun-

schweig, Vieweg, 1979.
[94] N. Rescher. Many-valued Logic. N.Y., 1969.
[95] T.T. Robinson. Independence of two nice sets of axioms for the propo-
sitional calculus. Journal of Symbolic Logic, 33, No. 2, 1968, 265270.
[96] R. Routley. Semantical analyses of propositional systems of Fitch and

Nelson. Studia Logica, 33, No. 3, 1974, 283298.
[97] K.F. Samokhvalov. Relatively constructive systems. Vychislitelnye sis-

temy, 124, Nowosibirsk, 1988, 99113 [in Russian].
[98] K. Segerberg. Propositional logics related to Heytings and Johans-

sons. Theoria, 34, No. 1, 1968, 2661.
[99] A. Sendlewski. Some investigations of varieties of N-lattices. Studia

Logica, 43, No. 3, 1984, 257280.
[100] A. Sendlewski. Nelson algebras through Heyting ones. Studia Logica,

49, No. 1, 1990, 106-126.
[101] A. Sendlewski. Axiomatic extensions of the constructive logic with

strong negation and disjunction property. Studia Logica, 55, No. 3,
1995, 377388.
[102] A.M. Sette. On the propositional calculus P 1 . Mathematica Japonicae,

18, No. 3, 1973, 173180.
[103] M.V. Stukacheva. On disjunction property in the class of extensions

of minimal logic. Algebra and Logic, 43, No. 2, 2004, 132141.
[104] M.V. Stukacheva. Some Remarks on Constrctive Extensions of Mini-

mal Logic. Vestnik NGU, Ser.: Mathematics, Mechanics and Informat-
ics, 5, No. 3, 2005, 316 (in Russian).
[105] M.V. Stukacheva. On canonical formulas for the extensions of minimal

logic. Siberian Electronic Mathematical Reports, 3, 2006, 312334.
[106] N.-Y. Suzuki. Constructing a continuum of predicate extensions of

each intermediate propositional logics. Studia Logica, 54, No. 2, 173
198 (1995).
Bibliography 235
[107] R. Thomason. A semantical study of constructive falsity. Zeitschrift

fur mathematische Logik und Grundlagen der Mathematik, 15, No. 3,
1969, 247257.
[108] I. Urbas. A note on Carnots logic. Bulletin of the Section of Logic,

23, No. 3, 1994, 118125.
[109] D. Vakarelov. Theory of negation in certain logical systems. Algebraic

and semantical approach. Ph.D. dissertation, University of Warsaw,
1976.
[110] D. Vakarelov. Notes on N -lattices and constructive logic with strong

negation. Studia Logica, 36, No. 12, 1977, 109125.
[111] D. Vakarelov. Consistency, completeness and negation. In: [87], 328

363.
[112] D. Vakarelov. Nelsons negation on the base of weaker versions of in-

tuitionistic negation. Studia Logica, 80, 2005, 393430.
[113] D. Vakarelov. Non-classical negation in the works by Helena Rasiowa

and their impact on the theory of negation. Studia Logica, 84, 2006,
105127.
[114] N.N. Vorobev. A constructive propositional calculus with strong nega-

tion. Doklady Akademii Nauk SSSR, 85, 1952, 465468 (in Russian).
[115] N.N. Vorobev. The problem of deducibility in constructive proposi-

tional calculus with strong negation. Doklady Akademii Nauk SSSR,
85, 1952, 689692 (in Russian).
[116] N.N. Vorobev. Constructive propositional calculus with strong nega-

tion. Transactions of Steklovs Institute, 72, 1964, 195227 (in
Russian).
[117] G. Wagner. Vivid logic. Knowledge-based reasoning with two kinds of

negation. Springer LNAI 764, 1994.
[118] H. Wansing. The logic of information structures. Springer LNAI 681,

1993.
[119] H. Wansing. Semantics-based nonmonotonic inference. Notre Dame

Journal of Formal Logic, 36, No. 1, 1995, 4454.
236 Bibliography
[120] H. Wansing. Negation. In: L. Goble (ed.), The Blackwell Guide to

Philosophical Logic, Basil Blackwell Publishers, Cambridge/MA, 2001,
415436.
[121] H. Wansing. Diamonds are a philosophers best friends. Journal of

Philosophical Logic, 31, No. 6, 2002, 591612.
[122] P. Wojtylak. Mutual interpretability of sentential logic. I. Reports on

Mathematical Logic, 11, 1981, 6989.
[123] P. Wojtylak. Mutual interpretability of sentential logic. II. Reports on

Mathematical Logic, 12, 1981, 5166.
[124] P.W. Woodru. A note on JP . Theoria, 36, No. 2, 1970, 183184.
[125] A. Wronski. The degree of completeness of some fragments of the intu-

itionistic propositional logic. Reports on Mathematical Logic, 2, 1974,
5562.
[126] A. Wronski. On the cardinalities of matrices strongly adequate for the

intuitionistic propositional logic. Reports on Mathematical Logic, 3,
1974, 6772.
Index
absurdity operator, 7, 102, 112 Deduction Theorem, 17

algebra, 21 deductive system, 15
Boolean algebra of regular denition equivalent, 103
elements, 25 disjunction property, 85
De Morgan algebra, 145 dual lattice isomorphism, 24
Heyting l-algebra, 66 Dummetts linearity axiom, 63, 177
Heyting algebra, 23
basic, 156 element
j-algebra, 23 dense, 25
lower, 42 distinguished, 22
negative, 24 regular, 25
negative l-algebra, 66 equational theory, 22
Peirce algebra, 26 extensional operator, 109, 115
Peirce-Johansson algebra, 33
F-structure, 135
pj-algebra, 33
saturated, 135
Stone algebra, 66
faithfully embedded via, 103
subdirectly ireducible, 25
lter, 24
upper, 42
generated by, 24
amalgam, 213
of completions, 160
amalgamation property, 213
of dense elements, 25
special lter of the rst kind, 169
canonical embedding, 156
sk, 169
complement, 23
fragment
conservative extension, 16
L1 -fragment, 16
contradiction operator, 6, 52, 109
positive, 16
counterpart
free combination, 51
explosive, 195
intuitionistic, 47, 48 Generalized Glivenkos Theorem, 58
negative, 47, 48
normal, 195 ideal, 161
normal explosive, 195 of contradictions, 161
Craig interpolation property, 213 implicative lattice, 23
237
238 Index
inference relation, 16 intermediate, 35

interpretation, 119 intuitionistic logic, 18
intuitionistic formula, 177 Johanssons logic, 17
isomorph, 35 logic of classical refutability,
18
j-frame, 28 maximal negative logic, 18
abnormal, 30 minimal logic, 17
closed, 30 minimal negative logic, 18
dense, 30 ne-logic, 178
identical, 30 negative, 35
normal, 30
Nelsons logic
separated, 30
explosive, 133
Jankov formula, 92
paraconsistent, 133
Jonssons theorem, 202
normal, 191
kernel, 169 normal explosive, 178
of nite codimension, 211
law of general form, 191
contraction law, 19 paraconsistent, 1
distributivity law, 19 positive logic, 16
ex contradictione quodlibet, 1 pretabular, 201
extended law of excluded properly paraconsistent, 35
middle, 17 s-logic, 178
identity law, 19 se-logic, 193
import and export of the sn-logic, 193
premiss, 19 special, 178
Peirce law, 17 special explosive, 193
permutation law, 19 special normal, 193
reductio ad absurdum, 17 tabular, 201
weak law of excluded middle, 63 trivial logic, 16
logic, 15 -logic, 113
A-logic, 112
C-logic, 109 matrix, 22
L
ukasiewiczs modal logic, 105 model, 22
characterized by, 29 characteristic, 22
classical logic, 18
classical positive logic, 17 nnf, 134
explosive, 1, 191 N3-lattice
extensional, 116 normal, 179
Glivenkos logic, 57 special, 193
relativized, 84 N4-lattice, 145
Index 239
N4 -lattice, 156 weak replacement rule, 135

special, 179 world
special normal, 193 abnormal, 28
negation normal form, 134 normal, 28
negatively equivalent, 83
negatively lesser, 81 2, 26
2 , 26
opremum, 25 2P , 26
4 , 34
particular case, 15
4L , 105
presentation
A B, 24
A-presentation, 113
A , 42
exact, 113
A , 139
C-presentation, 113
A3 , 193
exact, 113
A , 179
strong, 114
An , 193
-presentation, 114
A , 42
exact, 114
A , 145
prime theory, 120
A B, 22
propositional language, 15
A B, 202
relative pseudo-complement, 23 B f C, 61
repeatedless disjunction, 20 chn (k, ), 203
CIP , 213
splitting, 189 CLuN, 119
splitting element, 189 Con(A), 27
strong equivalence, 133 D(A), 92
strongly independent sequence, 93 D A , 147, 148
substitution instance, 15 (B), 162
sum of logics, 18 l (B), 164
theorem, 15 e , 164
twist-operations, 139 EL, 18
twist-structure, 139, 155 Eq(A), 22
full, 139, 155 Eq(K), 22
Lindenbaum twist-structure, 141 neg , 83
(L), 178
valuation, 22, 28 3 (L), 193
variety (L), 178
arithmetic, 27 n (L), 193
congruence distributive, 27 Exp, 190
congruence permutable, 27 Fd (A), 25
240 Index
F or, 133 |= , 139

F or , 133 |=F , 137
G(L1 , L2 ), 84 N3, 133
Gen, 191 N3 , 183
H(K), 22 N4, 133
h , 167 N4N , 190
h , 167 N4 , 134
I(B), 162 N4 C, 201
I(K), 22 N4C, 201
i , 164 (B), 162
Int, 35 , 172
J(A), 92 , 172
Jhn, 18 l (B), 164
Jhn+ , 18 (L), 88
K , 178 Neg, 35
Ker(h), 169 Nor, 190
L(V ), 153 P 1 , 123
L1 L2 , 50, 198 Par, 35
L
, 141 , 179
Le, 18 3 , 193
Le , 41 , 179
neg , 81 n , 193
Lexp , 195 R(A), 25
Lg, 57 S(K), 22
Li, 18 S (A), 139
Lint , 48 , 133
Lj , 17 (L), 177
Lj , 17 3 , 193
Lk, 18 n , 193
Lk+ , 17 iI Li , 18
Lmn, 18 Spec(L1 , L2 ), 50, 198
Ln, 18 A , 25
Lne , 195 T w(A, , ), 162
Lneg , 48 Up(K), 22
Lnor , 195 VN4 , 148
Lp, 16 V ar(L), 152
L
, 105
TRENDS IN LOGIC
1. G. Schurz: The Is-Ought Problem. An Investigation in Philosophical Logic. 1997

ISBN 0-7923-4410-3
2. E. Ejerhed and S. Lindstrom (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 of Fuzzy Logic. 1998 ISBN 0-7923-5238-6
5. H.J. Ohlbach and U. Reyle (eds.): Logic, Language and Reasoning. Essays in Honour
of Dov Gabbay. 1999 ISBN 0-7923-5687-X
6. K. Dosen: Cut Elimination in Categories. 2000 ISBN 0-7923-5720-5
7. R.L.O. Cignoli, I.M.L. DOttaviano and D. Mundici: Algebraic Foundations of many-
valued Reasoning. 2000 ISBN 0-7923-6009-5
8. E.P. Klement, R. Mesiar and E. Pap: Triangular Norms. 2000
ISBN 0-7923-6416-3
9. V.F. Hendricks: The Convergence of Scientic Knowledge. A View From the Limit.
2001 ISBN 0-7923-6929-7
10. J. Czelakowski: Protoalgebraic Logics. 2001 ISBN 0-7923-6940-8
11. G. Gerla: Fuzzy Logic. Mathematical Tools for Approximate Reasoning. 2001
ISBN 0-7923-6941-6
12. M. Fitting: Types, Tableaus, and Godels God. 2002 ISBN 1-4020-0604-7
13. F. Paoli: Substructural Logics: A Primer. 2002 ISBN 1-4020-0605-5
14. S. Ghilardi and M. Zawadowki: Sheaves, Games, and Model Completions. A Cat-
egorical Approach to Nonclassical Propositional Logics. 2002
ISBN 1-4020-0660-8
15. G. Coletti and R. Scozzafava: Probabilistic Logic in a Coherent Setting. 2002
ISBN 1-4020-0917-8; Pb: 1-4020-0970-4
16. P. Kawalec: Structural Reliabilism. Inductive Logic as a Theory of Justication. 2002
ISBN 1-4020-1013-3
17. B. Lowe, W. Malzkorn and T. Rasch (eds.): Foundations of the Formal Sciences
II. Applications of Mathematical Logic in Philosophy and Linguistics, Papers of a
conference held in Bonn, November 10-13, 2000. 2003 ISBN 1-4020-1154-7
18. R.J.G.B. de Queiroz (ed.): Logic for Concurrency and Synchronisation. 2003
ISBN 1-4020-1270-5
19. A. Marcja and C. Toffalori: A Guide to Classical and Modern Model Theory. 2003
ISBN 1-4020-1330-2; Pb 1-4020-1331-0
20. S.E. Rodabaugh and E.P. Klement (eds.): Topological and Algebraic Structures in
Fuzzy Sets. A Handbook of Recent Developments in the Mathematics of Fuzzy Sets.
2003 ISBN 1-4020-1515-1; Pb 1-4020-1516-X
21. V.F. Hendricks and J. Malinowski: Trends in Logic. 50 Years Studia Logica. 2003
ISBN 1-4020-1601-8
22. M. Dalla Chiara, R. Giuntini and R.Greechie: Reasoning in Quantum Theory. Sharp
and Unsharp Quantum Logics. 2004 ISBN 1-4020-1978-5
23.
B. Lowe,
B. Piwinger and T. Rasch (eds.): Classical and New Paradigms of Computa-
tion and their Complexity Hierarchies. Papers of the conference Foundations of the
Formal Sciences III held in Vienna, September 2124, 2001. 2004
ISBN 1-4020-2775-3
24. G. Jager:
Anaphora and Type Logical Grammar. 2005 ISBN 978-1-4020-3904-1
25. M. Winter: Goguen Categories. A Categorical Approach to L-fuzzy Relations. 2007
ISBN 978-1-4020-6163-9
26. S.P. Odintsov: Constructive Negations and Paraconsistency. 2008
ISBN 978-1-4020-6866-9
springer.com

Sergei P. Odintsov Constructive Negations and Paraconsistency

Enviado por

Dados do documento

Direitos autorais

Formatos disponíveis

Compartilhar este documento

Compartilhar ou incorporar documento

Opções de compartilhamento

Você considera este documento útil?

Este conteúdo é inapropriado?

Direitos autorais:

Formatos disponíveis

Sergei P. Odintsov Constructive Negations and Paraconsistency

Enviado por

Direitos autorais:

Formatos disponíveis

Constructive Negations and Paraconsistency

SCOPE OF THE SERIES

ISBN 978-1-4020-6866-9 e-ISBN 978-1-4020-6867-6

Library of Congress Control Number: 2007940855

2008 Springer Science+Business Media B.V.

Printed on acid-free paper.

2 Minimal Logic. Preliminary Remarks 15

3 Logic of Classical Refutability 31

4 The Class of Extensions of Minimal Logic 41

5 Adequate Algebraic Semantics for Extensions

6 Negatively Equivalent Logics 81

II Strong Negation 129

to study the phenomenon of contradiction itself. Namely this formal logical

Only in 1937 I. Johansson [41] questioned the using of ex contradictione

This means that the negation makes no sense in inconsistent Lj-theories,

constructive point of view. When the negation of a conjunction ( )

Jhn = Int Neg Par.

Note that negative logics are denition equivalent to positive ones.

The advantage of the language with intuitionistic negation becomes

In Chapter 6, we introduce the negative equivalence of logics (see above),

Minimal Logic. Preliminary

2.1 Denition of Basic Logics

If (1 , . . . , n ) is obtained from (p1 , . . . , pn ) by the substitution rule, we

In this case we say also that L1 is an L1 -fragment of L2 . In what follows by

2. (p (q r)) ((p q) (p r))

8. (p r) ((q r) ((p q) r))

Positive logic satises Deduction Theorem:

Moreover, Lj () for any formula .

and in the language L as follows:

Lk = Li+{p(p q)}, Le = Lj+{p(p q)}, Lmn = Ln+{p(p q)}.

Proposition 2.1.2 The following formulas are provable in Lp:

1. p p (the identity law).

2. (p q) (q p), (p q) (q p) (the commutativity of and ).

3. (p (q r)) ((p q) r), (p (q r)) ((p q) r)

4. (p (q r)) ((p q) (p r)), (p (q r)) ((p q) (p r))

5. (p (q r)) (q (p r)) (the permutation law).

6. (p (p q)) (p q) (the contraction law).

7. (p (q r)) ((p q) r) (import and export of the premiss).

8. ((p q) r) ((p r) (q r)).

9. ((p q) r) ((p r) (q r)) r.

Proposition 2.1.3 The following formulas are provable in Lj:

Proposition 2.1.4 [98]

1. The equality Lk = Lj + {} holds, where is one of the following

The next proposition shows how to construct axioms of an intersection

Proposition 2.1.5 Let L {Lp, Lj}, L1 = L + {i | i I}, and L2 :=

Proof. Suppose L1 L2 . By Deduction

substitution instances of i and j respectively. Using axiom 8 of positive

from which L + {i j | i I, j J}.

Proof. Let L {Lp, Lj}. We prove only that

L, Li EL. Assume L = L + and Li = L + i for i I.

2.2 Algebraic Semantics

A, f1A , . . . , fkA of the language L is a set, where the connectives of L are

of universes |A1 | . . . |An | and the operations are componentwise. Note

Since the operations on the direct product are componentwise, we have

L(M1 . . . Mn ) = LM1 . . . LMn .

Proposition 2.2.1 [93] Every Lj-extension has a characteristic model.

The reader is expected to be familiar with lattices and with distributive

Proposition 2.2.2 [92] Let A be an implicative lattice and a, b A. Then

By a j-algebra we mean an algebra A =

A negative algebra is a j-algebra with = 1. Obviously, negative algebras

and for a variety V Sub(Vj ), dene a logic

L(V ) := { | A |= for all A V }.

It is clear that for any L Jhn+ and V Sub(Vj ) we have

V (L) Sub(Vj ) and L(V ) Jhn+ .

then for every a A, a = 1, the set {a, A } is the universe of a subalgebra

|=x y W (x y ( |=y y Q)).

W, , Q abnormal if Q = W , i.e., if all worlds are abnormal. Finally, a

1 and [, 1]A = A, then A contains an element incomparable

Theorem 3.1.7 Let Le. There are three possible cases: