Escolar Documentos
Profissional Documentos
Cultura Documentos
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
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
9 8 7 6 5 4 3 2 1
springer.com
Contents
1 Introduction 1
I Reductio ad Absurdum 13
v
vi Contents
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
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
:= 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
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
(p1 , . . . , pn )
and .
(1 , . . . , n )
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 .
1. p (q p)
3. (p q) p
4. (p q) q
5. (p q) ((p r) (p (q r)))
6. p (p q)
7. q (p q)
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 + {};
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.
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
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)
20 2 Minimal Logic. Preliminary Remarks
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.
(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)}.
Proposition 2.1.6 The lattices ELp and ELj are distributive. Moreover,
the intersection distributes with the innite sum in these lattices.
L iI Li = iI (L Li ),
M1 . . . Mn =
A1 . . . An , DA1 . . . DAn .
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}.
2.2 Algebraic Semantics 23
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
V (L) := {A | A Vj , A |= L},
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.
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
26 2 Minimal Logic. Preliminary Remarks
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 2.2.9 [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.
hold in V.
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},
|=x p x v(p).
2.3 Kripke Semantics 29
And further,
|=x x Q.
LK := { | W K (W |= )}.
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 =
Logic of Classical
Refutability1
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.
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
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
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
1. Le + {} is trivial;
2. Le + {} = Lmn;
3. Le + {} = Lk.
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
36 3 Logic of Classical Refutability
x (x)
1 1
a 1
0 0
1 0
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
x (x)
1 1
a a
0 1
1 a
x y := (x y) = (x) (y), {, , },
x := (x).
38 3 Logic of Classical Refutability
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
x (x)
1 0
a 1
0 0
1 1
x y := (x y) = ( (x) (y)),
x := (x).
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
1. Lk Le ;
2. Lmn Le ;
3. Lmn Le ( ) Le .
v1 ((p1 , . . . , pn )) = v((p1 , . . . , pn )) = 1.
Le Le ( ) Le .
v( ) = v( ( )) = 1.
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.
41
42 4 The Class of Extensions of Minimal Logic
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.
(x a) (y a) (x y) a.
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.
1. L1 L .
2. 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.
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.
interesting logics, which do not belong to this interval. One of them is the
Glivenko logic treated in the beginning of the next chapter.
In() = (p0 , . . . , pn ).
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 .
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 |= .
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 },
Spec(L1 , L2 ) = [L1 L2 , L1 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.
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 ) ).
(( j 1 ) . . . ( j n )) ( ),
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.
eP1 ,P2 eP2 ,P3 = eP1 ,P3 , rP3 ,P2 rP2 ,P1 = rP3 ,P1 .
Proof. 1. We calculate
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
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.
Adequate Algebraic
Semantics for Extensions
of Minimal Logic1
P . (( p) ) = ( p).
57
58 5 Adequate Algebraic Semantics for Extensions of Minimal Logic
1. For any ,
Lk L .
2. L Lg and L Neg.
5.2 Representation of j-Algebras 59
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 ( ( y)) = x,
5.2 Representation of j-Algebras 61
All other relations, except the last, are trivial. Check the last relation
using the obtained formula for relative pseudo-complement. We have
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) ) )
(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
5.3 Segerbergs Logics and their Semantics 65
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
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).
5. A |= Lg if and only if fA (A ) Fd (A ).
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.
((x, y) (, )) (((x, y) (, )) (, )) = (x , 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 = .
68 5 Adequate Algebraic Semantics for Extensions of Minimal Logic
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
2. Ljx = Lk Ln.
5.3 Segerbergs Logics and their Semantics 69
3. For all L1 Int and L2 Neg, we have (L1 L2 )p , (L1 L2 )l
Spec(L1 , L2 ) and the following equality holds
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 .
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.
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
70 5 Adequate Algebraic Semantics for Extensions of Minimal Logic
f (a b) f (b a) = f (b) f (a) = = ,
Using Proposition 5.3.1 and Corollary 5.3.2 we can easily check that for all
logics L P ar the equalities
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
72 5 Adequate Algebraic Semantics for Extensions of Minimal Logic
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)
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.
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 ).
5.3 Segerbergs Logics and their Semantics 75
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
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
5.3 Segerbergs Logics and their Semantics 77
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
Figure 5.7
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.
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
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
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.
2
For a class of j-frames K, we dene
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
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.
1. L1 neg L2 .
L1 = L2 .
X L1 (1 . . . k ),
L1 (1 . . . m ) (1 . . . k ).
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 .
84 6 Negatively Equivalent Logics
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 ):
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 ).
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.
| ( p) (( ( p) and | ( p) ) | ).
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.
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.
(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+ .
1. L (L).
3. (L) Jhn+ .
4. (L) = (L).
v( ) v(( ) ) = v(( ( )) )
= v(( ) ) v( ) (A).
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
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 .
90 6 Negatively Equivalent Logics
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}.
Lg = Lj + { | Lk}.
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.
(p ) (p ) = In(p p) Lk Ln.
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.
92 6 Negatively Equivalent Logics
1. J(A) LB.
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
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.
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:
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
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.
96 6 Negatively Equivalent Logics
X := { | D(A), = 1 X}
X |=B
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)}).
Jhn+ / neg , neg
=
Spec(Lk, Ln) {F}, .
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 : 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
98 6 Negatively Equivalent Logics
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
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.
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. 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
L1 () L2 .
L1 () and L2 ().
((1 , . . . , n )) = (1 , . . . n )
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
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.
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
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.
108 7 Absurdity as Unary Operator
, Lj .
p (q p)
(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
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),
then C is extensional in L.
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.
(p (q (q C(q))) (p C(p)).
L (p C(p0 p0 )) p.
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
and, further,
p p C(p) p (p ) p .
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.
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))
() := C(); (C()) := ;
L := { | S} (7.3)
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},
L p p p.
(p) (p p) = (p p p).
L p p p.
LA := {A | L}.
L i LA A .
LA A( A ) (() A(())),
Proof. 1. The set C of formulas in the language L we dene as the least set
of formulas satisfying the following conditions:
3) if C, then C.
L := {C | LC },
p (p p) (p (p p)).
L L L C L ,
L2 L L1 .
L1
for any .
2
7.4 A- and C -Presentations 119
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
U . If () = 0, then () = 1.
120 7 Absurdity as Unary Operator
(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.
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 .
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
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
122 7 Absurdity as Unary Operator
( ) = ( ( )) = ().
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
U C . If () = 0, then (C()) = 0.
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).
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:
:= ( ) ( ),
:= ((( ) ) (( ) )) ( ).
124 7 Absurdity as Unary Operator
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.
7.4 A- and C -Presentations 125
1A . A( A) ;
2A . ( A) , where is molecular.
( ) (( ( A)) ( A).
( ( A)) ( A)
( ( ( A))) A ( ( A)) A
( A) ( A).
In particular, we have
L ( A) (( A) A( A)).
(( A) A( A)) , (7.4)
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).
1C ) C , where is molecular.
1C . If = 0, then C = 0.
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
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);
A3) (p q) p;
A4) (p q) q;
A6) p (p q);
A7) q (p q);
A9) p p;
A10) (p q) ( p q);
A11) (p q) ( p q);
A12) (p q) (p q).
A13) p (p q).
134 8 Semantical Study of Paraconsistent Nelsons Logic
N3 p, p .
p = p p = p
= # = #
( ) = ( ) =
( ) = =
N4(N4 ) .
Proposition 8.1.3 The logics N4 and N4 are closed under the weak re-
placement rule
0 1 0 1
.
(0 ) (1 )
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;
N4 .
[] [] := [ ] , 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.
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.
(1 , . . . , n ) |=A
F (1 , . . . , n ).
2
138 8 Semantical Study of Paraconsistent Nelsons Logic
1) N4 ;
2) |=F ;
3) |=A
F for any saturated F-structure A;
4) |=L
F .
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
A =
A 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
Na = {b A | (a, b) B}
for any a |A|. The correctness of this denition can be veried directly.
140 8 Semantical Study of Paraconsistent Nelsons Logic
|=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:
2
This section was originally published in [71].
8.3 Twist-structures 143
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
( ( )) = ( ) ( )
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 ) .
144 8 Semantical Study of Paraconsistent Nelsons Logic
( ) ( ( )) ( ( ))
( ) (( ) ( ) )
(( ( ) ) ) ( ( ( ) )) ( ) ( ) .
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
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].
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.
4. For any a, b A, (a b) a b.
First we check that the twist-structures belong to the just dened class
of lattices.
146 8 Semantical Study of Paraconsistent Nelsons Logic
h(a) := ([a] , [ a] ).
h( a) = ([ a] , [ a] ) = ([ a] , [a] ) = h(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 .
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.
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)
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 .
8.5 The Variety of N4-Lattices 149
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.
A VN4 A is an N4-lattice.
(a c) (b d) = (a (b d)) (c (b d)) DA ,
c c (a b) if and only if a c b a c,
8.5 The Variety of N4-Lattices 151
a) p p = p p;
b) (p p) q q;
c) q q (p q) (p q).
b b (a a) ( b ((a a) b)) = b b.
(a b) (a b) = (a b) ((a a) b) = [by 2N ]
a (b b) a (a b) b ( b a) b ( a a).
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
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).
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 ).
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.
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).
A =
A A, , , , ,
such that
A A, , , , is the full twist-structure over the im-
plicative lattice
A, , , , 1 and = (0, 1).
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
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
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
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.
Proof. First, we obtain a more convenient denition for I(B) and a repre-
sentation of B via I(B).
We now turn to the proof of the proposition and show that (B) is an
ideal of A. We need one more lemma.
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
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)).
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
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
= (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)) =
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 ).
1. If a, b D A , then a b DA .
2. If a A and b D A , then a $ b;
4. For any a, b A, a (a b) $ b.
a (a a) a (a a) a a.
9.2 Homomorphisms and Subdirectly Irreducible N4 -Lattices 171
(a a) a ( a a) a DA .
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.
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.
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
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 | = {[] | F},
174 9 N4 -Lattices
L := {([0 ] , [1 ] ) | 0 1 L}.
([] , [] ) ([ ] , [( ) ( )] )
[ ] F [ ] [( ) ( )] F .
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
9.2 Homomorphisms and Subdirectly Irreducible N4 -Lattices 175
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.
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
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.
177
178 10 The Class of N4 -Extensions
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
:=
.
180 10 The Class of N4 -Extensions
2. A
|= if and only if A |= .
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
Corollary 10.1.6 For any L Int+ , the logics (L) and (L) are conser-
vative extensions of L.
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.
182 10 The Class of N4 -Extensions
1. L is a special logic if and only if all rules of the form / are ad-
missible in L.
eL1 ,L2 eL2 ,L3 = eL1 ,L3 and rL3 ,L2 rL2 ,L1 = rL3 ,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
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
tq
@
@
q@ @q &
@
@q
f
M =
{f, t, &}, , , , , {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
{2 }, {2
R , 2L }, {2R }, {2L }, {2 }.
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
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.
10.2 The Lattice Structure of EN4 189
(2,q 0)
6
@
@
@
(0, 1) q -@q(1, 0)
@
@
@
@q?
(0, 2)
and
EN4 /{RM3 , L
3 } = N3 .
190 10 The Class of N4 -Extensions
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 ).
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:
2. Exp = EN3.
5. Nor = EN4N .
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.
10.2 The Lattice Structure of EN4 193
Denote
3 := Exp and n := Nor .
The mappings 3 : Int+ Exp and n : Int+ Nor are dened as follows.
For every L Int+ ,
n := (p p) .
pvar()
194 10 The Class of N4 -Extensions
1. A 3
3 |= if and only if A |= .
n |= if and only if A |= .
2. A n
K3 := {A
3 | A K} and Kn := {An | A K}.
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.
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}).
10.3 Explosive and Normal Counterparts 197
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,
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}.
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.
10.3 Explosive and Normal Counterparts 199
1. Spec(L1 , L2 ) 1 (L).
{exp | L1 } {nor | L2 } L
200 10 The Class of N4 -Extensions
( 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
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
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.
LA LB B HS({A}).
{[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 (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
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 ))
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
206 10 The Class of N4 -Extensions
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
En := pk pm .
1k<mn+1
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}.
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.
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
A1
@
i1 @ 1
@@
R
A0 A
@
i@
2 2
@
R
@
A2
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 }.
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
10.5 Some Transfer Theorems for the Class of N4 -Extensions 215
= if and only if v() = v() (*)
= , a Ai .
i (a) := [pia ]
:= {z A0 | a 1 z} and := {z A0 | z 2 b}.
F1 = {x A1 | a z 1 x for some z F0 }.
216 10 The Class of N4 -Extensions
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
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.
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,
218 10 The Class of N4 -Extensions
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 .
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
L | = {([]L , [ ]L ) | []L },
|L
L L.
([0 ]L , [0 ]L ) ([1 ]L , [1 ]L ) F .
[0 0 ]L , [1 1 ]L F.
a a = ([p0 p0 ]F , [p0 p0 ]F ),
220 10 The Class of N4 -Extensions
(p0 p0 ) (q0 , . . . , qn ) L.
F1 := {[]L | (p0 p0 ) L}
(p p) L or & (q q) L.
Both alternatives conict with our assumption. Let us consider the lter
Conclusion
223
224 11 Conclusion
p (p ) &.
,
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
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 ()
,
() ()
[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.
[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.
[13] M. Bozic and K. Dosen. Models for Normal Intuitionistic Modal Logics.
Studia Logica, 43, 1984, 217245.
[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.
[21] H.B. Curry. The system LD. Journal of Symbolic Logic, 17, No. 1,
1952, 3542.
[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.
[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.
[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.
[59] A.A. Markov. Constructive logic. Usp. Mat. Nauk, 5, No. 3, 1950,
187188 (in Russian).
[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.
[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.
[71] S.P. Odintsov. On the embedding of Nelsons logics. Bull. Sect. Log.,
Univ. Lodz, Dep. Log., 31, No. 4, 2002, 241250.
[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.
[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.
237
238 Index
springer.com