-
FEDERAL UNIVERSITY OF RIO GRANDE DO NORTECENTER OF EXACT AND
EARTH SCIENCES
DEPARTMENT OF INFORMATICS AND APPLIED MATHEMATICSPROGRAM OF
GRADUATE STUDIES IN SYSTEMS AND COMPUTING
DOCTORATE IN COMPUTER SCIENCE
On Algebras for Interval-Valued Fuzzy Logic
Antônia Jocivania Pinheiro
Natal-RNAugust, 2019
-
Antônia Jocivania Pinheiro
On Algebras for Interval-Valued Fuzzy Logic
This thesis was submitted to the PostgraduateProgram in Systems
and Computing of FederalUniversity of Rio Grande do Norte.Area of
Concentration: Fundations for Com-puting.
Advisor: Prof. Dr. Regivan Hugo Nunes Santiago
Natal-RN
August, 2019
-
Pinheiro, Antônia Jocivania. On algebras for interval-valued
fuzzy logic / AntoniaJocivania Pinheiro. - 2019. 135f.: il.
Tese (Doutorado)-Universidade Federal do Rio Grande do
Norte,Centro de Ciências Exatas e da Terra, Pós-Graduação em
Sistemase Computação, Natal, 2019. Orientador: Dr. Regivan Hugo
Nunes Santiago.
1. Interval Mathematics - Tese. 2. Fuzzy Logic - Tese. 3.
BCIAlgebras - Tese. 4. SBCI Algebras - Tese. 5. Fuzzy Implications-
Tese. I. Santiago, Regivan Hugo Nunes. II. Título.
RN/UF/BCZM CDU 004.032.26
Universidade Federal do Rio Grande do Norte - UFRNSistema de
Bibliotecas - SISBI
Catalogação de Publicação na Fonte. UFRN - Biblioteca Central
Zila Mamede
Elaborado por Raimundo Muniz de Oliveira - CRB-15/429
-
ABSTRACT
This work aims to introduce other approaches to the
interval-valued fuzzy logic. Thesenew approaches were inspired by
Lodwick and Chalco’s works on constraint intervals. Theseconstraint
intervals were used in this thesis to extend the fuzzy operators
into two modes, namedSingle-Level Constrained Interval Operators
and Constrained Interval Operators and studiedtheir properties. A
new algebra, called SBCI algebra, which arises from the
intervalization ofBCI-algebras, is also introduced. These algebras
aims to be the algebraic model for interval-valued fuzzy logics,
which take into account the notion of correctness.
A new class of fuzzy implications, called (T,N)-implications has
also been studied. Theauthor investigated the behavior of the
BCI/SBCI algebras and (T,N)-implications.Keywords: Interval-valued
fuzzy logic, Interval Mathematics, Fuzzy Logic, BCI algebras,
SBCIalgebras, Fuzzy Implications.
-
RESUMO
Este trabalho visa introduzir outras abordagens para a lógica
fuzzy com valores intervalares.Essas novas abordagens foram
inspiradas nos trabalhos de Lodwick e Chalco sobre
intervalosrestritos. Esses intervalos restritos foram usados para
estender os operadores fuzzy, nos quaiseles foram chamados
Operadores Intervalares Restritos de Nı́vel Único (C-operador) e
suaspropriedades foram estudadas. Além disso, esses operadores
foram estendidos a operadores cor-retos chamados Operadores
Intervalares Restritos. Uma nova álgebra, chamada SBCI
álgebra,que surge da intervalização de BCI álgebras, também é
introduzida. Essas álgebras têm comoobjetivo ser o modelo
algébrico para lógicas fuzzy com valores intervalares que levam
em contaa noção de correção.
Também foi estudada uma nova classe de implicações fuzzy,
chamada (T,N)-implicações.O autor investigou o comportamento das
BCI/SBCI álgebras e das (T,N)-implicações.Palavras-chave:
Lógica Fuzzy com Valores Intervalares, Matemática Intervalar,
Lógica Fuzzy,BCI álgebras, SBCI álgebras, Implicações
Fuzzy.
-
Contents
1 Introduction p. 8
I Preliminaries 12
2 Interval Arithmetics and Logics p. 13
2.1 Arithmetics . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . p. 13
2.1.1 Standard Interval Arithmetic . . . . . . . . . . . . . . .
. . . . . . . . . p. 13
2.1.1.1 Intervalization of Structures . . . . . . . . . . . . .
. . . . . p. 14
2.1.2 Constrained Interval Arithmetic and Single-Level
Constrained Inter-val Arithmetic . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . p. 15
2.2 Logics . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . p. 17
2.2.1 Fuzzy Connectives . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . p. 18
2.2.2 Interval Fuzzy Connectives . . . . . . . . . . . . . . . .
. . . . . . . . p. 22
2.3 Final Remarks . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . p. 23
3 BCI Algebras p. 24
3.1 Pseudo-BCI Algebras . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . p. 26
3.2 Final Remarks . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . p. 28
II Contributions 29
4 (T,N)-Implications p. 30
4.1 Definition and Basic Properties . . . . . . . . . . . . . .
. . . . . . . . . . . . . p. 30
4.2 Functional equations and (T,N)-implications . . . . . . . .
. . . . . . . . . . p. 424.3 Applying (T,N)-implications to
generate fuzzy subsethood measures . . . . p. 474.4 Generalization
of (T,N)-implications . . . . . . . . . . . . . . . . . . . . . . .
p. 52
4.4.1 Characterizations of (N ′, T,N)-Implications . . . . . . .
. . . . . . . p. 644.4.2 Aggregating (N ′, T,N)-Implications . . .
. . . . . . . . . . . . . . . p. 66
-
4.5 Final Remarks . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . p. 67
5 Intervalization of BCI algebras and Semi-BCI algebras p.
68
5.1 Intervalization of BCI algebras . . . . . . . . . . . . . .
. . . . . . . . . . . . . p. 68
5.2 Semi-BCI algebras . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . p. 74
5.3 Comparing Semi-BCI and Pseudo-BCI Algebras . . . . . . . . .
. . . . . . . p. 78
5.4 Final Remarks . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . p. 79
6 Single-Level Constraint Interval Fuzzy Connectives p. 80
6.1 Single-Level Constrained Interval Operators . . . . . . . .
. . . . . . . . . . . p. 82
6.2 Composition of C-Operators . . . . . . . . . . . . . . . . .
. . . . . . . . . . . p. 88
6.2.1 Properties of Some Interval Fuzzy Connectives with Respect
to EagerComposition . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . p. 89
6.2.2 Properties of Some Interval Fuzzy Connectives with Respect
to Single-Level Composition . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . p. 95
6.3 Final Remarks . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . p. 100
7 Constrained Interval Fuzzy Connectives p. 101
7.1 Some Properties of the Constrained Interval Fuzzy
Connectives with Respectto Order ≤
KM. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . p. 111
7.1.1 (T,N)-Implications Generated from Constrained Interval
Fuzzy Con-nectives . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . p. 115
7.2 Some Properties of the Constrained Interval Fuzzy
Connectives with Respectto Order ⊴ . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . p. 117
7.3 On the Extension of BCI and Semi-BCI Algebras via
Constrained IntervalFuzzy Operators . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . p. 120
7.3.1 On Extension of BCI and Semi-BCI Algebras with Respect to
EagerComposition . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . p. 121
7.3.2 On Extension of BCI and Semi-BCI Algebras with Respect to
Con-strained Punctual Composition . . . . . . . . . . . . . . . . .
. . . . . p. 122
7.4 Final Remarks . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . p. 123
8 Concluding remarks p. 125
References p. 129
-
8
1 Introduction
Interval-valued fuzzy logic was developed in order to deal with
the uncertainties concerningnot only on fuzzy rules, in which fuzzy
logic is very successful, but also on inputs-outputs. Incases in
which the fuzzification of inputs derives from choices made by
experts, fuzzy logic isnot always suitable, as the expert may be
unsure when defining the membership function. Thiswas one of the
motivations for the research on interval-valued fuzzy logic.
Another case liesin the situation in which the inputs represent
imprecise numerical data. Interval-valued fuzzylogic is a
particular case of interval type-2 fuzzy logic (ZADEH, 1975; LIANG;
MENDEL, 2000;BUSTINCE et al., 2015; CASTILLO et al., 2016).
Interval-valued fuzzy logic was applied in a wide variety of
domains, for example: Melinand Castillo (CERVANTES; CASTILLO;
MELIN, 2011) used it in the context of plant control;Figueroa et
al. (FIGUEROA et al., 2005) for non-autonomous robots in the
context of a robotfootball game; Lynch et al. (LYNCH; HAGRAS;
CALLAGHAN, 2005) built an interval control sys-tem for large marine
diesel engines; Chourasia et al. (CHOURASIA; TIWARI;
GANGOPADHYAY,2014) developed a new method for assessing fetal
health status based on interval type-2 fuzzylogic through fetal
phonocardiography; Nguyen et al. (NGUYEN et al., 2015) used the
waveletfeature in interval type-2 fuzzy logic system (IT2FLS) to
reduce the computation burden andtime of IT2FLS; Leow et al. (LEOW
et al., 2019) developed a hybrid of Generalized AdaptiveResonance
Theory (GART) and interval type-2 fuzzy logic system algorithm;
among manyothers.
The field of interval analysis has a long term development. The
idea of bounding round-ing errors by computing with intervals was
first given by Warmus (WARMUS, 1956), Sunaga(SUNAGA, 1958) and
Moore (MOORE, 1959). However, it can be said that interval
mathemat-ics and analysis began with the appearance of R. E.
Moore’s book, Interval Analysis in 1966(MOORE, 1966). It deals with
numerical data in the form of compact intervals in order to
encodecomputational errors or inaccuracies. The interval analysis
has been applied in several areas(JAULIN et al., 2001; KEARFOTT;
KREINOVICH, 2013), like: Electrical power systems (BARBOZA;DIMURO;
REISER, 2004), mechanical engineering (MUHANNA; ZHANG; MULLEN,
2007), chem-ical engineering (STADTHERR et al., 2007), artificial
intelligence (HU et al., 2008), multi-agentsystems (DIMURO; COSTA,
2004) and geophysics (AGUIAR; DIMURO; COSTA, 2004).
Warmus (WARMUS, 1956), Teruo Sunaga (SUNAGA, 1958) and Ramon
Moore (MOORE,1959, 1962) independently developed the interval
arithmetic. The Moore’s arithmetic is ac-cepted as the standard
approach and is called here standard interval arithmetic (SIA).
There aretwo most important criteria for an interval arithmetic,
namely: Correctness (accuracy) and op-timality (HICKEY; JU; EMDEN,
2001; MOORE, 1979). The first criterion establishes that the
resultof an interval computation must always contain the value of
the respective real function (see(MOORE, 1979, Theorem 3.1)).
Although correctness is a desirable property, not every
interval
-
9
method is correct. Santiago et al. (SANTIAGO; BEDREGAL; ACIÓLY,
2006) investigates the no-tion of correctness for interval
functions and its impact on some interval topological
viewpoints.They call correctness as interval representations, since
interval entities (algorithms and inter-vals) are seen as
linguistic entities which represent real entities (functions and
numbers). Aninterval function, F , is said to represent a real
function, f , whenever it satisfies the followingproperty: x ∈ [a,
b] ⇒ f(x) ∈ F ([a, b]) and F ([a, a]) = f(a). The second criterion
estab-lishes that the resulting interval of a computation should
not be greater than necessary, which iscaptured by the notion of
canonical interval representation.
The process of giving the correct and optimal interval version F
for a function f is calledintervalization. There are many proposals
of intervalization of algebraic structures further thanthat of real
numbers proposed by Moore and Sunaga, see for example the case for
Łukasiewiczalgebras and MV-algebras in (BEDREGAL; SANTIAGO, 2013)
and (CABRER; MUNDICI, 2014),respectively. In most cases, interval
algebras fail to satisfy some properties that are satisfied bythe
algebras from which they came.
In order to solve the algebraic incompatibility between the real
arithmetic and Moore arith-metic, Lodwick (LODWICK, 1999) defined a
new interpretation for intervals called constraintintervals. With
this new approach, Lodwick proposed an alternative to Moore
arithmetic onintervals in order to have X −X = [0,0], X ÷X = [1,1]
if 0 ∉ X and the distributive property.This new arithmetic, called
constrained interval arithmetic (CIA), is an extension of
Moore’sinterval arithmetic, in the sense that they coincide in the
case that there is no variable depen-dence and are distinct when
there are dependencies. In this case, the CIA arithmetic presents
asmaller width interval, thus improving the overestimation of
Moore’s arithmetic.
This thesis presents a new approach to interval-valued fuzzy
logic, in which interval oper-ators preserve some of the main
algebraic properties, the overestimation problem is mitigatedand
there is no loss of information. These operators were defined using
the constrained intervalintroduced by Lodwick (LODWICK, 1999). In
what follows, it is described how this study wasdone and also the
contributions that were studied or compared with this new approach.
Thereader can find a more detailed discussion in Part II.
The first contribution of this work lies on the investigation of
a new class of fuzzy implica-tions, called (T,N)-implications
(BEDREGAL, 2007), in which it is obtained from the compo-sition of
a fuzzy negation and a t-norm. It is not difficult to find in the
literature implicationsthat are obtained through other operators,
among which we can mention the (S,N)-, R- andQL-implications that
have been widely investigated. Many applications have already been
de-veloped using fuzzy implications, such as (MAS et al., 2007;
BACZYŃSKI, 2013; BACZYŃSKI et al.,2013), and they can still be
applied in areas of study such as approximate reasoning, control
the-ory, decision making theory, expert systems, diffuse
mathematical morphology (BLOCH, 2009;YAGER, 2004; BACZYŃSKI, 2013;
BANDLER; KOHOUT, 1980; BUSTINCE et al., 2013), among oth-ers. In
this document, the main properties of (T,N)-implications (PINHEIRO
et al., 2017, 2018a)with respect to different fuzzy negations were
studied. In addition, an application to fuzzysubsethood measure was
presented, in which a new subsethood measure was defined,
namelyPB-subsethood measure (PINHEIRO et al., 2018b), and it was
verified that it is possible to gen-erate this measure from a
family of (T,N)-implications. Finally, it was defined (N ′,
T,N)-implications (PINHEIRO et al., 2018), generalizes
(T,N)-implications. The author also presentsa characterization of
the (N ′, T,N)-implications and verify that it is possible to
aggregate afamily of (N ′, T,N)-implications and this aggregation
is still an (N ′, T,N)-implication.
-
10
The second contribution lies on the application of
intervalization on BCI algebras (ISÉKI,1966; HUANG, 2006). BCI
algebras are the algebraic counterpart of a common fragment
ofseveral important logics, like fuzzy logic, which is modeled by
BL algebras (a kind of BCIalgebra). Thus, the intervalization of
BCI algebras are important for the construction of analgebraic
model for some interval fuzzy logics. Here, a class of BCI algebras
was interval-ized and an investigation of the resulting structures
was provided. Like for MV-algebras andŁukasiewicz algebras, the
resulting interval structure does not belong to the same category
ofits starting algebra, it is a new mathematical structure. This
new structure is a generalization ofthe BCI algebras and came to be
called semi-BCI algebra (SANTIAGO et al., 2019). The semi-BCI
algebras were studied in detail and, in addition, the relationship
between the semi-BCI andpseudo-BCI algebras has been investigated.
It is verified that the only intersection between thetwo is the
class of BCI algebra.
The third contribution introduces a new approach to
interval-valued fuzzy logic, in whichthe notion of constraint
interval, proposed by Lodwick in (LODWICK, 1999), is applied. In
2014,Chalco-Cano et al. in (CHALCO-CANO; LODWICK; BEDE, 2014)
proposed a variant of constraintinterval operators that uses a
single parameter (level), instead of using two parameters
proposedby Lodwick. Following Lodwick and Chalco, the author
extends the fuzzy operators to theso-called Single-Level
Constrained Interval Operators (C-operators) and studied their
mainproperties. Also, it has been shown that the fuzzy connectives
are extended to their respectiveC-operators, however, not all are
correct. The composition of C-operators provides two methodsof
evaluation for interval compositions, called: Eager evaluation and
single-level evaluation.Important properties such as the exchange
principle, contraposition law (also, left and rightcontraposition
law), among others, were investigated by using both methods.
The main and final contribution of this thesis presents another
approach to interval-valuedfuzzy logic, in which the new operators,
called Constrained Interval Operators, are very closeto Moore’s
correctness, in fact, they satisfy a new correction that will be
suggested, in whichit is as efficient as Moore’s correction, being,
however, less demanding, which was called hereConstraint Interval
Correctness. Two methods of evaluating the compositions of these
opera-tors, namely: Eager and constrained punctual compositions,
have been defined. In addition tothe comparative study between
single-level constrained interval operators and constrained
inter-val operators, the main properties of this operator have been
verified in relation to two orders,namely: Kulisch-Miranker and
Moore order. It has been found that this new approach guar-antees
the extension of many algebraic properties and that its fuzzy
operators are extended totheir respective constrained interval
operators, when considering the order of Kulisch-Miranker.Also, a
special class of fuzzy implications was extended, called
(T,N)-implications. It hasbeen found that the (T,N)-implications
generated from constrained interval operators coincidewith the best
representation of the original (T,N)-implication. Finally, both
algebras, BCI andsemi-BCI, are extended to their respective
constrained interval algebras with respect to punctualcomposition.
Here correctness is maintained from the perspective of Constrained
Interval.
This thesis is organized as follows: Part I recalls some
definitions and concepts usedthroughout the text in order to
provide a self-contained document. It is divided into two
chap-ters: Chapter 2 presents interval arithmetics and logics; and
Chapter 3 introduces the BCI alge-bras. Part II presents the
contributions of this work, it is divided into five chapters:
Chapter 4provides a new class of fuzzy implications called
(T,N)-implication. Chapter 5 proposes a newalgebra called Semi-BCI
algebra, which generalizes BCI algebras. Chapter 6 proposes a
newway of making interval-valued fuzzy logics, in which the
operators were called Single-Level
-
11
Constrained Interval Operators and Chapter 7 proposes another
way of making interval-valuedfuzzy logics, in which the operators
were called Constrained Interval Operators, which gener-alize the
operators of the previous chapter. The last chapter includes some
conclusions, futureworks, and the bibliography.
-
12
Part I
Preliminaries
-
13
2 Interval Arithmetics and Logics
2.1 Arithmetics
The limited capacity of machines to store just a finite set of
finitely represented objects con-strains the automatic calculation
(computation) of structures in which a machine representationof
some objects exceeds such capacity. In the case of real numbers,
although programs oftenprovide highly accurate results, it can
happen that rounding errors built up during each step inthe
computation produce results which are not even meaningful. One of
the proposals to over-come this problem is due, almost
simultaneously, to M. Warmus (WARMUS, 1956), T. Sunaga(SUNAGA,
1958) and R. Moore (MOORE, 1959, 1962), with the development of the
so-calledinterval arithmetic, as the following section shows:
2.1.1 Standard Interval Arithmetic
Interval arithmetic is a set of operations on the set of all
closed intervals I(R) = {X ∣ X =[x,x];x,x ∈ R and x ≤ x}. The
operations are defined in the following way:
1. X + Y = [x + y, x + y],
2. X − Y = [x − y, x − y] ,
3. X ⋅ Y = [min{x ⋅ y, x ⋅ y, x ⋅ y, x ⋅ y},max{x ⋅ y, x ⋅ y, x
⋅ y, x ⋅ y}],
4. X/Y = [x,x] ⋅ ([1/y,1/y]), provided that 0 ∉ [y, y].
Observe that for each operation ∗ ∈ {+,−, ⋅, /}, X ⊛ Y = {x ∗ y
∈ R ∶ x ∈ X and y ∈ Y } isalways an interval.
Moore’s development of interval arithmetic is accepted as the
standard approach to intervalarithmetic, which will be called
standard interval arithmetic (SIA), and it is the approach
tointerval arithmetic in common use. Here are some properties
associated with SIA (see (MOORE,1979)), for X,Y and Z in I(R):
(1) X + (Y +Z) = (X + Y ) +Z – the associative law for
addition
(2) X ⋅ (Y ⋅Z) = (X ⋅ Y ) ⋅Z – the associative law for
multiplication
(3) X + Y = Y +X – the commutative law for addition
-
14
(4) X ⋅ Y = Y ⋅X – the commutative law for multiplication
(5) [0,0] +X =X + [0,0] =X – additive identity
(6) [1,1] ⋅X =X ⋅ [1,1] =X – multiplicative identity
(7) X ⋅ (Y +Z) ⊆X ⋅ Y +X ⋅Z – the subdistributive property
Moore’s interval arithmetic presents a problem of overestimation
associated with multipleoccurrences of the same variable in an
expression. Also, it is clear that X −X is never 0, unlessX is a
real number (a zero width interval) and X ÷X is never 1, unless X
is a real number (azero width interval).
In this theory, there are two important criteria called
correctness (accuracy) and optimality(HICKEY; JU; EMDEN, 2001;
MOORE, 1979), which were formalized by Santiago et al. in
(SAN-TIAGO; BEDREGAL; ACIÓLY, 2006), and was called interval
representation and best intervalrepresentation, respectively (see,
e.g., (BEDREGAL; SANTIAGO, 2013; BEDREGAL; TAKAHASHI,2005, 2006)).
Both are defined in the following subsection.
2.1.1.1 Intervalization of Structures
Assuming that the set X ⊛ Y = {x ∗ y ∈ R ∶ x ∈ X and y ∈ Y }
always corresponds toan interval, where ∗ ∈ {+,−, ⋅, /}, this
reveals two important properties of this arithmetic (a)Correctness
and (b) Optimality.
“Correctness. The criterion for correctness of a definition of
interval arithmeticis that the “Fundamental Theorem of Interval
Arithmetic” holds 1: when an ex-pression is evaluated using
intervals, it yields an interval containing all results ofpointwise
evaluations based on point values that are elements of the argument
in-tervals.
[. . . ]Optimality. By optimality, it is meant that the computed
floating-point interval
is not wider than necessary.”Hickey et.al(HICKEY; JU; EMDEN,
2001, p.1040)
The application of interval methods follows the following
paradigm: Enclosure in intervalsthe values which are not exact by
whatever reason (e.g. the value comes from an imprecisemeasurement)
and applying correct and optimal operations on such intervals in
order to obtainthe best interval which contains the desired
output.
The property of correctness was investigated in 2006 by Santiago
et al (SANTIAGO; BEDRE-GAL; ACIÓLY, 2006; BEDREGAL; SANTIAGO,
2013). Instead of correctness, they used the terminterval
representation, since an interval computation could be understood
not just as a machinerepresentation of real numbers, but also as a
mathematical representation of real numbers (thisidea is confirmed
by the Representation Theorems of Euclidean continuous functions in
(SANTI-AGO; BEDREGAL; ACIÓLY, 2006; BEDREGAL; SANTIAGO, 2013)).
Also, the notion of optimality
1Moore (MOORE, 1979, Theorem 3.1, p. 21): If F is an inclusion
monotonic interval extension of f , then→
f (X1, ...,Xn) ⊆ F (X1, ...,Xn), where→
f (X1, ...,Xn) = {f(x1, . . . , xn) ∶ xi ∈Xi}.
-
15
was named as best interval representation, or best
representation for short. And, in what fol-lows, this notion is
shown for binary operations: A binary interval operation ⊛
represents abinary real operation, ∗, whenever:
(x, y) ∈X × Y ⇒ x ∗ y ∈X ⊛ Y.
This can be easily extended to n-ary operations. The author
showed that this notion is moregeneral than what is stated by the
Fundamental Theorem of Interval Arithmetic, given that thereare
representations which are not inclusion monotonic (see (SANTIAGO;
BEDREGAL; ACIÓLY,2006, p. 238)). The formal definition
follows:
Definition 2.1. An interval operation F ∶ U([0,1])n → U([0,1])
is Moore-correct or intervalrepresentation with respect to a
function f ∶ [0,1]n → [0,1] whenever, for all (A1, . . . ,An)
∈U([0,1])n and ai ∈ Ai, f(a1, . . . , an) ∈ F (A1, . . . ,An). In
addition, F is best interval rep-resentation with respect to
function f , denoted by f̂ , if F (A1, . . . ,An) is the least
intervalcontaining the set {f(a1, . . . , an) ∣ ai ∈ Ai} for all Ai
∈ U([0,1]) and i ∈ {1, . . . , n}, i.e.,
f̂(A1, . . . ,An) = [inf{f(a1, . . . , an) ∣ ai ∈ Ai}, sup{f(a1,
. . . , an) ∣ ai ∈ Ai}] . (2.1)
The process of giving the correct and optimal interval version F
for a function f is calledintervalization. There are many proposals
of intervalization of algebraic structures further thanthat of real
numbers proposed by Moore, Warmus and Sunaga. In the literature,
the reader canfind proposals even for the field of Logic. For
example: The Łukasiewicz implicative algebra⟨[0,1],→LK ,1⟩, where x
→LK y = min(1,1 − x + y) interprets some many-valued logics andwas
intervalized by Bedregal et al in (BEDREGAL; SANTIAGO, 2013). Its
MV algebra counterpartwas intervalized by Cabrer et al in (CABRER;
MUNDICI, 2014), also, in order to overcome thesame problems already
stated for I(R). In both cases, the interval algebras did not
satisfy thesame properties that are satisfied by the algebras from
which they came.
In order to solve the algebraic incompatibility between the real
arithmetic and Moore arith-metic, Lodwick (LODWICK, 1999) defined a
new interpretation for intervals called constraintintervals, as
seen below below.
2.1.2 Constrained Interval Arithmetic and Single-Level
Constrained In-terval Arithmetic
In 1999, Lodwick (LODWICK, 1999) proposed an alternative to
Moore arithmetic on inter-vals in order to have X −X = [0,0] and X
÷X = [1,1] if 0 ∉X . For this purpose, he defined anew way of
interpreting intervals, called constrained intervals.
Definition 2.2. Given an interval X = [x,x] a constrained
interval associated to X is thefunction fX ∶ [0,1]→ [0,1], s.t for
0 ≤ λx ≤ 1,
fX(λx) = (1 − λx)x + λxx= x + λxωx,
where ωx = x − x (the width of X).
-
16
The resulting arithmetic, called constrained interval arithmetic
(CIA), is defined as follows:
fX(λx) ∗ fY (λy), (2.2)
for λx, λy ∈ [0,1] and ∗ ∈ {+,−,×,÷}, where the resulting
interval, X ⊛ Y , can be extractedfrom (2.2) by computing the
minimum and maximum,
[ min0≤λx,λy≤1
{fX(λx) ∗ fY (λy)}, max0≤λx,λy≤1
{fX(λx) ∗ fY (λy)}] .
This arithmetic retains the desirable properties, i.e., X − X =
[0,0], X ÷ X = [1,1] and thedistributive property X × (Y + Z) = (X
× Y ) + (X × Z). Note that this new arithmetic isan extension of
Moore’s interval arithmetic, in the sense that they coincide in the
case thatthere is no dependence and are distinct when there are
dependencies present. In this case, theCIA arithmetic presents a
smaller width interval, thus improving the overestimation of
Moore’sarithmetic. See the following example.
Example 2.1. Consider the expression X + Y −X for the intervals
X = [1,2] and Y = [−1,1].Using Moore’s interval arithmetic we
obtain X + Y −X = ([1,2] + [−1,1]) − [1,2] = [0,3] −[1,2] = [−2,2].
While by CIA arithmetic, fX(λx) = 1+λx and fY (λy) = −1+2λy, so
(fX(λx)+fY (λy)) − fX(λx) = fY (λy) and hence X + Y −X = Y =
[−1,1].
Instead of using independent lambdas for each variable,
Chalco-Cano et al. in (CHALCO-CANO; LODWICK; BEDE, 2014) proposed
another arithmetic, called single level constrained in-terval
arithmetic (SLCIA). This arithmetic is the constrained interval
arithmetic in which onlyone parameter, i.e. λy = λx. That is, if we
consider two intervals X and Y , we take a levelλ ∈ [0,1] for both
intervals instead for of a λx for X and λy for Y , i.e. we take
fX(λ) = (1 − λ)x + λx and fY (λ) = (1 − λ)y + λy.
After operating at all levels, the minimum and maximum are
calculated, i.e.:
X ⊛ Y = [min0≤λ≤1
{fX(λ) ∗ fY (λ)},max0≤λ≤1
{fX(λ) ∗ fY (λ)}].
The SLCIA arithmetic is a restriction of the CIA, in which it
maintains the desirable properties,that is, X −X = [0,0], X ÷X =
[1,1] and the distributive law. However, in this approach wealso
have X − Y = [0,0] when X = Y , unlike CIA.
Chalco-Cano extended the single-level interval arithmetic for
expressions with intervaloperands. The evaluation of an expression
is performed according to the following rule:
E(A1, . . . ,An) = [min0≤λ≤1
{E(fA1(λ), . . . , fAn(λ))},max0≤λ≤1
{E(fA1(λ), . . . , fAn(λ))}].
Considering this way of evaluating the expressions, they showed
the following algebraicproperties: For all interval X,Y,Z and α,β ∈
R,
(1) X ⊕ (−Y ) =X ⊖ Y ;
(2) X ⊖ (−Y ) =X ⊕ Y ;
(3) ⊕ is associative: (X ⊕ Y )⊕Z =X ⊕ (Y ⊕Z);
-
17
(4) ⊕ is commutative: X ⊕ Y = Y ⊕X;
(5) [0,0] is the only neutral element for ⊕: X ⊕ [0,0] =X;
(6) α⊙ (X ⊕ Y ) = (α⊙X)⊕ (α⊙ Y );
(7) (α + β)⊙X = (α⊙X)⊕ (β ⊙X);
(8) X ⊖X = [0,0];
(9) (X ⊕ Y )⊖ Y =X;
(10) (X ⊖ Y )⊖X = (−1)⊙ Y ;
(11) [0,0]⊖ (X ⊖ Y ) = Y ⊖X;
(12) X ⊖ Y = (−1)⊙ (Y ⊖X);
(13) X ⊖ Y = ((−1)⊙ Y )⊖ ((−1)⊙X);
(14) X ⊖ Y = Y ⊖X iff X ⊖ Y is symmetric;
(15) X ⊗ (Y ⊕Z) =X ⊗ Y ⊕X ⊗Z;
(16) (Y ⊕Z)⊗X = Y ⊗X ⊕Z ⊗X;
(17) X ⊗ (Y ⊖Z) =X ⊗ Y ⊖X ⊗Z;
(18) (Y ⊖Z)⊗X = Y ⊗X ⊖Z ⊗X .
Intervals not only provide a way to express a number
approximations, they can also be usedas a logical value. The next
section shows how intervals play this role.
2.2 Logics
In conventional or classical logic, a statement is either false
or true and can not be partiallyfalse and partially true. However,
in the real world, it is very common to meet complicatedproblems
that are not always bivalent, nor are they always made of
absolutely true or falsefacts. To model such problems, multivalent
logics have emerged, such as Fuzzy Logic, inwhich it reflects the
way people think, trying to shape their sense of words, decision
making, orcommon sense.
Fuzzy Logic was introduced by Lofti Zadeh in 1965 in the article
titled: Fuzzy Sets (ZADEH,1965). Fuzzy logic is a formalism
suitable for modeling the human capacity for approximatereasoning
and support for decision making in environments where there is
imperfect informa-tion and gradual set belonging, allowing a
sufficient variety of physical and mental tasks tobe performed
without any measure or computation. It has been applied in several
areas, suchas control systems (GUANRONG; TAT, 2001), decision
making (CHANG; WANG, 2009), expertsystems (SILER; BUCKLEY, 2005),
pattern recognition (CHOI; RHEE, 2009), etc.
-
18
2.2.1 Fuzzy Connectives
Among the most important operators in fuzzy logic, the author
highlights t-norms, t-conorms,fuzzy negations and fuzzy
implications. These operators are generalizations of the classical
con-junctions, disjunctions, negations and implications to fuzzy
logic, respectively . The definitionsare given as follows:
Definition 2.3. (SCHWEIZER; SKLAR, 1958, 1960, 1961) A function
T ∶ [0,1]2 → [0,1] is said tobe a triangular norm (t-norm, for
short) if it satisfies the following conditions, for all x, y, z
∈[0,1]:
(T1) Symmetry: T (x, y) = T (y, x);
(T2) Associativity: T (x,T (y, z)) = T (T (x, y), z);
(T3) Monotonicity: If x1 ≤ x2 and y1 ≤ y2 then T (x1, y1) ≤ T
(x2, y2);
(T4) 1-identity: T (x,1) = x. (boundary condition)
In fuzzy logic, the conjunction is often represented by a
t-norm. The standard fuzzy con-junction TM ∶ [0,1]2 → [0,1], given
by TM(x, y) = min{x, y}, called minimum t-norm, is theonly
idempotent t-norm (see (KLIR; YUAN, 1995) - Theorem 3.9). Another
example of t-normsis the product, denoted by TP .
Proposition 2.1. (BEDREGAL, 2007) Let T be a t-norm. Then T (0,
y) = 0 for each y ∈ [0,1].
Definition 2.4. A t-norm T is called positive if, for all x, y ∈
[0,1], it satisfies the condition:T (x, y) = 0 if and only if x = 0
or y = 0.
Definition 2.5. (SCHWEIZER; SKLAR, 1961) A triangular conorm
(t-conorm for short) is a binaryoperation S on the unit interval
[0,1], i.e., a function S ∶ [0,1]2 → [0,1], which, for all x, y, z
∈[0,1], satisfying (T1), (T2), (T3) and
(S4) S(x,0) = x. (boundary condition)
The standard fuzzy disjunction SM ∶ [0,1]2 → [0,1] given by
SM(x, y) = max{x, y},called maximum t-conorm, is the only
idempotent t-conorm (see (KLIR; YUAN, 1995) - Theorem3.14).
From an axiomatical point of view, t-norms and t-conorms differ
only with respect to theirboundary conditions.
In the following, the notion of fuzzy negation is recalled.
Definition 2.6. (FODOR; ROUBENS, 1994) A function N ∶ [0,1]→
[0,1] is a fuzzy negation if
(N1) N is antitonic, i.e. N(x) ≤ N(y) whenever y ≤ x;
(N2) N(0) = 1 and N(1) = 0.It is strict whenever
-
19
(N3) N is continuous and
(N4) N(x) < N(y) whenever y < x.It is strong if
(N5) N(N(x)) = x, for each x ∈ [0,1].A fuzzy negation N is crisp
if
(N6) N(x) ∈ {0,1}, for all x ∈ [0,1];A fuzzy negation N is
frontier if it satisfies the property:
(N7) N(x) ∈ {0,1} if and only if x = 0 or x = 1;A fuzzy negation
N is non-vanishing if
(N8) N(x) > 0 whenever x < 1.
Example 2.2. The least fuzzy negation, N⊥, and the greatest
fuzzy negation, N⊺, are defined,respectively, by
N⊥(x) = {1, if x = 00, if x > 0
and
N⊺(x) = {0, if x = 11, if x < 1 .
Definition 2.7. Given a t-norm T and a fuzzy negation N , it can
be said that the pair (T,N)satisfies the law of contradiction
whenever
T (x,N(x)) = 0, x ∈ [0,1]. (LC)
Definition 2.8. Let T be a t-norm, S be a t-conorm and N be a
fuzzy negation. Then S issaid to be N -dual to T if for all x, y ∈
[0,1] we have N(S(x, y)) = T (N(x),N(y)). It willbe denoted by ST .
Analogously, T is said to be N -dual to S if for all x, y ∈ [0,1]
we haveN(T (x, y)) = S(N(x),N(y)). It will be denoted by TS .
Fuzzy implication generalizes the usual material implications.
In what follows they arepresented with some of their
properties.
Definition 2.9. (FODOR; ROUBENS, 1994) A function I ∶ [0,1]2 →
[0,1] is a fuzzy implicationif the following properties are
satisfied, for all x, y, z ∈ [0,1]:
(I1) If x ≤ y then I(y, z) ≤ I(x, z);
(I2) If y ≤ z then I(x, y) ≤ I(x, z);
(I3) I(0, y) = 1;
(I4) I(x,1) = 1;
-
20
(I5) I(1,0) = 0.
The set of all fuzzy implications will be denoted by FI .
Definition 2.10. Let I ∈ FI . The function NI ∶ [0,1]→ [0,1]
defined by
NI(x) = I(x,0), x ∈ [0,1] (2.3)
is called the natural negation of I or the negation induced by I
.
In the following, some of the most important properties of some
fuzzy implications arepresented, which will be useful in this work
(see (SMETS; MAGREZ, 1987; TRILLAS; VALVERDE,1993; FODOR; ROUBENS,
1994)).
Definition 2.11. A fuzzy implication I is said to satisfy:
(i) the identity property, if, for all x ∈ [0,1]
I(x,x) = 1; (IP)
(ii) the left neutrality property, if, for all y ∈ [0,1]
I(1, y) = y; (NP)
(iii) the exchange principle, if, for all x, y, z ∈ [0,1]
I(x, I(y, z)) = I(y, I(x, z)). (EP)
(iv) the left-ordering property, if, for all x, y ∈ [0,1]
I(x, y) = 1 whenever x ≤ y; (LOP)
(v) the right-ordering property, if, for all x, y ∈ [0,1]
I(x, y) ≠ 1 whenever x > y. (ROP)
(vi) the order property iff I satisfy (LOP) and (ROP), i.e., for
all x, y ∈ [0,1]
I(x, y) = 1 iff x ≤ y; (OP)
(vii) the law of left self-distributivity2, if, for all x, y, z
∈ [0,1]
I(x, I(y, z)) = I(I(x, y), I(x, z)). (LSD)
Definition 2.12. Let I ∈ FI and let N be a fuzzy negation. I is
said to satisfy the:
(i) contraposition law with respect to N , if
I(x, y) = I(N(y),N(x)), for all x, y ∈ [0,1]; (CP)
(ii) left contraposition law with respect to N , if
I(N(x), y) = I(N(y), x), for all x, y ∈ [0,1]; (L-CP)2This law
was studied in (CRUZ; BEDREGAL; SANTIAGO, 2018) under the name
Boolean-Like. The name given
here is more appropriate because it is closer to self
left-distributive law in (FRINK, 1955).
-
21
(iii) right contraposition law with respect to N , if
I(x,N(y)) = I(y,N(x)), for all x, y ∈ [0,1]. (R-CP)
If I satisfies the (left, right) contraposition with respect to
a specific N , the following nota-tion will be used: L−CP (N), R−CP
(N) and CP (N), respectively.
Proposition 2.2. (BACZYŃSKI; JAYARAM, 2008) Let I ∶ [0,1]2 →
[0,1] be an any function andNI be a strong negation.
(i) If I satisfies CP (NI), then I satisfies (NP ).
(ii) If I satisfies (EP ), then I satisfies (I3), (I4), (I5),
(NP ) and (CP ) only with respectto NI .
Proposition 2.3. (BACZYŃSKI; JAYARAM, 2008) If a function I ∶
[0,1]2 → [0,1] satisfies (EP )and NI is a fuzzy negation, then I
satisfies R−CP (NI).
Proposition 2.4. (BACZYŃSKI; JAYARAM, 2008) If a function I ∶
[0,1]2 → [0,1] satisfies (R−CP )with respect to continuous fuzzy
negation N , then I satisfies (I1) if and only if it satisfies
(I2).
Definition 2.13. Let I ∈ FI and T be any t-norm and I be a fuzzy
implication for T . The pair(I, T ) satisfies the T-conditionality
property for T if, for each x, y ∈ [0,1],
T (x, I(x, y)) ≤ y . (TC)
Definition 2.14. Let I be a fuzzy implication and T be a t-norm.
It may be said that I satisfiesthe Law of importation (LI) with
respect to a t-norm T if
I(T (x, y), z) = I(x, I(y, z)), (2.4)
for all x, y, z ∈ [0,1].
It is well known that the fuzzy implications are generalizations
of the implications of classi-cal logic to fuzzy logic, just as a
t-norm and a t-conorm are generalizations of classical conjunc-tion
and disjunction, respectively. There are some ways to generate
fuzzy implications fromlogical connectives. The main fuzzy
implications are generalizations of the following tautolo-gies of
classical logic:
p→ q = ¬p ∨ q, p→ q = ¬p ∨ (p ∧ q) and p→ q = (¬p ∧ ¬q) ∨ q,
namely (S,N), QL and D-implications, respectively (see (FODOR,
1991; MAS; MONSERRAT;TORRENS, 2006; BACZYŃSKI; JAYARAM, 2008;
BACZYŃSKI, 2004)). In addition, there is animplication that arises
from the isomorphism that exists between the classical logic of
twovalues and the classical set theory by the following
identity
A′ ∪B = (A/B)′ =⋃{C ⊆X ∣ A ∩C ⊆ B},
where A and B are subsets of some universal set X and A′ is the
complement of set A. Fuzzyimplications obtained as generalization
of identity above form the family of residual implica-tions,
usually called in the literature of R-implications.
-
22
Definition 2.15. Let I be a fuzzy implication. It can be said
that I is called:
(i) (S,N)-implications, if I(x, y) = S(N(x), y) for a given
t-conorm S and a fuzzy negationN . If N is strong, then I is simply
called S-implications.
(ii) R-implications, if I(x, y) = sup{z ∈ [0,1] ∣ T (x, z) ≤ y}
for a given left-continuoust-norm T .
(iii) QL-implications, if I(x, y) = S(N(x), T (x, y)) for a
given t-conorm S, a t-norm T andthe greatest fuzzy negation N⊺.
(i) D-implications, if I(x, y) = S(T (N(x),N(y)), y) for a given
t-conorm S, a t-norm Tand a strong negation N .
Fuzzy connectives can be extended to take into account
imprecision. The next section showshow some of the above fuzzy
connectives can be extended to operate with interval values.
2.2.2 Interval Fuzzy Connectives
Let U([0,1]) be the set of closed intervals on [0,1], i.e
U([0,1]) = {[x,x] ∣ 0 ≤ x ≤ x ≤ 1}.Let ⟨U([0,1]),≤⟩ be a bounded
poset with [0,0] as bottom and [1,1] as top elements.
Definition 2.16. A function N ∶ U([0,1])→ U([0,1]) is an
interval fuzzy negation on ⟨U([0,1]),≤⟩ if it is decreasing and
satisfies N([0,0]) = [1,1] and N([1,1]) = [0,0]. If N(N(A)) = A,∀A
∈ U([0,1]), then N is called strong interval fuzzy negation.
Definition 2.17. A t-norm onU([0,1]), called interval triangular
norm (it-norm) on ⟨U([0,1]),≤⟩, is a commutative, associative,
increasing mapping T ∶ U([0,1])2 → U([0,1]) which satisfiesT(A,
[1,1]) = A, for all A ∈ U([0,1]).
Proposition 2.5. Let T ∶ U([0,1])2 → U([0,1]) be an it-norm on
the bounded poset ⟨U([0,1]),≤⟩. Then, T(A1, [0,0]) = [0,0] for all
A1 ∈ U([0,1]).
Proof. Indeed, for allA1 ∈ U([0,1]), we have that [0,0] ≤ A1 ≤
[1,1]. So, since T is increasing,T([0,0],A1) ≤ T([0,0], [1,1]).
From the commutativity and boundary condition of T, weobtain T(A1,
[0,0]) ≤ [0,0], therefore, T(A1, [0,0]) = [0,0].
Definition 2.18. A function I ∶ U([0,1])2 → U([0,1]) is an
interval fuzzy implication on⟨U([0,1]),≤⟩ if, for all A1,A2,A3 ∈
U([0,1]), I satisfies the following properties:
(I1) If A1 ≤ A2 then I(A2,A3) ≤ I(A1,A3) ;
(I2) If A2 ≤ A3 then I(A1,A2) ≤ I(A1,A3);
(I3) I([0,0],A2) = [1,1];
(I4) I(A1, [1,1]) = [1,1];
(I5) I([1,1], [0,0]) = [0,0].
-
23
In (BEDREGAL; TAKAHASHI, 2006), Bedregal and Takahashi presented
a characterizationfor the best interval representation of fuzzy
implications. See the following theorem:
Theorem 2.1. (BEDREGAL; TAKAHASHI, 2006, Theorem 6.2) Let I be a
fuzzy implication. Then,the best interval representation of I ,
denoted by Î , is given by:
Î(A1,A2) = [I(a1, a2), I(a1, a2)] ,
for all A1,A2 ∈ U([0,1]).
Following the notion of interval representation (formalized by
Santiago et al. in (SANTIAGO;BEDREGAL; ACIÓLY, 2006)), Bedregal et
al. (BEDREGAL; SANTIAGO, 2013) showed that theintervalization of
Łukasiewicz implication does not preserve (OP), however weakening
the rightside of (OP), gave rise to the pair of properties, in
which the implication of Łukasiewicz satisfies:
(*) the interval r-weak order property, for all A1,A2 ∈
U([0,1]), if A1 ≤ A2 then
I(A1,A2) = [1,1]; (IR-WOP)
(**) the interval l-order property, for all A1,A2 ∈ U([0,1]), if
I(A1,A2) = [1,1] then
A1 ≤ A2. (IL-OP)
Logics are usually interpreted by algebras; e.g. Classical
Propositional Logics is interpretedby the usual boolean algebra
{0,1}, Łukasiewicz Logics is interpreted by MV-algebras. Mostof the
algebraic models for logics are BCI algebras (ISÉKI, 1966) with
additional axioms. Thenext chapter introduces those algebras and an
extension for that. Chapter 5 provides anothergeneralization called
Semi-BCI algebras, which rises from the process of intervalization
of BCIalgebras which captures the properties (IR-WOP) and
(IL-OP).
2.3 Final Remarks
The definitions of some interval arithmetic in the literature
and their properties were pre-sented. Besides, the definitions and
main properties of fuzzy connectives and interval fuzzyconnectives
were also exposed. Still in this interval context, two of the main
concepts of inter-val theory were presented, namely: Correctness
and optimality, where the first states that theresult of an
interval computation must always contain the value of the
respective real functionand the second establishes that the
interval result should be as small as possible meeting
thecorrectness criterion.
In the next chapter we present the BCI-algebras, which will be
later intervalized (see Chap-ter 5) using the concepts of
correctness and optimality.
-
24
3 BCI Algebras
Artificial intelligence has the important task of making
computers simulate humans to dealwith certainty and uncertainty in
information. Certain information processing is based on
theclassical logic of two values, however, it is natural and
necessary to try to establish some rationallogical system as the
logical basis for the uncertain information processing. This type
of logicis an extension of two-valued logic. In order to construct
natural and efficient inference systemsto deal with uncertainty,
several types of non-classical logic systems have been developed,
suchas BCI-logic.
BCI algebras were introduced by Iséki (ISÉKI, 1966) in the 60s
and since then have beenextensively investigated. The term,
BCI-algebra, originates from the combinatories B, C, I
incombinatory logic. There are several (equivalent) definitions of
a BCI algebra, differing in typeand notation. Some of them contain
a ∗ symbol for the binary operation and the symbol 0 (or) for the
null element. Here the BCI algebras will be used as algebras
⟨A,→,⊺⟩ and one of theconvincing arguments for this notation is
that it makes obvious the connection with logic (theoriginal
approach is done in signature ⟨A,∗,0⟩, see e.g. (ISÉKI, 1966) and
(IMAI; ISÉKI, 1965)).
Definition 3.1. A BCI algebra is a structure C = ⟨A,→,⊺⟩, where→
is a binary operation on Aand ⊺ is an element of A, verifying, the
axioms: for all x, y, z ∈X ,
(C-1) (y → z)→ ((z → x)→ (y → x)) = ⊺,
(C-2) x→ ((x→ y)→ y) = ⊺,
(C-3) x→ x = ⊺,
(C-4) if x→ y = ⊺ and y → x = ⊺ then x = y.
On any such BCI algebra it is possible to define a partial order
“⪯”:
(C-5) x ⪯ y iff x→ y = ⊺.
Whenever x→ ⊺ = ⊺, i.e. x ⪯ ⊺, the BCI algebra C = ⟨A,→,⊺⟩ will
be called a BCK algebra.
This relation ⪯ is called induced order ofA. It is not mandatory
for ⊺ be the greatest elementof (A,⪯) (however, it is maximal),
contrary to the case of BCK-algebras.
BCI algebras are the stronghold of several algebras that model
important logics, includingthe fuzzy logics that are modeled by BL
algebras. Thus, the intervalization of BCI algebras areimportant
for the construction of an algebraic model for some interval fuzzy
logics.
-
25
Example 3.1.
(1) The Łukasiewicz implicative algebra ([0,1],→LK ,1), where
x→LK y =min(1,1−x+y), is a BCI algebra.
(2) Given an abelian group (G, ⋅, e) with e as the unit element,
(G,→, e) is a BCI algebra,where x→ y = y ⋅ x−1.
(3) Given a set A, consider the parts of A, denoted by P (A).
The structure (P (A),⇒,∅)is a BCI algebra, with⇒ such that X ⇒ Y =
Y ∩XC , where XC is the complement ofX .
Let us now recall some useful properties of BCI algebras (for
more details see (HUANG,2006)):
(A-1) ⊺ ⪯ x implies x = ⊺;
(A-2) x ⪯ y implies y → z ⪯ x→ z; (First place antitonicity)
(A-3) x ⪯ y implies z → x ⪯ z → y; (Second place
isotonicity)
(A-4) x ⪯ y and y ⪯ z implies x ⪯ z; (Transitivity)
(A-5) x→ (y → z) = y → (x→ z); (Exchange – EP)
(A-6) x ⪯ y → z implies y ⪯ x→ z;
(A-7) x→ y ⪯ (z → x)→ (z → y);
(A-8) ⊺→ x = x; (Left Neutrality)
(A-9) ((y → x)→ x)→ x = y → x;
(A-10) x→ y ⪯ (y → x)→ ⊺;
(A-11) (x→ y)→ ⊺ = (x→ ⊺)→ (y → ⊺);
(A-12) (x→ y)→ ⊺ = ((y → x)→ ⊺)→ ⊺;
(A-13) y → ((x→ ⊺)→ ⊺) = ((y → x)→ ⊺)→ ⊺;
(A-14) x→ ⊺ = ((x→ y)→ y)→ ⊺;
(A-15) x ⪯ y implies x→ ⊺ = y → ⊺;
(A-16) y → ⊺ ⪯ x implies x = y → ⊺.
Note that properties (A-7), (A-5) and (C-3) model the
combinators B, C and I of BCI Logic(HINDLEY; SELDIN, 1986).
Proposition 3.1. Let ⟨A,→,⊺⟩ be a BCI algebra. ⟨A,→,⊺⟩ is a BCK
algebra if and only if foreach x ∈ A there exists y ∈ A such that y
⪯ x and y ⪯ ⊺.
-
26
Proof. (⇒) Straightforward because in BCK algebras ⊺ is the
greatest element, i.e. x ⪯ ⊺ foreach x ∈ A.
(⇐) Suppose, by contradiction, that ⟨A,→,⊺⟩ is not a BCK
algebra. Then, there existsa ∈ A such that a /⪯ ⊺. By hypothesis
there exists b ∈ A such that b ⪯ a and b ⪯ ⊺. So, by (A-8)and the
definition of ⪯, (b→ ⊺)→ ((⊺→ a)→ (b→ a)) = ⊺→ (a→ ⊺) = a→ ⊺ ≠ ⊺.
Therefore,(C-1) fails.
BCK and BCI algebras have been extensively investigated by many
researchers (see (JUN;SHIM, 2005; LIU; XU; MENG, 2007; ZHAN; LIU,
2005; LIU et al., 2000; LIU; ZHANG, 1994)). Thereare three
important classes of BCI algebras: commutative BCI-algebras (MENG;
XIN, 1992a),implicative BCI-algebras (MENG; XIN, 1992b) and
positive implicative BCI-algebras (MENG;XIN, 1993). In addition to
important concepts such as ideals/filters (these are dual concepts
anddepend on the definition used, see e.g. (MENG, 1993; LIU; ZHANG,
1994; WEI; JUN, 1995)). Fora more detailed view of BCI-algebra, see
e.g. (HUANG, 2006).
There are several generalizations of the BCI algebras as shown
by, for example, Iorgulescuin (IORGULESCU, 2016a), in which he
found thirty-one new generalizations distinct from BCI orBCK
algebras and showed the hierarchies existing among these algebras
(see also (IORGULESCU,2016b)). In this work the author presents a
generalization (see the Pseudo-BCI algebra in thefollowing section)
in which it has the signature different from the one previously
mentioned,since it brings two binary operators instead of one, and
propose another generalization aim-ing to capture the
intervalization of point algebras that model logics (see Chapter
5). BecausePseudo-BCI has the same signature as the new
generalization of BCI algebra shown in Chapter5, the author decided
to show the relationship between them.
3.1 Pseudo-BCI Algebras
In (GEORGESCU; IORGULESCU, 2001a), G. Georgescu and A.
Iorgulescu introduced the no-tion of pseudo-BCK algebras as an
extension of BCK-algebras. The motivation was the fol-lowing: since
bounded commutative BCK algebra corresponds (is categorically
equivalent)to MV algebra (MUNDICI, 1986), they wanted to verify
which structure corresponds to thepseudo-MV algebra, which
pseudo-MV algebra is a non-commutative extension of MV al-gebras
(GEORGESCU; IORGULESCU, 1999, 2001b). Years later, W. A. Dudek and
Y. B. June(DUDEK; JUN, 2008) proposed a generalization of the BCI
algebras, called pseudo-BCI alge-bras, as an extension of
BCI-algebras.
Definition 3.2. A pseudo-BCI algebra, or PBCI algebra for short,
is a structure ⟨A,≤,→,↝,⊺⟩such that “≤” is a binary relation on the
setA, “→” and “↝” are binary operations onA, ⊺ ∈ Aand for all x, y,
z ∈ A:
(PB-1) x→ y ≤ (y → z)↝ (x→ z),
(PB-2) x↝ y ≤ (y ↝ z)→ (x↝ z),
(PB-3) x ≤ (x→ y)↝ y,
(PB-4) x ≤ (x↝ y)→ y,
-
27
(PB-5) x ≤ x,
(PB-6) if x ≤ y and y ≤ x, then x = y,
(PB-7) x ≤ y⇔ x→ y = ⊺⇔ x↝ y = ⊺.
Whenever x → ⊺ = ⊺, i.e. x ⪯ ⊺ for all x, the PBCI algebra
⟨A,≤,→,↝,⊺⟩ will be called aPBCK algebra.
Note that every PBCI algebra satisfying x→ y = x↝ y for all x, y
∈X is a BCI algebra.
Example 3.2.
(1) The structure A = ⟨R2,⪯,↠,→, (0,0)⟩, where (x1, y1) ↠ (x2,
y2) = (x2 − x1, (y2 −y1)e−x1) and (x1, y1) → (x2, y2) = (x2 − x1,
y2 − y1ex2−x1), is a PBCI algebra (proper,i.e. it is not a
PBCK-algebra).
(2) The structure A = ⟨(−∞,0] ⪯,↠,→, (0,0)⟩, where
x↠ y = { 0, if x ⪯ y2yπ arctan(ln(
yx)), if y < x
and
x→ y = { 0, if x ⪯ yye−tan(
πx2y
), if y < x
is a PBCK algebra.
Proposition 3.2. LetA = ⟨A,≤,→,↝,⊺⟩ be a PBCI algebra, then the
following properties holdsfor all x, y, z ∈ A:
(P-1) ⊺ ⪯ x implies x = ⊺;
(P-2) x ⪯ y implies y → z ⪯ x→ z and y ↝ z ⪯ x↝ z;
(P-3) x ⪯ y and y ⪯ z implies x ⪯ z;
(P-4) x→ (y ↝ z) = y → (x↝ z)
(P-5) x ⪯ y → z iff y ⪯ x↝ z;
(P-6) x→ y ⪯ (z → x)→ (z → y), x↝ y ⪯ (z ↝ x)↝ (z ↝ y);
(P-7) x ⪯ y implies z → x ⪯ z → y and z ↝ x ⪯ z ↝ y;
(P-8) ⊺→ x = ⊺↝ x = x;
(P-9) ((x→ y)↝ y)→ y = x→ y and ((x↝ y)→ y)↝ y = x↝ y;
(P-10) x→ y ⪯ (y → x)↝ ⊺ and x↝ y ⪯ (y ↝ x)→ ⊺;
(P-11) (x→ y)→ ⊺ = (x→ ⊺)↝ (y ↝ ⊺), (x↝ y)↝ ⊺ = (x↝ ⊺)→ (y →
⊺);
-
28
(P-12) x→ ⊺ = x↝ ⊺.
Since its definition, PBCI algebras have been investigated by
many researchers, in whichthey have defined new concepts and
applications. Among these researchers worth mentioningis Xiaohong
Zhang and Grzegorz Dymek, for their many published works on the
subject. Formore results see for example (ZHANG, 2010; LEE; PARK,
2009; HALAŠ; KÜHR, 2009; DYMEK,2012, 2013; ZHANG; PARK; WU, 2018;
ZHANG; LU; MAO, 2010; ZHANG; MA; SMARANDACHE,2017).
3.2 Final Remarks
The definition of BCI algebras and its main properties have been
presented; which algebrais well known in the literature. In
addition, it was exposed one of several generalizations of
BCIalgebras, namely: Pseudo-BCI algebras.
The previous section closes the part preliminary of the present
thesis. In what follows, thereader finds my published and submitted
contributions. They can be divided into two categories:(1) Fuzzy
Connectives and (2) An extension of BCI algebras, which was
proposed to capturethe process of intervalization of those algebras
(Semi-BCI algebras).
-
29
Part II
Contributions
-
30
4 (T,N)-Implications
In this chapter, it is introduced a class of fuzzy implication
called (T,N)-implication, ob-tained from the composition of a fuzzy
negation and a t-norm. It has been shown under whatconstraints,
(T,N)-implications preserve some main properties of fuzzy
implications, such asproperty of order, the principle of exchange,
the law of contraposition, among others. In ad-dition, we apply
this new implication class to generate a new fuzzy subsethood
measure. Andfinally the (T,N)-implications were extended to (N ′,
T,N)-implications.
4.1 Definition and Basic Properties
As well as the (S,N), QL and D-implications, defined in Chapter
2, are generalizationsof implications of classical logic for fuzzy
logic, the (T,N)-implications also satisfy the samefamily of
implications, in which it generalizes the tautology:
p→ q = ¬(p ∧ ¬q).
This session is focused on verifying if the classical law of
double negation, ¬(¬p) = p, is satis-fied thus the
(T,N)-implications coincide with (S,N)-implications. See below the
definitionof (T,N)-implication, initially defined by Bedregal in
(BEDREGAL, 2007).
Proposition 4.1. (BEDREGAL, 2007) Let T be a t-norm and N be a
fuzzy negation. Then thefunction INT ∶ [0,1]2 → [0,1] defined
by
INT (x, y) = N(T (x,N(y))) (4.1)
is a fuzzy implication.
Definition 4.1. Let T be a t-norm and N be a fuzzy negation. The
function INT defined by equa-tion (4.1) is called the
(T,N)-implication.
Proposition 4.2. (BEDREGAL, 2007) Let N be a strong fuzzy
negation and T be a t-norm. Then,
T (x, y) = N(INT (x,N(y))).
If I is a (T,N)-implication and N is a strong fuzzy negation,
then by (KLEMENT; MESIAR;PAP, 2000)(p.234), we get that I is an
S-implication, i.e.
I(x, y) = ST (N(x), y),
-
31
where, ST is as in Definition 2.8. The reciprocal is also true,
since by duality we have:
ST (N(x), y) = N(T (N(N(x)),N(y))) = N(T (x,N(y))).
Remark 4.1. If N is not a strong negation and I(x, y) = S(N(x),
y), then we say that I is a(S,N)-implication. It will be denoted by
I(S,N).
Proposition 4.3. Let INT be a (T,N)-implication. If N is a
strong fuzzy negation, then INTsatisfies (NP ), (EP ), R−CP (N) and
CP (N).
Proof. (NP) Since T is a t-norm, then by the symmetry and
1-identity properties, for any y ∈[0,1], we have
INT (1, y)(4.1)= N(T (1,N(y))) T1/T4= N(N(y)) N5= y.
(EP) Since N a strong fuzzy negation and T a t-norm, we have
INT (x, INT (y, z))(4.1)= N(T (x,N(N(T (y,N(z)))))) N5= N(T (x,T
(y,N(z))))T1= N(T (x,T (N(z), y))) T2= N(T (T (x,N(z)), y))T1= N(T
(y, T (x,N(z)))) N5= N(T (y,N(N(T (x,N(z))))))
(4.1)= INT (y, INT (x, z)).
(R-CP) Because N is strong and from the symmetry of T, we
have
INT (x,N(y))(4.1)= N(T (x,N(N(y)))) N5= N(T (x, y)) T1= N(T (y,
x))N5= N(T (y,N(N(x)))) (4.1)= INT (y,N(x)).
(CP) Again, because N is strong and from the symmetry of T, we
have
INT (N(y),N(x))(4.1)= N(T (N(y),N(N(x)))) N5= N(T (N(y), x))T1=
N(T (x,N(y))) (4.1)= INT (x, y).
Proposition 4.4. Given a (T,N)-implication INT , the following
properties are satisfied:
(i) NINT = N ;(ii) L−CP (N);
(iii) If N is strict, then R−CP (N−1).
Proof. (i) SinceN is a fuzzy negation, thenN(0) = 1, therefore
from the 1-identity propertyof the t-norm T , we have, for all x ∈
[0,1],
NINT (x)(2.3)= INT (x,0)
(4.1)= N(T (x,N(0))) N2= N(T (x,1)) T4= N(x).
-
32
(ii) From the symmetry of T , we have
INT (N(x), y)(4.1)= N(T (N(x),N(y))) T1= N(T (N(y),N(x))) (4.1)=
INT (N(y), x),
for all x, y ∈ [0,1].
(iii) Again, from the symmetry property, we have
INT (x,N−1(y))(4.1)= N(T (x,N(N−1(y)))) = N(T (x, y)) T1= N(T
(y, x))= N(T (y,N(N−1(x)))) (4.1)= INT (y,N−1(x)),
for all x, y ∈ [0,1].
Under some conditions, there are methods available for obtaining
t-conorms and t-norms ofa (T,N)-implication and a fuzzy negation.
The following propositions present these methods:
Proposition 4.5. Given a (T,N)-implication INT , define the
function SINT ∶ [0,1]2 → [0,1] by
SINT (x, y) = INT (N(x), y)
for all x, y ∈ [0,1]. Then:
(i) SINT (1, x) = SINT (x,1) = 1,∀x ∈ [0,1];
(ii) SINT is increasing in both the arguments, i.e., ∀x, y, z ∈
[0,1] with y ≤ z we haveSINT (x, y) ≤ SINT (x, z) and SINT (y, x) ≤
SINT (z, x);
(iii) SINT is commutative;
(iv) If N is strong, then SINT satisfies (S4), i.e., SINT (x,0)
= x;
(v) It N is strong, then SINT satisfies (S2), i.e., SINT (x,SINT
(y, z)) = SINT (SINT (x, y), z).
Proof. (i) As N is a fuzzy negation and T is a t-norm, then from
Proposition 2.1, we have
SINT (1, x) = INT (N(1), x)
N2= INT (0, x)(4.1)= N(T (0,N(x))) Prop.2.1= N(0) N2= 1
and
SINT (x,1) = INT (N(x),1)
(4.1)= N(T (N(x),N(1)) N2= N(T (N(x),0))T1= N(T (0,N(x)))
Prop.2.1= N(0) N2= 1.
(ii) For any x, y ∈ [0,1] we get
SINT (x, y) = INT (N(x), y)
(4.1)= N(T (N(x),N(y)))
-
33
andSINT (x, z) = I
NT (N(x), z)
(4.1)= N(T (N(x),N(z))),hence, by the monotonicity of the t-norm
T , we get that
y ≤ z N1⇒ N(z) ≤ N(y) T3⇒ T (N(x),N(z)) ≤ T (N(x),N(y)),
applying (N1), we have
N(T (N(x),N(y))) ≤ N(T (N(x),N(z)))
i.e.,SINT (x, y) ≤ SINT (x, z).
The other part follows similar.
(iii) By Proposition 4.4(ii), we have
SINT (x, y) = INT (N(x), y)
L−CP= INT (N(y), x) = SINT (y, x).
(iv) Being N a strong fuzzy negation, we have by Proposition 4.3
that
SINT (x,0)(iii)= SINT (0, x) = I
NT (N(0), x)
N2= INT (1, x)NP= x.
(v) By Proposition 4.4(ii), we have
SINT (x,SINT (y, z)) = INT (N(x), INT (N(y), z))
L−CP= INT (N(x), INT (N(z), y)).
However, being N strong, Proposition 4.3(ii) ensures that
INT (N(x), INT (N(z), y)) = INT (N(z), INT (N(x), y)),
thus:
SINT (x,SINT (y, z)) = INT (N(z), INT (N(x), y)) = SINT (z, SINT
(x, y))
(iii)= SINT (SINT (x, y), z).
From this proposition, it is possible to conclude that, if N is
strong, then SINT is a t-conorm.Thus, INT is an S-implication.
Proposition 4.6. Given a (T,N)-implication IN1T , define the
function TN2
IN1T
∶ [0,1]2 → [0,1] by
TN2IN1T
(x, y) = N2(IN1T (x,N2(y))), x, y ∈ [0,1],
where N2 is a fuzzy negation. Then, TN2IN1T
is a t-norm iff N2 is strict and N1 = N−12 .
-
34
Proof. Let us assume, firstly, that TN2IN1T
is a t-norm. To prove that N2 is strict, we must show
that N2 is decreasing and continues.
(i) N2 is decreasing.In fact, suppose there are x0, y0 ∈ [0,1]
com y0 < x0, such that N2(x0) = N2(y0). Thus,for (I2),
IN1T (1,N2(x0)) = IN1T (1,N2(y0))
N1⇒ TN2IN1T
(1, x0) = TN2IN1T
(1, y0)T1/T4⇒ x0 = y0.
Contradiction. Therefore, as N2 is a fuzzy negation, we have by
(N1) that N2 is decreas-ing.
(ii) N2 is continuous.In fact, first there is a need to check
that N2 is injective. Given x, y ∈ [0,1] with x ≠ ywe can suppose
that x < y. Then, as N2 is decreasing, we have that N2(y) <
N2(x), andtherefore, N2 is injective. N2 is also surjective, since
given y ∈ [0,1] any, as TN2
IN1T
is a
t-norm, we have
yT4= TN2
IN1T
(y,1) = N2(IN1T (y,N2(1)))N2= N2(IN1T (y,0))
(4.1)= N2(N1(T (y,1))) T4= N2(N1(y)),
therefore, there is N1(y) ∈ [0,1] such that
N2(N1(y)) = y, (4.2)
so, N2 is surjective. Thus, N2 is bijective. It then follows
that N2 is continuous be-cause otherwise there would be y ∈ [0,1]
such that y ≠ N2(x), for all x ∈ [0,1], whichcontradicts the
bijectivity of N2.
Finally, to ensure that N1 = N−12 , by Equation (4.2), just
suffice to show that N1(N2(y)) = y.Suppose N1(N2(y)) ≠ y. Then
N1(N2(y)) < y or N1(N2(y)) > y. If N1(N2(y)) < y, then
by(i), we have
N2(y) < N2(N1(N2(y)))Eq.(4.2)= N2(y).
Contradiction. Analogously, we come to a contradiction when
N1(N2(y)) > y. Let us assumenow that N2 is strict and N1 = N−12
. Then, for all x, y ∈ [0,1], we have
TN2IN1T
(x, y) = N2(IN1T (x,N2(y)))(4.1)= N2(N1(T (x,N1(N2(y)))))
N1=N−12= T (x, y).
Therefore, TN2IN1T
is a t-norm.
Lemma 4.1. If I ∈ FI satisfies (NP ), (EP ) and NI is a strong
fuzzy negation, then
TI(x, y) = NI(I(x,NI(y)))
is a t-norm.
-
35
Proof. By Propositions 2.2 and 2.3, we have that I satisfies CP
(NI) and R−CP (NI). It willverified now that TI satisfies the
conditions of the definition of t-norm:
(T1) From law of right contraposition, we get that
TI(x, y) = NI(I(x,NI(y))) R−CP= NI(I(y,NI(x))) = TI(y, x).
(T2) By virtue of NI is strong and I satisfies (EP ) and (R−CP
), we have
TI(x,TI(y, z)) N5= NI(I(x, I(y,NI(z)))) R−CP= NI(I(x,
I(z,NI(y))))(EP )= NI(I(z, I(x,NI(y)))) N5= TI(z, TI(x, y)) T1=
TI(TI(x, y), z).
(T3) It must be shown that, given x, y, z ∈ [0,1] with y ≤ z,
then TI(x, y) ≤ TI(x, z), i.e.,
NI(I(x,NI(y))) ≤ NI(I(x,NI(z))).
Indeed,
y ≤ z N2⇒ NI(z) ≤ NI(y)I2⇒ I(x,NI(z)) ≤ I(x,NI(y))
N2⇒ NI(I(x,NI(y))) ≤ NI(I(x,NI(z))).
(T4) As I satisfies (NP ), we have for x ∈ [0,1],
TI(x,1) T1= TI(1, x) = NI(I(1,NI(x))) NP= NI(NI(x)) N5= x.
Portanto, TI uma t-norm.
Theorem 4.1. For a function I ∶ [0,1]2 → [0,1], the following
statements are equivalent:
(i) I = INT is a (T,N)-implication, with N strong fuzzy
negation;
(ii) I satisfies (I1), (EP ) and (NI) is a strong fuzzy
negation.
Moreover, the representation of I = INT is unique in this
case.
Proof. (i)⇒ (ii) Because of Proposition 4.1 we get that I = INT
∈ FI , so (I1) is satisfies. Now,by Proposition 4.3, I satisfies
(EP ) and of Proposition 4.4(i) it can be concluded NI = N ,
andtherefore NI is strong.
(ii)⇒ (i) AsNI is strong, then by Propositions 2.2, 2.3 and 2.4
it can be concluded I ∈ FI .As I ∈ FI satisfies (NP ), (EP ) and NI
is a strong fuzzy negation, we have by the Lemma 4.1that TI is a
t-norm. Now, it will be demonstrated that I = INITI . In fact, for
all x, y ∈ [0,1],
INITI (x, y)(4.1)= NI(TI(x,NI(y))) = NI(NI(I(x,NI(NI(y))))) N5=
I(x, y).
-
36
Finally, to prove unity assume that there are two strong fuzzy
negations N1, N2 and twot-norms T1, T2 such that
I(x, y) = IN1T1 (x, y) = IN2T2
(x, y)for all x, y ∈ [0,1]. In particular, for y = 0 we obtain,
for all x ∈ [0,1],
NI(x)(2.3)= I(x,0) = IN1T1 (x,0) = I
N2T2
(x,0),
so,
NI(x)(2.3)= INiTi (x,0)
(4.1)= Ni(Ti(x,Ni(0))) N2= Ni(Ti(x,1)) T4= Ni(x), i = 1,2,
thus NI = N1 = N2. Now, since NI is strong, we obtain from
Proposition 4.2 that
Ti(x, y) = Ni(INiTi (x,Ni(y))), i = 1,2,
then, for all x, y ∈ [0,1],
T1(x, y) = N1(IN1T1 (x,N1(y)))hip= N1(IN2T2 (x,N1(y)))
N1=N2= N2(IN2T2 (x,N2(y))) = T2(x, y).
Therefore, the representation I = INT is unique.
Proposition 4.7. Let Nα(x) = {1, if x ≤ α0, if x > α , for
some α ∈ (0,1), and T be a t-norm. Then
INαT is the implication
INαT (x, y) =⎧⎪⎪⎨⎪⎪⎩
1, if y > α or x ≤ α0, otherwise
.
Proof. Indeed,
INαT (x, y) = Nα(T (x,Nα(y))) = {1, if y > α or x = 0
Nα(x), if y ≤ α e x > 0= { 1, if y > α or x ≤ α
0, otherwise.
Remark 4.2.
1. The t-norm T is irrelevant for INαT .
2. There is no (S,N)-implication such that INαT = I(S,N), since
if it exists and N is a fuzzynegation, then N(1) = 0 and S(N(1), y)
= y ∉ {0,1}, therefore S(N(x), y) ≠ INαT .
3. There is no t-norm T such that INαT = IT , since by
definition, for x = 1:
IT (1, y) = sup{z ∈ [0,1] ∣ T (1, z) ≤ y}= sup{z ∈ [0,1] ∣ z ≤
y}= y ≠ INαT (1, y), ∀y ∈ (0,1).
Therefore, IT ≠ INαT .
-
37
Theorem 4.2. Let T be a t-norm and N be a fuzzy negation. Then N
is strong iff INT is anS-implication with N as underlying
negation.
Proof. Assume that N is a strong fuzzy negation, then by
(KLEMENT; MESIAR; PAP, 2000)(p.234), we get that INT is an
S-implication. Conversely, let I
NT be an S-implication. Then
there exist a fuzzy negation N ′ and a t-conorm S such that INT
(x, y) = S(N ′(x), y), for allx, y ∈ [0,1]. Therefore,
x = S(N ′(1), x) = INT (1, x) = N(T (1,N(x))) = N(N(x)), ∀x ∈
[0,1],
i.e., N(N(x)) = x for all x ∈ [0,1]. Moreover,
N ′(x) (S4)= S(N ′(x),0) = INT (x,0) = N(T (x,N(0)))(N2)= N(T
(x,1)) (T4)= N(x) ∀x ∈ [0,1],
so, N ′ = N .
From Theorem 4.2, it is implied that every result which holds
for S-implications, where Nis a strong fuzzy negation, also holds
for INT . Therefore, for the purposes of this work, there isno
interest in this kind of outcomes (results with a strong N ). The
study keeps on focusing onproving results for non-strong fuzzy
negations.
Proposition 4.8. Let INT be a (T,N)-implication and let N be a
non-strong fuzzy negation:
(i) INT does not satisfy (NP );
(ii) If N is strict, then INT does not satisfy (EP );
(iii) If N is strict, then INT does not satisfy R−CP (N);
(iv) If N is strict, then INT does not satisfy CP (N).
Proof. (i) In fact, since N is not a strong fuzzy negation, then
there exists x ∈ [0,1] suchthat N(N(x)) ≠ x, so
INT (1, x) = N(T (1,N(x))) = N(N(x)) ≠ x.
(ii) Since N is not strong, then there exists x ∈ [0,1] such
that N(N(x)) ≠ x, so
INT (1, INT (x,0)) = N(T (1,N(N(T (x,N(0))))))N2= N(T (1,N(N(T
(x,1)))))
T4= N(T (1,N(N(x)))) T4= N(N(N(x)))
and
INT (x, INT (1,0)) = N(T (x,N(N(T (1,N(0))))))N2= N(T (x,N(N(T
(1,1)))))
T4= N(T (x,N(N(1)))) N2= N(T (x,1)) T4= N(x).
Therefore, by N being strict, INT (1, INT (x,0)) ≠ INT (x, INT
(1,0)).
-
38
(iii) Again, by hypothesis, there exists x ∈ [0,1] such that
N(N(x)) ≠ x, then
INT (x,N(1)) = N(T (x,1))T4= N(x)
and
INT (1,N(x)) = N(T (1,N(N(x))))T4= N(N(N(x))).
So, since N is strict, INT (x,N(1)) ≠ INT (1,N(x)).
(iv) Again, there exists x ∈ [0,1] such that N(N(x)) ≠ x,
then
INT (x,0) = N(T (x,1))T4= N(x)
and
INT (N(0),N(x)) = INT (1,N(x)) = N(T (1,N(N(x))))T4=
N(N(N(x))).
Therefore, once N is strict, INT (x,0) ≠ INT (N(0),N(x)).
Remark 4.3. As any QL-implication (R-implication, D-implication)
(see Definition 2.15) sat-isfies (NP), we have by Proposition
4.8(i) that if a (T,N)-implication is a QL-implication
(R-implication, D-implication), then N is strong and therefore it
is an S-implication.
Remark 4.4. By (DIMURO et al., 2017), a fuzzy negation N ∶ [0,1]
→ [0,1] is crisp if and onlyif there exists α ∈ [0,1[ such that N =
Nα or there exists α ∈ ]0,1] such that N = Nα, where
Nα(x) =⎧⎪⎪⎨⎪⎪⎩
0, if x > α1, if x ≤ α
and
Nα(x) =⎧⎪⎪⎨⎪⎪⎩
0, if x ≥ α1, if x < α.
Theorem 4.3. Let INT be a (T,N)-implication and let N be a crisp
fuzzy negation. Then:
(i) INT satisfies (EP );
(ii) INT satisfies R −CP (N);
(iii) INT satisfies CP (N);
(iv) INT does not satisfy (NP );
(v) INT does not satisfy (ROP );
(vi) INT satisfies (LOP ).
Proof. Consider N = Nα, for some α ∈ [0,1[, then:
-
39
(i) Given x, y, z ∈ [0,1]: (1) If z ≤ α, then Nα(z) = 1, so
INαT (x, INαT (y, z)) = Nα(T (x,Nα(Nα(T (y,Nα(z)))))) = Nα(T
(x,Nα(Nα(y))))
=⎧⎪⎪⎨⎪⎪⎩
Nα(T (x,Nα(0))), if y > αNα(T (x,Nα(1))), if y ≤ α
=⎧⎪⎪⎨⎪⎪⎩
Nα(x), if y > α1, if y ≤ α
=⎧⎪⎪⎨⎪⎪⎩
0, if x > α and y > α1, otherwise
and
INαT (y, INαT (x, z)) = Nα(T (y,Nα(Nα(T (x,Nα(z)))))) = Nα(T
(y,Nα(Nα(x))))
=⎧⎪⎪⎨⎪⎪⎩
Nα(y), if x > α1, if x ≤ α
=⎧⎪⎪⎨⎪⎪⎩
0, if x > α and y > α1, otherwise.
Therefore, for z ≤ α, INαT (x, INαT (y, z)) = INαT (y, INαT (x,
z)). (2) If z > α, thenNα(z) = 0,so:
INαT (x, INαT (y, z)) = Nα(T (x,Nα(Nα(T (y,Nα(z)))))) = Nα(T
(x,Nα(Nα(0))))= Nα(T (x,0)) = Nα(0) = 1
and
INαT (y, INαT (x, z)) = Nα(T (y,Nα(Nα(T (x,Nα(z)))))) = Nα(T
(y,Nα(Nα(0))))= Nα(T (y,0)) = Nα(0) = 1.
In any case, INαT (x, INαT (y, z)) = INαT (y, INαT (x, z)).
(ii) Given x, y ∈ [0,1]:
INαT (x,Nα(y)) = Nα(T (x,Nα(Nα(y)))) =⎧⎪⎪⎨⎪⎪⎩
Nα(x), if y > αNα(0), if y ≤ α
=⎧⎪⎪⎨⎪⎪⎩
0, if x > α and y > α1, otherwise
and
INαT (y,Nα(x)) = Nα(T (y,Nα(Nα(x)))) =⎧⎪⎪⎨⎪⎪⎩
Nα(y), if x > αNα(0), if x ≤ α
=⎧⎪⎪⎨⎪⎪⎩
0, if y > α and x > α1, otherwise.
Therefore, INαT (x,Nα(y)) = INαT (y,Nα(x)).
-
40
(iii) Given x, y ∈ [0,1]:
INαT (x, y) = Nα(T (x,Nα(y))) =⎧⎪⎪⎨⎪⎪⎩
Nα(x), if y ≤ αNα(0), if y > α
=⎧⎪⎪⎨⎪⎪⎩
0, if x > α and y ≤ α1, otherwise
and
INαT (Nα(y),Nα(x)) = Nα(T (Nα(y),Nα(Nα(x)))) =⎧⎪⎪⎨⎪⎪⎩
Nα(Nα(y)), if x > αNα(0), if x ≤ α
=⎧⎪⎪⎨⎪⎪⎩
0, if x > α and y ≤ α1, otherwise.
Therefore, INαT (x,Nα(y)) = INαT (y,Nα(x)).
(iv) Indeed,
INαT (1, y) = Nα(T (1,Nα(y))) =⎧⎪⎪⎨⎪⎪⎩
Nα(0), if y > αNα(1), if y ≤ α
=⎧⎪⎪⎨⎪⎪⎩
1, if y > α0, if y ≤ α.
Therefore, INαT (1, y) ≠ y for all y ∈ (0,1).
(v) Indeed, if y < x ≤ α, we have that INαT (x, y) = Nα(T
(x,Nα(y))) = Nα(T (x,1)) =Nα(x) = 1.
(vi) If x ≤ y, we have the following cases: (1) If x ≤ y ≤ α,
then Nα(x) = Nα(y) = 1 andNα(T (x,Nα(y))) = Nα(x) = 1; (2) If x ≤ α
< y, then Nα(x) = 1 and Nα(y) = 0,so Nα(T (x,Nα(y))) = Nα(0) =
1; (3) If α < x ≤ y, then Nα(x) = Nα(y) = 0, soNα(T (x,Nα(y))) =
Nα(0) = 1. In any case, INαT (x, y) = 1.
The proof follows analogously for N = Nα, for some α ∈
]0,1].
Remark 4.5. By item (v) of Theorem 4.3, it can be concluded INT
does not satisfy (OP) when Nis crisp.
Proposition 4.9. Let T be a positive t-norm and let N be a
frontier fuzzy negation. ThenINT (x, y) = 1, if and only if x = 0
or y = 1.
Proof. Assume INT (x, y) = 1, then N(T (x,N(y))) = 1 and,
therefore, T (x,N(y)) = 0, sinceN is a frontier fuzzy negation.
Now, take a positive T , so we have x = 0 or N(y) = 0, andtherefore
x = 0 or y = 1. Conversely, if x = 0 or y = 1 then by (vi) of
Theorem 4.3, it followsstraightforward.
Proposition 4.10. LetN be a fuzzy negation and let T be the
minimum t-norm. IfN(N(x)) ≤ x,then INT (x, INT (x, y)) = INT (x,
y).
-
41
Proof. By definition, INT (x, INT (x, y)) = N(T (x,N(N(T
(x,N(y)))))). Given a t-norm T ,T (x,N(y)) ≤ x, and so
N(N(T (x,N(y)))) ≤ N(N(x))hip.≤ x.
As T is the minimum t-norm, T (x,N(N(T (x,N(y))))) = N(N(T
(x,N(y)))), so we haveINT (x, INT (x, y)) = N(N(N(T (x,N(y))))).
But, since N is a fuzzy negation, we have thatN(N(x)) ≤ x implies
N(x) ≤ N(N(N(x))). On the other hand, taking y = N(x)
andsubstituting it in N(N(y)) ≤ y, we obtain N(N(N(x))) ≤ N(x). And
so, N(N(N(x))) =N(x). Therefore,
INT (x, INT (x, y)) = N(N(N(T (x,N(y))))) = N(T (x,N(y))) = INT
(x, y).
Proposition 4.11. Let I = INT be a (T,N)-implication. If N is a
crisp fuzzy negation, thenINT (x, INT (x, y)) = INT (x, y).
Proof. Suppose N = Nα, then:
INαT (x, INαT (x, y)) = Nα(T (x,Nα(Nα(T (x,Nα(y))))))
=⎧⎪⎪⎨⎪⎪⎩
Nα(T (x,Nα(Nα(x)))), if y ≤ αNα(T (x,0)) = 1, if y > α
=⎧⎪⎪⎨⎪⎪⎩
1, if y > α or (x ≤ α and y ≤ α)0, if x > α
and
INαT (x, y) = Nα(T (x,Nα(y))) =⎧⎪⎪⎨⎪⎪⎩
Nα(x), if y ≤ α1, if y > α
=⎧⎪⎪⎨⎪⎪⎩
1, if y > α or (x ≤ α and y ≤ α)0, if x > α.
Therefore, INαT (x, INαT (x, y)) = INαT (x, y). The result
follows analogously for N = Nα.
Proposition 4.12. Let I = INT be a (T,N)-implication. If N is a
strict fuzzy negation andINT (x, INT (x, y)) = INT (x, y), then N
is strong.
Proof. As INT (x, INT (x, y)) = INT (x, y), in particular we
have INT (1, INT (1, y)) = INT (1, y), i.e.N(T (1,N(N(T
(1,N(y)))))) = N(T (1,N(y))). Since T is a t-norm, N(N(N(N(y))))
=N(N(y)), and therefore, as N is strict, N(N(y)) = y,∀y ∈
[0,1].
Proposition 4.13. Let INT be a (T,N)-implication and let N be a
strong fuzzy negation. ThenINT (x, INT (x, y)) = INT (x, y) if and
only if T is idempotent.
Proof. Suppose INT (x, INT (x, y)) = INT (x, y), for y = 0 we
have:
INT (x,0) = N(T (x,N(0))) = N(T (x,1)) = N(x)
-
42
and
INT (x, INT (x,0)) = INT (x,N(x)) = N(T (x,N(N(x))))Nstrong= N(T
(x,x)),
so N(x) = N(T (x,x)). Since N is strong, T (x,x) = x.
Conversely, as T is idempotent, thenT is minimum, and therefore, by
Proposition 4.10, the result follows.
4.2 Functional equations and (T,N)-implications
As already mentioned, functional equations are the ones in which
the unknowns are func-tions instead of being a traditional
variable. This section investigates the validity of some
func-tional equations by the (T,N)-implication. In (BACZYŃSKI;
JAYARAM, 2008), Baczyński statesthat functional equations come up
as generalizations of the corresponding tautologies in classi-cal
logic involving boolean implications. The results presented in the
sequel consider the law ofimportation (LI), Equation 2.4, and four
basic distributive equations involving an implication,which will be
discussed later.
The exchange principle (EP) is one of the crucial properties of
fuzzy implications. Dueto the commutativity property of t-norm T ,
one of the conditions for an implication to satisfyit is that (LI)
is also satisfied. The well-known fuzzy implications called (S, N),
R, QL andD-implications satisfy (LI) under some conditions (see
(JAYARAM, 2008; MAS; MONSERRAT;TORRENS, 2009)). In addition, some
possible applications were pointed out in (JAYARAM, 2008).As
follows, the conditions under which (T,N)-implications satisfy
(LI).
Proposition 4.14. Let INT be a (T,N)-implication. Then:
(i) If N is strong then INT satisfies (LI) with respect to the
t-norm T ;
(ii) If N is continuous and INT satisfies (LI) with respect to
the t-norm T , then N is strong.
Proof. (i) Indeed, for all x, y, z ∈ [0,1], by the associativity
of T ,
INT (x, INT (y, z)) = N(T (x,N(N(T (y,N(z)))))) = N(T (x,T
(y,N(z))))= N(T (T (x, y),N(z))) = INT (T (x, y), z).
(ii) As INT satisfies (LI) with respect to the t-norm T , then,
for x = y = 1, INT (1, INT (1, z)) =INT (T (1,1), z) ⇒ N(T (1,N(N(T
(1,N(z)))))) = N(T (1,N(z))) for all z ∈ [0,1],still by the
boundary condition of T ,
N(N(N(N(z)))) = N(N(z)). (4.3)
Given that N is continuous, for all y ∈ [0,1] there exists x′ ∈
[0,1] such that N(x′) = y.Only for this x′ there exists x ∈ [0,1]
such thatN(x) = x′. Thus, for all y ∈ [0,1] there ex-ists x ∈ [0,1]
such thatN(N(x)) = y. Therefore, by Equation (4.3),N(N(N(N(x))))
=N(N(x))⇒ N(N(y)) = y, for all y ∈ [0,1].
-
43
Note that if N is continuous and non-strong then INT does not
satisfy (LI). However, thereare non-continuous negationsN such that
INT satisfies (LI) for some t-norm T . See the
followingexample:
Example 4.1. Take a crisp negation N given by N = Nα and the
minimum t-norm T , so
INαT (x, INαT (y, z)) = Nα(T (x,Nα(Nα(T (y,Nα(z))))))
=⎧⎪⎪⎨⎪⎪⎩
Nα(T (x,Nα(Nα(y)))), if z ≤ α1, if z > α
=⎧⎪⎪⎨⎪⎪⎩
Nα(x), if z ≤ α and y > α1, if z > α or y ≤ α
=⎧⎪⎪⎨⎪⎪⎩
0, if z ≤ α and y > α and x > α1, otherwise
and
INαT (T (x, y), z) = Nα(T (T (x, y),Nα(z))) =⎧⎪⎪⎨⎪⎪⎩
Nα(T (T (x, y),1)), if z ≤ α1, if z > α
=⎧⎪⎪⎨⎪⎪⎩
Nα(T (x, y)), if z ≤ α1, if z > α
=⎧⎪⎪⎨⎪⎪⎩
0, if z ≤ α and T (x, y) > α1, if z > α or T (x, y) ≤
α
=⎧⎪⎪⎨⎪⎪⎩
0, if z ≤ α and x > α and y > α1, otherwise
,
thus, INαT satisfies (LI).
Another example can be given by taking the crisp fuzzy negation
N = Nα with α = 0 andany t-norm T . In this case, by Proposition
2.1 we also have that INαT satisfies (LI).
In classical logic, the distributivity of binary operators over
one another can somehow definethe framework of the algebra imposed
by these operators. In fuzzy logic, one can find a varietyof
studies on the distributivity of t-norms over t-conorms
(BERTOLUZZA, 1993; BERTOLUZZA;DOLDI, 2004; CARBONELL et al., 1996;
KLEMENT; MESIAR; PAP, 2000). In this sense, taking intoaccount the
four basic distributive equations involving an implication,
Equations 4.4, 4.5, 4.6,4.7, in the next propositions, some
generalizations of them - which yield the distributivity
of(T,N)-implications over t-norms and t-conorms - are
presented.
I(T (x, y), z) = S(I(x, z), I(y, z)) (4.4)I(S(x, y), z) = T
(I(x, z), I(y, z)) (4.5)
I(x,S1(y, z)) = S2(I(x, y), I(x, z)) (4.6)I(x,T1(y, z)) =
T2(I(x, y), I(x, z)) (4.7)
Proposition 4.15. Let INT be a (T,N)-implication and S be a
t-conorm. Then:
(i) If T is N -dual of S and the range of N is a subset of the
idempotent elements of T thenINT satisfies Equation (4.4) with
respect to the t-norm T and to the t-conorm S;
-
44
(ii) If INT satisfies Equation (4.4) with respect to the t-norm
T and to the t-conorm S, then(1) T is N -dual of S and,(2) If N is
strict then the range of N is a subset of the idempotent elements
of T .
Proof. (i) As T is N -dual of S and the range of N is a subset
of the idempotent elements ofT , i.e., T (N(x),N(x)) = N(x) for all
x ∈ [0,1], then, for all x, y, z ∈ [0,1] :
S(INT (x, z), INT (y, z)) = S(N(T (x,N(z))),N(T (y,N(z))))= N(T
(T (x,N(z)), T (y,N(z))))
(T1) (T2)= N(T (T (x, y), T (N(z),N(z))))= N(T (T (x, y),N(z)))=
INT (T (x, y), z).
(ii) (1) As INT satisfies Equation (4.4) with respect to the
t-norm T and to the t-conorm S,then, for z = 0, N(T (T (x,
y),N(0))) = S(N(T (x,N(0))),N(T (y,N(0)))), so by(T4), N(T (x, y))
= S(N(x),N(y)) for all x, y ∈ [0,1] and(2) In particular, for x = y
= 1, S(INT (1, z), INT (1, z)) = N(T (T (1,1),N(z))), so by(T4),
S(N(N(z)),N(N(z))) = N(N(z)) for all z ∈ [0,1], since T is N -dual
of S wehave N(T (N(z),N(z))) = N(N(z)) N strict⇒ T (N(z),N(z)) =
N(z), for all z ∈ [0,1].
Corollary 4.1. Let N be a strict negation and T be a t-norm.
Then, INT satisfies Equation (4.4)if and only if T = TM and S = SM
.
In the previous corollary, the continuity ofN ensures that if
INT satisfies Equation (4.4) thenT is minimum. However, there are
non-continuous negations such that INT satisfies Equation(4.4) for
some t-norms. See the following example:
Example 4.2. Take a crisp negation N given by N = Nα and take T
as the minimum t-norm, so
S(INαT (x, z), INαT (y, z)) = S(Nα(T (x,Nα(z))),Nα(T
(y,Nα(z))))
=⎧⎪⎪⎨⎪⎪⎩
S(Nα(x),Nα(y)), if z ≤ α1, if z > α
=⎧⎪⎪⎨⎪⎪⎩
0, if z ≤ α and x > α and y > α1, otherwise
and, by Example 4.1
INαT (T (x, y), z) =⎧⎪⎪⎨⎪⎪⎩
0, if z ≤ α and x > α and y > α1, otherwise
thus, INαT satisfies Equation (4.4).
Another example can be given for any t-norm T . Just take the
crisp fuzzy negation N = Nαwith α = 0. Then, by Proposition 2.1 we
also have that IN0T satisfies Equation (4.4).
-
45
Proposition 4.16. Let INT be a (T,N)-implication. Then,
(i) INT satisfies Equation (4.5) for TM and SM , i.e.,
considering TM as T and SM as S inEquation (4.5);
(ii) If INT satisfies Equation (4.5) with respect to the t-norm
T and to the t-conorm S, then(1) S is N -dual of T and(2) If N is
strict then the range of N is a subset of the idempotent elements
of S.
Proof. (i) For all x, y, z ∈ [0,1], if x ≤ y then SM(x, y) = y
and, by (T3) and (N1), INT (y, z) ≤INT (x, z), so
TM(INT (x, z), INT (y, z)) = INT (y, z) = INT (SM(x, y), z).
Therefore, INT satisfies Equation (4.5). Similarly, if x > y
the result follows.
(ii) (1) As INT satisfies Equation (4.5) with respect to the
t-norm T and to the t-conorm S,then, for z = 0, N(T (S(x, y),N(0)))
= T (N(T (x,N(0))),N(T (y,N(0)))), so by(T4), N(S(x, y)) = T
(N(x),N(y)) for all x, y ∈ [0,1].(2) In particular, for x = y = 1,
T (INT (1, z), INT (1, z)) = INT (S(1,1), z), so by (T4),T
(N(N(z)),N(N(z))) = N(N(z)) for all z ∈ [0,1], since S is N -dual
of T we haveN(S(N(z),N(z))) = N(N(z)) N strict⇒ S(N(z),N(z)) =
N(z), for all z ∈ [0,1].
Corollary 4.2. Let N be a strict negation and T be a t-norm.
Then, INT satisfies Equation (4.5)if and only if T = TM and S = SM
.
Proposition 4.17. Let INT be a (T,N)-implication and S1 and S2
be t-conorms. Then:
(i) If S1 = S2 = SM then, for any t-norm T and any negation N ,
INT satisfies Equation (4.6);
(ii) If INT satisfies Equation (4.6) with respect to t-conorms
S1 and S2, then:(1) The range of N is a subset of the idempotent
elements of S2 and(2) If N is strict then S1 = S2 = SM .
Proof. (i) For all x, y, z ∈ [0,1], if y ≤ z then SM(y, z) = z
and, by (N1) and (T3), INT (x, y) ≤INT (x, z), so
SM(INT (x, y), INT (x, z)) = INT (x, z) = INT (x,SM(y, z)).
Therefore, INT satisfies Equation (4.6). Similarly, if y > z
the result follows.
(ii) (1) As INT satisfies Equation (4.6) then, in particular for
y = z = 0,
N(T (x,N(S1(0,0)))) = S2(N(T (x,N(0))),N(T (x,N(0)))),
so by (T4), N(x) = S2(N(x),N(x)), for all x ∈ [0,1]. (2) Since N
is strict andS2(N(x),N(x)) = N(x) for all x ∈ [0,1], then
S2(y, y) = S2(N(N−1(y)),N(N−1(y))) = N(N−1(y)) = y
-
46
for all y ∈ [0,1], so S2 = SM . On the other hand, doing x = 1
and z = y, we haveN(T (1,N(S1(y, y)))) = S2(N(T (1,N(y))),N(T
(1,N(y)))) for all y ∈ [0,1], so by(T4),
N(N(S1(y, y))) = S2(N(N(y)),N(N(y))) S2=SM= N(N(y)),
for all y ∈ [0,1]. Thus, S1(y, y) = y for all y ∈ [0,1], since N
is strict. Therefore,S1 = SM .
Corollary 4.3. Let N be a strict negation and T be a t-norm.
Then, INT satisfies Equation (4.6)if and only if S1 = S2 = SM .
Proposition 4.18. Let INT be a (T,N)-implication and T1 and T2
be t-norms. Then:
(i) If T1 = T2 = TM then, for any t-norm T and any negation N ,
INT satisfies Equation (4.7);
(ii) If INT satisfies Equation (4.7) with respect to t-norms T1
and T2, then:(1) The range of N is a subset of the idempotent
elements of T2 and(2) If N is strict then T1 = T2 = TM .
Proof. (i) For all x, y, z ∈ [0,1], if y ≤ z then TM(y, z) = y
and, by (N1) and (T3), INT (x, y) ≤INT (x, z), so
TM(INT (x, y), INT (x, z)) = INT (x, y) = INT (x,TM(y, z)).
Therefore, INT satisfies Equation (4.7). Similarly, if y > z
the result follows.
(ii) (1) As INT satisfies Equation (4.7) then, in particular for
y = z = 0,
N(T (x,N(T1(0,0)))) = T2(N(T (x,N(0))),N(T (x,N(0)))),
so by (T4), N(x) = T2(N(x),N(x)), for all x ∈ [0,1]. (2) Since N
is strict andthe range of N a subset of the idempotent elements of
T2, we have that T2(x,x) =T2(N(N−1(x)),N(N−1(x))) = N(N−1(x)) = x.
On the other hand, for x = 1 andz = y, N(T (1,N(T1(y, y)))) =
T2(N(T (1,N(y))),N(T (1,N(y)))), so by (T4),
N(N(T1(y, y))) = T2(N(N(y)),N(N(y))) T2=TM= N(N(y)),
for all y ∈ [0,1]. Thus, T1(y, y) = y for all y ∈ [0,1], since N
is strict. Therefore,T1 = TM .
There are other conditions for t-norms and negations that imply
that a (T,N)-implicationsatisfies equation (4.7). The following
example ensures that if we take T1 = TM and the crispnegation N ,
given by N = Nα with α ∈ [0,1), then, independently from t-norms T
and T2, INTsatisfies Equation (4.7).
-
47
Example 4.3. Take the crisp negation N given by N = Nα and take
T1 as the minimum t-norm,so
T2(INαT (x, y), INαT (x, z)) = T2(Nα(T (x,Nα(y))),Nα(T
(x,Nα(z))))
=⎧⎪⎪⎨⎪⎪⎩
Nα(T (x,Nα(y))), if z > αT2(Nα(T (x,Nα(y))),Nα(x)), if z ≤
α
=
⎧⎪⎪⎪⎪⎪⎪⎪⎨⎪⎪⎪⎪⎪⎪⎪⎩
1, if z > α and y > αNα(x), if z > α and y ≤ αNα(x), if
z ≤ α and y > αT2(Nα(x),Nα(x)), if z ≤ α and y ≤ α
=⎧⎪⎪⎨⎪⎪⎩
0, if x > α and T1(y, z) ≤ α1, otherwise
and
INαT (x,T1(y, z)) = Nα(T (x,Nα(T1(y, z)))) =⎧⎪⎪⎨⎪⎪⎩
1, if T1(y, z) > αNα(x), if T1(y, z) ≤ α
=⎧⎪⎪⎪⎪⎨⎪⎪⎪⎪⎩
1, if T1(y, z) > α0, if T1(y, z) ≤ α and x > α1, if T1(y,
z) ≤ α and x ≤ α
=⎧⎪⎪⎨⎪⎪⎩
0, if T1(y, z) ≤ α and x > α1, otherwise
thus, INT satisfies Equation (4.7).
4.3 Applying (T,N)-implications to generate fuzzy subset-hood
measures
Fuzzy subsethood measures determine up to what extent a fuzzy
set is included into anotherfuzzy set. Formally, given two fuzzy
sets A,B ∈ F (X), it can be said that A is included in B,written A
≤ B, if the inequality: A(x) ≤ B(x) holds for every x ∈ X (ZADEH,
1965), where ≤defines a partial order in F (X) which extends the
linear order between real numbers.
Clearly, this form