chapter 6 multi-fuzzy logic
TRANSCRIPT
CHAPTER 6
Multi-fuzzy Logic
We analyze in this chapter the notion of multi-fuzzy logic on residuated lattices
using multi-fuzzy sets as evaluations of logic formulae. We derive some equations and
inequalities, which are useful in diverse areas like multi dimensional reasoning, pat-
tern recognition, image processing, taste characterization and approximate reasoning.
We study some theorems in basic logic using multi-fuzzy evaluations and show that
multi-fuzzy logic is a tool for approximate reasoning. Also we develop the theory of
multi-fuzzy filters of residuated lattices and some characterizations of multi-fuzzy im-
plicative filters and multi-fuzzy regular filters. It seems that, multi-fuzzy filter theory
is parallel to the filter theory in general residuated lattices. In the last section we
propose theories of lattice implication algebra on multi-fuzzy sets, and implication
relation between propositions with different value domains.
6.1 IntroductionHajek [29] introduced the concepts of Basic Logic and BL-algebras as the logic sys-
tem of fuzzy logic. MV -algebras, Godel algebras and product algebras are the most
6Some results of this chapter are included in the paper Multi-fuzzy logic, which is communicatedfor publication.
113
6.2. MULTI-FUZZY LOGIC 114
known structural extensions of BL-algebras. Hajek [29] proposed the notions of filters
and prime filters in BL-algebras and later Xu [94] introduced the notion of lattice
implication algebras and Wang [90] proved that lattice implication algebras are cate-
gorically equivalent to MV -algebras. Theory of filters plays an important role in the
study of BL-algebras. Various filters correspond to various deduction systems and
in other words, they are the sets of various provable formulae. Turunen [86] intro-
duced the notion of Boolean filters and the notion of implicative filters. He proved
that implicative filters are equivalent to Boolean filters in BL-algebras. Afterwards,
Xu and Qin [96, 97] proposed the notions of positive implicative filters and fuzzy
positive implicative filters in lattice implication algebras. Jun et al. [33, 34, 74] de-
rived some characterizations of fuzzy positive implicative filters of lattice implication
algebras. Liu and Li [42, 43] introduced the notions of fuzzy filters, fuzzy Boolean
filters, fuzzy implicative filters, fuzzy positive implicative filters, fuzzy prime filters
and the cosets of fuzzy filters in BL-algebras. They derived several characterizations
of them and proved that fuzzy filters are useful tools to obtain results on classical
filters of BL-algebras. Also they proved that fuzzy Boolean filters are equivalent to
fuzzy implicative filters and fuzzy Boolean filters are fuzzy positive implicative filters
in BL-algebras. Recently, Zhu and Xu [111] developed the filter theory of general
residuated lattices and extended the above-mentioned types of fuzzy filters to resid-
uated lattices. They proposed the notions of regular filters and fuzzy regular filters
in general residuated lattices, and derived some of their characterizations.
6.2 Multi-fuzzy LogicIn this section we propose the concepts of multi-fuzzy logic and study some basic
properties of basic logic BL with multi-fuzzy sets as its evaluation. Also we study
the algebras of logic in multi-fuzzy sets and propose some relations in multi-fuzzy
6.2. MULTI-FUZZY LOGIC 115
logic. Throughout this section the constant propositions 0 and 1 the contradiction
and the tautology respectively. An order relation in∏j∈J
Lj means the product order
(see 1.2.8).
Lemma 6.2.1. If J is an indexing set and {(Lj,∧j,∨j,⊗j,⇒j, 0j, 1j) : j ∈ J} is a
family of residuated lattices, then (∏j∈J
Lj,∧,∨,⊗,⇒, 0, 1) is a residuated lattice with
respect to the operations, for every x = (xj)j∈J and y = (yj)j∈J in∏j∈J
Lj:
x⊗ y = (xj ⊗j yj)j∈J ;
x⇒ y = (xj ⇒j yj)j∈J ;
x ∧ y = (xj ∧j yj)j∈J ;
x ∨ y = (xj ∨j yj)j∈J ,
where 0 = (0j)j∈J , 1 = (1j)j∈J and 0j, 1jxj, yj ∈ Lj for each j ∈ J.
Proof. Follows from the definitions.
Lemma 6.2.2. If J is an indexing set and {(Lj,∧j,∨j,⊗j,⇒j, 0j, 1j) : j ∈ J} is
a family of BL algebra, then (∏j∈J
Lj,∧,∨,⊗,⇒, 0, 1) is a BL algebra with respect to
the operations defined in 6.2.1.
Proof. By Lemma 6.2.1, we have (∏j∈J
Lj,∧,∨,⊗,⇒, 0, 1) is a residuated lattice. We
need to prove the properties of pre-linearity divisibility. For every x = (xj)j∈J and
y = (yj)j∈J in∏j∈J
Lj:
(x⇒ y) ∨ (y ⇒ x) = ((xj ⇒j yj) ∨j (yj ⇒ xj))j∈J = (1j)j∈J = 1.
x⊗ (x⇒ y) = (xj ⊗j (xj ⇒j yj))j∈J = (xj ∧j yj)j∈J = x ∧ y.
6.2. MULTI-FUZZY LOGIC 116
Definition 6.2.3. Let X be the set of all formulae (including propositional variables)
in a logic system. A multi-fuzzy set A in X (with a residuated lattice∏j∈J
Lj as its
value domain) is a multi-fuzzy evaluation, if it satisfies the following conditions:
A(0) = 0;
A(φ→ ψ) =A(φ)⇒ A(ψ),∀φ, ψ ∈ X;
A(φ& ψ) =A(φ)⊗ A(ψ), ∀φ, ψ ∈ X.
Theorem 6.2.4. Let A,B,C be multi-fuzzy sets in X with same value domain∏j∈J
Lj.
If A,B,C are evaluations of a BL system, then for every φ, ψ, χ ∈ X :
(1) 0 ≤∧φ∈X
A(φ);
(2)∨ψ∈X
(A(φ)⊗B(ψ)) ≤ A(φ);
(3) (A(φ)⊗B(ψ)) = (B(ψ)⊗ A(φ));
(4) (A(φ)⊗ (B(ψ)⊗ C(χ))) = ((A(φ)⊗B(ψ))⊗ C(χ))).
Proof. (1) to (4) follow from the definition of ⊗ in the product lattice.
Note 6.2.5. Following results are immediately obtained from the definitions of join,
meet, equivalence, negation and 1-tautology mentioned in the subsection 1.6.1.
(1) A(¬φ) = A(φ→ 0) = A(φ)⇒ 0;
(2) A(1) = A(¬0) = A(0)⇒ 0 = 0⇒ 0 = 1;
(3) A(φ ≡ ψ) = A((φ→ ψ) & (ψ → φ)) = (A(φ)⇒ A(ψ))⊗ (A(ψ)⇒ A(φ));
(4) A(φ ∧ ψ) = A(φ& (φ→ ψ)) = A(φ)⊗ (A(φ)⇒ A(ψ)) = A(φ) ∧ A(ψ), since the
part (e) of Definition 1.6.7 ;
(5) A(φ ∨ ψ) = A(((φ→ ψ)→ ψ) ∧ ((ψ → φ)→ φ))
= A(((φ→ ψ)→ ψ)) ∧ A(((ψ → φ)→ φ))
6.2. MULTI-FUZZY LOGIC 117
= ((A(φ)⇒ A(ψ))⇒ A(ψ)) ∧ ((A(ψ)⇒ A(φ))⇒ A(φ))
= A(φ) ∨ A(ψ), by the part (5) of Lemma 1.6.8.
Proposition 6.2.6. [29] Each BL-algebra is a distributive lattice.
Theorem 6.2.7. If multi-fuzzy sets A,B,C in X (with same value domain) are
evaluations of a BL system, then for every φ, ψ, χ ∈ X :
(1)∨ψ∈X
(A(φ) ∧B(ψ)) ≤ A(φ);
(2) A(φ) ≤∧ψ∈X
(A(φ) ∨B(ψ));
(3) (A(φ) ∧B(ψ)) = (B(ψ) ∧ A(φ));
(4) (A(φ) ∨B(ψ)) = (B(ψ) ∨ A(φ));
(5) (A(φ) ∧ (B(ψ) ∧ C(χ))) = ((A(φ) ∧B(ψ)) ∧ C(χ)));
(6) (A(φ) ∨ (B(ψ) ∨ C(χ))) = ((A(φ) ∨B(ψ)) ∨ C(χ)));
(7)∧ψ∈X
(A(φ) ∨ (A(φ) ∧B(ψ))) = A(φ) =∨ψ∈X
(A(φ) ∧ (A(φ) ∨B(ψ)));
(8) A(φ) ∨ (B(ψ) ∧ C(χ)) = (A(φ) ∨B(ψ)) ∧ (A(φ) ∨ C(χ));
(9) A(φ) ∧ (B(ψ) ∨ C(χ)) = (A(φ) ∧B(ψ)) ∨ (A(φ) ∧ C(χ)).
Proof. (1) to (7) follow from the definition of lattice, and (8) and (9) follow from the
distributive property of BL algebra (see 6.2.6 ).
Note 6.2.8. If φ→ ψ is a tautology in basic logic BL and if a multi-fuzzy set A is
an evaluation of formulae, then A(φ) ⇒ A(ψ) = 1. Lemma 6.2.2 and Lemma 1.6.8
implies A(φ) ≤ A(ψ).
Theorem 6.2.9. Let A be a multi-fuzzy set in X. If A is an evaluation of a BL
system, then for every φ, ψ, χ ∈ X :
(1) (A(φ)⇒ A(ψ)) ≤∧χ∈X
((A(ψ)⇒ A(χ))⇒ (A(φ)⇒ A(χ)));
6.2. MULTI-FUZZY LOGIC 118
(2) ((A(φ)⇒ A(ψ))⇒ A(χ)) ≤ (((A(ψ)⇒ A(φ))⇒ A(χ))⇒ A(χ));
(3) (A(φ)⊗ (A(φ)⇒ A(ψ))) = (A(ψ)⊗ (A(ψ)⇒ A(φ)));
(4) (A(φ)⇒ (A(ψ)⇒ A(χ))) = ((A(φ)⊗ A(ψ))⇒ A(χ)));
(5) (A(φ)⇒ A(ψ)) ∨ (A(ψ)⇒ A(φ)) = 1.
Proof. (1) Since (φ→ ψ)→ ((ψ → χ)→ (φ→ χ)) is a 1-tautology,
A(φ → ψ) ≤ A(((ψ → χ) → (φ → χ)))(see Definition 1.6.1 and Note 6.2.8).
Therefore (A(φ) ⇒ A(ψ)) ≤ ((A(ψ) ⇒ A(χ)) ⇒ (A(φ) ⇒ A(χ))),∀χ ∈ X. Hence
(A(φ)⇒ A(ψ)) ≤∧χ∈X
((A(ψ)⇒ A(χ))⇒ (A(φ)⇒ A(χ))).
Similarly (2) to (4) follow immediately from Definition 1.6.1 and Note 6.2.8, and (5)
follows from 6.2.2.
Proposition 6.2.10. (see [29,78,86,111]). If L is a BL algebra and x, y ∈ L, then:
(1) x ≤ (x′ ⇒ y);
(2) x ≤ (x′)′;
(3) (x⇒ y) ≤ (y′ ⇒ x′);
(4) (x⊗ x′) = 0;
(5) x ≤ (1⊗ x);
(6) (x ∨ y)′ = (x′ ∧ y′);
(7) (x ∧ y)′ = (x′ ∨ y′).
Theorem 6.2.11. Let A be a multi-fuzzy set in X. If A is an evaluation of a BL
system, then for every φ, ψ ∈ X :
(1) A(φ) ≤∧ψ∈X
(A′(φ)⇒ A(ψ));
(2) A(φ) ≤ (A′(φ))′;
6.2. MULTI-FUZZY LOGIC 119
(3) (A(φ)⇒ A(ψ)) ≤ (A′(ψ)⇒ A′(φ));
(4) (A(φ)⊗ A′(φ) = 0;
(5) A(φ) ≤ (1⊗ A(φ));
(6) (A(φ) ∨ A(ψ))′ = (A′(φ) ∧ A′(ψ));
(7) (A(φ) ∧ A(ψ))′ = (A′(φ) ∨ A′(ψ)).
Proof. (1) to (7) are obtained if we replace x by A(φ) and y by A(ψ) in the Proposi-
tion 6.2.10.
Proposition 6.2.12. (see [29]) Basic logic satisfies the following relations:
(1) (φ& (φ→ ψ))→ ψ;
(2) φ→ (ψ → (φ& ψ));
(3) φ→ (ψ → φ);
(4) (φ→ (ψ → χ))→ (ψ → (φ→ χ));
(5) (φ→ ψ)→ ((φ& χ)→ (ψ & χ));
(6) (φ→ ψ) & (χ→ ω)→ ((φ& χ)→ (ψ & ω));
(7) (φ→ ψ)→ (φ→ (φ ∧ ψ));
(8) ((φ→ ψ) ∧ (φ→ χ))→ (φ→ (ψ ∧ χ));
(9) (φ→ ψ)→ ((φ ∨ ψ)→ φ);
(10) ((φ→ χ) ∧ (ψ → χ))→ ((φ ∨ ψ)→ χ);
(11) φ& (ψ ∧ χ) ≡ (φ& ψ) ∧ (φ& χ);
(12) φ& (ψ ∨ χ) ≡ (φ& ψ) ∨ (φ& χ).
6.2. MULTI-FUZZY LOGIC 120
Theorem 6.2.13. Let A be a multi-fuzzy set in X. If A is an evaluation of a BL
system, then for every φ, ψ, χ, ω ∈ X :
(1)∨φ∈X
(A(φ)⊗ (A(φ)⇒ A(ψ))) ≤ A(ψ);
(2) A(φ) ≤∧ψ∈X
(A(ψ)⇒ (A(φ)⊗ A(ψ)));
(3) A(φ) ≤∧ψ∈X
(A(ψ)⇒ A(φ));
(4) (A(φ)⇒ (A(ψ)⇒ A(χ))) ≤ (A(ψ)⇒ (A(φ)⇒ A(χ)));
(5) (A(φ)⇒ A(ψ)) ≤∧χ∈X
((A(φ)⊗ A(χ))⇒ (A(ψ)⊗ A(χ)));
(6) (A(φ)⇒ A(ψ))⊗ (A(χ)⇒ A(ω)) ≤ ((A(φ)⊗ A(χ))⇒ (A(ψ)⊗ A(ω)));
(7) (A(φ)⇒ A(ψ)) ≤ (A(φ)⇒ (A(φ) ∧ A(ψ)));
(8) (A(φ)⇒ A(ψ)) ∧ (A(φ)⇒ A(χ)) ≤ (A(φ)⇒ (A(ψ) ∧ A(χ)));
(9) (A(φ)⇒ A(ψ)) ≤ ((A(φ) ∨ A(ψ))⇒ A(φ));
(10) ((A(φ)⇒ A(χ)) ∧ (A(ψ)⇒ A(χ))) ≤ ((A(φ) ∨ A(ψ))⇒ A(χ));
(11) A(φ)⊗ (A(ψ) ∧ A(χ)) = (A(φ)⊗ A(ψ)) ∧ (A(φ)⊗ A(χ));
(12) A(φ)⊗ (A(ψ) ∨ A(χ)) = (A(φ)⊗ A(ψ)) ∨ (A(φ)⊗ A(χ)).
Proof. (1)-(12) follow from the definitions and the properties of basic logic BL in [29]
(see 6.2.12).
Note 6.2.14. Important generalized inference rules in multi-fuzzy logic are the
following:
(1) Generalized modus ponens, that is,∨φ∈X
(A(φ) ∧ (A(φ)⇒ B(ψ))) = B(ψ),∀ψ ∈ X;
6.3. DEDUCTIVE SYSTEMS OF RESIDUATED LATTICES 121
(2) Generalized modus tollens, that is,∨ψ∈X
(B′(ψ) ∧ (A(φ)⇒ B(ψ))) = A′(φ),∀φ ∈ X;
(3) Generalized hypothetical syllogism, that is,∨ψ∈X
(A(φ)⇒ B(ψ)) ∧ (B(ψ)⇒ C(χ)) = (A(φ)⇒ C(χ)),∀φ, χ ∈ X.
These inference rules are suitable for approximate reasoning.
6.3 Deductive Systems of Residuated LatticesIn this section we introduce the notions of multi-fuzzy filter, multi-fuzzy implicative
filter and multi-fuzzy regular filter, and derive some equivalent conditions for them.
Proposition 6.3.1. [29, 78, 86] Let L be a residuated lattice. For any x, y, z ∈ L,
the following properties hold:
(RL1) (x⊗ y)→ z = x→ (y → z);
(RL2) x→ (y → z) = y → (x→ z);
(RL3) y → z ≤ (x→ y)→ (x→ z);
(RL4) x→ y ≤ (y → z)→ (x→ z);
(RL5) x→ y ≤ y′ → x′;
(RL6) x ≤ y implies z → x ≤ z → y;
(RL7) x ≤ y implies y → z ≤ x→ z and y′ ≤ x′;
(RL8) 1→ x = x, x→ x = 1;
(RL9) 1′ = 0, 0′ = 1;
(RL10) x ≤ y if and only if x→ y = 1;
(RL11) x→ (y ∧ z) = (x→ y) ∧ (x→ z);
6.3. DEDUCTIVE SYSTEMS OF RESIDUATED LATTICES 122
(RL12) (x ∨ y)→ z = (x→ z) ∧ (y → z);
(RL13) x→ (y → (x⊗ y)) = 1;
(RL14) x⊗ (y ∨ z) = (x⊗ y) ∨ (x⊗ z);
(RL15) (x ∨ y)′ = x′ ∧ y′;
(RL16) x⊗ y ≤ x ∧ y, x⊗ x′ = 0;
(RL17) y ≤ x→ y, x′ ≤ x→ y;
(RL18) x′ = x′′′, x ≤ x′′;
(RL19) x ∨ x′ = 1 implies x ∧ x′ = 0;
(RL20) x ∨ y ≤ ((x→ y)→ y) ∧ ((y → x)→ x).
6.3.1 Filters of Residuated LatticesDefinition 6.3.2. [36, 37, 42, 111] A non-empty subset F of a residuated lattice L
is called a filter of L, if it satisfies the axioms, for all x, y ∈ L :
(1) x, y ∈ F, implies x⊗ y ∈ F ;
(2) x ∈ F and x ≤ y implies y ∈ F.
Remark 6.3.3. Throughout this subsection we assume that elements of F are
nonzero.
Proposition 6.3.4. [78] Let F be a subset of L containing 1. The following
assertions are equivalent, for all x, y, z ∈ L :
(1) F is a filter of L;
(2) x, x→ y ∈ F implies y ∈ F ;
(3) x→ y, y → z ∈ F implies x→ z ∈ F ;
6.3. DEDUCTIVE SYSTEMS OF RESIDUATED LATTICES 123
(4) x→ y, x⊗ z ∈ F implies y ⊗ z ∈ F ;
(5) x, y ∈ F and x ≤ y → z implies z ∈ F.
Definition 6.3.5. [30, 35, 86, 95, 109–111] Let F be a filter of a residuated lattice
L. For all x, y ∈ L :
(1) F is a Boolean filter (B-filter), if x ∨ x′ ∈ F.
(2) F is a G-filter (G-filter), if x2 → y ∈ F implies x→ y ∈ F, where x2 = x⊗ x.
(3) F is a MV filter (MV-filter), if y → x ∈ F implies ((x→ y)→ y)→ x ∈ F.
(4) F is a regular filter (R-filter), if x′′ → x ∈ F, where x′′ = (x′)′.
6.3.2 Fuzzy FilterDefinition 6.3.6. [36,37,42,111] A fuzzy set µ of a residuated lattice L is called a
multi-fuzzy filter of L, if it satisfies the axioms, for all x, y ∈ L :
(1) min{µ(x), µ(y)} ≤ µ(x⊗ y);
(2) x ≤ y implies µ(x) ≤ µ(y).
Remark 6.3.7. Throughout this subsection we assume that µ(0) = 0.
Definition 6.3.8. [30, 31, 35–37, 42, 86, 95, 108–111] Let µ be a fuzzy filter of a
residuated lattice L. For all x, y ∈ L :
(1) µ is a fuzzy Boolean filter (FB-filter), if µ(x ∨ x′) = µ(1);
(2) µ is a fuzzy G-filter (FG-filter), if µ(x→ (x→ y)) ≤ µ(x→ y);
(3) µ is a fuzzy MV filter (FMV-filter), if µ(y → x) ≤ µ(((x→ y)→ y)→ x);
(4) µ is a fuzzy regular filter (FR-filter), if µ(x′′ → x) = µ(1);
(5) µ is a fuzzy implicative filter, if µ(x→ (z′ → y)) ∧ µ(y → z) ≤ µ(x→ z).
6.3. DEDUCTIVE SYSTEMS OF RESIDUATED LATTICES 124
6.3.3 Multi-fuzzy FilterMulti-fuzzy filters are the deductive systems of multi-fuzzy logic. In this subsection
we introduce the basic notions of multi-fuzzy filters and propose some results. Some
of the results included in this subsection are similar to the respective results in fuzzy
filters proposed by Liu and Li [42], and Zhu and Xu [111].
Definition 6.3.9. A multi-fuzzy set A of a residuated lattice L is called a multi-fuzzy
filter of L, if it satisfies the axioms, for all x, y ∈ L :
(MFF1) A(x) ∧ A(y) ≤ A(x⊗ y);
(MFF2) x ≤ y implies A(x) ≤ A(y).
Remark 6.3.10. Throughout this subsection we assume that A(0) = 0.
Theorem 6.3.11. Let A be a multi-fuzzy set in a residuated lattice L. Then the
following assertions are equivalent, for all x, y, z ∈ L :
(1) A is a multi-fuzzy filter of L;
(2) x ≤ y → z implies A(x) ∧ A(y) ≤ A(z);
(3) A(x) ≤ A(1) and A(x→ y) ∧ A(x) ≤ A(y).
Proof. (1)⇒ (2). Assume that, A is a multi-fuzzy filter of L.
x ≤ y → z implies x⊗ y ≤ z, by the adjoint property
implies A(x⊗ y) ≤ A(z), by (MFF2)
impliesA(x) ∧ A(y) ≤ A(z), by (MFF1).
(2)⇒ (3). Put y = x and z = 1 in (2), we have A(x) ∧ A(x) ≤ A(1), since
x ≤ x → 1. That is, A(x) ≤ A(1), ∀x ∈ L. Since (x → y) ≤ x → y, ∀x ∈ L and by
the assumption (2) we have A(x→ y) ∧ A(x) ≤ A(y).
6.3. DEDUCTIVE SYSTEMS OF RESIDUATED LATTICES 125
(3)⇒ (1). By the assumption (3), we have A(x) ≤ A(1), and if x ≤ y, then
x → y = 1. Hence A(x) = A(x) ∧ A(1) = A(x) ∧ A(x → y) ≤ A(y). That is, x ≤ y
implies A(x) ≤ A(y). The adjoint property of residuated lattices implies,
x→ (y → (x⊗ y)) = (x⊗ y)→ (x⊗ y) = 1. (6.i)
A(x) ∧ A(y) = (A(x) ∧ A(1)) ∧ A(y)
= (A(x) ∧ A(x→ (y → (x⊗ y)))) ∧ A(y), by the equation (6.i)
≤A(y → (x⊗ y)) ∧ A(y), by the assumption (3)
≤A(x⊗ y), by the assumption (3).
Hence A is a multi-fuzzy filter.
Definition 6.3.12. A multi-fuzzy filter A of a residuated lattice L is called a multi-
fuzzy implicative filter, if it satisfies the axiom:
(MFF3) A(x→ (z′ → y)) ∧ A(y → z) ≤ A(x→ z),∀x, y, z ∈ L.
Theorem 6.3.13. Let A be a multi-fuzzy filter of L. Then the following assertions
are equivalent, for all x, y, z ∈ L :
(1) A is a multi-fuzzy implicative filter of L;
(2) A(x→ (z′ → z)) ≤ A(x→ z);
(3) A(x→ (z′ → z)) = A(x→ z);
(4) A(y → (x→ (z′ → z))) ∧ A(y) ≤ A(x→ z).
Proof. (1) ⇒ (2). Assume that, A is a multi-fuzzy implicative filter. In (MFF3)
put y = z, we have A(x → (z′ → z)) ∧ A(z → z) ≤ A(x → z). That is,
A(x→ (z′ → z)) ∧ A(1) ≤ A(x→ z). Hence A(x→ (z′ → z)) ≤ A(x→ z).
6.3. DEDUCTIVE SYSTEMS OF RESIDUATED LATTICES 126
(2) ⇒ (3). Suppose that (2) holds. Proposition 6.3.1 (see (RL17) and (RL2))
implies
x→ z ≤ z′ → (x→ z) = x→ (z′ → z). (6.ii)
(6.ii) and (MFF2) imply that A(x → z) ≤ A(x → (z′ → z)). This and the
assumption (2) together imply A(x→ (z′ → z)) = A(x→ z).
(3) ⇒ (4). Suppose that (3) holds. Theorem 6.3.11 (see part 3 of the theorem)
and the assumption (3) together imply
A(y → (x→ (z′ → z))) ∧ A(y) ≤ A(x→ (z′ → z)) = A(x→ z).
(4) ⇒ (1). Suppose that (4) holds for every x, y, z ∈ L. Replace y by y → z in (4),
we have
A((y → z)→ (x→ (z′ → z))) ∧ A(y → z) ≤ A(x→ z). (6.iii)
Using Proposition 6.3.1 (see (RL2)), we have that
(y → z)→ (x→ (z′ → z)) =x→ ((y → z)→ (z′ → z))
=x→ (z′ → ((y → z)→ z)). (6.iv)
y ≤ y ∨ z ≤ ((y → z)→ z) ∧ ((z → y)→ y) ≤ (y → z)→ z. (6.v)
(6.v) and Proposition 6.3.1 (see (RL6)) together imply
z′ → y ≤ z′ → ((y → z)→ z).
6.3. DEDUCTIVE SYSTEMS OF RESIDUATED LATTICES 127
Again by Proposition 6.3.1 (see (RL6)), we have
x→ (z′ → y) ≤ x→ (z′ → ((y → z)→ z)). (6.vi)
(6.vi), (MFF2) and (6.iv) together imply
A(x→ (z′ → y))≤A(x→ (z′ → ((y → z)→ z)))
=A((y → z)→ (x→ (z′ → z))). (6.vii)
(6.vii) and (6.iii) together imply
A(x→ (z′ → y)) ∧ A(y → z)≤A((y → z)→ (x→ (z′ → z))) ∧ A(y → z)
≤A(x→ z).
This proves that the condition (1) holds.
Definition 6.3.14. A multi-fuzzy filter A of a residuated lattice L is called a
multi-fuzzy regular filter (MFR-filter), if it satisfies the axiom:
(MFF4) A(x′′ → x) = A(1), for all x ∈ L.
Theorem 6.3.15. Let A be a multi-fuzzy filter of L. Then the following assertions
are equivalent, for all x, y, z ∈ L :
(1) A is an MFR-filter of L;
(2) A(x′ → y′) ≤ A(y → x);
(3) A(x′ → y′) = A(y → x);
(4) A(x′ → y) ≤ A(y′ → x);
(5) A(x′ → y) = A(y′ → x).
6.3. DEDUCTIVE SYSTEMS OF RESIDUATED LATTICES 128
Proof. (1) ⇒ (2). Suppose thatA is an MFR-filter. Proposition 6.3.1 (see (RL5), (RL7)
and (RL18)) implies x′ → y′ ≤ y′′ → x′′ ≤ y → x′′ and hence
x′′ → x ≤ (y → x′′)→ (y → x) ≤ (x′ → y′)→ (y → x).
Thus, by (MFF2) we have
A(x′′ → x) ≤ A((x′ → y′)→ (y → x)). (6.viii)
A(x′ → y′) =A(x′ → y′) ∧ A(1).
=A(x′ → y′) ∧ A(x′′ → x), by Definition 6.3.14
≤A(x′ → y′) ∧ A((x′ → y′)→ (y → x)), by (6.viii)
≤A(y → x), ∀x, y ∈ L by Theorem 6.3.11. (6.ix)
(2) ⇒ (3). Suppose that A satisfies the the condition (2). By Proposition 6.3.1
(see (RL5)) and (MFF2), we have
A(y → x) ≤ A(x′ → y′) (6.x)
(6.x) and the assumption (2) together imply A(y → x) = A(x′ → y′).
(3) ⇒ (4). Suppose that (3) holds. Using Proposition 6.3.1 (see (RL18) and
(RL6)), we have
x′ → y ≤ x′ → y′′ = x′ → (y′)′. (6.xi)
6.4. MULTI-FUZZY LOGIC IN LIA 129
The assumption (3) and 6.xi together imply
A(x′ → y) ≤ A(x′ → (y′)′) = A(y′ → x).
(4) ⇒ (5). Suppose that A satisfies the condition (4) for all x, y ∈ L. Replace x
by y and y by x in (4), we have
A(y′ → x) ≤ A(x′ → y),∀x, y ∈ L. (6.xii)
(6.xii) and the assumption (4) together imply A(x′ → y) = A(y′ → x).
(5) ⇒ (1). Suppose that (5) holds. Using Proposition 6.3.1 (see (RL8)) and the
assumption (5), we have
A(1) = A(x′ → x′) = A(x′ → (x′)) = A((x′)′ → x) = A(x′′ → x).
Hence A is an MFR-filter.
6.4 Multi-fuzzy Logic in LIAIn this section we introduce and study the concept of lattice implication algebra on
multi-fuzzy sets and implication relation between propositions with different value
domains.
Definition 6.4.1. [94] A bounded lattice (L,∨,∧, 0, 1) with order-reversing involu-
tion ′ and a binary operation→ is a lattice implication algebra (LIA) if the following
axioms hold for every x, y, z ∈ L :
(1) x→ (y → z) = y → (x→ z);
(2) x→ x = 1;
6.4. MULTI-FUZZY LOGIC IN LIA 130
(3) x→ y = y′ → x′;
(4) x→ y = y → x = 1⇒ x = y;
(5) (x→ y)→ y = (y → x)→ x;
(6) (x ∨ y)→ z = (x→ z) ∧ (y → z);
(7) (x ∧ y)→ z = (x→ z) ∨ (y → z).
Proposition 6.4.2. Let {(Lj,∨j,∧,′j ,→j, 0j, 1j) : j ∈ J} be a family of lattice
implication algebra (LIA), then (∏Lj,t,u,′ ,→, 0, 1) is an LIA with respect to the
following operations. For every x, y ∈∏Lj :
x ∨ y = (xj ∨j yj)j∈J ;
x ∧ y = (xj ∧j yj)j∈J ;
x→ y = (xj →j yj)j∈J ;
(x)′ = (x′j)j∈J .
Proof. Follows immediately from the definition.
Proposition 6.4.3. Let {(Lj,∨j,∧,′j ,→j, 0j, 1j) : j ∈ J} be a family of lattice im-
plication algebra (LIA). For every x, y ∈∏Lj, the operations x∨y, x∧y, (x)′, x→ y
defined as in Proposition 6.4.2 and
x⊕ y = x′ → y, x⊗ y = (x→ y′)′.
Thenx⊕ y = (xj ⊕j yj)j∈J and x⊗ y = (xj ⊗j yj)j∈J ,
wherexj ⊕j yj = x′j →j yj and xj ⊗j yj = (xj →j y
′j)′.
6.4. MULTI-FUZZY LOGIC IN LIA 131
Proof. For every x, y ∈∏Lj,
x⊕ y=x′ → y
= (x′j)j∈J → (yj)j∈J
= (x′j →j yj)j∈J
= (xj ⊗j yj)j∈J .
Similarly
x⊗ y= (x→ y′)′
= ((xj)j∈J → (yj)′j∈J)′
= ((xj →j y′j)′)j∈J
= (xj ⊗j yj)j∈J .
Example 6.4.4. Let X be a nonempty set and {(Lj,∨j,∧,′j ,→j, 0j, 1j) : j ∈ J}
be a family of lattice implication algebra (LIA). For every x ∈ X and A,B ∈∏LXj ;
define
(A tB)(x) = A(x) ∨B(x),
(A uB)(x) = A(x) ∧B(x),
(A→ B)(x) = A(x)⇒ B(x)
and
A′(x) = (A(x))′.
As usual we can verify that (∏LXj ,t,u,′ ,→,0X ,1X) is an LIA. Note that A→ B is
6.4. MULTI-FUZZY LOGIC IN LIA 132
the multi-fuzzy set {(x,A(x) ⇒ B(x)) : x ∈ X}. If A and B are crisp subsets of X,
then A→ B is the set compliment of A \B.
Theorem 6.4.5. [98] Let (L,∨,∧,′ ,→, 0, 1) be a lattice implication algebra, then
(L,∨,∧) is a distributive lattice.
Theorem 6.4.6. [98] Let (L,∨,∧,′ ,→, 0, 1) be a lattice implication algebra, the
lattice operations disjunction ∨, conjunction ∧, implication → and the complement ′
has the following relationships, for all x, y, z ∈ L :
(1) x ∨ y = (x→ y)→ y;
(2) x ∧ y = (x′ ∨ y′)′;
(3) x ≤ y if and only if x→ y = 1;
(4) x→ 0 = x′, 0→ x = 1, 1→ x = x, x→ 1 = 1;
(5) x→ y = 0 if and only if x = 1 and y = 0;
(6) x→ y ≤ (y → z)→ (x→ z);
(7) x′ ∨ y ≤ x→ y.
Theorem 6.4.7. [99] Let (L,∨,∧,′ ,→, 0, 1) be an LIA, for every x, y ∈ L, if x ≤ y,
then the following conditions are equivalent:
(1) y → z ≤ x→ z,∀z ∈ L;
(2) z → x ≤ z → y,∀z ∈ L;
(3) x→ z = x→ (y ∧ z), ∀z ∈ L;
(4) z → y = (z ∨ x)→ y,∀z ∈ L.
Definition 6.4.8. [98] A mapping f : L1 → L2 from lattice implication algebras L1
to L2 is called a lattice implication homomorphism if it satisfies the following axioms,
for every x, y ∈ L1 :
6.4. MULTI-FUZZY LOGIC IN LIA 133
(1) f(x→ y) = f(x)→ f(y);
(2) f(x ∨ y) = f(x) ∨ f(y);
(3) f(x ∧ y) = f(x) ∧ f(y);
(4) f(x′) = f(x)′.
For more details of lattice implication algebra, we refer readers to [94,97,98].
Theorem 6.4.9. Let L and M be lattice implication algebras. If the mapping
h : M → L is a lattice implication homomorphism, then:
(1) h(1M) = 1L;
(2) h(0M) = 0L.
Proof. For every x ∈M,
(1) h(1M) = h(x→M x) = h(x)→L h(x) = 1L.
(2) h(x)→ 0L = (h(x))′ = h(x′) = h(x→ 0) = h(x)→ h(0).
Therefore,
h(0)→ 0L = h(0)→ h(0) = 1L
and implies h(0) ≤ 1L. But 0L ≤ h(x), for every x ∈M and so h(0M) = 0L.
Theorem 6.4.10. Let h :∏Mi →
∏Lj be a lattice implication homomorphism and
x, y ∈∏Mi, then:
(1) h(x⊕M y) = h(x)⊕L h(y);
(2) h(x⊗M y) = h(x)⊗L h(y);
(3) h(1M) = 1L;
(4) h(0M) = 0L,
6.4. MULTI-FUZZY LOGIC IN LIA 134
where 0M , 1M , 0L and 1L are the least element of∏Mi, greatest element of
∏Mi,
least element of∏Li and greatest element of
∏Li respectively.
Proof. For every x, y ∈∏Mi,
(1) h(x⊕M y) = h(x′ →M y) = h(x′)→L h(y) = h(x)′ →L h(y) = h(x)⊕L h(y).
(2) h(x⊗M y) = h((x→M y′)′) = (h(x→M y′))′ = (h(x)→L h(y′))′
= (h(x)→L h(y)′)′ = h(x)⊗L h(y).
(3) Similar to part 1 of Theorem 6.4.9.
(4) Similar to part 2 of Theorem 6.4.9.
6.4.1 Implication Between Different Value DomainsThroughout this subsection M =
∏Mi and L =
∏Lj. The relations →M and →L
represent the implication relations in M and L respectively.
Definition 6.4.11. Let and h :∏Mi →
∏Lj be a lattice implication homomor-
phism. For every x ∈∏Mi and y ∈
∏Lj, an implication relation ’x implies y’ under
the bridge function h denoted by x →h y and defined by h(x) →L y , where →L is
the implication operation in L.
Lemma 6.4.12. Let f : X → Y be a crisp function, A ∈∏MX
i , B ∈∏LYj , 1L be
the greatest element of∏Lj and the mapping h :
∏Mi →
∏Lj be a bridge function
for the multi-fuzzy extension of f . If y = f(x), then A(x)→h B(y) = 1L.
Proof. Since x ∈ f−1(y) and h(A(x)) ≤∨
t∈f−1(y)
h(A(t)),
A(x)→h B(y) = h(A(x))→L B(y) = h(A(x))→L
∨t∈f−1(y)
h(A(t)) = 1L.
6.4. MULTI-FUZZY LOGIC IN LIA 135
If y = f(x), then A(x)→h B(y) = 1L, for every bridge function h and so we write
A(x) ⇒ B(y) = 1L.
Theorem 6.4.13. Let y ∈∏Lj, z ∈
∏Nk and x, x1, x2 ∈
∏Mi. If h1 :
∏Mi →∏
Nk and h2 :∏Lj →
∏Nk are lattice implication homomorphisms, then:
(1) x→h1 (y →h2 z) = y →h2 (x→h1 z);
(2) x→h1 y ≤ h1(h−11 (y′)→M x′);
(3) h1((y →h−11 x)→M x) ≤ (x→h1 y)→L y;
(4) x1 ∨ x2 →h1 y = (x1 →h1 y) ∧ (x2 →h1 y);
(5) x1 ∧ x2 →h1 y = (x1 →h1 y) ∨ (x2 →h1 y).
Proof. 1. For every x ∈∏Mi, y ∈
∏Lj and z ∈
∏Nk,
x→h1 (y →h2 z) =h1(x)→N (h2(y)→N z)
=h2(y)→N (h1(x)→N z)
= y →h2 (x→h1 z).
2. For every x ∈∏Mi and y ∈
∏Lj,
x→h1 y= h1(x)→L y
= y′ →L (h1(x))′
= y′ →L h1(x′)
≤h1(h−11 (y′))→L h1(x′),
6.4. MULTI-FUZZY LOGIC IN LIA 136
since h1(h−11 (y′)) ≤ y′. That is,
x→h1 y ≤ h1(h−11 (y′))→L h1(x
′) = h1(h−11 (y′)→M x′).
Equality holds, if h is surjective.
3. For every x ∈∏Mi and y ∈
∏Lj,
h1((y →h−11 x)→M x) = h1((h
−11 (y)→M x)→M x)
= h1(h−11 (y)→M x)→L h1(x)
= (h1(h−11 (y))→L h1(x))→L h1(x)
≤ (y →L h1(x))→L h1(x)
= (h1(x)→L y)→L y
= (x→h1 y)→L y.
Equality holds, if h is surjective.
4. For every x1, x2 ∈∏Mi and y ∈
∏Lj,
x1 ∨ x2 →h1 y=h1(x1 ∨ x2)→L y
= (h1(x1) ∨ h1(x2))→L y
= (h1(x1)→L y) ∧ (h1(x2)→L y)
= (x1 →h1 y) ∧ (x2 →h1 y).
6.4. MULTI-FUZZY LOGIC IN LIA 137
5. For every x1, x2 ∈∏Mi and y ∈
∏Lj,
x1 ∧ x2 →h1 y=h1(x1 ∧ x2)→L y
= (h1(x1) ∧ h1(x2))→L y
= (h1(x1)→L y) ∨ (h1(x2)→L y)
= (x1 →h1 y) ∨ (x2 →h1 y).
∞