Академический Документы
Профессиональный Документы
Культура Документы
www.elsevier.com/locate/ins
Abstract
*
Corresponding author. Tel.: +98 341 322 1079; fax: +98 341 322 1080.
E-mail address: mashinchi@mail.uk.ac.ir (M. Mashinchi).
0020-0255/$ - see front matter 2004 Elsevier Inc. All rights reserved.
doi:10.1016/j.ins.2004.08.007
104 Gh. Khaledi et al. / Information Sciences 174 (2005) 103–122
1. Introduction
In fuzzy set theory the concepts of t-norm and t-conorm play an important
role in generalizing the AND and OR aggregation operators [9]. Uninorms
were studied by Yager and Rybalov [10], among others, as a generalization
for t-norm and t-conorm. From a theoretical point of view [2], it is interesting
to notice that uninorms U with a neutral element in (0, 1) are just those binary
operators which make the structures ([0, 1], sup, U) and ([0, 1], inf, U) distribu-
tive semirings in the sense of Golan [5]. A uninorm U play an important role
in constructing e-implications and pseudo-e-implications and also in introduc-
ing special classes of e-implications and pseudo-e-implications where e is the
neutral element of U. Then many other researchers developed this notion
[1,2,4,6–8]. In this generalization a neutral element is introduced that can
be any number from the unit interval. The uninorm becomes a t-norm or a
t-conorm, if this neutral element is 1 or 0, respectively. Fodor et al. [4] studied
the structure of uninorms. The structure of the uninorm U with neutral ele-
ment e 2 (0, 1) on the squares [0, e]2 and [e, 1]2 is closely related to t-norms
and t-conorms. They have investigated the existence of representable uni-
norms. This representation is in terms of a single-variable function. In fact,
this is similar to representation of continuous Archimedean t-norm and t-con-
orm [9].
The algebraic structures of t-norm and t-conorm were studied by Ziaie et al.
[11]. Baczynski et al. [3] have introduced fuzzy implications as a special case of
fuzzy relations in [0, 1], and have shown that under some simple assumption the
sup–min composition of fuzzy implications also gives a fuzzy implication
which leads to a semigroup structure on fuzzy implications.
In this paper for any e in the real interval (0, 1], we introduce e-implication
and pseudo-e-implication. Then we will show that under appropriate opera-
tions on the set of all e-implications and on the set of all pseudo-e-implications
we will get monoid structures. Moreover we will show that the set of all
pseudo-e-implications is both complete lattice and De Morgan algebra. We
introduce a class of e-implications such that it becomes a De Morgan algebra.
Also contrapositive, self-dual pseudo-e-implication and contrapositive e-impli-
cation are disscussed.
2. Preliminaries
In this section we will review some known definitions and results which we
need throughout the paper. For more details see [4,10].
Remark. As explained in [4], Definition 2.2 is rather general one. The following
important particular cases are included. If e = 1, then Definition 2.2 gives back
the notion of t-norms, which are well-accepted models for AND in fuzzy logic.
If e = 0, then the class of t-conorms is obtained. These operations model logical
OR in fuzzy set theory. Therefore in the sequel we assume 0 < e < 1, where e is
neutral element of uninorm U, unless otherwise stated explicitly.
Let R ¼ R [ f1; 1g, denotes the extended real numbers.
Theorem 2.3 [4]. Consider e2(0, 1) and a strictly increasing continuous mapping
h : ½0; 1 ! R with h(e) = 0,h(0) = 1, h(1) = + 1. The binary operator U
defined by
2
U ðx; yÞ ¼ h1 ðhðxÞ þ hðyÞÞ8ðx; yÞ 2 ½0; 1 n fð0; 1Þ; ð1; 0Þg
if and only if
2
ð8ðx; yÞ 2 ½0; 1 n fð0; 1Þ; ð1; 0ÞgÞðU ðx; yÞ ¼ N ðU ðN ðxÞ; N ðyÞÞÞÞ:
Definition 3.1. Let e be a real number in the interval (0, 1]. The mapping
Ie : [0, 1]2 ! [0, 1] such that
Example 3.2. Let e be a real number in the interval (0, 1]. Then
MinfMaxfe x þ y; 0g; 1g ðx; yÞ 2 ð0; 1 ½0; 1Þ;
I e ðx; yÞ ¼
1 elsewhere
is an e-implication.
Remark. We can see that Ie in Example 3.2 is a Lukasiewicz implication [3] for
e = 1.
Proof
(i) We have y P 0. Since Ie is an e-implication, by Definition 3.1(iv)(2) we
have
I e ð0; yÞ P I e ð0; 0Þ ¼ 1 or I e ð0; yÞ ¼ 1:
(ii) Since x 6 1 and Ie is an e-implication, by Definition 3.1(iv)(1) we have
I e ðx; 1Þ P I e ð1; 1Þ ¼ 1 or I e ðx; 1Þ ¼ 1:
(iii) Since 0 < e 6 1, so by Definition 3.1 parts (iv) and (iii) we will have
I e ð1; 0Þ 6 I e ðe; 0Þ ¼ 0 or I e ð1; 0Þ ¼ 0:
Definition 3.5. Let e be a real number in the interval (0, 1]. The mapping
Ie : [0, 1]2 ! [0, 1] is called a pseudo-e-implication if it satisfies conditions (i)–(iii)
of Definition 3.1 and the following condition:
Example 3.6. Let e be a real number in the interval (0, 1]. Then
(
2
MinfMaxfe x þ y; 0g; 1g ðx; yÞ 2 ½0; 1 n fð0; 0Þ; ð1; 1Þg;
I e ðx; yÞ ¼
1 elsewhere
is a pseudo-e-implication.
Example 3.7. Let U be a representable uninorm with neutral element e 2 (0, 1).
Define
(
U ðN U ðxÞ; yÞ ðx; yÞ 2 ½0; 12 n fð0; 0Þ; ð1; 1Þg;
I e ðx; yÞ ¼
1 elsewhere:
Then by Theorems 2.4 and 2.5 Ie is a pseudo-e-implication.
Let e be a real number in the interval (0, 1]. The set of all e-implications and
the set of all pseudo-e-implications are denoted by Ie and PIe, respectively.
108 Gh. Khaledi et al. / Information Sciences 174 (2005) 103–122
where Min and Max are the usual minimum and maximum operations on real
numbers.
Lemma 3.8
(a) ^ and _ are closed operations on Ie(PIe)
(b) ^ and _ are associative on Ie(PIe).
Proof. (a) Due to similarity we only give the proof for Ie. First we prove that Ie
is closed under the operation ^. To this end we show that Ie ^ Je satisfies the
conditions of Definition 3.1.
Let e be a real number in the interval (0, 1]. Define the mappings 1e,0e,
ue, ze : [0, 1]2 ! [0, 1] as follows:
Lemma 3.9. Let e be a real number in the interval (0, 1]. Then
Proof. (a) Due to similarity we only prove that 1e 2 Ie. It is clear that 1e is well-
defined. To show that 1e 2 Ie, we prove the conditions of Definition 3.1 hold.
Theorem 3.11. Let e be a real number in the interval (0, 1]. Then (Ie, ^), (Ie, _),
(PIe, ^) and (PIe, _) are monoids.
In the sequel we let 6 be the usual ordering of the real functions on Ie(PIe).
That is
2
I e 6 J e () ð8ðx; yÞ 2 ½0; 1 ÞðI e ðx; yÞ
6 J e ðx; yÞÞ for all I e ; J e 2 IeðI e ; J e 2 PIeÞ:
Definition 3.12 [9]. A lattice is a partially ordered set (V, 6 ) in which every
pair of elements of V has a sup and an inf in V. Moreover a complete lattice is a
lattice, if every subset of V has both a sup and an inf in V.
Lemma 3.13. Let e be a real number in the interval (0, 1]. Then (Ie, _, ^, 6) and
(PIe, _, ^, 6) are complete lattices.
So I e 6 V 0e .
Now suppose that L0e 2 Ie is such that
ð8I e 2 AÞ ðI e 6 L0e Þ:
So, we have
2
ð8ðx; yÞ 2 ½0; 1 Þ ðI e ðx; yÞ 6 L0e ðx; yÞÞ:
Therefore
2
ð8ðx; yÞ 2 ½0; 1 Þ ðV 0e ðx; yÞ 6 L0e ðx; yÞÞ:
So V 0e 6 L0e .
Hence V 0e is the sup of A. Similarly we can show that A has inf. h
Let e be a real number in the interval (0, 1]. Assume u : [0, 1] ! [0, 1] be an
increasing bijection with fixed point e, and let Ie 2 Ie(Ie 2 PIe). Define Iue as
the following:
I ue : ½0; 12 ! ½0; 1;
I ue ðx; yÞ ¼ u1 ðI e ðuðxÞ; uðyÞÞÞ:
One can find many examples of such functions u such as the following:
Example 3.14. Let e be a real number in the interval (0, 1]. Define
u : [0, 1] ! [0, 1] as the following:
ð1=eÞx2 0 6 x < e;
uðxÞ ¼
x e 6 x 6 1:
Then u is an increasing bijection such that u(e) = e.
Lemma 3.15. Let e be a real number in the interval (0, 1] and u : [0, 1] ! [0, 1]
be an increasing bijection with fixed point e. Then Iue 2 Ie(Iue 2 PIe) for every
Ie 2 Ie(Ie 2 PIe).
Lemma 3.16. Let e be a real number in the interval (0, 1]. Then
Remark. Note that Definition 3.1 (Definition 3.5) could be more general by
requiring only the condition
(i) be changed to
(ii) 0 x 2 (0, 1) Ie(x, x) P e.
Then, with this definition, all the results proved in the paper until now remain
true. However, with this change one can easily check that a new important
family of implicators, characterized by Proposition 7 of [2], obtained by resid-
uation from any uninorm U with neutral element e2(0, 1) such that U(x,0) = 0
for all x 2 [0, 1) in the form of
I e ðx; yÞ ¼ supfz 2 ½0; 1jU ðx; zÞ 6 yg
satisfy the generalized Definition 3.1 (Definition 5.1). But in this paper we will
not adopt this generalization.
Let e be a real number in the interval (0, 1) and N : [0, 1] ! [0, 1] be a strong
negation with fixed point e and Ie 2 PIe(Ie 2 Ie), then we define I Ne as the
following:
Gh. Khaledi et al. / Information Sciences 174 (2005) 103–122 113
8 N 2
< I e : ½0; 1 ! ½0; 1
>
I Ne ðx; yÞ ¼ N fI e ðN ðxÞ; N ðyÞÞg; ðx; yÞ 2 ½0; 12 n fð0; 0Þ; ð1; 1Þg;
>
:
1; elsewhere:
Lemma 3.17. Let e be a real number in the interval (0, 1) and N : [0, 1] ! [0, 1]
be a strong negation with fixed point e and Ie 2 PIe. Then
(1) I Ne 2 PIe
(2) (a) ðI Ne ÞN ¼ I e
(b) ð8I e ; J e 2 PIe ðI e 6 J e ) J Ne 6 I Ne Þ.
(3) (De Morgan properties)
N
(i) ð8I e ; J e 2 PIe ððI e _ J e Þ ¼ I Ne ^ J Ne Þ,
N
(ii) ð8I e ; J e 2 PIe ððI e ^ J e Þ ¼ I Ne _ J Ne Þ,
(b) Let Ie, Je 2 PIe, Ie 6 Je and (x, y) 2 [0, 1]2. Consider the following two
cases:
(i) (x, y) 2 [0, 1]2n{(0, 0), (1, 1)}.
Since Ie 6 Je we have
I e ðN ðxÞ; N ðyÞÞ 6 J e ðN ðxÞ; N ðyÞÞ:
Hence,
N ðJ e ðN ðxÞ; N ðyÞÞÞ 6 N ðI e ðN ðxÞ; N ðyÞÞÞ:
So
J Ne ðx; yÞ 6 I Ne ðx; yÞ:
(ii) (x, y) 62 [0, 1]26{(0, 0),(1, 1)}.
We have
J Ne ðx; yÞ ¼ 1 6 1 ¼ I Ne ðx; yÞ:
Therefore, J Ne 6 I Ne .
(3) (De Morgan properties)
(i) Let Ie, Je 2 PIe and (x, y) 2 [0, 1]2. The following two cases occur
(a) (x, y) 2 [0, 1]2n{(0, 0), (1, 1)}.
We have
N
ðI e _ J e Þ ðx; yÞ ¼ N ððI e _ J e ÞðN ðxÞ; N ðyÞÞÞ
¼ N ðMaxfI e ðN ðxÞ; N ðyÞÞ; J e ðN ðxÞ; N ðyÞÞgÞ
¼ MinfN ðI e ðN ðxÞ; N ðyÞÞÞ; N ðJ e ðN ðxÞ; N ðyÞÞÞg
since N is strictly decreasing
¼ MinfI Ne ðx; yÞ; J Ne ðx; yÞg ¼ ðI Ne ^ J Ne Þðx; yÞ:
(b) (x, y) 62 [0, 1]2n{(0, 0), (1, 1)}.
We have (Ie _ Je)N(x, y) = 1.
On the other hand we have:
ðI Ne ^ J Ne Þðx; yÞ ¼ MinfI Ne ðx; yÞ; J Ne ðx; yÞg ¼ Minf1; 1g ¼ 1:
N
Therefore, ðI e _ J e Þ ¼ I Ne ^ J Ne .
(ii) The proof follows from Lemma 3.17 (2-a) and Lemma 3.17 (3-i). h
Proof
(i) Obviously, ze 6 uNe . From Lemma 3.17 (2) we have
ðuNe ÞN 6 zNe ) ue 6 zNe ) ue ¼ zNe :
(ii) Proof is similar to (i). h
Proof. The proof follows from Lemma 3.17 and Theorem 3.18. h
Example 3.21. Let U be a representable uninorm with neutral element e 2 (0, 1).
Ie is self-dual with respect to NU, where Ie(x, y) = U(NU(x),y);x, y 2 [0, 1].
Example 3.22. Let e = 0.5 and N(x) = 1x. Hence, N is a strong negation with
fixed point e. Consider Ie in Example 3.1. We know that I0.5 2 I0.5, but I N0:5 62 I 0:5 ,
since Definition 3.1(i)(iv) does not hold for t = 1, x2 = 1 and x1 = 0.75.
In the following we give a characterization for the case that Ie is closed un-
der negation operator I Ne .
Lemma 3.23. Let e be a real number in the interval (0, 1), N : [0, 1] ! [0, 1] be a
strong negation with fixed point e and Ie 2 Ie. Then I Ne 2 Ie if and only if
ð8x 2 ½0; 1ÞðI Ne ðx; 1Þ ¼ 1Þ:
Proof. It is clear that 1C 2 Ie. To complete the proof we show the following
statement:
ð8x 2 ½0; 1Þð1C ðx; 1Þ ¼ 1Þ:
If x = 1, the statement holds, since 1C 2 Ie. Let x 2 [0, 1) then we have
" #
1NC ðx; 1Þ ¼ N ½1C ðN ðxÞ; N ð1ÞÞ ¼ N sup I e ðN ðxÞ; N ð1ÞÞ
I e 2CN
e
Proof. Since the proof of the other parts are similar to the proof of Lemma
3.17, we only prove part (i-2). Since I e 2 CNe , so I Ne 2 Ie by Lemma 3.23. To
complete the proof we must prove the following statement:
N
ðx 2 ½0; 1ÞððI Ne Þ ðx; 1Þ ¼ 1Þ:
We have
N
ðI Ne Þ ðx; 1Þ ¼ I e ðx; 1Þby part ði 1Þ ¼ 1 by Lemma 3.4 ðiiÞ:
(i) 0NC ¼ 1C .
(ii) 1NC ¼ 0C .
118 Gh. Khaledi et al. / Information Sciences 174 (2005) 103–122
Proof. The proof follows from Lemma 3.26 and Theorem 3.27. h
(i) uk2U.
Moreover
(ii) I 2 UN U , where I is identity function.
(iii) uk 2 UN U , since
Lemma 3.30
(i) u 2 UN ) u1 2 UN.
(ii) u1, u2 2 UN)u1 u2 2 UN.
Proof. It is clear. h
Theorem 3.31. (UN, ) is a group.
Theorem 3.32. Let e be a real number in the interval (0, 1), N be a strong
negation with fixed point e and u 2 UN be a function with fixed point e. If Ie 2 PIe
is self-dual with respect to N, then Iue is self-dual with respect to N.
Gh. Khaledi et al. / Information Sciences 174 (2005) 103–122 119
Proof. Let (x, y) 2 [0, 1]2. We consider the following two cases:
Therefore I Nue ¼ I ue . h
Definition 3.33. Let e be a real number in the interval (0, 1] and N be a strong
negation. Ie 2 Ie (Ie 2 PIe) has the contrapositive property with respect to N if
2
ð8ðx; yÞ 2 ½0; 1 ÞðI e ðx; yÞ ¼ I e ðN ðyÞ; N ðxÞÞÞ:
Example 3.35. Let U be a representable uninorm with neutral element e2(0, 1).
Then Ie inPIe in the Example 3.7 has contrapositive property with respect to NU.
Theorem 3.36. Let e be a real number in the interval (0, 1] and N be a strong
negation. Suppose Ie 2 Ie (Ie 2 PIe) has the contrapositive property with respect
to N and u 2 UN be a function with fixed point e, then Iue has the contrapositive
property with respect to N.
Proof. We first prove that ^ is closed. To this end we show that Ie ^ Je has
contrapositive property with respect to NU if Ie and Je be in CPIe.
ðI e ^ J e Þðx; yÞ ¼ MinfI e ðx; yÞ; J e ðx; yÞg
¼ MinfI e ðN U ðyÞ; N U ðxÞÞ; J e ðN U ðyÞ; N U ðxÞÞg
ðsince I e and J e are in CPIe:Þ ¼ ðI e ^ J e ÞðN U ðyÞ; N U ðxÞÞ:
Similarly we can prove that _ is closed. h
Proof. It is clear that 1Ce and 0Ce are well-defined. To show that 1Ce 2 CPIe,
we prove the condition of Definition 3.33, holds.
1Ce ðx; yÞ ¼ sup I e ðx; yÞ ¼ sup I e ðN U ðyÞ; N U ðxÞÞ ðsince I e 2 CPIe:Þ
I e 2CPIe I e 2CPIe
Lemma 3.39. 1Ce and 0Ce are the neutral elements of operators ^ and _
respectively.
Lemma 3.40. (CPIe, ^) and (CPIe, _) are submonoids of (PIe, ^) and (PIe, _),
respectively.
Proof. Let (x, y)2[0, 1]2. Consider the following two cases:
Gh. Khaledi et al. / Information Sciences 174 (2005) 103–122 121
(ii) (x, y) 62 [0, 1]2n{(0, 0), (1, 1)}. Then we consider the following two sub cases:
Lemma 3.43
(i) 0NCeU ¼ 1Ce
(ii) 1NCeU ¼ 0Ce .
Proof. The proof follows from Lemma 3.17 (2–3), Lemmas 3.42 and 3.43. h
4. Conclusion
In this paper, we have studied the structure of the set of all e-implications
and the structure of the set of all pseudo-e-implications. We showed that these
sets are a complete lattice, where the set of all pseudo-e-implications becomes a
De Morgan algebra. Moreover it is shown that a class of e-implications is a De
Morgan algebra. Also it is shown that by considering a representable uninorm
with neutral element e 2 (0, 1), the set of all contrapositive pseudo-e-implica-
tions with respect to the involutive negator of the uninorm U is a complete sub-
lattice of the set of all pseudo-e-implications which is also a De Morgan
algebra.
Acknowledgments
The authors would like to thank the referees for their many constructive
comments which evolved the paper thoroughly and Mahani Mathematical
122 Gh. Khaledi et al. / Information Sciences 174 (2005) 103–122
References
[1] B. De Baets, Idempotent uninorms, European Journal of Operational Research 118 (1999)
631–642.
[2] B. De Baets, J. Fodor, Residual operators of uninorms, Soft Computing 3 (1999) 89–100.
[3] M. Baczynski, J. Drewniak, J. Sobera, Semigroups of fuzzy implications, Department of
Mathematics, Silesian University, Poland, Preprint.
[4] J. Fodor, R.R. Yager, A. Rybalov, Structure of uninorms, International Journal of
Uncertainity of Fuzziness Knowledge-Based Systems 5 (1997) 411–427.
[5] J. Golan, The theory of semirings with applications in mathematics and theoretical computer
science, Pitman Monographs and Surveys in Pure and Applied Mathematics, vol. 54,
Longman Scientific and Technical, 1992.
[6] Yong-Ming Li, Zhong-Ke Shi, Weak uninorm aggregation operators, Information Sciences
124 (2000) 317–323.
[7] M. Mas, G. Mayor, J. Torrens, The modularity condition for uninorms and t-operators,
Fuzzy Sets and Systems 126 (2002) 207–218.
[8] M. Mas, G. Mayor, J. Torrens, The distributivity condition for uninorms and t-operators,
Fuzzy Sets and Systems 128 (2002) 209–225.
[9] H.T. Nguyen, E.A. Walker, A First Course in Fuzzy Logic, second ed., Chapman and Hall,
London, 2000.
[10] R.R. Yager, A. Rybalov, Uninorm aggregation operators, Fuzzy Sets and Systems 80 (1996)
111–120.
[11] S.A. Ziaie, S. Ray, M. Mashinchi, R. Kamran, Boolean structure of triangular norms,
Mathware and Soft Computing 6 (1999) 63–78.