on the characterizations of fuzzy implications satisfying
TRANSCRIPT
![Page 1: On the characterizations of fuzzy implications satisfying](https://reader036.vdocument.in/reader036/viewer/2022082617/57501f631a28ab877e95786b/html5/thumbnails/1.jpg)
Information Sciences 177 (2007) 2954–2970
www.elsevier.com/locate/ins
On the characterizations of fuzzy implicationssatisfying Iðx; yÞ ¼ Iðx; Iðx; yÞÞ
Y. Shi a,*, D. Ruan a,b, E.E. Kerre a
a Fuzziness and Uncertainty Modelling Research Unit, Department of Applied Mathematics and Computer Science, Ghent University,
Krijgslaan 281 (S9), 9000 Gent, Belgiumb Belgium Nuclear Research Centre (SCK•CEN), 2400 Mol, Belgium
Received 27 October 2005; received in revised form 20 December 2006; accepted 26 January 2007
Abstract
Iterative boolean-like laws in fuzzy logic have been studied by Alsina and Trillas [C. Alsina, E. Trillas, On iterativeboolean-like laws of fuzzy sets, in: Proc. 4th Conf. Fuzzy Logic and Technology, Barcelona, Spain, 2005, pp. 389–394]for functional equations with boolean background in which only fuzzy conjunctions, fuzzy disjunctions and fuzzy nega-tions are contained. In this paper we study an iterative boolean-like law with fuzzy implications, more precisely we derivecharacterizations of some classes of fuzzy implications satisfying Iðx; yÞ ¼ Iðx; Iðx; yÞÞ, for all ðx; yÞ 2 ½0; 1�2. Our discussionmainly focuses on the three important classes of implications: S-implications, R-implications and QL-implications. Weprove the sufficient and necessary conditions for an S-implication generated by any t-conorm and any fuzzy negation,an R-implication generated by a left-continuous t-norm, a QL-implication generated by a continuous t-conorm, a contin-uous t-norm and a strong fuzzy negation to satisfy Iðx; yÞ ¼ Iðx; Iðx; yÞÞ, for all ðx; yÞ 2 ½0; 1�2.� 2007 Elsevier Inc. All rights reserved.
Keywords: Iterative boolean-like law; Fuzzy implications; S-implications; R-implications; QL-implications
1. Introduction
In [14, Section 4], the authors analyzed some non-standard aspects in the constructions of fuzzy set theoriesand dealt with the derived boolean properties of fuzzy operations, among which the iterative boolean-like laws
[1] were considered as derived boolean laws not valid in any standard fuzzy set theories. In [1], the authorsstudied a class of functional equations [1, Definition 4.1] with the boolean background [1, Definition 4.2], namedas iterative boolean-like laws which are formulated in fuzzy logic where some variables appear several timesbecause they come from boolean identities where no simplifications such as the application of idempotency ordistributivity, absorption, etc. have been made. In [1, Section 5], the authors analyzed some standard iterativeboolean-like laws in fuzzy logic such as ðA [ AÞ \ ðA \ AÞco ¼ ;, A [ B ¼ ðA \ BÞ [ ½ðA [ BÞ \ ðA \ BÞco�,
0020-0255/$ - see front matter � 2007 Elsevier Inc. All rights reserved.
doi:10.1016/j.ins.2007.01.026
* Corresponding author. Tel.: +32 9 264 47 72; fax: +32 9 264 49 95.E-mail address: [email protected] (Y. Shi).
![Page 2: On the characterizations of fuzzy implications satisfying](https://reader036.vdocument.in/reader036/viewer/2022082617/57501f631a28ab877e95786b/html5/thumbnails/2.jpg)
Y. Shi et al. / Information Sciences 177 (2007) 2954–2970 2955
ðA [ B [ BÞ [ ðA \ B \ BÞ ¼ A [ B, and solve the functional equations derived from them. But only laws withfuzzy conjunctions, fuzzy disjunctions and fuzzy negations were considered. Since fuzzy implications are alsofuzzy operations that play an important role in fuzzy logic, we will analyze some iterative boolean-like lawswith fuzzy implications.
In this paper we will consider the derived boolean law p ! ðp! qÞ ¼ p! q. In classical logic, the impli-cation operator ‘!’ used to gain the value of the proposition ‘if p, then q’ (p ! q) is defined through ‘:p _ q’(in which �denotes the negation and _denotes the disjunction) uniquely and the proposition ‘p ! ðp ! qÞ’ isequivalent to ‘p ! q’. However, this equivalence does not hold for every fuzzy implication derived from clas-sical logic, which is generated by t-norms, t-conorms and fuzzy negations. Our main concern of this paper is tofind out the conditions under which the following iterative boolean-like law in fuzzy logic holds:
Iðx; yÞ ¼ Iðx; Iðx; yÞÞ; 8x; y 2 ½0; 1�2; ð1Þ
where I denotes a fuzzy implication from the classes of S-, R- or QL-implications.In Section 2 we give some basic notations and definitions of fuzzy conjunctions, fuzzy disjunctions and
fuzzy negations. In Section 3 we focus on some general properties of fuzzy implications and list a table of somewidely used implications from the three classes to see the properties including (1) they satisfy. In Section 4 wework out the sufficient and necessary conditions for the three classes of fuzzy implications to satisfy (1).Finally we conclude our main results in Section 5.
2. Preliminaries
2.1. Fuzzy negations
In fuzzy logic, a mapping n: ½0; 1� ! ½0; 1� is a fuzzy negation if it satisfies:
n1. boundary conditions: nð0Þ ¼ 1 and nð1Þ ¼ 0,n2. monotonicity: x 6 y ) nðxÞP nðyÞ.
For any continuous fuzzy negation n, there exists a unique equilibrium point e 2�0; 1½ such that nðeÞ ¼ e andfor all x < e, nðxÞ > e > x, for all x > e, nðxÞ < e < x [4, Section 1.1]. A fuzzy negation is said to be strict if n isa continuous and strictly decreasing mapping. A fuzzy negation is said to be involutive if nðnðxÞÞ ¼ x, for allx 2 ½0; 1�. Fuzzy negations that are involutive are called strong fuzzy negations. A strong fuzzy negation isstrict and it is a continuous mapping.
Next we introduce concepts of automorphism and conjugate, which will be useful in this paper.An automorphism of the interval ½a; b� � R is a continuous, strictly increasing mapping u from ½a; b� to
½a; b� with boundary conditions uðaÞ ¼ a, uðbÞ ¼ b [4, Definition 0].Two mappings F ;G : ½0; 1�2 ! ½0; 1� are conjugate, if there exists an automorphism u of the unit interval
such that G ¼ F u, where F uðx; yÞ ¼ u�1ðF ðuðxÞ;uðyÞÞÞ, x; y 2 ½0; 1� [2, Definition 2].
Proposition 1 [18]. A fuzzy negation n is strong iff there exists an automorphism u of the unit interval such that
nðxÞ ¼ u�1ð1� uðxÞÞ 8x 2 ½0; 1�:
2.2. Triangular norms
The conjunction in fuzzy logic is often represented by a triangular norm (t-norm for short). A mappingT : ½0; 1�2 ! ½0; 1� is a t-norm if for all x; y; z 2 ½0; 1� it satisfies:
T1. boundary condition: T ðx; 1Þ ¼ x,T2. monotonicity: y 6 z implies T ðx; yÞ 6 T ðx; zÞ,T3. commutativity: T ðx; yÞ ¼ T ðy; xÞ,T4. associativity: T ðx; T ðy; zÞÞ ¼ T ðT ðx; yÞ; zÞ.
![Page 3: On the characterizations of fuzzy implications satisfying](https://reader036.vdocument.in/reader036/viewer/2022082617/57501f631a28ab877e95786b/html5/thumbnails/3.jpg)
2956 Y. Shi et al. / Information Sciences 177 (2007) 2954–2970
There are also some additional requirements of t-norms. Here we mention seven of them:
T5. continuity: T is continuous,T6. left-continuity: the partial mappings of T are left-continuous,T7. idempotency: T ðx; xÞ ¼ x, for all x 2 ½0; 1�,T8. subidempotency: T ðx; xÞ < x, for all x 2�0; 1½ ,
T9. Archimedean property: for all ðx; yÞ 2 �0; 1½2, there exists n 2 N such that T ðx; x; . . . ; x|fflfflfflfflfflffl{zfflfflfflfflfflffl}n times
Þ < y [10, Defini-tion 2.9 (iv)],
T10. strictness: T is continuous and 0 < y < z < 1) T ðx; yÞ < T ðx; zÞ, for all x; y; z 2�0; 1�,T11. nilpotency: T is continuous and for all x 2�0; 1½ , there exists y 2�0; 1½ such that T ðx; yÞ ¼ 0.
Remark 1. According to [10, Definition 2.1 (i)], an element x 2 ½0; 1� such that T ðx; xÞ ¼ x is called anidempotent element of T. The numbers 0 and 1, which are idempotent elements for each t-norm T, are calledtrivial idempotent elements of T. Each idempotent element in �0; 1½ will be called a non-trivial idempotentelement of T. Thus T satisfying subidempotency is equivalent to T having only trivial idempotent elements.Hence according to [10, Fig. 2.2], the subidempotency is equivalent to the Archimedean property only forcontinuous t-norms.
There are four very important t-norms [10, Example 1.2]:
(1) T Mðx; yÞ ¼ minðx; yÞ, (minimum)(2) T P ðx; yÞ ¼ xy, (product)(3) T –Lðx; yÞ ¼ maxðxþ y � 1; 0Þ, (Łukasiewicz t-norm)
(4) T Dðx; yÞ ¼minðx; yÞ x ¼ 1 or y ¼ 10 otherwise:
�(drastic product)
The standard fuzzy conjunction TM is the only idempotent t-norm [11, Theorem 3.9].
Theorem 1 [11, Theorem 3.11]. T is a continuous Archimedean t-norm iff there exists a decreasing generator f
such that
T ðx; yÞ ¼ f ð�1Þðf ðxÞ þ f ðyÞÞ; 8x; y 2 ½0; 1�: ð2Þ
In this theorem, a decreasing generator f is defined as a continuous and strictly decreasing mapping from½0; 1� to ½0;1� such that f ð1Þ ¼ 0. The pseudo-inverse of f, f ð�1Þ: ½0;1� ! ½0; 1� is defined as:
f ð�1ÞðxÞ ¼ f �1ðxÞ x 2 ½0; f ð0Þ�;0 x 2�f ð0Þ;1�;
�ð3Þ
where f�1 denotes the ordinary inverse of f.It has been proved that a continuous Archimedean t-norm is either strict or nilpotent [10, p. 33].
Theorem 2 [10, Corollary 5.7]
(i) A t-norm T is strict iff it is conjugate with the product TP.
(ii) A t-norm T is nilpotent iff it is conjugate with the Łukasiewicz t-norm T –L.
Theorem 3 [4, Theorem 2]. A continuous t-norm T is such that T ðx; nðxÞÞ ¼ 0 holds for all x 2 ½0; 1� with a strict
fuzzy negation n iff T is conjugate with the Łukasiewicz t-norm T Ł, i.e., there exists an automorphism / of the unit
interval such that
T ðx; yÞ ¼ /�1ðmaxð/ðxÞ þ /ðyÞ � 1; 0ÞÞ
andnðxÞ 6 /�1ð1� /ðxÞÞ:
![Page 4: On the characterizations of fuzzy implications satisfying](https://reader036.vdocument.in/reader036/viewer/2022082617/57501f631a28ab877e95786b/html5/thumbnails/4.jpg)
Y. Shi et al. / Information Sciences 177 (2007) 2954–2970 2957
Continuous t-norms have been well studied. We have the next definition and theorem. Suppose thatf½am; bm�g is a countable family of non-overlapping, closed, proper subintervals of ½0; 1�, denoted by I andwith each ½am; bm� 2 I associate a continuous Archimedean t-norm Tm. Let T be a mapping defined on½0; 1�2 via
T ðx; yÞ ¼ am þ ðbm � amÞT mx�am
bm�am; y�am
bm�am
� �ðx; yÞ 2 ½am; bm�2;
minðx; yÞ otherwise:
(ð4Þ
Then T is called the ordinal sum of f½am; bm�; T mg and each Tm is called a summand.
Theorem 4 [8, Section 1.3.4], [10, Theorem 5.11]. T is a continuous t-norm iff
(i) T ¼ T M or
(ii) T is continuous Archimedean or
(iii) there exists a family f½am; bm�; T mg where f½am; bm�g is a countable family of non-overlapping, closed, proper
subintervals of ½0; 1� with each Tm being a continuous Archimedean t-norm such that T is the ordinal sum of
this family.
2.3. Triangular conorms
The disjunction in fuzzy logic is often represented by a triangular conorm (t-conorm for short). A mappingS : ½0; 1�2 ! ½0; 1� is a t-conorm if for all x; y; z 2 ½0; 1� it satisfies:
S1. boundary condition: Sðx; 0Þ ¼ x,S2. monotonicity: y 6 z implies Sðx; yÞ 6 Sðx; zÞ,S3. commutativity: Sðx; yÞ ¼ Sðy; xÞ,S4. associativity: Sðx; Sðy; zÞÞ ¼ SðSðx; yÞ; zÞ. Similar to t-norms, we mention here six other properties of t-
conorms:S5. continuity: S is continuous,S6. idempotency: Sðx; xÞ ¼ x, for all x 2 ½0; 1�,S7. subidempotency: Sðx; xÞ < x, for all x 2�0; 1½ ,S8. Archimedean property: for all ðx; yÞ 2 �0; 1½ 2, there exists n 2 N such that Sðx; x; . . . ; x|fflfflfflfflfflffl{zfflfflfflfflfflffl}
n times
Þ > y [10, Remark2.20 (AP*)],
S9. strictness: S is continuous and 0 < y < z < 1) Sðx; yÞ < Sðx; zÞ, for all x; y; z 2 ½0; 1½ ,S10. nilpotency: S is continuous and for all x 2�0; 1½ , there exists y 2�0; 1½ such that Sðx; yÞ ¼ 1.
Remark 2. Similar to that for t-norms, the subidempotency is equivalent to the Archimedean property onlyfor continuous t-conorms. [10, Remark 2.20, Fig. 2.2].
There are four very important t-conorms [10, Example 1.14]:
(1) SMðx; yÞ ¼ maxðx; yÞ, (maximum)(2) SP ðx; yÞ ¼ xþ y � xy, (probabilistic sum)(3) S–Lðx; yÞ ¼ minðxþ y; 1Þ, (Łukasiewicz t-conorm, bounded sum)
(4) SDðx; yÞmaxðx; yÞ x ¼ 0 or y ¼ 01 otherwise:
�(drastic sum)
The standard fuzzy disjunction SM is the only idempotent t-conorm [11, Theorem 3.14].
Theorem 5 [11, Theorem 3.16]. S is a continuous Archimedean t-conorm iff there exists an increasing generator
g such that
Sðx; yÞ ¼ gð�1ÞðgðxÞ þ gðyÞÞ; 8x; y 2 ½0; 1�: ð5Þ
![Page 5: On the characterizations of fuzzy implications satisfying](https://reader036.vdocument.in/reader036/viewer/2022082617/57501f631a28ab877e95786b/html5/thumbnails/5.jpg)
2958 Y. Shi et al. / Information Sciences 177 (2007) 2954–2970
In this theorem, an increasing generator g is defined as a continuous and strictly increasing mapping from½0; 1� to ½0;1� such that gð0Þ ¼ 0. The pseudo-inverse of g, gð�1Þ : ½0;1� ! ½0; 1� is defined as:
gð�1ÞðxÞ ¼ g�1ðxÞ x 2 ½0; gð1Þ�;1 x 2�gð1Þ;1�;
�ð6Þ
where g�1 denotes the ordinary inverse of g.Similar to t-norms, it has been concluded that a continuous Archimedean t-conorm is either strict or nil-
potent [10, Remark 2.20].
Theorem 6 [8, Theorem 1.8, Theorem 1.9]
(i) A t-conorm S is strict iff it is conjugate with the probabilistic sum SP.(ii) A t-conorm S is nilpotent iff it is conjugate with the Łukasiewicz t-conorm S–L.
Theorem 7 [4, Theorem 1]. A continuous t-conorm S is such that Sðx; nðxÞÞ ¼ 1 holds for all x 2 ½0; 1� with a
strict fuzzy negation n iff S is conjugate with the Łukasiewicz t-conorm S–L, i.e., there exists an automorphism
/ of the unit interval such that
Sðx; yÞ ¼ /�1ðminð/ðxÞ þ /ðyÞ; 1ÞÞ
andnðxÞP /�1ð1� /ðxÞÞ:
And similar to continuous t-norms, we have the next definition and theorem for continuous t-conorms.Suppose that f½am; bm�g is a countable family of non-overlapping, closed, proper subintervals of ½0; 1�, denotedby I. With each ½am; bm� 2 I associated a continuous Archimedean t-conorm Sm. Let S be a mapping definedon ½0; 1�2 via
Sðx; yÞ ¼ am þ ðbm � amÞSmx�am
bm�am; y�am
bm�am
� �ðx; yÞ 2 ½am; bm�2;
maxðx; yÞ otherwise:
(ð7Þ
S is called the ordinal sum of f½am; bm�; Smg and each Sm is called a summand.
Theorem 8 [8, Section 1.4.4], [10, Corollary 3.58]. S is a continuous t-norm iff
(i) S ¼ SM or
(ii) S is continuous Archimedean or(iii) there exists a family f½am; bm�; Smg where f½am; bm�g is a countable family of non-overlapping, closed, proper
subintervals of ½0; 1� with each Sm being a continuous Archimedean t-conorm such that S is the ordinal sum
of this family.
If for all x; y; z 2 ½0; 1�, Sðx; T ðy; zÞÞ ¼ T ðSðx; yÞ; Sðx; zÞÞ, we say the t-conorm S is distributive over the t-normT [10, Proposition 2.22]. Similarly, if for all x; y; z 2 ½0; 1�, T ðx; Sðy; zÞÞ ¼ SðT ðx; yÞ; T ðx; zÞÞ, we say the t-norm T
is distributive over the t-conorm S. The only distributive pair is TM and SM [10, Proposition 2.22].
3. Fuzzy implications
3.1. Definitions and properties
Fuzzy implications have been widely studied in the fuzzy literature and they play important roles in differ-ent domains. In Zadeh’s composition rule of fuzzy inference (fuzzy modus ponens, fuzzy modus tollens, etc.),a fuzzy implication represents the fuzzy relation between two variables [9, Chapter 5,12,13,16,20]. In imageprocessing, a fuzzy implication is used to define the subsethood measure between two fuzzy sets [5, Theorem1]. And in data mining, a fuzzy implication is used to express the relationship between two items in an asso-ciation rule [22, Eq. (2)]. Also the set of axioms of fuzzy implications has been well studied [2,17].
![Page 6: On the characterizations of fuzzy implications satisfying](https://reader036.vdocument.in/reader036/viewer/2022082617/57501f631a28ab877e95786b/html5/thumbnails/6.jpg)
Y. Shi et al. / Information Sciences 177 (2007) 2954–2970 2959
Originally, a fuzzy implication is a fuzzy operator which is used to represent IF–THEN rules in fuzzy logic.The truth value of the proposition ‘if p, then q’ could be obtained through IðvðpÞ; vðqÞÞ, in which I refers to afuzzy implication and vðpÞ and vðqÞ refer to the truth value of p and q, respectively.
In contrast to t-norms and t-conorms, there exist several definitions of fuzzy implications [4, Section 2], [8,Definition 1.15]. We define a fuzzy implication as a ½0; 1�2 ! ½0; 1� mapping at least satisfying the boundaryconditions:
I1. Ið0; 0Þ ¼ Ið0; 1Þ ¼ Ið1; 1Þ ¼ 1, Ið1; 0Þ ¼ 0.
Hence a fuzzy implication extends the classical implication.Many other potential properties could be found in [2,4,7,9,19,21]. The most important properties of fuzzy
implications are [4,8,15,19]:
I2. first place antitonicity: x 6 y ) Iðx; zÞP Iðy; zÞ, for all x; y; z 2 ½0; 1�,I3. second place isotonicity: y 6 z) Iðx; yÞ 6 Iðx; zÞ, for all x; y; z 2 ½0; 1�,I4. left neutrality of truth: Ið1; xÞ ¼ x, for all x 2 ½0; 1�,I5. exchange principle: Iðx; Iðy; zÞÞ ¼ Iðy; Iðx; zÞÞ, for all x; y; z 2 ½0; 1�,I6. ordering property: Iðx; yÞ ¼ 1 iff x 6 y, for all x; y 2 ½0; 1�,I7. contrapositive law: Iðx; yÞ ¼ IðnðyÞ; nðxÞÞ, for all x; y 2 ½0; 1�, where n denotes a strong fuzzy negation,I8. continuity.
These properties are not independent from each other, it was proved in [2, Section 4] that a fuzzy implica-tion that satisfies I5, I6 and I8 also satisfies the other five properties I1, I2, I3, I4 and I7.
Some implications do not satisfy the property I8, but they are right continuous mappings in the second argu-
ment [2, Section 3].
3.2. Implications generated by t-norms, t-conorms and fuzzy negations
It was mentioned in [4, Section 2] that all fuzzy implications are obtained by generalizing the implications ofclassical logic. There are three important ways to generate fuzzy implications through classical logic: S-impli-
cations, R-implications and QL-implications.Let S be a t-conorm and n be a fuzzy negation. An S-implication is defined by:
Iðx; yÞ ¼ SðnðxÞ; yÞ; 8x; y 2 ½0; 1�: ð8Þ
It should be mentioned here that in some references, n is assumed to be a strong fuzzy negation [4, Section 4],[8, Definition 1.16]. But in other references, n is not necessary supposed to be strong, even not necessary to becontinuous [10, Definition 11.5], [3] (where the authors call the S-implications generated by a fuzzy negation Nand a t-conorm S (S,N)-implications). In this paper, we assume n to be any fuzzy negation, so as [10, Defini-tion 11.5].
Let T be a t-norm. An R-implication is defined by:
Iðx; yÞ ¼ supft 2 ½0; 1�jT ðx; tÞ 6 yg; 8x; y 2 ½0; 1�: ð9Þ
The definition of R-implication comes from a residuation concept in intuitionistic logic. Actually, this defini-tion is only reasonable for left-continuous t-norms [2, Section 3]. And when T is a left-continuous t-norm, (9)is equivalent to [2, Section 3]:Iðx; yÞ ¼ maxft 2 ½0; 1�jT ðx; tÞ 6 yg; 8x; y 2 ½0; 1�: ð10Þ
Let I be an R-implication generated by a left-continuous t-norm via (9) (or (10)), then a t-norm T can be gen-erated by I through:
T ðx; yÞ ¼ infft 2 ½0; 1�jIðx; tÞP yg; 8x; y 2 ½0; 1�: ð11Þ
The following important theorem and its corollary will be useful in our further discussion.![Page 7: On the characterizations of fuzzy implications satisfying](https://reader036.vdocument.in/reader036/viewer/2022082617/57501f631a28ab877e95786b/html5/thumbnails/7.jpg)
2960 Y. Shi et al. / Information Sciences 177 (2007) 2954–2970
Theorem 9 [8, Theorem 1.14]. A mapping I : ½0; 1�2 ! ½0; 1� is an R-implication based on some left-continuous t-
norm iff I satisfies I3, I5, I6 and it is right continuous in the second argument.
If we denote the R-implication generated by a fixed left-continuous t-norm T through (9) (or (10)) as IT andthe t-norm generated by an R-implication I, which satisfies I3, I5, I6 and right-continuity in the second argu-ment through (11) as TI, then we will find in the proof of Theorem 1.14 in [8] that IT I ¼ I and T IT ¼ T . Thefollowing corollary is based on these facts.
Corollary 1 [2, Corollary 10]. A mapping T : ½0; 1�2 ! ½0; 1� is a left-continuous t-norm iff T can be represented
by:
TableFuzzy
Name
Kleene
Reiche
Most S
Least
Łukas
R0
Godel
Gogue
Early
Klir an
Klir anYua
T ðx; yÞ ¼ minft 2 ½0; 1�jIðx; tÞP yg ð12Þ
for some mapping I : ½0; 1�2 ! ½0; 1� which satisfies I3, I5, I6 and right-continuity in the second argument.Let S be a t-conorm, n be a strong fuzzy negation and T be a t-norm. A QL-implication is defined by:
Iðx; yÞ ¼ SðnðxÞ; T ðx; yÞÞ; 8x; y 2 ½0; 1�: ð13Þ
Some authors assume n to be any fuzzy negation, not necessary strong, e.g., [6, Eq. (1.1)]. But in this paper weassume n to be strong.
3.3. An overview of some fuzzy implications and their properties
In this paper, we will discuss fuzzy implications of the three important classes that satisfy Eq. (1). In Table 1we list several widely used fuzzy implications from the three classes and in the last column we indicate theirproperties. Actually an S-implication satisfies I1, I2, I3, I4 and I5 [3, Proposition 1]. An R-implication basedon a left-continuous t-norm satisfies I1, I2, I3, I4 I5 and I6 [8, Lemma 1.3, Theorem 1.14]. A QL-implicationsatisfy I1, I3 and I4 [4, Proposition 2(i)].
1implications and their properties
Symbol Iðx; yÞ Class Properties
-Dienes Ib maxð1� x; yÞ S, QL I1, I2, I3I4, I5, I7, I8, (1)
nbach Ir 1� xþ xy S, QL I1, I2, I3I4, I5, I7, I8
trict IM1 x ¼ 0y otherwise
�S I1, I2, I3
I4, I5, (1)
ILS
y x ¼ 11� x y ¼ 01 otherwise
8<: S I1, I2, I3
I4, I5, I7
Strict ILRy x ¼ 11 otherwise
�S, R, QL I1, I2, I3
I4, I5, I6, (1)iewicz Ia minð1� xþ y; 1Þ S, R, QL I1, I2, I3
I4, I5, I6, I7, I8Imin0
1 x 6 ymaxð1� x; yÞ otherwise
�S, R I1, I2, I3
I4, I5, I6, I7Ig
1 x 6 yy otherwise
�R I1, I2, I3
I4, I5, I6, (1)n ID
1 x 6 yy=x otherwise
�R I1, I2, I3
I4, I5, I6Zadeh Im maxð1� x; minðx; yÞÞ QL I1, I3, I4
I8, (1)d Yuan 1 Ip 1� xþ x2y QL I1, I3, I4, I8
d Iq
y x ¼ 11� x x 6¼ 1; y 6¼ 11 x 6¼ 1; y ¼ 1
8<: QL I1, I3, I4, (1)
n 2
![Page 8: On the characterizations of fuzzy implications satisfying](https://reader036.vdocument.in/reader036/viewer/2022082617/57501f631a28ab877e95786b/html5/thumbnails/8.jpg)
Y. Shi et al. / Information Sciences 177 (2007) 2954–2970 2961
4. Solutions of (1) in S-, R- and QL-implications
Now we analyze under which conditions the three classes of fuzzy implications satisfy (1).
4.1. S-implications
Theorem 10. An S-implication I generated by a t-conorm S and a fuzzy negation n satisfies (1) iff the range of n is
a subset of the idempotent elements of S.
Proof. (: According to (8), Iðx; yÞ ¼ SðnðxÞ; yÞ and by S4,
Iðx; Iðx; yÞÞ ¼ SðnðxÞ; SðnðxÞ; yÞÞ ¼ SðSðnðxÞ; nðxÞÞ; yÞ:
If the range of n is a subset of the idempotent elements of S, then for all x 2 ½0; 1�, SðnðxÞ; nðxÞÞ ¼ nðxÞ andhence I satisfies (1).): By S1, for all x 2 ½0; 1�, Iðx; 0Þ ¼ SðnðxÞ; 0Þ ¼ nðxÞ andIðx; Iðx; 0ÞÞ ¼ SðnðxÞ; SðnðxÞ; 0ÞÞ ¼ SðnðxÞ; nðxÞÞ:
If I satisfies (1), then Iðx; 0Þ ¼ Iðx; Iðx; 0ÞÞ, and hence nðxÞ ¼ SðnðxÞ; nðxÞÞ, for all x 2 ½0; 1�. So for ally 2 rngðnÞ, Sðy; yÞ ¼ y. hThe next corollary is a strong result for the condition that n refers to a continuous fuzzy negation in theabove theorem.
Corollary 2. An S-implication I generated by a t-conorm S and a continuous fuzzy negation n satisfies (1) iff
S ¼ SM .
Proof. (: Straightforward.): If n is continuous, then the range of n is ½0; 1�. According to Theorem 10, I satisfies (1) iff the subset of
the idempotent elements of S is ½0; 1� i.e., S is an idempotent t-conorm. Since max is the only idempotent t-conorm, S ¼ SM . h
Example 1. Consider the Godel fuzzy negation NGðxÞ ¼1 x ¼ 00 otherwise
�.
Because the range of n is f0; 1g and for every t-conorm, 0 and 1 are idempotent elements, when the S-implication I is generated by the Godel fuzzy negation and any t-conorm, I ¼ IM and (1) holds.
Example 2. The dual t-conorm of the well-known left-continuous t-norm nilpotent minimum [7, Section 3] with
the strong fuzzy negation nðxÞ ¼ 1� x is Sðx; yÞ ¼ 1 xþ y P 1maxðx; yÞ otherwise
�.
The set of its idempotent elements is ½0; 0:5½[f1g. Consider the fuzzy negation8
nðxÞ ¼1 x ¼ 0;
f ðxÞ x 2�0; 0:5�;0 x 2�0:5; 1�;
><>:
where f denotes a strictly decreasing mapping satisfying f ð0Þ ¼ 0:5, f ð0:5Þ ¼ 0, for all x 2 ½0; 0:5�. The range ofn is equal to the set of the idempotent elements of S. So the S-implication generated by S and n satisfies (1).
4.2. R-implications
Theorem 11. A mapping I : ½0; 1�2 ! ½0; 1� satisfies I3, I5, I6, right-continuity in the second argument and (1) iff Iis the Godel implication i.e.,
Iðx; yÞ ¼ Igðx; yÞ ¼1 x 6 y;
y otherwise:
�
![Page 9: On the characterizations of fuzzy implications satisfying](https://reader036.vdocument.in/reader036/viewer/2022082617/57501f631a28ab877e95786b/html5/thumbnails/9.jpg)
2962 Y. Shi et al. / Information Sciences 177 (2007) 2954–2970
Proof. (: It is easy to see that the Godel implication Ig satisfies I3, I5, I6, right-continuity in the second argu-ment and (1).
): Since I satisfies I3, I5, I6 and right-continuity in the second argument, according to Corollary 1, a left-continuous t-norm T can be generated through:
T ðx; yÞ ¼ minft 2 ½0; 1�jIðx; tÞP yg; 8ðx; yÞ 2 ½0; 1�2:
And if we denote IT ðx; yÞ ¼ supft 2 ½0; 1�jT ðx; tÞ 6 yg by (9), then according to the proof of Theorem 1.14 in[8], IT ¼ I .
We will first show that T ðx; xÞ ¼ x, for all x 2 ½0; 1�.Indeed, from the formula (12), we have for all x 2 ½0; 1�, T ðx; xÞ ¼ minft 2 ½0; 1�jIðx; tÞP xg. Obviously,
when t 2 ½x; 1�, Iðx; tÞ ¼ 1 P x holds by I6. Assume that there exists t0 2 ½0; x½ such that Iðx; t0ÞP x. Then byI6, Iðx; Iðx; t0ÞÞ ¼ 1. Since we have for all ðx; yÞ 2 ½0; 1�2, Iðx; yÞ ¼ Iðx; Iðx; yÞÞ, Iðx; t0Þ ¼ Iðx; Iðx; t0ÞÞ ¼ 1. Andby I6, x 6 t0. This is a contradiction with our assumption that t0 < x. So if t0 2 ft 2 ½0; 1�jIðx; tÞP xg, thent0 P x i.e.,
T ðx; xÞ ¼ minft 2 ½0; 1�jIðx; tÞP xg ¼ x; 8x 2 ½0; 1�:
That is to say, T ¼ T M . Hence,
Iðx; yÞ ¼ supft 2 ½0; 1�jminðx; tÞ 6 yg ¼1 x 6 y
y otherwise
�¼ Igðx; yÞ: �
Corollary 3. An R-implication I generated by a left-continuous t-norm T satisfies (1) iff T ¼ T M .
Proof. (: The R-implication generated by T ¼ T M through (9) is the Godel implication Ig and hence it sat-isfies (1).): According to Theorem 9, an R-implication I satisfying I3, I5, I6 and right-continuity in the second
argument is generated by a left-continuous t-norm T through (9). From Theorem 11, if I satisfies (1)additionally, it is the Godel implication Ig. Denote T Ig as the t-norm generated by Ig via (11) and IT Ig
as the R-implication generated by T Ig via (9), then according to Theorem 9 and its corollary, Ig ¼ IT Ig
i.e., Ig isgenerated by T Ig through (9) where T Ig is the t-norm generated by Ig through (11), that is T Ig ¼ T M . h
Remark 3. A continuous t-norm is of course left-continuous. So Theorem 11 and its corollary are strongresults for R-implications generated by a left-continuous t-norm. They are also proper for those R-implica-tions generated by a continuous t-norm.
Remark 4. In [2], Theorem 1, which was first proposed by Smets and Magrez [17], shows that a mappingI : ½0; 1�2 ! ½0; 1� is continuous and satisfies I3, I5 and I6 iff I is conjugate with Ia, which means the fuzzy impli-cations being conjugate with Ia are the ones and only the ones to be continuous and to satisfy I3, I5 and I6.Since TM is conjugate only with itself, by Proposition 12 in [2], we know that Ig is also conjugate with itself.Thus Theorem 11 shows that the fuzzy implications being conjugate with Ig are the ones and only the ones tobe right-continuous in the second argument and to satisfy I3, I5 and I6. Thus Theorem 11 is an analogousresult to Theorem 1 in [2].
4.3. QL-implications
According to (13), a QL-implication I is generated by a t-conorm S, a t-norm T and a strong fuzzy negationn. In this section, let n denote any strong fuzzy negation. We suppose both S and T are continuous and intendto investigate all possible combinations of S and T and find the sufficient and necessary conditions for I tosatisfy (1). In Table 2 we list all possible combinations of S and T.
![Page 10: On the characterizations of fuzzy implications satisfying](https://reader036.vdocument.in/reader036/viewer/2022082617/57501f631a28ab877e95786b/html5/thumbnails/10.jpg)
Table 2Nine possible combinations of S and T to generate a QL-implications
S ¼ SM S is continuous Archimedean S is ordinal sum
T ¼ T M 1 4 7T is continuous 2 5 8ArchimedeanT is ordinal sum 3 6 9
Y. Shi et al. / Information Sciences 177 (2007) 2954–2970 2963
Theorem 12. A QL-implication I generated by the t-conorm SM, a continuous t-norm T and a strong fuzzy
negation n satisfies (1) iff
(i) T ¼ T M or
(ii) T is the ordinal sum of a family f½am; bm�; T mg where f½am; bm�g is a countable family of non-overlapping,
closed, proper subintervals of ½0; 1� with each Tm being a continuous Archimedean t-norm, and for all
½am; bm�, bm 6 e, where e denotes the equilibrium point of n.
Proof. Here we have for all ðx; yÞ 2 ½0; 1�2, Iðx; yÞ ¼ maxðnðxÞ; T ðx; yÞÞ and
Iðx; Iðx; yÞÞ ¼ maxðnðxÞ; T ðx;maxðnðxÞ; T ðx; yÞÞÞÞ:
(:(i) Because SM is distributive over TM, we obtain Iðx; Iðx; yÞÞ ¼ maxðnðxÞ;minðx;maxðnðxÞ;minðx; yÞÞÞÞ ¼minðmaxðnðxÞ; xÞ;maxðnðxÞ;maxðnðxÞ;minðx; yÞÞÞÞ ¼ minðmaxðnðxÞ; xÞ;maxðnðxÞ;minðx; yÞÞÞ by S4 andthe idempotency of max, ¼ minðmaxðnðxÞ; xÞÞ;minðmaxðnðxÞ; xÞ;maxðnðxÞ; yÞÞ ¼ minðmaxðnðxÞ; xÞ;maxðnðxÞ; yÞÞ by T4 and the idempotency of min ¼ maxðnðxÞ;minðx; yÞÞ ¼ Iðx; yÞ.
(ii) Suppose T is the ordinal sum of the corresponding family f½am; bm�; T mg with each bm 6 e. For an arbi-trary pair ðx0; y0Þ 2 ½0; 1�
2, if for all ½am; bm�, ðx0; y0Þ 62 ½am; bm�2, then according to (4), T ðx0; y0Þ ¼minðx0; y0Þ, from the proof above we can see that Iðx0; y0Þ ¼ Iðx0; Iðx0; y0ÞÞ. If on the contrary there exists½am; bm� such that ðx0; y0Þ 2 ½am; bm�2, then since for x0 6 bm 6 e, nðx0ÞP x0 holds, we havenðx0ÞP x0 P T ðx0; y0Þ and nðx0ÞP x0 P T ðx0;maxðnðx0Þ; T ðx0; y0ÞÞÞ. Thus Iðx0; Iðx0; y0ÞÞ ¼ nðx0Þ ¼Iðx0; y0Þ. Hence Iðx; Iðx; yÞÞ ¼ Iðx; yÞ, for all ðx; yÞ 2 ½0; 1�2.
): According to Theorem 4, if T is a continuous t-norm, then T is either TM, or continuous Archimedean,or the ordinal sum of a family f½am; bm�; T mg. Thus we prove this part through proving that if T is a continuousArchimedean t-norm or the ordinal sum of a family f½am; bm�; T mg and there exists ½am; bm� such that bm > e,then (1) does not hold.
(i) Let T be continuous Archimedean. Since n is continuous, there always exists an x0 2 �0; 1½ such thatnðx0Þ < x0. For y = 1, we have Iðx0; 1Þ ¼ maxðnðx0Þ; x0Þ ¼ x0 and Iðx0; Iðx0; 1ÞÞ ¼ maxðnðx0Þ; T ðx0; x0ÞÞ.Since T is continuous Archimedean, T ðx0; x0Þ < x0. Thus (1) does not hold.
(ii) Let T be the ordinal sum of a family f½am; bm�; T mg and suppose there exists ½am; bm� such that bm > e.Take y ¼ bm and x0 2 �maxðam; eÞ; bm½, then nðx0Þ < e < x0. Thus according to (4), Iðx0; bmÞ ¼maxðnðx0Þ; T ðx0; bmÞÞ ¼ max nðx0Þ; am þ ðbm � amÞT m
x0�ambm�am
; bm�ambm�am
� �� �¼ maxðnðx0Þ; x0Þ ¼ x0.
Since Tm is continuous Archimedean and according to (4), T ðx0; x0Þ ¼ am þ ðbm � amÞT mx0�ambm�am
; x0�ambm�am
� �< am þ ðbm � amÞ x0�am
bm�am¼ x0. Thus Iðx0; Iðx0; bmÞÞ ¼ maxðnðx0Þ; T ðx0; x0ÞÞ < x0. Hence (1) does not
hold. h
Remark 5. Theorem 12 treats the 3 possible combinations in the first column of Table 2:
CASE 1: a QL-implication generated by SM and TM satisfies (1);CASE 2: a QL-implication generated by SM and a continuous Archimedean T does not satisfy (1);
![Page 11: On the characterizations of fuzzy implications satisfying](https://reader036.vdocument.in/reader036/viewer/2022082617/57501f631a28ab877e95786b/html5/thumbnails/11.jpg)
2964 Y. Shi et al. / Information Sciences 177 (2007) 2954–2970
CASE 3: a QL-implication generated by SM and an ordinal sum T satisfies (1) iff T satisfies some extra con-dition as mentioned in Theorem 12.
Theorem 13. A QL-implication I generated by a continuous t-conorm S, the t-norm TM and a strong fuzzy nega-
tion n satisfies (1) iff
(i) S ¼ SM or
(ii) S is the ordinal sum of a family f½am; bm�; Smg where f½am; bm�g is a countable family of non-overlapping,
closed, proper subintervals of ½0; 1� with each Sm being a continuous Archimedean t-conorm, and for all
½am; bm�, am P e, where e denotes the equilibrium point of n.
Proof. Here we have for all ðx; yÞ 2 ½0; 1�2, Iðx; yÞ ¼ SðnðxÞ;minðx; yÞÞ and
Iðx; Iðx; yÞÞ ¼ SðnðxÞ;minðx; SðnðxÞ;minðx; yÞÞÞÞ
(:
(i) Straightforward from the proof of Theorem 12.(ii) Suppose S is the ordinal sum of the corresponding family f½am; bm�; Smg with each am P e. For an arbi-
trary nðx0Þ 2 ½0; 1�, if for all ½am; bm�, nðx0Þ 62 ½am; bm�, then according to (7), for all y 2 ½0; 1�, we haveIðx0; yÞ ¼ maxðnðx0Þ;minðx0; yÞÞ and
Iðx0; Iðx0; yÞÞ ¼ maxðnðx0Þ;minðx0;maxðnðx0Þ;minðx0; yÞÞÞÞ:
Thus according to the proof of Theorem 12, Iðx0; yÞ ¼ Iðx0; Iðx0; yÞÞ. If on the contrary there exists ½am; bm�such that nðx0Þ 2 ½am; bm�, then we have minðx0; yÞ 6 x0 6 e 6 am 6 nðx0Þ, for all y 2 ½0; 1�.
(a) If minðx0; yÞ ¼ e, then x0 ¼ e ¼ nðx0Þ ¼ am. Thus according to (7), Iðx0; yÞ ¼ Sðe; eÞ ¼ e andIðx0; Iðx0; yÞÞ ¼ Sðe;minðx0; eÞÞ ¼ Sðe; eÞ ¼ e ¼ Iðx0; yÞ.
(b) If minðx0; yÞ < e, then minðx0; yÞ < am. According to (7), Iðx0; yÞ ¼ maxðnðx0Þ;minðx0; yÞÞ ¼ nðx0Þ andIðx0; Iðx0; yÞÞ ¼ Sðnðx0Þ;minðx0; nðx0ÞÞÞ ¼ Sðnðx0Þ; x0Þ. If x0 ¼ e, then am ¼ e ¼ nðx0Þ. Thus according to(7) Iðx0; Iðx0; yÞÞ ¼ Sðe; eÞ ¼ e ¼ nðx0Þ ¼ Iðx0; yÞ. If x0 < e, then x0 < am. Thus according to (7)Iðx0; Iðx0; yÞÞ ¼ maxðnðx0Þ; x0Þ ¼ nðx0Þ ¼ Iðx0; yÞ.
Hence Iðx; yÞ ¼ Iðx; Iðx; yÞÞ, for all ðx; yÞ 2 ½0; 1�2.): According to Theorem 8, if S is a continuous t-conorm, then S is either SM, or continuous
Archimedean, or the ordinal sum of a family f½am; bm�; Smg. Thus we prove this part through proving that if Sis continuous Archimedean or the ordinal sum of a family f½am; bm�; Smg and there exists ½am; bm� such thatam < e, then (1) does not hold.
(i) Let S be continuous Archimedean. Since n is continuous, there always exists nðx0Þ 2 �0; 1½ such thatnðx0Þ < x0. For y ¼ 0, we have Iðx0; 0Þ ¼ nðx0Þ and Iðx0; Iðx0; 0ÞÞ ¼ Sðnðx0Þ;minðx0; nðx0ÞÞÞ ¼Sðnðx0Þ; nðx0ÞÞ. Since S is continuous Archimedean, Sðnðx0Þ; nðx0ÞÞ > nðx0Þ. Thus (1) does not hold.
(ii) Let S be the ordinal sum of a family f½am; bm�; Smg and suppose there exists ½am; bm� such that am < e.Since n is continuous, there always exist nðx0Þ 2 �am;minðbm; eÞ½ and x0 > e > nðx0Þ. Take y ¼ 0, sinceSm is continuous Archimedean and according to (7),
Iðx0; Iðx0; 0ÞÞ ¼ Sðnðx0Þ;minðx0; nðx0ÞÞÞ ¼ Sðnðx0Þ; nðx0ÞÞ ¼ am þ ðbm � amÞSmnðx0Þ � am
bm � am;nðx0Þ � am
bm � am
� �
> am þ ðbm � amÞnðx0Þ � am
bm � am¼ nðx0Þ:
Since Iðx0; 0Þ ¼ nðx0Þ, (1) does not hold. h
![Page 12: On the characterizations of fuzzy implications satisfying](https://reader036.vdocument.in/reader036/viewer/2022082617/57501f631a28ab877e95786b/html5/thumbnails/12.jpg)
Y. Shi et al. / Information Sciences 177 (2007) 2954–2970 2965
Remark 6. Theorem 13 treats the three possible choices in the first row of Table 2: CASE 1 (so as mentioned inRemark 5); CASE 4: a QL-implication generated by a continuous Archimedean S and TM does not satisfy (1);CASE 7: a QL-implication generated by an ordinal sum S and TM satisfies (1) iff S satisfies some extra con-dition as mentioned in Theorem 13.
Lemma 1. A necessary condition for a QL-implication I generated by a continuous Archimedean t-conorm S, a
continuous t-norm T and a strong fuzzy negation n to satisfy (1) is T ðx; nðxÞÞ ¼ 0, for all x 2 ½0; 1�.
Proof. Suppose there exists x0 2 ½0; 1� such that T ðx0; nðx0ÞÞ > 0, then x0 > 0 and nðx0Þ > 0. Since n is strong,x0 > 0 implies nðx0Þ < 1. According to (5) and (6), there exists a strictly increasing generator g and its pseudo-inverse gð�1Þ such that
Sðnðx0Þ; T ðx0; nðx0ÞÞÞ ¼ gð�1Þðgðnðx0ÞÞ þ gðT ðx0; nðx0ÞÞÞÞ ¼g�1ðgðnðx0ÞÞ þ gðT ðx0; nðx0ÞÞÞÞ gðnðx0ÞÞ þ gðT ðx0; nðx0ÞÞÞ 2 ½0; gð1Þ�1 otherwise:
�
If Sðnðx0Þ; T ðx0; nðx0ÞÞÞ ¼ 1, then Sðnðx0Þ; T ðx0; nðx0ÞÞÞ > nðx0Þ.If Sðnðx0Þ; T ðx0; nðx0ÞÞÞ < 1, then
Sðnðx0Þ; T ðx0; nðx0ÞÞÞ ¼ g�1ðgðnðx0ÞÞ þ gðT ðx0; nðx0ÞÞÞÞ:
Since g and g�1 are strictly increasing, T ðx0; nðx0ÞÞ > 0 impliesg�1ðgðnðx0ÞÞ þ gðT ðx0; nðx0ÞÞÞÞ > g�1ðgðnðx0ÞÞÞ ¼ nðx0Þ;
namely, Sðnðx0Þ; T ðx0; nðx0ÞÞÞ > nðx0Þ. According to (13), Iðx0; 0Þ ¼ nðx0Þ and Iðx0; Iðx0; 0ÞÞ ¼ Sðnðx0Þ;T ðx0; nðx0ÞÞÞ. That is to say, Iðx0; 0Þ < Iðx0; Iðx0; 0ÞÞ, and hence, I does not satisfy (1). h
Remark 7. According to Theorem 3, a continuous t-norm T satisfying T ðx; nðxÞÞ ¼ 0, for all x 2 ½0; 1� iff T isconjugate with the Łukasiewicz t-norm T –L. Thus Lemma 1 treats the case 6 of Table 2: a QL-implication gen-erated by a continuous Archimedean S and an ordinal sum T does not satisfy (1).
Lemma 2. A necessary condition for a QL-implication I generated by a continuous Archimedean t-conorm S, a
continuous Archimedean t-norm T and a strong fuzzy negation n to satisfy (1) is Sðx; nðxÞÞ ¼ 1, for all x 2 ½0; 1�.
Proof. Suppose there exists x0 2 ½0; 1� such that Sðx0; nðx0ÞÞ < 1, then x0 2 �0; 1½. According to (2) and (3),
T ðx0; Sðnðx0Þ; x0ÞÞ ¼ f ð�1Þðf ðx0Þ þ f ðSðnðx0Þ; x0ÞÞÞ ¼f �1ðf ðx0Þ þ f ðSðnðx0Þ; x0ÞÞÞ f ðx0Þ þ f ðSðnðx0Þ; x0ÞÞ 2 ½0; f ð0Þ�;0 otherwise:
�
Since f is strictly decreasing and f ð1Þ ¼ 0, Sðnðx0Þ; x0Þ < 1 implies f ðSðnðx0Þ; x0ÞÞ > 0. If f ðx0Þþf ðSðnðx0Þ; x0ÞÞ 2 ½0; f ð0Þ�, then
T ðx0; Sðnðx0Þ; x0ÞÞ ¼ f �1ðf ðx0Þ þ f ðSðnðx0Þ; x0ÞÞÞ < f �1ðf ðx0ÞÞ ¼ x0:
If f ðx0Þ þ f ðSðnðx0Þ; x0ÞÞ 2 �f ð0Þ;1�, then T ðx0; Sðnðx0Þ; x0ÞÞ ¼ 0 < x0. Thus T ðx0; Sðnðx0Þ; x0ÞÞ < x0 alwaysholds.
According to (5) and (6), since Sðnðx0Þ; x0Þ < 1, it can be expressed by g�1ðgðnðx0ÞÞ þ gðx0ÞÞ and we havegðnðx0ÞÞ þ gðx0Þ 2 ½0; gð1Þ½.
Because T ðx0; Sðnðx0Þ; x0ÞÞ < x0 and g is strictly increasing, gðT ðx0; Sðnðx0Þ; x0ÞÞÞ < gðx0Þ. Thus
gðnðx0ÞÞ þ gðT ðx0; Sðnðx0Þ; x0ÞÞÞ < gðnðx0ÞÞ þ gðx0Þ;
which implies gðnðx0ÞÞ þ gðT ðx0; Sðnðx0Þ; x0ÞÞÞ 2 ½0; gð1Þ½ . Thus we obtain
Sðnðx0Þ; T ðx0; Sðnðx0Þ; x0ÞÞÞ ¼ g�1ðgðnðx0ÞÞ þ gðT ðx0; Sðnðx0Þ; x0ÞÞÞÞ:
![Page 13: On the characterizations of fuzzy implications satisfying](https://reader036.vdocument.in/reader036/viewer/2022082617/57501f631a28ab877e95786b/html5/thumbnails/13.jpg)
2966 Y. Shi et al. / Information Sciences 177 (2007) 2954–2970
Since g�1 is also strictly increasing and gðnðx0ÞÞ þ gðx0Þ 2 ½0; gð1Þ½,
g�1ðgðnðx0ÞÞ þ gðx0ÞÞ > g�1ðgðnðx0ÞÞ þ gðT ðx0; Sðnðx0Þ; x0ÞÞÞÞ;
namely, Sðnðx0Þ; x0Þ > Sðnðx0Þ; T ðx0; Sðnðx0Þ; x0ÞÞÞ. According to (13), Iðx0; 1Þ ¼ Sðnðx0Þ; x0Þ and Iðx0; Iðx0; 1ÞÞ ¼Sðnðx0Þ; T ðx0; Sðnðx0Þ; x0ÞÞÞ. That is to say, Iðx0; 1Þ > Iðx0; Iðx0; 1ÞÞ, and hence, I does not satisfy (1). h
Corollary 4. If a QL-implication I generated by a continuous Archimedean t-conorm S, a continuous Archime-dean t-norm T and a strong fuzzy negation n satisfies (1), then neither S nor T can be strict.
Proof. Straightforward from Lemmas 1, 2, Theorems 3 and 7. h
Lemma 3. If a QL-implication I generated by a continuous t-conorm S, a continuous Archimedean t-norm T and a
strong fuzzy negation n satisfies (1), then there does not exist a countable family f½am; bm�g of non-overlapping,
closed, proper subintervals of ½0; 1� such that S can be the ordinal sum of the corresponding family
f½am; bm�; Smg with each Sm being a continuous Archimedean t-norm.
Proof. According to (13), take y ¼ 1, then Iðx; 1Þ ¼ SðnðxÞ; xÞ and
Iðx; Iðx; 1ÞÞ ¼ SðnðxÞ; T ðx; SðnðxÞ; xÞÞÞ:
We will prove this lemma by contraposition.Let e denote the equilibrium point of n and suppose there exists a countable family f½am; bm�g such that Scan be the ordinal sum of the corresponding family f½am; bm�; Smg with each Sm being a continuousArchimedean t-norm.
If for all ½am; bm�, ½0; e� 6� ½am; bm�, then since n is continuous, there always exists nðx0Þ 2 �0; e½ such thatnðx0Þ 62 ½am; bm� and x0 > e > nðx0Þ. Thus Iðx0; 1Þ ¼ Sðnðx0Þ; x0Þ ¼ maxðnðx0Þ; x0Þ ¼ x0. And since T iscontinuous Archimedean and x0 2 �e; 1½, T ðx0; x0Þ < x0. Thus by (13) and (7), Iðx0; Iðx0; 1ÞÞ ¼Sðnðx0Þ; T ðx0; x0ÞÞ ¼ maxðnðx0Þ; T ðx0; x0ÞÞ < x0.
Hence (1) does not hold.If on the contrary there exists ½am; bm� such that ½0; e� � ½am; bm�, then we have am ¼ 0 and bm 2 ½e; 1½ . Take
x1 2 �bm; 1½ , then nðx1Þ < nðbmÞ 6 bm < x1.And since T is continuous Archimedean, T ðx1; x1Þ < x1.Thus by (13) and (7), we have Iðx1; 1Þ ¼ Sðnðx1Þ; x1Þ ¼ maxðnðx1Þ; x1Þ ¼ x1, and
Iðx1; Iðx1; 1ÞÞ ¼ Iðx1; x1Þ ¼ Sðnðx1Þ;T ðx1; x1ÞÞ ¼am þ ðbm � amÞSmðnðx1Þ�am
bm�am; T ðx1;x1Þ�am
bm�amÞ T ðx1; x1Þ 2 ½am; bm�;
maxðnðx1Þ;T ðx1; x1ÞÞ otherwise:
(
If T ðx1; x1Þ 2 ½am; bm�, then Iðx1; Iðx1; 1ÞÞ 2 ½am; bm�, i.e., Iðx1; Iðx1; 1ÞÞ < x1.If T ðx1; x1Þ 62 ½am; bm�, then Iðx1; Iðx1; 1ÞÞ ¼ maxðnðx1Þ; T ðx1; x1ÞÞ < x1. Therefore we always have
Iðx1; Iðx1; 1ÞÞ < Iðx1; 1Þ.Hence (1) does not hold. h
Remark 8. Lemma 3 treats the case 8 of Table 2: a QL-implication generated by an ordinal sum S and a con-tinuous Archimedean T does not satisfy (1).
Hence besides the theorems and lemmas above, there are only two combinations of S and T to be investi-gated: the case 5 and the case 9. And from Corollary 4, we know that for the case 5, it is necessary that both S
and T are nilpotent.
Theorem 14. Let / denote an automorphism of the unit interval. A QL-implication I generated by the nilpotent
t-conorm which is expressed by Sðx; yÞ ¼ /�1ðminð/ðxÞ þ /ðyÞ; 1ÞÞ, the nilpotent t-norm which is expressed by
T ðx; yÞ ¼ /�1ðmaxð/ðxÞ þ /ðyÞ � 1; 0ÞÞ and a strong fuzzy negation n satisfies (1) iff nðxÞ ¼ /�1ð1� /ðxÞÞ, for
all x 2 ½0; 1�.
![Page 14: On the characterizations of fuzzy implications satisfying](https://reader036.vdocument.in/reader036/viewer/2022082617/57501f631a28ab877e95786b/html5/thumbnails/14.jpg)
Y. Shi et al. / Information Sciences 177 (2007) 2954–2970 2967
Proof. According to Lemmas 1 and 2, I satisfies (1) iff for all x 2 ½0; 1�, T ðx; nðxÞÞ ¼ 0 and SðnðxÞ; xÞ ¼ 1. Andaccording to Theorems 3 and 7, T ðx; nðxÞÞ ¼ 0 and SðnðxÞ; xÞ ¼ 1 iff nðxÞ ¼ /�1ð1� /ðxÞÞ. h
Remark 9. Actually, if there exists an automorphism / of the unit interval such that the QL-implication I isgenerated by the t-conorm Sðx; yÞ ¼ /�1ðminð/ðxÞ þ /ðyÞ; 1ÞÞ, the t-norm T ðx; yÞ ¼ /�1ðmaxð/ðxÞþ/ðyÞ � 1; 0ÞÞ and the strong fuzzy negation nðxÞ ¼ /�1ð1� /ðxÞÞ, then Iðx; yÞ ¼ SðnðxÞ; T ðx; yÞÞ ¼maxð/�1ð1� /ðxÞÞ; yÞ ¼ maxðnðxÞ; yÞ, i.e., I is an S-implication generated by SM and the strong fuzzy nega-tion n, and hence according to Corollary 2, it satisfies (1).
Theorem 15. Let u and / denote two different automorphisms of the unit interval. The QL-implication I gener-
ated by the nilpotent t-conorm which is expressed by Sðx; yÞ ¼ u�1ðminðuðxÞ þ uðyÞ; 1ÞÞ, the nilpotent t-norm
which is expressed by T ðx; yÞ ¼ /�1ðmaxð/ðxÞ þ /ðyÞ � 1; 0ÞÞ and a strong fuzzy negation n satisfies (1) iff n
satisfies
u�1ð1� uðxÞÞ 6 nðxÞ 6 /�1ð1� /ðxÞÞ; 8x 2 ½0; 1� ð14Þ
and I is expressed byIðx; yÞ ¼nðxÞ T ðx; yÞ ¼ 0;
y 0 < T ðx; yÞ < u�1ð1� uðnðxÞÞÞ;1 T ðx; yÞP u�1ð1� uðnðxÞÞÞ:
8><>: ð15Þ
Proof. Here we have for all ðx; yÞ 2 ½0; 1�2, Iðx; yÞ ¼ SðnðxÞ; T ðx; yÞÞ and
Iðx; Iðx; yÞÞ ¼ SðnðxÞ; T ðx; SðnðxÞ; T ðx; yÞÞÞÞ: ð16Þ
): Suppose that I satisfies (1), then according to Lemmas 1, 2, Theorems 3 and 7, we haveu�1ð1� uðxÞÞ 6 nðxÞ 6 /�1ð1� /ðxÞÞ:
Hence (14) holds. Next we will prove that I satisfy (15).According to (13), T ðx; yÞ ¼ 0 implies Iðx; yÞ ¼ nðxÞ. And T ðx; yÞP u�1ð1� uðnðxÞÞÞ implies Iðx; yÞ ¼ 1.Thus in order to prove that I satisfies (15), we only need to prove:
0 < T ðx; yÞ < u�1ð1� uðnðxÞÞÞ ) Iðx; yÞ ¼ y:
Indeed, if T ðx; yÞ < u�1ð1� uðnðxÞÞÞ, then
Iðx; yÞ ¼ u�1ðuðnðxÞÞ þ uðT ðx; yÞÞÞ < 1:
If I satisfies (1), then Iðx; Iðx; yÞÞ < 1. Thus according to (16), we have
Iðx; Iðx; yÞÞ ¼ u�1ðuðnðxÞÞ þ uðT ðx;u�1ðuðnðxÞÞ þ uðT ðx; yÞÞÞÞÞÞ:
Since Iðx; yÞ ¼ Iðx; Iðx; yÞÞ and both u and u�1 are strictly increasing mappings, we haveT ðx; yÞ ¼ T ðx;u�1ðuðnðxÞÞ þ uðT ðx; yÞÞÞÞ:
Because T ðx; yÞ > 0, T ðx; yÞ ¼ /�1ð/ðxÞ þ /ðyÞ � 1Þ. AndT ðx;u�1ðuðnðxÞÞ þ uðT ðx; yÞÞÞÞ ¼ /�1ð/ðxÞ þ /ðu�1ðuðnðxÞÞ þ uðT ðx; yÞÞÞÞ � 1Þ:
Since both / and /�1 are strictly increasing mappings, we have y ¼ u�1ðuðnðxÞÞ þ uðT ðx; yÞÞÞ, namelyIðx; yÞ ¼ y. Hence (15) holds.(: Suppose n and I satisfy (14) and (15), then we will prove that I satisfies (1). According to (15), ifT ðx; yÞ ¼ 0, then Iðx; yÞ ¼ nðxÞ. And
Iðx; Iðx; yÞÞ ¼ Iðx; nðxÞÞ ¼ SðnðxÞ; T ðx; nðxÞÞÞ:
Since n satisfies (14), T ðx; nðxÞÞ ¼ 0, for all x 2 ½0; 1�. ThusSðnðxÞ; T ðx; nðxÞÞÞ ¼ SðnðxÞ; 0Þ ¼ nðxÞ ¼ Iðx; yÞ:
![Page 15: On the characterizations of fuzzy implications satisfying](https://reader036.vdocument.in/reader036/viewer/2022082617/57501f631a28ab877e95786b/html5/thumbnails/15.jpg)
2968 Y. Shi et al. / Information Sciences 177 (2007) 2954–2970
If T ðx; yÞP u�1ð1� uðnðxÞÞÞ, then Iðx; yÞ ¼ 1. Thus Iðx; Iðx; yÞÞ ¼ Iðx; 1Þ ¼ SðnðxÞ; xÞ. Since n satisfies (14),SðnðxÞ; xÞ ¼ 1 for all x 2 ½0; 1�. Thus Iðx; Iðx; yÞÞ ¼ 1 ¼ Iðx; yÞ.
If 0 < T ðx; yÞ < u�1ð1� uðnðxÞÞÞ, then Iðx; yÞ ¼ y. Thus Iðx; Iðx; yÞÞ ¼ Iðx; yÞ. Hence I satisfies (1). h
Remark 10. Corollary 4, Theorems 14 and 15 treat the case 5 of Table 2: a QL-implication generated by acontinuous Archimedean S and a continuous Archimedean T satisfies (1) providing both are nilpotent andsatisfy some extra conditions as stated.
Theorem 16. Let f½am; bm�g and f½cm; dm�g denote two countable families of non-overlapping, closed, proper sub-
intervals of ½0; 1�. And let T be a t-norm which is the ordinal sum of the corresponding family f½am; bm�; T mg with
each Tm being a continuous Archimedean t-norm and S be a t-conorm which is the ordinal sum of the correspond-
ing family f½cm; dm�; Smg with each Sm being a continuous Archimedean t-conorm. Then the QL-implication I gen-
erated by S, T and a strong fuzzy negation n satisfies (1) iff for all ½am; bm�, bm 6 e and for all ½cm; dm�, cm P e,where e denotes the equilibrium point of n.
Proof. We will first derive some special instances of Iðx; Iðx; yÞÞ.According to (13) and (16), for y ¼ 1, we obtain Iðx; 1Þ ¼ SðnðxÞ; xÞ and Iðx; Iðx; 1ÞÞ ¼ SðnðxÞ;
T ðx; SðnðxÞ; xÞÞÞ. And for y ¼ 0, we obtain Iðx; 0Þ ¼ nðxÞ and Iðx; Iðx; 0ÞÞ ¼ SðnðxÞ; T ðx; nðxÞÞÞ.(: In order to prove the ‘(’ part, we will consider three cases according to the position of x w.r.t. e:
(i) For all x < e, we have nðxÞ > e, x < nðxÞ, and x < cm, for all cm. Since for all y 2 ½0; 1�, T ðx; yÞ 6 x < cm
and T ðx; SðnðxÞ; T ðx; yÞÞÞ 6 x < cm, according to (7), we have:
Iðx; yÞ ¼ SðnðxÞ; T ðx; yÞÞ ¼ maxðnðxÞ; T ðx; yÞÞ ¼ nðxÞand
Iðx; Iðx; yÞÞ ¼ SðnðxÞ; T ðx; SðnðxÞ; T ðx; yÞÞÞÞ ¼ maxðnðxÞ; T ðx; SðnðxÞ; T ðx; yÞÞÞÞ ¼ nðxÞ:Thus Iðx; yÞ ¼ Iðx; Iðx; yÞÞ.
(ii) For all x > e, we have nðxÞ < e, x > nðxÞ, x > bm, for all bm and nðxÞ < cm, for all cm. Then by (7) and (4),for all y 2 ½0; 1�, Iðx; yÞ ¼ maxðnðxÞ;minðx; yÞÞ and
Iðx; Iðx; yÞÞ ¼ maxðnðxÞ;minðx;maxðnðxÞ;minðx; yÞÞÞÞ:Thus according to the proof of Theorem 12, Iðx; yÞ ¼ Iðx; Iðx; yÞÞ.
(iii) For x ¼ e, we have nðxÞ ¼ e and Iðx; yÞ ¼ Iðe; yÞ ¼ Sðe; T ðe; yÞÞ.Two subcases will be considered depend-ing now on the position of the variable y:
(1) y 6 e.If for all ½am; bm�, ðe; yÞ 62 ½am; bm�2, then according to (4), T ðe; yÞ ¼ minðe; yÞ ¼ y.If on the contrary there exists ½am; bm� such that ðe; yÞ 2 ½am; bm�2, then bm ¼ e. According to (4),T ðe; yÞ ¼ am þ ðe� amÞT mðe� am
e� am;y � am
e� amÞ ¼ y:
That is to say, T ðe; yÞ ¼ y; for all y 6 e. Thus Iðx; yÞ ¼ Sðe; yÞ:And if for all ½cm; dm�, ðe; yÞ 62 ½cm; dm�2, then according to (7), Sðe; yÞ ¼ maxðe; yÞ ¼ e.If on the contrary there exists ½cm; dm� such that ðe; yÞ 2 ½cm; dm�2, then cm ¼ y ¼ e. According to (7),
Sðe; yÞ ¼ eþ ðdm � eÞSme� e
dm � e;
y � edm � e
� �¼ y ¼ e:
That is to say, Iðx; yÞ ¼ Iðe; yÞ ¼ e, for all y 6 e.(2) y > e.In this case y > bm. Thus according to (4), T ðe; yÞ ¼ minðe; yÞ ¼ e and hence Iðx; yÞ ¼ Sðe; T ðe; yÞÞ ¼Sðe; eÞ.Now we look at the position of cm (P e) w.r.t. e.
![Page 16: On the characterizations of fuzzy implications satisfying](https://reader036.vdocument.in/reader036/viewer/2022082617/57501f631a28ab877e95786b/html5/thumbnails/16.jpg)
Y. Shi et al. / Information Sciences 177 (2007) 2954–2970 2969
If cm > e, then according to (7), Sðe; eÞ ¼ maxðe; eÞ ¼ e. If cm ¼ e, then according to (7),
Sðe; eÞ ¼ eþ ðdm � eÞSme� e
dm � e;
e� edm � e
� �¼ e:
That is to say, Iðe; yÞ ¼ e, for all y > e.Hence Iðe; yÞ ¼ e, for all y 2 ½0; 1�. Therefore we have
Iðx; Iðx; yÞÞ ¼ Iðe; Iðe; yÞÞ ¼ Sðe; T ðe; Iðe; yÞÞÞ ¼ Sðe; T ðe; eÞÞ:Finally we consider the position of the upper bound bm (6 e) w.r.t. e.If bm < e, then according to (4), T ðe; eÞ ¼ minðe; eÞ ¼ e. If bm ¼ e, then according to (4),
T ðe; eÞ ¼ am þ ðe� amÞT me� am
e� am;e� am
e� am
� �¼ e:
Thus Sðe; T ðe; eÞÞ ¼ Sðe; eÞ ¼ e, according to the proof above. Therefore Iðx; Iðx; yÞÞ ¼ e ¼ Iðe; yÞ ¼Iðx; yÞ.
Hence we can conclude that Iðx; Iðx; yÞÞ ¼ Iðx; yÞ, for all ðx; yÞ 2 ½0; 1�2, i.e., I satisfies (1).): The reverse implication (‘Rightarrow’) will be proved by contraposition.
(i) First assume there exists ½am; bm� such that bm > e.If for all ½cm; dm�, cm > nðbmÞ, then since n is continu-ous, there exists x0 such that nðx0Þ 2 �nðbmÞ;minðcm; eÞ½ and x0 2 �maxðam; eÞ; bm½ . Thus Iðx0; 1Þ ¼Sðnðx0Þ; x0Þ ¼ maxðnðx0Þ; x0Þ ¼ x0 and Iðx0; Iðx0; 1ÞÞ ¼ Sðnðx0Þ; T ðx0; Sðnðx0Þ; x0ÞÞÞ ¼ Sðnðx0Þ; T ðx0; x0ÞÞ ¼maxðnðx0Þ; T ðx0; x0ÞÞ:Since Tm is continuous Archimedean and according to (4),
T ðx0; x0Þ ¼ am þ ðbm � amÞT mx0 � am
bm � am;
x0 � am
bm � am
� �< x0:
Thus Iðx0; Iðx0; 1ÞÞ < x0. Hence (1) does not hold. If on the contrary there exists ½cm; dm� such thatcm 6 nðbmÞ, then cm < e. Since n is continuous, there exists x1 such that nðx1Þ 2 �cm;minðdm; eÞ½ andx1 2 �bm; 1½ . Thus Iðx1; Iðx1; 0ÞÞ ¼ Sðnðx1Þ; T ðx1; nðx1ÞÞÞ ¼ Sðnðx1Þ;minðx1; nðx1ÞÞÞ ¼ Sðnðx1Þ; nðx1ÞÞ. SinceSm is continuous Archimedean and according to (7),
Sðnðx1Þ; nðx1ÞÞ ¼ cm þ ðdm � cmÞSmnðx1Þ � cm
dm � cm;nðx1Þ � cm
dm � cm
� �> nðx1Þ:
And Iðx1; 0Þ ¼ nðx1Þ < Iðx1; Iðx1; 0ÞÞ. Thus (1) does not hold. So we have proved that if there exists½am; bm� such that bm > e, then (1) does not hold.
(ii) Second assume that there exists ½cm; dm� such that cm < e and for all ½am; bm�, bm 6 e.Since n is continuous, there exists x0 such that nðx0Þ 2 �cm;minðdm; eÞ½ and x0 > bm. Then Iðx0; 0Þ ¼ nðx0Þand according to (4), Iðx0; Iðx0; 0ÞÞ ¼ Sðnðx0Þ; T ðx0; nðx0ÞÞÞ ¼ Sðnðx0Þ;minðx0; nðx0ÞÞÞ ¼ Sðnðx0Þ; nðx0ÞÞ.Since Sm is continuous Archimedean and according to (7),
Sðnðx0Þ; nðx0ÞÞ ¼ cm þ ðdm � cmÞSmnðx0Þ � cm
dm � cm;nðx0Þ � cm
dm � cm
� �> nðx0Þ:
Thus (1) does not hold. h
Remark 11. Theorem 16 treats the case 9 of Table 2: a QL-implication generated by an ordinal sum S and anordinal sum T satisfies (1) providing both satisfy some extra conditions as stated.
![Page 17: On the characterizations of fuzzy implications satisfying](https://reader036.vdocument.in/reader036/viewer/2022082617/57501f631a28ab877e95786b/html5/thumbnails/17.jpg)
2970 Y. Shi et al. / Information Sciences 177 (2007) 2954–2970
5. Conclusions
Iterative boolean-like laws are proposed by the authors in [1]. In this paper, we studied an iterative boolean-like law in which fuzzy implications are concerned, namely whether Iðx; Iðx; yÞÞ ¼ Iðx; yÞ holds for allðx; yÞ 2 ½0; 1�2 or not where I denotes a fuzzy implication derived from classical logic, which is generated byt-norms, t-conorms and fuzzy negations.
In Section 4 we gave sufficient and necessary conditions for an S-implication generated by any t-conormand any fuzzy negation to satisfy (1) in Theorem 10 and its corollary, for an R-implication generated by a leftcontinuous t-norm in to satisfy (1) in Theorem 11 and its corollary and for a QL-implication generated by acontinuous t-conorm, a continuous t-norm and a strong fuzzy negation to satisfy (1) from Theorem 12–16.
The standard t-norm TM and the standard t-conorm SM play important roles in the results of the investi-gated Eq. (1). An S-implication generated by a t-conorm S and a continuous fuzzy negation n satisfies Eq. (1)iff S is SM. And an R-implication generated by a left-continuous t-norm T satisfies Eq. (1) iff T is TM. But for aQL-implication generated by a fuzzy negation n, a t-norm T and a t-conorm S to satisfy Eq. (1), it is sufficientbut not necessary for T to be TM and S to be SM. Instead, the equilibrium point of the strong fuzzy negationplay an important role in the solutions for QL-implications to satisfy (1).
Acknowledgement
The authors want to thank the referees for their helpful comments that substantially improved this paper.
References
[1] C. Alsina, E. Trillas, On iterative boolean-like laws of fuzzy sets, in: Proc. 4th Conf. Fuzzy Logic and Technology, Barcelona, Spain2005, pp. 389–394.
[2] M. Baczynski, Residual implications revisited. Notes on the Smets–Magrez Theorem, Fuzzy Sets and Systems 145 (2004) 267–277.[3] M. Baczynski, B. Jayaram, On the characterizations of ðS;NÞ-implications generated from continuous negations, in: Proc. 11th Conf.
Information Processing and Management of Uncertainty in Knowledge-based Systems, Paris, France 2006, pp. 436–443.[4] H. Bustince, P. Burillo, F. Soria, Automorphisms, negations and implication operators, Fuzzy Sets and Systems 134 (2003) 209–229.[5] H. Bustince, M. Pagolaa, E. Barrenechea, Construction of fuzzy indices from fuzzy DI-subsethood measures: application to the global
comparison of images, Information Sciences 177 (2007) 906–929.[6] J.C. Fodor, On fuzzy implication operators, Fuzzy Sets and Systems 42 (1991) 293–300.[7] J.C. Fodor, Contrapositive symmetry of fuzzy implications, Fuzzy Sets and Systems 69 (1995) 141–156.[8] J.C. Fodor, M. Roubens, Fuzzy Preference Modelling and Multicriteria Decision Support, Kluwer Academic Publishers,
Netherlands, 1994.[9] E.E. Kerre, C. Huang, D. Ruan, Fuzzy Set Theory and Approximate Reasoning, Wu Han University Press, Wu Chang, 2004.
[10] E.P. Klement, R. Mesiar, E. Pap, Triangular Norms, Kluwer Academic Publishers, Netherlands, 2000.[11] G.J. Klir, B. Yuan, Fuzzy Sets and Fuzzy Logic, Theory and Applications, Prentice Hall, USA, 1995.[12] H.W. Liua, G.J. Wang, Unified forms of fully implicational restriction methods for fuzzy reasoning, Information Sciences 177 (2007)
956–966.[13] N.N. Morsia, Elewa M. Roshdy, Issues on adjointness in multiple-valued logics, Information Sciences 176 (2006) 2886–2909.[14] A. Pradera, E. Trillas, E. Renedo, An overview on the construction of fuzzy set theories, New Mathematics and Natural Computation
1 (2005) 329–358.[15] D. Ruan, A critical study of widely used fuzzy implication operators and their influence on the inference rules in fuzzy expert systems,
Ph.D. Thesis, Gent University, 1990.[16] D. Ruan, E.E. Kerre, Fuzzy implication operators and generalized fuzzy method of cases, Fuzzy Sets and Systems 54 (1993) 23–37.[17] P. Smets, P. Magrez, Implication in fuzzy logic, International Journal of Approximate Reasoning 1 (1987) 327–347.[18] E. Trillas, sobre funciones de negacion en la teora de conjuntos difusos, Stochastica 3 (1979) 47–84.[19] R.R. Yager, On global requirements for implication operators in fuzzy modus ponens, Fuzzy Sets and Systems 106 (1999) 3–10.[20] R.R. Yager, On some new classes of implication operators and their role in approximate reasoning, Information Sciences 167 (2004)
193–216.[21] R.R. Yager, On the implication operator in fuzzy logic, Information Sciences 31 (1983) 141–164.[22] P. Yan, G. Chen, Discovering a cover set of ARsi with hierarchy from quantitative databases, Information Sciences 173 (2005) 319–
336.