International Journal of Approximate Reasoning 88 (2017) 91–109

Contents lists available at ScienceDirect

International Journal of Approximate Reasoning

www.elsevier.com/locate/ijar

Interval-valued implications and interval-valued strong

equality index with admissible orders
H. Zapata a , H. Bustince b,c,∗ , S. Montes d , B. Bedregal e , G.P. Dimuro f,c ,
Z. Takáč g , M. Baczyński h , J. Fernandez b,c
a
Facultad de Ciencias, Universidad Central de Venezuela, Avenida los ilustres, Caracas 1020, Venezuela
b
Departamento de Automática y Computación, Universidad Pública de Navarra, Campus Arrosadía s/n, P.O. Box 31006, Pamplona, Spain
c
Institute of Smart Cities, Universidad Pública de Navarra, Pamplona, Spain
d
Department of Statistics and O.R., University of Oviedo, Spain
e
Departamento de Informática e Matemática Aplicada, Universidade Federal do Rio Grande do Norte, Campus Universitário, s/n – Lagoa
Nova, 59078-900, Natal, Brazil
f
Centro de Ciências Computacionais, Universidade Federal do Rio Grande, Avenida Itália, Campus Carreiros, 96203-900, Rio Grande, Brazil
g
Institute of Information Engineering, Automation and Mathematics, Slovak University of Technology in Bratislava, Radlinskeho 9, Bratislava,
Slovakia
h
Institute of Mathematics, University of Silesia, ul. Bankowa 14, 40-007 Katowice, Poland

a r t i c l e i n f o a b s t r a c t

Article history: In this work we introduce the definition of interval-valued fuzzy implication function with
Received 9 February 2017 respect to any total order between intervals. We also present different construction meth-
Received in revised form 19 May 2017 ods for such functions. We show that the advantage of our definitions and constructions
Accepted 22 May 2017
lays on that we can adapt to the interval-valued case any inequality in the fuzzy setting,
Available online 29 May 2017
as the one of the generalized modus ponens. We also introduce a strong equality measure
Keywords: between interval-valued fuzzy sets, in which we take the width of the considered inter-
Interval-valued fuzzy implications vals into account, and, finally, we discuss a construction method for this measure using
Admissible order implication functions with respect to total orders.
Interval-valued generalized modus ponens © 2017 Elsevier Inc. All rights reserved.
Interval-valued strong equality index

1. Introduction

Fuzzy implication functions are widely used to build similarity measures, entropies, erosion operators in fuzzy morphol-
ogy, inclusion measures, etc. [6,8,13,19]. However, one of the most important applications of such operators is in fuzzy
inference through the generalized modus ponens (GMP), as follows (see [2,25,26]): if I is a fuzzy implication function, T ,
a t-norm and N, a fuzzy negation, the GMP with respect to T and N is written by means of the inequality:

T (x, I (x, y )) ≤ y for all x, y ∈ [0, 1] . (1)

Besides, in 1971 Zadeh, in his paper Quantitative fuzzy semantics ([34]), settles that the construction of the fuzzy sets
(FSs), that is, the determination of the membership degree of each element to a set, is the biggest problem for using fuzzy
sets theory in applications. This fact led him to suggest, in 1975 [35], the notion of interval-valued fuzzy set (IVFS). The

*
Corresponding author.
E-mail addresses: [email protected] (H. Zapata), [email protected] (H. Bustince), [email protected] (S. Montes), [email protected]
(B. Bedregal), [email protected] (G.P. Dimuro), [email protected] (Z. Takáč), [email protected] (M. Baczyński),
[email protected] (J. Fernandez).

http://dx.doi.org/10.1016/j.ijar.2017.05.009
0888-613X/© 2017 Elsevier Inc. All rights reserved.
92 H. Zapata et al. / International Journal of Approximate Reasoning 88 (2017) 91–109

definition of IVFS was given in the same year by Sambuc [29] as follows: the membership degree of the elements to the
interval-valued fuzzy set is given by a closed subinterval of the unit interval [0, 1].
There exist many interpretations of IVFSs, and we are going to consider the most usual among them, which consists of
considering that pointwise membership degree of the element to the considered set is a point of the considered subinter-
val [14].
Since 1975 a lot has been written about interval-valued fuzzy sets. We consider that, when dealing with IVFSs, we should
take into account the following:

1. The output of any information measure, any logical connective, etc., defined on IVFSs, must always be a closed subin-
terval of [0, 1];
2. From our interpretation of an interval-valued membership degree, FSs are a particular case of IVFSs, since they can be
considered as intervals of null width;
3. Each interval has an associated width, which provides information that must be used in our theoretical and applied
developments;
4. In the application, in order to recover fuzzy results, i.e., when the width of the considered membership intervals is
equal to zero, we should use total orders between intervals since, if we just use the classical partial order relationship,
[x, x]  [ y , y ] iff x ≤ y and x ≤ y, it may happen that the resulting intervals are not comparable to each other. For
instance, in the case of the GMP with intervals, it may happen that, if I is an interval-valued implication function and
T is an interval-valued t-norm, both with respect to the partial order , the result of T ([x, x], I ([x, x], [ y , y ]) is not
comparable to the interval [ y , y ], whereas this does not happen in the fuzzy case.

We should remark that items 3 and 4 are crucial, as they make the difference between using real numbers or using
intervals to represent the membership degree of the elements to the set.
These considerations has led us to the following objectives in this paper:

(A) To define and study the concept of interval-valued implication function for every order relationship between intervals
(either partial or total).
(B) To introduce methods for constructing interval-valued implication functions.
(C) To analyze the importance of preserving the width of the intervals under appropriate conditions.
(D) To analyze under which conditions the interval-valued GMP holds.
(E) To study strong equality measures built using interval-valued implication functions defined with respect to total orders.

To make this study, we are going to use the notion of admissible order which was introduced in [16] and the studies of
aggregation functions for total orders developed in [17].
This paper is organized as follows: In Preliminaries, we recall the basic concepts that we are going to use. Next we
introduce the definition of fuzzy negation for intervals using linear orders. In Section 4, we define interval-valued implication
functions imposing decreasingness in the first component and increasingness in the second one with respect to a total order.
Then we present several construction methods. In Section 6 we analyze the case of Xu and Yager and the lexicographical
orders. Next we analyze the GMP and its properties when we use interval-valued t-norms and implication functions defined
by means of total orders. We analyze under which conditions the widths of the intervals are preserved and in the following
section we introduce the concept of interval-valued strong equality index and we present a construction method for such
indexes using implication functions defined with respect to total orders. We finish with some conclusions and references.

2. Preliminaries

In this section, we introduce several well known notions and results which are necessary for our subsequent develop-
ments. We are going to consider closed subintervals of the unit interval [0, 1]. In this sense, we denote:

L ([0, 1]) = {[ X , X ] | 0 ≤ X ≤ X ≤ 1}.

We will use capital letters to denote elements in L ([0, 1]). Another key notion in this work is that of order relation.
We recall here its definition, adapted for the case of L ([0, 1]).

Definition 2.1. An order relation on L ([0, 1]) is a binary relation ≤ on L ([0, 1]) such that, for all X , Y , Z ∈ L ([0, 1]),

(i) X ≤ X , (reflexivity),
(ii) X ≤ Y and Y ≤ X imply X = Y , (antisymmetry),
(iii) X ≤ Y and Y ≤ Z imply X ≤ Z , (transitivity).

An order relation on L ([0, 1]) is called total or linear if any two elements of L ([0, 1]) are comparable, i.e., if for every
X , Y ∈ L ([0, 1]), X ≤ Y or Y ≤ X . An order relation on L ([0, 1]) is partial if it is not total.
 H. Zapata et al. / International Journal of Approximate Reasoning 88 (2017) 91–109 93

We will denote by  L the partial order relation on L ([0, 1]) induced by the usual partial order in R2 , that is:

[ X , X ]  L [Y , Y ] if X ≤ Y and X ≤ Y . (2)

This is the order relation most widely used in the literature [20].
We denote by ≤ L any order in L ([0, 1]) (which can be partial or total) with 0 L = [0, 0] as its minimal element (that is,
0 L ≤ L X for all X ∈ L ([0, 1])) and 1 L = [1, 1] as its maximal element (that is, X ≤ L 1 L for all X ∈ L ([0, 1])). To denote a total
order in L ([0, 1]) with 0 L and 1 L as minimal and maximal elements, respectively, we use the notation ≤ T L .

Example 2.2.

1. A total order on L ([0, 1]) is, for example, the Xu and Yager’s order (see [32]):

X + X < Y + Y or
[ X , X ] ≤ X Y [Y , Y ] if (3)
X + X = Y + Y and X − X ≤ Y − Y .

This definition of Xu and Yager’s order was originally provided for Atanassov intuitionistic fuzzy pairs [32].
2. Another example of total order is provided by the lexicographical orders with respect to the first variable, ≤lex1 and
with respect to the second variable, ≤lex2 , which are defined, respectively, by:

X < Y or
[ X , X ] ≤lex1 [Y , Y ] if
X = Y and X ≤ Y ,

X < Y or
[ X , X ] ≤lex2 [Y , Y ] if
X = Y and X ≤ Y .

Regarding total orders in L ([0, 1]), we are going to consider the so-called admissible orders, whose definition we recall
now.

Definition 2.3. [16] An admissible order on L ([0, 1]) is a total order ≤ T L in L ([0, 1]) such that it refines the classical partial
order  L . That is, for every X , Y ∈ L ([0, 1]), if X  L Y then X ≤ T L Y .

Note that, for any admissible order, 1 L is its maximal element and 0 L is its minimal element.
An interesting feature of admissible orders is that they can be built using aggregation functions, as stated in the following
result.

Proposition 2.4. ([16]) Let M 1 , M 2 : [0, 1]2 −→ [0, 1] be two aggregation functions (i.e., increasing functions with M 1 (0, 0) =
M 2 (0, 0) = 0 and M 1 (1, 1) = M 2 (1, 1) = 1, see [27]) such that for all X , Y ∈ L ([0, 1]), the equalities M 1 ( X , X ) = M 1 (Y , Y ) and
M 2 ( X , X ) = M 2 (Y , Y ) can only hold simultaneously if X = Y .
The order ≤ M 1 , M 2 on L ([0, 1]) given by

M 1 ( X , X ) < M 1 (Y , Y ) or
X ≤ M 1 , M 2 Y if
M 1 ( X , X ) = M 1 (Y , Y ) and M 2 ( X , X ) ≤ M 2 (Y , Y )

is an admissible order on L ([0, 1]).

Example 2.5.

x+ y
1. Xu and Yager’s order is an example of admissible order with M 1 (x, y ) = 2 and M 2 (x, y ) = y.
2. The lexicographical orders ≤lex1 (≤lex2 ) are also examples of admissible orders with M 1 (x, y ) = x (M 1 (x, y ) = y) and
M 2 (x, y ) = y (M 2 (x, y ) = x).
3. More generally, if, for α ∈ [0, 1] we define the aggregation function

K α (x, y ) = α x + (1 − α ) y

(which is the weighted mean), then, for α , β ∈ [0, 1] with α = β , we can obtain an admissible order ≤α β just taking
M 1 (x, y ) = K α (x, y ) and M 2 (x, y ) = K β (x, y ). See [16] for more details.

From now on, whenever we speak of a total order in L ([0, 1]), we mean an admissible order in L ([0, 1]).
94 H. Zapata et al. / International Journal of Approximate Reasoning 88 (2017) 91–109

2.1. Interval-valued aggregation functions with respect to a partial order

The definition of aggregation function has been extended to the interval-valued setting with respect to the order  L in
a straightforward way [24].

Definition 2.6. Let n ≥ 2. An (n-dimensional) interval-valued (IV) aggregation function in L ([0, 1]) with respect to  L is a
mapping M : ( L ([0, 1]))n → L ([0, 1]) which verifies:

(i) M (0 L , · · · , 0 L ) = 0 L .
(ii) M (1 L , · · · , 1 L ) = 1 L .
(iii) M is an increasing function with respect to  L .

Remark 2.7. Note that this definition does not fully recover the usefulness of the usual definition of aggregation functions
in the real setting since there may exist intervals which are not comparable by means of the order  L , so the full meaning
of monotonicity is lost.

Example 2.8. ([28]) If A : [0, 1]2 → [0, 1] is an aggregation function, then the function M A : L ([0, 1])2 → L ([0, 1]) given by

M A ([ X , X ], [Y , Y ]) = [ A ( X , Y ), A ( X , Y )],

is an IV aggregation function in L ([0, 1]) with respect to the order  L .

Moreover, if A , B : [0, 1]2 → [0, 1] are two aggregation functions such that A (x, y ) ≤ B (x, y ) for each x, y ∈ [0, 1], then

M A , B ([ X , X ], [Y , Y ]) = [ A ( X , Y ), B ( X , Y )],

is an IV aggregation function in L ([0, 1]) with respect to the order  L .

Example 2.9. The following functions are IV aggregation functions in L ([0, 1]) with respect to the order  L .

• M ([ X , X ], [Y , Y ]) = [( X Y ) p , ( X Y ) p ] for every p > 0,

• M ([ X , X ], [Y , Y ]) = [ X Y 1/2 , ( X + Y )/2].

2.2. Interval-valued implication functions with respect to a partial order

In the literature, the definition of an implication in the IV setting has been provided with respect to the partial order  L
(cf. [7,9–11,30]). This definition reads as follows.

Definition 2.10. An interval-valued (IV) implication function in L ([0, 1]) with respect to  L is a function I : ( L ([0, 1]))2 →
L ([0, 1]) which verifies the following properties:

(i) I is a decreasing function in the first component and an increasing function in the second component with respect to
the order  L .
(ii) I (0 L , 0 L ) = I (1 L , 1 L ) = 1 L .
(iii) I (1 L , 0 L ) = 0 L .

An IV implication function can satisfy the following relevant properties, for which we follow the notations in [15] for the
properties that do not have a specific name:

(I4) I ( X , Y ) = 0 L ⇔ X = 1 L and Y = 0 L .
(I5) I ( X , Y ) = 1 L ⇔ X = 0 L or Y = 1 L .
(N P ) Left-neutrality: I (1 L , Y ) = Y for all Y ∈ L ([0, 1]).
(E P ) Exchange principle: I ( X , I (Y , Z )) = I (Y , I ( X , Z )) for all X , Y , Z ∈ L ([0, 1]).
(O P ) Ordering property: I ( X , Y ) = 1 L ⇔ X  L Y .
(S N) Strong negation principle: N ( X ) = I ( X , 0 L ) is a strong IV negation for all X ∈ L ([0, 1]) (for the notion of IV-negation
see Definition 3.1).
(I10) Y  L I ( X , Y ) for all X , Y ∈ L ([0, 1]).
(I P ) Identity principle: I ( X , X ) = 1 L for all X ∈ L ([0, 1]).
(C P ) Contrapositivity I ( X , Y ) = I ( N (Y ), N ( X )) for all X , Y ∈ L ([0, 1]), where N is an IV negation.
(I14) I ( X , N ( X )) = N ( X ) for all X ∈ L ([0, 1]), where N is an IV negation.
 H. Zapata et al. / International Journal of Approximate Reasoning 88 (2017) 91–109 95

Remark 2.11. The definition of IV implication functions has never been done in the literature for a total order. However, in
the real case, in order to consider monotonicity (i.e., decreasingness in the first variable and increasingness in the second
variable), the usual order between real numbers is considered, and this is clearly a total order. So, for fully translating the
notion of monotonicity to the interval-valued setting, a total order must be considered.
Furthermore, note that the ordering property, the strong negation principle and property I10, in general, only make full
sense as long as any two intervals in L ([0, 1]) are comparable, i.e., as long as we are dealing with a total (in our case,
admissible) order in L ([0, 1]).

3. Interval-valued negations

The notion of negation may be extended to the interval-valued setting when an arbitrary order is considered as follows.

Definition 3.1. Let ≤ L be an order relation in L ([0, 1]). A function N : L ([0, 1]) → L ([0, 1]) is an interval-valued negation
function (IV negation) with respect to ≤ L if it is a decreasing function with respect to the order ≤ L such that N (0 L ) = 1 L
and N (1 L ) = 0 L . A negation N is called strong negation if N ( N ( X )) = X for every X ∈ L ([0, 1]).

Note that, if N is a strong IV negation, then for every Y ∈ L ([0, 1]) there exists X ∈ L ([0, 1]) such that N ( X ) = Y , since it
is enough to take X = N (Y ).
As an example of results which are already known for negations in L ([0, 1]) we recall the following proposition.

Proposition 3.2. ([12, Theorem 2]) A function N : L ([0, 1]) → L ([0, 1]) is a strong negation with respect to  L if and only if there
exists a strong negation n : [0, 1] → [0, 1] (i.e., a decreasing function such that n(0) = 1, n(1) = 0 and n(n(x)) = x for every x ∈ [0, 1]),
such that

N ( X ) = [n( X ), n( X )]

for every X ∈ L ([0, 1]).

Remark 3.3. Note that, in general, Proposition 3.2 does not hold if N is not assumed to be a strong negation. For instance,
take X 0 in L ([0, 1]) with X 0 = X 0 . Then, the function:
⎧
⎪
⎨1 L if X = 0 L ;
N(X) = 0L if X = 1 L ;
⎪
⎩
X0 in other case,

is an IV negation (which is not strong) with respect to  L (in fact, with respect to any order relation ≤ L ). However, it does
not exist a negation n : [0, 1] → [0, 1] such that N ( X ) = [n( X ), n( X )].

Example 3.4.

X+X X−X
1. Consider Xu and Yager’s order ≤ X Y . Given X = [ X , X ] ∈ L ([0, 1]), denote c = 2
,r= 2
and α = ∧(c , 1 − c ). Consider
the function

c = 1 − c,
N ( X ) = [c − r , c + r ] with
r =α −r .

Then N is a strong IV negation with respect to Xu and Yager’s order.

2. Let n : [0, 1] → [0, 1] be a negation. Then, the function:

1L if [ X , X ] = 0 L ,
N ([ X , X ] =
[0, n( X )] in other case,

is an example of IV negation with respect to the order ≤lex2 . Analogously, the function

0L if [ X , X ] = 1 L ,
N ([ X , X ] =
[n( X ), 1] in other case,

is an example of IV negation with respect to the order ≤lex1 . Note that these are not strong IV negations.

The extension to the interval-valued setting of negations when we must work with total orders is not straightforward.
96 H. Zapata et al. / International Journal of Approximate Reasoning 88 (2017) 91–109

Proposition 3.5. Let n : [0, 1] → [0, 1] be a strong negation. Let N : L ([0, 1]) → L ([0, 1]) be the function defined by

N ([ X , X ]) = [n( X ), n( X )] .
If N is an IV negation with respect to ≤ L in L ([0, 1]), then ≤ L is not a total order.

Proof. Let x, y ∈ [0, 1] be such that x = y, x ≤ n(x) and y ≤ n( y ). Suppose that ≤ L is a total order. Then X = [x, n(x)] and
Y = [ y , n( y )] are comparable. Without loss of generality we can assume that X ≤ L Y . If N is a negation with respect to
the order ≤ L then, by antitonicity, N (Y ) ≤ L N ( X ). But, N ( X ) = [n(n(x)), n(x)] = [x, n(x)] = X and, analogously, N (Y ) = Y .
Therefore X = Y , which is a contradiction. Hence ≤ L is not a total order. 2

Remark 3.6. From Proposition 3.5, it follows that an analogous of Proposition 3.2 can not hold whatever the total order
≤ T L is.

A deep study on IV negations with respect to admissible orders can be found in [3].

4. Interval-valued implication functions with respect to a total order

As stated in Remark 2.11, in order to fully translate into the interval-valued setting the notion of monotonicity as it
is considered in the real setting, it is necessary to introduce a total order in the definition of an implication. As we have
already said, in this work we consider the specific case of admissible orders. In this way, the definition of an interval-valued
implication function with respect to a total order ≤ T L reads as follows.

Definition 4.1. Let ≤ T L be a total order in L ([0, 1]). An interval-valued (IV) implication function in L ([0, 1]) with respect to
≤ T L is a function I : ( L ([0, 1]))2 → L ([0, 1]) which verifies the following properties:

(i) I is a decreasing function in the first component and an increasing function in the second component with respect to
the order ≤ T L .
(ii) I (0 L , 0 L ) = I (1 L , 1 L ) = 1 L .
(iii) I (1 L , 0 L ) = 0 L .

Formally, this definition is the same as the one discussed in Definition 2.10, which is reasonable. However, now the
properties:

(O P ) Ordering property: I ( X , Y ) = 1 L ⇔ X ≤ T L Y .
(S N) Strong negation principle: N ( X ) = I ( X , 0 L ) is a strong IV negation for all X ∈ L ([0, 1]).
(I10) I ( X , Y ) ≥ T L Y for all X , Y ∈ L ([0, 1]).

make full sense, since any two intervals in L ([0, 1] are now comparable. Furthermore, also property

(C P ) Contrapositivity I ( X , Y ) = I ( N (Y ), N ( X )) for all X , Y ∈ L ([0, 1]), where N is an IV negation.

can be fully exploited in this setting, as strong negations also rely heavily on the notion of monotonicity.

Example 4.2. Given an order ≤ L , and analogously to the [0, 1] case, it is possible to provide some straightforward examples
of IV implication functions with respect to ≤ L . For instance, the mapping I : L ([0, 1])2 → L ([0, 1]) given by

0L if X = 1 L and Y = 0 L ,
I(X, Y ) =
1L in other case,

is an example of IV implication function for ≤ L . In fact, and analogously to the [0, 1] case, this is the greatest IV implication
function.

Of course, this example is trivial. In the next section, we discuss general methods to build IV implication functions. We
focus on the case of admissible orders.

5. Construction of IV implication functions in L ([0, 1])

The construction of IV implication functions with respect to an arbitrary order ≤ L directly from its definition can be
rather difficult. However, analogously to the [0, 1] case, it is well-known that IV implication functions can be defined in
terms of IV aggregation functions and IV negations, as in the case of usual (real-valued) implication functions [28]. For this
 H. Zapata et al. / International Journal of Approximate Reasoning 88 (2017) 91–109 97

reason, we start extending to the interval-valued setting with a total order the definition of IV aggregation function, which
formally reads the same as the one for a partial order.

Definition 5.1. Let ≤ T L be a total order in L ([0, 1]) and let n ≥ 2. An (n-dimensional) interval-valued (IV) aggregation func-
tion in L ([0, 1]) with respect to ≤ T L is a mapping M : ( L ([0, 1]))n → L ([0, 1]) which verifies:

(i) M (0 L , · · · , 0 L ) = 0 L .
(ii) M (1 L , · · · , 1 L ) = 1 L .
(iii) M is an increasing function in each component with respect to ≤ T L .

Remark 5.2. Note that, from the definition of admissible order, every IV aggregation function with respect to an admissible
order is also an IV aggregation function with respect to the partial order  L .

Example 5.3.

• Given a total order ≤ T L , the ∧ (minimum) and the ∨ (maximum) (with respect to ≤ T L ) are aggregation functions with
respect to this total order ≤ T L .
• Let p > 0. Consider the function M ([ X , X ], [Y , Y ]) = [( X Y ) p , ( X Y ) p ]. Then, M is an IV aggregation function with respect
to the lexicographical order ≤lex2 .

Proposition 5.4. The function M : L ([0, 1])2 → L ([0, 1]) given by:
 
XY + XY XY + XY
M ([ X , X ], [Y , Y ]) = , (4)
2 2

is an IV aggregation function with respect to Xu and Yager’s order. Furthermore, this function verifies that

M ([ X , X ], [Y , Y ]) = [0, 0] if and only if [ X , X ] = [0, 0] or [Y , Y ] = [0, 0]

and

M ([ X , X ], [Y , Y ]) = [1, 1] if and only if [ X , X ] = [Y , Y ] = [1, 1].

Proof. The proof follows from a straightforward calculation. 2

Remark 5.5. Example 5.3 and Proposition 5.4 provides the first example in the literature of non-trivial IV aggregation func-
tions with respect to a total order.

With the definition of an IV aggregation function at hand, we can provide the construction explicited in the following
Proposition, whose proof we also include for the sake of completeness. Note that the method works properly regardless of
whether we are considering a partial or a total order.

Proposition 5.6. Let ≤ L be an order in L ([0, 1]). Let M : L ([0, 1])2 → L ([0, 1]) be an IV aggregation function such that

M (1 L , 0 L ) = M (0 L , 1 L ) = 0 L
and let N be an IV negation in L ([0, 1]), both with respect to the same order ≤ L .
Then the function I M : L ([0, 1])2 → L ([0, 1]) given by

I M ( X , Y ) = N ( M ( X , N (Y )))
is an IV implication function with respect to ≤ L .

Proof. (i) Since M is an increasing function and N is a decreasing function with respect to ≤ L , it is clear that I M is a
decreasing function in the first component and an increasing function in the second component.
(ii) I M (0 L , 0 L ) = I M (0 L , 1 L ) = I M (1 L , 1 L ) = 1 L .
(iii) I M (1 L , 0 L ) = 0 L . 2

Theorem 5.7. In the same setting of Proposition 5.6, with N a strong IV negation, the following statements hold.

(i) If I M verifies EP and M ( X , 1 L ) = X for all X ∈ L ([0, 1]), then M is a symmetric function.
(ii) If M ( X , Y ) ≤ L Y for all X , Y ∈ L ([0, 1]), then I M verifies the property I10.
(iii) If M is a symmetric function and I M verifies NP: I M (1 L , X ) = X for all X ∈ L ([0, 1]), then I M verifies SN.
98 H. Zapata et al. / International Journal of Approximate Reasoning 88 (2017) 91–109

(iv) I M verifies NP if and only if M (1 L , Y ) = Y for all Y ∈ L ([0, 1]).

(v) If I M verifies OP, then I M verifies IP.

Proof. (i)

N ( M ( X , Y )) = N ( M ( X , M (Y , 1 L ))) = I M ( X , N ( M (Y , 1 L ))) = I M ( X , I M (Y , 0 L ))
= I M (Y , I M ( X , 0 L )) = N ( M (Y , N ( I M ( X , 0 L )))) = N ( M (Y , M ( X , 1 L )))
= N ( M (Y , X )).
(ii) Since M ( X , N (Y )) ≤ L N (Y ), then

I M ( X , Y ) = N ( M ( X , N (Y ))) ≥ L N ( N (Y )) = Y .
(iii) Since I M (1 L , X ) = X and I M (1 L , N ( X )) = N ( X ) for every X ∈ L ([0, 1]), it follows that M (1 L , X ) = X , and thus

I M ( X , 0 L ) = N ( M ( X , 1 L )) = N ( M (1 L , X )) = N ( X ).
Therefore, I M ( X , 0 L ) is a strong negation.
(iv) (⇒) Since I M (1 L , Y ) = Y for all Y ∈ L ([0, 1]) and there exists X ∈ L ([0, 1]) such that Y = N ( X ), then M (1 L , X ) = X .
(⇐) If M (1 L , Y ) = Y for all Y ∈ L ([0, 1]) then, since for all X ∈ L ([0, 1]) there exists Y ∈ L ([0, 1]) such that X = N (Y ),
we have

I M (1 L , X ) = I M (1 L , N (Y )) = N ( M (1 L , Y )) = N (Y ) = X .
(v) The proof is straightforward. 2

Note that it is possible to build IV implication functions with respect to an order ≤ L just modifying a bit the procedure
in Proposition 5.6.

Proposition 5.8. Let M : L ([0, 1])2 → L ([0, 1]) be an IV aggregation function such that M (1 L , 0 L ) = M (0 L , 1 L ) = 1 L and let N be an
IV negation in L ([0, 1]), both with respect to the same order ≤ L , which can be partial or total. Then the function I M : L ([0, 1])2 →
L ([0, 1]) given by

I M ( X , Y ) = M ( N ( X ), Y )
is an IV implication function.

Proof. It follows from a straightforward calculation analogous to that of Proposition 5.6. 2

Observe that, if N is a strong negation, these two procedures are closely related.

Proposition 5.9. Let N : L ([0, 1]) → L ([0, 1]) be an IV strong negation with respect to an order ≤ L . Then for every IV aggregation
function (with respect to the same ≤ L ) such that M (1 L , 0 L ) = M (0 L , 1 L ) = 0 L , it holds that

I M ( X , Y ) = I MN ( X , Y )
for every X , Y ∈ L ([0, 1]), where M N ( X , Y ) = N ( M ( N ( X ), N (Y ))) is the IV aggregation function dual to M with respect to N.

Proof. The calculation follows the same guidelines as the usual fuzzy case. Note that:

I M N ( X , Y ) = M N ( N ( X ), Y ) = N ( M ( N ( N ( X )), N (Y ))) = N ( M ( X , N (Y )))

so the result follows. 2

Example 5.10. We consider in L ([0, 1]) the order  L . Take the IV aggregation function M ( X , Y ) = [ X Y , X Y ] and the IV
negation N ( X ) = [1 − X , 1 − X ], so

I M ( X , Y ) = [1 − X + X Y , 1 − X + X Y ] = [ I R ( X , Y ), I R ( X , Y )] ,
where I R denotes the Reichenbach implication. Then I M is an IV implication which also verifies the properties:

• P 1 : I M ( X , Y ) = 0L ⇔ X = 1L and Y = 0L ,
• P 2 : I M ( X , Y ) = 1L ⇔ X = 0L or Y = 1L .
 H. Zapata et al. / International Journal of Approximate Reasoning 88 (2017) 91–109 99

Example 5.11. Let us consider the order  L . Let I M ( X , Y ) = N ( M ( X , N (Y ))), where M : L ([0, 1])2 → L ([0, 1]) is the IV ag-
gregation function (see [1])
⎧
⎪
⎪ 2XY 2XY
, if X + Y = 0
⎪
⎪ X +Y
, X +Y
⎪
⎨
M( X , Y ) = 2XY
⎪
⎪ 0, , if X + Y = 0 and X + Y = 0
⎪
⎪
X +Y
⎪
⎩
0L , if X = Y = 0 L .
Let N be a strong IV negation. Then, I M is an IV implication which also verifies the property I M ( X , N ( X )) = N ( X ) because
I M ( X , N ( X )) = N ( M ( X , X )) = N ( X ).
Moreover, I M verifies P 1 and P 2 .
However, it is straightforward that properties NP, EP, OP, SN, I10 and IP do not hold.

If we make use of an IV aggregation function to build an IV implication function, some of the properties of the former
translate into the fulfillment of some properties by the latter. For instance.

Theorem 5.12. Let M : L ([0, 1])2 → L ([0, 1]) be a symmetric IV aggregation function (i.e., such that M ( X , Y ) = M (Y , X ) for every
X , Y ∈ L ([0, 1])) such that M (1 L , 0 L ) = 0 L and let N be an IV strong negation in L ([0, 1]), both with respect to the same order ≤ L ,
which can be total or partial. Then the following properties are equivalent:

(i) I M verifies EP.

(ii) M is associative.

Proof. (i ) ⇒ (ii ) For any X , Y , Z ∈ L ([0, 1]) we have that:

N ( M ( X , M (Y , Z ))) = N ( M ( M ( Z , Y ), X )) = I M ( M ( Z , Y ), N ( X )) = I M ( X , N ( M ( Z , Y )))
= I M ( X , I M ( Z , N (Y ))) = I M ( Z , I M ( X , N (Y )))= N ( M ( Z , N ( I M ( X , N (Y )))))
Note, regarding the third identity above, that
I M ( N (Y ), N ( X )) = N ( M ( N (Y ), X )) = N ( M ( X , N (Y ))) = I M ( X , Y ) .
Since N is strong
N ( I M ( X , N (Y ))) = M ( X , Y )
so it holds
N ( M ( X , M (Y , Z ))) = N ( M ( Z , M ( X , Y ))) .
Since N is strong, it is bijective and therefore, from the symmetry, M is associative.
(ii ) ⇒ (i )
I M ( X , I M (Y , Z )) = N ( M ( X , M (Y , N ( Z )))) = N ( M ( X , M ( N ( Z ), Y )))
= N ( M ( M ( X , N ( Z )), Y )) = N ( M (Y , M ( X , N ( Z )))) = I M (Y , I M ( X , Z )). 2

Let us introduce now the definition of an IV triangular norm (t-norm) with respect to an order ≤ L .

Definition 5.13. An IV t-norm in L ([0, 1]) with respect to an order ≤ L is a mapping T : L ([0, 1])2 → L ([0, 1]) such that:

1. T is symmetric.
2. T is increasing with respect to ≤ L .
3. T ( X , 1 L ) = T (1 L , X ) = X for every X ∈ L ([0, 1]).
4. T is associative, i.e., T ( X , T (Y , Z )) = T ( T ( X , Y ), Z ) for every X , Y , Z ∈ L ([0, 1]).

Note that an IV t-norm is a special case of an IV aggregation function.

Example 5.14. Given an order ≤ L in L ([0, 1]), the function

T ( X , Y ) = ∧( X , Y )
where the minimum is calculated with respect to ≤ L is an example of IV t-norm (with respect to ≤ L ).
100 H. Zapata et al. / International Journal of Approximate Reasoning 88 (2017) 91–109

With this concept at our disposal, we have the following Corollary of Theorem 5.12.

Corollary 5.15. Let T : L ([0, 1])2 → L ([0, 1]) be an IV t-norm and let N be an IV strong negation in L ([0, 1]), both with respect to the
same order ≤ L , which can be total or partial. Then I T ( X , Y ) = N ( T ( X , N (Y ))) is an IV implication function such that I T verifies EP.

Proof. It follows straightforwardly from Theorem 5.12 and the fact that IV t-norms are, in particular, IV aggregation func-
tions. 2

Note that this IV implication functions are analogous to (N,T)-implication in the [0, 1]-case.
The IV implication functions defined by aggregation functions and negations do not verify, in general, some properties;
in particular, and due to its relevance from the point of view of applications, OP. For this reason we consider now another
construction method for IV implication functions.

Proposition 5.16. Let ≤ T L be a total order in L ([0, 1]), and let N be an IV negation function with respect to that order. The function
I : L ([0, 1])2 → L ([0, 1]) defined by

1L , if X ≤ T L Y ,
I(X, Y ) =
∨( N ( X ), Y ), if X >T L Y ,
is an IV implication function.

Proof. It is clear that the function I is an increasing function in the second component and a decreasing function in the
first component. Moreover
I (0 L , 0 L ) = I (1 L , 1 L ) = 1 L and I (1 L , 0 L ) = 0 L . 2

Remark 5.17. Note that in this way we recover Kleene–Dienes implication in the interval-valued setting for any order ≤ L .
Furthermore, this is analogous to a particular instance of (S,N)-implication in the [0, 1]-case.

This result can be further generalized as follows.

Proposition 5.18. Let ≤ T L be a total order in L ([0, 1]), and let N be an IV negation function with respect to that order. If
M : L ([0, 1])2 → L ([0, 1]) is an IV aggregation function, then the function I : L ([0, 1])2 → L ([0, 1]) defined by

1L , if X ≤ T L Y ,
I(X, Y ) =
M ( N ( X ), Y ), if X > T L Y ,

is an IV implication function.

Observe that now Proposition 5.16 is just an example of Proposition 5.18.

Another way to obtain IV implication function is by residuation. Given a t-norm, the definition of residual function is
well known in the fuzzy case. We extend it to the interval-valued case, considering a total order.

Definition 5.19. Let ≤ T L be a total order in L ([0, 1]) and let T be an IV t-norm with respect to the same order. The residual
operation associated to T is the function I R T : L ([0, 1])2 → L ([0, 1]) given by
I R T ( X , Y ) = sup { Z ∈ L ([0, 1]) | T ( X , Z ) ≤ T L Y }.

Example 5.20. Let ≤ T L be any total order in L ([0, 1]) and take T = ∧ with respect to that order. Then

1L if X ≤ T L Y
I RT (X, Y ) =
Y if X > T L Y

is the corresponding residual implication.

The residual operation associated with an IV t-norm T is an implication function. We have, in fact, the following result:

Theorem 5.21. [21] Let ≤ T L be a total order in L ([0, 1]) and let T be an IV t-norm with respect to the same order. Then the residual
operator I R T is an implication function.

Proof. It follows straightforwardly from the fact that T ( X , 0 L ) = 0 L for every X ∈ ([0, 1]). 2
 H. Zapata et al. / International Journal of Approximate Reasoning 88 (2017) 91–109 101

6. The case of Xu and Yager and the lexicographical orders

In the previous section we have discussed construction methods for IV implications which can be applied when we are
dealing, in general, with any partial or total order. In order to provide more specific examples, in this section we focus
on the two main examples of admissible order that can be found in the literature: namely, Xu and Yager’s order and the
lexicographical orders. We start with the former. Our idea is to build IV implication functions using appropriate IV negations
and aggregation functions. For this reason, we start constructing the negation, and later we build the aggregation function.

6.1. Construction of IV implication functions with respect to Xu and Yager’s order

We start discussing the construction of IV implication functions for Xu and Yager’s order. Our first step is to provide a
new IV negation with respect to this order.

Theorem 6.1. The function N X Y : L ([0, 1]) → L ([0, 1]) defined by

⎧
⎪
⎨ N X Y ([0, 0]) = [1, 1],
1− X
N X Y ([ X , X ]) = [ 2 , 1 − X2 ], if [0, 0] =
 [X, X] =
 [1, 1], (5)
⎪
⎩
N X Y ([1, 1]) = [0, 0],
is an IV negation with respect to the order ≤ X Y .

Proof. It is clear that if X + X < Y + Y , then

1− X X 3 − (X + X) 3 − (Y + Y ) 1−Y Y
+1− = > = +1−
2 2 2 2 2 2
and if X + X = Y + Y and X − X ≤ Y − Y , then

2− X 1− X 1 − (X − X) 1 − (Y − Y ) 2 − Y − (1 − Y )
− = ≥ =
2 2 2 2 2
and hence [ X , X ] ≤ X Y [Y , Y ] ⇒ N X Y ([ X , X ]) ≥ X Y N X Y ([Y , Y ]). 2

Remark 6.2. The negation given by Equation (5), N X Y , is not a strong negation since
 
1+ X 2+ X
N X Y ( N X Y ([ X , X ])) = , for [0, 0] =
 [X, X] =
 [1, 1] .
4 4

We provide now an example of IV aggregation function with respect to Xu and Yager’s order.

Proposition 6.3. For all α ∈ [0, 1] the function

M α ([ X , X ], [Y , Y ]) = [α X + (1 − α )Y , α X + (1 − α )Y ]
is an IV aggregation function with respect to the order ≤ X Y .

Proof. If [ X , X ] ≤ X Y [Y , Y ], then we have that

M α ([ Z , Z ], [ X , X ]) ≤ X Y M α ([ Z , Z ], [Y , Y ]).
Indeed, we have

M α ([ Z , Z ], [ X , X ]) = [α Z + (1 − α ) X , α Z + (1 − α ) X ],
M α ([ Z , Z ], [Y , Y ]) = [α Z + (1 − α )Y , α Z + (1 − α )Y ].
If X + X < Y + Y , then

α ( Z + Z ) + (1 − α )( X + X ) < α ( Z + Z ) + (1 − α )(Y + Y ).
If X + X = Y + Y and X − X ≤ Y − Y then

α ( Z + Z ) + (1 − α )( X + X ) = α ( Z + Z ) + (1 − α )(Y + Y )
and
α ( Z − Z ) + (1 − α )( X − X ) ≤ α ( Z − Z ) + (1 − α )(Y − Y ).
Similarly it can be proved that M α ([ X , X ], [ Z , Z ]) ≤ X Y M α ([Y , Y ], [ Z , Z ]). 2
102 H. Zapata et al. / International Journal of Approximate Reasoning 88 (2017) 91–109

Remark 6.4. Note that an analogous calculation shows that, if we take α1 , . . . , αn ∈]0, 1] with α1 + · · · + αn = 1, then the
function M : L ([0, 1])n → L ([0, 1]) given by

M ( X 1 . . . , X n ) = [α1 X 1 + · · · + αn X n , α1 X 1 + · · · + αn X n )]

is an IV aggregation function with respect to Xu and Yager’s order.

The function defined in Proposition 6.3 does not verify that M α (0 L , 1 L ) = 0 L = M α (1 L , 0 L ). Thus, in order to define the
∗ as follows.
implication as in Proposition 5.6, we redefine the function M α

Corollary 6.5. Let α ∈ [0, 1]. Then, the function


∗ [α X + (1 − α )Y , α X + (1 − α )Y ], if [ X , X ] = 0 L = [Y , Y ]
M α ([ X , X ], [Y , Y ]) =
0L , if [ X , X ] = 0 L or [Y , Y ] = 0 L

is an IV aggregation function with respect to Xu and Yager’s order.

Proof. It is straightforward from Proposition 6.3. 2

In this way, we are able to build an example of IV implication function with respect to Xu and Yager’s order.

Proposition 6.6. Let α ∈]0, 1[. The function

⎧
⎪
⎪[0, 0] if X = 1 L and Y = 0 L
⎪
⎪
⎨[1, 1] if X = 0 L or Y = 1 L
I α ([ X , X ], [Y , Y ]) =
⎪
⎪[ 12 − α X +(21−α ) , 2−α X −(
2
1 −α )
] if Y = 0 L and 0 L = X = 1 L
⎪
⎪
⎩ 2α (1− X )+(1−α )(1+Y ) 2α (2− X )+(1−α )(2+Y )
4
, 4
in other case

is an IV implication function with respect to Xu and Yager’s order.

∗ defined in Corol-
Proof. It is enough to note that I α is built using Proposition 5.6 with the IV aggregation function M α
lary 6.5 and the IV negation defined in Theorem 6.1. 2

Remark 6.7. Note that this is the first example of a non-trivial IV implication function with respect to a total order that can
be found in the literature.

6.2. Construction of IV implication functions with respect to the lexicographical orders

In the same way as in the previous subsection, we intend now to build IV implication functions for the lexicographical
orders, building first appropriate IV negations and IV aggregation functions. However, this work is easier now, since we can
take advantage of our developments for the Xu and Yager’s case.

Theorem 6.8. Let N X Y be the function defined in Eq. (5). Then N X Y , is also an IV negation with respect to the lexicographical orders

1− X 2− X
Proof. Since N X Y ([ X , X ]) = [ 2
, 2 ] we have

1− X 1− Y
1. if X < Y , then 2
> 2
,
1− X 1− Y
2. if X = Y and X ≤ Y , then 2
= 2
and 2−2 X ≥ 2−2 Y .

Therefore

[ X , X ] ≤ Lex1 [Y , Y ] ⇒ N X Y ([ X , X ]) ≥ Lex1 N X Y ([Y , Y ]).

In a similar way,

[ X , X ] ≤ Lex2 [Y , Y ] ⇒ N X Y ([ X , X ]) ≥ Lex2 N X Y ([Y , Y ]). 2

 H. Zapata et al. / International Journal of Approximate Reasoning 88 (2017) 91–109 103

Analogously, we have the following results.

Proposition 6.9. Let α ∈ [0, 1]. Let M α be the IV function defined in Proposition 6.3. Then, M α is also an IV aggregation function with
respect to the orders ≤ Lex1 and ≤ Lex2 .

Proof. If [ X , X ] ≤ Lex1 [Y , Y ], then we have

M α ([ Z , Z ], [ X , X ]) ≤ Lex1 M α ([ Z , Z ], [Y , Y ]) .
Indeed, we obtain

M α ([ Z , Z ], [ X , X ]) = [α Z + (1 − α ) X , α Z + (1 − α ) X ],
M α ([ Z , Z ], [Y , Y ]) = [α Z + (1 − α )Y , α Z + (1 − α )Y ].
If X < Y , then α Z + (1 − α ) X < α Z + (1 − α )Y .
If X = Y and X ≤ Y , then α Z + (1 − α ) X = α Z + (1 − α )Y and α Z + (1 − α ) X ≤ α Z + (1 − α )Y .
Similarly it can be proved that

M α ([ X , X ], [ Z , Z ]) ≤ Lex1 M α ([Y , Y ], [ Z , Z ])
The proof for ≤ Lex2 is similar. 2

So the following corollary is straightforward.

∗ defined in Corollary 6.5 is an IV aggregation function with respect to the

Corollary 6.10. Let α ∈ [0, 1]. Then, the IV function M α
lexicographical orders.

So finally we also have this result straightforwardly.

Proposition 6.11. Let α ∈]0, 1[. The function

⎧
⎪[0, 0]
⎪
⎪
if X = 1 L and Y = 0 L
⎪
⎨[1, 1] if X = 0 L or Y = 1 L
I α ([ X , X ], [Y , Y ]) = 1 α X +(1−α ) 2−α X −(1−α )
⎪[ 2
⎪
⎪
− 2
, 2
] if Y = 0 L and 0 L = X = 1 L
⎪
⎩ 2α (1− X )+(1−α )(1+Y ) 2α (2− X )+(1−α )(2+Y )
4
, 4
in other case

is an IV implication function with respect to the lexicographical orders.

Remark 6.12. Xu and Yager’s and lexicographical orders are particular cases of a more general class of admissible orders,
defined as follows: The results in this section can be extended to the K α ,β orders easily.

7. Generalized modus ponens in the interval-valued setting with a total order

It is well known that in approximate reasoning, fuzzy inference procedures are done using the generalized modus ponens
(GMP) [4]. In the fuzzy setting, the GMP inference rule is satisfied if there exists a continuous t-norm t : [0, 1]2 → [0, 1] and
an implication function i : [0, 1]2 → [0, 1] such that the inequality

t (x, i (x, y )) ≤ y
holds for every x, y ∈ [0, 1], which corresponds to the so-called GMP functional equation.
If we want to extend the GMP equation to the interval-valued setting, once an order ≤ L , an IV t-norm T and an IV
implication I have been fixed, it is clear that we must require that T ( X , I ( X , Y )) and Y are always comparable to each
other. In order to do so, the easiest way is to consider a total order.
With this consideration in mind, we propose the following extension of the GMP functional equation to the interval-
valued setting.

Definition 7.1. Let ≤ T L be a total order in L ([0, 1]). We say that an IV t-norm T and an IV implication function I (both with
respect to ≤ T L ) satisfy the generalized modus ponens (GMP) with respect to ≤ T L if

T ( X , I ( X , Y )) ≤ T L Y
for every X , Y ∈ L ([0, 1]).
104 H. Zapata et al. / International Journal of Approximate Reasoning 88 (2017) 91–109

As a first result, we can say the following.

Proposition 7.2. Let ≤ T L be a total order in L ([0, 1]). Let T be an IV t-norm, M be an idempotent IV aggregation function, i.e., M sat-
isfies that M ( X , X ) = X for all X ∈ L ([0, 1]) and N be an IV negation, all with respect to ≤ T L . Let I be the IV implication function
defined as in Proposition 5.8 and assume that the inequality

T ( X , I ( X , Y )) ≤ T L Y

holds for some fixed X , Y ∈ L ([0, 1]). Then, for every X ∈ L ([0, 1]) such that X ≤ T L X and N ( X ) ≤ T L Y , the inequality

T ( X , I ( X , Y )) ≤ T L Y

holds, too.

Proof. If X ≤ T L Y , then X ≤ T L Y and T ( X , I ( X , Y )) ≤ T L T ( X , 1 L ) = X ≤ T L Y .

If X > T L Y , then I ( X , Y ) = M ( N ( X ), Y ) ≤ T L M (Y , Y ) = Y and T ( X , I ( X , Y )) ≤ T L T ( X , Y ) ≤ T L Y . 2

Remark 7.3. In general, if Y < T L X and Y ≤ T L N ( X ), it is not always true that T ( X , I ( X , Y )) ≤ T L Y , even if M ( X , X ) ≤ T L X ,
as we can see in the following cases, where we consider T = ∧, Xu and Yager’s order and the strong negation N given by:
X+X X−X
If X = [ X , X ], c = 2
, r= α = ∧(c , 1 − c ),
2
and

c = 1 − c,
N ( X ) = [c − r , c + r ] with
r =α −r .

If Y = [1/2, 1/2] < X Y X = [1/3, 2/3] < X Y N ( X ) = [1/6, 5/6], then

(i) M ( X , Y ) = ∨( X , Y ).
If I ( X , Y ) = M ( N ( X ), Y ) = N ( X ) and T ( X , N ( X )) = X > T L Y .
X +Y
(ii) M ( X , Y ) = X +
2
Y
= 2
, X +Y
2
.
I ( X , Y ) = M ( N ( X ), Y ) = X = [1/3, 2/3] and T ( X , X ) = X > T L Y .

Clearly, the following result in [21] and [31] also hold:

Theorem 7.4. Let ≤ T L be a total order in L ([0, 1]). Let T be an IV t-norm and let I be an IV implication function, both with respect to
≤ T L . Then T and I verify the GMP with respect to ≤ T L if and only if I ( X , Y ) ≤ T L I R T ( X , Y ) for every X , Y ∈ L ([0, 1]).

Example 7.5. Let’s consider Xu and Yager’s order, ≤ X Y . Let N be any IV negation with respect to ≤ X Y and take T = ∧.
Consider the implication function
⎧
⎨1 L if X ≤ X Y Y
I(X, Y ) = N ( X )Y + N ( X )Y N ( X )Y + N ( X )Y
⎩ 2
, 2
in other case.

Then, if for X > X Y Y it holds that N ( X ) ≤ X Y Y , then the GMP functional equation holds.

Regarding the analysis of the verification of the GMP functional equation for some other specific cases, as S-implications,
the analysis in the interval-valued case when are considering total orders is much more complicate than in the real case,
since for the latter, characterization results depends on the continuity of the involved t-norms. For this reason, in future
works we intend to develop an analysis of the notion of continuity in L ([0, 1]) which allows us to obtain such characteriza-
tion results.

8. Preservation of the width of the intervals

In the fuzzy setting, the use of intervals in applications such as approximate reasoning is usually justified when the
experts have problems to provide exact numerical values for the considered membership functions [10]. In this sense,
the width of a given membership interval can be seen as a measure of the uncertainty or ignorance of the expert to
provide the exact value. From this point of view, it is reasonable to suppose that, in general, if we start with a given
uncertainty (measured by the width of the intervals that we are considering), this uncertainty can not decrease when we
make calculations, and we can only expect that it does not increase, at least, if all the initial intervals have the same width
(and hence, correspond to the same uncertainty) [5,23]. In particular, if we are dealing with an IV implication function I
 H. Zapata et al. / International Journal of Approximate Reasoning 88 (2017) 91–109 105

and if we take two intervals X , Y ∈ L ([0, 1]), we can not in principle expect that the width of the interval I ( X , Y ) is strictly
smaller than the width of the intervals X and Y , and the best situation we can expect is that uncertainty does not increase.
In this section, taking into account the previous discussion, we make a very first study on how the width of the inputs
intervals is affected by the IV implication function that we consider when all of them have the same width. This is a
situation that arises quite often, and to make our study, we will first analyze what happens for the different IV negations
and IV aggregation functions that we have defined, since they are the blocks which eventually lead to the IV implication
function. of course, this is only a first step in the analysis of those IV implication functions which do not lead to an increase
of uncertainty.
Note that, for Xu and Yager’s order, an interval is smaller than another one if the medium point of the former is smaller
than the medium point of the latter, or, if both medium points are the same, and if the width of the first interval is smaller
than the width of the second one. In general, any total order between intervals takes into account, in one way or another,
the width of the intervals.
We denote the width of an interval X ∈ L ([0, 1]) by w ( X ) = X − X .
Let us start with the case of negations.

Proposition 8.1. Let N X Y be the IV negation defined by Eq. 5. Then, N X Y preserves the width of intervals of width 1/3.

1− X 1− X 1−( X − X ) 1−1/3
Proof. If X = [ X , X ] with w ( X ) = 1/3, then N ( X ) = 2
, 1 − X2 and 1 − X
2
− 2
= 2
= 2
= 1/3. 2

We can consider a slightly modified version of this result.

Proposition 8.2. Let m, n ∈ N with 1 ≤ m ≤ n/2. Consider the function: N m,n : L ([0, 1]) → L ([0, 1]) defined by
⎧
⎪
⎪ N ([0, 0]) = [1, 1],
⎨ m,n
N m,n ([1, 1]) = [0, 0],
⎪
⎪
⎩ Nm,n ([ X , X ]) = m−m X , n−m X , if X ∈ L ([0, 1])\{[0, 0], [1, 1]}
n n

−m
Then, N m,n is an IV negation function with respect to Xu and Yager’s order, and preserves the width of intervals of width nn+m
.

m(1− X ) n−m−m nn−m

Proof. If X = [ X , X ] with w ( X ) = nn−m
+m , then N ( X ) = n
, 1 − mnX and 1 − mX
n
− m(1n− X ) = n−m−m( X − X )
n
= n
+m
=
n−m
n+m
. 2

Remark 8.3. For intervals of width greater than or equal to 1/3, we can find values of n and m which approximates the
width of intervals preserved by the negation N m,n to the desired width.
As 1 ≥ nn−m n λ−1 1+l
+m = l ≥ 1/3, let λ = m ≥ 2 and we have λ+1 = l, so λ = 1−l . Since l ∈ [0, 1] and rational numbers are dense,
p n q+ p
we can approximate l ∼ q
∈ Q and m
= q− p
, so n = q + p and m = q − p.

∗ be the IV aggregation function defined in Corollary 6.5. Then, M ∗ preserves the width of intervals of the same
Proposition 8.4. Let M α α
width.

Proof. If the width of the intervals X and Y is the same, X − X = Y − Y = l, then the width of the interval M ∗ ( X , Y ) is
α X + (1 − α )Y − (α X + (1 − α )Y ) = αl + (1 − α )l = l. 2

Corollary 8.5. Let X , Y ∈ L ([0, 1]) be two intervals such that w ( X ) = w (Y ). Let N be an IV negation with respect to some order ≤ L
and let M be a bivariate IV aggregation function with respect to the same order ≤ L . Assume that both N ( X ), N (Y ) and M ( X , Y ) are of
the same width as X and Y . Then, the IV implication functions I M and I M also preserve the width of X and Y ; that is, w ( I M ( X , Y )) =
w ( I M ( X , Y )) = w ( X ) = w (Y ).

Proof. It follows from the definitions I M ( X , Y ) = N ( M ( X , N (Y ))), I M ( X , Y ) = M ( N ( X ), Y ) and I α ( X , Y ) = N ( M ∗ ( X ,

N (Y ))). 2

The following corollary is straightforward.

Corollary 8.6. Under the same conditions of Corollary 8.5, if w ( X ) = w (Y ) = 1/3 and I α ( X , Y ) is defined as in Proposition 6.6, then
w ( I α ( X , Y )) = 1/3.
106 H. Zapata et al. / International Journal of Approximate Reasoning 88 (2017) 91–109

9. Interval-valued strong equality index

Let us consider a finite referential set of n elements, U = {u 1 , . . . , un }. We denote by I V F S (U ) the set of all interval-
valued fuzzy sets over U . Recall that, as we have already said, an interval-valued fuzzy set A over U is a mapping
A : U → L ([0, 1]) (see e.g. [14]). Note that an order ≤ L induces a partial order ≤ L in I V F S (U ) given, for A , B ∈ I V F S (U ),
by

A ≤ L B if A (u i ) ≤ L B (u i ) for every u i ∈ U .

This order extends Zadeh’s notion of inclusion between fuzzy sets, see [33]. However, a stronger notion of inclusion between
fuzzy sets was proposed by Dubois et al. [22]. In this section, we extend the latter notion to the interval-valued setting,
adding a property that ensures the prevalence of the width of the intervals.
First of all, let us define the notion of IV strong equality function.

Definition 9.1. Let ≤ L be an order in L ([0, 1]). An IV strong equality function is a mapping  : L ([0, 1])2 → L ([0, 1]) such
that:

1. ( X , Y ) = 1L if and only if X = Y ∈ {0L , 1L };

2. ( X , Y ) = 0L if and only if { X , Y } = {0L , 1L };
3. ( X , Y ) = (Y , X ) for every X , Y ∈ L ([0, 1]);
4. If X , Y ∈ L ([0, 1]) such that w ( X ) = w (Y ), then w (( X , Y )) = w ( X ).

Note that the fourth condition holds trivially in the fuzzy setting.

Example 9.2. The function

⎧
⎪
⎪1 if X = Y ∈ {0 L , 1 L }
⎨ L
( X , Y ) = 0L if { X L , Y L } = {0 L , 1 L }
⎪
⎪
⎩ X +Y
, X +Y
in other case,
2 2

is an IV strong equality function.

IV strong equality functions can be characterized in the following way (see [18] for an analogous result in the real case).

Theorem 9.3. Let  : L ([0, 1])2 → L ([0, 1]) be a mapping. The following statements are equivalent:

1.  is an IV strong equality function.

2. There exist two mappings M : L ([0, 1])2 → L ([0, 1]) and I : L ([0, 1])2 → L ([0, 1]) such that
• M is symmetric, M ( X , Y ) = 1 L if and only if X = Y = 1 L and M ( X , Y ) = 0L if and only if X = 0L or Y = 0L , and if w ( X ) =
w (Y ), then w ( M ( X , Y )) = w ( X ).
• I satisfies the properties P 1 , P 2 , and if w ( X ) = w (Y ), then w (I ( X , Y )) = w ( X )
such that

( X , Y ) = M (I ( X , Y ), I (Y , X )) for every X , Y ∈ L ([0, 1]) .

Proof. Assume that (1) holds, so  is an IV strong equality index. Define:

⎧
⎪
⎨0 L if X = 1 L and Y = 0 L
I( X, Y ) = 1L if X = 0 L or Y = 1 L
⎪
⎩
( X , Y ) in other case.

Take as M, for instance, the minimum ∧ with respect to any admissible order ≤ T L . Then clearly M satisfies all the conditions
in (2) of the Theorem. Let us see that I satisfies properties P 1 , P 2 and that if w ( X ) = w (Y ), then w (I ( X , Y )) = w ( X ).
Property P 2 : Assume first that I ( X , Y ) = 1 L . Then there are two possibilities:

(i) X = 0 L or Y = 1 L ;
(ii) ( X , Y ) = 1 L . But in this case, since we are assuming that  is an IV strong equality function, it follows that X = Y ∈
{0L , 1L }, and we should be in fact in the previous case.

The converse implication in property P 2 is obvious.

 H. Zapata et al. / International Journal of Approximate Reasoning 88 (2017) 91–109 107

Property P 1 : Assume that I ( X , Y ) = 0 L . Then, again, there are two possibilities:

(i) X = 1 L and Y = 0 L ;
(ii) ( X , Y ) = 0 L . But in this case, since we are assuming that  is an IV strong equality function, it follows that X = 1 L
and Y = 0 L , or X = 0 L and Y = 1 L , which can not happen unless we are in the previous case.

The fact that w (I ( X , Y )) = w ( X ) for every X , Y ∈ L ([0, 1]) such that w ( X ) = w (Y ) is straightforward from the equality
w (0 L ) = w (1 L ) and the fact that since  is a strong equality function then if w ( X ) = w (Y ), w (( X , Y )) = w ( X ).
Finally, from the choice of M and the construction of I , it is straightforward that

( X , Y ) = M (I ( X , Y ), I (Y , X )) for every X , Y ∈ L ([0, 1]) .

So we have (2). The fact that (2) implies (1) is straightforward. 2

We now provide a specific, non-trivial example of IV strong equality function. To do so, we start with the following
result.

Proposition 9.4. For every β ∈]0, 1[, the function Iβ defined as:
⎧
⎪
⎨[0, 0] if X = 1 L and Y = 0 L
Iβ ([ X , X ], [Y , Y ]) = [1, 1] if X = 0 L or Y = 1 L
⎪
⎩ 
β X + (1 − β)Y , β X + (1 − β)Y in other case

satisfies the conditions in (2) of Theorem 9.3.

Proof. It is straightforward. 2

Note that we are not requiring that Iβ should be an IV implication, since we are not imposing any monotonicity condi-
tion. So we can now build a family of IV strong equality functions as follows.

Proposition 9.5. The function  : L ([0, 1])2 → L ([0, 1]) given by

( X , Y ) = M ∗1 (Iβ ( X , Y ), Iβ (Y , X ))
2

with M ∗1 defined as in Corollary 6.5 and Iβ defined as in Proposition 9.4, is an IV strong equality function.
2

Proof. It follows straightforwardly from Theorem 9.3. 2

Analogously, we can form another family of IV strong equality functions:

Proposition 9.6. The function  : L ([0, 1])2 → L ([0, 1]) given by

( X , Y ) = ∧(Iβ ( X , Y ), Iβ (Y , X ))
where ∧ denotes the minimum with respect to any total order ≤ T L and Iβ is defined as in Proposition 9.4, is an IV strong equality
function.

Proof. It follows from Theorem 9.3. 2

Now we introduce the concept of strong equality index in the interval-valued setting.

Definition 9.7. A function S : I V F S (U )2 → L ([0, 1]) is an interval-valued (IV) strong equality index if:

1. S ( A , B ) = 1L if and only if A = B and A (u i ), B (u i ) ∈ {0L , 1L } for every u i ∈ U .

2. S ( A , B ) = 0L if and only if, for every u i ∈ U , A (u i ) = 1L and B (u i ) = 0L , or A (u i ) = 0L and B (u i ) = 1L .
3. S ( A , B ) = S ( B , A ) for every A , B ∈ I V F S (U ).
4. If w ( A (u i )) = w ( B (u i )) = l for every u i ∈ U , then w (S ( A , B )) = l.

Then we have the following construction result.

108 H. Zapata et al. / International Journal of Approximate Reasoning 88 (2017) 91–109

Theorem 9.8. Let M : L ([0, 1])n → L ([0, 1]) be a mapping such that M ( X 1 , . . . , X n ) = 0 L if and only if X 1 = · · · = X n = 0 L ,
M ( X 1 , . . . , X n ) = 1 L if and only if X 1 = · · · = X n = 1 L and if w ( X i ) = w ( X j ) for 1 ≤ i , j ≤ n then w ( M ( X 1 , . . . , X n )) = w ( X 1 ).
Let  : L ([0, 1])2 → L ([0, 1]) be an IV strong equality function. Then, the function S : I V F S (U )2 → L ([0, 1]) given by:

S ( A , B ) = M (( A (u 1 ), B (u 1 )), . . . , ( A (un ), B (un )))

is an IV strong equality index.

Proof. It is straightforward. 2

Combining Theorems 9.3 and 9.8, we get the following Corollary.

Corollary 9.9. Let M : L ([0, 1])n → L ([0, 1]) be a mapping such that M ( X 1 , . . . , X n ) = 0 L if and only if X 1 = · · · = X n = 0 L ,
M ( X 1 , . . . , X n ) = 1 L if and only if X 1 = · · · = X n = 1 L and if w ( X i ) = w ( X j ) for 1 ≤ i , j ≤ n then w ( M ( X 1 , . . . , X n )) = w ( X 1 ).
Let I : L ([0, 1])2 → L ([0, 1]) be a mapping which fulfills properties P 1 , P 2 and if w ( X ) = w (Y ), then w (I ( X , Y )) = w ( X ). Then, the
function S : I V F S (U )2 → L ([0, 1]) given by:

S ( A , B ) = M (∧(I ( A (u 1 ), B (u 1 )), I ( B (u 1 ), A (u 1 ))), . . . , ∧(I ( A (un ), B (un )), I ( B (un ), A (un )))
where ∧ denotes the minimum with respect to any total order ≤ T L , is an IV strong equality index.

Now we can provide a family of interval-valued strong equality indexes.

Proposition 9.10. The function S : I V F S (U )2 → L ([0, 1]) given by

 
1 1
n n
S ( A, B ) = M ∗1 (Iβ ( A (u i ), B (u i )), Iβ ( B (u i ), A (u i ))), M ∗1 (Iβ ( A (u i ), B (u i )), Iβ ( B (u i ), A (u i )))
n 2 n 2
i =1 i =1

with M ∗1 defined as in Corollary 6.5 and Iβ defined as in Proposition 9.4, is an IV strong equality index.
2

Proof. The result follows from Proposition 9.5 and from Theorem 9.8 taking as M the mapping:
 
1 1
n n
M ( X 1 , . . . , Xn ) = Xi , Xi 2
n n
i =1 i =1

10. Conclusions

In this paper we have defined, for the first time in the literature, interval-valued implication functions with respect to a
total order. This definition has allowed us to extend in a proper way some well-known results from real-valued implication
functions. The use of total orders between intervals in the definition of interval-valued implication functions and in the
definition of interval-valued t-norms has allowed us to extend the GMP to the interval-valued setting. Moreover, we have
also presented the definition of strong equality index between interval-valued fuzzy sets, where we have considered the
properties usually required to these indexes, but also, for the interval-valued case, the preservation of the widths of the
intervals.
As we have said in the introduction, the use of interval to represent the membership degree of the elements to the set
has two main advantages:
1.- The possibility of using total or partial orders;
2.- The information provided by the width of the intervals can be taken into account.
These two considerations have been taken into account in this work, whereas this is not usually the case in the literature
when interval-valued fuzzy sets are considered.
Although it would also be of interest to analyze some other types of IV implication functions, as D-implications and
QL-implications, we leave this study for a future work, since it would result too long for the present paper. We also leave
for future works a comprehensive study of IV triangular norms with respect to a total order.

Acknowledgements

This work was partially supported by the Spanish Ministry of Science and Technology under projects TIN2016-77356-P
and TIN2014-59543-P, by grant VEGA 1/0420/15 and by the Brazilian funding agency CNPq under processes 306970/2013-9
and 307681/2012-2, and by Caixa and Fundacion Caja Navarra of Spain. The work on this paper for M. Baczyński was
supported by the National Science Centre, Poland, under Grant No. 2015/19/B/ST6/03259.
 H. Zapata et al. / International Journal of Approximate Reasoning 88 (2017) 91–109 109

References

[1] J. Aczél, On Applications and Theory of Functional Equations, Birkhäuser, Basel, Stuttgart, 1969.
[2] C. Alsina, E. Trillas, When ( S , N )-implications are ( T , T 1 )-conditional functions?, Fuzzy Sets Syst. 134 (2003) 305–310.
[3] M.J. Asiaín, H. Bustince, R. Mesiar, A. Kolesárová, Z. Takáč, Negations with respect to admissible orders in the interval-valued fuzzy set theory, IEEE
Trans. Fuzzy Syst. (2017), http://dx.doi.org/10.1109/TFUZZ.2017.2686372, in press.
[4] M. Baczyński, B. Jayaram, Fuzzy Implications, Studies in Fuzziness and Soft Computing, vol. 231, Springer, Berlin, 2008.
[5] J.F. Baldwin, B.W. Pilsworth, Axiomatic approach to implication for approximate reasoning with fuzzy logic, Fuzzy Sets Syst. 3 (1980) 193–219.
[6] W. Bandler, L. Kohout, Fuzzy power sets and fuzzy implication operators, Fuzzy Sets Syst. 4 (1980) 13–30.
[7] B. Bedregal, G. Dimuro, R. Santiago, R. Reiser, On interval fuzzy S-implications, Inf. Sci. 180 (8) (2010) 1373–1389.
[8] B. De Baets, E. Kerre, M. Gupta, The fundamentals of fuzzy mathematical morphology 1. Basic concepts, Int. J. Gen. Syst. 23 (2) (1995) 155–171.
[9] B. Bedregal, G. Dimuro, R. Reiser, An approach to interval-valued R-implications and automorphisms, in: Proceedings of IFSA-EUSFLAT 2009, 2009,
pp. 1–6.
[10] H. Bustince, Indicator of inclusion grade for interval-valued fuzzy sets. Application to approximate reasoning based on interval-valued fuzzy sets, Int. J.
Approx. Reason. 23 (3) (2000) 137–209.
[11] H. Bustince, E. Barrenechea, V. Mohedano, Intuitionistic fuzzy implication operators. An expression and main properties, Int. J. Uncertain. Fuzziness
Knowl.-Based Syst. 12 (3) (2004) 387–406.
[12] H. Bustince, E. Barrenechea, M. Pagola, Generations of interval-valued fuzzy and Atanassov’s intuitionistic fuzzy connectives and from K α operators:
laws for conjunctions and disjunctions, amplitude, Int. J. Intell. Syst. 23 (2008) 680–714.
[13] H. Bustince, E. Barrenechea, M. Pagola, Relationship between restricted dissimilarity functions, restricted equivalence functions and normal
E N -functions: image thresholding invariant, Pattern Recognit. Lett. 29 (4) (2008) 525–536.
[14] H. Bustince, E. Barrenechea, M. Pagola, J. Fernández, Z. Xu, B. Bedregal, J. Montero, H. Hagras, F. Herrera, B. De Baets, A historical account of types of
fuzzy sets and their relationship, IEEE Trans. Fuzzy Syst. 24 (1) (2016) 179–194.
[15] H. Bustince, P. Burillo, F. Soria, Automorphisms, negations and implication operators, Fuzzy Sets Syst. 134 (2003) 209–229.
[16] H. Bustince, J. Fernandez, A. Kolesárová, R. Mesiar, Generation of linear orders for intervals by means of aggregation functions, Fuzzy Sets Syst. 220
(2013) 69–77.
[17] H. Bustince, M. Galar, B. Bedregal, A. Kolesárová, R. Mesiar, A new approach to interval-valued Choquet integrals and the problem of ordering in
interval-valued fuzzy set applications, IEEE Trans. Fuzzy Syst. 21 (6) (2013) 1150–1162.
[18] H. Bustince, J. Fernandez, J. Sanz, M. Baczyński, R. Mesiar, Construction of strong equality index from implication operators, Fuzzy Sets Syst. 211 (2013)
15–33.
[19] H. Bustince, V. Mohedano, E. Barrenechea, M. Pagola, Definition and construction of fuzzy DI-subsethood measures, Inf. Sci. 176 (21) (2006) 3190–3231.
[20] C. Cornelis, G. Deschrijver, E.E. Kerre, Implication in intuitionistic fuzzy and interval-valued fuzzy set theory: construction, classification, application,
Int. J. Approx. Reason. 35 (1) (2004) 55–95.
[21] B. De Baets, J. Fodor, Residual operators of uninorms, Soft Comput. 3 (1999) 89–100.
[22] D. Dubois, H. Prade, Fuzzy Sets and Systems: Theory and Applications, Academic Press, New York, 1980.
[23] S. Fukami, M. Mizumoto, K. Tanaka, Some considerations on fuzzy conditional inference, Fuzzy Sets Syst. 4 (1980) 243–273.
[24] M. Komorníková, R. Mesiar, Aggregation functions on bounded partially ordered sets and their classification, Fuzzy Sets Syst. 175 (1) (2011) 48–56.
[25] M. Mas, M. Monserrat, J. Torrens, Modus ponens and modus tollens in discrete implications, Int. J. Approx. Reason. 49 (2008) 422–435.
[26] M. Mas, M. Monserrat, J. Torrens, E. Trillas, A survey on fuzzy implication functions, IEEE Trans. Fuzzy Syst. 15 (6) (2007) 1107–1121.
[27] R. Mesiar, M. Komorníková, Aggregation functions on bounded posets, in: C. Cornelis, et al. (Eds.), 35 Years of Fuzzy Sets Theory, in: Studies in Fuzziness
and Soft Computing, vol. 261, Springer, Berlin, 2010, pp. 3–17.
[28] A. Pradera, G. Beliakov, H. Bustince, B. De Baets, A review of the relationship between implication, negation and aggregation functions from the point
of view of material implication, Inf. Sci. 329 (2016) 357–380.
[29] R. Sambuc, Φ -Flous, Application a láide au Diagnostic en Pathologie Thyroidienne, Thése de Doctorát en Médicine, Univ. Marseille, Marseille, France,
1975.
[30] R.H.S. Reiser, G.P. Dimuro, B.C. Bedregal, R.H.N. Santiago, Interval valued QL-implications, in: Logic, Language, Information and Computation, in: Lecture
Notes in Computer Science, Springer, Berlin, 2007, pp. 307–321.
[31] E. Trillas, C. Alsina, A. Pradera, On MPT-implication functions for fuzzy logic, Rev. R. Acad. Cienc. Ser. A Mat. (RACSAM) 98 (2004) 259–271.
[32] Z.S. Xu, R.R. Yager, Some geometric aggregation operators based on intuitionistic fuzzy sets, Int. J. Gen. Syst. 35 (2006) 417–433.
[33] L.A. Zadeh, Fuzzy sets, Inf. Control 8 (1965) 338–353.
[34] L.A. Zadeh, Quantitative fuzzy semantics, Inf. Sci. 3 (1971) 159–176.
[35] L.A. Zadeh, The concept of a linguistic variable and its application to approximate reasoning, Inf. Sci. 8 (1975) 199–249.