67583268
67583268
com
OR CLICK HERE
DOWLOAD EBOOK
ebooknice.com
https://ebooknice.com/product/vagabond-vol-29-29-37511002
ebooknice.com
https://ebooknice.com/product/advances-in-insect-physiology-
vol-29-1308058
ebooknice.com
ebooknice.com
(Ebook) Advances in Child Development and Behavior 29 by
Robert V. Kail and Hayne W. Reese (Eds.) ISBN
9780120097296, 012009729X
https://ebooknice.com/product/advances-in-child-development-and-
behavior-29-1797872
ebooknice.com
https://ebooknice.com/product/29-single-and-nigerian-53599780
ebooknice.com
ebooknice.com
https://ebooknice.com/product/boeing-b-29-superfortress-1573658
ebooknice.com
Studies in Fuzziness and Soft Computing
Tofigh Allahviranloo
Soheil Salahshour Editors
Advances in
Fuzzy Integral
and Differential
Equations
Studies in Fuzziness and Soft Computing
Volume 412
Series Editor
Janusz Kacprzyk, Systems Research Institute, Polish Academy of Sciences,
Warsaw, Poland
The series “Studies in Fuzziness and Soft Computing” contains publications on
various topics in the area of soft computing, which include fuzzy sets, rough sets,
neural networks, evolutionary computation, probabilistic and evidential reasoning,
multi-valued logic, and related fields. The publications within “Studies in Fuzziness
and Soft Computing” are primarily monographs and edited volumes. They cover
significant recent developments in the field, both of a foundational and applicable
character. An important feature of the series is its short publication time and
world-wide distribution. This permits a rapid and broad dissemination of research
results.
Indexed by SCOPUS, DBLP, WTI Frankfurt eG, zbMATH, SCImago.
All books published in the series are submitted for consideration in Web of Science.
© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature
Switzerland AG 2022
This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether
the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse
of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and
transmission or information storage and retrieval, electronic adaptation, computer software, or by similar
or dissimilar methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, expressed or implied, with respect to the material contained herein or for any
errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.
This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Contents
v
vi Contents
Abstract This chapter presents the theory of differential and integral calculus for
autoregressive fuzzy processes, that is, fuzzy number-valued functions F such that
F(t) and F(t + h) are interactive fuzzy numbers for every |h| sufficiently small.
Recall that interactivity between fuzzy variables are described in terms of joint pos-
sibility distributions and plays a similar role to that of dependence between random
variables. We present a theory of differential calculus for certain autoregressive fuzzy
processes based on a special type of interactivity called F-interactivity. Next, we
introduce a Banach space, denoted by RF(A) , which is composed by fuzzy numbers
that are interactive with each other, such that the vector addition corresponds to an
interactive operation. The concepts of differential and integral for mappings from
R to RF(A) correspond to the classical notions of Fréchet derivative and Riemann
integral in Banach spaces. Thus, the theory of differential and integral calculus for
functions taking values in RF(A) is different from those found in fuzzy literature that
are based in (generalized) Hukuhara differentiability and fuzzy Aumann (or Rie-
mann) integral, which involve non-interactive arithmetic. In contrast to the usual
fuzzy Aumann integral, the diameter of a fuzzy function defined in terms of the
Riemann integral in RF(A) may decrease. We also study fuzzy ordinary and partial
differential equations in RF(A) .
1 Introduction
The first approach to calculus theory for fuzzy number-valued function was pro-
posed by Puri and Ralescu where they introduced the notion of fuzzy Hukuhara
differentiability or, for short, H -differentiability [34]. The concept of fuzzy Aumann
integrability was introduced by [35]. Subsequently, Kaleva established a fundamen-
tal theorem of calculus using these two concepts [27]. However, from a practical
point of view, the application of such a notion of differentiability is quite limited.
For example, let A be a fuzzy number and F(t) = g(t)A be such that g is a differ-
entiable in R. The fuzzy function F is not H -differentiable at t0 if |g| is a strictly
decreasing function at t0 . Stefanini and Bede proposed the concept of generalized
Hukuhara differentiability (or, for short, g H -differentiability) which extends the H -
differentiability and is the most common and well-known [47]. The g H -derivative
of the function F(t) = g(t)A at t0 is given by g (t0 )A which is an intuitive result.
One the other hand, when we are dealing with a fuzzy function F that describes
an uncertain process, it is reasonable to assume that the value of this process in a near
future F(t + h) is close (or related) to the value of the present F(t), even though the
quantities are inaccurate and/or uncertain. In addition, depending on the phenomenon
described by the fuzzy process F, one can expect that the level of uncertainty at
instant t + h should be reduced or increased in comparison to the instant t. In fuzzy
set theory, interactivity (or dependence) between F(t) and F(t + h) can be described
by the notion of joint possibility distributions. The concept of interactivity for fuzzy
variables is analogous to concept of dependence for random variables that arises from
joint probability distributions. Intuitively, according Zadeh, two fuzzy quantities “are
interactive if the assignment of a value to one affects the fuzzy restrictions placed
on the other” [52]. Once the interactivity between two fuzzy quantities is known,
we can perform arithmetic operations taking into account it. This approach leads
to the so-called interactive arithmetic (see Sect. 3). The g H -derivative is given in
terms of the g H -difference of F(t + h) and F(t) which, in its turn, implicitly takes
into account the existence of a type of interactivity described by a certain joint
possibility distribution [49]. Thus, the g H -difference can be viewed as an interactive
subtraction. If we assume other types of interactivity between F(t) and F(t + h),
then we obtain other interactive differences of F(t) and F(t + h) and, consequently,
other notions of differentiability for fuzzy functions F [6]. The theory of calculus
based on interactive arithmetic operations is called of interactive fuzzy calculus. In
this chapter, we focus on a type of interactivity between fuzzy numbers, namely, the
F-interactivity, which leads to the notion of F-interactive differentiability.
Despite the fact that g H -difference is a particular case of interactive fuzzy subtrac-
tion, many researchers consider the other arithmetic operations on fuzzy numbers,
namely, addition, multiplication, and division, as being the standard operations which
is an extension of interval arithmetic. Recall that, in this case, the sum of two uncer-
tain quantities given by fuzzy numbers never results in a real number. However, in
many situations and phenomena, we may have the sum of two fuzzy numbers (or
uncertain quantities) is a real number. For example, the weight of a pregnant woman
Differential and Integral Calculus … 5
is a real number but consist of the sum of the weights of the woman and her fetus
that are uncertain approximations.
In the context of fuzzy dynamical system, interactivity also arises naturally. A
damped harmonic oscillator may be governed by an equation of the form a X +
bX + cX = 0 where X, X , and X are uncertain quantities. Salgado
et al. studied these types of systems with fuzzy interactivity [38, 39]. Another
interesting dynamic system where the usual arithmetic can not be properly applied is
the compartmental epidemiological model without vital dynamics whose the initial
conditions and/or parameters are uncertain [11]. For example, in the susceptible-
infected (SI) model without vital dynamics, we have that the sum of the infected indi-
viduals with the successive ones is equal to the total population, that is, S + I = N .
Note that, from a practical point of view, it is reasonable to assume that the intial con-
ditions are uncertain, since it may be impossible to determine the number of infected
due to various factors such as false positives, the existence asymptomatic individ-
uous, impossibility to test all population at same time, etc. Let S and I be fuzzy
numbers, we have that the sum of S and I is equal to the real number N . To ensure
that the total population is a real number, the initial conditions must be interactive
fuzzy numbers. A similar situation appears in chemical reactions where at each stage
of the reaction we have that the mass of intermediate reagents is uncertain but the
total is known [50]. In [41], authors investigate some boundary value problems with
boundary conditions given by interactive fuzzy numbers. These examples suggest the
use of interactive arithmetic operations and, consequently, the interactive calculus to
model and study fuzzy process described by such fuzzy dynamical equations.
Given a fuzzy process F, determining g H -derivative of F at a point t may be
a hard task. Most of this difficulty lies in verifying the hypothesis of the Stack-
ing theorem. The calculation of the F-interactive derivative assumes that F(t) and
F(t + h) are F-interactive which may be difficult to verify in practice too. In a
further step, obtaining solutions for fuzzy differential equations given in terms of
g H -differentiability may also be a hard and costly task from both practical and ana-
lytical point of views. Recall that the calculation of such solutions may require the
analysis of types of g H -differentiability, existence of switch points [13], application
of Stacking theorem [32], and to solve two differential equations for each α ∈ [0, 1].
Again, in general, the verification of the hypothesis of Stacking theorem may be
difficult in practical problems. Similarly, solving fuzzy differential equations using
F-interactive differentiability is not a easy task as well. In contrast, the concept
A-linear interactivity, which is a special type of F-interactive relation with respect
to a fixed fuzzy number A, can be used to develop a simple and intuitive theory of
calculus for functions taking values in a certain subclass of fuzzy numbers denoted
by RF(A) . A fuzzy function from R to RF(A) is also called an A-linearly interactive
fuzzy process. The simplicity of such a theory of calculus is due to the fact the
space RF(A) forms a Banach space that isomorphic to R2 [21]. Roughly speaking,
the calculation of the derivative (or integral) of a A-linearly interactive fuzzy process
is given in terms of the classical derivatives (or integrals) of a certain pair of two
real functions. For example, the function F(t) = g(t)A, that we considered at the
beginning of Introduction, corresponds to an A-linearly interactive fuzzy process.
6 L. C. de Barros et al.
Using this theory of calculus, we have that for every t0 such that g (t0 ) exists, the
function F has derivative equal to g (t0 )A.
This chapter is organized as follows. Section 2 provides some mathematical foun-
dations used throughout this chapter. Section 3 presents the concept of interactivity
for fuzzy numbers and the arithmetic operations with interactive operands which
are the basis for the development of the interactive fuzzy calculus. A particular case
of interactive differential calculus for certain fuzzy functions, namely, F-interactive
fuzzy processes, is presented in Sect. 4. Section 5 deals with another particular case
of interactive calculus that is simpler to understand and use. Thus, beginner readers
can choose to read this section first, before the previous ones which deal with more
general cases.
2 Mathematical Background
This section reviews some basic concepts and notations from fuzzy set theory that
are used throughout this chapter.
A fuzzy (sub)set A of the universe X = ∅ is described by its membership function
μ A : X −→ [0, 1], where μ A (x) means the degree to which x belongs to A. For
notational convenience, we write A(x) instead of μ A (x). We use the symbol F(X )
to denote the set of all fuzzy subsets of X . The α-levels of the fuzzy subset A are
classical subsets defined as follows:
[A]α = {x ∈ X : A(x) ≥ α}
Definition 1 (Fuzzy number) [4, 20] The fuzzy subset A of R is a fuzzy number if
Proposition 1 [14] (RF , d∞ ) is a complete metric space and moreover, the following
properties hold true:
A = d∞ (A, 0).
where f −1 (y) stands for the preimage of y under f , that is, f −1 (y) = {x ∈
X | f (x) = y}. By definition, sup ∅ = 0. The Zadeh extension principle can also be
defined for functions with multiple arguments as follows. Let f : X 1 × . . . × X n →
Y and Ai ∈ F(X i ), i = 1, . . . , n. The Zadeh extension of f at (A1 , . . . , An ) is the
fuzzy set fˆ(A1 , . . . , An ) ∈ F(Y ) given by
The next theorem characterizes the α-levels of the Zadeh extension of a continuous
functions in terms of the α-levels of the corresponding arguments.
Theorem 1 [5, 33]
(a) If f : R → R is continuous and A ∈ RF , then fˆ(A) ∈ RF , and
The Zadeh extension principle can be used to define the standard arithmetic oper-
ations for fuzzy numbers as follows.
Definition 3 (Standard Arithmetic) Let A, B ∈ RF .
The next proposition reveals the space of fuzzy numbers with the standard arith-
metic and the scalar product does not satisfy the algebraic structure of a vector space.
1. A + B = A + B.
2. A + (B + C) = (A + B) + C.
3. A + 0 = A.
4. there exists D ∈ RF such that A + D = 0 iff A ∈ R.
5. ( p + q)A = p A + q A if pq ≥ 0.
6. p(A + B) = p A + p B.
7. ( pq)A = p(q A).
Note that if A − H B exits, then len[A]α > len[B]α , for all α ∈ [0, 1]. This nec-
essary condition brings a considerable restriction for the use of this fuzzy difference
in practical problems. The notion of generalized Hukuhara difference extends the
Hukuhara difference and does not require such conditions.
Definition 4 [47] Let A, B ∈ RF , the generalized Hukuhara difference (or, for short,
g H -difference) is the fuzzy number D, if it exists, such that
A=B+D
A −g H B = D ⇐⇒ ,
or B = A − D
where “+” and “−” represent, respectively, the standard addition and subtraction of
fuzzy numbers given by of the Zadeh’s extension principle.
If the g H -difference of two fuzzy number A and B exists, then [8]
[B −g H A]α = [min{bα− − aα− , bα+ − aα+ }, max{bα− − aα− , bα+ − aα+ }]. (3)
In order to compare the method described in Sect. 4 to solve FIVPs with the
method based on g H , we recall this notion of g H -differentiability.
Definition 5 [47] Let F : [a, b] → RF . We say that F is generalized Hukuhara
differentiable (g H - differentiable for short) at t0 ∈ [a, b] if there exists an element
Fg H (t0 ) ∈ RF , such that
1
Fg H (t0 ) = lim F(t0 + h) −g H F(t0 ) ,
h→0 h
where the limit is taken in the metric d∞ . At the end points, we consider only the
one-sided derivative.
ii. F is g H2 -differentiable at t0 , if
Fg H2 (t0 ) α
= [( f α+ ) (t0 ), ( f α− ) (t0 )], ∀α ∈ [0, 1].
10 L. C. de Barros et al.
The notion of fuzzy Aumann integral was introduced by Puri and Ralescu [35].
Fundamental theorems of calculus considering this notion of integrability and the
H -differentiability is given in [27].
Definition 7 (Aumann integral) [27, 35] Let F : [a, b] → RF be a fuzzy function.
b
The fuzzy Aumann integral of F over [a, b], denoted by (F A) a F(t)dt, is defined
by α-levels:
b b b
(F A) F(t)dt = (A) [F(t)]α dt = (A) [ f α− (t), f α+ (t)]dt
a α a a
b
= y(t)dt | y : [a, b] → R is a measurable selection of [F(·)]α ,
a
(4)
for all α ∈ [0, 1], provided that (4) defines a fuzzy numbers.
n
d∞ F(tk )(xk − xk−1 ), S < ,
k=1
The next proposition states that the notion of fuzzy Aumann integrability and of
fuzzy Riemann integrability coincide if the integrand function is continuous.
Fg H (t) = A, ∀t ∈ R
and
1 1 1
F(t)dt = f α− (t)dt, f α+ (t)dt
−1 α −1 −1
0 1 0 1
= taα+ dt + taα− dt, taα− dt + taα+ dt
−1 0 −1 0
1 −
= aα − aα+ , aα+ − aα− .
2
for all α ∈ [0, 1]. Thus, we obtain
1
1
F(t)dt = (A − A) = 0.
−1 2
The concept of interactivity between two fuzzy sets was proposed by Zadeh [52].
This concept plays an analogous role to the concept of dependence and independence
for random variables, where the joint probability distribution reveals whether there
is dependence or independence between the random variables. Similarly, the notion
of interactivity between two fuzzy variables arises from a given joint possibility (or
joint membership) distribution, where the uncertainty involved is given by means of
membership functions instead of probability density functions such as in the case of
random variables [15].
The operations between fuzzy numbers given by the Zadeh extension principle,
which is defined in terms of the minimum t-norm, can be generalized by considering
a general t-norms in Eq. (2), which, in its turn, is a particular case of a joint possibility
distribution. The concept of joint possibility distributions is inspired by the one of
joint probability distributions for random variables.
According to [12, 23], a fuzzy set J ∈ F(Rn ) is said to be a joint possibility
distribution (JPD) for A1 , . . . , An ∈ RF if each Ai is the projections of J , that is,
Pi (J ) = Ai , ∀i = 1, . . . , n
where
Pi (J )(y) = sup J (x), ∀y ∈ R.
x∈Rn ,xi =y
max J (x1 , x2 ) = A(x1 ) and max J (x1 , x2 ) = B(x2 ), for any x1 , x2 ∈ R. (8)
x2 x1
In this case, A and B are also called marginal possibility distributions of J . Figure 1
illustrates an arbitrary joint possibility distribution J with marginals possibility dis-
tributions A and B.
where f is the joint probability density function of the random variables A and B.
Differential and Integral Calculus … 13
A + J B = + J (A, B),
A − J B = − J (A, B),
A · J B = · J (A, B),
A ÷ J B = ÷ J (A, B),
J (x, y) = A(x)χ{(u,v); v=F (u)} (x, y) = B(y)χ{(u,v); v=F (u)} (x, y), (11)
where χ{(u,v):v=F (u)} denotes the characteristic function of the graph of the function
F. Since J satisfies P2 (J ) = B, one can show that (11) is equivalent to B = F̂(A),
that is
B(y) = sup μ A (z) (12)
z∈F −1 (y)
for all z ∈ R. Thus, if F is a continuous function, from Theorem 1, the α-levels are
given by
[B]α = F([A]α ), ∀α ∈ [0, 1].
and
⎧ −
⎨ bα − aα− , bα+ − aα+ , if F (x) ≥ 1
[B −F A]α = bα+ − aα+ , bα− − aα− , if 0 ≤ F (x) < 1 , ∀z ∈ [A]α . (16)
⎩ −
bα − aα+ , bα+ − aα− , if F (x) < 0
From Eq. (3), we can observe that the F-interactive difference B −F A coincides
to g H -difference B −g H A if F (x) ≥ 0 for all x ∈ [A]0 . This fact corroborates
to the fact that the g H -difference can be viewed as particular case of interactive
subtraction [49]. On the other hand, if F (x) < 0 for all x ∈ [A]0 , we have that
B −F A coincides with the standard difference. Similarly, from Proposition 2, if
F (x) ≥ 0 for all x ∈ [A]0 , then B +F A coincides with the standard sum B + A.
However, as we can note, B ·F A and B ÷F A are different from standard operations
in the all cases.
An interesting subtype of F-interactivity arise if we only consider functions F
of the form F(x) = q x + r . This restriction leads us to linear interactivity relation
among fuzzy numbers. This concept is formalized in the next definition.
Definition 11 [12] The A, B be fuzzy numbers are linearly interactive (or com-
pletely correlated) if there exists q, r ∈ R such that the corresponding joint possibility
distribution J of A and B is given by
[J ]α = {(x, q x + r ) | x ∈ [A]α }.
and ⎧ −
⎨ bα − aα− , bα+ − aα+ , if q ≥ 1
[B −L A]α = bα+ − aα+ , bα− − aα− , if 0 ≤ q < 1 .
⎩ −
bα − aα+ , bα+ − aα− , if q < 0
Using the concept of F-derivative, we can now consider fuzzy initial value prob-
lems (FIVPs) of the form
X F (t) = F(t, X (t))
, (17)
X (0) = X 0
where 0 < λ ∈ R represents the force of the imitative behavior in the market and
is called the coefficient of imitation. We are assuming that X is a locally F-
autocorrelated fuzzy process, that is,
for all α ∈ [0, 1] and for each t ∈ R, where Ft,h : R → R is a family of continuous
functions that represents the interactivity of the process at instants t and t + h, for
sufficiently small |h|.
Note that λX (t) and 1 − X (t) are L-interactive with L(x) = 1 + λx . Therefore,
we consider the L-interactivity arithmetic in the field F(t, X (t)) of FIVP (20). So,
[F(t, X (t))]α = [ f α− (t, xα− (t), xα+ (t)), f α+ (t, xα− (t), xα+ (t))].
• (Contractive market model) To consider the case where the uncertainty disap-
pears over time we must consider 0 < Ft,h (z) < 1. Although we do not know the
exact value of Ft,h (z), we just need to get it between 0 and 1. In this case, by
Theorem 3, for all α ∈ [0, 1], we have
[(xα+ ) (t), (xα− ) (t)] = [ f α− (t, xα− , xα+ ), f α+ (t, xα− , xα+ )] = [F(t, X )]α
− + . (21)
[X (0)]α = [x0α , x0α ]
• (Expansive market model) When we do not have much information about the
market it is reasonable to assume that the uncertainty in the solution increases over
time which corresponds to impose that Ft,h (z) > 1. In this case, by Theorem 3,
we have, for all α ∈ [0, 1]
[(xα− ) (t), (xα+ ) (t)] = [ f α− (t, xα− , xα+ ), f α+ (t, xα− , xα+ )] = [F(t, X )]α
− + . (22)
[X (0)]α = [x0α , x0α ]
A solution of the system (23) with λ = 0.9 and X 0 = (0.1; 0.27; 0.42) can be seen
in Fig. 3a.
• (via g H1 derivative) Solving the same problem via g H1 -deferenciability, we have
[(xα+ ) (t), (xα− ) (t)] = [xα+ (t)λ(1 − xα− (t)), xα− (t)λ(1 − xα+ (t))]
− + . (24)
[X (0)]α = [x0α , x0α ]
A solution of the system (24) with λ = 0.9 and X 0 = (0.1; 0.27; 0.42) can be
viewed in Fig. 3b.
In Fig. 3a, we can see that the fuzzy solution via g H2 -differentiability is nar-
rower than the solution via F-interactivity. This fact occurs due to the different
approaches used in the multiplication operation at λX (t) · (1 − X (t)). On the other
hand, in Fig. 3b, we can see that the fuzzy solution via g H1 -differentiability is wider
than that obtained via F-interactivity. In general, when using gH-differentiability,
the standard multiplication operation is used. In our opinion, this is inconsis-
tent, since non-interactivity is given through the Zadeh extension principle, and in
Differential and Integral Calculus … 21
Fig. 2 The continuous line stands the 1-level, whereas the dashed-dotted lines stand for the 0-level
of the solutions of systems (21) and (22), respectively. The parameters used were λ = 0.9, and
X 0 = (0.1; 0.27; 0.42)
Fig. 3 The continuous line stands the 1-level of solutions of the systems (21), (23), (22), and (24),
respectively, whereas the dashed lines stand for the 0-level of the solutions of systems (23) and (24),
respectively, and the dashed-dotted lines stand for the 0-level of the solutions of systems (21) and
(22), respectively. The parameters used were λ = 0.9, and X 0 = (0.1; 0.27; 0.42)
this case the only fuzzy process differentiable is the real function, i.e., F(t) ∈ R,
∀t ∈ [a, b]. So, we believe that the solution via F-interactivity is more consistent
than the one obtained via g H2 -differentiability.
F(t + h) “ − ”F(t)h
lim
h→0 .
Usual approaches to fuzzy calculus consider such a limit, however, the choice of the
subtraction “−” at the limit is not unanimous in fuzzy calculus theory.
The next section introduces another approach to fuzzy calculus that considers
Banach spaces composed by fuzzy numbers that are interactive with each other, so
that the vector addition corresponds to an interactive operation. The operations in
this space also correspond to the arithmetic operations for interactive fuzzy numbers
but a different way than that presented in this section. The concepts of differential
and integral in this space correspond to the classical notions of Fréchet derivative
and Riemann integral in Banach spaces.
We begin this section by considering a simple fuzzy function such as that of the
introductory section. Let g : R → R be a differentiable function and let F(t) =
g(t)A, A ∈ RF \ R. The fuzzy function F is not Hukuhara differentiable if |g(·)| is
a strictly decreasing function. Thus, Hukuhara differentiability may not be useful
in simple situations. In contrast, the function F is g H -differentiable [8] and F-
differentiable at t for g(t) = 0 (see Example 3). These derivatives coincide and are
given by g (t)A, which is very intuitive result. On the other hand, the usual notions of
fuzzy Aumann and Riemann integrals of the function F coincide since the function
F is continuous with respect to the metric d∞ . Despite the fact the g H/F-derivative
of F is given in terms of the derivative of g, usually, these fuzzy integrals are not
given in terms of the integral of the function g. For example, if g(t) = t, then we
have that fuzzy Aumann integral of F over [−1, 1] is given by 21 (A − A) = 0 (see
1
Example 1) and not by A −1 sds = 0. This last observation reveals that the same
intuitive result for the g H -derivative is not valid for usual notions of fuzzy integrals.
This example is one of the motivation to development a calculus theory for certain
fuzzy functions, based on the notions of Fréchet derivative and Riemann integral
on Banach spaces, in which the corresponding derivative and integral of F is given
in terms of the derivative and integral of g, respectively. The concepts and results
presented in this section can be found in [21, 43, 44].
This section deals with fuzzy functions of the form F(t) = q(t)A + r (t), where
A ∈ RF and q, r : R → R. For a fixed fuzzy number A, consider the operator ψ A :
R2 → RF that associates each vector (q, r ) ∈ R2 to the fuzzy number
Differential and Integral Calculus … 23
ψ A (q, r ) = q A + r, ∀q, r ∈ R.
We denote the range of the operator ψ A by the symbol RF(A) , that is,
Note that R can be embedded in RF(A) since every real number r can be identified
with the fuzzy number ψ A (0, r ) ∈ RF(A) . Thus, we have R ⊆ RF(A) . Moreover, one
can prove that if B ∈ RF(A) R, then RF(A) = RF(B) [21].
The next proposition establishes that the operator ψ A does not satisfies the linearity
property with respect to the standard addition.
[J ]α = {(q x + r, ux + v) | x ∈ [A]α }
for all α ∈ [0, 1]. Since every [J ]α is a non-empty, connected, and compact set of
R2 , from Theorem 2, we have that
for all α ∈ [0, 1]. Hence, the J -interactive sum B + J C is equal to ψ A (q + u, r + v).
The next corollary states that, for every non-symmetric fuzzy number, we can define
a Banach space that is isomorphic to the real Banach space (R2 , · ).
Corollary 4 [21] Let · be a norm of R2 and A ∈ RF be non-symmetric. The real
vector space (RF(A) , +ψ A , ·) forms a Banach space with the norm · ψ A given by
B ψA = ψ A−1 (B)
For each fuzzy function F : [a, b] −→ RF(A) , we can define a family of multi-
valued functions Fα : [a, b] → K given by Fα (t) = [F(t)]α , where K denotes the
family of compact subsets of R. That is,
ω(h) ψA
with lim = 0. At the end points, we consider only the one-sided of the
h→0 |h|
limit.
26 L. C. de Barros et al.
From Definition 14, one can show that if F is Fréchet differentiable at t then
F [t](h) = hC, where C = F [t](1).
The next theorem characterizes the Fréchet derivative of an A-linearly interactive
fuzzy process.
Theorem 4 [21] Let A ∈ RF be non-symmetric and F : [a, b] → RF(A) such that
F(t) = q(t)A + r , for all t ∈ [a, b]. The function F is Fréchet differentiable at t ∈
(a, b) if, and only if, p : [a, b] → R2 given by p(t) = (q(t), r (t)) is differentiable
at t. Additionally, f [t](h) = q (t)h A + r (t)h for all h ∈ R.
The Fréchet derivative of functions between Banach spaces satisfy some desired
properties that are stated in the next two propositions [21, 53].
Proposition 12 Let A ∈ RF be non-symmetric. If F : [a, b] −→ RF(A) is Fréchet
differentiable in [a, b], then F is continuous in [a, b].
Proposition 13 Let A ∈ RF be non-symmetric. If F, G : [a, b] −→ RF(A) are
Fréchet differentiable in [a, b], then (F +ψ A G) [t] = F [t] +ψ A G [t] and (λ ·
F) [t] = λ · F [t] for all t ∈ [a, b] and λ ∈ R.
Since (RF(A) , +ψ A , ·) is a Banach space, one can define a derivative of F at a
point t by means of a limit such as in calculus for real-valued functions [44]. More
precisely, let A ∈ RF be non-symmetric and F : [a, b] → RF(A) such that F(t) =
q(t)A + r (t). We say that F is ψ-differentiable or, simply, differentiable at t ∈ [a, b]
if there exists F (t) ∈ RF(A) such that
1
lim · (F(t + h) −ψ A F(t)) = F (t), (26)
h→0 h
in sense of dψ A . Again, we consider only the one-sided of the limit at the endpoints
a and b. We say that F is (ψ-)differentiable in [a, b] if F is (ψ-)differentiable at
every t ∈ [a, b]. As we can observe in the next theorem, the notions of Fréchet
differentiability and ψ-differentiability have a close relation.
Theorem 5 [19] A function F : [a, b] → RF(A) is F-differentiable at t ∈ [a, b] if,
and only if, F is ψ-differentiable at t. Moreover,
F (n) (t) = ψ A (q (n) (t), r (n) (t)) = q (n) (t)A + r (n) (t).
From comments above, we have that the function F(t) = g(t)A, that is one of
the motivations to the development of this theory of calculus on RF(A) , has derivative
given by F (t) = g (t)A if g is differentiable at t.
Since RF(A) forms a Banach space when A is non-symmetric, one can define inte-
gral for A-linearly interactive fuzzy process by means of the definition of Riemann
integral on Banach spaces [26].
Definition 15 (Riemann Integral in RF(A) ) [43] Let A be a non-symmetric fuzzy
number. A function F : [a, b] → RF(A) is said to be Riemann integrable if there
exists S ∈ RF(A) such that for every > 0 there exists δ > 0 such that for all partition
a = x0 < x1 < . . . < xn = b with xk − xk−1 < δ, k = 1, . . . , n, we have
n
F(tk )(xk − xk−1 ) −ψ A S <
k=1 ψA
F (t) = A + et , ∀t ∈ [−1, 1 ]
The next theorem can be viewed as a version of the mean value theorem for
integrals of the classical theory of calculus.
b
F(t)dt = (b − a)C,
a
is differentiable with
G (t) = F(t), ∀t ∈ [a, b].
The next example emphasizes the difference between the concepts of interactive
and non-interactive fuzzy integrals in light of Theorem 8.
30 L. C. de Barros et al.
Example 6 Let A be a non-symmetric fuzzy number and let F(t) = cos(t)A for
0 ≤ t ≤ π.
• Via interactive Riemann integral (Definition 15): From Theorem 6, we have
t
G 1 (t) = F(s)ds = sin(t)A, 0 ≤ t ≤ π.
0
In this case, the diameter of G 1 is strictly decreasing in π2 , π . Such a property
is not possible when we consider the fuzzy Aumann integral [27]. Moreover, the
function G 1 is not Hukuhara differentiable in π2 , π .
• Via non-interactive Riemann integral (Definition 8): Note that F is a continuous
function and, for all α ∈ [0, 1], we have
− +
π
[F(t)]α = cos(t)aα+ , cos(t)aα− , 0π ≤ t ≤ 2 .
cos(t)aα , cos(t)aα , 2 < t ≤ π
From Proposition 5, we have that the fuzzy Aumann integral and the non-interactive
fuzzy
Riemann
integral (given in Definition 8) coincide. Thus, by Eq. (5), for all
t ∈ 0, π2 and α ∈ [0, 1] we obtain
t t
[G 2 (t)]α = cos(s)aα− ds, cos(s)aα+ ds
0 0
= sin(t)aα− , sin(t)aα+ .
π
Now, for every t ∈ 2
, π and α ∈ [0, 1] we obtain [G 2 (t)]α =
π π
2 t 2 t
= cos(s)aα− ds + cos(s)aα+ ds, cos(s)aα+ ds + cos(s)aα− ds
π π
0 0
2 2
As we can note in the example above, the theory of calculus for A-linearly inter-
active fuzzy processes presented in this subsection is indeed different from that based
on Hukuhara derivative and Aumann integral [27, 34, 35].
F(t, X ) −ψ A F(t, Y ) ψA ≤ L X −ψ A Y ψA ,
32 L. C. de Barros et al.
for t ∈ [a, a + δ] and x, y ∈ Cρ . Then the FIVP (27) has a unique solution X (·)
defined in an interval [a, a + k], where k > 0 is sufficiently small, and X (t) is given
by Eq. (28) for all t ∈ [a, a + k].
In what follows, we present some methods to solve fuzzy linear differential equa-
tions in RF(A) .
which is the variation of the constants formula for two classic differential equations.
Thus, the solution (30) can be written by
t t s
( a(s) ds) (− t a(u) du)
X (t) = e t0
q0 + g1 (s)e 0 ds A
t0
t s
t
( a(s) ds) (− a(u) du)
+ e t0 r0 + g2 (s)e t0 ds .
t0
To ilustrate the formula (30) we return to the market model of Frank Bass.
Another way of examining the uncertainty in forecasting the number of consumers
of a product to be launched is to consider the Bass model (19) in space RF(A) , where
A is a non-symmetric fuzzy number. Hence,
Example 7 (Markets without feedback) This model is a market model that is not
concerned with competition, but with how the market evolves as a whole. Thus, in
this case, all suppliers are seen as a single supplier [3]. An example of this type of
model is the mobile and internet markets. The rate at which customers are signing
up for subscriptions is given by
where a ∈ R is the adaptation rate for the service and 1̃ = 1A + 1. The adaptation
rate is constant and is the same for all potential customers.
Note that, in this model, the desire to buy a product does not depend on how many
already own the product, that is, the market feedback does not exist or is so weak
that it can be neglected [3]. If X (t) = q(t)A + r (t), then, by Eq. (30), the solution
of (31) is given by
−at −at
X (t) = 1 − (1 − q0 )e A + 1 − (1 − r0 )e . (32)
Figure 4 exhibits the solution X of the system (31) given by Eq. (32) with a = 0.15,
q0 = 0.5, r0 = 0.3 and A = (0.1; 0.27; 0.42). We can see that the evolution of the
market (i.e. lim X (t)) is such that X (t) = A + 1 = 1̃.
t→∞
If X (t) = q(t)A + r (t), then, by Eq. (30), the solution of FIVP (33) is given by
−
a0
(1−e−γ t ) −
a0
(1−e−γ t )
γ γ
X (t) = 1 − e A+ 1−e . (34)
Figure 5 exhibits the solution X given as in Eq. (34) with a0 = 0.15, γ = 0.1, and
A = (0.1; 0.27; 0.42).
34 L. C. de Barros et al.
Fig. 4 Graphical representation of the fuzzy process X given as in Eq. (32) with a = 0.15, q0 = 0.5,
r0 = 0.3 and A = (0.1; 0.27; 0.42)
Fig. 5 Graphical representation of the fuzzy process X given as in Eq. (34) with a0 = 0.15, γ = 0.1,
and A = (0.1; 0.27; 0.42)
Differential and Integral Calculus … 35
where a, b, c ∈ R and X (t) ∈ RF(A) . Let us seek for solutions of the type X (t) =
q(t)A + r (t), where q(t) = r (t) = eγ t . Recall that the second order derivative of
X (t) = q(t)A + r (t) is given by X (t) = q (t)A + r (t). Thus, the fuzzy differen-
tial equation in FIVP (35), can be rewritten as follows
aγ 2 + bγ + c = 0. (36)
X (t) = emt (c1 cos nt + c2 sin nt)A + emt (d1 cos nt + d2 sin nt). (39)
36 L. C. de Barros et al.
where a, b, c ∈ R and X (t), G(t) ∈ RF(A) , for all t. Note that X 0 and X 0 are inter-
active [49].
Indeterminate Coefficient Method is a systematic way of determining the general
form of the particular solution X p (t) based on the nonhomogeneous term G(t) =
g1 (t)A + g2 (t) in (40).
This method basically consists of using the indeterminate coefficient method in
the following differential equations that are obtained considering a solution X of the
form X (t) = q(t)A + r (t). Thus, q and r are obtained by solving
where q(t) = qh (t) + q p (t) and r (t) = rh (t) + r p (t) with qh and rh being the homo-
geneous solutions and q p and r p being the particular solutions of (41) and (42),
respectively.
Next, we study the equilibrium in the market model.
Example 9 The development and launch of new products are characterized by the
lack of the necessary information of value and physical indicators for accurate pricing.
The statistical data on the demand for such a product in the market is very limiting
due to its novelty [30]. In this example, we will study the market equilibrium in
terms of fuzzy price function P : R → RF(A) . Market equilibrium occurs when the
quantity of demand
S(t) = − f +ψ A g · P(t),
occurs [48]. We will establish the time trajectory of the market price, assuming the
equilibrium in each instant t, that is, D(t) = S(t), ∀t ∈ R.
Simplifying the system, we obtain
To solve the linear differential equation (43) with constants coefficients, we must
first find the particular solution based on G(t) = 0 A − (a + f ). In this case, a par-
ticular solution is given by the function Pp (t) = 0 A + a+ f
b+g
.
Now, we need to find the homogeneous solution. To this end, we need the roots
of the characteristic function dγ 2 + cγ − (b + g) = 0.
1. If c2 > −4d(b + g), then we have two distinct real roots, say γ1 and γ2 . Thus,
by Eq. (37), the solution of homogeneous equation is
2. If c2 = −4d(b + g), then we have a single real root γ . Thus, by Eq. (38), the
solution of homogeneous equation is
Fig. 6 Graphical representation of fuzzy price function (P) given as in Eq. (44) with A =
(0.1; 0.27; 0.42)
The price fluctuation is the result of the assimilation of new information and
speculation from other markets. These can be broad in the beginning, as is the case
in Fig. 6 and then disappear over time, resulting in nothing. Or it can generate a
growing perception about the new price level as new information is assimilated,
which would be the case where the fluctuation grows over time.
Remark 4 Note that if d > 0, then −4d(b + g) is always negative and, conse-
quently, only the first case with two distinct real roots is possible. If c < 0, all three
cases are possible. If d < 0 in the second case, both roots will be negative. When we
have c, d < 0 in the second case, we have that the root is negative. In third case, this
condition produces a negative m as well.
In this section we present the fuzzy Laplace transform for A-linearly interactive
fuzzy processes, which was initially studied by [40]. The Laplace transform is given
+∞
by means of improper integrals of the form a F(t)dt which is defined as being
b
the limit of the definite integral a F(t)dt when b → ∞.
Differential and Integral Calculus … 39
where L denotes the usual Laplace transform for real-valued functions. The next
proposition states that the Laplace transform for A-linearly interactive fuzzy pro-
cesses is a linear integral operator.
Proposition 16 [40] Let F, G : [0, ∞[ → RF(A) be functions such that their fuzzy
Laplace transform exist and let c a real number, then:
L F(t) +ψ A cG(t) = L {F(t)} +ψ A cL {G(t)} .
Our website is not just a platform for buying books, but a bridge
connecting readers to the timeless values of culture and wisdom. With
an elegant, user-friendly interface and an intelligent search system,
we are committed to providing a quick and convenient shopping
experience. Additionally, our special promotions and home delivery
services ensure that you save time and fully enjoy the joy of reading.
ebooknice.com