1. Introduction
Fuzzy set theory is a robust tool for modeling uncertain problems, making it widely applicable to various natural phenomena. Specifically, fuzzy fractional differential equations (FFDEs) are commonly used in fields such as population modeling, weapon-system evaluation, civil engineering, and electro-hydraulics. The concept of fractional derivatives is crucial in fuzzy calculus, drawing significant attention from mathematicians and engineers.
The work on FFDEs was conducted by Agarwal et al. [
1], who defined the Riemann-Liouville differentiability concept under Hukuhara differentiability to solve these equations. Recently, fractional calculus has proven effective in addressing complex mathematical and engineering problems, including aerodynamics, control systems, signal processing, and biomathematics.
Numerous authors have studied FFDEs, employing various methods to solve them. For instance, Hoa explored FFDEs using Caputo gH-differentiability, while Agarwal et al. [
2] surveyed the topic to highlight its relevance to optimal control problems. Long et al. [
3] demonstrated the solvability of FFDEs, and Salahshour et al. [
4] applied fuzzy Laplace transforms to solve these equations.
The well-known theory of fuzzy sets is a very strong tool in mathematics, utilized for modeling problems involving uncertainty. It has been effectively applied to a widespread range of natural phenomena, which have been adoptly reformulated in fuzzy language. The well-known fuzzy fractional differential equation (FFDE), in particular, plays a pivotal role across various scientific disciplines, including but not limited to population models [
5], evaluation of the systems of weapon [
6], civil engineering [
7], and electro-hydraulic systems modeling [
8].
Fractional derivatives hold a significant place in fuzzy calculus, leading to the heightened interest in FFDEs among both mathematicians and engineers. In the early 1970s, Chang and Zadeh [
9] introduced a very interesting notion of derivatives using fuzzy set theory, which paved the way for extensive subsequent researches. The first formal study in FFDEs, however, was conducted in 2010 by Agarwal et al. [
2]. Indeed, they defined the differentiability of Riemann-Liouville in the context of the differentiability of Hukuhara and applied this novel approach to solve FFDEs. Since then, numerous researchers have delved into fractional calculus. This fascinating area of study has shown great applicability in solving complex problems in different fields such as biomathematics, aerodynamics, signal processing, and control systems (see, e.g., [
2,
10,
11,
12,
13]).
In fuzzy mathematics, fractional differential equations (FDEs) have been extensively explored, with significant contributions from many authors, including the second author of this paper (see, e.g., [
14,
15,
16,
17,
18,
19]). Notable works include Salahshour, Allahviranloo, and Abbasbandy’s utilization of the fuzzy Laplace transform for solving FFDEs in 2012 [
4], Hoa’s study of FFDEs through Caputo gH-differentiability in 2015 [
20], and Long, Son and Tam’s demonstration of the solvability of FFDEs in 2017 [
3]. More recently, in 2018, Agarwal et al. [
1] conducted a survey on FFDEs, revealing their connection to a unique type of optimal control problems, specifically nonlocal evolution optimal control equations.
There are various numerical methods for reformulating the FFDEs into crisp problems to solve them (see, e.g., [
21,
22]). A fuzzy method based on the generalized fuzzy Taylor expansion, introduced in [
16], is a one-step method. However, in the present paper, a novel multi-step fuzzy method is developed to find a numerical solution to the FFDEs directly in the fuzzy form without converting them to a crisp form. The Taylor series expansion and the Taylor-collocation method are two well-known, strong, and useful tools for solving both linear and nonlinear problems (see, e.g., [
16,
19,
23,
24,
25]).
In the present work, first, the generalized fuzzy Taylor expansion using the notion of Caputo’s fractional differentiability is extended. Second, the Adams methods are applied to solve the FFDEs. Additionally, the local truncation error, consistency, convergence, and stability of these methods are demonstrated. Finally, several examples with switching points are provided and solved using these methods. The numerical simulations and results reported here show that the generalized fuzzy Adams methods have high accuracy and can be effectively applied to solve the FFDEs.
2. Basic Preliminaries
Here, some elementary notions and theorems that are needed in the paper are recalled. For a widespread discussion, the interested reader may see [
16,
26,
27,
28].
Definition 1 ([
16])
. Suppose that and denotes the family of all functions , which satisfy the following statements: - 1.
t is normal in the sense that for some ;
- 2.
for all and all ;
- 3.
t is upper semi-continuous at a point , i.e., - 4.
is compact in with the Euclidean metric.
The family introduced above is the space consisting of fuzzy numbers, and each is a fuzzy number. Also, for all and all , the β-level set of t is Remark 1 ([
16])
. It follows from Statements 1–4 in Definition 1 that the β-level sets of a fuzzy number are all nonempty, bounded and closed (thus, compact) intervals. Definition 2 ([
16])
. Every fuzzy set in defines a triangular fuzzy number t characterized by , where as well as the lower bound and the upper bound of t are the endpoints of the β-level sets for all . Remark 2 ([
16])
. A crisp number k is also called singleton. For simplicity, the singleton k is represented by
Two well-known operations, namely the addition ⊕
and the scalar multiplication ⊙
in , can also be introduced and formulated naturally. In fact, if , then the addition of t and s is Moreover, if μ is a scalar, then the scalar multiplication of μ in t is The Hausdorff distance of two fuzzy numbers is given by the function defined by the rulewhere . Note that the function satisfies the following statements: - 1.
For all we have - 2.
For all and all we have - 3.
For all we have
Also, defines a metric on . It can be verified that is a complete, separable, and locally compact metric space.
Definition 3 ([
16])
. Suppose that . - 1.
If for some , then r is called the Hukuhara difference (shortly, H-difference) of t and s. The H-difference of t and s is denoted by . In other words, - 2.
If or for some , then r is called the generalized Hukuhara difference (shortly, gH-difference) of t and s. The gH-difference of t and s is simply denoted by . In other words,
- 1.
It can be easily verified that Statements (i) and (ii) in the definition of gH-difference are valid if r is crisp and vice versa.
- 2.
The interested reader can see some other conditions that imply the existence of the gH-difference in [26]. However, for simplicity, we assume here that every two fuzzy numbers have a gH-difference.
Before going on the notion of continuity in the space of fuzzy numbers, notice that by a fuzzy-valued function, assume a function
such that
is the
-level set or the parametric shape of
f for all
.
Definition 4 ([
16])
. A fuzzy-valued function is called continuous (regarding the Hausdorff metric ) at a point if for any arbitrary , there is a such that implies for all . If f is continuous at each point , then f is called continuous on the whole interval . The family of all continuous fuzzy-valued functions f, whose domain is the interval , may be denoted by . Remark 4 ([
16])
. A continuous function (regarding the Hausdorff metric ) is also integrable on the interval . Moreover, the functionis continuous on . Moreover,for all . Definition 5 ([
16])
. Suppose for a fuzzy-valued function that both and are differentiable at a point for all and (i.e., the gH-derivative of f) exists. - 1.
f is called FD1-gH-differentiable at whenever - 2.
f is called FD2-gH-differentiable at whenever
Now are ready to go on fuzzy fractional generalized Hukuhara derivative.
Definition 6 ([
16], see also [
29])
. Suppose that is a fuzzy Lebesgue integrable fuzzy-valued function and . The α-th order fuzzy Reimann-Liouville fractional (shortly, FRLF) integral of f is Definition 7 ([
16], see also [
29])
. Suppose for a fuzzy-valued function that (i.e., the m-th gH-derivative of f) is integrable on for all . The fuzzy fractional Caputo derivative (shortly, FFCD) of f isfor all , all and all . Remark 5 ([
16])
. Suppose for a fuzzy-valued function that (i.e., the gH-derivative of f) is integrable on . Consider the α-th order fuzzy Caputo generalized Hukuhara derivative (shortly, FC-gH-derivative) of f for some . The FC-gH-derivative can be formulated by In this paper, authors are interested in working on the -th order FC-gH-derivatives of fuzzy-valued functions.
Lemma 1 ([
16])
. For a continuous fuzzy-valued function , the function is also continuous on the interval for all . Definition 8 ([
16], see also [
29])
. Let be a fuzzy-valued function and . The point is called a switching point for the FC-gH-derivative of f if, for every neighborhood V of , there are with such that - (I)
f is FC1-gH-differentiable at but it is not FC2-gH-differentiable at , and
f is FC2-gH-differentiable at but it is not FC1-gH-differentiable at ;
or
- (II)
f is FC2-gH-differentiable at but it is not FC1-gH-differentiable at , and
f is FC1-gH-differentiable at but it is not FC2-gH-differentiable at .
Theorem 1 ([
16], fuzzy generalized Taylor)
. Suppose that and is a fuzzy-valued function with for . - 1.
If is FC1-gH-differentiable for and the type of the fuzzy Caputo differentiability remains unchanged on the interval , then - 2.
If is FC2-gH-differentiable for and the type of the fuzzy Caputo differentiability remains unchanged on the interval , then - 3.
If exists for and the type of the fuzzy Caputo generalized Hukuhara differentiability changes on the interval , then - 4.
Suppose that for all , and f is FC2-gH-differentiable on the subinterval and is FC1-gH-differentiable on the subinterval , i.e., c is a Type (II) switching point for the α-th order derivative of f. Suppose further that there is a such that the derivative of f of order at a point has a Type (I) switching point. If the type of the fuzzy Caputo differentiability remains unchanged on the interval , then
3. Fuzzy Generalized Adams Methods
In this section, are delved into some intriguing aspects of the fuzzy generalized Taylor theorem (Theorem 1), with a specific focus on the fuzzy generalized Adams methods. Our approach begins with establishing the fuzzy generalized Adams–Bashforth (shortly, A-B) method and the fuzzy generalized Adams–Moulton (shortly, A-M) method, which is grounded in the concept of fuzzy fractional gH-differentiability.
To lay the groundwork for these methods, first are introduced the concepts of forward finite differences and backward finite differences using gH-differences. Following this, are formulated Newton’s forward interpolation polynomial and backward interpolation polynomial. Utilizing these foundations, we derived the fuzzy generalized A-B method and the fuzzy generalized A-M method.
Prior to embarking on our main discussion, are presented a concise overview of the formulation of linear multi-step methods. This introduction is aimed at providing a fundamental understanding necessary for grasping the subsequent advanced concepts. In the general case, a linear multi-step method for a continuous fuzzy-valued function
is formulated by
where
is an integer,
and
’s and
’s are real constants such that
and
. It is noted that the family of Adams methods is a subfamily of the family of linear multi-step methods with
3.1. Forward Finite Differences, Backward Finite Differences and Newton’s Interpolation Polynomials
Definition 9 ([
19])
. A fuzzy interpolation polynomial of some data is a fuzzy-valued function satisfying the following statements: - 1.
for ;
- 2.
p is continuous on the whole set ;
- 3.
The interpolation polynomial p is also crisp whenever the data are crisp.
To construct a fuzzy interpolation polynomial, i.e., a fuzzy-valued function p satisfying Conditions 1–3 in Definition 9, it is necessary to introduce forward finite differences and backward finite differences using the gH-differences between and . Next do this in two separated subsections and then construct Newton’s fuzzy forward and backward interpolation polynomials that are used frequently in our methods. Newton’s fuzzy interpolation polynomials are two well-known examples of fuzzy interpolation polynomials.
3.1.1. Newton’s Fuzzy Forward Interpolation Polynomial
Consider a continuous fuzzy-valued function
such that the values of
f at
are known, where
. For a fixed integer
and any
, put
Also, the
-level sets of
are defined by
Similarly, for
we have
Continuing this argument yields
and so on. The fuzzy values
, where
, are called the forward finite differences based on the generalized Hukuhara differences (shortly, forward finite gH-differences) of
f.
Using these notations, Newton’s fuzzy forward interpolation polynomial may be defined as follows:
Definition 10 ([
19])
. For any , Newton’s fuzzy forward interpolation polynomial at the point is Remark 6 ([
19])
. The variable θ in Equation (2) is crisp since and x, ’s and h are all crisp variables. 3.1.2. Newton’s Fuzzy Backward Interpolation Polynomial
Similar to Newton’s fuzzy forward interpolation polynomial, one may formulate Newton’s fuzzy backward interpolation polynomial. To this end, consider again a continuous fuzzy-valued function
such that the values of
f at
are known, where
. For a fixed integer
and any
, put
Also, the
-level sets of
are defined by
Similarly, for
we obtain
Continuing this arguments yields
and so on. The fuzzy values
, where
, are called the backward finite differences based on the generalized Hukuhara differences (shortly, backward finite gH-differences) of
f.
Using these notations, Newton’s fuzzy backward interpolation polynomial may be defined as follows:
Definition 11 ([
19])
. For any , Newton’s fuzzy backward interpolation polynomial at the point is Remark 7 ([
19])
. Similar to Equation (2), the variable θ in Equation (4) is crisp since and x, ’s and h are all crisp variables. 3.2. Fuzzy Generalized A-B Method and Fuzzy Generalized A-M Method
Consider the FFIVP
where
y is the unknown fuzzy-valued function of a crisp variable
x,
is a continuous fuzzy-valued function,
and
is the
-th order fuzzy fractional Caputo derivative of
y with a finitely many switching points in
. By separating the closed interval
into
N subintervals with the same step length
, obtain a partition
of this interval, where
Please note that
. Further is assumed that the FFIVP (
5) has a unique solution, and the authors wish to solve it using three fuzzy generalized Adams methods, namely:
fuzzy generalized A-B two-step method with backward finite gH-differences,
fuzzy generalized A-M two-step method with forward finite gH-differences, and
fuzzy generalized A-M three-step method with forward finite gH-differences.
3.2.1. Fuzzy Generalized A-B Two-Step Method with Backward Finite gH-Differences
To solve the FFIVP (
5) using the fuzzy generalized A-B two-step method with backward finite gH-differences, suppose that
is a fixed integer,
and
are fuzzy initial values, i.e.,
and
are known fuzzy numbers and also,
In particular,
and we obtain
Then the Newton’s fuzzy backward interpolation polynomial at
and
is
Now, assume that Equation (
3)(i) holds. Taking integral over the closed interval
from the first equality in the FFIVP (
5) leads us to
and so
Thus, by substituting the fuzzy-valued function
f with Newton’s fuzzy backward interpolation polynomial
at the points
and
in Equation (
6) as well as using the definition of backward finite gH-differences obtain
3.2.2. Fuzzy Generalized A-M Two-Step Method with Forward Finite gH-Differences
To solve the FFIVP (
5) using the fuzzy generalized A-M two-step method with forward finite gH-differences, suppose again that
is a fixed integer,
,
and
are fuzzy initial values, i.e.,
,
and
are known fuzzy numbers and also,
In particular,
and we obtain
Then the Newton’s fuzzy forward interpolation polynomial at
,
and
is
Now, assume that Equation (
1)(i) holds. Taking integral over the closed interval
from the first equality in the FFIVP (
5) leads us to
and so
Thus, by substituting the fuzzy-valued function
f with the Newton’s fuzzy forward interpolation polynomial
at the points
,
and
in Equation (
8) as well as using the definition of forward finite gH-differences we obtain
3.2.3. Fuzzy Generalized A-M Three-Step Method with Forward Finite gH-Differences
To solve the FFIVP (
5) using the fuzzy generalized A-M three-step method with forward finite gH-differences, suppose one more time that
is a fixed integer,
,
,
and
are fuzzy initial values, i.e.,
,
,
and
are known fuzzy numbers and also,
In particular,
and we obtain
Then the Newton’s fuzzy forward interpolation polynomial at
,
,
and
is
Now, assume that Equation (
1)(i) holds. Taking integral over the closed interval
from the first equality in the FFIVP (
5) leads us to
and so
Thus, by substituting the fuzzy-valued function
f with Newton’s fuzzy forward interpolation polynomial
at the points
,
,
and
in Equation (
10) as well as using the definition of forward finite gH-differences obtain
4. Analysis of the Fuzzy Generalized Adams Methods
Now, are explored the local truncation error (LTE) and the global truncation error (GTE) associated with the fuzzy generalized Adams methods. Our aim is to establish the consistency, the convergence as well as the stability of the developed methods. It is important to note that our investigation here is limited to the fuzzy generalized A-B two-step method. Here is focused on this method because the proofs for other methods in this category are quite similar and would follow analogous lines of reasoning.
4.1. Local Truncation Error (LTE)—Consistency
Let the FFIVP (
5) have a unique solution
y. First, define the residual of
y. Therefore, consider the two types of differentiability of
y separately.
Definition 12. Let y be the unique solution of the FFIVP (5). Define the residual under two different cases: - 1.
In the case that y is FC1-gH-differentiable on the interval and the type of the differentiability remains unchanged, have - 2.
In the case that y is FC2-gH-differentiable on the interval and the type of the differentiability remains unchanged, have
Now go on the local truncation error (LTE) as well as the consistency of y.
Definition 13. Let y be the unique solution of the FFIVP (5) and is its residual. - 1.
The local truncation error (LTE) is denoted by and is defined as - 2.
The fuzzy generalized Adams methods are consistent if
Remark 8. Let y be the unique solution of the FFIVP (5). For , depending upon the type of the differentiability of y, the LTE and the residual are formulated as follows: - 1.
In the case that y is FC1-gH-differentiable on the interval and the type of the differentiability remains unchanged, have - 2.
In the case that y is FC2-gH-differentiable on the interval and the type of the differentiability remains unchanged, have
Now, it is ready to prove the consistency of the fuzzy generalized Adams methods. To this end, let
where
. There are two cases:
Case One. If
y is FC1-gH-differentiable on the interval
and the type of the differentiability remains unchanged, then
Case Two. If y is FC2-gH-differentiable on the interval and the type of the differentiability remains unchanged, then a similar argument proves the consistency of the methods.
It should be noted that the fuzzy generalized Adams methods are consistent until the solution y lies in .
4.2. Global Truncation Error (GTE)—Convergence
Lemma 2 ([
30])
. For every , we have . Definition 14. Let y be the unique solution of the FFIVP (5) and is its residual. - 1.
The global truncation error (GTE) is denoted by and is defined as the accumulation of the LTE on all iterations, considering good information of y at the initial time step.
- 2.
The fuzzy generalized Adams methods are convergent if the GTE tends to 0 whenever the step size approaches 0, i.e.,
Remark 9. Let y be the unique solution of the FFIVP (5). Independent from the type of differentiability of y, the GTE can be written by . Now, start the proof of the convergence of our developed fuzzy methods. To this end, let
exist and
f satisfies the Lipschitz condition on the region
We have two cases:
Case One. If
y is FC1-gH-differentiable on the interval
and the type of the differentiability remains unchanged, then
On the other hand,
and
where
is the Lipschitz constant of
f. Therefore, putting
gets
Because the last inequality is true for every positive integer
k, hence
Repeating this argument yields
and from the summation formula
of geometric sequences obtain
Now, set
in Lemma 2. Then
where
for
. Thus, in (
12) find
If the initial value is chosen accurately enough, then we obtain
. Hence
Finally, letting , find .
Case Two. If y is FC2-gH-differentiable on the interval and the type of the differentiability remains unchanged, then a similar argument proves the convergence of the methods.
5. Stability
Now, is investigated the stability of our developed fuzzy methods. First, have a definition.
Definition 15. Let , for be the solution obtained from the fuzzy generalized Adams methods and be the solution obtained from that method where determines the perturbed fuzzy initial condition. The fuzzy generalized Adams methods are stable whenever there are positive constants and such that for all with and and all , implies .
Now, prove the stability of our developed fuzzy methods. There are two cases:
Case One. If
y is FC2-gH-differentiable on the interval
and the type of the differentiability remains unchanged, then the perturbed problem can be formulated as
Here, assume that the Hausdorff metric specifications hold. Therefore, applying the Lipschitz condition obtain
Redoing this argument and using Lemma 2 yield
for all
with
and
and all
, where
.
Case Two. If y is FC1-gH-differentiable on the interval and the type of the differentiability remains unchanged, then a similar argument proves the stability of the methods.
Theorem 2. Let be a solution to the fuzzy fractional differential equation (FFDE)with the initial condition . If is invariant under the transformation and , then the solution exhibits symmetry about the origin, i.e., . Proof. Consider the fuzzy fractional differential equation
with the initial condition
.
Apply the transformation
and
. Let
. Then,
Using the property of the fractional derivative, we have
Substitute
in the original equation:
Given that
is invariant under the transformation
and
, we have
Since
satisfies the same differential equation as
, and given the initial condition
, it follows that
. Hence,
Thus, the solution exhibits symmetry about the origin. □