1. Introduction
As an important tool for modelling many phenomena in various fields of science, fractional calculus and the fractional differential equations have been intensively investigated in the last decades. It seems that many natural systems can be represented more accurately via the memory data included through the fractional derivative formulation. For more details on fractional calculus theory and fractional differential equations, see the monographs of Kilbas et al. [
1] and Podlubny [
2]. For distributed order fractional differential equations, we refer Jiao at al. [
3] and an application-oriented exposition is given in Diethelm [
4]. The important case of impulsive differential and functional differential equations with fractional derivative and some applications are considered in the monograph of Stamova and Stamov [
5].
From practical experience, it is well known that the stability of a process (in particular, of a stationary state) is the ability of the process to resist a priory unknown, small influences (disturbances). If such disturbances do not essentially change it, the process is said to be stable. It turns out that the investigation of stability is of utmost importance. So, the integral representations of the solutions of the studied models as main tools for investigation of different kinds of stability properties are an important theme for research. For deep information and a good historical overview of the stability properties for delayed and neutral systems with integer order derivatives we recommend the books by Hale and Lunel [
6] and Kolmanovskii and Myshkis [
7]. For the case of fractional-order derivatives, we prefer the surveys [
8,
9,
10,
11]. From the newest results, we note [
12,
13] (for fundamental theory) and [
14,
15,
16] (for applications).
In the present work, we considered an initial (Cauchy) problem (IP) for a linear system with derivatives in Caputo’s sense of incommensurate order, distributed delays and discontinuous initial functions. The motivation to study distributed delay systems is because this type of delay includes as a special case, all types of delays (it follows from the Riesz’s theorem applied to the functional of Krasovskii), and in this sense, it is most appropriate to obtain results valid for all types of delays. To study the considered initial problem, we introduce a new approach via an equivalent Volterra–Stieltjes integral system, which allows us to obtain an integral representation of the solutions of the Cauchy problem for the studied class systems. This approach is based on the existence and uniqueness of a global fundamental matrix for the corresponding homogeneous system, which allows us to prove straightforwardly that the corresponding resolvent system possesses a unique measure resolvent kernel. As a main result, we obtain a relation between the global fundamental matrix of the homogeneous system and the measure resolvent kernel, which is the unique solution of the resolvent system. Then, we establish two equivalent integral representations, where one of them is applicable even in cases when we interpret the space of the initial functions as state space. Then, using the obtained integral representations we study the relations between the stability of the zero solution of the homogeneous system and different kinds of boundedness of its other solutions.
The paper is organised as follows. In
Section 2, as usual we recall the most-used definitions of Riemann–Liouville and Caputo-type fractional derivatives and some of their properties. In the same section, we introduce the statement of the problem, as well as some necessary comments and auxiliary results used later. In
Section 3, we introduce our approach for establishing of integral representation of the solutions of the initial problem for the studied systems with discontinuous initial functions. The obtained integral representation is applicable even in the case when the space of initial functions is treated as state space. In
Section 4, using the results obtained in the previous section, we study the relations between the stability (uniform stability) of the zero solution of the system (
3) and the different kinds of boundedness of its other solutions.
Section 5 is devoted to some conclusions and comments.
2. Preliminaries and Problem Statement
We start with recall of the basic definitions of the fractional integral and derivative in the Riemann–Liouville sense, the Caputo-type fractional derivative as well as some of their properties. For additional details and other properties, we recommend [
1,
2].
Let
be an arbitrary number. With
we will denote the linear space of all locally Lebesgue integrable functions
and by
its subspace of all locally bounded functions. Then, for arbitrary
, each
and
the left-sided fractional integral operator and the Riemann–Liouville and Caputo-type left-sided fractional derivatives of order
are defined via
and the following relations (see [
1]) hold:
The following notations will be used, too: , , , , denotes the real linear space of the square matrices with dimension , are the identity and zero matrix, respectively, and denotes the zero vector-column.
Everywhere below when we speak about the integer case we understand the same system in the form when , i.e., the corresponding system with first-order derivatives.
Let , . With is denoted the linear space of the matrix valued functions with bounded variation in on every compact interval for every , i.e., and . For , , , we use the notations .
Let
be arbitrary. With
we denote the Banach space of all vector functions
which are bounded and Lebesgue measurable (piecewise continuous) on the interval
with norm
. With
we denote the set of all jump points of
and as usual
is the subspace of all continuous functions, i.e.,
.
Let
be arbitrary fixed and consider for
the inhomogeneous fractional linear delayed system with incommensurate-type differential orders and distributed delays in the following general form:
with initial condition for each
as follows:
where
,
(the symbol
⊤ means transposition),
,
;
;
; , ; .
The Lebesgue decomposition in
of
has the form
with
,
,
, where
,
,
are the jump, the absolutely continuous and the singular part, respectively, in the decomposition and the integral in (
1) is understood in the Lebesgue–Stieltjes sense. The homogeneous system of (
1) (i.e.,
,
) written in detail for any
has the form:
Definition 1 ([
6,
7,
17])
. We say that for the kernel the conditions (S) are fulfilled, if for each , and any the following conditions hold:- (S1)
The functions , and are measurable in , continuous from left in θ on , normalised so that for , for , where , , , and when .
- (S2)
The kernels , , , where , ,
and , .
- (S3)
, , where is the Heaviside function and the matrix .
- (S4)
For each , the following relations hold: For any , we denote the set of its all jump points by and assume that the set does not have limit points.
Remark 1. As it is standard, the presented form of the initial condition (
2)
, is chosen to consider the function , as a prolongation from right of the function defined for . Note that some authors, instead the more weak condition (S4), assume that the kernels are continuous in t, but this assumption excludes from considerations the important case of systems with variable concentrated delays (i.e., the system (
1)
can have only constant delays). We emphasise that for any from Condition (S4) it follows that the set is at most finite. Definition 2. The vector function is a solution of the IP (
1)
, (
2)
in for some , if satisfies the system (
1)
for all and the initial condition (
2)
for each . (When misunderstanding is not possible, we will write only .) Applying to both sides of (
1) the left-sided fractional integral operator
, we obtain the following auxiliary system
where
.
Definition 3. The vector function is a solution of the IP (
4)
, (
2)
in if satisfies the system (
4)
for all and the initial condition (
2)
for each . In our exposition below, we need some auxiliary results for the case when the initial function ( is discontinuous).
Theorem 1 (see [
18])
. Let the following conditions hold:- 1.
Conditions (S) hold.
- 2.
The function .
Then, for arbitrary fixed initial point the following statements hold:
- (i)
Every solution of IP (
1)
, (
2)
is a solution of the IP (
4)
, (
2)
and vice versa. - (ii)
For any initial function and , the IP (
4)
, (
2)
has a unique solution in .
For any , define the following initial functions , with (where is the l-th column of the identity matrix and for .
Let
be the low terminal and initial point. Consider for
the IP (
4), (
2) in the homogeneous case of (
4) (i.e., when
,
) with initial function
,
,
as follows:
Corollary 1. Let the conditions (S) hold.
Then, the IP (
5)
, (
6)
has a unique solution for any and arbitrary fixed number . Proof. Since the statement of Theorem 1 holds for any fixed initial point , then the assertion of Corollary 1 follows immediately from Theorem 1 for any . □
Definition 4. The matrix-valued functionis called the fundamental (or Cauchy) matrix of the system (
3)
if for any its l-th column is a solution of the IP (
5)
, (
6)
. As usual, everywhere below we will assume that is prolonged as continuous in t function on for arbitrary fixed .
Below we formulate as a corollary the important properties of the fundamental matrix needed later in our exposition.
Corollary 2 (see [
19,
20])
. Let the conditions (S) hold.Then, the following statements hold:
- (i)
The system (
3)
has a unique fundamental matrix , which is absolutely continuous in t on every compact subinterval of , i.e., for any fixed , i.e., . - (ii)
For arbitrary fixed initial point and each fixed , is continuous at , right continuous at , left continuous at and has jumps of the first kind for .
- (iii)
The fundamental matrix has bounded variation in s on for arbitrary and its total variation in s on is bounded in t, .
Theorem 2 ([
21], Corollary 2)
. Suppose that and the following conditions hold:- 1.
The functions and , are nondecreasing.
- 2.
The inequality holds for .
Then, for any we have that .
Corollary 3. Let the conditions (S) hold. Then, the fundamental matrix has the following a priory estimate: Proof. Introduce the notations
,
and since the function
,
has a local minimum at
, where
we obtain the following estimations:
where
for
and
when
.
From Corollary 1, it follows that the fundamental matrix
satisfies the following matrix system:
and since
then from (
9), it follows that
Then, the estimation (
7) follows from (
10) and Theorem 2. □
3. Main Results
The main goal of this section is to introduce a new approach to study the IP (
5), (
6) via an equivalent Volterra–Stieltjes system. It is based on the existence and uniqueness of a global fundamental matrix of the system (
3), which is a well known and often studied problem.
Consider for arbitrary initial function
and
the Volterra–Stieltjes system
where
Theorem 3. Let the following conditions hold:
- 1.
Conditions (S) hold.
- 2.
The initial function and the function are arbitrary.
Then, every solution of IP (
1)
, (
2)
(IP (
4)
, (
2)
) is a continuous solution of the system (
11)
for and vice versa. Proof. Let
be the corresponding unique solution of the IP (
1), (
2) existing according Theorem 1 for the
and
mentioned in condition 2 of Theorem 3. According the first statement of Theorem 1, the solution
satisfies system (
4) and then substituting this solution in (
4) and splitting off in system (
4) the parts that explicitly depend on the initial data, we obtain
Let us define the kernel
via the equality
Applying Fubini’s theorem to the right side of (
13), we have that
and then using (
14), system (
15) can be rewritten as the Volterra–Stieltjes Equation (
11).
The vice-versa statement can be proved in the reverse way. □
Definition 5 ([
22])
. We say that is a Volterra measure kernel of type on (we denote ) if the following conditions hold:- (i)
For any the is a matrix of scalar real Borel measures on with support in ;
- (ii)
;
- (iii)
For each Borel set the function is Borel measurable.
Definition 6. We say that is a Volterra–Stieltjes measure kernel of type on and denote , if is a matrix of scalar real Volterra–Stieltjes measures on with support in and the following conditions hold:
- (i)
For the kernel the conditions (S) hold;
- (ii)
.
We say that the function () is a local kernel of type () on , and denote by (), if the conditions of Definition 5 (Definition 6) hold on any compact subsets of .
Lemma 1. Let .
Then, we have that .
Proof. Let . Then, the conditions (S) imply that the kernel is a matrix with elements that are scalar Lebesgue–Stieltjes measures and hence they are also Borel measures. Then, from (S1) and (S2) it follows that the kernel satisfies the conditions (i) and (iii) in Definition 5. Taking into account that the conditions (ii) in the Definition 5 and Definition 6 coincide, we can conclude that . □
Lemma 2. Let the following conditions hold:
- 1.
.
- 2.
The functions are bounded.
Then, we have that .
Proof. The statement follows immediately from Lemma 1. □
Lemma 3. Let the following conditions hold:
- 1.
The kernel is defined via (
14)
. - 2.
The kernel appearing in (
9)
satisfies the conditions (S).
Then, and for every fixed with we have that .
Proof. Since
is defined via (
14) and
satisfies the conditions
(S), then from the conditions
(S) it follows that that the kernel
is a matrix with elements that are scalar Lebesgue–Stieltjes measures and hence they are also Borel measures. Then, the assertion follows immediately from Lemma 1 and (
14). □
The next simple corollary gives an explicit condition that guaranties that condition (ii) of Definition 6 holds.
Corollary 4. Let the following conditions hold:
- 1.
The conditions of Lemma 3 are fulfilled.
- 2.
The functions are bounded.
Then, we have that .
Proof. Condition 2 of the corollary implies that condition (ii) holds, i.e.,
Then, applying Lemma 1 we obtain that the statement of Corollary 4 holds. □
The keystone of the approach introduced in the remarkable book [
22] for solving linear systems having the same form as (
6) with nonconvolution measure-valued Volterra kernels
, as well as obtaining an integral representation of their solutions, is based essentially on the possibility of solving the corresponding resolvent system for the kernel
in the form
The solution
of the system (
16) if it exists is called measure resolvent for the kernel
.
The proof in [
22] of the problem of the existence of a measure resolvent
is based on non-trivial techniques using Banach algebras. Here, we introduce a new approach for establishing of this existence, without the necessity to use Banach algebra techniques. This approach is based on the existence and uniqueness of a global fundamental matrix for the system (
3) (see Corollary 1 in [
18]), which is a well known and often studied problem. The key result in our approach is to establish a relation between the existence of global fundamental matrix of the homogeneous system (
3) and the existence of a measure resolvent kernel
for the measure kernel
; this problem is treated in the next theorem.
Remark 2. It must be noted that if is a solution of system (
16)
, then for every fixed with we have that . Let
be arbitrary and
be the unique fundamental matrix of (
3). Introduce the relation
where
is the indicator function.
Theorem 4. Let the following conditions hold:
- 1.
The kernel is defined via (
14)
. - 2.
The kernel appearing in (
9)
satisfies the conditions (S).
Then, the matrix defined via the relation (
17)
is a solution of the resolvent Equation (
16)
if and only if, when the matrix appearing in (
17)
is the unique fundamental matrix of (
3)
. Proof. Sufficiency.
Let
be an arbitrary fixed number. Then, for any
the solution of the IP (
5), (
6) denoted by
exists according Corollary 1, i.e., for any
we have that
where
is defined via (
14) and hence the fundamental matrix.
is the unique solution of the following matrix system:
For the function
defined via (
17), taking into account the properties of
mentioned in Corollary 2, we have that
for any fixed
,
.
Substituting (
17) in (
16) and integrating by parts for
, we obtain that
and hence the function
defined via (
17) is a solution of the resolvent Equation (
16).
To complete the proof of the sufficiency, we must prove that .
From (
17), it follows that for any fixed
with
we have that
, and since the kernel
is with support in
then the same is true for
, too. According to Corollary 2, the fundamental matrix
has bounded variation in
s on
for any
, and its total variation in
s on
is bounded in
t,
. Then, from (
17) it follows that
has the same properties and hence
.
Necessity. Let
be a solution of (
16). Then, using (
17) in a reverse way via (
16) we obtain that
satisfies (
19) for
. If we assume that (
16) has two different solutions
, then via (
17) we can obtain that (
19) has two different solutions
, which is impossible and hence the system (
16) has a unique solution. □
The next corollary is an adaptation of Theorem 2.5 in [
22], Ch.10, in a form more convenient for practical application, in partial for our case too.
Corollary 5. Let the following conditions hold:
- 1.
The kernel is defined via (
14)
. - 2.
The kernel appearing in (
9)
satisfies the conditions (S).
- 3.
The function .
Then, for any initial function and each the IP (
6)
, (
2)
has a unique solution that possesses the following integral representation (variation of constants formula):where the function is defined via (
12)
and is the unique solution of (
16)
defined via (
17)
. Proof. From Lemma 1 and Lemma 3 it follows that the kernel
, and then applying Theorem 4 we obtain that the resolvent system (
16) has a unique solution
. Then, the statement of Corollary 5 follows from Theorem 2.5 in [
22], Ch.10. □
Theorem 5. Let the conditions of Corollary 5 hold.
Then, for any initial function and each the unique solution of the IP (
6)
, (
2)
has the following integral representation:where the function is defined via the Equality (
12)
and is the fundamental matrix of (
3)
. Proof. Let the functions
and
be arbitrary. Then, according Theorems 1 and 3 the IP (
11), (
2) has a unique solution
in
and hence the system (
3) has a unique fundamental matrix
. Since the kernel
is defined via (
9) and the kernel
appearing in (
14) satisfies the conditions
(S), then according to Lemma 3 we have that
. From Theorem 4, it follows that the kernel
defined via (
17) is the unique solution of the resolvent Equation (
16) and
. Applying Corollary 5, we obtain that for the unique solution of the IP (
11), (
2)
the integral representation (
21) holds. Integrating by parts the equality (
21), we obtain that
which completes the proof. □
Remark 3. The obtained above results are a generalisation of the results in [23] obtained via another approach. Now, we will establish an equivalent, but more applicable version of the integral representation (
22), interpreting
as state space to characterise the operators appearing in the obtained version. In our point of view, the new version of this formula will have great advantages in comparison with (
22). The interpretation that uses
as state space is widely accepted, and it has proven it is fruitful in many sections of the theory, especially sections related to asymptotic behaviour and periodicity of solutions.
Let , be an arbitrary real Banach space and with .
Definition 7 ([
24])
. A two-parameter family of bounded linear operators , is called a forward evolutionary system on whenever the following hold: for any ;
for any .
Everywhere below we will assume that the conditions of Theorem 1 hold. Then, for any
the IP (
1), (
2) has a forward non-continuable unique solution
and let for any
define the family of linear maps
via the relation:
Proposition 1. Let the conditions of Theorem 1 hold and the kernel is defined via (
14)
. Then, the operators defined via (
23)
are continuous for any . Proof. Let
be arbitrary and
,
are the corresponding unique solutions of IP (
4), (
2). Using the same notations as in Corollary 3, we obtain that the estimations (
8) hold and
Substituting both solutions in (
4), subtracting both equations and then from the estimations (
8), (
24) and Theorem 2 it follows that
holds for any fixed
.
This completes the proof. □
Remark 4. We note that if the initial function , then for any , but if , then when . Moreover, Theorem 1 implies that the shifting operator (
23)
defines an evolutionary system on when the initial function (). Define the matrix
as
, where
,
are the same functions as in the initial condition (
6). It is clear that for
we have
and
when
. Since each of its columns
,
, then from (
23) it follows that for
As in the integer case, the map defined via (
23) will be called a solution map or shifting operator (along the solutions trajectories) for the system (
1). In the homogeneous case, i.e., when the forced term
in
, then
denotes the corresponding unique solution of the IP (
3), (
2). It must be noted that the solution map has a key role in the abstract-type representations of the solutions in both cases (homogeneous and inhomogeneous systems).
According to (
23), for any
the unique solution of the IP (
3), (
2)
has the form
and the unique solution
of the IP (
1), (
2) with initial function
,
(see [
19,
20]) has the form
for any
.
Then, for
we have the following representation
Hence the superposition principle implies that for any fixed
we have that
and therefore from (
23) and (
26) it follows that for any fixed
we obtain
where the integral in the right side of (
27) is understood as a family of Euclidean space integrals parametrised via
.
Our aim is to give an interpretation of (
27) when we use
as state space. Note that this problem in the fractional case is more complicated in comparison with the integer case and cannot be realised via one family of compact linear operators parametrised via
, allowing representation as vector-valued Lebesgue–Stieltjes integrals.
Integrating by parts the integral in the right side in (
26), we have
Define for arbitrary fixed
,
and
the following two families of operators (parametrised via
):
and hence for any fixed
the defined operators
are linear. The equalities (
29) and (
28) show that in the fractional case the situation is essentially different compared with the integer case.
Theorem 6. Let the conditions of Corollary 5 hold. Then, the operators defined via (
29)
of both families for any are bounded and compact. Proof. Let
be fixed,
be arbitrary and
. Then, according Lemma 2.1 and (2.8) in [
1], the operator
is bounded and hence from (
29) and Corollary 2, point 1, it follows that the operator
is bounded too. Furthermore, from (
29) it follows that if the operator
is compact, then the operator
is compact too.
Introduce for arbitrary fixed
the set
To apply the Arzela–Ascoli theorem, it is necessary to check that the set is uniformly bounded and equicontinuous.
From (
24) for arbitrary
, we have that
and hence the set
is uniformly bounded.
Since the function is uniformly continuous for , then for any there exists such that if then we have
and hence
Thus, the set
is equicontinuous and therefore the operator
is compact for any
. □
Finally, when we interpret
as a state space, from (
27), (
28) and (
29) we obtain the following representation:
where, according to Theorem 6, the families of operators
are compact for any
.
5. Comments and Conclusions
In the present work a Cauchy (initial) problem for a fractional linear system with distributed delays, Caputo-type derivatives of incommensurate order and discontinuous initial functions was considered.
As a main result, first, a new straightforward approach was introduced to study the considered initial problem via an equivalent Volterra–Stieltjes integral system, which allowed us to obtain an integral representation of the solutions of the Cauchy problem for the studied class system. This approach is based on the existence and uniqueness of a global fundamental matrix for the corresponding homogeneous system; this fact allowed us to prove that the corresponding resolvent system possesses a unique measure resolvent kernel. The proof in [
22] of the problem of the existence of a measure resolvent
is based on non-trivial Banach algebra techniques. Here, we have introduced a new approach for establishing this existence, without the necessity to use these techniques. In our point of view, an approach based on the existence and uniqueness of a global fundamental matrix for the system (
3) is not too complicated, since it is based on a well-known and often studied problem. The key result in our approach is the established relation between the global fundamental matrix of the homogeneous system (
3) and the measure resolvent kernel
for the measure kernel
.
As a consequence of this result, we have obtained an integral representation of the solutions of the studied systems. The integral representation of the solutions of the considered Cauchy problem obtained via the introduced approach is applicable even in the case when the initial functions are only locally bounded and Lebesgue measurable. With (
30) we have established an equivalent, but more applicable version of the obtained integral representation (
22) in cases when we interpret
as state space. The interpretation that uses
as state space is nowadays widely accepted and it has proven fruitful in many sections of the theory, especially sections related to asymptotic behaviour and periodicity of solutions. Analysing the operators appearing in the second addend in the right side in (
30), we can conclude that in the fractional case the situation is essentially different compared with these in the integer case. More preciously speaking, it is clear that the fractional case is more complicated in comparison with the integer case and cannot be realised as in the integer case via the one-parameter family of compact linear operators parametrised via
, each of them allowing representation as a vector-valued Lebesgue–Stieltjes integral.
Then, using the integral representations obtained in the previous section, we have obtained relations between the stability of the zero solution of the homogeneous system and different kinds of boundedness of its other solutions. We emphasise that in contrast with the integer case, the assertion of Corollary 6 is not equivalent with assertions (i)–(iii) of Theorem 7 and this is caused from the singular kernels of the fractional operators that impact on the boundedness of the operator defined via (
14). In the integer case, this operator is bounded on
t and then the assertion of Corollary 6 is equivalent to those of Theorem 7, but in the fractional case this is not enough as is illustrated via the example in [
25].