Mittag-Leffler Type Stability of BAM Neural Networks Modeled by the Generalized Proportional Riemann-Liouville Fractional Derivative

Ravi P. Agarwal; Snezhana Hristova; Donal O'Regan

doi:10.20944/preprints202305.1659.v1

Submitted:

22 May 2023

Posted:

23 May 2023

You are already at the latest version

Abstract

The main goal of the paper is to use a generalized proportional Riemann-Liouville fractional derivative (GPRLFD) to model BAM neural networks and to study some stability properties of the equilibrium. Initially, several properties of the GPRLFD are proved such as the fractional derivative of a squared function. Also some comparison results for GPRLFD are provided. Two types of equilibrium of the BAM model with GPRLFD are defined. In connection with the applied fractional derivative and its singularity at the initial time the Mittag-Leffler exponential stability in time of the equilibrium is introduced and studied. An example is given illustrating the meaning of the equilibrium as well as its stability properties.

Keywords:

BAM neural networks

;

Mittag-Leffler type stability

;

fractional differential equations

;

generalized proportional Riemann-Liouville fractional derivative.

Subject:

Computer Science and Mathematics - Applied Mathematics

MSC: 34A34; 34A08; 34D20

1. Introduction

One of the main qualitative properties of the solutions of differential equations is stability. There are various types of stability defined, studied and applied to different types of differential equation, esspecially to fractional differential equation. The stability of Hadamard fractional differential equations is studied in [1]. The stability of Caputo type fractional derivatives are studied by many authors and many sufficient conditions are obtained (for example, see Mittag-Leffler stability in [2], the application of Lyapunov functions in [3]). Concerning fractional differential equations with Riemann-Liouville fractional derivatives the stability of linear systems is studied in [4], nonlinear systems in [5,6], Lyapunov functions are applied and comparison results are established in [14], practical stability is studied in [8], existence and Ulam stability in [9]. Note the initial condition for fractional differential equations with the Riemann-Liouville type fractional derivative is totally different than the initial condition to ordinary differential equations or to fractional differential equations with Caputo type derivatives. Some of the authors did not take this into account and consequently the study of stability has a gap. Concerning the basic concepts of the stability for Riemann-Liouville fractional differential equations we note [10] where several up-to-date types of fractional derivatives are defined, studied and applied to differential equations. Recently, the so called generalized proportional fractional integrals and derivatives were defined (see, [11,12]). Similar to classical fractional derivatives there are two main types of generalized proportional fractional derivatives: Caputo type and Riemann-Liouville type. Several results concerning existence (see, for example,[13,14]), integral presentation of the solutions (see, for example, [15]), stability properties (see, for example, [16,7]) and applications to some models (see, for example,[7]) are considered with the Caputo type of generalized proportional fractional derivatives. Also there are some results concerning Riemann-Liouville type. Some existence results are obtained in [18]. In [19,20] the oscillation properties of the fractional differential equations with a generalized proportional Riemann-Liouville fractional derivative is studied. The existence and uniqueness of a coupled system is studied in [21] in the case of three-point generalized fractional integral boundary conditions. In this paper initially we prove some comparison results for generalized proportional Riemann-Liouville fractional derivatives. Also, we discuss the behavior of the solutions on small enough intervals about the initial time. Some examples are given illustrating the necessity of excluding the initial time when the stability is studied. The obtained results are a basis for studying a stability property of the equilibrium of a model of neural networks. The models of neural networks are important issues due to their successive application in pattern recognition, artificial intelligence, automatic control, signal processing, optimization and etc. In the past decades, several types of fractional derivatives are applied to the models of neural networks to describe more adequately the dynamics of the neurons. Many qualitative properties of their equilibriums are studied. In this paper we apply the generalized proportional Riemann-Liouville fractional derivative to the BAM model of neutral networks. One of the main properties of the applied fractional derivative is its singularity at the initial time. In connection with this we define in an appropriate way an exponential Mittag-Leffler stability in time excluding the initial time. Also, two types of equilibrium deeply connected with the applied fractional derivative are defined. Sufficient conditions based on the new comparison results are obtained and illustrated with examples. The rest of this paper is organized as follows. In Section 2, some notes on fractional calculus are provided, the basic definitions of the generalized proportional fractional integrals and derivatives are given in the case when the order of fractional derivative is in the interval

(0, 1)

and the parameter is in

(0, 1]

. The connection with the tempered fractional integrals and the derivatives is discussed. In Section 3, we prove some comparison results for generalized Riemann-Liouville fractional derivatives. In Section 4, the model of BAM neural networks with GPRLFD is set up and studied. Two types of equilibriums are defined. These definitions are deeply connected with the applied GPRLFD and its properties which are totally different than the ones of ordinary derivative and the Caputo type fractional derivatives. Mittag-Lefller exponential stability in time of both types of equilibriums is defined and studied. Finally, an example is given to illustrate the theoretical results and statements.

2. Some notes on Fractional Calculus

The main idea of fractional calculus is the generalization of the differential operator to an operator with any real or complex number order. The most standard of these operators are the Riemann-Liouville fractional integral and derivatives (for basic definitions and properties see, for example, the classical books [22,23,24]). In the last decades many different definitions were proposed. Some of them are equivalent to the classical ones, some of them are generalizations. One of the ways to generalize the classical definitions is to include an exponential factor in the kernel. For more information about the definitions of fractional integrals and derivatives with exponential kernal, called tempered fractional integral and derivative, and some applications to stochastic process, Brownian motion, etc. we refer the reader to [25]. Recently, [11,12], generalized tempered fractional calculus was considered by fractionalising the power of the exponential function and these were called generalized proportional fractional ones. In this way, the used parameter in the exponential kernel gives us more detailed information.

We recall some basic definitions and properties relevant to the generalized proportional fractional derivative and integral. The terms and notations are adopted from [11,12].

Definition 1.

[11] (The generalized proportional fractional integral) (GPFI) Let

υ : [a, b] \to R, b \leq \infty,

and

ρ \in (0, 1]

,

q \geq 0

. We define the GPFI of the function υ by

(_{a} I^{0, ρ} υ) (t) = υ (t)

and

\begin{matrix} (_{a} I^{q, ρ} υ) (t) = & \frac{1}{ρ^{q} Γ (q)} \int_{a}^{t} e^{\frac{ρ - 1}{ρ} (t - s)} {(t - s)}^{q - 1} υ (s) d s, t \in (a, b] . \end{matrix}

(1)

Definition 2.

[11] (The generalized proportional Riemann-Liouville fractional derivative) (GPRKFD) Let

υ : [a, b] \to R, b \leq \infty,

and

ρ \in (0, 1]

,

q \in (0, 1)

. Define the GPRLFD of the function υ by

\begin{matrix} (_{a}^{R} D^{q, ρ} υ) (t) & = \frac{1}{ρ^{1 - q} Γ (1 - q)} ((1 - ρ) \int_{a}^{t} e^{\frac{ρ - 1}{ρ} (t - s)} {(t - s)}^{- q} υ (s) d s \\ + ρ \frac{d}{d t} \int_{a}^{t} e^{\frac{ρ - 1}{ρ} (t - s)} {(t - s)}^{- q} υ (s) d s), t \in (a, b] . \end{matrix}

(2)

Remark 1.

The parameter q in Definitions 1 and 2 is interpreted as an order of integration and differentiation, respectively. The parameter ρ is connected with the power of the exponential function. In the case

ρ = 1

the given fractional integral and derivative reduce to the classical Riemann-Liouville fractional integral

a I_{t}^{q} υ (t) = \frac{1}{Γ (q)} \int_{a}^{t} {(t - s)}^{q - 1} υ (s) d s,

(3)

and the Riemann-Liouville fractional derivative

{}_{a}^{RL}D_{t}^{q} υ (t) = \frac{1}{Γ (1 - q)} \frac{d}{d t} \int_{a}^{t} {(t - s)}^{- q} υ (s) d s,

(4)

The relation between GPRLFD with the Riemann-Liouville fractional derivative is given in the following Lemma.

Lemma 1.

Let

ρ \in (0, 1], q \in (0, 1)

, and

υ \in C ([a, b]), b \leq \infty

. Then

(_{a} D^{q, ρ} υ) (t) = ρ^{q} e^{\frac{ρ - 1}{ρ} t} ({}_{a}^{RL}D_{t}^{q} (e^{\frac{1 - ρ}{ρ} t} υ (t))), t \in (a, b] .

(5)

Proof.

From Eq. (2) and (4) we have

\begin{matrix} ({}_{a}^{R}D^{q, ρ} υ) (t) & = \frac{1}{ρ^{1 - q} Γ (1 - q)} ((1 - ρ) \int_{a}^{t} e^{\frac{ρ - 1}{ρ} (t - s)} {(t - s)}^{- q} υ (s) d s \\ + ρ \frac{d}{d t} e^{\frac{ρ - 1}{ρ} t} \int_{a}^{t} e^{\frac{1 - ρ}{ρ} s} {(t - s)}^{- q} υ (s) d s) \\ = \frac{1}{ρ^{- q} Γ (1 - q)} e^{\frac{ρ - 1}{ρ} t} \frac{d}{d t} \int_{a}^{t} e^{\frac{1 - ρ}{ρ} s} {(t - s)}^{- q} υ (s) d s \\ = ρ^{q} e^{\frac{ρ - 1}{ρ} t} ({}_{a}^{RL}D_{t}^{q} e^{\frac{1 - ρ}{ρ} t} υ (t)) . \end{matrix}

□

Remark 2.

The equality (5) gives us an opportunity to apply some of the properties known in the literature for Riemann-Liouville fractional derivatives to GPRLFD. However it does not allow us to use directly properties of solutions of fractional differential equations with Riemann-Liouville fractional derivatives to the ones with GPRLFD. That is why it is absolutely necessarily to study independently differential equations with GPRLFD and to obtain sufficient conditions for some qualitative properties of their solutions such as various types of stability.

Define the set

C_{q, ρ} ([a, b], R^{n}) = {υ : [a, b] \to R^{n} : f o r a n y t \in (a, b] t h e r e e x i s t s ({}_{a}^{RL}D^{q, ρ} υ) (t) < \infty} .

We will provide some results which are partial cases of the obtained ones in [12] and which will be used in our further considerations.

Lemma 2.(semigroup property) (Theorem 3.8, Corollary 3.10, Theorem 3.11, Lemma 3.12 [12]) If

ρ \in (0, 1], R e (q) > 0, R e (β) > 0

, and

υ \in C ([a, b]), b \leq \infty

, we have the following:

\begin{matrix} {}_{a}I^{q, ρ} ({}_{a}I^{β, ρ} υ) (t) = {}_{a}I^{β, ρ} ({}_{a}I^{q, ρ} υ) (t) = ({}_{a}I^{β, ρ} υ) (t) \\ ({}_{a}^{R}D^{β, ρ} {}_{a}I^{β, ρ} υ) (t) = {}_{a}I^{q - β, ρ} υ) (t), 0 < β < q, \\ ({}_{a}^{R}D^{q, ρ} {}_{a}^{R}D^{β, ρ} υ) (t) = υ (t) \\ {}_{a}I^{q, ρ} ({}_{a}^{R}D^{q, ρ} υ) (t) = υ (t) - \frac{({}_{a}I^{1 - q, ρ} υ) (a)}{ρ^{q - 1} Γ (q)} e^{\frac{ρ - 1}{ρ} (t - a)} {(t - a)}^{q - 1} . \end{matrix}

(6)

Lemma 3.(Lemma 2 [15] ) Let

ρ \in (0, 1], q \in (0, 1)

and

y \in C ([a, b], R)

.

(i): Let there exist a limit ${lim}_{t \to a +} (e^{\frac{1 - ρ}{ρ} t} {(t - a)}^{1 - q} y (t)) = c < \infty$ . Then $(_{a} I^{1 - q, ρ} y) (a) = c \frac{Γ (q)}{ρ^{1 - q}} e^{\frac{ρ - 1}{ρ} a} .$
(ii): Let $(_{a} I^{1 - q, ρ} y) (a +) = b < \infty$ . If there exists the limit $l i m_{t \to a +} (e^{\frac{1 - ρ}{ρ} t} {(t - a)}^{1 - q} y (t))$ , then ${lim}_{t \to a +} (e^{\frac{1 - ρ}{ρ} t} {(t - a)}^{1 - q} y (t)) = \frac{b ρ^{1 - q} e^{\frac{1 - ρ}{ρ} a}}{Γ (q)} .$

Lemma 4.

[11, Example 4.4] The solution of the initial value problem (IVP) for the scalar linear GPRLFDE

({}_{a}^{RL}D^{q, ρ} u) (t) = ρ^{q} λ u (t) + f (t), (_{a} I^{1 - q, ρ} u) (a +) = u_{0}, q \in (0, 1), ρ \in (0, 1]

has a solution

υ \in C_{q, ρ} ([a, \infty))

given by

\begin{matrix} u (t) & = u_{0} ρ^{1 - q} e^{\frac{ρ - 1}{ρ} (t - a)} {(t - a)}^{q - 1} E_{q, q} (λ {(t - a)}^{q}) \\ + ρ^{- q} \int_{a}^{t} E_{q, q} (λ {(t - s)}^{q}) e^{\frac{ρ - 1}{ρ} (t - s)} {(t - s)}^{q - 1} f (s) d s, \end{matrix}

where

E_{q, q} (t)

is the Mittag-Leffler function of two parameters,

λ \in R

.

Corollary 1.

{}_{a}^{RL}D^{q, ρ} (e^{\frac{ρ - 1}{ρ} (t - a)} {(t - a)}^{q - 1}) = 0, t > a

.

The proof of Corollary 1 follows from Lemma 4 with

λ = 0, f (t) \equiv 0

and the equality

E_{q, q} (0) = \frac{1}{Γ (q)}

.

Proposition 1.([11, Proposition 3.7]).

{}_{a}^{RL}D^{q, ρ} (e^{\frac{ρ - 1}{ρ} (t - a)}) = \frac{1}{ρ^{q} Γ (1 - q)} e^{\frac{ρ - 1}{ρ} (t - a)} {(t - a)}^{- q}, t > a

.

Remark 3.

In Theorem 2.1 [26] it is proved that tempered fractional integrals and derivatives could be theoretically expressed as an infinite series of classical Riemann-Liouville fractional integrals and derivatives. The same is true for GPFI and GPRLFD. However the applications of infinite series practically is very difficult. It requires independent study of differential equations with GPRLFD and finding applicable sufficient conditions for properties of their solutions.

3. Comparison results for GPRLFD

Lemma 5.

Let

υ \in C ([a, b], R)

,

a < b < \infty

be Lipshitz, and there exists a point

T \in (a, b]

such that

υ (T) = 0

, and

υ (t) < 0

, for

a \leq t < T .

Then, if the GPRLFD of υ exists for

t = T

with

q \in (0, 1), ρ \in (0, 1]

, then the inequality

({}_{a}^{RL}D^{q, ρ} υ) (t) |_{t = T} \geq 0

holds.

Proof.

Let

H (t) = \int_{a}^{t} e^{\frac{ρ - 1}{ρ} (t - s)} {(t - s)}^{- q} υ (s) d s

for

t \in [a, b]

. According to (2) we have

\begin{matrix} ({}_{a}^{R}D^{q, ρ} υ) (T) & = \frac{1}{ρ^{1 - q} Γ (1 - q)} ((1 - ρ) H (T) + ρ lim_{h \to 0 +} \frac{H (T - h) - H (T)}{h}) \\ = \frac{1}{ρ^{1 - q} Γ (1 - q)} lim_{h \to 0 +} ((1 - ρ) H (T) + ρ \frac{H (T - h) - H (T)}{h}) . \end{matrix}

(7)

There exists a constant

K > 0

such that

0 > v (s) = v (s) - v (T) \geq K (s - T)

for

s \in [T - h, T)

,

h > 0

, and

\begin{matrix} \int_{T - h}^{T} e^{\frac{1 - ρ}{ρ} s} {(T - s)}^{- q} υ (s) d s \geq - K \int_{T - h}^{T} e^{\frac{1 - ρ}{ρ} s} {(T - s)}^{1 - q} d s \\ = \frac{K e^{\frac{1 - ρ}{ρ} T}}{{(\frac{1 - ρ}{ρ})}^{2 - q}} (Γ (2 - q, h \frac{1 - ρ}{ρ}) - Γ (2 - q)) \equiv M (h), \end{matrix}

(8)

where

Γ (., .)

is the incomplete Gamma function and

lim_{h \to 0 +} \frac{Γ (2 - q, h \frac{1 - ρ}{ρ}) - Γ (2 - q)}{h} = 0 .

(9)

Thus, using

e^{\frac{ρ - 1}{ρ} h} {(T - h - s)}^{q} < {(T - s)}^{q}

for

s \in [T - h, T), h > 0, ρ \in (0, 1],

and

υ (s) < 0

on

[a, T)

we get

\begin{matrix} H (T - h) - H (T) = \int_{a}^{T} (e^{\frac{ρ - 1}{ρ} (T - h - s)} {(T - h - s)}^{- q} - e^{\frac{ρ - 1}{ρ} (T - s)} {(T - s)}^{- q}) υ (s) d s \\ - \int_{T - h}^{T} e^{\frac{ρ - 1}{ρ} (T - s)} (e^{\frac{1 - ρ}{ρ} h} {(T - h - s)}^{- q} - {(T - s)}^{- q}) υ (s) d s \\ + \int_{T - h}^{T} e^{\frac{ρ - 1}{ρ} (T - s)} {(T - s)}^{- q} υ (s) d s \\ \geq \int_{a}^{T} (e^{\frac{ρ - 1}{ρ} (T - h - s)} {(T - h - s)}^{- q} - e^{\frac{ρ - 1}{ρ} (T - s)} {(t - s)}^{- q}) υ (s) d s + M (h) e^{\frac{ρ - 1}{ρ} T} . \end{matrix}

(10)

Using (8), (9), and (10) we obtain

\begin{matrix} lim_{h \to 0 +} ((1 - ρ) H (T) + \frac{ρ}{h} (H (t) - H (T - h))) \\ \geq (1 - ρ) \int_{a}^{T} e^{\frac{ρ - 1}{ρ} (T - s)} {(t - s)}^{- q} υ (s) d s \\ + ρ \int_{a}^{T} lim_{h \to 0 +} \frac{e^{\frac{ρ - 1}{ρ} (T - h - s)} {(T - h - s)}^{- q} - e^{\frac{ρ - 1}{ρ} (T - s)} {(T - s)}^{- q}}{h} υ (s) d s \\ + lim_{h \to 0 +} \frac{M (h)}{h} ρ e^{\frac{ρ - 1}{ρ} T} \\ = (1 - ρ) \int_{a}^{T} e^{\frac{ρ - 1}{ρ} (T - s)} {(T - s)}^{- q} υ (s) d s + ρ \int_{a}^{T} \frac{d}{d T} (e^{\frac{ρ - 1}{ρ} (T - s)} {(T - s)}^{- q}) υ (s) d s \\ = (1 - ρ) \int_{a}^{T} e^{\frac{ρ - 1}{ρ} (T - s)} {(T - s)}^{- q} υ (s) d s \\ + \int_{a}^{T} ((ρ - 1) e^{\frac{ρ - 1}{ρ} (T - s)} {(T - s)}^{- q} - q ρ e^{\frac{ρ - 1}{ρ} (T - s)} {(t - s)}^{- 1 - q}) υ (s) d s \\ = - q ρ \int_{a}^{T} e^{\frac{ρ - 1}{ρ} (T - s)} {(T - s)}^{- 1 - q} υ (s) d s > 0 . \end{matrix}

(11)

□

Example 1.

Consider

υ (t) = e^{\frac{ρ - 1}{ρ} t} (t - 2)

for

t \in [0, 2]

,

ρ = 0.5

. Note

υ (t) < 0

for

t \in [0, 2)

,

υ (2) = 0

and for any

q \in (0, 1)

we have

\begin{matrix} ({}_{0}^{R}D^{q, ρ} υ) (t) |_{t = 2} = \frac{1}{0 . 5^{1 - q} Γ (1 - q)} (0.5 \int_{0}^{2} e^{- (2 - s)} {(s - 2)}^{- q} e^{- s} (2 - s) d s \\ + 0.5 \frac{d}{d t} \int_{0}^{t} e^{- (t - s)} {(t - s)}^{- q} e^{- s} {(s - 2) d s |}_{t = 2}) \\ = \frac{1}{0 . 5^{- q} Γ (1 - q)} (- e^{- 2} \int_{0}^{2} {(2 - s)}^{1 - q} d s + \frac{d}{d t} e^{- t} \int_{0}^{t} {(t - s)}^{- q} {(s - 2) d s |}_{t = 2}) \\ = \frac{1}{0 . 5^{- q} Γ (1 - q)} (- \frac{2^{2 - q}}{(2 - q) e^{2}} + \frac{d}{d t} (\frac{e^{- t} t^{1 - q} (t + 2 q - 4)}{2 - 3 q + q^{2}}) |_{t = 2}) \\ = \frac{1}{0 . 5^{- q} Γ (1 - q)} (- \frac{2^{2 - q}}{(2 - q) e^{2}} + 2^{- q} \frac{4 - 2 q^{2}}{(2 - 3 q + q^{2}) e^{2}}) > 0 . \end{matrix}

(12)

Remark 4.

A similar claim to Lemma 5 but for the Riemann-Liouville fractional derivatives is proved in [7].

Lemma 6.

Let

g \in C ([t_{0}, b] \times R, R)

, the functions

μ, ν \in C_{q, ρ} ([t_{0}, b], R)

be Lipshitz and satisfy the inequalities

({}_{t 0}^{R L}D^{q, ρ} μ) (t) < g (t, μ (t)), t \in (t_{0}, b], lim_{t \to t_{0} +} (e^{\frac{1 - ρ}{ρ} (t - t_{0})} {(t - t_{0})}^{1 - q} μ (t)) = μ_{0} \frac{ρ^{q - 1}}{Γ (q)},

(13)

and

({}_{t 0}^{R L}D^{q, ρ} ν) (t) \geq g (t, ν (t)), t \in (t_{0}, b], lim_{t \to t_{0} +} (e^{\frac{1 - ρ}{ρ} (t - t_{0})} {(t - t_{0})}^{1 - q} ν (t)) = ν_{0} \frac{ρ^{q - 1}}{Γ (q)} .

(14)

Then if

μ_{0} < ν_{0}

the inequality

μ (t) < ν (t), t \in (t_{0}, b]

holds.

Proof.

Suppose the contrary. Since,

μ_{0} < ν_{0}

and the functions

e^{\frac{1 - ρ}{ρ} (t - t_{0})} {(t - t_{0})}^{1 - q} μ (t)

and

e^{\frac{1 - ρ}{ρ} (t - t_{0})} {(t - t_{0})}^{1 - q} ν (t)

are continuous, there exists a point

τ \in (t_{0}, b]

such that

μ (t) < ν (t), t \in [t_{0}, τ)

and

μ (τ) = ν (τ)

. According to Lemma 5 for

υ = μ - ν, a = t_{0}

we obtain

0 = g (τ, μ (τ)) - g (τ, ν (τ)) > ({}_{t 0}^{R L}D^{q, ρ} μ) (t) |_{t = τ} - ({}_{t 0}^{R L}D^{q, ρ} ν) (t) {|_{t = τ} = ({}_{t 0}^{R L}D^{q, ρ} μ - ν) (t) |}_{t = τ} \geq 0

.

The obtained contradiction proves the claim. □

In the case when the initial condition contains the generalized proportional fractional integral we obtain the following result.

Corollary 2.

Let

g \in C ([t_{0}, b] \times R, R)

, the functions

μ, ν \in C_{q, ρ} ([t_{0}, b], R)

be Lipshitz and satisfy the inequalities

({}_{t 0}^{R L}D^{q, ρ} μ) (t) < g (t, μ (t)), t \in (t_{0}, b], ({}_{t 0}I^{1 - q, ρ} μ) (t) |_{t = t_{0}} = μ_{0},

(15)

and

({}_{t 0}^{R L}D^{q, ρ} ν) (t) \geq g (t, ν (t)), t \in (t_{0}, b], ({}_{t 0}I^{1 - q, ρ} ν) (t) |_{t = t_{0}} = ν_{0} .

(16)

Then if

μ_{0} < ν_{0}

the inequality

μ (t) < ν (t), t \in (t_{0}, b]

holds.

Corollary 3.

Let the functions

μ, ν \in C_{q, ρ} ([t_{0}, b], R)

be Lipshitz and satisfy the inequalities

\begin{matrix} ({}_{t 0}^{R L}D^{q, ρ} μ) (t) < ({}_{t 0}^{R L}D^{q, ρ} ν) (t), t \in (t_{0}, b], \\ lim_{t \to t_{0} +} (e^{\frac{1 - ρ}{ρ} (t - t_{0})} {(t - t_{0})}^{1 - q} μ (t)) < lim_{t \to t_{0} +} (e^{\frac{1 - ρ}{ρ} (t - t_{0})} {(t - t_{0})}^{1 - q} ν (t)) . \end{matrix}

(17)

Then the inequality

μ (t) < ν (t), t \in (t_{0}, b]

holds.

Lemma 7.

Let the function

ν \in C_{q, ρ} ([t_{0}, b], R)

and

ν^{2} \in C_{q, ρ} ([t_{0}, b], R)

. Then the inequality

({}_{t 0}^{R L}D^{q, ρ} ν^{2}) (t) \leq 2 ν (t) ({}_{t 0}^{R L}D^{q, ρ} ν) (t), t \in (t_{0}, b] .

(18)

holds.

Proof.

Fix a point

T \in (t_{0}, b]

and define the function

μ (s) = {(ν (T) - ν (s))}^{2}

for all

s \in [t_{0}, T]

. The function

(- μ (s))

satisfies all the conditions of Lemma 5 for

υ = - ν, a = t_{0}

and we obtain

({}_{t 0}^{R L}D^{q, ρ} (- μ) (t) |_{t = T} \geq 0

, i.e. applying Definition 2 we get

({}_{t 0}^{R L}D^{q, ρ} (μ) (t) |_{t = T} = \frac{1}{ρ^{1 - q} Γ (1 - q)} lim_{h \to 0 +} ((1 - ρ) H (T) + ρ \frac{H (T - h) - H (T)}{h}) \leq 0,

(19)

where

H (t) = \int_{t_{0}}^{t} e^{\frac{ρ - 1}{ρ} (t - s)} {(t - s)}^{- q} μ (σ) d σ, t \in [t_{0}, b]

.

Define the functions

P (t) = \int_{t_{0}}^{t} e^{\frac{ρ - 1}{ρ} (t - s)} {(t - s)}^{- q} ν (s) d s, t \in [t_{0}, b]

and

W (t) = \int_{t_{0}}^{t} e^{\frac{ρ - 1}{ρ} (t - s)} {(t - s)}^{- q} ν^{2} (s) d s, t \in [t_{0}, b] .

According to Definition 2 we have

\begin{matrix} ({}_{t 0}^{R L}D^{q, ρ} ν) (t) & = \frac{1}{ρ^{1 - q} Γ (1 - q)} ((1 - ρ) P (t) + ρ lim_{h \to 0 +} \frac{P (t - h) - P (t)}{h}) \\ = \frac{1}{ρ^{1 - q} Γ (1 - q)} lim_{h \to 0 +} ((1 - ρ) P (t) + ρ \frac{P (t - h) - P (t)}{h}) \end{matrix}

(20)

and

\begin{matrix} ({}_{t 0}^{R L}D^{q, ρ} ν^{2}) (t) & = \frac{1}{ρ^{1 - q} Γ (1 - q)} lim_{h \to 0 +} ((1 - ρ) W (t) + ρ \frac{W (t - h) - W (t)}{h}) . \end{matrix}

(21)

Note

ν^{2} (s) - 2 ν (T) ν (s) = {(ν (T) - ν (s))}^{2} - ν^{2} (s) = μ (s) - ν^{2} (s) \leq μ (s), s \in [t_{0}, T],

(22)

and

\begin{matrix} W (T) - 2 ν (T) P (T) = \int_{t_{0}}^{T} e^{\frac{ρ - 1}{ρ} (T - s)} {(T - s)}^{- q} (ν^{2} (σ) - 2 ν (T) ν (σ)) d σ \\ \leq \int_{t_{0}}^{T} e^{\frac{ρ - 1}{ρ} (T - s)} {(T - s)}^{- q} μ (σ) d σ = H (T), \\ W (T - h) - 2 ν (T) P (T - h) = \int_{t_{0}}^{T - h} e^{\frac{ρ - 1}{ρ} (T - h - s)} {(T - h - s)}^{- q} (ν^{2} (σ) - 2 ν (T) ν (σ)) d σ \\ \leq \int_{t_{0}}^{T - h} e^{\frac{ρ - 1}{ρ} (T - h - s)} {(T - h - s)}^{- q} μ (σ) d σ = H (T - h) . \end{matrix}

(23)

Then

\begin{matrix} ({}_{t 0}^{R L}D^{q, ρ} ν^{2}) (T) - 2 ν (T) ({}_{t 0}^{R L}D^{q, ρ} ν) (T) \\ = \frac{1}{ρ^{1 - q} Γ (1 - q)} lim_{h \to 0 +} ((1 - ρ) (W (T) - 2 ν (T) P (T)) \\ + ρ \frac{(W (T - h) - ν (T) P (T - h) - (W (T) - ν (T) P (T))}{h}) \\ = \frac{1}{ρ^{1 - q} Γ (1 - q)} lim_{h \to 0 +} ((1 - ρ) (W (T) - 2 ν (T) P (T)) \\ + ρ \frac{(W (T - h) - ν (T) P (T - h)) - (W (T) - ν (T) P (T))}{h}) \\ \leq \frac{1}{ρ^{1 - q} Γ (1 - q)} lim_{h \to 0 +} ((1 - ρ) H (T) + ρ \frac{H (T - h) - H (T)}{h}) \\ = ({}_{t 0}^{R L}D^{q, ρ} μ) (T) \leq 0 . \end{matrix}

(24)

Since

T \in (t_{0}, b]

is an arbitrary point, the claim is proved. □

Corollary 4.

Let the functions

ν_{i} \in C_{q, ρ} ([t_{0}, b], R)

and

ν_{i}^{2} \in C_{q, ρ} ([t_{0}, b], R), i = 1, 2, \dots, n

. Then the inequality

({}_{t 0}^{R L}D^{q, ρ} \sum_{i = 1}^{n} ν_{i}^{2} (\cdot)) (t) \leq 2 \sum_{i = 1}^{n} ν_{i} (t) ({}_{t 0}^{R L}D^{q, ρ} ν_{i} (\cdot)) (t), t \in (t_{0}, b] .

(25)

holds.

Remark 5.

Note several authors ([27],) used the inequality (25) for the Riemann-Liouville fractional derivative to prove the main results, citing the results from [3,28] which concern the Caputo fractional derivative.

4. BAM neural networks modeled by GPRLFD

The general model of the fractional-order BAM neural networks with the GPRLFD is described by the following state equations

\begin{matrix} ({}_{0}^{R L}D^{q, ρ} x_{i}) (t) = - a_{i} (t) x_{i} (t)) + \sum_{k = 1}^{m} b_{i, k} (t) f_{k} (y_{k} (t)) + I_{i} (t), t > 0, i = 1, 2, \dots, n, \\ ({}_{0}^{R L}D^{q, ρ} y_{j}) (t) = - c_{j} (t) y_{j} (t)) + \sum_{k = 1}^{n} d_{j, k} (t) g_{k} (y_{k} (t)) + J_{j} (t), t > 0, j = 1, 2, \dots, m, \end{matrix}

(26)

where

x_{i} (t)

and

y_{j} (t)

are the state variables of i-th neuron in the first layer at time t and the state variables of j-th neuron in the second layer at time t, respectively, n and m are the numbers of units in first and the second layer in the neural network,

{}_{0}^{R L}D^{q, ρ}

denotes the GPRLFD of order

q \in (0, 1), ρ \in (0, 1]

,

f_{i} (u)

and

g_{j} (u)

denote the activation functions,

b_{i, k} (t), d_{i, k} (t) : [0, \infty) \to R

denote the connection weight coefficients of the neurons,

a_{i} (t), c_{j} (t) : [0, \infty) \to (0, \infty)

represent the decay coefficients of signals at time t, and

I_{i} (t), J_{j} (t)

denotes the external inputs of the first and the second layers respectively at time t.

The initial conditions associated with the model (26) can be written in the form

({}_{0}I^{1 - q, ρ} x_{i}) (t) {|_{t = 0} = x_{i}^{0}, ({}_{0}I^{1 - q, ρ} y_{j}) (t) |}_{t = 0} = y_{j}^{0}, i = 1, 2 \dots, n, j = 1, 2, \dots, m .

(27)

Remark 6.

According to Lemma 3 the initial conditions (27) could be replaced by initial conditions of the type

lim_{t \to 0 +} (e^{\frac{1 - ρ}{ρ} t} t^{1 - q} x_{i} (t)) = x_{i}^{0} \frac{ρ^{q - 1}}{Γ (q)}, lim_{t \to 0 +} (e^{\frac{1 - ρ}{ρ} t} t^{1 - q} y_{i} (t)) = y_{i}^{0} \frac{ρ^{q - 1}}{Γ (p)} .

(28)

The goal of this paper is to study a special type of stability of the model (26) with initial conditions (27) or their equivalent (28).

Initially we will consider an Example to discuss some properties of the solutions of equations with the generalized proportional Riemann-Liouville fractional derivative.

Example 2.

Consider the initial value problem for the scalar differential equation with GPRLFD

({}_{0}^{R L}D^{q, ρ} u) (t) = - u (t), ({}_{0}I^{1 - q, ρ} u) (0 +) = u_{0},

where

q \in (0, 1), ρ \in (0, 1]

. According to Lemma 4 with

λ = - \frac{1}{ρ^{q}}, f (t, u) \equiv 0

, the solution is given by

u (t; u_{0}) = u_{0} ρ^{1 - q} e^{\frac{ρ - 1}{ρ} t} t^{q - 1} E_{q, q} (- {(\frac{t}{ρ})}^{q}) .

For any nonzero initial value we have

{lim}_{t \to 0 +} u (t; u_{0}) = \infty

and

{lim}_{t \to \infty} u (t; u_{0}) = 0

. Then for any

ϵ > 0

there exists

T = T (ϵ, u_{0})

such that

| u (t; u_{0}) | < ϵ

for

t > T

, but we could not find a nonzero initial value

u_{0}

such that

| u (t; u_{0}) | < ϵ

for

t \geq 0

.

The above example illustrates that any type of stability for differential equations with GPRLFD has to be defined in a different way than the ones for ordinary differential equations or for the differential equations with the Caputo type fractional derivative. The initial time has to be excluded. Some authors do not exclude the initial time (it is usually 0) and they do not note that for order

q \in (0, 1)

of the Riemann-Liouville fractional derivative of a constant depends on the expressions

t^{- q}

and

t^{q - 1}

which are not bounded for points close enough to the initial time 0 (see, for example [29,6,31]). Note the main concepts of stability of Riemann-Liouville fractional derivative are discussed and studied in [10].

We now introduce the class

Λ

of Lyapunov-like functions which will be used to investigate the stability of the model (26).

Definition 3.

Let

Δ \subset R^{n}, 0 \in Δ

. We will say that the function

V (x) : Δ \to R_{+}

belongs to the class

Λ (Δ)

if

V (x) \in C (Δ)

and it is locally Lipschitzian.

Remark 7.

Lyapunov functions could be applied with the quadratic function

V (x) = \sum_{i = 1}^{n} x_{i}^{2}, x = (x_{1}, x_{2}, \dots, x_{n})

for which Corollary 4 could be applied.

Note some authors when applying Lyapunov functions to fractional differential equations use the equality

({}_{0}^{R L}D^{q} | ν |) (t) = s i g n (ν (t)) ({}_{0}^{R L}D^{q} ν) (t)

(see, for example (31)[32]). However this equality is not true for all continuous functions

ν

.

Example 3.

Let

ν (t) = t - 1, t \in [0, 2]

,

q = 0.3, t_{0} = 0

. Then for

t \in (1, 2)

we get

\begin{matrix} {}_{0}^{R L}D_{t}^{0.3} | t - 1 | = \frac{1}{Γ (0.7)} \frac{d}{d t} \int_{0}^{t} {(t - s)}^{- 0.3} | s - 1 | d s \\ = \frac{1}{Γ (0.7)} \frac{d}{d t} \int_{0}^{t} {(t - s)}^{- 0.3} s i g n (s - 1) (s - 1) d s \\ = \frac{1}{Γ (0.7)} \frac{d}{d t} (- \int_{0}^{1} {(t - s)}^{- 0.3} (s - 1) d s + \int_{1}^{t} {(t - s)}^{- 0.3} (s - 1) d s) \\ \neq \frac{1}{Γ (0.7)} \frac{d}{d t} \int_{0}^{t} {(t - s)}^{- 0.3} (s - 1) d s = s i g n (t - 1) ({}_{0}^{R L}D_{t}^{0.3} (t - 1)) . \end{matrix}

(29)

In connection with above remark and example we will use the quadratic function as a Lyapunov function.

We will define the equilibrium of the neural network (26),(27). Usually, the equilibrium is a point, whose derivative is zero, and satisfies an appropriate algebraic equation. In the case when the generalized proportional derivatives (Caputo or Riemann-Liouville type) is taken for a nonzero constant, then the result is not equal to zero (which is true for the ordinary derivative and the Caputo derivative). For generalized proportional Caputo fractional derivative the equilibrium is defined by

C e^{\frac{ρ - 1}{ρ} t}

and studied for some types of stability in [17]. In the case of the Riemann-Liouville fractional derivative the equilibrium is defined as a constant in [30], but since

{}_{0}^{R L}D_{t}^{q} 1 = \frac{t^{- q}}{Γ (1 - q)}

, the algebraic system (12) [30] could not be satisfied for all

t \geq 0

since the right hand side part does not depend on t but the left hand side part depends on the variable

t^{- q}

which has no bound as

t \to 0 +

.

A similar situation occurs with the GPRLFD. We will study the stability behaviour of the model (26) in several cases.

4.1. General case of the model

Consider the model (26) in the general case when at least one of the of the coefficients and the external inputs in both layers are variable in time.

4.1.1. Variable in time equilibrium

Applying Corollary 1 with

a = 0

we will define the equilibrium of (26):

Definition 4.

The function

U^{*} (t) = (x^{*} (t), y^{*} (t)) : (0, \infty) \to R^{n + m},

where

x^{*} (t) = C e^{\frac{ρ - 1}{ρ} t} t^{q - 1}

and

y^{*} (t) = K e^{\frac{ρ - 1}{ρ} t} t^{q - 1}

with

C = (C_{1}, C_{2}, \dots, C_{n})

,

K = (K_{1}, K_{2}, \dots, K_{m}), C_{i} = c o n s t, i = 1, 2, \dots, n, K_{j} = c o n s t, j = 1, 2, \dots, m,

is called an equilibrium of the model of fractional order BAM neural networks (26) if the equalities

\begin{matrix} a_{i} (t) C_{i} e^{\frac{ρ - 1}{ρ} t} t^{q - 1} = \sum_{k = 1}^{m} b_{i, k} (t) f_{k} (K_{k} e^{\frac{ρ - 1}{ρ} t} t^{q - 1}) + I_{i} (t), t \geq 0, i = 1, 2 \dots, n \\ b_{j} (t) K_{j} e^{\frac{ρ - 1}{ρ} t} t^{q - 1} = \sum_{k = 1}^{n} d_{j, k} (t) g_{k} (C_{k} e^{\frac{ρ - 1}{ρ} t} t^{q - 1}) + J_{j} (t), t \geq 0, j = 1, 2 \dots, m \end{matrix}

(30)

hold.

Note that

{lim}_{t \to 0 +} (e^{\frac{1 - ρ}{ρ} t} t^{1 - q} U^{*} (t)) = U^{0}

where

U^{0} = (C, K)

and therefore, the equilibrium

U^{*} (t)

is a solution of the model (26), (27) with

x_{0} = C \frac{Γ (q)}{ρ^{1 - q}}

and

y_{0} = K \frac{Γ (q)}{ρ^{1 - q}}

.

Let

U^{*} (t)

be an equilibrium of (26) defined by Definition 4. Consider the change of variables

u (t) = x (t) - x^{*} (t), v (t) = y (t) - y^{*} (t), t \geq 0

, in system (26). Then we obtain

\begin{matrix} ({}_{0}^{R L}D^{q, ρ} u_{i}) (t) = - a_{i} (t) u_{i} (t) + \sum_{k = 1}^{m} b_{i, k} (t) F_{k} (t, v_{k} (t)), t > 0, i = 1, 2, \dots, n, \\ ({}_{0}^{R L}D^{q, ρ} v_{j}) (t) = - b_{j} (t) v_{j} (t) + \sum_{k = 1}^{n} d_{j, k} (t) G_{k} (t, u_{k} (t)), t > 0, j = 1, 2, \dots, m, \end{matrix}

(31)

where

F_{j} (t, u) = f_{j} (u + y_{j}^{*} (t)) - f_{j} (y_{j}^{*} (t))

,

G_{i} (t, u) = g_{i} (u + x_{i}^{*} (t)) - g_{i} (x_{i}^{*} (t))

,

i = 1, 2, \dots, n

,

j = 1, 2, \dots, m

for

t > 0, u \in R

.

The initial conditions associated with the revised model (31) can be written in the form

\begin{matrix} ({}_{0}I^{1 - q, ρ} u_{i}) (t) |_{t = 0} = x_{i}^{0} - C_{i} \frac{Γ (q)}{ρ^{1 - q}}, i = 1, 2 \dots, n, \\ ({}_{0}I^{1 - q, ρ} v_{j}) (t) |_{t = 0} = y_{j}^{0} - K_{j} \frac{Γ (q)}{ρ^{1 - q}}, j = 1, 2, \dots, m . \end{matrix}

(32)

Note the system (31) has a zero solution (with zero initial values).

Definition 5.

Let

α \in (0, 1)

and

ρ \in (0, 1]

. The equilibrium

U^{*} (t)

of (26) is called Mittag-Leffler exponentially stable in timeif there exists

T > 0

such that for any solution

U (t) = (x (t), y (t))

of (26), (27) the inequality

∥U (t) - U^{*} (t)∥ \leq Ξ (∥v^{0} - U^{0} \frac{Γ (q)}{ρ^{1 - q}}∥) e^{λ \frac{ρ - 1}{ρ} t} E_{q, q} (- λ t^{q}), t \geq T,

holds, where

v^{0} = (x^{0}, y^{0})

,

λ > 0

is a constant,

Ξ (s) \geq 0, Ξ (0) = 0,

is a given locally Lipschitz function.

Remark 8.

The Mittag-Leffler exponential stability in time of the equilibrium

(x^{*} (t), y^{*} (t))

of (26) implies that every solution

(x (t), y (t))

of the model (26) satisfies

{lim}_{t \to \infty} ∥ x (t) - x^{*} (t) ∥ = 0

,

{lim}_{t \to \infty} ∥ y (t) - y^{*} (t) ∥ = 0

for any initial values.

Theorem 1.

Let the following assumptions hold:

1.: $q \in (0, 1)$ and $ρ \in (0, 1]$ .
2.: The functions $a_{i}, c_{j} \in C (R_{+}, (0, \infty)), b_{i, j}, d_{j, i}, I_{i}, J_{j} \in C (R_{+}, R)$ , $i = 1, 2, \dots, n, j = 1, 2, \dots, m .$
3.: The activation functions $f_{i}, g_{j} \in C (R, R)$ , and there exist positive constants $μ_{i}, η_{j} i = 1, 2, \dots, n$ , such that $| f_{i} (v) - f_{i} (w) | \leq μ_{i} | v - w |$ and $| g_{j} (v) - g_{j} (w) | \leq η_{j} | v - w |$ for $v, w \in R$ , $i = 1, 2, \dots, n, j = 1, 2, \dots, m .$
4.: There exist constants $C_{i}, K_{j}, i = 1, 2, \dots, n, j = 1, 2, \dots, m,$ such that the algebraic system (30) is satisfied for all $t \geq 0 .$
5.: There exist constants $λ_{i}, μ_{j} > 0, i = 1, 2, \dots, n, j = 1, 2, \dots, m$ , such that the inequalities

$2 a_{i} (t) - \sum_{k = 1}^{m} | b_{i, k} (t) | - η_{i}^{2} \sum_{j = 1}^{m} | d_{j, i} (t) | \geq λ_{i}, t \geq 0, i = 1, 2, \dots, n$

$2 c_{j} (t) - \sum_{k = 1}^{n} | d_{j, k} (t) | - μ_{j}^{2} \sum_{i = 1}^{n} | b_{i, j} (t) | \geq μ_{j}, t \geq 0, j = 1, 2, \dots, m .$

hold.

Then, the equilibrium

U^{*} (t) = (C_{1}, C_{2}, \dots, C_{n}, K_{1}, K_{2}, \dots, K_{m}) e^{\frac{ρ - 1}{ρ} t} t^{q - 1}

of model (26) is Mittag-Leffler exponentially stable.

Remark 9.

Condition 4 of Theorem 1 guarantee the existence of the equilibrium

U^{*} (t)

of (26).

Proof.

Consider the Lyapunov function

V (x, y) = 0.5 \sum_{i = 1}^{n} x_{i}^{2} + 0.5 \sum_{j = 1}^{m} y_{j}^{2}, x \in R^{n}, y \in R^{m}

.

Let

U (\cdot) = (x (\cdot), y (\cdot)) \in R^{n + m}

be a solution of (26), (27) and let

X (t) = x (t) - x^{*} (t), Y (t) = y (t) - y^{*} (t), t \geq 0

where

U^{*} (\cdot) = (x^{*} (\cdot), y^{*} (\cdot))

.

Then according to Corollary 4 we get

\begin{matrix} ({}_{0}^{R L}D^{q, ρ} V (X (\cdot), Y (\cdot))) (t) = 0.5 \sum_{i = 1}^{n} ({}_{0}^{R L}D^{q, ρ} X_{i}^{2} (\cdot)) (t) + 0.5 \sum_{j = 1}^{m} ({}_{0}^{R L}D^{q, ρ} Y_{j}^{2} (\cdot)) (t) \\ \leq \sum_{i = 1}^{n} X_{i} (t) ({}_{0}^{R L}D^{q, ρ} X_{i} (\cdot)) (t) + \sum_{j = 1}^{m} Y_{j} (t) ({}_{0}^{R L}D^{q, ρ} Y_{j} (\cdot)) (t) \\ = \sum_{i = 1}^{n} (- a_{i} (t) X_{i}^{2} (t) + \sum_{k = 1}^{m} b_{i, k} (t) X_{i} (t) F_{k} (t, Y_{k} (t))) \\ + \sum_{j = 1}^{n} (- c_{j} (t) Y_{j}^{2} (t) + \sum_{k = 1}^{n} d j, k (t) Y_{j} (t) G_{k} (t, X_{k} (t))) \\ \leq \sum_{i = 1}^{n} (- a_{i} (t) X_{i}^{2} (t) + \sum_{k = 1}^{m} | b_{i, k} (t) | 0.5 (X_{i}^{2} (t) + F_{k}^{2} (t, Y_{k} (t))) \\ + \sum_{j = 1}^{m} (- c_{i} (t) Y_{j}^{2} (t) + \sum_{k = 1}^{n} | d_{j, k} (t) | 0.5 (Y_{j}^{2} (t) + G_{k}^{2} (t, X_{k} (t))) \\ \leq \sum_{i = 1}^{n} (- a_{i} (t) + 0.5 \sum_{k = 1}^{m} | b_{i, k} (t) | + 0.5 η_{i}^{2} \sum_{j = 1}^{m} | d_{j, i} (t) |) X_{i}^{2} (t) \\ + \sum_{j = 1}^{m} (- c_{j} (t) + 0.5 \sum_{k = 1}^{n} | d_{j, k} (t) | + 0.5 μ_{j}^{2} \sum_{i = 1}^{n} | b_{i, j} (t) |) Y_{j}^{2} (t) \\ \leq - γ V (X (t), Y (t)), \end{matrix}

(33)

where

γ = {min}_{i = 1, 2, \dots, n, j = 1, 2, \dots, m} {λ_{i}, μ_{j}}

.

Also, we have

\begin{matrix} lim_{t \to 0 +} (e^{\frac{1 - ρ}{ρ} t} t^{1 - q} V (X (t), Y (t))) = 0.5 lim_{t \to 0 +} (e^{\frac{1 - ρ}{ρ} t} t^{1 - q} (\sum_{i = 1}^{n} X_{i}^{2} + \sum_{j = 1}^{m} Y_{j}^{2})) \\ = 0.5 \sum_{i = 1}^{n} {(x_{i}^{0} \frac{ρ^{1 - q}}{Γ (q)} - C_{i})}^{2} + 0.5 \sum_{j = 1}^{m} {(y_{j}^{0} \frac{ρ^{1 - q}}{Γ (q)} - K_{j})}^{2} \\ = 0.5 {(\frac{ρ^{1 - q}}{Γ (q)})}^{2} {(∥v^{0} - U^{0} \frac{Γ (q)}{ρ^{1 - q}}∥)}^{2} < u_{0} \frac{ρ^{1 - q}}{Γ (q)}, \end{matrix}

(34)

where

u_{0} = \frac{ρ^{1 - q}}{Γ (q)} {(∥v^{0} - U^{0} \frac{Γ (q)}{ρ^{1 - q}}∥)}^{2}

,

v^{0} = (x^{0}, y^{0}), U^{0} = (C, K)

.

Consider the scalar equation

({}_{0}^{R L}D^{q, ρ} u (\cdot)) (t) = - γ u (t)

with the initial condition

({}_{0}I^{1 - q, ρ} u) (t) |_{t = 0} = u_{0}

. According to Lemma 4 it has a solution

u (t) = u_{0} ρ^{1 - q} e^{\frac{1 - ρ}{ρ} t} t^{q - 1} E_{q, q} (- γ {(\frac{t}{ρ})}^{q}) .

Since

{lim}_{t \to \infty} t^{q - 1} = 0

there exists

T = T (q) > 0

such that

t^{q - 1} \leq 1

for

t \geq T

.

According to Corollary 3 we obtain for

t \geq T

V (X (t), Y (t)) < u (t) \leq \frac{ρ^{2 - 2 q}}{Γ (q)} {(∥v^{0} - U^{0} \frac{Γ (q)}{ρ^{1 - q}}∥)}^{2} e^{\frac{1 - ρ}{ρ} t} E_{q, q} (- γ {(\frac{t}{ρ})}^{q}) .

Thus, the equilibrium

U^{*} (\cdot)

is Mittag-Leffler exponentially stable with

Ξ (u) = \frac{ρ^{2 - 2 q}}{Γ (q)} u^{2}

.

□

4.1.2.Constant equilibrium

We define the equilibrium of the model (26) as a constant vector in the form

V^{*} = (C_{1}, C_{2}, \dots, C_{n + m})

.

From Eq. (5) using CAS Wolfram Mathematica we obtain

({}_{a}D^{q, ρ} 1) (t) = ρ^{q} e^{\frac{ρ - 1}{ρ} t} ({}_{a}^{R L}D_{t}^{q} (e^{\frac{1 - ρ}{ρ} t})) = {(1 - ρ)}^{q} (1 - \frac{Γ (- q, \frac{1 - ρ}{ρ} t)}{Γ (- q)})

(35)

where

Γ (a, x) = \int_{x}^{\infty} t^{a - 1} e^{- t} d t

is the upper incomplete gamma function. It is clear that

{lim}_{t \to 0} \frac{Γ (- q, \frac{1 - ρ}{ρ} t)}{Γ (- q)} = \infty

and

{lim}_{t \to \infty} \frac{Γ (- q, \frac{1 - ρ}{ρ} t)}{Γ (- q)} = 0

for

q \in (0, 1)

and

ρ \in (0, 1]

.

Based on (35) we will define the constant equilibrium of (26):

Definition 6.

The constant vector

V^{*} = (C_{1}, C_{2}, \dots, C_{n + m})

is called a constant equilibrium of the model of fractional order BAM neural networks (26) if the equalities

\begin{matrix} C_{i} ({(1 - ρ)}^{q} (1 - \frac{Γ (- q, \frac{1 - ρ}{ρ} t)}{Γ (- q)}) + a_{i} (t)) = \sum_{k = 1}^{m} b_{i, k} (t) f_{k} (C_{n + k}) + I_{i} (t), t \geq 0, i = 1, 2 \dots, n \\ C_{n + j} ({(1 - ρ)}^{q} (1 - \frac{Γ (- q, \frac{1 - ρ}{ρ} t)}{Γ (- q)}) + b_{j} (t)) = \sum_{k = 1}^{n} d_{j, k} (t) g_{k} (C_{k}) + J_{j} (t), t \geq 0, j = 1, 2 \dots, m \end{matrix}

(36)

hold.

Note that

{lim}_{t \to 0 +} (e^{\frac{1 - ρ}{ρ} t} t^{1 - q} V^{*}) = 0

and therefore, the equilibrium

V^{*}

is a solution of the model (26), (27) with

x_{0} = y_{0} = 0

.

Let

V^{*}

be a constant equilibrium of (26) defined by Definition 6. Consider the change of variables

u_{i} (t) = x_{i} (t) - C_{i}, v_{j} (t) = y_{j} (t) - C_{n + j}, t \geq 0

, in system (26). Then applying (35) and (36) we obtain

\begin{matrix} ({}_{0}^{R L}D^{q, ρ} u_{i}) (t) = - a_{i} (t) u_{i} (t) + \sum_{k = 1}^{m} b_{i, k} (t) F_{k} (v_{k} (t)), t > 0, i = 1, 2, \dots, n, \\ ({}_{0}^{R L}D^{q, ρ} v_{j}) (t) = - b_{j} (t) v_{j} (t) + \sum_{k = 1}^{n} d_{j, k} (t) G_{k} (u_{k} (t)), t > 0, j = 1, 2, \dots, m, \end{matrix}

(37)

where

F_{j} (u) = f_{j} (u + C_{n + j}) - f_{j} (C_{n + j})

,

G_{i} (u) = g_{i} (u + C_{i}) - g_{i} (C_{i})

,

u \in R

,

i = 1, 2, \dots, n

,

j = 1, 2, \dots, m

,

u \in R

.

Note the system (31) has a zero solution (with zero initial values).

Definition 7.

Let

α \in (0, 1)

and

ρ \in (0, 1]

. The constant equilibrium

V^{*}

of (26) is called Mittag-Leffler exponentially stable in timeif there exists

T > 0

such that for any solution

U (t) = (x (t), y (t))

of (26), (27) the inequality

∥U (t) - V^{*}∥ \leq Ξ (∥v^{0}∥) e^{λ \frac{ρ - 1}{ρ} t} E_{q, q} (- λ t^{q}), t \geq T,

holds, where

v^{0} = (x^{0}, y^{0})

,

λ > 0

is a constant,

Ξ (s) \geq 0, Ξ (0) = 0,

is a given locally Lipschitz function.

Theorem 2.

Let the conditions of Theorem 1 be satisfied. Then, the constant equilibrium

V^{*} = (C_{1}, C_{2}, \dots, C_{n + m})

of model (26) is Mittag-Leffler exponentially stable.

The proof is similar to the one in Theorem 1 so we omit it.

0

section.2Partial case - constant coefficient and constant inputs in the model

Let all coefficients in both layers as well as the external inputs are constants, i.e.

a_{i} (t) \equiv a_{i}, c_{j} (t) \equiv c_{j}, b_{i, k} (t) \equiv b_{i, k}, d_{j, k} (t) \equiv d_{j, k}, I_{i} (t) \equiv I_{i}, J_{j} (t) \equiv J_{j}, i = 1, 2, \dots, n, j = 1, 2, \dots, m

.

Then for variable in time equilibrium the algebraic system (30) reduces to

\begin{matrix} a_{i} C_{i} e^{\frac{ρ - 1}{ρ} t} t^{q - 1} = \sum_{k = 1}^{m} b_{i, k} f_{k} (K_{k} e^{\frac{ρ - 1}{ρ} t} t^{q - 1}) + I_{i}, t \geq 0, i = 1, 2 \dots, n, \\ b_{j} K_{j} e^{\frac{ρ - 1}{ρ} t} t^{q - 1} = \sum_{k = 1}^{n} d_{j, k} g_{k} (C_{k} e^{\frac{ρ - 1}{ρ} t} t^{q - 1}) + J_{j}, t \geq 0, j = 1, 2 \dots, m . \end{matrix}

(38)

The system (38) could have a solution

(C_{1}, C_{2}, \dots, C_{n}, K_{1}, \dots, K_{m})

, i.e. the model (26) could have a variable in time equilibrium.

For constant equilibrium the algebraic system (36) reduces to

\begin{matrix} C_{i} {(1 - ρ)}^{q} (1 - \frac{Γ (- q, \frac{1 - ρ}{ρ} t)}{Γ ((- q)}) = - a_{i} C_{i} + \sum_{k = 1}^{m} b_{i, k} f_{k} (C_{n + k}) + I_{i}, t \geq 0, i = 1, 2 \dots, n \\ C_{n + j} {(1 - ρ)}^{q} (1 - \frac{Γ (- q, \frac{1 - ρ}{ρ} t)}{Γ (- q)}) = - b_{j} C_{n + j} + \sum_{k = 1}^{n} d_{j, k} g_{k} (C_{k}) + J_{j}, t \geq 0, j = 1, 2 \dots, m \end{matrix}

(39)

If there is no external inputs, i.e.

I_{i} = 0, J_{j} = 0

and

f_{i} (0) = 0, g_{j} (0) = 0, i - 1, 2, \dots, n, j = 1, 2, \dots, m,

then the system (39) has a zero solution

C_{k} = 0, k = 1, 2, \dots, n + m

, i.e. the model (26) has a zero equilibrium.

If there is external inputs, i.e. at least one of

I_{i}, J_{j}

are nonzero, then the system (39) has no solution, thus the model has no constant equilibrium.

5. Examples

Example 4.

Consider the following BAM neural networks of two layers with two neurons with the GPRLFD:

\begin{matrix} ({}_{0}^{R L}D^{α, ρ} x_{1}) (t) = - x_{1} (t) + \frac{0.1}{1 + e^{- y_{1} (t)}} - 0.05, \\ ({}_{0}^{R L}D^{α, ρ} x_{2}) (t) = - (1 + e^{\frac{ρ - 1}{ρ} t}) x_{2} (t) - e^{\frac{ρ - 1}{ρ} t} \frac{1}{1 + e^{- y_{2} (t)}} + e^{\frac{ρ - 1}{ρ} t}, \\ ({}_{0}^{R L}D^{α, ρ} y_{1}) (t) = - (1 + 0.5 e^{\frac{ρ - 1}{ρ} t}) y_{1} (t) - e^{\frac{ρ - 1}{ρ} t} \frac{1}{1 + e^{- x_{1} (t)}} + \frac{1}{1 + e^{- x_{2} (t)}} + 0.5 (e^{\frac{ρ - 1}{ρ} t} - 1) \\ ({}_{0}^{R L}D^{α, ρ} y_{2}) (t) = - (1.5 + e^{\frac{ρ - 1}{ρ} t}) y_{2} (t) - \frac{1}{1 + e^{- y_{2} (t)}} + 0.5, \end{matrix}

(40)

with coefficients

a_{1} (t) = 1, a_{2} (t) = 1 + e^{\frac{ρ - 1}{ρ} t}, c_{1} (t) = 1 + 0.5 e^{\frac{ρ - 1}{ρ} t}, c_{2} (t) = 1.5 + e^{\frac{ρ - 1}{ρ} t}

, the activation functions

f_{k} (u), g_{k} (u) = \frac{1}{1 + e^{- u}} > 0, k = 1, 2, u \in R,

are equal to the sigmoid function with

μ_{k} = η_{k} = 0.25

, the external inputs are given by

I_{1} (t) = - 0.05, I_{2} (t) = e^{\frac{ρ - 1}{ρ} t}, J_{1} (t) = 0.5 (e^{\frac{ρ - 1}{ρ} t} - 1), J_{2} (t) = 0.5,

and

B = {b_{i, k} (t)} = [\begin{matrix} 0.1 & 0 \\ 0 & - e^{\frac{ρ - 1}{ρ} t} \end{matrix}], D = {d_{i, k} (t)} = [\begin{matrix} - e^{\frac{ρ - 1}{ρ} t} & 1 \\ 0 & - 1 \end{matrix}] .

Then the algebraic system (30) reduces to

\begin{matrix} a_{1} (t) C_{1} e^{\frac{ρ - 1}{ρ} t} t^{q - 1} = \frac{b_{1, 1}}{1 + e^{- K_{1} e^{\frac{ρ - 1}{ρ} t} t^{q - 1}}} + I_{1} (t), t \geq 0, \\ a_{2} (t) C_{2} e^{\frac{ρ - 1}{ρ} t} t^{q - 1} = \frac{b_{2, 2}}{1 + e^{- K_{2} e^{\frac{ρ - 1}{ρ} t} t^{q - 1}}} + I_{2} (t), t \geq 0, \\ c_{1} (t) K_{1} e^{\frac{ρ - 1}{ρ} t} t^{q - 1} = d_{1, 1} (t) \frac{1}{1 + e^{- C_{1} e^{\frac{ρ - 1}{ρ} t} t^{q - 1}}} + d_{1, 2} (t) \frac{1}{1 + e^{- C_{2} e^{\frac{ρ - 1}{ρ} t} t^{q - 1}}} + J_{1} (t), \\ c_{2} (t) K_{2} e^{\frac{ρ - 1}{ρ} t} t^{q - 1} = d_{2, 1} (t) \frac{1}{1 + e^{- C_{1} e^{\frac{ρ - 1}{ρ} t} t^{q - 1}}} + d_{2, 2} (t) \frac{1}{1 + e^{- C_{2} e^{\frac{ρ - 1}{ρ} t} t^{q - 1}}} + J_{2} (t), t \geq 0 . \end{matrix}

(41)

The system (41) has a zero solution

C_{1} = C_{2} = K_{1} = K_{2} = 0

.

Then, for

ρ \in (0, 1], q \in (0, 1)

, system (40) has the equilibrium

U^{*} (t) = (0, 0, 0, 0)

.

Also, condition 5 of Theorem 1 is satisfied because of the inequalities

2 a_{1} (t) - | b_{1, 1} (t) | - | b_{1, 2} (t) | - η_{1}^{2} | d_{1, 1} (t) | - η_{2}^{2} | d_{2, 1} (t) | \geq λ_{1} = 1.8375, t \geq 0,

2 a_{2} (t) - | b_{2, 1} (t) | - | b_{2, 2} (t) | - η_{1}^{2} | d_{1, 2} (t) | - η_{2}^{2} | d_{2, 2} (t) | \geq λ_{2} = 1.875, t \geq 0,

2 c_{1} (t) - | d_{1, 1} (t) | - | d_{1, 2} (t) | - μ_{1}^{2} | b_{1, 1} (t) | + μ_{2}^{2} | b_{2, 1} (t) | \geq μ_{1} = 0.99375, t \geq 0,

2 c_{2} (t) - | d_{2, 1} (t) | - | d_{2, 2} (t) | - μ_{2}^{2} | b_{1, 2} (t) | + μ_{2}^{2} | b_{2, 2} (t) | \geq μ_{2} = 1, t \geq 0,

According to Theorem 2 the zero equilibrium of (40) is Mittag-Leffler exponentially stable, i.e. every solution

(x_{1} (\cdot), y_{2} (\cdot), y_{1} (\cdot), y_{2} (\cdot))

of (40) with the initial condition

({}_{0}I^{1 - q, ρ} x_{i}) (t) {|_{t = 0} = x_{i}^{0}, ({}_{0}I^{1 - q, ρ} y_{j}) (t) |}_{t = 0} = y_{j}^{0}, i, j = 1, 2,

satisfies the inequality

\sqrt{x_{1}^{2} (t) + x_{2}^{2} (t) + y_{1}^{2} (t) + y_{2}^{2} (t)} \leq \frac{ρ^{2 - 2 q}}{Γ (q)} ({(x_{1}^{0})}^{2} + {(x_{2}^{0})}^{2} + {(y_{1}^{0})}^{2} + {(y_{2}^{0})}^{2}) E_{q, q} (- \frac{0.99375}{ρ^{q}} t^{q})

with

γ = m i n (1.8375, 1.875, 0.99375, 1)

.

Author Contributions

Conceptualization, R.P.A, S.H, and D. O’R.; methodology, R.P.A, S.H, and D. O’R.; formal analysis, R.P.A, S.H, and D. O’R.; investigation, R.P.A, S.H, and D. O’R.; writing—original draft preparation, R.P.A, S.H, and D. O’R.; writing—review and editing, R.P.A, S.H, and D. O’R.. All authors have read and agreed to the published version of the manuscript.

Funding

S.H. is supported by the Bulgarian National Science Fund under Project KP-06-PN62/1.

Institutional Review Board Statement

Not applicable.

Informed Consent Statement

Not applicable.

Conflicts of Interest

The authors declare no conflict of interest.

References

G. Wang, K. Pei, Y. Q. Chen, Stability analysis of nonlinear Hadamard fractional differential system, J. Franklin Inst.356, 12, (2019), 6538-654. [CrossRef]
Li Y., Chen Y., Podlubny I. Stability of fractional-order nonlinear dynamic systems: Lyapunov direct method and generalized Mittag Leffler stability, Comput. Math. Appl. 59 (2010) 1810–1821. [CrossRef]
N. Aguila-Camacho, M. A. Duarte-Mermoud, J. A. Gallegos, Lyapunov functions for fractional order systems, Comm. Nonlinear Sci. Numer. Simul., 19, (2014) 2951-2957. [CrossRef]
D. Qian, Ch. Li, R. P. Agarwal, P. J.Y. Wong, Stability analysis of fractional differential system with Riemann–Liouville derivative, Math. Comput. Modell. 52 (2010) 862–874. [CrossRef]
Hristova, S.; Tersian, S.; Terzieva, R. Lipschitz Stability in Time for Riemann–Liouville Fractional Differential Equations, Fractal Fract., 5, 37, (2021). [CrossRef]
Liu S., Wu X., Zhou X.F., Jiang W. Asymptotical stability of Riemann-Liouville fractional nonlinear systems,Nonlinear Dynamics, 86 (2016), 65–71. [CrossRef]
Devi, J.V.; Rae, F.A.M.; Drici, Z. Variational Lyapunov method for fractional differential equations, Comput. Math. Appl. 64, (2012) 2982–2989. [CrossRef]
Agarwal, R., Hristova, S., O’Regan, D. Practical stability for Riemann–Liouville delay fractional differential equations. Arab. J. Math., 10, 271–283 (2021). [CrossRef]
Benchohra, M.; Bouriah, S.; Nieto, J. J. Existence and Ulam stability for nonlinear implicit differential equations with Riemann-Liouville fractional derivative, Demonstratio Mathematica, 52, 1, (2019), 437–450. [CrossRef]
Agarwal, R.; Hristova, S.; O’Regan, D. Stability Concepts of Riemann-Liouville Fractional-Order Delay Nonlinear Systems. Mathematics 2021, 9, 435. [Google Scholar] [CrossRef]
Jarad, F.; Abdeljawad, T.; Alzabut, J. Generalized fractional derivatives generated by a class of local proportional derivatives. Eur. Phys. J. Spec. Top. 2017, 226, 3457–3471. [Google Scholar] [CrossRef]
Jarad, F.; Abdeljawad, T. Generalized fractional derivatives and Laplace transform. Discret. Contin. Dyn. Syst. Ser. S 2020, 13, 709–722. [Google Scholar] [CrossRef]
Abbas, M.I.; Hristova, S. On the Initial Value Problems for Caputo-Type Generalized Proportional Vector-Order Fractional Differential Equations. Mathematics 2021, 9, 2720. [Google Scholar] [CrossRef]
D. Boucenna, D. Baleanu, A. B. Makhlouf, A.M. Nagy, Analysis and numerical solution of the generalized proportional fractional Cauchy problem, Applied Numerical Mathematics 167 (2021) 173–186. [CrossRef]
Hristova, S.; Abbas, M.I. Explicit Solutions of Initial Value Problems for Fractional Generalized Proportional Differential Equations with and without Impulses. Symmetry, 13, (2021), 996. [CrossRef]
M. I. Abbas, Controllability and Hyers-Ulam stability results of initial value problems for fractional differential equations via generalized proportional-Caputo fractional derivative, Miskolc Mathematical Notes 22 2, (2021), 491–502. [CrossRef]
Almeida, R.; Agarwal, R. P.; Hristova, S.; O’Regan, D., Quadratic Lyapunov Functions for Stability of the Generalized Proportional Fractional Differential Equations with Applications to Neural Networks. Axioms, 10, 4, (2021) 322. [CrossRef]
S. Hristova; M. I. Abbas, Fractional differential equations with anti-periodic fractional integral boundary conditions via the generalized proportional fractional derivatives, AIP Conference Proceedings 2459, 030014 (2022). [CrossRef]
J. Alzabut, J. Viji, V. Muthulakshmi, W. Sudsutad, Oscillatory Behavior of a Type of Generalized Proportional Fractional Differential Equations with Forcing and Damping Terms, Mathematics 2020, 8, 1037. [CrossRef]
W. Sudsutad, J. Alzabut, C. Tearnbucha, C.Thaiprayoon, On the oscillation of differential equations in frame of generalized proportional fractional derivatives, AIMS Mathematics, 5 (2) (2020) 856–871. [CrossRef]
M.I. Abbas, M. Ghaderi, S. Rezapour, S. T. M. Thabet, On a coupled system of fractional differential equations via the generalized proportional fractional derivatives, J. Function Spaces, 2022, (2022), Article ID 4779213. [CrossRef]
Sh. Das, Functional Fractional Calculus, Springer-Verlag Berlin Heidelberg, 2011.
I. Podlubny,Fractional Differential Equations, Academic Press, San Diego, 1999.
G. Samko, A.A. Kilbas, O.I. Marichev, Fractional Integrals and Derivatives: Theory and Applications, Gordon and Breach, 1993.
F. Sabzikar, M.M. Meerschaert, J. Chen, Tempered fractional calculus,J. Comput. Phys. 293 (2015) 14–28. [CrossRef]
A. Fernandez, C. Ustaoglu, On some analytic properties of tempered fractional calculus, J. Comput. Appl. Math. 366 (2020) 112400. [CrossRef]
S. Liu, X. Wu, Y.-J. Zhang, Asymptotical stability of Riemann–Liouville fractional neutral systems, Appl. Math. Letters, 69, 2017, 168-173. [CrossRef]
M. A. Duarte-Mermoud, N. Aguila-Camacho, J. A. Gallegos, R. Castro-Linares, Using general quadratic Lyapunov functions to prove Lyapunov uniform stability for fractional order systems, Commun. Nonlinear Sci. Numer. Simulat., 22, 2015, 650-659. [CrossRef]
R. Zhang, Sh. Yang, Sh. Feng, Stability analysis of a class of nonlinear fractional differential systems with Riemann-Liouville derivative, EE/CAA J. Autom. Sinica. [CrossRef]
J. Alidousti, R. Khoshsiar Ghaziani, A. B. Eshkaftaki, Stability analysis of nonlinear fractional differential order systems with Caputo and Riemann–Liouville derivatives, Turk. J. Math., bf 41, (2017) 1260–1278. [CrossRef]
Z. Qin, R. Wu, Y. Lu, Stability analysis of fractionalorder systems with the Riemann–Liouville derivative, Systems Sci. Control Eng.: An Open Access J., 2, 2014, 727–731. [CrossRef]
H. Zhang, R. Ye, J. Cao, Ah. Alsaedi, Existence and Globally Asymptotic Stability of Equilibrium Solution for Fractional-Order Hybrid BAM Neural Networks with Distributed Delays and Impulses, Complexity, 2017, (2017), Art. ID 6875874. [CrossRef]
Bohner, M., Hristova, S. Stability for generalized Caputo proportional fractional delay integro-differential equations. Bound Value Probl 2022, 14 (2022). [CrossRef]
A. Fernandez, M.A. Ozarslan, D. Baleanu, On fractional calculus with general analytic kernels, Appl. Math. Comput. 354 (2019) 248–265. [CrossRef]
Gu, C.-Y.; Zheng, F.-H.; Shiri, B. Mittag–Leffler stability analysis of tempered fractional neural networks with short memory and variable-order, Fractals, bf 8, (2021), 2140029. [CrossRef]
Meerschaert, M.M.; Sabzikar, F.; Phanikumar, M.S.; Zeleke, A. Tempered fractional time series model for turbulence in geophysical flows, J. Stat. Mech. Theory Exper., 9, (2014), 9023. [CrossRef]
Syed Ali M., Narayanan G., Shekher V., Alsaedi A. and Ahmad B., Global Mittag–Leffler stability analysis of impulsive fractional-order complex-valued BAM neural networks with time varying delays, Commun. Nonlinear Sci. Numer. Simul., 83 (2020) Art. 105088. [CrossRef]

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

© 2023 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

Copyright: This open access article is published under a Creative Commons CC BY 4.0 license, which permit the free download, distribution, and reuse, provided that the author and preprint are cited in any reuse.

Mittag-Leffler Type Stability of BAM Neural Networks Modeled by the Generalized Proportional Riemann-Liouville Fractional Derivative

Abstract

Keywords:

Subject:

1. Introduction

2. Some notes on Fractional Calculus

3. Comparison results for GPRLFD

4. BAM neural networks modeled by GPRLFD

4.1. General case of the model

4.1.1. Variable in time equilibrium

4.1.2.Constant equilibrium

0

5. Examples

Author Contributions

Funding

Institutional Review Board Statement

Informed Consent Statement

Conflicts of Interest

References

MDPI Initiatives

Important Links

Subscribe