Analysis of Fractional-Order Nonlinear Dynamic Systems with General Analytic Kernels: Lyapunov Stability and Inequalities

mathematics

Article Analysis of Fractional-Order Nonlinear Dynamic Systems with General Analytic Kernels: Lyapunov Stability and Inequalities

Oscar Martínez-Fuentes 1,† , Fidel Meléndez-Vázquez 2,† , Guillermo Fernández-Anaya 2,*,† and José Francisco Gómez-Aguilar 3,†

1 School of Engineering, Universidad Anáhuac Veracruz, Campus Xalapa, Circuito Arco Sur s/n, Col. Lomas Verdes, Xalapa 91098, Mexico; [email protected] 2 Departamento de Física y Matemáticas, Universidad Iberoamericana Ciudad de México, Prol. Paseo de la Reforma 880, Lomas de Santa Fe, Alvaro Obregon, Mexico City 01219, Mexico; [email protected] 3 CONACyT-Tecnológico Nacional de México/CENIDET, Interior Internado Palmira s/n, Col. Palmira, Cuernavaca 62490, Mexico; [email protected] * Correspondence: [email protected] † These authors contributed equally to this work.

Abstract: In this paper, we study the recently proposed fractional-order operators with general analytic kernels. The kernel of these operators is a locally uniformly convergent power series that can be chosen adequately to obtain a family of fractional operators and, in particular, the main existing fractional derivatives. Based on the conditions for the Laplace transform of these operators, in this paper, some new results are obtained—for example, relationships between Riemann–Liouville and Caputo derivatives and inverse operators. Later, employing a representation for the product of two functions, we determine a form of calculating its fractional derivative; this result is essential Citation: Martínez-Fuentes, O.; due to its connection to the fractional derivative of Lyapunov functions. In addition, some other Meléndez-Vázquez, F.; new results are developed, leading to Lyapunov-like theorems and a Lyapunov direct method that Fernández-Anaya, G.; serves to prove asymptotic stability in the sense of the operators with general analytic kernels. The Gómez-Aguilar, J.F. Analysis of FOB-stability concept is introduced, which generalizes the classical Mittag–Lefﬂer stability for a wide Fractional-Order Nonlinear Dynamic class of systems. Some inequalities are established for operators with general analytic kernels, which Systems with General Analytic Kernels: Lyapunov Stability and generalize others in the literature. Finally, some new stability results via convex Lyapunov functions Inequalities. Mathematics 2021, 9, 2084. are presented, whose importance lies in avoiding the calculation of fractional derivatives for the https://doi.org/10.3390/math9172084 stability analysis of dynamical systems. Some illustrative examples are given.

Academic Editor: Duarte Valério Keywords: nonlinear fractional-order systems; general analytic kernels; FOB-stability; generalized lyapunov stability; inequalities Received: 24 July 2021 Accepted: 26 August 2021 Published: 28 August 2021 1. Introduction Publisher’s Note: MDPI stays neutral Fractional calculus generalizes the well-known classical calculus to operators with with regard to jurisdictional claims in non-integer orders. In this sense, this theory has had as a main purpose to extend classical published maps and institutional affil- mathematical results and develop a new formulation of calculus; this has been done along iations. with the development of integer calculus. However, fractional calculus did not begin to have practical applications until the 1970s. The first theoretical and applied results in fractional calculus involve the Riemann–Liouville integral and the Riemann–Liouville and Caputo derivatives, which are still widely studied and used [1–10]; nevertheless, other Copyright: © 2021 by the authors. definitions of non-integer operators have been developed. Some recent classifications of Licensee MDPI, Basel, Switzerland. families of operators appear in [11–15]. This article is an open access article Moreover, some fractional operators have been proposed, which seek to generalize distributed under the terms and several other existing operators. Some of these generalizations have been given in [16–21]. conditions of the Creative Commons In this sense, Fernandez et al. introduced a family of operators that generalizes a wide vari- Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ ety of fractional operators [22], such as the classical Riemann–Liouville and Caputo opera- 4.0/). tors, and the operators with a non-singular kernel; among these, the Caputo–Fabrizio [23]

Mathematics 2021, 9, 2084. https://doi.org/10.3390/math9172084 https://www.mdpi.com/journal/mathematics Mathematics 2021, 9, 2084 2 of 29

and Atangana–Baleanu [24] derivatives are found between other definitions. An interesting peculiarity of the operators with non-singular kernel is that they are simple, yet they include many operators that are currently widely used; furthermore, they can be expressed in terms of the classical Riemann–Liouville operators by means of power series expansions. The fractional operators with a non-singular kernel have been widely studied recently [25–30] and used in multiple applications and modeling of systems and phenomena [31–39]. Thus, inspired by the general analytic kernel (GAK) fractional operators defined in [22], in this work, we extend several well-known results for classical operators (such as the Caputo and Riemann–Liouville derivatives) to these new operators, including the Leibniz rule and some widely used inequalities in Lyapunov-type stability. With these tools, we are able to extend classical asymptotic stability results for nonlinear dynamical systems modeled with the GAK operators of Riemann–Liouville (GAKRL) and Caputo (GAKC) type. In addition, a new stability definition is introduced, which generalizes previous definitions of stability for fractional systems; moreover, some generalizations of stability for GAK operators using convex Lyapunov functions are presented. The rest of the paper is organized as follows: In Section2, we introduce the mathematical background, some definitions concerning the GAK operators, and we show how other fractional operators are obtained as particular cases of them, along with some results necessary for the following sections. In Section3, we establish the relationship between the GAKRL and GAKC fractional operators and present a novel form of the rule for the product of functions with these operators, avoiding the infinite series that appears in the Leibniz rule for classical fractional calculus. Section4 addresses some results related to the Laplace transform of the GAKRL and GAKC operators, their relationship, and their application in the solution of fractional-order differential equations. Furthermore, some consequences for the generalized Lyapunov direct method with these operators are given through the Fernandez-Özarslan–Baleanu stability. The stability analysis task is more straightforward with adequate tools, and, in Section5, we prove some functional inequalities for quadratic Lyapunov functions. Moreover, in Section6, we use locally convex functions for generaliz- ing stability results via convex Lyapunov functions, and, in Section7, some representative examples are given. Finally, in Section8, we present the conclusions of the work. Throughout the paper, the following notation is used: • I denotes fractional integrals. The left superscript of I denotes the type of integral, while the right superscript represents the order, and the right subscript indicates the lower limit of the integral: RL α – I a+ is the Riemann–Liouville fractional integral of order α. AB α – I a+ is the Atangana–Baleanu fractional integral of Riemann–Liouville type of order α. A α,β – I a+ is the fractional-order integral operator with general analytic kernel with orders α and β. • D denotes fractional derivatives. The left superscript of D denotes the type of derivative, while the right superscript represents the order, and the right subscript indicates the lower limit of the set where the operator is being applied: RL α – Da+ is the Riemann–Liouville fractional differential operator of order α. C α – Da+ is the Caputo differential operator of order α. CF α – Da+ is the Caputo–Fabrizio differential operator of order α. ABC α – Da+ is the Atangana–Baleanu differential operator of Caputo type of order α. ABR α – Da+ is the Atangana–Baleanu differential operator of Riemann–Liouville type of order α. • In the case of the differential operators with general analytic kernel, both left super- and subscripts are considered: Mathematics 2021, 9, 2084 3 of 29

A α,β – RLDa+ is the Riemann–Liouville fractional differential operator with general analytic kernel with orders α and β. A α,β – CDa+ is the Caputo fractional differential operator with general analytic kernel with orders α and β. ∂V • For a candidate Lyapunov function V(t, x(t)), ∇|V(t, x(t)) = [40]. ∂x

2. Preliminaries As we mentioned, in this work, we propose new tools for the stability analysis of a class of fractional-order systems, where the main characteristic of those systems is the general analytic kernel in their derivative. This fractional derivative covers a wide variety of fractional operators, and, as particular cases, we ﬁnd the classical Caputo, Riemann–Liouville, and Atangana–Baleanu derivatives. Therefore, we start this section with some basic deﬁnitions from the theory of fractional calculus and, subsequently, their generalization by using the general analytic kernel operators.

Z ∞ + − −t Deﬁnition 1. [41]. Let α ∈ R and Γ(α) = tα 1e dt be the Gamma function. The Riemann– 0 Liouville fractional integral of order α is given by

Z t RL α 1 α−1 I a+ f (t) = (t − τ) f (τ) dτ, (1) Γ(α) a

RL α where a ≤ t ≤ b and I a+ f (t) ∈ L1[a, b].

+ Deﬁnition 2. [41]. Let α ∈ R and consider n = min{k ∈ N | k > α}. Based on the Riemann– Liouville integral, the Riemann–Liouville fractional derivative of order α is deﬁned by

h i 1 dn Z t RLDα ( ) = n RLIn−α ( ) = ( − )n−α−1 ( ) a+ f t D a+ f t n t τ f τ dτ, (2) Γ(n − α) dt a

+ where Dn is the usual derivative of order n ∈ Z .

+ Deﬁnition 3. [41]. Let α ∈ R and consider n = min{k ∈ N | k > α}. Whenever Dn ∈ L1[a, b], the Caputo derivative of fractional order α is given by

Z t C α RL n−α n 1 n−α−1 (n) Da+ f (t) = I a+ [D f (t)] = (t − τ) f (τ) dτ. (3) Γ(n − α) a

Deﬁnition 4. [23]. Let 0 < α < 1 and consider a function f ∈ H1(a, b) with b > a. The operator described by Z t CF α M(α) 0 α Da+ f (t) = f (τ) exp − (t − τ) dτ (4) 1 − α a 1 − α represents a fractional derivative with a nonsingular kernel called the Caputo–Fabrizio fractional derivative, where the normalization function M(α) satisﬁes M(0) = M(1) = 1.

Deﬁnition 5. [24]. The operator deﬁned by

Z t ABR α B(α) d −α α Da+ f (t) = f (τ)Eα (t − τ) dτ, (5) 1 − α dt a 1 − α

is called the Atangana–Baleanu Riemann–Liouville (ABR) fractional derivative for f ∈ L1(a, b), 0 < α < 1, a < t < b. Mathematics 2021, 9, 2084 4 of 29

Deﬁnition 6. [24]. Let f be a differentiable function on [a, b] such that f 0 ∈ L1(a, b). The operator deﬁned by Z t ABC α B(α) 0 −α α Da+ f (t) = f (τ)Eα (t − τ) dτ, (6) 1 − α a 1 − α is called the Atangana–Baleanu–Caputo (ABC) fractional derivative of f , where 0 < α < 1 and a < t < b.

+ Deﬁnition 7. [24]. Let α ∈ R . Based on the classical Riemann–Liouville fractional integral, the AB fractional integral of f (t) can be deﬁned by

1 − α α ABIα f (t) = f (t) + RLIα f (t), (7) a+ B(α) B(α) a+

where a < t < b and f ∈ L1(a, b).

Remark 1. The operators (5)–(7) use a real-valued normalization function B(α) that satisfies B(α) > 0 and B(0) = B(1) = 1. In addition, Eα(·) represents the Mittag–Leffler function with one parameter α > 0 defined by the convergent series [42]:

∞ zk E (z) = . (8) α ∑ ( + ) k=0 Γ αk 1

A generalization of function (8) is the Mittag–Lefﬂer function with two parameters α > 0, β > 0 deﬁned by the convergent series [42]:

∞ zk E (z) = . (9) α,β ∑ ( + ) k=0 Γ αk β

Now that the previous fractional derivatives and integrals have been deﬁned, we may ask ourselves if there is a way to describe a family of derivatives to which operators (2) to (7) belong. In this sense, Fernández et al. proposed a simple integral model based on an analytical kernel and the Riemann–Liouville integral (1) that generalizes the already known operators [22]. Some useful deﬁnitions and results concerning these new operators are presented below.

+ Deﬁnition 8. [22]. Let [a, b] ⊂ R, α, β ∈ C with Re(α) > 0, Re(β) > 0, and R ∈ R such ( ) that R > (b − a)Re β . A general analytic kernel is a complex function analytic on the disc D(0, R) deﬁned by the following locally uniformly convergent power series

∞ k A(x) = ∑ akx , (10) k=0

where the coefﬁcients ak = ak(α, β) may have dependence on α and β.

Deﬁnition 9. [22]. Let A be a general analytic kernel satisfying Deﬁnition8 and f ∈ L1[a, b]. The fractional-order integral operator with general analytic kernel (GAK) is given by:

Z t A α,β α−1 β I a+ f (t) = (t − τ) A (t − τ) f (τ) dτ. (11) a

Definition 10. [22]. Let A be a general analytic kernel satisfying Definition8. The transformed function AΓ is defined as follows:

∞ k AΓ(x) = ∑ akΓ(βk + α)x . (12) k=0 Mathematics 2021, 9, 2084 5 of 29

Theorem 1. [22]. Let A be a general analytic kernel satisfying Deﬁnition8. Then, for any function f ∈ L1[a, b], the integral (11) is equivalent to the locally uniformly convergent series on [a, b]:

∞ A α,β RL α+kβ I a+ f (t) = ∑ akΓ(βk + α) I a+ f (t), (13) k=0

RL α+kβ where I a+ is the Riemann–Liouville integral.

Definition 11. [22]. Let a, b, α, β, A satisfying Definition8, and let f be a function f ∈ L1[a, b] with sufficient differentiability properties. The differential operators of Riemann–Liouville and Caputo type with general analytic kernel (GAKRL and GAKC, respectively) are given by

m α,β d α0,β0 AD f (t) = AI f (t) , (14) RL a+ dtm a+ m α,β α0,β0 d AD f (t) = AI f (t) , (15) C a+ a+ dtm + 0 0 respectively, where m ∈ Z , and the orders α and β depend on α and β.

Example 1. Let x0 ∈ R; then, by deﬁnition (15), one has that

A α,β CDa+ x0 = 0.

On the other hand, by direct calculation employing Equation (14):

∞ 0 0 α,β Γ(α + β k) 0+ 0 − AD x = x a (t − a)α β k m. RL a+ 0 0 ∑ k ( 0 + 0 + − ) k=0 Γ α β k 1 m

Remark 2. Some classical fractional operators may be obtained using different values of α and β in deﬁnitions (11), (14), (15), as follows: 1 (a) If α0 = α, β0 = 0, m = 1 and A (t − τ)0 = , then Γ(α)

Z t A α,0 1 α−1 RL α I a+ f (t) = (t − τ) f (τ) dτ = I a+ f (t), (16) Γ(α) a

represents the Riemann–Liouville integral operator of order α. 1 (b) If α0 = m − α, β0 = 0, m = 1 and A (t − τ)0 = , then Γ(1 − α)

dm ADα,β f (t) = AIm−α,0 f (t) RL a+ dtm a+ (17) 1 dm Z t = ( − )m−α−1 ( ) = RLDα ( ) m t τ f τ dτ a+ f t , Γ(n − α) dt a

dm ADα,β f (t) = AIm−α,0 f (t) C a+ a+ dtm (18) Z t 1 m−α−1 (m) C α = (t − τ) f (τ) dτ = Da+ f (t), Γ(m − α) a represent the Riemann–Liouville fractional derivative and the Caputo fractional derivative, respectively. Mathematics 2021, 9, 2084 6 of 29

M(α) α (c) If α0 = 1, β0 = 1, m = 1 and A((t − τ)) = exp − (t − τ) , then 1 − α 1 − α

d ADα,β f (t) = AI1,1 f (t) C a+ a+ dt (19) Z t M(α) 0 α CF α = f (τ) exp − (t − τ) dτ = Da+ f (t), 1 − α a 1 − α

is the Caputo–Fabrizio derivative. B(α) α (d) If α0 = 1, β0 = α, m = 1 and A(t − τ)α = E − (t − τ)α , then 1 − α α 1 − α

d ADα,β f (t) = AI1,α f (t) RL a+ dt a+ Z t (20) B(α) d −α α ABR α = f (τ)Eα (t − τ) dτ = Da+ f (t), 1 − α dt a 1 − α

d ADα,β f (t) = AI1,α f (t) C a+ a+ dt (21) Z t B(α) 0 −α α ABC α = f (τ)Eα (t − τ) dτ = Da+ f (t), 1 − α a 1 − α represent the ABR and ABC fractional derivatives, respectively.

3. Some Results with the General Analytic Kernel Operators In this section, we relate the general analytic kernel differential operators (14) and (15) to each other, in a similar manner to the relationship between the classical Riemann– Liouville and Caputo derivatives. Moreover, the main result of the section consists of a new formula to calculate the fractional-order GAK derivative of the product of two functions.

Theorem 2. Let a, b, A satisfying Deﬁnition8, α, β ∈ C with Re(α) > 0, Re(β) > 0, and f ∈ L1[a, b]. Then

m−1 ∞ 0 0 α,β α,β Γ(β k + α ) − + 0+ 0 AD f (t) = AD f (t) + a (t − a)j m α kβ f (j)(a). (22) RL a+ C a+ ∑ ∑ k ( − + 0 + 0 + ) j=0 k=0 Γ j m α kβ 1

Proof. From Deﬁnition (14) and the representation (13) for the fractional integral operator, we have: " # dm h 0 0 i dm ∞ 0 0 ADα,β ( ) = AIα ,β ( ) = 0 + 0 RLIα +kβ ( ) RL a+ f t m a+ f t m ∑ akΓ β k α a+ f t . dt dt k=0

The series for A is assumed to be locally uniformly convergent, so the order of summation can be swapped, to get:

∞ dm 0 0 ADα,β ( ) = 0 + 0 RLIα +kβ ( ) RL a+ f t ∑ akΓ β k α m a+ f t , k=0 dt

i.e.,

∞ 0 0 ∞ 0 0 A α,β 0 0 RL α +kβ −m 0 0 RL m−α −kβ RLDa+ f (t) = ∑ akΓ β k + α I a+ f (t) = ∑ akΓ β k + α Da+ f (t). k=0 k=0 Mathematics 2021, 9, 2084 7 of 29

Moreover, considering that the classical Riemann–Liouville and Caputo operators are related to each other [3], then

∞ " m−1 j−m+α0+kβ0 # α,β m−α0−kβ0 (t − a) AD f (t) = a Γβ0k + α0 CD f (t) + f (j)(a) RL a+ ∑ k a+ ∑ ( − + 0 + 0 + ) k=0 j=0 Γ j m α kβ 1

∞ 0 0 0 0 RL m−(m−α −kβ ) (m) = ∑ akΓ β k + α I a+ f (t) k=0 0 0 ∞ m−1 (t − a)j−m+α +kβ + a Γβ0k + α0 f (j)(a) ∑ k ∑ ( − + 0 + 0 + ) k=0 j=0 Γ j m α kβ 1

∞ 0 0 0 0 RL α +kβ (m) = ∑ akΓ β k + α I a+ f (t) k=0 m− 0 0 1 ∞ Γ(β k + α ) 0 0 + a (t − a)j−m+α +kβ f (j)(a) ∑ ∑ k ( − + 0 + 0 + ) j=0 k=0 Γ j m α kβ 1 m− 0 0 0 0 dm 1 ∞ Γ(β k + α ) 0 0 = AIα ,β f (t) + a (t − a)j−m+α +kβ f (j)(a) a+ m ∑ ∑ k ( − + 0 + 0 + ) dt j=0 k=0 Γ j m α kβ 1 m−1 ∞ 0 0 α,β Γ(β k + α ) − + 0+ 0 = AD f (t) + a (t − a)j m α kβ f (j)(a). C a+ ∑ ∑ k ( − + 0 + 0 + ) j=0 k=0 Γ j m α kβ 1

Theorem 3. Let a, b, A satisfying Deﬁnition8, α, β ∈ C with Re(α) > 0, Re(β) > 0, f ∈ ( ) + L1[a, b] with sufﬁcient differentiability properties such that f j (a) ≥ 0, 0 ≤ j ≤ m − 1, m ∈ Z . 0 0 If ak > 0 and j − m + α + kβ + 1 > 0 for all j, k, then

A α,β A α,β CDa+ f (t) ≤ RLDa+ f (t). (23)

Proof. From the Theorem2, we have

m−1 ∞ 0 0 α,β α,β Γ(β k + α ) − + 0+ 0 AD f (t) = AD f (t) − a (t − a)j m α kβ f (j)(a). C a+ RL a+ ∑ ∑ k ( − + 0 + 0 + ) j=0 k=0 Γ j m α kβ 1

(j) 0 0 Because f (a) ≥ 0, if ak > 0 and j − m + α + kβ + 1 > 0, we get

A α,β A α,β CDa+ f (t) ≤ RLDa+ f (t).

This concludes the proof.

Lemma 1. Let a, b, A satisfying Deﬁnition8, α, β ∈ C with Re(α) > 0, Re(β) > 0, f ∈ L1[a, b] with sufﬁcient differentiability properties, and the following relationship holds:

A α,βn A α,β o I a+ RLDa+ f (t) ∞ " m m−j α+βl−j+1 # d α+βl−m+α0,β0 d α0,β0 (t − a) = a Γ(α + βl) AI f (t) − AI f (t) . ∑ l a+ ∑ m−j a+ ( + + − ) l=0 dt j=1 dt t=a Γ 2 α βl j

Proof.

∞ A α,βn A α,β o RL α+βk A α,β I a+ RLDa+ f (t) = ∑ akΓ(α + βk) I a+ RLDa+ f (t) k=0 ! ∞ ∞ 0 0 RL α+βk 0 0 RL α +β n−m = ∑ akΓ(α + βk) I a+ ∑ anΓ α + β n I a+ f (t) . k=0 n=0 Mathematics 2021, 9, 2084 8 of 29

By hypothesis, the series is locally uniformly convergent, hence the order of integration and the summation can be changed. Considering the properties between the Riemann–Liouville fractional derivative and integral [1], one has that

∞ ∞ 0 0 A α,βn A α,β o 0 0 RL α+βkRL m−α −β n I a+ RLDa+ f (t) = ∑ ∑ akanΓ(α + βk)Γ α + β n I a+ Da+ f (t) k=0 n=0 ∞ ∞ 0 0 0 0 nRL m−α −β n−α−βk = ∑ ∑ akanΓ(α + βk)Γ α + β n Da+ f (t) k=0 n=0 ) q 0 0 α+βk−j hRL m−α −β n−j i (t − a) − Da+ f (t) ∑ t=a ( + + − ) j=1 Γ 1 α βk j m d + 0 + 0 = AIα α ,β β f (t) dtm a+ ∞ ∞ q 0 0 α+βk−j 0 0 hRL m−α −β n−j i (t − a) − akanΓ(α + βk)Γ α + β n Da+ f (t) . ∑ ∑ ∑ t=a ( + + − ) k=0 n=0 j=1 Γ 1 α βk j

It is well known that the fractional derivative of a product of functions obtained with the classical fractional operators is difficult to calculate. Therefore, we present a result that helps to solve this problem by using the GAKRL derivative. According to the integer-order definition, the first order derivative of the function f (t) is given by

d f (t) − f (t − ε) f (t) = lim . (24) dt ε→0 ε Similarly, d2 1 f (t) = lim [ f (t) − 2 f (t − ε) + f (t − 2ε)], (25) dt2 ε→0 ε2 d3 1 f (t) = lim [ f (t) − 3 f (t − ε) + 3 f (t − 2ε) − f (t − 3ε)]. (26) dt3 ε→0 ε3 Iterating this process n−times, we get

dm 1 m m f (t) = lim (−1)j f (t − jε). (27) m ε→0 m ∑ dt ε j=0 j

On the other hand, observe that the product of two functions can be expressed as follows:

u(τ)v(τ) = (u(τ) − u(t))(v(τ) − v(t)) + u(t)v(τ) + u(τ)v(t) − u(t)v(t). (28)

Based on the previous comments, we prove the following result.

Lemma 2. Let a, b, A satisfying Deﬁnition8, α, β ∈ C with Re(α) > 0, Re(β) > 0, u(t), v(t) ∈ L1[a, b], u(t)v(t) ∈ L1[a, b]. If α0 + β0k − m > 0 for all j, k, then

A α,β A α,β A α,β RLDa+ (u(t)v(t)) = u(t) RLDa+ v(t) + v(t) RLDa+ u(t) ∞ 0 0 Γ(α + β k) Z t 0 0 + a (t − τ)α +β k−1−m[(u(τ) − u(t))(v(τ) − v(t))] dτ ∑ k ( 0 + 0 − ) k=0 Γ α β k m a (29) ∞ 0 0 Γ(α + β k) 0+ 0 − − u(t)v(t) a (t − a)α β k m. ∑ k ( 0 + 0 + − ) k=0 Γ α β k 1 m Mathematics 2021, 9, 2084 9 of 29

Proof. Combining Deﬁnition (14) and expression (28), the derivative of the product between u(t) and v(t) can be expressed as follows:

dm Z t 0 0 ADα,β( ( ) ( )) = ( − )α −1 ( − )β ( ) ( ) RL a+ u t v t m t τ A t τ u τ v τ dτ dt a ∞ dm Z t 0 0 = ( − )α +β k−1 ( ) ( ) ∑ ak m t τ u τ v τ k=0 dt a = I1 + I2 + I3 − I4,

where

∞ dm Z t 0 0 = ( − )α +β k−1( ( ) − ( ))( ( ) − ( )) I1 ∑ ak m t τ u τ u t v τ v t dτ , (30) k=0 dt a ∞ dm Z t 0 0 = ( − )α +β k−1 ( ) ( ) I2 ∑ ak m t τ u t v τ dτ , (31) k=0 dt a ∞ dm Z t 0 0 = ( − )α +β k−1 ( ) ( ) I3 ∑ ak m t τ u τ v t dτ , (32) k=0 dt a ∞ dm Z t 0 0 = ( − )α +β k−1 ( ) ( ) I4 ∑ ak m t τ u t v t dτ . (33) k=0 dt a

By deﬁnition (14), it follows that

A α,β A α,β I2 + I3 = u(t) RLDa+ v(t) + v(t) RLDa+ u(t), (34)

and the integral (30) can be expressed by using the form (27), i.e.,

∞ m Z t−jε 1 j m α0+β0k−1 I1 = ak lim (−1) (t − jε − τ) [(u(τ) − u(t))(v(τ) − v(t))] dτ. ∑ ε→0 m ∑ k=0 ε j=0 j a

m By a simple calculation in the integral I1, observe that lim ε = 0 and ε→0

m t−jε m Z 0+ 0 − lim (−1)j (t − jε − τ)α β k 1[(u(τ) − u(t))(v(τ) − v(t))] dτ ε→0 ∑ j=0 j a m t m Z 0+ 0 − = ∑(−1)j (t − τ)α β k 1[(u(τ) − u(t))(v(τ) − v(t))] dτ j=0 j a t Z 0+ 0 − = (1 + (−1))m (t − τ)α β k 1[(u(τ) − u(t))(v(τ) − v(t))] dτ a = 0. Mathematics 2021, 9, 2084 10 of 29

To solve the problem with the indeterminate form, by using the L’Hôpital’s rule:

∞ m (−1)j(−j)m m Γ(α0 + β0k) I = a × 1 ∑ k ∑ ( + ) ( 0 + 0 − ) k=0 j=1 Γ m 1 j Γ α β k m

Z t 0 0 (t − τ)α +β k−1−m[(u(τ) − u(t))(v(τ) − v(t))] dτ a ∞ m! m (−1)j(−j)m Γ(α0 + β0k) = a × ∑ k ( + ) ∑ ( − ) ( 0 + 0 − ) k=0 Γ m 1 j=1 j! m j ! Γ α β k m

Z t 0 0 (t − τ)α +β k−1−m[(u(τ) − u(t))(v(τ) − v(t))] dτ a ∞ 0 0 t Γ(α + β k) Z 0+ 0 − − = a (t − τ)α β k 1 m[(u(τ) − u(t))(v(τ) − v(t))] dτ, ∑ k ( 0 + 0 − ) k=0 Γ α β k m a

where we have used the fact that

m (−1)j(−j)m = 1. ∑ ( − ) j=1 j! m j !

Finally, the integral (33) is solved by direct calculation:

∞ dm Z t 0 0 = ( ) ( ) ( − )α +β k−1 I4 u t v t ∑ ak m t τ dτ k=0 dt a 0 0 ∞ Γ(α + β k) dm 0 0 = u(t)v(t) a (t − a)α +β k ∑ k ( 0 + 0 + ) m k=0 Γ α β k 1 dt ∞ 0 0 Γ(α + β k) 0+ 0 − = u(t)v(t) a (t − a)α β k m. ∑ k ( 0 + 0 + − ) k=0 Γ α β k 1 m

Therefore,

A α,β A α,β A α,β RLDa+ (u(t)v(t)) = u(t) RLDa+ v(t) + v(t) RLDa+ u(t) 0 0 ∞ Γ(α + β k) Z t 0 0 + a (t − τ)α +β k−1−m[(u(τ) − u(t))(v(τ) − v(t))] dτ ∑ k ( 0 + 0 − ) k=0 Γ α β k m a ∞ 0 0 Γ(α + β k) 0+ 0 − − u(t)v(t) a (t − a)α β k m. ∑ k ( 0 + 0 + − ) k=0 Γ α β k 1 m

A α,β A α,β Lemma 3. If the assumptions in Lemma2 are satisﬁed, but replacing RLDa+ by CDa+ , then

A α,β A α,β A α,β CDa+ u(t)v(t) = u(t) CDa+ v(t) + v(t) CDa+ u(t) m− 0 0 1 ∞ Γ(β k + α ) 0 0 + u(t) a (t − a)j−m+α +kβ v(j)(a) ∑ ∑ k ( − + 0 + 0 + ) j=0 k=0 Γ j m α kβ 1 m− 0 0 1 ∞ Γ(β k + α ) 0 0 + v(t) a (t − a)j−m+α +kβ u(j)(a) ∑ ∑ k ( − + 0 + 0 + ) j=0 k=0 Γ j m α kβ 1 0 0 ∞ Γ(α + β k) Z t 0 0 + a (t − τ)α +β k−1−m[(u(τ) − u(t))(v(τ) − v(t))] dτ ∑ k ( 0 + 0 − ) k=0 Γ α β k m a 0 0 ∞ Γ(α + β k) 0 0 − u(t)v(t) a (t − a)α +β k−m ∑ k ( 0 + 0 + − ) k=0 Γ α β k 1 m m−1 ∞ 0 0 Γ(β k + α ) − + 0+ 0 − a (t − a)j m α kβ (uv)(j)(a). ∑ ∑ k ( − + 0 + 0 + ) j=0 k=0 Γ j m α kβ 1 Mathematics 2021, 9, 2084 11 of 29

Proof. The result can be directly obtained from Theorem2 and Lemma2.

4. Laplace Transform and Generalized Lyapunov Direct Method As with their integer-order counterpart, Laplace transform has been one of the most useful tools for the solution of fractional differential equations. Thus, in this section, we establish some results related to the Laplace transform of the differential operators of Riemann–Liouville and Caputo type with general analytic kernels, the relationship between them, and their application to the analysis of existence and uniqueness of certain types of differential equations. In addition, some results of a generalized Lyapunov direct method for these operators are presented.

Theorem 4. Ref. [22] Let f ∈ L2[0, b], b > 0 with Laplace transform F(s). Let α, β A satisfying A α,β Deﬁnition8. The Laplace transform of the fractional integral I0+ f (t) is given by:

nA α,β o −α −β L I0+ f (t) = s AΓ(s )F(s), (35)

where the function AΓ satisﬁes Deﬁnition 10.

The Laplace transforms of (14) and (15) are determined as follows.

Theorem 5 (Laplace transform for the GAKRL derivative). Let f ∈ L2[0, b], b > 0 with Laplace transform F(s). Let α0, β0, A satisfy Deﬁnition8. The Laplace transform of the GAKRL A α,β derivative RLD0+ f (t) is given by:

n A α,β o −β0 m−α0 L RLD0+ f (t) = AΓ s s F(s). (36)

Proof. From Deﬁnition (14) and its representation (13),

∞ 0 0 A α,β 0 0 RL α +kβ −m RLDa+ f (t) = ∑ akΓ β k + α I0+ f (t). k=0

0 0 The series is locally uniformly convergent, due to 0 ≤ |(t − τ)β | ≤ bRe(β ) < R, and A is locally uniformly convergent by hypothesis on D(0, R). This allows for interchanging the order between the summation and the integration. Therefore, by using the Laplace transform for the classical Riemann–Liouville integral, we have that ( ) ∞ 0 0 n A α,β o 0 0 RL α +kβ −m L RLDa+ f (t) = L ∑ akΓ β k + α I0+ f (t) k=0 ∞ 0 0 0 0 nRL α +kβ −m o = ∑ akΓ β k + α L I0+ f (t) k=0 ∞ F(s) = a Γβ0k + α0 . ∑ k α0+kβ0−m k=0 s

Finally, considering the representation given in Deﬁnition (10), the Laplace transform for the GAKRL derivative is reduced to

∞ k n A α,β o 0 0 −β0 m−α0 −β0 m−α0 L RLDa+ f (t) = ∑ akΓ β k + α s s F(s) = AΓ s s F(s). k=0 Mathematics 2021, 9, 2084 12 of 29

Theorem 6 (Laplace transform for GAKC derivative). Let f ∈ L2[0, b], b > 0 with Laplace transform F(s). Let α, β A satisfying Deﬁnition8. The Laplace transform of the GAKC derivative A α,β CD0+ f (t) is given by:

" m−1 # nA α,β o −β0 −α0 m m−1−j (j) L CD0+ f (t) = AΓ s s s F(s) − ∑ s f (0) . (37) j=0

Proof. The operator (15) can be rewritten in its integral form as follows:

Z t A α,β α0−1 β0 (m) α0−1 β0 (m) CD0+ f (t) = (t − τ) A (t − τ) f (τ) dτ = t A t ∗ f (t), (38) 0 where ∗ is the convolution operator. Applying the Laplace transform to the above equation, one has that

( ∞ ) nA α,β o n α0−1 β0 o n (m) o α0−1 β0k n (m) o L CD0+ f (t) = L t A t · L f (t) = L t ∑ akt · L f (t) . k=0

The series for A is assumed to be locally uniformly convergent, which allows for interchanging the order of the integration and the summation, i.e.,

" ∞ # nA α,β o n α0+β0k−1o n (m) o L CD0+ f (t) = ∑ akL t · L f (t) k=0 " # ∞ Γ(α0 + β0k) h i = a smF(s) − sm−1 f (0) − ... − f (m−1)(0) ∑ k α0+β0k k=0 s ∞ k 0 0 −β0 −α0 h m m−1 (m−1) i = ∑ akΓ α + β k s s s F(s) − s f (0) − ... − f (0) . k=0

Finally, by Deﬁnition 10 and rewriting the summation in square brackets, the proof is completed.

Example 2. Let α0 = 1, β0 = α, m = 1; then,

A 1,1+α A 1,α 0 ABC α CD0+ f (t) = I0+ f (t) = D0+ f (t) Z t 1−1 B(α) α α 0 = (t − τ) Eα − (t − τ) f (τ) dτ, (39) 0 1 − α 1 − α | {z } A((t−τ)α)

ABC α where D0+ f (t) is the Atangana–Baleanu derivative [24] in the Caputo sense, with

∞ 1 B(α) α k A((t − τ)α) = − ((t − τ)α)k. (40) ∑ ( + ) − − k=0 Γ αk 1 1 α 1 α | {z } ak

On the other hand, from expression (37), with the parameters established previously,

∞ nABC α o −αk −1 L D0+ f (t) = ∑ akΓ(1 + αk) s s [sF(s) − f (0)]. (41) k=0

Now, by using ak as in the Equation (40) and the geometric series, one has that

α−1 nABC α o B(α) s L D0+ f (t) = α [sF(s) − f (0)]. (42) 1 − α sα + 1 − α Mathematics 2021, 9, 2084 13 of 29

This result coincides with the Laplace transform for the ABC derivative. In a similar manner, we can obtain the Laplace transform pairs for other fractional operators.

In this paper, we consider fractional-order nonlinear systems with a general analytic kernel of the form α,β AD x(t) = f (t, x(t)), (43) C t0+ where the initial conditions have the form

dkx(0) x(k)(0) = = γ , k = 0, 1, . . . , m − 1. (44) dt k The set of initial conditions is necessary to specify the unique solution to the system n (43); moreover, f : [t0, ∞) × Ω → R is piecewise continuous in t and locally Lipschitz in n x on [t0, ∞) × Ω, where Ω ⊂ R is a domain that contains the equilibrium point x = 0. The equilibrium point of (43) is deﬁned as follows.

Deﬁnition 12. An equilibrium point of the fractional-order system (43) is a constant x0 that satisﬁes f (t, x0) = 0.

Remark 3. Without loss of generality, let the equilibrium point be x = 0. If the equilibrium point of system (43) is x¯ 6= 0, let y(t) = x(t) − x.¯ Then,

α,β α,β α,β AD y(t) = AD (x(t) − x¯) = AD x(t) = f (t, x(t)) = f (t, y(t) + x¯) = g(t, y(t)), C t0+ C t0+ C t0+ where g(t, 0) = 0. A direct implication of this analysis is that the system with the variable y(t) satisﬁes Deﬁnition 12 so that its equilibrium point lies at the origin.

Now, consider the following assumptions, and deﬁne a function that will be useful for the analysis of the solution x(t).

−β0 −β0 −α0+m Assumption 1. (1) The series AΓ s is uniformly convergent and satisﬁes AΓ s s 6= 0 for Re(s) > 0. (2) f (t, x(t)) ∈ L1[0, ∞).

0 0 Deﬁnition 13. Let α, β ∈ C with Re(α) > 0, Re(β) > 0. Let α , β be parameters that depend + 0 0 on α, β, respectively, and m ∈ Z . The function A (t; α , β , m) deﬁned by ( ) 0 0 − 1 A t; α , β , m = L 1 , (45) λ −β0 −α0+m AΓ s s + λ

whenever the inverse Laplace transform L −1{·} converges, is called the Fernandez–Özarslan– Baleanu function (FOB-function).

Theorem 7. If the conditions stated in Assumption1 hold, then the solution of system (43) can be rewritten as m−1 (j) 0 0 x (0) j x(t) = x(0) + A t; α , β , m ∗ f (t, x(t)) + t . (46) 0 ∑ ( + ) j=1 Γ j 1

Proof. Under Assumption1, we apply the Laplace transform to (43):

" m−1 # −β0 −α0 m m−1−j (j) AΓ s s s X(s) − ∑ s x (0) = F(s), (47) j=0 Mathematics 2021, 9, 2084 14 of 29

where F(s) = L { f (t, x(t))}. Rearranging and taking inverse Laplace transforms, it follows that

m−1 −β0 m−α0 −β0 m−α0−1−j (j) AΓ s s X(s) = F(s) + AΓ s ∑ s x (0) j=0 F(s) m−1 X(s) = + s−1−jx(j)(0) −β0 m−α0 ∑ AΓ s s j=0 ( ) 1 m−1 1 L −1{X(s)} = L −1 · F(s) + L −1 x(j)(0) −β0 −α0+m ∑ j+1 AΓ s s j=0 s ( ) 1 m−1 1 x(t) = L −1 ∗ f (t, x(t)) + tjx(j)(0), −β0 −α0+m ∑ ( + ) AΓ s s j=0 Γ j 1

where ∗ denotes the standard convolution operation. Finally, by using the FOB-function and rearranging the summation, the proof is completed.

Example 3. If we set some particular values to m, α0 and β0 in the FOB-function, then the solution of (43) recovers interesting results previously reported in the literature. According to the Remark2 for the Caputo derivative, if m = 1, α0 = 1 − α, β0 = 0, then

1 1 tα−1 ( − ) = −1 = −1 = A0 t; 1 α, 0, 1 L 0 α L α , AΓ(s )s s Γ(α)

where

∞ k ∞ k 1 A s0 = a Γ(α) s0 = Γ(α) a s0 = Γ(α)A s0 = Γ(α) = 1. Γ ∑ k ∑ k ( ) k=0 k=0 Γ α

Then, by Theorem7, the solution of a fractional-order system CDα x(t) = f (t, x(t)) can be t0+ determined by solving the following equivalent problem:

tα−1 x(t) = A (t; 1 − α, 0, 1) ∗ f (t, x(t)) + x(0) = x(0) + ∗ f (t, x(t)). 0 Γ(α)

According to the convolution operation and by the Riemann–Liouville integral (1), we obtain the following Volterra integral equation:

Z t 1 α−1 RL α x(t) = x(0) + (t − τ) f (τ, x(τ)) dτ = x(0) + I0+ f (t, x(t)). Γ(α) 0

On the other hand, for the Atangana–Baleanu-Caputo derivative, we set m = 1, α0 = 1, β0 = α; then, 1 ( ) = −1 A0 t; 1, α, 1 L −α . AΓ(s )

Considering ak as in Equation (40), we have

∞ ∞ k α−1 k B(α) α B(α) s A s−α = a Γ(αk + 1)s−α = − s−α = s. (48) Γ ∑ k − ∑ − − α k=0 1 α k=0 1 α 1 α sα + 1 − α

Then, the solution of ABCDα x(t) = f (t, x(t)) is the solution of the Volterra integral equation: t0+

x(t) = A0(t; 1, α, 1) ∗ f (t, x(t)) + x(0). Mathematics 2021, 9, 2084 15 of 29

By Deﬁnition 13 and Equation (48), one has that        1  x(t) = L −1 ∗ f (t, x(t)) + x(0). B( ) sα−1  α   α s   1 − α sα +  1 − α Applying the inverse Laplace transform to the previous expression, we have that

α tα−1 1 − α x(t) = x(0) + ∗ f (t, x(t)) + δ(t) ∗ f (t, x(t)), B(α) Γ(α) B(α)

where δ(t) is the Dirac delta function. Finally, considering the convolution operator and its properties, we obtain the following Volterra integral equation:

t t α 1 Z − 1 − α Z x(t) = x(0) + (t − τ)α 1 f (τ, x(τ)) dτ + δ(t − τ) f (τ, x(τ)) dτ, B(α) Γ(α) 0 B(α) 0

that is,

t α 1 Z − 1 − α x(t) = x(0) + (t − τ)α 1 f (τ, x(τ)) dτ + f (t, x(t)), B(α) Γ(α) 0 B(α)

and rearranging terms, AB α x(t) = x(0) + I0+ f (t, x(t)), AB α where I0+ f (t, x(t)) represents the Atangana–Baleanu integral [24]. A similar analysis for different values of parameters α0, β0 and m can be done to obtain the solutions of fractional differential equations that involve different fractional operators.

If we use the GAKRL fractional derivative instead of the GAKC operator, the solution of

α,β AD x(t) = f (t, x(t)), (49) RL t0+

takes the form (46), where all the initial conditions x(j)(0), j = 0, 1, ... , m − 1 are set to zero. An equilibrium point of a system (49) can be deﬁned.

Deﬁnition 14. An equilibrium point of the fractional-order system (49) is a constant x0 that satisﬁes α,β AD x = f (t, x ). (50) RL t0+ 0 0

Remark 4. Similarly to Remark3, suppose the equilibrium point for (49) is x¯ 6= 0 and consider the same change of variable y(t) = x(t) − x.¯ Then,

∞ 0 0 α0+β0−m α,β α,β Γ(α + β k)(t − t ) AD y(t) = AD (x(t) − x¯) = f (t, y(t) + x¯) − x¯ a 0 RL t0+ RL t0+ ∑ k ( 0 + 0 + − ) k=0 Γ α β k 1 m = g(t, y(t)),

where g(t, 0) = 0. In terms of the new variable, the system has an equilibrium at the origin.

Lemma 4. (Comparison Lemma) Let a, b, A satisfying Deﬁnition8. Let x(0) = y(0),..., (m−1) (m−1) A α,β A α,β x (0) = y (0) and CD0+ x(t) ≥ CD0+ y(t), where m ∈ N, α, β ∈ C with Re(α) > 0 0 0, Re(β) > 0. If the FOB-function A0(t; α , β , m) takes non-negative functions into non-negative functions, then x(t) ≥ y(t). Mathematics 2021, 9, 2084 16 of 29

A α,β A α,β Proof. From inequality CD0+ x(t) ≥ CD0+ y(t), it follows that there exists a nonnegative function M(t) that satisﬁes the following equation:

A α,β A α,β CD0+ x(t) = CD0+ y(t) + M(t). (51)

Applying the Laplace transform (37) to Equation (51), we have that

−β0 −α0 h m m−1 (m−1) i AΓ s s s X(s) − s x(0) − ... − x (0) = (52) −β0 −α0 h m m−1 (m−1) i AΓ s s s Y(s) − s y(0) − ... − y (0) + M(s).

By hypothesis, x(0) = y(0),..., x(m−1)(0) = y(m−1)(0), then equality (52) is reduced to:

−β0 −α0+m −β0 −α0+m AΓ s s X(s) = AΓ s s Y(s) + M(s). (53)

−β0 −α0+m Dividing by AΓ s s produces

M(s) X(s) = Y(s) + . (54) −β0 −α0+m AΓ s s

Taking the inverse Laplace transform of (54) and using the Convolution Theorem, one has that 0 0 x(t) = y(t) + A0 t; α , β , m ∗ m(t). (55) The second term of the right-hand side of (55) is non-negative because M(t) is non-negative. Based on this reasoning, the proof is completed and x(t) ≥ y(t).

0 0 Deﬁnition 15. Let α, β ∈ C with Re(α) > 0, Re(β) > 0. Let α , β be parameters that depend + 0 + 0 −β0 on α, β, respectively, and m ∈ Z , γ ∈ Z ∪ {0}, λ > 0 and β such that AΓ s converges. The function deﬁned by

 0 0 0  A s−β sm−1−α −γ G γ0 0 0 −  Γ  A t; α , β , m = L 1 , (56) λ −β0 −α0+m  AΓ s s + λ 

whenever the inverse Laplace transform L −1{·} converges, is called the generalized Fernandez– Özarslan–Baleanu function (A − FOB-function).

Deﬁnition 16 (FOB-Stability). The solution of (43) is said to be Fernandez–Özarslan–Baleanu stable (FOB-stable) if

b "m−1 # k ( )k ≤ ( ( )) G j − 0 0 x t ∑ νj x t0 A λ t t0; α , β , m , (57) j=0

for all t ≥ t0, where ν(x) ≥ 0 is a locally Lipschitz continuous function on x, with ν(0) = 0, 0 + 0 −β0 m − α − j > 0, m ∈ Z , λ > 0, b > 0 and β such that AΓ s converges.

Remark 5. The A − FOB-function represents a family of Mittag–Lefﬂer functions and their generalizations. Note that, for α0 = 1 − α, β0 = 0, m = 1, and γ0 = 0 ( ) A s0 sα−1 sα−1 G 0 ( − ) = −1 Γ = −1 = (− α) A λ t; 1 α, 0, 1 L 0 α L α Eα λt , (58) AΓ(s )s + λ s + λ Mathematics 2021, 9, 2084 17 of 29

and the FOB-stability coincides with the classical Mittag–Lefﬂer stability [43]:

b h G 0 i α b kx(t)k ≤ ν0(x(t0)) A λ(t − t0; 1 − α, 0, 1) = [ν0(x(t0))Eα(−λ(t − t0) )] . (59)

Based on the previously established deﬁnitions and results, we can then analyze the behavior of fractional-order systems and, in particular, prove asymptotic stability by means of an extension of the Lyapunov direct method, which is directly related to FOB-stability.

Theorem 8. Let x = 0 be an equilibrium point for system (43) and D ⊂ Rn be a domain containing the origin. Let V(t, x(t)) : [0, ∞) × D → R be a continuously differentiable function and locally Lipschitz with respect to x such that V(j)(0, x(0)) > 0, 0 ≤ j ≤ m − 1, and

a ab α1kxk ≤ V(t, x(t)) ≤ α2kxk , (60) A α,β ab CD0+ V(t, x(t)) ≤ −α3kxk , (61)

where t ≥ 0, x ∈ D, α, β ∈ C with Re(α) > 0, Re(β) > 0, α1, α2, α3, a and b are arbitrary positive constants. Then, x = 0 is FOB-stable and asymptotically stable.

Proof. From inequalities (60) and (61), one has that

A α,β α3 CD0+ V(t, x(t)) ≤ − V(t, x(t)). (62) α2

Let M(t) be a non-negative function. Based on this function, the inequality (62) can be rewritten as A α,β CD0+ V(t, x(t)) + λV(t, x(t)) + M(t) = 0, (63) −1 with λ = α3α2 . Applying the Laplace transform to Equation (63), we have that

" m−1 # −β0 −α0 m m−1−j (j) AΓ s s s V(s) − ∑ s V (0, x(0)) + λV(s) + M(s) = 0. j=0

Rearranging this equation and solving for V(s), we have

m−1 h −β0 −α0+m i −β0 m−1−α0−j (j) V(s) AΓ s s + λ = AΓ s ∑ s V (0, x(0)) − M(s). j=0

Therefore,

−β0 m−1−α0−j m−1 AΓ s s M(s) V(s) = V(j)(0, x(0)) − . (64) ∑ −β0 −α0+m −β0 −α0+m j=0 AΓ s s + λ AΓ s s + λ

Once the algebraic problem is solved, we apply the inverse Laplace transform of (64) to obtain

m−1 ( ( )) = (j)( ( )) G j 0 0 − 0 0 ∗ ( ) V t, x t ∑ V 0, x 0 A λ t; α , β , m Aλ t; α , β , m M t . (65) j=0

0 0 0 0 Considering Aλ(t; α , β , m) ≥ 0, ∀ t ≥ 0, then Aλ(t; α , β , m) ∗ M(t) ≥ 0 and thus

m−1 ( ( )) ≤ (j)( ( )) G j 0 0 V t, x t ∑ V 0, x 0 A λ t; α , β , m . (66) j=0 Mathematics 2021, 9, 2084 18 of 29

Combining (66) with condition stated in inequality (60) yields

" #1/a m−1 V(j)(0, x(0)) k ( )k ≤ G j 0 0 x t ∑ A λ t; α , β , m , j=0 α1

V(j)(0, x(0)) V(j)(0, x(0)) where > 0 for x(0) 6= 0. Letting Kj = ≥ 0, then we have α1 α1

a "m−1 #1/ k ( )k ≤ G j 0 0 x t ∑ Kj A λ t; α , β , m , (67) j=0

where Kj = 0 if and only if x(0) = 0. Because V(t, x(t)) is locally Lipschitz with respect to x, its derivatives are bounded and V(j)(0, x(0)) = 0 if and only if x(0) = 0, then it follows that Kj is also Lipschitz with respect to x(0) and Kj(0) = 0; this implies the FOB stability of system (43). Furthermore, by using the ﬁnal value theorem on the right-hand side of (66) for m − α0 − j > 0, we get

−β0 m−α0−j m−1 AΓ s s (j) lim V(t, x(t)) ≤ lim 0 0 V (0, x(0)) = 0. (68) t→∞ s→0 ∑ −β −α +m j=0 AΓ s s + λ

Combining inequalities (60), (68), and considering that V(t, x(t)) ≥ 0 for all t yields

a lim α1kxk ≤ lim V(t, x(t)) = 0. (69) t→∞ t→∞

a It follows from (69) that lim α1kxk ≤ 0. Finally, due to α1, a > 0, then t→∞

lim kx(t)k = 0. t→∞

This proves that the origin of system (43) is asymptotically stable.

Remark 6. Note that FOB-stability implies asymptotic stability.

A α,β A α,β Theorem 9. If the assumptions in Theorem8 are satisﬁed except replacing CD0+ by RLD0+ , then the origin x = 0 of system (49) is asymptotically stable.

Proof. By using the inequality of Theorem3 and V(t, x(t)) ≥ 0, we obtain

A α,β A α,β ab CD0+ V(t, x(t)) ≤ RLD0+ V(t, x(t)) ≤ −α3kxk .

Following the proof of Theorem8, the proof is completed.

5. Useful Inequalities for Lyapunov Stability Analysis In this section, some inequalities are established. These results help develop tools for the stability analysis of fractional-order nonlinear systems, employing the generalized Lyapunov direct method shown in the previous section. We start with some signiﬁcant lemmas that will help us prove the main results.

Lemma 5. [44]. Let u(t) be a continuous and differentiable real-valued function. Then, for any time instant t ≥ t0 and for all 0 < α < 1:

RLDα u2(t) ≤ u(t) RLDα u(t) t0+ 2 t0+ . Mathematics 2021, 9, 2084 19 of 29

Lemma 6. [45]. Let u(t) be a continuous and differentiable real-valued function. Then, for any time instant t ≥ t0 and for all 0 < α < 1:

C Dα u2(t) ≤ u(t) C Dα u(t) t0+ 2 t0+ .

Lemma 7. Let u(t) ∈ R be a continuously differentiable and monotonically increasing function 0 0 + and u(t) ≥ 0. Then, for all t ≥ t0, α + β k > m, m ∈ Z , the following inequality holds:

α0+β0k−m α0+β0k−m RLI u2(t) ≤ 2u(t) RLI u(t). (70) t0+ t0+ Proof. Rewriting the inequality (70), one has that

α0+β0k−m α0+β0k−m RLI u2(t) − 2u(t) RLI u(t) ≤ 0. (71) t0+ t0+ Differentiating inequality (71) with respect to time, we have

m+1−α0−β0k m+1−α0−β0k α0+β0k−m RLD u2(t) − 2u(t) RLD u(t) ≤ 2u˙(t) RLI u(t). (72) t0+ t0+ t0+

By the hypotheses on function u, and using the fact that α0 + β0k > m,

α0+β0k−m u˙(t) RLI u(t) ≥ 0. (73) t0+ Then, expression (72) is clearly true due to Lemma5. This concludes the proof.

0 0 Remark 7. If we set ak > 0 for all k ≥ 0, α > 0, β > 0; then, from the previous Lemma,

0 0 a Γα0 RLIα −mu2(t) ≤ 2a Γα0u(t) RLIα −mu(t), 0 t0+ 0 t0+ α0+β0−m α0+β0−m a Γα0 + β0 RLI u2(t) ≤ 2a Γα0 + β0u(t) RLI u(t), 1 t0+ 1 t0+ α0+2β0−m α0+2β0−m a Γα0 + 2β0 RLI u2(t) ≤ 2a Γα0 + 2β0u(t) RLI u(t), 2 t0+ 2 t0+ . . α0+β0 j−m α0+β0 j−m a Γα0 + jβ0 RLI u2(t) ≤ 2a Γα0 + jβ0u(t) RLI u(t). j t0+ j t0+

Then, under the conditions that permit local uniform convergence on a disc D(0, R) such that R > (b − a)Re(β), it follows that

j 0 0 j 0 0 0 0 RL α +β k−m 2 0 0 RL α +β k−m lim akΓ α + β k It + u (t) ≤ lim akΓ α + β k 2u(t) It + u(t). j→∞ ∑ 0 j→∞ ∑ 0 k=0 k=0

This idea can be summarized in the following corollary.

0 0 0 0 Corollary 1. Let a, b, A satisfying Deﬁnition8, α , β ∈ C with Re(α ) > 0, Re(β ) > 0. 0 0 + If ak > 0 for all k ≥ 0, α + β k > m, m ∈ Z . Then,

∞ ∞ α0+β0k−m α0+β0k−m a Γα0 + β0k RLI u2(t) ≤ a Γα0 + β0k2u(t) RLI u(t). (74) ∑ k t0+ ∑ k t0+ k=0 k=0

1 0 0 Theorem 10. Let u ∈ L [a, b]. If ak > 0 and α + β k − m > 0 for all k, then, for any time instant t ≥ t0, the following inequality holds:

α,β α,β AD u2(t) ≤ 2u(t) AD u(t). RL t0+ RL t0+ Mathematics 2021, 9, 2084 20 of 29

Proof. From Lemma2 with u(t) = v(t), we have

α,β α,β AD u2(t) = 2u(t) AD u(t) RL t0+ RL t0+ ∞ 0 0 t Γ(α + β k) Z 0+ 0 − − h i + a (t − τ)α β k 1 m u2(τ) − 2u(τ)u(t) dτ ∑ k ( 0 + 0 − ) k=0 Γ α β k m t0 0 0 ∞ Γ(α + β k) Z t 0 0 + u2(t) a (t − τ)α +β k−1−m dτ ∑ k ( 0 + 0 − ) k=0 Γ α β k m t0 ∞ 0 0 Γ(α + β k) 0+ 0 − − u2(t) a (t − t )α β k m. ∑ k ( 0 + 0 − + ) 0 k=0 Γ α β k m 1

Since Z t 0 0 α0+β0k−m−1 α0+β0k−m Γ(α + β k − m) (t − τ) dτ = (t − t0) 0 0 , t0 Γ(α + β k − m + 1) and, by using the classical Riemann–Liouville integral, one has that

α,β α,β AD u2(t) = 2u(t) AD u(t) RL t0+ RL t0+ ∞ h α0+β0k−m α0+β0k−m i + a Γα0 + β0k RLI u2(t) − 2u(t) RLI u(t) . ∑ k t0+ t0+ k=0

0 0 Since the integral converges for all k, then, from Lemma7 by setting ak > 0 for α + β k − m > 0, the series is convergent and satisﬁes (74). Then, it follows that

α,β α,β AD u2(t) ≤ 2u(t) AD u(t), RL t0+ RL t0+ and the proof is completed.

1 0 0 Theorem 11. Let u ∈ L [a, b]. If ak > 0 and α + β k − m > 0 for all k, then, for any time instant t ≥ t0, the following inequality holds:

α,β α,β AD u2(t) ≤ 2u(t) AD u(t). (75) C t0+ C t0+ Proof. Proving that inequality (75) is true, is equivalent to prove that

α,β α,β 2u(t) AD u(t) − AD u2(t) ≥ 0. (76) C t0+ C t0+ Using Deﬁnition (15) and representation (13), the expression (76) can be written as

∞ m m α0+kβ0 d α0+kβ0 d a Γβ0k + α0 2u(t) RLI u(t) − RLI u2(t) ≥ 0. (77) ∑ k t0+ m t0+ m k=0 dt dt

Furthermore, from the classical Caputo derivative (Deﬁnition3), the inequality takes the following form:

∞ h m−α0−β0k m−α0−β0k i a Γβ0k + α0 2u(t) C D u(t) − C D u2(t) ≥ 0. (78) ∑ k t0+ t0+ k=0

Since the Riemann–Liouville integral converges for all k, α0 + β0k > 0; then, from Lemma6, by setting ak > 0, the series is convergent, and this concludes the proof.

Corollary 2. Assume the conditions of Theorems 10 and 11. Mathematics 2021, 9, 2084 21 of 29

1 (a) If α0 = 1 − α, β0 = 0, m = 1 and A (t − τ)0 = , then for all t ≥ t Γ(1 − α) 0

RLDα u2(t) ≤ u(t) RLDα u(t) t0+ 2 t0+ , (79)

C Dα u2(t) ≤ u(t) C Dα u(t) t0+ 2 t0+ . (80) M(α) α (b) If α0 = 1, β0 = 1, m = 1 and A((t − τ)) = exp − (t − τ) , then for all 1 − α 1 − α t ≥ t0 CF Dα u2(t) ≤ u(t) CF Dα u(t) t0+ 2 t0+ . (81) B(α) α (c) If α0 = 1, β0 = α, m = 1 and A(t − τ)α = E − (t − τ)α , then for all 1 − α α 1 − α t ≥ t0 ABRDα u2(t) ≤ u(t) ABRDα u(t) t0+ 2 t0+ , (82) ABC Dα u2(t) ≤ u(t) ABC Dα u(t) t0+ 2 t0+ . (83) Proof. The proof is straightforward by a simple substitution of α0, β0 and A(·) of Deﬁnitions (14) and (15) in Theorems 10 and 11.

Remark 8. We presented two general cases for inequalities related to quadratic forms. Particular cases for classical operators are treated in Corollary2. All these inequalities were previously treated as the main results in different papers [35,44–46].

6. Convex Lyapunov Functions and Stability In this section, the stability results presented formerly are extended considering convex Lyapunov functions. Previously, in [47,48], the convex analysis was treated for the classic Caputo and Riemann–Liouville derivatives. Now, we extend the results considering general analytic kernel operators, which represent a generalization of earlier studies. In this section, we will assume that Ω is a convex and compact set [49].

Deﬁnition 17. Let f (x) be a continuously differentiable function. The function f (x) is said to be convex in a convex domain if it satisﬁes

f (y) ≥ f (x) + h f 0(x), y − xi (84)

for all x, y in its domain.

n Theorem 12. Let V(t, x(t)) : Ω → R, where Ω = [0, ∞) × D, D ⊂ R , and x(t) : [t0, ∞) → D be two continuous and differential functions. Suppose that V(t, x(t)) is convex over Ω, 0 0 with V(t0, x(t0)) = 0, ak > 0 and m − α − β k > 0 for all k. Then, for all t ≥ t0,

α,β α,β AD V(t, x(t)) ≤ ∇|V(t, x(t)) AD x(t). (85) RL t0+ RL t0+ Proof. The function V(t, x(t)) can be expressed as

Z t V(t x(t)) = V(t x(t )) + V˙ ( x( )) d = V(t x(t )) + RLI1 V˙ (t x(t)) , 0, 0 τ, τ τ 0, 0 t0+ , . (86) t0 Taking the GAKRL (14) in both sides of (86), then

∞ 0 0 A α,β Γ(α + β k) α0+β0k−m D V(t, x(t)) = V(t0, x(t0)) a (t − t0) RL t0+ ∑ k ( 0 + 0 + − ) k=0 Γ α β k 1 m ∞ (87) α0+kβ0+1−m + a Γβ0k + α0 RLI V˙ (t, x(t)). ∑ k t0+ k=0 Mathematics 2021, 9, 2084 22 of 29

Following the same idea for the function x(t), we have that

∞ 0 0 A α,β Γ(α + β k) α0+β0k−m D x(t) = x(t0) a (t − t0) RL t0+ ∑ k ( 0 + 0 + − ) k=0 Γ α β k 1 m ∞ (88) α0+kβ0+1−m + a Γβ0k + α0 RLI x˙(t). ∑ k t0+ k=0

Thus, to justify the proposed inequality, we can rewrite inequality (85) as follows:

∞ 0 0 Γ(α + β k) 0+ 0 − (∇|V(t, x(t))x(t ) − V(t , x(t ))) a (t − t )α β k m 0 0 0 ∑ k ( 0 + 0 + − ) 0 k=0 Γ α β k 1 m (89) ∞ 0 0 t Γ(α + β k) Z 0+ 0 − + a (t − τ)α β k m[∇|V(t, x(t)) − ∇|V(τ, x(τ))]x˙(τ) dτ ≥ 0. ∑ k ( 0 + 0 + − ) k=0 Γ α β k 1 m t0 Let ϕ(τ) = V(t, x(t)) − V(τ, x(τ)) − ∇|V(t, x(t))(x(t) − x(τ)). By hypothesis, the function V(t, x(t)) is convex; then, according to Deﬁnition 17, it follows that ϕ(τ) ≤ 0. Now, integration by parts can be applied in the second term of (89) with u = ϕ(τ), du = ϕ˙ (τ) dτ, 0+ 0 − 0+ 0 − − v = (t − τ)α β k m, dv = (m − α0 − β0k)(t − τ)α β k m 1 dτ. The inequality (89) can be written as

∞ 0 0 Γ(α + β k) 0+ 0 − (∇|V(t, x(t))x(t ) − V(t , x(t ))) a (t − t )α β k m 0 0 0 ∑ k ( 0 + 0 + − ) 0 k=0 Γ α β k 1 m ∞ 0 0 " Γ(α + β k) ϕ(τ) ϕ(t0) + ak lim 0 0 − 0 0 (90) ∑ ( 0 + 0 + − ) τ→t m−α −β k m−α −β k k=0 Γ α β k 1 m (t − τ) (t − t0) t Z 0+ 0 − − +(m − α0 − β0k) −ϕ(τ)(t − τ)α β k m 1 dτ ≥ 0. t0

Considering that α0 + β0k + 1 − m > 0, then the limit is calculated by applying the L’Hôpital’s rule, i.e.,

ϕ(τ) [∇|V(t, x(t)) − ∇|V(τ, x(τ))]x˙(τ) lim 0 0 = lim 0 0 = 0. τ→t (t − τ)m−α −β k τ→t (m − α0 − β0k)(t − τ)m−α −β k−1

Evaluating ϕ(t0) and considering that V(t0, x(t0) = 0, we have

∞ 0 0 Γ(α + β k) n 0+ 0 − a (t − t )α β k m[∇|V(t, x(t))x(t) − V(t, x(t))] ∑ k Γ(α0 + β0k + 1 − m) 0 k=0 (91) t Z 0+ 0 − − +(m − α0 − β0k) −ϕ(τ)(t − τ)α β k m 1 dτ ≥ 0. t0

Finally, due to the convexity of V(t, x(t)), ∇|V(t, x(t))x(t) − V(t, x(t)) ≥ 0. Furthermore, ϕ(t) ≤ 0, ak > 0 for all k such that the series converges. Then, the inequality (91) is true, and the proof is completed.

Corollary 3. Assume the conditions of Theorem 12: 1 (a) If α0 = 1 − α, β0 = 0, m = 1 and A (t − τ)0 = , then, for all t ≥ t Γ(1 − α) 0

RLDα V(t x(t)) ≤ ∇|V(t x(t)) RLDα x(t) t0+ , , t0+ . (92)

B(α) α (b) If α0 = 1, β0 = α, m = 1 and A(t − τ)α = E − (t − τ)α , then for all 1 − α α 1 − α t ≥ t0 ABRDα V(t x(t)) ≤ ∇|V(t x(t)) ABRDα x(t) t0+ , , t0+ . (93) Mathematics 2021, 9, 2084 23 of 29

Proof. The proof is immediate and is omitted.

n Theorem 13. Let V(t, x(t)) : Ω → R, where Ω = [0, ∞) × D, D ⊂ R , and x(t) : [t0, ∞) → D be two continuous and differential functions. Suppose that V(t, x(t)) is convex over Ω, (j) 0 0 with V (t0, x(t0)) = 0 for all j > 0, ak > 0 and m − α − β k > 0 for all k. Then, for all t ≥ t0, α,β α,β AD V(t, x(t)) ≤ ∇|V(t, x(t)) AD x(t). (94) C t0+ C t0+ Proof. The proof is obtained directly by combining Theorem2 and Theorem 12, considering (j) that V (t0, x(t0)) = 0 for all j.

Corollary 4. Assume the conditions of Theorem 13. 1 (a) If α0 = 1 − α, β0 = 0, m = 1, and A (t − τ)0 = , then, for all t ≥ t , Γ(1 − α) 0

C Dα V(t x(t)) ≤ ∇|V(t x(t)) C Dα x(t) t0+ , , t0+ . (95)

M(α) α (b) If α0 = 1, β0 = 1, m = 1 and A((t − τ)) = exp − (t − τ) , then, for all 1 − α 1 − α t ≥ t0, CF Dα V(t x(t)) ≤ ∇|V(t x(t)) CF Dα x(t) t0+ , , t0+ . (96) B(α) α (c) If α0 = 1, β0 = α, m = 1 and A(t − τ)α = E − (t − τ)α , then, for all 1 − α α 1 − α t ≥ t0, ABC Dα V(t x(t)) ≤ ∇|V(t x(t)) ABC Dα x(t) t0+ , , t0+ . (97) Proof. The proof is straightforward by a simple substitution of α0, β0 and A(·) of Deﬁnitions (14) and (15) in Theorem 13.

Theorem 14. Let x = 0 be an equilibrium point for the fractional-order system (43). Let V(t, x(t)) such that it satisﬁes Theorem 13, locally Lipschitz with respect to x such that the following inequalities hold:

a ab α1kxk ≤ V(t, x(t)) ≤ α2kxk , (98) | ab ∇ V(t, x(t)) f (t, x(t)) ≤ −α3kxk , (99)

where α1, α2, α3, a, and b are arbitrary positive constants. Then, x = 0 is asymptotically stable. Proof. From Theorem 13 and inequality (99), the fractional-order derivative of V(t, x(t)) satisﬁes

α,β α,β AD V(t, x(t)) ≤ ∇|V(t, x(t)) AD x(t) = ∇|V(t, x(t)) f (t, x(t)) ≤ −α kxkab. C t0+ C t0+ 3 From this analysis and due to V(t, x(t)) satisfying (98), it follows that conditions of Theorem8 are satisﬁed. Therefore, the origin of system (43) is asymptotically stable.

Theorem 15. Let x = 0 be an equilibrium point for the fractional-order system (49). Let V(t, x(t) be as in Theorem 12, locally Lipschitz with respect to x such that

a ab α1kxk ≤ V(t, x(t)) ≤ α2kxk , (100) | ab ∇ V(t, x(t)) f (t, x(t)) ≤ −α3kxk , (101)

where α1, α2, α3, a, and b are arbitrary positive constants. Then, x = 0 is asymptotically stable. Proof. The proof follows the same outline as the one from Theorem 14 and is omitted. Mathematics 2021, 9, 2084 24 of 29

Remark 9. If we set different values of α0 and β0, the conclusions of Theorems (14) and (15) hold for different fractional-order derivatives. For example, for α0 = β0 = 1, m = 1 and A((t − τ)) = M(α) α exp − (t − τ) , then, for all t ≥ t , we obtain the conclusion for systems with a 1 − α 1 − α 0 Caputo–Fabrizio derivative [50].

7. Some Representative Examples 7.1. Scalar Systems In this ﬁrst example, we study scalar systems. Consider the system given by [51]:

A α,β p CD0+ x = ax + g(x), (102)

where the Caputo derivative of general analytic kernel (15) is considered. In system (102), − the parameter p > 0 is odd, a < R and g(x) is bounded as follows:

|g(x)| ≤ k|x|p+1, (103)

where the inequality (103) is valid in some neighborhood of the origin x = 0. To carry out the stability analysis and show the validity, in practical cases, of the inequalities demonstrated in previous sections, let us consider the Lyapunov candidate function V(x) = 1 x2. Then, applying Theorem 11 and considering system (102), one has that 2

A α,β A α,β p p+1 CD0+ V ≤ x CD0+ x = x[ax + g(x)] ≤ ax + |x||g(x)|.

Hence, due to the bound (103), the fractional-order derivative of V(x) is bounded as follows: A α,β p+1 p+2 CD0+ V ≤ ax + k|x| .

p+1 A α,β p+2 Near the origin, the term ax is dominant, and then CD0+ V ≤ k|x| . This implies that the origin is Fernandez–Özarslan–Baleanu stable and therefore asymptotically stable.

7.2. Second Order Systems There is a wide class of systems of second order. In particular, consider the following system modeled using the Caputo derivative of general analytic kernel (15):

A α,β CD + x1 = −x1 + x1x2, 0 (104) A α,β CD0+ x2 = −x2.

It is not difﬁcult to show that the equilibrium point of system (104) is (0, 0). To analyze 1 2 2 stability, consider V(x) = 2 x1 + x2 . Applying the inequality given in Theorem 11 and considering system (104), one has that

A α,β A α,β A α,β 2 CD0+ V ≤ x1 CD0+ x1 + x2 CD0+ x2 = x1(−x1 + x1x2) − x2.

2 To analyze the last inequality, consider the ball kxk2 ≤ r where |x1| ≤ r. If we restrict the analysis to this set, then

α,β r AD V ≤ −x2 − x2 + r|x ||x | = − |x |2 − 2 · |x ||x | + |x |2 . C 0+ 1 2 1 2 1 2 1 2 2 Mathematics 2021, 9, 2084 25 of 29

The right-hand side of the above inequality can be rewritten by using a matrix form, so that | A α,β |x1| 1 −r/2 |x1| CD0+ V ≤ − . |x2| −r/2 1 |x2| | {z } M

By a simple calculation, it is not difﬁcult to show that the associated eigenvalues of M r + 2 −r + 2 are λ = , λ = ; then, if we choose r < 2 and applying Theorem8, it follows 1 2 2 2 that the origin is asymptotically stable.

7.3. A Spacecraft Modeled by Generalized Dynamics A rotating rigid spacecraft is studied in [52] by using the well-known Euler equations. The differential equations can be generalized considering the Caputo (or Riemann–Liouville (14)) derivative of general analytic kernel (15). This allows for considering different kinds of analysis in engineering applications. For this case, consider the following set of differential equations: A α,β −1 CD0+ ω1 = J1 [(J2 − J3)ω2ω3 + u1], A α,β −1 CD0+ ω2 = J2 [(J3 − J1)ω3ω1 + u2], (105) A α,β −1 CD0+ ω3 = J2 [(J1 − J2)ω1ω2 + u3],

where the scalar components of the vector ω are, respectively, ω1, ω2 and ω3. On the other hand, some torque inputs, denoted by u1, u2 and u3, are considered and applied about the principal axes. In addition to the dynamical analysis, the components J1, J2 and J3 represent the moments of inertia. The stability analysis will be done by considering the controlled system (105). Suppose that the torque inputs apply the feedback control ui = −kiωi, where ki > 0 and 1 ≤ i ≤ 3. Then, the close-loop system is given by

A α,β −1 CD0+ ω1 = J1 [(J2 − J3)ω2ω3 − k1ω1], A α,β −1 CD0+ ω2 = J2 [(J3 − J1)ω3ω1 − k2ω2], (106) A α,β −1 CD0+ ω3 = J3 [(J1 − J2)ω1ω2 − k3ω3], 1 Now, taking V(ω) = J ω2 + J ω2 + J ω2 as a Lyapunov function candidate and 2 1 1 2 2 3 3 using the inequality (75), it is not difﬁcult to obtain

A α,β 2 2 2 CD0+ ≤ −k1ω1 − k2ω2 − k3ω3.

Thus, by applying Theorem8, the origin is Fernandez–Özarslan–Baleanu stable and therefore asymptotically stable.

7.4. Financial Analysis Financial systems are employed to analyze different situations in society, and many points of view have been proposed to model ﬁnancial scenarios [53]. Consider, for example, the following fractional-order system modeled by using the Caputo derivative with general analytic kernel (15):

A α,β CD0+ x1(t) = x3(t) + (x2(t) − a)x1(t) + u1(t), A α,β 2 CD0+ x2(t) = 1 − bx2(t) − x1(t) + u2(t), (107) A α,β CD0+ x3(t) = −x1(t) − cx3(t) + u3(t), Mathematics 2021, 9, 2084 26 of 29

where x1(t) represents the interest rate, x2(t) the investment demand, and x2(t) the price index, and the constants a = 1 (the saving amount), b = 0.1 (the cost per investment), and c = 1 (the elasticity of demand of commercial markets). The stability analysis considers the equilibrium point of the closed-loop system considering the control laws given by

2/3 1/3 5/3 1/3 u1(t) = x1 (t)x3 (t) + x1 (t)x2 (t), 4/3 2/3 u2(t) = −1 − x2(t) − x1 (t)x2 (t), (108) 1/3 2/3 u3(t) = −x1 (t)x3 (t).

On the other hand, let V(x(t)) be the candidate Lyapunov function described by

4/3 4/3 4/3 V(x(t)) = x1 (t) + x2 (t) + x3 (t). (109)

It is not difﬁcult to show, by using the Hessian matrix, that the function (109) is convex. This function can be rewritten as follows:

2 2/3 2 2/3 2 2/3 V(x(t)) = (x1(t)) + (x2(t)) + (x3(t)) ,

and, considering a well known inequality in control systems [54], the function V(x(t)) is bounded as follows:

2/3 2/3 2 2 2 1/3 2 2 2 x1(t) + x2(t) + x3(t) ≤ V(x(t)) ≤ 3 x1(t) + x2(t) + x3(t) kxk4/3 ≤ V(x(t)) ≤ 31/3kxk4/3. (110)

The advantage of Theorem 14 is that we do not need to calculate fractional derivatives of the Lyapunov function, and we only need to restrict the calculation to determine the gradient. Then, one has that

∂V | 4 f (x, t) = x1/3(t) x (t) + (x (t) − a)x (t) + x2/3(t)x1/3(t) + x5/3(t)x1/3(t) ∂x 3 1 3 2 1 1 3 1 2 4 + x1/3(t) 1 − bx (t) − x2(t) − 1 − x (t) − x4/3(t)x2/3(t) (111) 3 2 2 1 2 1 2 4 + x1/3(t) −x (t) − cx (t) − x1/3(t)x2/3(t) . 3 3 1 3 1 3 Finally, after some algebraic reductions and from (110), we have

∂V | 4 f (x, t) = − ax1/3(t) + (1 + b)x1/3(t) + cx1/3(t) ∂x 3 1 2 3 4 ≤ − x1/3(t) + x1/3(t) + x1/3(t) 3 1 2 3 (112) ≤ −V(x(t) ≤ −kxk4/3.

Finally, according to Theorem 14, it is concluded that the origin of system (107) with the control law (108) is Fernandez–Özarslan–Baleanu stable and asymptotically stable.

Remark 10. As proposed throughout the text, the stability results shown generalize the results previously established in the literature. In the presented examples, when considering particular values for α and β, different dynamics can be obtained, which allows for comparing and analyzing dynamical behaviors that permit to choose one fractional derivative or propose one that adapts to the objectives of the research. Mathematics 2021, 9, 2084 27 of 29

8. Conclusions In this paper, we studied some properties of fractional-order derivatives and integrals with general analytic kernels. Some results concerning the Laplace transform were proved and used to establish some remarks on the solutions of fractional-order differential equations that involve these fractional operators, along with a generalized comparison principle. We proposed the FOB-function and the concept of FOB stability, which generalizes Mittag–Lefﬂer stability for a comprehensive family of systems with different fractional-order derivatives. Moreover, one of the main results presented consists of the generalization of the Lyapunov direct method, which is directly related to the FOB-stability and the boundedness of solutions. In addition, some inequalities for quadratic forms have been proposed. These results allow using the Lyapunov direct method for the stability analysis of fractional-order systems with the operators considered. Furthermore, since the inequalities are established for operators with general analytic kernels, we have shown that the inequalities previously presented in the literature emerge from our work as particular cases. In addition, as an extension of the stability analysis, we have treated the stability problem and its solution through convex Lyapunov functions; some theorems are obtained directly from this part. Finally, we provided some illustrative examples to demonstrate the applicability of the proposed approach.

Author Contributions: Conceptualization, G.F.-A. and O.M.-F.; methodology, G.F.-A. and O.M.-F.; validation, G.F.-A., O.M.-F. and F.M.-V.; formal analysis, G.F.-A., O.M.-F. and F.M.-V.; investigation, G.F.-A., O.M.-F. and F.M.-V.; writing—original draft preparation, O.M.-F. and F.M.-V.; writing— review and editing, G.F.-A., O.M.-F., F.M.-V. and J.F.G.-A.; visualization, G.F.-A., O.M.-F., F.M.-V. and J.F.G.-A.; supervision, G.F.-A., O.M.-F. and J.F.G.-A.; project administration, G.F.-A.; funding acquisition, G.F.-A. All authors have read and agreed to the published version of the manuscript. Funding: Guillermo Fernández-Anaya acknowledges the support provided by the División de Investigación y Posgrado de la Universidad Iberoamericana. Universidad Iberoamericana Ciudad de México, Departamento de Física y Matemáticas, Prolongación Paseo de la Reforma 880, Colonia Lomas de Santa Fe, Álvaro Obregón, Ciudad de México 01219, México. Institutional Review Board Statement: Not applicable. Informed Consent Statement: Not applicable. Data Availability Statement: Not applicable. Acknowledgments: Guillermo Fernández-Anaya acknowledges the support provided by Universi- dad Iberoamericana, Ciudad de México. José Francisco Gómez-Aguilar acknowledges the support provided by CONACyT: cátedras CONACyT para jóvenes investigadores 2014. Conﬂicts of Interest: The authors declare no conﬂict of interest.

References 1. Podlubny, I. Fractional Differential Equations; Academic Press: Cambridge, MA, USA, 1999. 2. Samko, S.G.; Kilbas, A.A.; Marichev, O. Fractional Integrals and Derivatives: Theory and Applications; Gordon and Breach Sci. Publishers: Singapore, 1993. 3. Kilbas, A.A.; Srivastava, H.M.; Trujillo, J.J. Theory and Applications of Fractional Differential Equations; Elsevier: Amsterdam, The Netherlands, 2006; Volume 204. 4. Hilfer, R. Applications of Fractional Calculus in Physics; World Scientiﬁc Singapore: Singapore, 2000. 5. Oldham, K.; Spanier, J. The Fractional Calculus Theory and Applications of Differentiation and Integration to Arbitrary Order; Elsevier: Amsterdam, The Netherlands, 1974. 6. Tavazoei, M.; Asemani, M.H. On Robust Stability of Incommensurate Fractional-Order Systems. Commun. Nonlinear Sci. Numer. Simul. 2020, 90, 105344. [CrossRef] 7. Liu, S.; Yang, R.; Zhou, X.F.; Jiang, W.; Li, X.; Zhao, X.W. Stability analysis of fractional delayed equations and its applications on consensus of multi-agent systems. Commun. Nonlinear Sci. Numer. Simul. 2019, 73, 351–362. [CrossRef] 8. Lenka, B.K.; Banerjee, S. Sufﬁcient conditions for asymptotic stability and stabilization of autonomous fractional order systems. Commun. Nonlinear Sci. Numer. Simul. 2018, 56, 365–379. [CrossRef] Mathematics 2021, 9, 2084 28 of 29

9. Lenka, B.K. Fractional comparison method and asymptotic stability results for multivariable fractional order systems. Commun. Nonlinear Sci. Numer. Simul. 2019, 69, 398–415. [CrossRef] 10. Martínez-Fuentes, O.; Martínez-Guerra, R. A high-gain observer with Mittag–Leffler rate of convergence for a class of nonlinear fractional-order systems. Commun. Nonlinear Sci. Numer. Simul. 2019, 79, 104909. [CrossRef] 11. Baleanu, D.; Fernandez, A. On fractional operators and their classifications. Mathematics 2019, 7, 830. [CrossRef] 12. De Oliveira, E.C.; Tenreiro Machado, J.A. A review of definitions for fractional derivatives and integral. Math. Probl. Eng. 2014, 2014.[CrossRef] 13. Fernández-Anaya, G.; Nava-Antonio, G.; Jamous-Galante, J.; Muñoz-Vega, R.; Hernández-Martínez, E. Asymptotic stability of distributed order nonlinear dynamical systems. Commun. Nonlinear Sci. Numer. Simul. 2017, 48, 541–549. [CrossRef] 14. Srivastava, H.M.; Fernandez, A.; Baleanu, D. Some new fractional-calculus connections between Mittag–Leffler functions. Mathematics 2019, 7, 485. [CrossRef] 15. Hilfer, R.; Luchko, Y. Desiderata for fractional derivatives and integrals. Mathematics 2019, 7, 149. [CrossRef] 16. Jarad, F.; U˘gurlu,E.; Abdeljawad, T.; Baleanu, D. On a new class of fractional operators. Adv. Differ. Equ. 2017, 2017, 247. [CrossRef] 17. Ren, J.; Zhai, C. Stability analysis for generalized fractional differential systems and applications. Chaos Solitons Fractals 2020, 139, 110009. [CrossRef] 18. Akkurt, A.; Yildirim, M.; Yildirim, H. A new Generalized fractional derivative and integral. Konuralp J. Math. 2017, 5, 248–259. 19. Zhao, D.; Luo, M. General conformable fractional derivative and its physical interpretation. Calcolo 2017, 54, 903–917. [CrossRef] 20. Restrepo, J.E.; Ruzhansky, M.; Suragan, D. Explicit solutions for linear variable–coefficient fractional differential equations with respect to functions. Appl. Math. Comput. 2021, 403, 126177. 21. Luchko, Y. General fractional integrals and derivatives with the Sonine kernels. Mathematics 2021, 9, 594. [CrossRef] 22. Fernandez, A.; Özarslan, M.A.; Baleanu, D. On fractional calculus with general analytic kernels. Appl. Math. Comput. 2019, 354, 248–265. [CrossRef] 23. Caputo, M.; Fabrizio, M. A new definition of fractional derivative without singular kernel. Prog. Fract. Differ. Appl. 2015, 1, 73–85. 24. Atangana, A.; Baleanu, D. New fractional derivatives with nonlocal and non-singular kernel: theory and application to heat transfer model. Therm. Sci. 2016, 20, 763–769. [CrossRef] 25. Owolabi, K.; Atangana, A. Numerical approximation of nonlinear fractional parabolic differential equations with Caputo-Fabrizio derivative in Riemann–Liouville sense. Chaos Solitons Fractals 2017, 99, 171–179. [CrossRef] 26. Al-Refai, M.; Jarrah, A. Fundamental results on weighted Caputo-Fabrizio fractional derivative. Chaos Solitons Fractals 2019, 126, 7–11. [CrossRef] 27. Zheng, X.; Wang, H.; Fu, H. Well-posedness of fractional differential equations with variable-order Caputo-Fabrizio derivative. Chaos Solitons Fractals 2020, 138, 109966. [CrossRef] 28. Kumar, A.; Pandey, D.N. Existence of mild solution of Atangana–Baleanu fractional differential equations with non-instantaneous impulses and with non-local conditions. Chaos Solitons Fractals 2020, 132, 109551. [CrossRef] 29. Yadav, S.; Pandey, R.K. Numerical approximation of fractional burgers equation with Atangana–Baleanu derivative in Caputo sense. Chaos Solitons Fractals 2020, 133, 109630. [CrossRef] 30. Sadeghi, S.; Jafari, H.; Nemati, S. Operational matrix for Atangana–Baleanu derivative based on Genocchi polynomials for solving FDEs. Chaos Solitons Fractals 2020, 135, 109736. [CrossRef] 31. Shaikh, A.; Nisar, K. Transmission dynamics of fractional order Typhoid fever model using Caputo-Fabrizio operator. Chaos Solitons Fractals 2019, 128, 355–365. [CrossRef] 32. Ali, F.; Ali, F.; Sheikh, N.A.; Khan, I.; Nisar, K.S. Caputo–Fabrizio fractional derivatives modeling of transient MHD Brinkman nanoliquid: Applications in food technology. Chaos Solitons Fractals 2020, 131, 109489. [CrossRef] 33. Baleanu, D.; Jajarmi, A.; Mohammadi, H.; Rezapour, S. A new study on the mathematical modelling of human liver with Caputo-Fabrizio fractional derivative. Chaos Solitons Fractals 2020, 134, 109705. [CrossRef] 34. Sadeghi, S.; Jafari, H.; Nemati, S. Comparing the Atangana-Baleanu and Caputo-Fabrizio derivative with fractional order: Allen Cahn model. Chaos Solitons Fractals 2016, 89, 552–559. 35. Taneco-Hernández, M.; Vargas-De-León, C. Stability and Lyapunov functions for systems with Atangana-Baleanu Caputo derivative: An HIV/AIDS epidemic model. Chaos Solitons Fractals 2020, 132, 109586. [CrossRef] 36. Wei, Q.; Zhou, H.; Yang, S. Non-Darcy flow models in porous media via Atangana-Baleanu derivative. Chaos Solitons Fractals 2020, 141, 110335. [CrossRef] 37. Ali, G.; Gómez-Aguilar, J.; Kamrana. Approximation of partial integro differential equations with a weakly singular kernel using local meshless method. Alex. Eng. J. 2020, 59, 2091–2100. 38. Hoan, L.V.C.; Akinlar, M.A.; Inc, M.; Gómez-Aguilar, J.; Chu, Y.M.; Almohsen, B. A new fractional-order compartmental disease model. Alex. Eng. J. 2020, 59, 3187–3196. [CrossRef] 39. Hidalgo-Reyes, J.; Gomez-Aguilar, J.; Alvarado-Martinez, V.; Lopez-Lopez, M.; Escobar-Jimenez, R. Battery state-of-charge estimation using fractional extended Kalman filter with Mittag–Leffler memory. Alex. Eng. J. 2020, 59, 1919–1929. [CrossRef] 40. Glendinning, P. Stability, Instability and Chaos: An Introduction to the Theory of Nonlinear Differential Equations; Cambridge University Press: Cambridge, UK, 1994. Mathematics 2021, 9, 2084 29 of 29

41. Diethelm, K. The Analysis of Fractional Differential Equations: An Application-Oriented Exposition Using Differential Operators of Caputo type; Springer: Berlin/Heidelberg, Germany, 2010. 42. Gorenflo, R.; Kilbas, A.A.; Mainardi, F.; Rogosin, S.V. Mittag-Leffler Functions, Related Topics and Applications; Springer: Berlin, Germany, 2014. 43. Li, Y.; Chen, Y.; Podlubny, I. Mittag–Leffler stability of fractional order nonlinear dynamic systems. Automatica 2009, 45, 1965–1969. [CrossRef] 44. Liu, S.; Wu, X.; Zhou, X.; Jiang, W. Asymptotical stability of Riemann-Liouville fractional nonlinear systems. Nonlinear Dyn. 2016, 86, 65–71. [CrossRef] 45. Aguila-Camacho, N.; Duarte-Mermoud, M.A.; Gallegos, J.A. Lyapunov functions for fractional order systems. Commun. Nonlinear Sci. Numer. Simul. 2014, 19, 2951–2957. [CrossRef] 46. Martínez-Fuentes, O.; Delfín-Prieto, S. Stability of Fractional Nonlinear Systems with Mittag-Leffler kernel and Design of State Observers; John Wiley & Sons Inc.: Hoboken, NJ, USA, 2020. 47. Chen, W.; Dai, H.; Song, Y.; Zhang, Z. Convex Lyapunov functions for stability analysis of fractional order systems. IET Control Theory Appl. 2017, 11, 1070–1074. [CrossRef] 48. Badri, V.; Tavazoei, M. Stability analysis of fractional order time-delay systems: constructing new Lyapunov functions from those of integer order counterparts. IET Control Theory Appl. 2019, 13, 2476–2481. [CrossRef] 49. Nesterov, Y. Introductory Lectures on Convex Optimization: A Basic Course; Springer Science & Business Media: New York, 2003; Volume 87. 50. Salahshour, S.; Ahmadian, A.; Salimi, M.; Panserad, B.A.; Ferrarad, M. A new Lyapunov stability analysis of fractional-order systems with nonsingular kernel derivative. Alex. Eng. J. 2020, 59, 2985–2990. [CrossRef] 51. Khalil, H.K. Nonlinear Control; Pearson Higher Ed.: Hoboken, NJ, USA, 2014. 52. Perkins, W.R.; Cruz, J.B. Engineering of Dynamic Systems; John Wiley & Sons: New York, NY, USA, 1969. 53. Dadras, S.; Momeni, H.R. Control of a fractional-order economical system via sliding mode. Phys. A Stat. Mech. Appl. 2010, 389, 2434–2442. [CrossRef] 54. Qian, C.; Lin, W. Non-Lipschitz continuous stabilizers for nonlinear systems with uncontrollable unstable linearization. Syst. Control Lett. 2001, 42, 185–200. [CrossRef]