Loading [MathJax]/jax/output/SVG/jax.js
Research article Special Issues

Stability analysis of stochastic fractional-order competitive neural networks with leakage delay

  • This article, we explore the stability analysis of stochastic fractional-order competitive neural networks with leakage delay. The main objective of this paper is to establish a new set of sufficient conditions, which is for the uniform stability in mean square of such stochastic fractional-order neural networks with leakage. Specifically, the presence and uniqueness of arrangements and stability in mean square for a class of stochastic fractional-order neural systems with delays are concentrated by using Cauchy-Schwartz inequality, Burkholder-Davis-Gundy inequality, Banach fixed point principle and stochastic analysis theory, respectively. Finally, four numerical recreations are given to confirm the hypothetical discoveries.

    Citation: M. Syed Ali, M. Hymavathi, Bandana Priya, Syeda Asma Kauser, Ganesh Kumar Thakur. Stability analysis of stochastic fractional-order competitive neural networks with leakage delay[J]. AIMS Mathematics, 2021, 6(4): 3205-3241. doi: 10.3934/math.2021193

    Related Papers:

    [1] Gerardo Sánchez Licea . Sufficiency for singular trajectories in the calculus of variations. AIMS Mathematics, 2020, 5(1): 111-139. doi: 10.3934/math.2020008
    [2] Jun Moon . The Pontryagin type maximum principle for Caputo fractional optimal control problems with terminal and running state constraints. AIMS Mathematics, 2025, 10(1): 884-920. doi: 10.3934/math.2025042
    [3] Yanfei Chai . Robust strong duality for nonconvex optimization problem under data uncertainty in constraint. AIMS Mathematics, 2021, 6(11): 12321-12338. doi: 10.3934/math.2021713
    [4] Savin Treanţă, Muhammad Bilal Khan, Soubhagya Kumar Sahoo, Thongchai Botmart . Evolutionary problems driven by variational inequalities with multiple integral functionals. AIMS Mathematics, 2023, 8(1): 1488-1508. doi: 10.3934/math.2023075
    [5] Asaf Khan, Gul Zaman, Roman Ullah, Nawazish Naveed . Optimal control strategies for a heroin epidemic model with age-dependent susceptibility and recovery-age. AIMS Mathematics, 2021, 6(2): 1377-1394. doi: 10.3934/math.2021086
    [6] Min Wang, Jiao Teng, Lei Wang, Junmei Wu . Application of ADMM to robust model predictive control problems for the turbofan aero-engine with external disturbances. AIMS Mathematics, 2022, 7(6): 10759-10777. doi: 10.3934/math.2022601
    [7] Beyza Billur İskender Eroǧlu, Dilara Yapışkan . Generalized conformable variational calculus and optimal control problems with variable terminal conditions. AIMS Mathematics, 2020, 5(2): 1105-1126. doi: 10.3934/math.2020077
    [8] Murugan Suvinthra, Krishnan Balachandran, Rajendran Mabel Lizzy . Large Deviations for Stochastic Fractional Integrodifferential Equations. AIMS Mathematics, 2017, 2(2): 348-364. doi: 10.3934/Math.2017.2.348
    [9] H. M. Barakat, Magdy E. El-Adll, M. E. Sobh . Bootstrapping m-generalized order statistics with variable rank. AIMS Mathematics, 2022, 7(8): 13704-13732. doi: 10.3934/math.2022755
    [10] Caicai Feng, Saratha Sathasivam, Nurshazneem Roslan, Muraly Velavan . 2-SAT discrete Hopfield neural networks optimization via Crow search and fuzzy dynamical clustering approach. AIMS Mathematics, 2024, 9(4): 9232-9266. doi: 10.3934/math.2024450
  • This article, we explore the stability analysis of stochastic fractional-order competitive neural networks with leakage delay. The main objective of this paper is to establish a new set of sufficient conditions, which is for the uniform stability in mean square of such stochastic fractional-order neural networks with leakage. Specifically, the presence and uniqueness of arrangements and stability in mean square for a class of stochastic fractional-order neural systems with delays are concentrated by using Cauchy-Schwartz inequality, Burkholder-Davis-Gundy inequality, Banach fixed point principle and stochastic analysis theory, respectively. Finally, four numerical recreations are given to confirm the hypothetical discoveries.


    In this paper we establish and prove two new sufficiency theorems for weak and strong minima for an optimal control problem of Lagrange with fixed end-points, nonlinear dynamics, nonlinear isoperimetric inequality and equality constraints and pointwise mixed nonlinear time-state-control inequality and equality restrictions. The proof of the sufficiency theorems is independent of classical methods used to obtain sufficiency in optimal control problems of this type, see for example [32], where the insertion of the original optimal control problem in a Banach space is a fundamental component in order to obtain the corresponding sufficiency theory; [16], where the construction of a bounded solution of a matrix Riccati equation is crucial in this sufficiency approach; or [8,19], where a verification function and a quadratic function satisfying a Hamilton-Jacobi inequality is an indispensable tool in the sufficiency treatments of these theories. Concretely, the sufficiency theorems of this article state that if an admissible process satisfies a first order sufficient condition related with Pontryagin maximum principle, a similar hypothesis of the necessary Legendre-Clebsch condition, the positivity of a quadratic integral on the cone of critical directions, and several conditions of Weierstrass of some functions, where one of them plays a similar role to the Hamiltonian of the problem, then, the previously mentioned admissible process is a strict local minimum. The set of active indices of the corresponding mixed time-state-control inequality constraints must be piecewise constant on the underlying time interval of consideration, the Lagrange multipliers of the inequality constraints must be nonnegative and in fact they have to be zero whenever the associated index of the Lagrange multiplier is inactive. Additionally, the proposed optimal controls need not be continuous on the underlying interval of time but only measurable, see for example [7,8,9,13,14,15,16,17,18,19,21,22,25,26,27,29,32], where the continuity of the proposed optimal control is a crucial assumption in some sufficiency optimal control theories having the same degree of generality as the problems studied in this article. In contrast, in Examples 2.3 and 2.4, we show how two purely measurable optimal controls comprised with the proposed optimal processes satisfy all the hypotheses of Theorems 2.1 and 2.2 becoming in this way strict local minima.

    Additionally, it is worth mentioning that in these new sufficiency theorems for local minima presented in this paper, all the premises that must be satisfied by an admissible process to become an optimal process, are imposed in the hypotheses established in the theorems, in contrast, with other second order necessary and sufficiency theories which depend upon the verifiability of some preliminary assumptions, see for example [2,3,5,6,11,24], where the necessary second order conditions for optimality depend on some previous hypotheses involving the full rankness of a matrix whose nature arises from the linear independence of vectors whose role are the gradients of the active inequality and equality constraints and where further assumptions involving some notions of regularity or normality of a solution are fundamental hypotheses; or [28], where the corresponding sufficiency theory for optimality depends upon the existence of a continuous function dominating the norm of one of the partial derivatives of the dynamic of the problem. Another remarkable feature presented in this theory concerns the fact that our sufficiency treatment not only provides sufficient conditions for strict local minima but they allow measuring the deviation between admissible costs and optimal costs. This deviation involves a functional playing the role of the square of a norm of the Banach space L1, see for example [1,23], where similar estimations of the growth of the objective functional around the optimal control are established.

    On the other hand, it is worth pointing out the existence of some recent optimal control theories which also study optimal control problems with functional inequality or equality restrictions such as the isoperimetric constraints of this paper. Concretely, in [20], necessary optimality conditions for a Mayer optimal control problem involving semilinear unbounded evolution inclusions and inequality and equality Lipschitzian restrictions are obtained by constructing a sequence of discrete approximations and proving that the optimal solutions of discrete approximations converge uniformly to a given optimal process for the primary continuous-time problem. In [12], necessary and sufficient optimality conditions of Mayer optimal control problems involving differential inclusions and functional inequality constraints are presented and the authors study Mayer optimal control problems with higher order differential inclusions and inequality functional constraints. The necessary conditions for optimality obtained in [12], are important generalizations of associated problems for a first order differential inclusions of optimality settings established in [4,10,20]. The sufficiency conditions obtained in [12] include second order discrete inclusions with inequality end-point constraints. The use of convex and nonsmooth analysis plays a crucial role in this related sufficiency treatment. Moreover, one of the fundamental novelties of the work provided in [12] concerns the derivation of sufficient optimality conditions for Mayer optimal control problems having m-th order ordinary differential inclusions with m3.

    The paper is organized as follows. In Section 2, we pose the problem we shall deal with together with some basic definitions, the statement of the main results and two examples illustrating the sufficiency theorems of the article. Section 3 is devoted to state one auxiliary lemma on which the proof of Theorem 2.1, given in the same section, is strongly based. Section 4 is dedicated to state another auxiliary result on which the proof of Theorem 2.2, once again given in the same section, is based.

    Suppose we are given an interval T:=[t0,t1] in R, two fixed points ξ0, ξ1 in Rn, functions L, Lγ (γ=1,,K) mapping T×Rn×Rm to R, two functions f and φ=(φ1,,φs) mapping T×Rn×Rm to Rn and Rs respectively. Let

    A:={(t,x,u)T×Rn×Rmφα(t,x,u)0(αR),φβ(t,x,u)=0(βS)}

    where R:={1,,r} and S:={r+1,,s} (r=0,1,,s). If r=0 then R= and we disregard statements involving φα. Similarly, if r=s then S= and we disregard statements involving φβ.

    Let {Λn} be a sequence of measurable functions and let Λ be a measurable function. We shall say that the sequence of measurable functions {Λn} converges almost uniformly to a function Λ on T, if given ϵ>0, there exists a measurable set ΥϵT with m(Υϵ)<ϵ such that {Λn} converges uniformly to Λ on TΥϵ. We will also denote uniform convergence by ΛnuΛ, almost uniform convergence by ΛnauΛ, strong convergence in Lp by ΛnLpΛ and weak convergence in Lp by ΛnLpΛ. From now on we shall not relabel the subsequences of a given sequence since this fact will not alter our results.

    It will be assumed throughout the paper that L, Lγ (γ=1,,K), f and φ have first and second derivatives with respect to x and u. Also, if we denote by b(t,x,u) either L(t,x,u), Lγ(t,x,u) (γ=1,,K), f(t,x,u), φ(t,x,u) or any of their partial derivatives of order less or equal than two with respect to x and u, we shall assume that if B is any bounded subset of T×Rn×Rm, then |b(B)| is a bounded subset of R. Additionally, we shall assume that if {(Φq,Ψq)} is any sequence in AC(T;Rn)×L(T;Rm) such that for some ΥT measurable and some (Φ0,Ψ0)AC(T;Rn)×L(T;Rm), (Φq,Ψq)L(Φ0,Ψ0) on Υ, then for all qN, b(,Φq(),Ψq()) is measurable on Υ and

    b(,Φq(),Ψq())Lb(,Φ0(),Ψ0()) on Υ.

    Note that all conditions given above are satisfied if the functions L, Lγ (γ=1,,K), f and φ and their first and second derivatives with respect to x and u are continuous on T×Rn×Rm.

    The fixed end-point optimal control problem we shall deal with, denoted by (P), is that of minimizing the functional

    I(x,u):=t1t0L(t,x(t),u(t))dt

    over all couples (x,u) with x:TRn absolutely continuous and u:TRm essentially bounded, satisfying the constraints

    {˙x(t)=f(t,x(t),u(t))(a.e. in T).x(t0)=ξ0,x(t1)=ξ1.Ii(x,u):=t1t0Li(t,x(t),u(t))dt0(i=1,,k).Ij(x,u):=t1t0Lj(t,x(t),u(t))dt=0(j=k+1,,K).(t,x(t),u(t))A(tT).

    Denote by X the space of all absolutely continuous functions mapping T to Rn and by Uc:=L(T;Rc) (cN). Elements of X×Um will be called processes and a process (x,u) is admissible if it satisfies the constraints. A process (x,u) solves (P) if it is admissible and I(x,u)I(y,v) for all admissible processes (y,v). An admissible process (x,u) is called a strong minimum of (P) if it is a minimum of I with respect to the norm

    x:=suptT|x(t)|,

    that is, if for any ϵ>0, I(x,u)I(y,v) for all admissible processes (y,v) satisfying yx<ϵ. An admissible process (x,u) is called a weak minimum of (P) if it is a minimum of I with respect to the norm

    (x,u):=x+u,

    that is, if for any ϵ>0, I(x,u)I(y,v) for all admissible processes (y,v) satisfying (y,v)(x,u)<ϵ. It is a strict minimum if I(x,u)=I(y,v) only in case (x,u)=(y,v). Note that the crucial difference between strong and weak minima is that in the former, if I affords a strong minimum at (x0,u0), then, if (x,u) is admissible and it is sufficiently close to (x0,u0), in the sense that the quantity xx0 is sufficiently small, then I(x,u)I(x0,u0), meanwhile for the latter, if (x,u) is admissible and it is sufficiently close to (x0,u0), in the sense that the quantities xx0, uu0 are sufficiently small, then I(x,u)I(x0,u0).

    The following definitions will be useful in order to continue with the development of this theory.

    For any (x,u)X×Um we shall use the notation (˜x(t)) to represent (t,x(t),u(t)). Similarly (˜x0(t)) represents (t,x0(t),u0(t)). Throughout the paper the notation "" will denote transpose.

    Given K real numbers λ1,,λK, consider the functional I0:X×UmR defined by

    I0(x,u):=I(x,u)+Kγ=1λγIγ(x,u)=t1t0L0(˜x(t))dt,

    where L0:T×Rn×RmR is given by

    L0(t,x,u):=L(t,x,u)+Kγ=1λγLγ(t,x,u).

    For all (t,x,u,ρ,μ)T×Rn×Rm×Rn×Rs, set

    H(t,x,u,ρ,μ):=ρf(t,x,u)L0(t,x,u)μφ(t,x,u).

    Given ρX and μUs define, for all (t,x,u)T×Rn×Rm,

    F0(t,x,u):=H(t,x,u,ρ(t),μ(t))˙ρ(t)x

    and let

    J0(x,u):=ρ(t1)ξ1ρ(t0)ξ0+t1t0F0(˜x(t))dt.

    Consider the first variations of J0 and Iγ (γ=1,,K) with respect to (x,u)X×Um over (y,v)X×L2(T;Rm) which are given, respectively, by

    J0((x,u);(y,v)):=t1t0{F0x(˜x(t))y(t)+F0u(˜x(t))v(t)}dt,
    Iγ((x,u);(y,v)):=t1t0{Lγx(˜x(t))y(t)+Lγu(˜x(t))v(t)}dt.

    The second variation of J0 with respect to (x,u)X×Um over (y,v)X×L2(T;Rm) is given by

    J0((x,u);(y,v)):=t1t02Ω0(˜x(t);t,y(t),v(t))dt

    where, for all (t,y,v)T×Rn×Rm,

    2Ω0(˜x(t);t,y,v):=yF0xx(˜x(t))y+2yF0xu(˜x(t))v+vF0uu(˜x(t))v.

    Denote by E0 the Weierstrass excess function of F0, given by

    E0(t,x,u,v):=F0(t,x,v)F0(t,x,u)F0u(t,x,u)(vu).

    Similarly, the Weierstrass excess function of Lγ (γ=1,,K) corresponds to

    Eγ(t,x,u,v):=Lγ(t,x,v)Lγ(t,x,u)Lγu(t,x,u)(vu).

    For all (x,u)X×L1(T;Rm) let

    D(x,u):=max{D1(x),D2(u)}

    where

    D1(x):=V(x(t0))+t1t0V(˙x(t))dtandD2(u):=t1t0V(u(t))dt,

    where V(π):=(1+|π|2)1/21 with π:=(π1,,πn)Rn or π:=(π1,,πm)Rm.

    Finally, for all (t,x,u)T×Rn×Rm, denote by

    Ia(t,x,u):={αRφα(t,x,u)=0},

    the set of active indices of (t,x,u) with respect to the mixed inequality constraints. For all (x,u)X×Um, denote by

    ia(x,u):={i=1,,kIi(x,u)=0},

    the set of active indices of (x,u) with respect to the isoperimetric inequality constraints. Given (x,u)X×Um, let Y(x,u) be the set of all (y,v)X×L2(T;Rm) satisfying

    {˙y(t)=fx(˜x(t))y(t)+fu(˜x(t))v(t)(a.e. in T),y(ti)=0(i=0,1).Ii((x,u);(y,v))0(iia(x,u)),Ij((x,u);(y,v))=0(j=k+1,,K).φαx(˜x(t))y(t)+φαu(˜x(t))v(t)0(a.e. in T,αIa(˜x(t))).φβx(˜x(t))y(t)+φβu(˜x(t))v(t)=0(a.e. in T,βS).

    The set Y(x,u) is called the cone of critical directions along (x,u).

    Now we are in a position to state the main results of the article, two sufficiency results for strict local minima of problem (P). Given an admissible process (x0,u0) where the proposed optimal controls u0 need not be continuous but only measurable, the hypotheses include, two conditions related with Pontryagin maximum principle, a similar assumption of the necessary Legendre-Clebsch condition, the positivity of the second variation on the cone of critical directions and some conditions involving the Weierstrass functions delimiting problem (P). It is worth observing that the sufficiency theorems not only give sufficient conditions for strict local minima but also provides some information concerning the deviation between optimal and feasible costs. In the measure of this deviation are involved the functionals Di (i=1,2) which play the role of the square of the norm of the Banach space L1.

    The following theorem provides sufficient conditions for a strict strong minimum of problem (P).

    Theorem 2.1 Let (x0,u0) be an admissible process. Assume that Ia(˜x0()) is piecewise constant on T, suppose that there exist ρX, μUs with μα(t)0 and μα(t)φα(˜x0(t))=0 (αR,tT), two positive numbers δ,ϵ, and multipliers λ1,,λK with λi0 and λiIi(x0,u0)=0 (i=1,,k) such that

    ˙ρ(t)=Hx(˜x0(t),ρ(t),μ(t)) (a.e.in T),
    Hu(˜x0(t),ρ(t),μ(t))=0 (tT),

    and the following holds:

    (ⅰ) Huu(˜x0(t),ρ(t),μ(t))0 (a.e.in T).

    (ⅱ) J0((x0,u0);(y,v))>0 for all (y,v)(0,0), (y,v)Y(x0,u0).

    (ⅲ) If (x,u) is admissible with xx0<ϵ, then

    a. E0(t,x(t),u0(t),u(t))0 (a.e.in T).

    b. t1t0E0(t,x(t),u0(t),u(t))dtδmax{t1t0V(˙x(t)˙x0(t))dt,t1t0V(u(t)u0(t))dt}.

    c. t1t0E0(t,x(t),u0(t),u(t))dtδ|t1t0Eγ(t,x(t),u0(t),u(t))dt| (γ=1,,K).

    In this case, there exist θ1,θ2>0 such that if (x,u) is admissible with xx0<θ1,

    I(x,u)I(x0,u0)+θ2D(xx0,uu0).

    In particular, (x0,u0) is a strict strong minimum of (P).

    The theorem below gives sufficient conditions for weak minima of problem (P).

    Theorem 2.2 Let (x0,u0) be an admissible process. Assume that Ia(˜x0()) is piecewise constant on T, suppose that there exist ρX, μUs with μα(t)0 and μα(t)φα(˜x0(t))=0 (αR,tT), two positive numbers δ,ϵ, and multipliers λ1,,λK with λi0 and λiIi(x0,u0)=0 (i=1,,k) such that

    ˙ρ(t)=Hx(˜x0(t),ρ(t),μ(t)) (a.e.in T),
    Hu(˜x0(t),ρ(t),μ(t))=0 (tT),

    and the following holds:

    (ⅰ) Huu(˜x0(t),ρ(t),μ(t))0 (a.e.in T).

    (ⅱ) J0((x0,u0);(y,v))>0 for all (y,v)(0,0), (y,v)Y(x0,u0).

    (ⅲ) If (x,u) is admissible with (x,u)(x0,u0)<ϵ, then

    a. t1t0E0(t,x(t),u0(t),u(t))dtδt1t0V(u(t)u0(t))dt.

    b. t1t0E0(t,x(t),u0(t),u(t))dtδ|t1t0Eγ(t,x(t),u0(t),u(t))dt| (γ=1,,K).

    In this case, there exist θ1,θ2>0 such that if (x,u) is admissible with (x,u)(x0,u0)<θ1,

    I(x,u)I(x0,u0)+θ2D2(uu0).

    In particular, (x0,u0) is a strict weak minimum of (P).

    Examples 2.3 and 2.4 illustrate Theorems 2.1 and 2.2 respectively. It is worth mentioning that the sufficiency theory of [28] cannot be applied in both examples. Indeed, if f denotes the dynamic of the problems, as one readily verifies, in both examples, we have that

    fu(t,x,u)=(u2,u1) for all (t,x,u)[0,1]×R×R2,

    and hence, it does not exist a continuous function ψ:[0,1]×RR such that

    |fu(t,x,u)|ψ(t,x) for all (t,x,u)[0,1]×R×R2.

    Example 2.3 Let u02:[0,1]R be any measurable function whose codomain belongs to the set {1,1}.

    Consider problem (P) of minimizing

    I(x,u):=10{sinh(u1(t))+u21(t)cos(2πu2(t))x2(t)}dt

    over all couples (x,u) with x:[0,1]R absolutely continuous and u:[0,1]R2 essentially bounded satisfying the constraints

    {x(0)=x(1)=0.˙x(t)=u1(t)u2(t)+12x(t) (a.e.in[0,1]).I1(x,u):=10{14x2(t)+x(t)u1(t)u2(t)}dt0.(t,x(t),u(t))A (t[0,1])

    where

    A:={(t,x,u)[0,1]×R×R2u10, (u2u02(t))21, u22=1}.

    For this case, T=[0,1], n=1, m=2, r=2, s=3, k=K=1, ξ0=ξ1=0,

    L(t,x,u)=sinh(u1)+u21cos(2πu2)x2,f(t,x,u)=u1u2+12x,
    L1(t,x,u)=14x2+xu1u2,L0(t,x,u)=sinh(u1)+u21cos(2πu2)x2+λ1[14x2+xu1u2],
    φ1(t,x,u)=u1,φ2(t,x,u)=(u2u02(t))21,φ3(t,x,u)=u221.

    Clearly, L, L1, f and φ=(φ1,φ2,φ3) satisfy the hypotheses imposed in the statement of the problem.

    Also, as one readily verifies, the process (x0,u0)=(x0,u01,u02)(0,0,u02) is admissible.

    Moreover,

    H(t,x,u,ρ,μ)=ρu1u2+12ρxsinh(u1)u21cos(2πu2)+x2λ1[14x2+xu1u2]+μ1u1μ2[(u2u02(t))21]μ3[u221],
    Hx(t,x,u,ρ,μ)=12ρ+2xλ1[12x+u1u2],
    Hu(t,x,u,ρ,μ)=(ρu2cosh(u1)2u1cos(2πu2)λ1xu2+μ1ρu1+2πu21sin(2πu2)λ1xu12μ2(u2u02(t))2μ3u2).

    Therefore, if we set ρ0, μ11, μ2=μ30 and λ1=0, we have

    ˙ρ(t)=Hx(˜x0(t),ρ(t),μ(t)) (a.e. in T),Hu(˜x0(t),ρ(t),μ(t))=(0,0) (tT),

    and hence the first order sufficient conditions involving the Hamiltonian of problem (P) are verified. Moreover, if we set R:={1,2}, observe that

    λ10,λ1I1(x0,u0)=0,
    μα(t)0,μα(t)φα(˜x0(t))=0(αR,tT).

    Additionally, Ia(˜x0()){1} is constant on T. Also, it is readily seen that for all tT,

    Huu(˜x0(t),ρ(t),μ(t))=(2000),

    and so condition (ⅰ) of Theorem 2.1 is satisfied. Observe that, for all tT,

    fx(˜x0(t))=12,fu(˜x0(t))=(u02(t),0),L1x(˜x0(t))=0,L1u(˜x0(t))=(0,0),
    φ1x(˜x0(t))=0,φ1u(˜x0(t))=(1,0),φ3x(˜x0(t))=0,φ3u(˜x0(t))=(0,2u02(t)).

    Thus, Y(x0,u0) is given by all (y,v)X×L2(T;R2) satisfying

    {y(0)=y(1)=0.˙y(t)=12y(t)+u02(t)v1(t) (a.e. in T).v1(t)0 (a.e. in T).2u02(t)v2(t)=0 (a.e. in T).

    Moreover, note that, for all (t,x,u)T×R×R2,

    F0(t,x,u)=H(t,x,u,ρ(t),μ(t))˙ρ(t)x=sinh(u1)+u21cos(2πu2)x2u1,

    and so, for all tT,

    F0xx(˜x0(t))=2,F0xu(˜x0(t))=(0,0),F0uu(˜x0(t))=(2000).

    Consequently, we have

    12J0((x0,u0);(y,v))=10{v21(t)y2(t)}dt=10{(˙y(t)12y(t))2y2(t)}dt=10{˙y2(t)y(t)˙y(t)34y2(t)}dt=10{˙y2(t)34y2(t)}dt>0

    for all (y,v)(0,0), (y,v)Y(x0,u0). Hence, condition (ⅱ) of Theorem 2.1 is verified.

    Additionally, observe that for all (x,u) admissible and all tT,

    E0(t,x(t),u0(t),u(t))=sinh(u1(t))+u21(t)cos(2πu2(t))u1(t)u21(t)cos(2πu02(t))=u21(t)0,

    and then, condition (ⅲ)(a) of Theorem 2.1 is satisfied for any ϵ>0. Now, if (x,u) is admissible, then

    uu0=(u1u01,u2u02)=(u1,u02u02)=(u1,0)

    and so, if (x,u) is admissible,

    10E0(t,x(t),u0(t),u(t))dt10u21(t)dt10V(u1(t))dt=10V(u(t)u0(t))dt.

    Also, if (x,u) is admissible, then

    10E0(t,x(t),u0(t),u(t))dt10u21(t)dt=10{(u1(t)u2(t)+12x(t))2x(t)u1(t)u2(t)14x2(t)}dt=10{˙x2(t)x(t)u1(t)u2(t)14x2(t)}dt10˙x2(t)dt10V(˙x(t)˙x0(t))dt.

    Therefore, if (x,u) is admissible, then

    10E0(t,x(t),u0(t),u(t))dtmax{10V(˙x(t)˙x0(t))dt,10V(u(t)u0(t))dt},

    and hence, condition (ⅲ)(b) of Theorem 2.1 is verified for any ϵ>0 and δ=1. Finally, if (x,u) is admissible, note that

    |10E1(t,x(t),u0(t),u(t))dt|=|10x(t)u1(t)u2(t)dt|=|10{x(t)˙x(t)12x2(t)}dt|=|1012x2(t)dt|=1210x2(t)dt10˙x2(t)dt=10(u1(t)u2(t)+12x(t))2dt=10u21(t)dt+10{x(t)u1(t)u2(t)+14x2(t)}dt10u21(t)dt+10x(t)˙x(t)dt=10u21(t)dt10E0(t,x(t),u0(t),u(t))dt,

    implying that condition (ⅲ)(c) of Theorem 2.1 holds for any ϵ>0 and δ=1. By Theorem 2.1, (x0,u0) is a strict strong minimum of (P).

    Example 2.4 Consider problem (P) of minimizing

    I(x,u):=10{sinh(u1(t)+u1(t)x3(t))+12u21(t)cos(2πu2(t))cosh(x(t))+1}dt

    over all couples (x,u) with x:[0,1]R absolutely continuous and u:[0,1]R2 essentially bounded satisfying the constraints

    {x(0)=x(1)=0.˙x(t)=u1(t)u2(t)+x(t) (a.e.in[0,1]).I1(x,u):=10{sin(u1(t))sinh(u1(t)+u1(t)x3(t))}dt0.(t,x(t),u(t))A (t[0,1])

    where

    A:={(t,x,u)[0,1]×R×R2sin(u1)0, u22=1}.

    For this case, T=[0,1], n=1, m=2, r=1, s=2, k=K=1, ξ0=ξ1=0,

    L(t,x,u)=sinh(u1+u1x3)+12u21cos(2πu2)cosh(x)+1,f(t,x,u)=u1u2+x,
    L1(t,x,u)=sin(u1)sinh(u1+u1x3),
    L0(t,x,u)=sinh(u1+u1x3)+12u21cos(2πu2)cosh(x)+1+λ1[sin(u1)sinh(u1+u1x3)],
    φ1(t,x,u)=sin(u1),φ2(t,x,u)=u221.

    Clearly, L, L1, f and φ=(φ1,φ2) satisfy the hypotheses imposed in the statement of the problem.

    Let u02:TR be any measurable function whose codomain belongs to the set {1,1}.

    Clearly, the process (x0,u0)=(x0,u01,u02)(0,0,u02) is admissible.

    Moreover,

    H(t,x,u,ρ,μ)=ρu1u2+ρxsinh(u1+u1x3)12u21cos(2πu2)+cosh(x)1λ1[sin(u1)sinh(u1+u1x3)]+μ1sin(u1)μ2[u221],
    Hx(t,x,u,ρ,μ)=ρ3x2u1cosh(u1+u1x3)+sinh(x)+3λ1x2u1cosh(u1+u1x3),
    Hu1(t,x,u,ρ,μ)=ρu2[1+x3]cosh(u1+u1x3)u1cos(2πu2)λ1[cos(u1){1+x3}cosh(u1+u1x3)]+μ1cos(u1),
    Hu2(t,x,u,ρ,μ)=ρu1+πu21sin(2πu2)2μ2u2.

    Therefore, if we set ρ0, μ11, μ20 and λ1=0, we have

    ˙ρ(t)=Hx(˜x0(t),ρ(t),μ(t)) (a.e. in T),Hu(˜x0(t),ρ(t),μ(t))=(0,0) (tT),

    and hence the first order sufficient conditions involving the Hamiltonian of problem (P) are verified. Additionally, observe that

    λ10,λ1I1(x0,u0)=0,
    μ1(t)0,μ1(t)φ1(˜x0(t))=0(tT).

    Also, Ia(˜x0()){1} is constant on T. Moreover, it is readily seen that for all tT,

    Huu(˜x0(t),ρ(t),μ(t))=(1000),

    and so condition (ⅰ) of Theorem 2.2 is satisfied. Observe that, for all tT,

    fx(˜x0(t))=1,fu(˜x0(t))=(u02(t),0),L1x(˜x0(t))=0,L1u(˜x0(t))=(0,0),
    φ1x(˜x0(t))=0,φ1u(˜x0(t))=(1,0),φ2x(˜x0(t))=0,φ2u(˜x0(t))=(0,2u02(t)).

    Thus, Y(x0,u0) is given by all (y,v)X×L2(T;R2) satisfying

    {y(0)=y(1)=0.˙y(t)=y(t)+u02(t)v1(t) (a.e. in T).v1(t)0 (a.e. in T).2u02(t)v2(t)=0 (a.e. in T).

    Also, note that, for all (t,x,u)T×R×R2,

    F0(t,x,u)=H(t,x,u,ρ(t),μ(t))˙ρ(t)x=sinh(u1+u1x3)+12u21cos(2πu2)cosh(x)+1sin(u1),

    and so, for all tT,

    F0xx(˜x0(t))=1,F0xu(˜x0(t))=(0,0),F0uu(˜x0(t))=(1000).

    Consequently, we have

    J0((x0,u0);(y,v))=10{v21(t)y2(t)}dt=10{(˙y(t)y(t))2y2(t)}dt=10{˙y2(t)2y(t)˙y(t)}dt=10˙y2(t)dt>0

    for all (y,v)(0,0), (y,v)Y(x0,u0). Hence, condition (ⅱ) of Theorem 2.2 is verified.

    Additionally, observe that for any ϵ(0,1), all (x,u) admissible satisfying (x,u)(x0,u0)<ϵ and all tT,

    E0(t,x(t),u0(t),u(t))=sinh(u1(t)+u1(t)x3(t))+12u21(t)cos(2πu2(t))sin(u1(t))=sinh(u1(t)+u1(t)x3(t))+12u21(t)cos(2πu02(t))sin(u1(t))=sinh(u1(t)+u1(t)x3(t))+12u21(t)sin(u1(t)).

    Therefore, for any ϵ(0,1) and all (x,u) admissible satisfying (x,u)(x0,u0)<ϵ,

    10E0(t,x(t),u0(t),u(t))dt=10{sinh(u1(t)+u1(t)x3(t))sin(u1(t))+12u21(t)}dt1012u21(t)dt10V(u1(t))dt=10V(u(t)u0(t))dt,

    and hence condition (ⅲ)(a) of Theorem 2.2 is satisfied for any ϵ(0,1) and δ=1.

    Finally, if (x,u) is admissible, note that

    10E0(t,x(t),u0(t),u(t))dt|10{sinh(u1(t)+u1(t)x3(t))sin(u1(t))}dt|=|10E1(t,x(t),u0(t),u(t))dt|,

    implying that condition (ⅲ)(b) of Theorem 2.2 holds for any ϵ>0 and δ=1. By Theorem 2.2, (x0,u0) is a strict weak minimum of (P).

    In this section we shall prove Theorem 2.1. We first state an auxiliary result whose proof is given in Lemmas 2–4 of [31].

    In the following lemma we shall assume that we are given z0:=(x0,u0)X×L1(T;Rm) and a subsequence {zq:=(xq,uq)} in X×L1(T;Rm) such that

    limqD(zqz0)=0anddq:=[2D(zqz0)]1/2>0(qN).

    For all qN, set

    yq:=xqx0dqandvq:=uqu0dq.

    For all qN, define

    Wq:=max{W1q,W2q}

    where

    W1q:=[1+12V(˙xq˙x0)]1/2andW2q:=[1+12V(uqu0)]1/2.

    As we mentioned in the introduction, we do not relabel the subsequences of a given sequence since as one readily verifies this fact will not alter our results.

    Lemma 3.1

    a. For some v0L2(T;Rm) and some subsequence of {zq}, vqL1v0 on T. Even more, uqauu0 on T.

    b. There exist ζ0L2(T;Rn), ˉy0Rn, and some subsequence of {zq}, such that ˙yqL1ζ0 on T. Moreover, if y0(t):=ˉy0+tt0ζ0(τ)dτ (tT), then yquy0 on T.

    c. Let ΥT be measurable and suppose that Wqu1 on Υ. Let Rq,R0L(Υ;Rm×m), assume that RquR0 on Υ, R0(t)0 (tΥ), and let v0 be the function considered in condition (a) of Lemma 3.1. Then,

    lim infqΥvq(t)Rq(t)vq(t)dtΥv0(t)R0(t)v0(t)dt.

    Proof. The proof of Theorem 2.1 will be made by contraposition, that is, we shall assume that for all θ1,θ2>0, there exists an admissible process (x,u) such that

    xx0<θ1andI(x,u)<I(x0,u0)+θ2D(xx0,uu0). (1)

    Also, we are going to assume that all the hypotheses of Theorem 2.1 are satisfied with the exception of hypothesis (ⅱ) and we will obtain the negation of condition (ⅱ) of Theorem 2.1. First of all, note that since

    μα(t)0 (αR,tT)andλi0 (i=1,,k),

    if (x,u) is admissible, then I(x,u)J0(x,u). Also, since

    μα(t)φα(˜x0(t))=0 (αR,tT)andλiIi(x0,u0)=0 (i=1,,k),

    then I(x0,u0)=J0(x0,u0). Thus, (1) implies that for all θ1,θ2>0, there exists (x,u) admissible with xx0<θ1 and

    J0(x,u)<J0(x0,u0)+θ2D(xx0,uu0). (2)

    Let z0:=(x0,u0). Note that, for all admissible processes z=(x,u),

    J0(z)=J0(z0)+J0(z0;zz0)+K0(z)+E0(z) (3)

    where

    E0(x,u):=t1t0E0(t,x(t),u0(t),u(t))dt,
    K0(x,u):=t1t0{M0(t,x(t))+[u(t)u0(t)]N0(t,x(t))}dt,

    and the functions M0 and N0 are given by

    M0(t,y):=F0(t,y,u0(t))F0(˜x0(t))F0x(˜x0(t))(yx0(t)),
    N0(t,y):=F0u(t,y,u0(t))F0u(˜x0(t)).

    We have,

    M0(t,y)=12[yx0(t)]P0(t,y)(yx0(t)),N0(t,y)=Q0(t,y)(yx0(t)),

    where

    P0(t,y):=210(1λ)F0xx(t,x0(t)+λ[yx0(t)],u0(t))dλ,
    Q0(t,y):=10F0ux(t,x0(t)+λ[yx0(t)],u0(t))dλ.

    Now, as in [28], choose ν>0 such that for all z=(x,u) admissible with xx0<1,

    |K0(x,u)|νxx0[1+D(zz0)]. (4)

    Now, by (2), for all qN there exists zq:=(xq,uq) admissible such that

    xqx0<ϵ,xqx0<1q,J0(zq)J0(z0)<1qD(zqz0). (5)

    The last inequality of (5) implies that zqz0 and so for all qN,

    dq:=[2D(zqz0)]1/2>0.

    Since

    ˙ρ(t)=Hx(˜x0(t),ρ(t),μ(t)) (a.e. in T),Hu(˜x0(t),ρ(t),μ(t))=0 (tT),

    it follows that J0(z0;(y,v))=0 for all (y,v)X×L2(T;Rm). With this in mind, by (3), condition (ⅲ)(b) of Theorem 2.1, (4) and (5),

    J0(zq)J0(z0)=K0(zq)+E0(zq)νxqx0+D(zqz0)(δνxqx0).

    By (5), for all qN,

    D(zqz0)(δ1qνq)<νq

    and hence

    limqD(zqz0)=0.

    For all qN, define

    yq:=xqx0dqandvq:=uqu0dq.

    By condition (a) of Lemma 3.1, there exist v0L2(T;Rm) and a subsequence of {zq} such that vqL1v0 on T. By condition (b) of Lemma 3.1, there exist ζ0L2(T;Rn), ˉy0Rn and a subsequence of {zq} such that, if for all tT, y0(t):=ˉy0+tt0ζ0(τ)dτ, then yquy0 on T.

    We claim that

    ⅰ. J0(z0;(y0,v0))0, (y0,v0)(0,0).

    ⅱ. ˙y0(t)=fx(˜x0(t))y0(t)+fu(˜x0(t))v0(t) (a.e. in T), y0(ti)=0 (i=0,1).

    ⅲ. Ii(z0;(y0,v0))0 (iia(z0)), Ij(z0;(y0,v0))=0 (j=k+1,,K).

    ⅳ. φαx(˜x0(t))y0(t)+φαu(˜x0(t))v0(t)0 (a.e. in T,αIa(˜x0(t))).

    ⅴ. φβx(˜x0(t))y0(t)+φβu(˜x0(t))v0(t)=0 (a.e. in T,βS).

    Indeed, the equalities y0(ti)=0 (i=0,1) follow from the definition of yq, the admissibility of zq and the fact that yquy0 on T.

    For all qN, we have

    K0(zq)d2q=t1t0{M0(t,xq(t))d2q+vq(t)N0(t,xq(t))dq}dt.

    By condition (b) of Lemma 3.1,

    M0(,xq())d2qL12y0()F0xx(˜x0())y0(),
    N0(,xq())dqLF0ux(˜x0())y0(),

    both on T and, since vqL1v0 on T,

    12J0(z0;(y0,v0))=limqK0(zq)d2q+12t1t0v0(t)F0uu(˜x0(t))v0(t)dt. (6)

    We have,

    lim infqE0(zq)d2q12t1t0v0(t)F0uu(˜x0(t))v0(t)dt. (7)

    Indeed, by condition (a) of Lemma 3.1, we are able to choose ΥT measurable such that uquu0 on Υ. Since zq is admissible, then recalling the definition of Wq given in the beginning of this section, as one readily verifies, Wqu1 on Υ. Moreover, for all tΥ and qN,

    1d2qE0(t,xq(t),u0(t),uq(t))=12vq(t)Rq(t)vq(t)

    where

    Rq(t):=210(1λ)F0uu(t,xq(t),u0(t)+λ[uq(t)u0(t)])dλ.

    Clearly,

    Rq()uR0():=F0uu(˜x0()) on Υ.

    By condition (ⅰ) of Theorem 2.1, R0(t)0 (tΥ). Additionally, by condition (ⅲ)(a) of Theorem 2.1, for all qN,

    E0(t,xq(t),u0(t),uq(t))0(a.e. in T),

    and so, by condition (c) of Lemma 3.1,

    lim infqE0(zq)d2q=lim infq1d2qt1t0E0(t,xq(t),u0(t),uq(t))dtlim infq1d2qΥE0(t,xq(t),u0(t),uq(t))dt=12lim infqΥvq(t)Rq(t)vq(t)dt12Υv0(t)R0(t)v0(t)dt.

    As Υ can be chosen to differ from T by a set of an arbitrarily small measure and the function

    tv0(t)R0(t)v0(t)

    belongs to L1(T;R), this inequality holds when Υ=T and this establishes (7). By (3) and (5)–(7),

    12J0(z0;(y0,v0))limqK0(zq)d2q+lim infqE0(zq)d2q=lim infqJ0(zq)J0(z0)d2q0.

    If (y0,v0)=(0,0), then

    limqK0(zq)d2q=0

    and so, by condition (ⅲ)(b) of Theorem 2.1,

    12δlim infqE0(zq)d2q0,

    which contradicts the positivity of δ.

    For all qN, we have

    ˙yq(t)=Aq(t)yq(t)+Bq(t)vq(t) (a.e. in T),yq(t0)=0,

    where

    Aq(t)=10fx(t,x0(t)+λ[xq(t)x0(t)],u0(t))dλ,
    Bq(t)=10fu(t,xq(t),u0(t)+λ[uq(t)u0(t)])dλ.

    Since

    Aq()uA0():=fx(˜x0()),Bq()uB0():=fu(˜x0()),

    yquy0 and vqL1v0 all on Υ, it follows that ˙yqL1A0y0+B0y0 on Υ. By condition (b) of Lemma 3.1, ˙yqL1ζ0=˙y0 on Υ. Therefore,

    ˙y0(t)=A0(t)y0(t)+B0(t)v0(t)(tΥ).

    As Υ can be chosen to differ from T by a set of an arbitrarily small measure, then there cannot exist a subset of T of positive measure in which the functions y0 and v0 do not satisfy the differential equation ˙y0(t)=A0(t)y0(t)+B0(t)v0(t). Consequently,

    ˙y0(t)=A0(t)y0(t)+B0(t)v0(t)(a.e. in T)

    and (ⅰ) and (ⅱ) of our claim are proved.

    Finally, in order to obtain (ⅲ)–(ⅴ) of our claim it is enough to copy the proofs of [28] from Eqs (8)–(15).

    In this section we shall prove Theorem 2.2. We first state an auxiliary result which is an immediate consequence of Lemmas 3.1 and 3.2 of [30].

    In the following lemma we shall assume that we are given u0L1(T;Rm) and a sequence {uq} in L1(T;Rm) such that

    limqD2(uqu0)=0andd2q:=[2D2(uqu0)]1/2>0(qN).

    For all qN define

    v2q:=uqu0d2q.

    Lemma 4.1

    a. For some v02L2(T;Rm) and a subsequence of {uq}, v2qL1v02 on T.

    b. Let AqL(T;Rn×n) and BqL(T;Rn×m) be matrix functions for which there exist constants m0,m1>0 such that Aqm0, Bqm1 (qN), and for all qN denote by Yq the solution of the initial value problem

    ˙y(t)=Aq(t)y(t)+Bq(t)v2q(t) (a.e.in T),y(t0)=0.

    Then there exist σ0L2(T;Rn) and a subsequence of {zq}, such that ˙YqL1σ0 on T, and hence if Y0(t):=tt0σ0(τ)dτ (tT), then YquY0 on T.

    Proof. As we made with the proof of Theorem 2.1, the proof of Theorem 2.2 will be made by contraposition, that is, we shall assume that for all θ1,θ2>0, there exists an admissible process (x,u) such that

    (x,u)(x0,u0)<θ1andI(x,u)<I(x0,u0)+θ2D2(uu0). (8)

    Once again, as we made with the proof of Theorem 2.1, (8) implies that for all θ1,θ2>0, there exists (x,u) admissible with

    (x,u)(x0,u0)<θ1andJ0(x,u)<J0(x0,u0)+θ2D2(uu0). (9)

    Let z0:=(x0,u0). As in the proof of Theorem 2.1, for all admissible processes z=(x,u),

    J0(z)=J0(z0)+J0(z0;zz0)+K0(z)+E0(z)

    where E0 and K0 are given as in the proof of Theorem 2.1.

    Now, by (9), for all qN there exists zq:=(xq,uq) admissible such that

    zqz0<1q,J0(zq)J0(z0)<1qD2(uqu0). (10)

    Since zq is admissible, the last inequality of (10) implies that uqu0 and so

    d2q:=[2D2(uqu0)]1/2>0(qN).

    By the first relation of (10), we have

    limqD2(uqu0)=0.

    For all qN, define v2q as in Lemma 4.1 and

    Yq:=xqx0d2qandW2q:=[1+12V(uqu0)]1/2.

    By condition (a) of Lemma 4.1, there exist v02L2(T;Rm) and a subsequence of {zq} such that v2qL1v02 on T. As in the proof of Theorem 2.1, for all qN,

    ˙Yq(t)=Aq(t)Yq(t)+Bq(t)v2q(t),Yq(t0)=0(a.e. in T).

    We have the existence of m0,m1>0 such that Aqm0 and Bqm1 (qN). By condition (b) of Lemma 4.1, there exist σ0L2(T;Rn) and a subsequence of {zq} such that, if Y0(t):=tt0σ0(τ)dτ (tT), then YquY0 on T. We claim that

    ⅰ. J0(z0;(Y0,v02))0, (Y0,v02)(0,0).

    ⅱ. ˙Y0(t)=fx(˜x0(t))Y0(t)+fu(˜x0(t))v02(t) (a.e. in T), Y0(ti)=0 (i=0,1).

    ⅲ. Ii(z0;(Y0,v02))0 (iia(z0)), Ij(z0;(Y0,v02))=0 (j=k+1,,K).

    ⅳ. φαx(˜x0(t))Y0(t)+φαu(˜x0(t))v02(t)0 (a.e. in T,αIa(˜x0(t))).

    ⅴ. φβx(˜x0(t))Y0(t)+φβu(˜x0(t))v02(t)=0 (a.e. in T,βS).

    Indeed, for all qN, we have

    K0(zq)d22q=t1t0{M0(t,xq(t))d22q+v2q(t)N0(t,xq(t))d2q}dt.

    Also, we have

    M0(,xq())d22qL12Y0()F0xx(˜x0())Y0(),
    N0(,xq())d2qLF0ux(˜x0())Y0(),

    both on T and, since v2qL1v02 on T,

    12J0(z0;(Y0,v02))=limqK0(zq)d22q+12t1t0v02(t)F0uu(˜x0(t))v02(t)dt. (11)

    Now, for all tT and qN,

    1d22qE0(t,xq(t),u0(t),uq(t))=12v2q(t)Rq(t)v2q(t)

    where

    Rq(t):=210(1λ)F0uu(t,xq(t),u0(t)+λ[uq(t)u0(t)])dλ.

    Clearly,

    Rq()LR0():=F0uu(˜x0()) on T.

    Since zqz00 as q, it follows that W2qL1 on T and, by condition (ⅰ) of Theorem 2.2, R0(t)0 (a.e. in T). Consequently,

    lim infqE0(zq)d22q12t1t0v02(t)R0(t)v02(t)dt. (12)

    On the other hand, since

    ˙ρ(t)=Hx(˜x0(t),ρ(t),μ(t)) (a.e. in T),Hu(˜x0(t),ρ(t),μ(t))=0 (tT),

    we have that J0(z0;(y,v))=0 for all (y,v)X×L2(T;Rm). With this in mind, (10)–(12),

    12J0(z0;(Y0,v02))limqK0(zq)d22q+lim infqE0(zq)d22q=lim infqJ0(zq)J0(z0)d22q0.

    If (Y0,v02)=(0,0), then

    limqK0(zq)d22q=0

    and so, by condition (ⅲ)(a) of Theorem 2.2,

    12δlim infqE0(zq)d22q0,

    which contradicts the positivity of δ and this proves (ⅰ) of our claim.

    Now, we also claim that

    ˙Y0(t)=fx(˜x0(t))Y0(t)+fu(˜x0(t))v02(t)(a.e. in T),Y0(ti)=0(i=0,1).

    Indeed, the equalities Y0(ti)=0 (i=0,1) follow from the definition of Yq, the admissibility of zq and the fact that YquY0 on T. Also, observe that since YquY0,

    Aq()LA0():=fx(˜x0()),
    Bq()LB0():=fu(˜x0()),

    and v2qL1v02 all on T, then ˙YqL1A0Y0+B0v02 on T. By condition (b) of Lemma 4.1, ˙YqL1σ0=˙Y0 on T, which accordingly implies that

    ˙Y0(t)=A0(t)Y0(t)+B0(t)v02(t)(a.e. in T)

    and our claim is proved.

    Finally, in order to prove (ⅲ)–(ⅴ) of our claim it is enough to copy the proofs given in [28] from Eqs (8)–(15) by replacing y0 by Y0, v0 by v02 and Υ by T.

    In this article, we have provided sufficiency theorems for weak and strong minima in an optimal control problem of Lagrange with fixed end-points, nonlinear dynamics, inequality and equality isoperimetric restrictions and inequality and equality mixed time-state-control constraints. The sufficiency treatment studied in this paper does not need that the proposed optimal controls be continuous but only purely measurable. The sufficiency results not only provide local minima but they also measure the deviation between optimal and admissible costs by means of a functional playing a similar role of the square of the classical norm of the Banach space L1. Additionally, all the crucial sufficiency hypotheses are included in the theorems, in contrast, with other necessary and sufficiency theories which strongly depend upon some preliminary assumptions not embedded in the corresponding theorems of optimality. Finally, our sufficiency technique is self-contained because it is independent of some classical sufficient approaches involving Hamilton-Jacobi inequalities, matrix-valued Riccati equations, generalizations of Jacobi's theory appealing to extended notions of conjugate points or insertions of the original problem in some abstract Banach spaces.

    The author is thankful to Dirección General de Asuntos del Personal Académico, Universidad Nacional Autónoma de México, for the support provided by the project PAPIIT-IN102220. Moreover, the author thanks the two anonymous referees for the encouraging suggestions made in their reviews.

    The author declares no conflict of interest.



    [1] K. Diethelm, The analysis of fractional differential equations, Springer, 2010.
    [2] C. Huang, B. Liu, New studies on dynamic analysis of inertial neural networks involving non-reduced order method, Neurocomputing, 325 (2019), 283–287.
    [3] J. Wang, X. Chen, L. Huang, The number and stability of limit cycles for planar piecewise linear systems of node-saddle type, J. Math. Anal. Appl., 469 (2019), 405–427.
    [4] Y. Zuo, Y. Wang, X. Liu, Adaptive robust control strategy for rhombus-type lunar exploration wheeled mobile robot using wavelet transform and probabilistic neural network, Comput. Appl. Math., 37 (2018), 314–337.
    [5] C. Song, S. Fei, J. Cao, C. Huang, Robust synchronization of fractional-order uncertain chaotic systems based on output feedback sliding mode control, Mathematics, 7 (2019), 599.
    [6] C. Huang, J. Cao, F. Wen, X. Yang, Stability analysis of SIR model with distributed delay on complex networks, PLOS One, 11 (2016).
    [7] I. Podlubny, Geometric and physical interpretation of fractional integration and fractional differentiation, Fract. Calc. Appl. Anal., 5 (2002), 367–386.
    [8] Z. Rashidnejad, P. Karimaghaee, Synchronization of a class of uncertain chaotic systems utilizing a new finite-time fractional adaptive sliding mode control, Chaos, Solitons and Fractals: X, 5 (2020), 100042.
    [9] Y. Zhou, F. Jiao, J. Pecaric, On the Cauchy problem for fractional functional differential equations in Banach spaces, Topol. Method. Nonl. An., 42 (2013), 119–136.
    [10] A. Chauhan, J. Dabas, Local and global existence of mild solution to an impulsive fractional functional integro-differential equation with nonlocal condition, Commun. Nonlinear Sci., 19 (2014), 821–829.
    [11] L. Xu, X. Chu, H. Hu, Exponential ultimate bounded ness of non-autonomous fractional differential systems with time delay and impulses, Appl. Math. Lett., 99 (2020), 106000.
    [12] J. T. Edwards, N. J. Ford, A. C. Simpson, The numerical solution of linear multi-term fractional differential equations: systems of equations, J. Comput. Appl. Math., 148 (2002), 401–418.
    [13] A. Arbi, J. Cao, Pseudo-almost periodic solution on time-space scales for a novel class of competitive neutral-type neural networks with mixed time-varying delays and leakage delays, Neural Process. Lett., 46 (2017), 719–745.
    [14] Y. Liu, W. Liu, M. A. Obaid, I. A. Abbas, Exponential stability of Markovian jumping Cohen-Grossberg neural networks with mixed mode-dependent time-delays, Neurocomputing, 177 (2016), 409–415.
    [15] I. Stamova, T. Stamov, X. Li, Global exponential stability of a class of impulsive cellular neural networks with Supremums, Int. J. Adapt. Control, 28 (2014), 1227–1239.
    [16] X. Li, S. Song, Impulsive control for existence, uniqueness and global stability of periodic solutions of recurrent neural networks with discrete and continuously distributed delays, IEEE T. Neur. Net. Lear., 24(2013), 868–877.
    [17] Q. Zhu, J. Cao, Stability analysis of Markovian jump stochastic BAM neural networks with impulse control and mixed time delays, IEEE T. Neur. Net. Lear., 23 (2012), 467–479.
    [18] M. Syed Ali, P. Balasubramaniam, F. A. Rihan, S. Lakshmanan, Stability criteria for stochastic Takagi-Sugeno fuzzy Cohen-Grossberg BAM neural networks with mixed time-varying delays, Complexity, 21 (2016), 143–154.
    [19] H. Zhang, M. Ye, J. Cao, A. Alsaedi, Synchronization control of Riemann-Liouville fractional competitive network systems with time-varying delay and different time scales, Int. J. Control Autom., 16 (2018), 1–11.
    [20] P. Liu, X. Nie, J. Liang, J. Cao, Multiple Mittag-Leffler stability of fractional-order competitive neural networks with gaussian activation functions, Neural Networks, 108 (2018), 452–465.
    [21] L. J. Banu, P. Balasubramaniam, Robust stability analysis for discrete-time neural networks with time-varying leakage delays and random parameter uncertainties, Neurocomputing, 179 (2016), 126–134.
    [22] D. J. Lu, C. J. Li, Exponential stability of stochastic high-order BAM neural networks with time delays and impulsive effects, Neural Comput. Appl., 23 (2013), 1–8.
    [23] X. Lv, X. Li, Finite time stability and controller design for nonlinear impulsive sampled-data systems with applications, ISA T., 70 (2017), 30–36.
    [24] A. L. Wu, Z. G. Zeng, X. G. Song, Global Mittag-Leffler stabilization of fractional-order bidirectional associative memory neural networks, Neurocomputing, 177 (2016), 489–496.
    [25] F. Wang, Y. Q. Yang, X. Y. Xu, L. Li, Global asymptotic stability of impulsive fractional-order BAM neural networks with time delay, Neural Comput. Appl., 28 (2017), 345–352.
    [26] W. Chen, W. Zheng, Robust stability analysis for stochastic neural networks with time-varying delay, IEEE T. Neural Networks, 21 (2010), 508–514.
    [27] H. Gu, Adaptive synchronization for competitive neural networks with different time scales and stochastic perturbation, Neurocomputing, 73 (2009), 350–356.
    [28] O. M. Kwon, S. M. Lee, J. H. Park, Improved delay-dependent exponential stability for uncertain stochastic neural networks with time-varying delays, Phys. Lett. A, 374 (2010), 1232–1241.
    [29] J. H. Park, O. M. Kwon, Analysis on global stability of stochastic neural networks of neutral type, Mod. Phys. Lett. B, 22 (2008), 3159–3170.
    [30] J. H. Park, O. M. Kwon, Synchronization of neural networks of neutral type with stochastic perturbation, Mod. Phys. Lett. B, 23 (2009), 1743–1751.
    [31] J. H. Park, S. M. Lee, H. Y. Jung, LMI optimization approach to synchronization of stochastic delayed discrete-time complex networks, J. Optimiz. Theory Appl., 143 (2009), 357–367.
    [32] W. Su, Y. Chen, Global robust stability criteria of stochastic Cohen-Grossberg neural networks with discrete and distributed time-varying delays, Commun. Nonlinear Sci., 14 (2009), 520–528.
    [33] X. Yang, J. Cao, Y. Long, R. Wei, Adaptive lag synchronization for competitive neural networks with mixed delays and uncertain hybrid perturbations, IEEE T. Neural Networks, 21 (2010), 1656–1667.
    [34] X. Yang, C. Huang, J. Cao, An LMI approach for exponential synchronization of switched stochastic competitive neural networks with mixed delays, Neural Comput. Appl., 21 (2012), 2033–2047.
    [35] D. G. Hobson, L. C. G. Rogers, Complete models with stochastic volatility, Mathematical Finance, 8 (1998), 27–48.
    [36] J. Hull, A. White, The pricing of options on assets with stochastic volatilities, Journal of Finance, 42 (1987), 281–300.
    [37] I. Karatzas, Steven E. Shreve, Brownian motion and stochastic calculus, Springer-Verlag, 1991.
    [38] N. Laskin, Fractional market dynamics, Physica A, 287 (2000), 482–492.
    [39] C. Andoh, Stochastic variance models in discrete time with feed forward neural networks, Neural Computation, 21 (2009), 1990–2008.
    [40] S. Giebel, M. Rainer, Neural network calibrated stochastic processes: forecasting financial assets, Cent. Eur. J. Oper. Res., 21 (2013), 277–293.
    [41] J. Cao, Q. Yang, Z. Huang, Q. Liu, Asymptotically almost periodic solutions of stochastic functional differential equations, Appl. Math. Comput., 218 (2011), 1499–1511.
    [42] V. B. Kolmanovskii, A. Myshkis, Applied Theory of Functional Differential Equations, Kluwer Academic Publishers, Norwell, 1992.
    [43] Y. Ren, L. Chen, A note on the neutral stochastic functional differential equations with infinite delay and Possion jumps in an abstract space, J. Math. Phys., 50 (2009), 082704.
    [44] Y. Ren, N. Xia, Existence uniqueness and stability of the solutions to neutral stochastic functional differential equations with infinite delay, Appl. Math. Comput., 210 (2009), 72–79.
    [45] R. Sakthivel, J. Luo, Asymptotic stability of impulsive stochastic partial differential equations with infinite delays, J. Math. Anal. Appl., 356 (2009), 1–6.
    [46] R. Sakthivel, J. Luo, Asymptotic stability of nonlinear impulsive stochastic differential equations, Stat. Probabil. Lett., 79 (2009), 1219–1223.
    [47] R. Sakthivel, J. J. Nieto, N. I. Mahmudov, Approximate controllability of nonlinear deterministic and stochastic systems with unbounded delay, Taiwan. J. Math., 14 (2010), 1777–1797.
    [48] B. Xie, Stochastic differential equations with non-lipschitz coefficients in Hilbert spaces, Stoch. Anal. Appl., 26 (2008), 408–433.
    [49] L. Arnold, Stochastic differential equations: theory and applications, Wiley, 1974.
    [50] I. I. Gihman, A. V. Skorohod, Stochastic differential equations, Springer, 1972.
    [51] G. S. Ladde, V. Lakshmikantham, Random differential inequalities, New York: Academic Press, 1980.
    [52] A. Friedman, Stochastic differential equations and Applications, Academic Press, 1975.
    [53] H. Holden, B. Oksendal, J. Uboe, T. Zhang, Stochastic partial differential equations: A modeling, white noise functional approach, Boston: BirkhMauser, 1996.
    [54] F. Biagini, Y. Hu, B. Oksendal, T. Zhang, Stochastic calculus for fractional Brownian motion and applications, Springer, 2008.
    [55] D. He, L. Xu, Boundedness analysis of stochastic integro-differential systems with Levy noise, J. Taibah Univ. Sci., 14 (2020), 87–93.
    [56] A. G. Ladde, G. S. Ladde, Dynamic processes under random Environment, Bulletin of the Marathwada Mathematical Society, 8 (2007), 96–123.
    [57] J. Chen, Z. Zeng, P. Jiang, Global mittag-leffler stability and synchronization of memristor-based fractional-order neural networks, Neural Networks, 51 (2014), 1–8.
    [58] P. L. Butzer, U. Westphal, An Introduction to Fractional Calculus, World Scientific: Singapore, 2001.
    [59] X. Yang, Q. Song, Y. Liu, Z. Zhao, Finite-time stability analysis of fractional-order neural networks with delay, Neurocomputing, 152 (2015), 19–26.
    [60] L. Chen, R. Wu, D. Pan, Mean square exponential stability of impulsive stochastic fuzzy cellular neural networks with distributed delays, Expert Syst. Appl., 38 (2011), 6294–6299.
  • Reader Comments
  • © 2021 the Author(s), licensee AIMS Press. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0)
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Metrics

Article views(4122) PDF downloads(378) Cited by(13)

Figures and Tables

Figures(4)

/

DownLoad:  Full-Size Img  PowerPoint
Return
Return

Catalog