Abstract
In a Hilbert space, we consider a class of nonlinear fractional equations having the Caputo fractional derivative of the time variable t and the space fractional function of the self-adjoint positive unbounded operator. We consider various cases of global Lipschitz and local Lipschitz source with time-singular coefficient. These sources are generalized of the well–known fractional equations such as the fractional Cahn–Allen equation, the fractional Burger equation, the fractional Cahn–Hilliard equation, the fractional Kuramoto–Sivashinsky equation, etc. Under suitable assumptions, we investigate the existence, uniqueness of maximal solution, and stability of solution of the problems with respect to perturbed fractional orders. We also establish some global existence and prove that the global solution can be approximated by known asymptotic functions as \(t\rightarrow \infty \).
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
1.1 Statement of the problem
Let H be a Hilbert space, \(A : D(A) \subset H \rightarrow H\) be a self-adjoint positive definite unbounded operator, and \(f: [0, +\infty ) \times D(A^s) \rightarrow H\) with \(s\ge 0\). For \(\alpha \in (0, 1]\), \(\beta >0\), we consider the problem to find a function \(u: [0,T]\rightarrow H\) satisfying
where \( D_t^\alpha \) is the Caputo fractional derivative
and the fractional power of the operator \(A^\beta \) will be defined later. The equation (1.1) is a general form of a lot of well-known equations such as the Ginzburg–Landau equation (\( \alpha =1, A=-\varDelta , f(t,u)=au-bu^3\)), Burger equation (\( \alpha =1, A=-\varDelta , f(t,u)=u u_x\)), and Kuramoto–Sivashinsky equation (\( \alpha =1, A=\varDelta ^2, f(t,u)=\nabla ^2 u + (1/2)\Vert \nabla u \Vert ^2)\)). In the present paper, we will investigate the stability of solution of the initial value and the final value problems for (1.1).
The equation (1.1) subject to the initial data
is called the fractional initial value (or the Cauchy, the forward) problem (FIVP).
By the definition of the spectral resolution of the operator A and the Laplace transform, we can rewrite the FIVP as
where \( { E}_{\alpha , \beta }(z, t, \tau ) = (t-\tau )^{\alpha -1} E_{\alpha , \alpha }(-z^\beta (t-\tau )^\alpha )\) is expressed by the Mittag-Leffler function, and the operators \(E_\alpha \big (- t^\alpha A^\beta \big ), { E}_{\alpha , \beta }(A, t, \tau )\) will be defined in Sect. 2.
A function u that satisfies Eq. (1.3) is called a mild solution of the FIVP and denote by \(u=u_{\zeta ,\alpha ,\beta }\).
1.2 History and motivation
The abstract parabolic equation \(u_t+Au=f\) was considered for the last thirty years in many works on this area. The readers can see the classical book by Cazenave and Haraux [5] and references therein. The FIVP was also studied by lot of researchers. Xing et al. [24] discussed the existence, uniqueness, analyticity and the long-time asymptotic behavior of solutions of space-time fractional reaction–diffusion equations in \(\mathbb {R}^n\)
subject to the initial condition \(u(x, 0) =a(x)\). Existence and uniqueness of the maximal solutions of some linear and nonlinear fractional problems were investigated in [2, 6, 10, 12, 21]. The blow–up and global solution of time–fractional nonlinear diffusion–reaction equations were studied recently with some kind of nonlinear sources as \(f(t,u)=au+u^p\) (see Cao et al. [4]), \(f(t,u)=|u|^{p-1}u\) (see [27]), \(f(t,u)=|u|^p\) (Zhang [26]), Asogwa et al. [3] studied finite time blow up results for a version of equation (1.1).
The first motivation of our paper is to study the existence and uniqueness of maximal solution of the initial value problem with respect to a singular source. In many practical situation, the source is often assumed to satisfy \( {\left| \!\left| f(t,u)-f(t,v)\right| \!\right| } \le K(t, M)\Vert u-v\Vert \ \mathrm{for \ } \Vert u\Vert ,\Vert v\Vert \le M, \) where \(K: (0,\infty )\times (0,\infty )\rightarrow \mathbb {R}\) is the Lipschitz coefficient. If K(t, M) is dependent (independent) on t, we say that the coefficient is t-dependent (t-independent). Generally, in a lot of papers, the coefficient is assumed to be t-independent. A time-dependent coefficient which can be unbounded in a time interval (0, T), i.e. \(\sup _{0<t<T}K(t,M)=\infty \), is rarely studied. In the present paper, we consider a generalized form of singular source
with \(K(t,M)=\kappa _0(t)L(M)\) for \(\lim _{t\rightarrow 0^+}\kappa _0(t)=\infty \). In the special case \(\kappa _0(t)=t^{-\nu }\kappa (t)\), the condition of K(t, M) is similar to the generalized Nagumo condition (see, e.g., [13], Ch. 7 or [11]). In particular, the time-dependent source promises to bring many interesting global properties to the solution of the problem.
The second motivation of our paper is of studying the continuity of solution with respect to the fractional orders \(\alpha , \beta \) and the initial data \(\zeta \). In the papers mentioned above, the parameters \(\alpha ,\beta \) are assumed to be perfectly known. But in the real word of applications, the fractional orders can only be approximated from the mathematical model or statistical methods. In [1, 7], the Caputo derivatives can be identified approximately from observation data \(u(x_0,t)\) with \(t>0\), or u(x, T) with \(x\in \Omega \subset \mathbb {R}^n\). Besides, Kateregga [15] used statistical methods as the quantiles, logarithmic moments method, maximum likelihood, and the empirical characteristic function method to identify the parameters of the Lévy process. In these examples, the fractional orders are obtained only as approximate values. Hence, a natural question is that whether the solutions of fractional equations is continuous with respect to the perturbed orders. The papers devoted to these questions are still rare. We can list here some papers. Li and Yamamoto [16] investigated the solution \(u_{\gamma , D}\) of the problem
subject to the Neumann condition \(u_x(0, t) =u_x(1, t)=0\) and the initial condition \(u(x, 0)=f(x)\). They proved that
Dang et al. [8] studied the continuity of solutions of some linear fractional PDEs with perturbed orders. In the references [9, 19, 22], the authors considered the stability of solution of some class of nonlinear space–fractional diffusion problems taking into account the disturbance of parameters. To the best of our knowledge, until now, there is very few papers devoted to the stability of solutions of time-fractional problems with respect to the fractional time and space derivative parameters: \(\alpha , \beta \). In [20, 23], the stability was considered for the global Lipschitz source. The problem with the local time-singular Lipschitz source is still a topic of investigation. This stability result will serve as the foundation for numerical computation schemes for equations with imprecise fractional derivative parameters.
1.3 Outline of the paper
Summarizing the discussion of the FIVP, in the present paper, we will:
-
Investigate the existence and uniqueness of the maximal solution of the nonlinear FIVP with respect to the singular nonlinear source on the maximal interval \([0,T_{\zeta ,\alpha ,\beta })\). To solve the problem, we have to establish an appropriate Gronwall-type inequality which also has a specific merit in investigating other fractional problems.
-
Study the stability of the nonlinear FIVP with respect to the perturbed orders \(\alpha ,\beta \). In fact, we establish
$$\begin{aligned} u_{\zeta ',\alpha ',\beta '}\rightarrow u_{\zeta ,\alpha ,\beta }~~\ \text {as}\ ~~ (\zeta ',\alpha ',\beta ')\rightarrow (\zeta ,\alpha ,\beta ) \end{aligned}$$in an appropriate norm. Especially, for \(\alpha \rightarrow 1^-\), we will prove that the solution of the nonlinear FIVP tends to that of classical nonlinear initial value parabolic problem.
-
Study the existence of global solution on \([0,\infty )\) and prove decay estimates for the global solution. To illustrate for our results, we present an asymptotic result. Under some conditions, we will prove that
$$\begin{aligned} \lim _{t\rightarrow \infty } (1+t)^{\alpha }\Vert u_{\zeta ,\alpha ,\beta }(t)\Vert _s=\frac{1}{\varGamma (1-\alpha )}\Vert \zeta \Vert _{s-\beta } \end{aligned}$$for every \(\zeta \in D(A^s), \zeta \not =0\). The result shows that the decay of the FIVP is of polynomial order.
The rest of the paper is organized as follows. Section 2 gives the main results of our paper without proofs. In Sect. 3 the proofs of these results are presented.
2 Main results
2.1 Notations
To state our problem precisely, we will give some definitions. Firstly, in the paper, we always denote by \(C,C'\) generic constants which could be different from line to line. We denote the inner product in Hilbert space H by \(\langle .,.\rangle \) and the associated norm by \(\Vert .\Vert \).
We recall (see, e.g., [17, page 61, Ch.4]) that a resolution of the identity on a Hilbert space H is a one-parameter family {\(S_\lambda : \lambda \in R\)} of orthogonal projections on H such that
-
(i)
\(S_\lambda \le S_{\lambda '}\) if \(\lambda \le \lambda '\) (monotonicity),
-
(ii)
\(\lim _{\lambda '\rightarrow \lambda ^+}S_{\lambda '}\zeta = S_\lambda \zeta \) for \(\zeta \in H\) (strong right continuity),
-
(iii)
\(\lim _{\lambda \rightarrow -\infty }S_\lambda \zeta = 0\) and \(\lim _{\lambda \rightarrow +\infty }S_\lambda \zeta = \zeta \) for \(\zeta \in H\).
Assume that \(\theta >0\) is the lower bound of the spectrum of the operator A. Let us denote by \(\{{S_\lambda }\}\) the spectral resolution of the identity associated to operator A such that \(A=\int _\theta ^\infty \lambda dS_\lambda \). We follow [25, page 29] (see also [17, page 92]) to define the power of the self-adjoint positive definite unbounded operator as
Generally, for a continuous function \(h: \mathbb {R} \rightarrow \mathbb {R}\), we denote the domain of h(A) to be
If \(w \in D(h(A))\), we define the linear operator
Particularly, if \(h(z)=z^s\) for \(z \ge 0, s\in \mathbb {R}\), we have the Hilbert space \(D(A^s)\) with the norm \( {\left| \!\left| w\right| \!\right| }_s = \left( \int _\theta ^{+\infty } \lambda ^{2 s} \mathrm {d} {\left| \!\left| {S_\lambda }w\right| \!\right| }^2 \right) ^{1/2}\). For \(v\in C([0,T]; D(A^s))\) we denote \(|v|_{s,t}=\sup _{0 \le \tau \le t} \Vert v(\tau )\Vert _s\). Let \(0 \le s_* \le s^*\) and \(s_1, s_2 \in [s_*, s^*]\), \(s_2\le s_1\). It is easy to see that
For \({ M}, s>0\), we put
In this section, we also remind the Mittag–Leffler function and its properties which play important roles in the proof of main results of current paper. We recall also the Gamma and Beta functions,
The Mittag–Leffler function with two parameters is defined as
Definition 1
A function \(u\in C([0,T);D(A^s)\) is a maximal solution of Problem \(P_{\zeta ,\alpha ,\beta }\) if u satisfies \(P_{\zeta ,\alpha ,\beta }\) on the interval [0, T) such that \(T=\infty \) or that \(T<\infty \), \(\limsup _{t\rightarrow T^-}\Vert u(t)\Vert _s=\infty \). In the case \(T=\infty \), we say that u is aglobal solution of \(P_{\zeta ,\alpha ,\beta }\). The global solution u is said to have
-
the sub-polynomial decay rate if there are \(\rho \), \(C>0\) such that
$$\begin{aligned} \Vert u(t)\Vert _s\le C(1+t)^{-\rho }~~\text {for every}~t>0. \end{aligned}$$ -
the asymptotically polynomial decay if there is \(\rho , C_\rho >0\) such that
$$\begin{aligned} \lim _{t\rightarrow \infty }(1+t)^{\rho } \Vert u(t)\Vert _s=C_\rho . \end{aligned}$$
2.2 The global Lipschitz source
Using the notations defined, we can state precisely the assumption for the singular source. In fact, we consider the source function of the problem satisfying the following assumptions.
Assumption F1(\(\alpha \)) Let \(T>0\), \(\alpha \in (0,1]\) and \(f\in C( (0,\infty ) \times D(A^s); H)\). We assume that
Assumption G1 Let \(T>0\), \(s >0\), \(\nu \le \alpha /2\) and \(f\in C( (0,\infty ) \times D(A^s); H)\), \(\kappa \in C([0,\infty );[0,\infty ))\). We assume that
Remark 1
Assumption F1(\(\alpha \)) holds in many cases. For example, as shown in Lemma 2 (see the part of proofs), if \(\Vert f(t,0)\Vert \le t^{-\nu _f}\kappa _f(t)\) for \( 2\nu _f<\alpha , \kappa _f\in C([0,\infty );\mathbb {R})\), then Assumption F1(\(\alpha \)) holds.
Using the assumptions, we can obtain the following existence result.
Theorem 1
Let \(\alpha \in (0,1), \beta >0\), \(s \in [0, \beta /2]\), \(\zeta \in D(A^s)\) and let \(f\in C( (0,\infty )\times D(A^s); H)\). Assume that Assumption F1(\(\alpha \)) and Assumption G1 hold. For \(\nu <\alpha /2\), the equation (1.3) has a unique solution \(u=u_{\zeta ,\alpha ,\beta }\in C([0,\infty );D(A^s))\). Moreover, for a \(T>0\), if we put
then \(g \in C\left( [0, T], D(A^s)\right) \) and there is a constant C independent of t such that
where we recall \(|g|_{s,t}=\sup _{0\le \tau \le t}\Vert g(\tau )\Vert _s\).
For \(\nu = \alpha /2\), \(\kappa _{T,s} < \theta ^{\beta /2-s} \left( \varGamma (1-\alpha )\right) ^{-1/2}\), the equation (1.3) has a unique solution \(u\in C([0,T];D(A^s))\).
Remark 2
We can use the Edelstein fixed point theorem (see, e.g., [13], Ch. 7) to obtain the desired result for the case \(\nu =\alpha /2, \kappa _{T,s}=\theta ^{\beta /2-s} \left( \varGamma (1-\alpha )\right) ^{-1/2}\). We note that if we put \(u_{n+1}=\mathcal {F}(u_n)\), where \(\mathcal {F}(u)\) is the right hand side of equation (1.3), then the sequence \((u_n)\) converges to the solution u in C([0, T], \(D(A^s))\).
From the existence result stated, we can obtain an interesting global existence result. Moreover, we also give an asymptotically polynomial result for decay estimates.
Theorem 2
Let \(\alpha \in (0,1), \beta >0\), \(s \in [0, \beta /2]\), \(\zeta \in D(A^s)\), \(\nu ,\nu _f>0\), \( \nu <\alpha /2\), \( \nu _f<\alpha /2\) and let \(f\in C( (0,\infty )\times D(A^s); H)\). Assume :
-
Assumption G1 holds and there are \(\kappa _1>0\), \(\ell \in \mathbb {R}\) such that
$$\begin{aligned} 0\le \kappa (t)\le \kappa _1 (1+t)^{\ell }, \end{aligned}$$ -
There exist \(\kappa _f>0,\nu _f>0\), \(\ell _f\in \mathbb {R}\) such that
$$\begin{aligned} \Vert f(t,0)\Vert _s\le \kappa _f t^{-\nu _f}(1+t)^{\ell _f}. \end{aligned}$$
With these assumptions, we have the following two results :
-
1.
If \(\ell ,\ell _f<\nu \), \(\alpha >\frac{1}{2}-\nu \), then, for every \(\zeta \in D(A^s)\), the problem \(P_{\zeta ,\alpha ,\beta }\) has a unique global solution \(u_{\zeta ,\alpha ,\beta }\in C([0,\infty );D(A^s))\) which has the sub-polynomial decay rate. More explicitly, for every \(\omega \) satisfying \(0<\omega <\min \left\{ \frac{1}{2},\frac{1}{2}-\ell ,\frac{1}{2}-\ell _f\right\} \), we can find \(C_\omega >0\) such that
$$\begin{aligned} \Vert u_{\zeta ,\alpha ,\beta }(t)\Vert _s\le C(1+t)^{-\min \{\alpha ,\omega \}}~~~\text {for}\, ~t>0. \end{aligned}$$ -
2.
If \(\ell ,\ell _f< \frac{1}{2}-\alpha \), \(0<\alpha <1/2\) then, for every \(\zeta \in D(A^s)\), the problem \(P_{\zeta ,\alpha ,\beta }\) has a unique global solution \(u_{\zeta ,\alpha ,\beta }\in C([0,\infty );D(A^s))\) which has the sub-polynomial decay rate. More explicitly, for every \(\alpha<\omega <\min \left\{ \frac{1}{2}, \frac{1}{2}-\ell ,\frac{1}{2}-\ell _f\right\} \) we can find \(C_{\alpha ,\omega }>0\) such that
$$\begin{aligned} \Vert u_{\zeta ,\alpha ,\beta }(t)-E_{\alpha }(-t^\alpha A^\beta )\zeta \Vert _s\le C_{\alpha ,\omega }\Vert \zeta \Vert _s(1+t)^{-\omega }. \end{aligned}$$In addition, the solution \(u_{\zeta ,\alpha ,\beta }\) has the asymptotically polynomial decay rate
$$\begin{aligned} \lim _{t\rightarrow \infty } (1+t)^{\alpha }\Vert u_{\zeta ,\alpha ,\beta }(t)\Vert _s=\frac{1}{\varGamma (1-\alpha )}\Vert \zeta \Vert _{s-\beta }. \end{aligned}$$
Remark 3
As mentioned, the set \(D(A^s)\) is dense in H. Hence the polynomial decay rate of the solution holds for almost every \(\zeta \in H\).
In the next two theorems, we state some results on stability of solution of the initial problem with respect to the initial data and the fractional orders. In the following stability results, we recall that \(u_{\zeta , \alpha ,\beta }\) is the solution of Problem \(P_{\zeta ,\alpha ,\beta }\) in (1.3) corresponding to the initial data \(\zeta \) and the orders \(\alpha ,\beta \). We first have the classical stability with respect to the fixed orders \(\alpha ,\beta \). More precisely, for \(\alpha _*,\alpha ^*,\beta _*,\beta ^*\) satisfying \(0<\alpha _*<\alpha ^*<2\alpha _*\le 2\), and \(0<\beta _*<\beta ^*\), we consider
Theorem 3
Let \(\zeta , \xi \in D(A^s)\) be two initial data with \(s \in [0, \beta /2]\), and let \(T>0\), \( ({\alpha }, {\beta })\in \varDelta \) be as in (2.4). Let the source function f satisfy Assumption F1(\(\alpha \)) and Assumption G1. Then the problems \(P_{\xi ,\alpha ,\beta }, P_{\zeta ,\alpha ,\beta } \) have the unique solutions \( u_{\xi , {\alpha }, {\beta }}, u_{\zeta , {\alpha }, {\beta }} \in C([0,T],D(A^s)) \), respectively. Moreover, there exists a positive constant \(P_1\) independent of \(\zeta ,\xi \) such that
Hence, letting \(\zeta ,\zeta _k\in D(A^s)\), \(\zeta _k\rightarrow \zeta \) in \(D(A^s)\), we obtain
Moreover, we also have the stability results with respect to the perturbed orders \(\alpha ,\beta \). To investigate the stability of the solution of the problem (FIVP), we will restrict the value \((\alpha , \beta )\) in the bounded domain.
Theorem 4
Let \((\alpha , \beta ), (\alpha _k, \beta _k)\in \varDelta \) (defined in (2.4)) such that \((\alpha _k, \beta _k) \rightarrow (\alpha , \beta )\), and let \(\zeta , \, \zeta _k \in D(A^{s})\) such that \(\zeta _k \rightarrow \zeta \) in \(D(A^{s})\) as \(k\rightarrow \infty \). Let the source function f satisfy Assumption F1(\(\alpha _*\)) and Assumption G1. Then the problems \(P_{\zeta ,\alpha ,\beta }, P_{\zeta _k,\alpha _k,\beta _k} \) have the unique solutions
respectively. In addition, the following results hold :
-
(i)
If \(s \in [0, \beta /2)\), then
$$\begin{aligned} \lim _{k\rightarrow \infty } |u_{\zeta _k, \alpha _k, \beta _k} - u_{\zeta , \alpha , \beta }|_{s, T} = 0. \end{aligned}$$(2.5) -
(ii)
If \(s = \beta /2\) and \(\beta _k\ge \beta \) for \(k\rightarrow \infty \), then
$$\begin{aligned} \lim _{k\rightarrow \infty } |u_{\zeta _k, \alpha _k, \beta _k} - u_{\zeta , \alpha , \beta }|_{s, T} = 0. \end{aligned}$$(2.6) -
(iii)
If \(\beta _k \le \beta \) as \(k \rightarrow \infty \), then
$$\begin{aligned} \lim _{k\rightarrow \infty } |u_{\zeta _k, \alpha _k, \beta _k} - u_{\zeta , \alpha , \beta }|_{\beta _k/2, T} = 0. \end{aligned}$$(2.7) -
(iv)
If we suppose further that \(\zeta _k, \zeta \in D\left( A^{\beta ^*/2+r_1}\right) \) such that \(\zeta _k \rightarrow \zeta \) in \(D\left( A^{\beta ^*/2+r_1}\right) \) for some \(r_1>0\). We also suppose that \(s \!\in \! \left[ \frac{\beta _*}{2}, \frac{\beta }{2}\right) \), \(f(., u_{\zeta ,\alpha ,\beta }(.))\) \(\in C([0, T], D\left( A^{r_2}\right) )\) for some \(r_2>0\). Then, there exists constants \(L_0, L_1\) independent of \(\zeta , \zeta _k\) such that
$$\begin{aligned} {\left| u_{\zeta _k, \alpha _k, \beta _k} - u_{\zeta , \alpha , \beta }\right| }_{s, T} \le L_0\Vert \zeta - \zeta _k\Vert _s + L_1 (|\alpha - \alpha _k| +|\beta - \beta _k|) ^{\frac{\gamma _2}{2(\gamma _1+\gamma _2+2)}}, \end{aligned}$$(2.8)where \(L_0, L_1\) is depended on \((\alpha _*,\alpha ^*, \beta _*, \beta ^*, T)\), \(\gamma _1 = 2\beta ^*\), and \(\gamma _2=\min \{\beta ^*+2(r_1-s), 2r_2\}\).
Remark 4
Theorem 4 shows that if \(\alpha \rightarrow 1^-, \beta \rightarrow 1\), then the solutions of the fractional equation (1.1)–(1.2) tend to the solution of the classical equation
2.3 The local Lipschitz source
Assumption G2 Let \(s \in [0, \beta /2]\), \(\nu \le \alpha /2\), \(\kappa \in C([0,\infty );\mathbb {R})\), \(\kappa (t)\ge 0\) for \(t\ge 0\), and \(f\in C( (0,\infty ) \times D(A^s); H)\). For every \(T, { M}>0\), we assume that there is an \(L_T({ M})\) such that
Put \(\Xi =(0, 1)\), \(H=L^2(\Xi )\), we can directly check that some common sources of the following equations satisfy: the Ginzburg Landau equation and the Burger equation for \(\nu =0, s=1\), the Cahn–Hilliard and Kuramoto–Sivashinsky equations for \(\nu =0, s=2\).
In this section, we investigate the existence and uniqueness of the solution of the problem with local source defined in (2.2). In addition, we study the dependence of the solution with respect to the fractional order \(\alpha , \beta \) and the initial data \(\zeta \). To emphasize the dependence of the solution u on these given data, let us write it by \(u_{\zeta , \alpha , \beta }\). We have the following theorem.
Theorem 5
Let \(\alpha \in (0, 1)\), \(\beta \in (0, +\infty )\), \(s\in [0, \beta /2]\), \(\nu <\alpha /2\), and let \(\zeta \) be the initial data defined in (1.2) such that \(\zeta \in D(A^{\beta /2})\). Let the source function f satisfy Assumption F1(\(\alpha \)) and Assumption G2.
Then, for any \({ M}> 2 {\left| \!\left| \zeta \right| \!\right| }_{\beta /2}\), we have :
-
(i)
(Local existence) There exists a \(T_{ M}>0\) such that the FIVP has a unique mild solution \(u_{\zeta , \alpha , \beta }\) which belongs to \(C([0, T_{ M}]; D(A^s))\).
-
(ii)
(Uniqueness) If \(V,W\in C([0,T];D(A^s))\) are solutions of (1.3) on [0, T], then \(V=W\).
-
(iii)
(Maximal existence) Let
$$\begin{aligned} T_{\zeta , \alpha , \beta }=\sup \{T>0:\ 1.3 \quad \text {has a unique solution on}\ [0,T]\}. \end{aligned}$$Then the equation (1.3) has a unique solution \(u_{\zeta ,\alpha ,\beta }\in C([0,T_{\zeta ,\alpha ,\beta });D(A^s))\). Moreover, we have either \(T_{\zeta , \alpha , \beta } = +\infty \) or \(T_{\zeta , \alpha , \beta }<+\infty \) and \(\Vert u_{\zeta , \alpha , \beta }(t)\Vert _s \rightarrow \infty \) as \(t \rightarrow T_{\zeta , \alpha , \beta }^-\). Besides, if \(u_{\zeta , \alpha , \beta } \in B_{s, T}({ M})\) then
$$\begin{aligned} \Vert u_{\zeta , \alpha , \beta }(t) \Vert _s^2 \le 2 \varGamma (1-2\nu ) | g|_{s, t}^2 E_{\alpha -2\nu ,1-2\nu } \left( 2\theta ^{2s-\beta } L_T^2({ M}) t^{\alpha -2\nu }\right) , \end{aligned}$$for any \(t\in [0,T]\).
-
(iv)
Let \(\zeta ,\zeta _k\in D(A^s), \alpha ,\alpha _k\in [\alpha _*,\alpha ^*], \beta ,\beta _k\in [\beta _*,\beta ^*]\) satisfy
$$\begin{aligned} \zeta _k\rightarrow \zeta ~\, \mathrm{in}\, ~D(A^s),~~\alpha _k\rightarrow \alpha ,~~\beta _k\rightarrow \beta ~~~~{as} \ ~k\rightarrow \infty . \end{aligned}$$
Assume in addition that Assumption F1(\(\alpha _*\)) hold and \( s=\beta /2\), \(\beta _k\ge \beta \) as \(k\rightarrow \infty \) or \(0\le s<\beta /2\). Then for every \(T\in (0,T_{\zeta ,\alpha ,\beta })\) we can find a \(k_0>0\) such that \(T_{\zeta _k,\alpha _k,\beta _k}>T\) for every \(k>k_0\) and
Moreover, we have
If \(T_{\zeta ,\alpha ,\beta }=\infty \) then \( \lim _{k\rightarrow \infty }T_{\zeta _k,\alpha _k,\beta _k}=\infty \).
Using Theorem 5, we can obtain global existence and polynomial decay results. To state precisely the theorem, we state the following
Assumption G3 For \(\nu >0\), \(f: C((0,\infty ) \times D(A^s) ; H)\), \(\kappa \in C([0,\infty );[0,\infty ))\) we assume that there is an \(L_\infty ({ M})>0\) for every \( { M}>0\) such that
and there are constants \(\kappa _1,\eta >0,\ell \in \mathbb {R}\) such that
Theorem 6
Let \((\alpha ,\beta )\in \varDelta \) be as in (2.4), let \(0\le 2\nu \le \alpha <1\), \(\eta ,\omega >0\), \(s \in [0, \beta /2]\), \(\ell \in \mathbb {R}\) and suppose that Assumption G3 hold. Assume that
-
(i)
\(\eta >\max \left\{ 2(\ell -\nu ),0\right\} \) and \( \max \left\{ \frac{\ell -\nu }{\eta },0\right\}<\omega <1/2,\)
-
(ii)
\( m^2_{\infty ,\alpha ,\omega }:= \sup _{t \ge 0}(1+t)^{2\omega } \int _0^t (t-\tau )^{\alpha -1}E_{\alpha ,\alpha }(-\theta ^\beta (t-\tau )^\alpha ) \Vert f(\tau ,0)\Vert ^2 \mathrm {d}\tau \) \( <\infty .\)
Then there exists \(\delta _0>0\) such that Problem \(P_{\zeta ,\alpha ,\beta }\) has a unique solution \(u_{\zeta ,\alpha ,\beta }\in C([0,\infty );D(A^s))\) which has the optimal decay rate for
Moreover, if \(\frac{\ell -\nu }{\eta }<\alpha <1/2\), then
Remark 5
From the condition (i), the global result holds for \(\nu<\ell <1/2\). In this case the Lipschitz coefficient can be unbounded as \(t\rightarrow \infty \) since \(\lim _{t\rightarrow \infty }\kappa _1 t^{-\nu }(1+t)^\ell =\infty \).
3 Proofs
3.1 Preliminary lemmas
Lemma 1
[see [18]] Letting \(\lambda>0, p>0\) and \(k\in \mathbb {N}\), we have
Lemma 2
Let \(0<p_*<p^*<2\) such that \(p^*<2p_*\), and \(r_*>0\). Then for any \(p, p_0 \in [p_*, p^*]\), and \(r, r_0 \ge r_*\), and \(\lambda \ge 0\), we have :
-
(a)
There exists a constant \(C=C(p_*, p^*, r_*)>0\) such that
$$\begin{aligned} {\left| E_{p, r}(-\lambda )\right| } + {\left| \frac{\partial E_{p, r}}{\partial p}(-\lambda )\right| } + {\left| \frac{\partial E_{p, r}}{\partial r}(-\lambda )\right| } \le \frac{C}{1+\lambda }. \end{aligned}$$We also have
$$\begin{aligned} 0\le E_\alpha (-z)\le 1,\ \ 0\le E_{\alpha ,\alpha }(-z)\le \frac{1}{\varGamma (\alpha )}\ \ \ \text {for}\ \ z\ge 0. \end{aligned}$$(3.1) -
(b)
Let \(0<p_*<p^*<1\). There exist two constants \(C_1, C_2\) which depend only on \(p_*, p^*\) such that
$$\begin{aligned} \frac{1}{\varGamma (1-p)} \frac{C_1}{1+\lambda } \le E_{p}(- \lambda ) \le \frac{1}{\varGamma (1-p)} \frac{C_2}{1+\lambda } . \end{aligned}$$Moreover, we have \(\lim _{\lambda \rightarrow +\infty }\lambda E_p(-\lambda )=\frac{1}{\varGamma (1-p)}\).
-
(c)
There exists a constant \(C=C(p_*, p^*)\) such that
$$\begin{aligned} \left| E_{p}(-\lambda ^r t^p) - E_{p_0}(-\lambda ^{r_0} t^{p_0}) \right| \le C \lambda ^{r^*} (1+\ln \lambda ) \left( |p - p_0| +|r - r_0 | \right) , \, \, \forall \, \lambda \ge 1. \end{aligned}$$ -
(d)
We denote
$$\begin{aligned} { E}_{a, b}(\lambda , t, \tau ) = (t-\tau )^{a-1} E_{a, a}(-\lambda ^b (t-\tau )^a). \end{aligned}$$Then, there exists a constant \(C=C(p_*, p^*, r_*)\) such that
$$\begin{aligned} \int _0^t \left| { E}_{p, r}(\lambda , t, \tau ) - { E}_{p_0, r_0}(\lambda , t, \tau ) \right| \mathrm {d}\tau \le C \left( (1+\lambda ^r)|p- p_0| +|\lambda ^r - \lambda ^{r_0} | \right) . \end{aligned}$$ -
(e)
Put
$$\begin{aligned} I_{\alpha ,\nu ,\omega }(t)=\int _{0}^{t} (t - \tau )^{\alpha - 1}E_{\alpha ,\alpha }(-\theta ^\beta (t-\tau )^\alpha ) \frac{\tau ^{-2\nu }}{(1+\tau )^{2\omega }} \mathrm {d}\tau . \end{aligned}$$For \(t\ge 1\), \(\nu +\omega '<1/2\), \(0<\omega '\le \omega \), there is a constant \(D_{\omega ,\omega '}>0\) such that
$$\begin{aligned} 0\le I_{\alpha ,\nu ,\omega }(t)\le D_{\omega ,\omega '}t^{-2\nu -2\omega '}. \end{aligned}$$
Proof
We only prove (3.1) and (b), (e). The readers can see the proof of other cases in [8]. From the complete monotonicity of the Mittag-Leffler function \(E_{\alpha }(-z)\) for \(z\ge 0\) (see [14], Ch. 3) we have \((-1)^n\frac{d^n}{dz^n}E_\alpha (-z)\ge 0\) for \(z\ge 0\). Hence we have \(E_\alpha (-z), E_{\alpha ,\alpha }(-z)\) is decreasing which give \(0\le E_{\alpha }(-z)\le 1\), \(0\le E_{\alpha ,\alpha }(-z)\le \frac{1}{\varGamma (\alpha )}\) for \(z\ge 0\).
We prove (b). Using the asymptotic expansion in [14, page 19, Ch. 3] we have \(\lambda E_p(-\lambda )=\frac{1}{\varGamma (1-p)}+O(\lambda ^{-1})\). Hence \(\lim _{\lambda \rightarrow +\infty }\lambda E_p(-\lambda )=\frac{1}{\varGamma (1-p)}\).
We prove (e) next. In fact, noting that \(\sup _{\tau \ge 0}\frac{\tau ^{2\omega '}}{(1+\tau )^{2\omega }}\le 1\), we have
where
Using Lemma 2 (a) and estimating directly \(J_1\) gives
Similarly, by Lemma 1, we have
From the estimation of \(J_1, J_2\) we complete the proof of Part (e). \(\square \)
In this paper, we also need the following useful inequality.
Lemma 3
Let \(\alpha , q\in \mathbb {R}\), \(0<\alpha \le 1\), \(q<\alpha \), and let \(v, g\in C[0, T]\). Then the equation
has a unique solution \(u\in C[0,T]\) which satisfies
for \(t \in [0,T]\). As a consequence, if \(w\in C[0,T]\) satisfies
and if \(g(t)\ge 0\) for \(t\in [0,T]\), then
where \(C=\max _{k \ge 1} d_k\) with \(d_1=\varGamma ((\alpha -q)+1-q)/\varGamma (\alpha +1-q)\) and \(d_k=\varGamma (k(\alpha -q)+1-q)/\varGamma (k(\alpha -q)+1)\).
Remark 6
We note that \(\varGamma (a) \le \varGamma (b)\) for any \(2 \le a \le b\), therefore, \(\varGamma (k(\alpha -q)+1-q)/\varGamma (k(\alpha -q)+1) \le 1\) for k large enough or \(d_{k+1} \le d_k\). This implies that \(C=\max _{k \ge 1} d_k<+\infty \).
Proof of Lemma 3
Put
Using the similar technique as in Theorem 1, we can prove that there exists \(k_0 \in \mathbb {N}\) such that \(S^{k_0}\) is contraction in C[0, T]. Consequently, there exists a unique \(u \in C[0, T]\) such that \(u=Su\).
We put \(u_0=0, u_{n+1}=Su_{n}\). The function can be represented by the series \(u=\sum _{n=0}^\infty (u_{n+1}-u_{n})\) . The Weierstrass theorem shows that the series converges in C[0, T] and
Now, we prove the final inequality. Put \(w_0=Sw\), \(w_{n+1}=Sw_n\). Since \(g(t)\ge 0\) for \(t\in [0,T]\), we have \(Sw_1(t)\le Sw_2(t)\) for \(w_1(t)\le w_2(t)\), \(t\in [0,T]\). We note that \(w\le w_0\), hence, by induction we obtain \(w_{n}\le w_{n+1}\). Using the contraction principle we obtain \(\lim _{n\rightarrow \infty }\Vert w_{n}-u\Vert _{C[0,T]}=0\). Since \(w_n\le w_{n+1}\) for every \(n=0,1,\ldots \), we obtain \(w(t)\le w_0(t)\le u(t)\) for \(t\in [0,T]\). From (3.2) we obtain the desired inequality. \(\square \)
We also need the following results.
Lemma 4
Let \(T, \theta>0, \alpha \in (0, 1], \beta >0\), \(s \in [0, \beta /2]\), \(r \ge 0\), \(t\in (0,T]\), \(w\in C([0,T]; D(A^r))\).
(i) For \(\zeta \in D(A^{s+r})\), we have \( E_\alpha (-t^\alpha A^\beta )\zeta \in D(A^{s+r})\) and \(\Vert E_\alpha (-t^\alpha A^\beta )\zeta \Vert _{s+r}\) \(\le \Vert \zeta \Vert _{s+r}\), and
For every \(0\le \omega \le \alpha /2\), we also have
and
(ii) Put
where
If \(w\not \equiv 0\) on [0, t], then
where
Proof
We first prove (i). One has
We next prove (3.3). In fact, we have
Using Lemma 2 we have \(\lim _{t\rightarrow \infty } (1+t)^{\alpha }E_\alpha (-t^\alpha \lambda ^\beta ) =\frac{1}{\lambda ^\beta \varGamma (1-\alpha )}\) and
Hence, applying the Lebesgue dominated convergence theorem yields
From (3.3), we deduce (3.4). Put \(\xi =\zeta /\Vert \zeta \Vert _s\) we obtain (3.5).
Now, we consider Part (ii). We have \(E_{\alpha ,\alpha }(z)\ge 0\) (see [14], Ch. 3). Hence, Lemma 1 yields
By the Hölder inequality, Lemma 2 and (2.1), we obtain for \(t\in (0,T]\)
Noting that \(E_{\alpha ,\alpha }(-\lambda ^\beta (t-\tau )^\alpha )\le E_{\alpha ,\alpha }(-\theta ^\beta (t-\tau )^\alpha )\) for \(\lambda \ge \theta \), we obtain
\(\square \)
3.2 Proof of Theorem 1
For \(w\in C([0,T],D(A^s))\), we put
Choosing \(r=0\) in Lemma 4 (ii) gives \(\sup _{\lambda \ge \theta }\lambda ^{2s-\beta }H_0(\lambda ,t_1,t_2)\le \theta ^{2s-\beta }\) and
So we have
where \(d_1=\varGamma ((\alpha -2\nu )+1-2\nu )/\varGamma (\alpha +1-2\nu )\). We consider the case \(\nu <\alpha /2\). For \(w_1,w_2\in C([0,T],D(A^s))\), using the similar technique as in [8], we can prove by induction that
where \(d_k=\varGamma (k(\alpha -2\nu )+1-2\nu )/\varGamma (k(\alpha -2\nu )+1) d_{k-1}\) with \(k \ge 2\) and \(C=\max _{k \ge 1} d_k\). This gives
Hence there is a \(k_0\in \mathbb {N}\) such that
which gives
i.e., \({ F}^{k_0}\) is a contraction in \(C([0,T],D(A^s))\). Hence, the exists a unique fixed point \(u\in C([0,T],D(A^s)) \) satisfying \(u={ F}^{k_0}(u)\). We deduce that \({ F}u={ F}^{k_0}({ F}u)\), i.e., \({ F}u\) is also a fixed point of the operator \({ F}^{k_0}\). Hence \(u={ F}u\).
We give the estimate of u. In fact, from (3.10) we obtain
Hence
Using (3.2) of Lemma 3, we obtain the inequality of the theorem.
Finally, we consider the case \(\nu =\alpha /2\). We can find a \(\xi \in (0,T]\) such that \( {\left| \!\left| { F}(w_1)(\xi )-{ F}(w_2)(\xi )\right| \!\right| }^2_s=\sup _{0\le t\le T} {\left| \!\left| { F}(w_1)(t)-{ F}(w_2)(t)\right| \!\right| }^2_s\). Lemma 4 gives
If \(\kappa _{T,s} < \theta ^{\beta /2-s} \left( \varGamma (1-\alpha )\right) ^{-1/2}\) then \({ F}\) is a contraction in \(C([0,T],D(A^s))\). Consequently, the problem (1.3) has a unique solution in \(C([0,T],D(A^s))\).
3.3 Proof of Theorem 2
In the proof we denote \(u=u_{\zeta ,\alpha ,\beta }\) for short. We verify that \(\sup _{t\ge 0}(1+t)^\omega \Vert u(t)\Vert _s<\infty \). Assume by contradiction that \(\sup _{t\ge 0}(1+t)^\omega \Vert u(t)\Vert _s=\infty \). For every \(\lambda >0\), we put
By the continuity of u, we have
We note that
As in the proof of Theorem 1, choosing \(r=0\) in Lemma 4 gives
and we have
For every \(\lambda >0\), denoting \(w(t)=(1+t)^\omega u(t)\) and using (3.13) yield
where \(0<t\le T_\lambda \) and
To prove the theorem, we will use two necessary inequalities. In fact, we can find two constants \(\gamma _1,\gamma _2>0\) such that
The proof of these inequalities will be postponed to the end of the proof of the theorem.
-
(i)
We first consider the case \(\ell ,\ell _f<\nu \), \(\alpha >\frac{1}{2}-\nu \). Using the inequalies (3.14), (3.15) yield
$$\begin{aligned} \Vert w(t)\Vert _s^2&\le 4\varLambda _\omega ^2\Vert \zeta \Vert _s^2+ Ct^{-2\gamma _1}+Ct^{-2\gamma _2}. \end{aligned}$$Choosing \(t=T_\lambda \), we obtain
$$\begin{aligned} \lambda ^2&\le 4\varLambda _\omega ^2\Vert \zeta \Vert _s^2+C'\lambda ^2 T_\lambda ^{-2\gamma _1}+C' T_\lambda ^{-2\gamma _2} \end{aligned}$$which implies
$$\begin{aligned} 4\lambda ^{-2} \varLambda _\omega ^2\Vert \zeta \Vert _s^2+C'T_\lambda ^{-2\gamma _1}+C' \lambda ^{-2}T_\lambda ^{-2\gamma _1}\ge 1. \end{aligned}$$(3.16)Noting that \(\lim _{\lambda \rightarrow \infty }T_\lambda =\infty \), we obtain in view of (3.16) that \(0\ge 1\) which is a contradiction.
-
(ii)
We consider the case \(\ell ,\ell _f\le \frac{1}{2}-\alpha \). For the upper bound of \(\Vert u(t)\Vert _s\) we can use the same argument as Part (i) with \(\omega >\alpha \). We verify the asymptotic value for \((1+t)^{\alpha }\Vert u(t)\Vert _s\). Using Lemma 2 yields
$$\begin{aligned} \lim _{t\rightarrow \infty }(1+t)^{2\alpha } {\left| \!\left| E_\alpha (-t^\alpha A^\beta )\zeta \right| \!\right| }^2_{s} =\frac{1}{\varGamma ^2(1-\alpha )}\Vert \zeta \Vert ^2_{s-\beta }. \end{aligned}$$Choose \(\omega =\alpha \), denoting \(w(t)=(1+t)^\omega u(t)\) and using (3.13) yield
$$\begin{aligned}&{\left| \!\left| w(t)-(1+t)^\omega E_\alpha \big (- t^\alpha A^\beta \big ) \zeta \right| \!\right| }_s\\&\le \left( 4\theta ^{2s-\beta } \int _0^t (t -\tau )^{\alpha -1} \tau ^{-2\nu } \kappa _1^2(1+\tau )^{2\ell -2\omega }E_{\alpha ,\alpha }(-\theta ^\beta (t-\tau )^\alpha ) {\left| \!\left| w(\tau )\right| \!\right| }_s^2 \mathrm {d}\tau \right. \\&\left. \quad +4 \theta ^{2s-\beta }(1+t)^{2\omega } \int _0^t (t -\tau )^{\alpha -1} E_{\alpha ,\alpha }(-\theta ^\beta (t-\tau )^\alpha ) \Vert f(\tau ,0)\Vert ^2 \mathrm {d}\tau \right) ^{1/2} \\&\le \sqrt{(1+t)^{2\omega }V_1^2+(1+t)^{2\omega }V_2^2}. \end{aligned}$$Combining the latter inequality with (3.15) yields
$$\begin{aligned} {\left| \!\left| w(t)-(1+t)^\omega E_\alpha \big (- t^\alpha A^\beta \big ) \zeta \right| \!\right| }_s\le \sqrt{Ct^{-2\gamma _1}+Ct^{-2\gamma _2}}. \end{aligned}$$Hence
$$\begin{aligned} \lim _{t\rightarrow \infty }\Vert w(t)\Vert _s=\lim _{t\rightarrow \infty }\Vert (1+t)^\alpha E_\alpha \big (- t^\alpha A^\beta \big ) \zeta \Vert _s = \frac{1}{\varGamma (1-\alpha )}\Vert \zeta \Vert _{s-\beta }. \end{aligned}$$Finally, we prove (3.15). Since the proof for the case \(\ell \le 0\) is different from the one of the case \(\ell >0\), we divide the proof into two cases.
In the case \(\ell \le 0\), since \(0<\omega <1/2,\nu >0\) we obtain \( \omega -\nu <\min \{1/2-\nu ,\omega -\ell \}.\) Hence we can choose \(\omega '\) such that
which gives \(\omega '<\omega -\ell \), \(\nu +\omega '<1/2\), \(\gamma _1:=-\omega +\nu +\omega '>0\). We can use (3.14) to obtain
where \(I_{\alpha ,\nu ,\omega }\) is defined in Lemma 2. Applying Lemma 2 (e) we can find a \(C>0\) such that
Next, we consider the case \(0\le \ell <\nu \) we have \(\omega +\ell -\nu <1/2-\nu \) and \(\omega +\ell -\nu <\omega \). Hence we can find \(\omega '\) such that \( \omega +\ell -\nu<\omega '<\min \{1/2-\nu ,\omega \}.\) It follows that \(\omega '+\nu <1/2, \omega '\le \omega \) and \(\gamma _2:=-\omega -\ell +\nu +\omega '>0\). Using Lemma (2) (e) gives
In the second case, if \(0< \ell \le \frac{1}{2}-\alpha \), we can choose \(\omega =\alpha \).
Similarly, we can prove that there is a \(\gamma _2>0\) such that
This completes the proof of the theorem.
3.4 Proof of Theorem 3
We denote
where \({ E}_{a, b}(., t, \tau )\) defined in Lemma 4. Using (3.7) and direct computations, one has
Since \( u_{\xi , \widetilde{\alpha }, \widetilde{\beta }}\) and \(u_{\zeta , \widetilde{\alpha }, \widetilde{\beta }} \) are solution of equations \( { F}_{\xi , \widetilde{\alpha }, \widetilde{\beta }, A}(w) = w \) and \( { F}_{\xi , \widetilde{\alpha }, \widetilde{\beta }, A}(v) \) \( = v\), respectively, by Lemma 3, we conclude that
This leads to the result of Theorem 3.
3.5 Proof of Theorem 4
We first state the following lemma necessary to prove the theorem. The proof of this lemma is postponed to the next subsection.
Lemma 5
Let \(T>0\), \(\zeta \in D(A^{s})\), \( \alpha , \, \widetilde{\alpha } \in [\alpha _*, \alpha ^*] \) , \( \beta , \, \widetilde{\beta }\in [\beta _*, \, \beta ^*]. \) Let the source function f satisfy Assumption F1(\(\alpha _*\)) and Assumption G1. Then the initial problems have the unique solutions \( u_{\zeta , \widetilde{\alpha }, \widetilde{\beta }} , u_{\zeta , \alpha , \beta } \in C([0,T], D(A^s)) \) with \(s\in [0, \min \{\beta /2, \widetilde{\beta }/2\}]\). Then, for any \(\epsilon >0\), there exist two constants \(P, \, P_\epsilon >0\) which are independent of \(\alpha , \widetilde{\alpha }, \, \beta , \widetilde{\beta }\) and t such that
for every \( t\in [0, T]\). Suppose further that \(\zeta \in D\left( A^{\beta ^*/2+r_1}\right) \) for some \(r_1>0\) and \(f(., u_{\zeta ,\alpha ,\beta }(.)) \in C([0, T], D\left( A^{r_2}\right) )\). Then, there exists a constant \(Q_0>0\) which is independent of \(\alpha , \widetilde{\alpha }, \, \beta , \widetilde{\beta }, \, N\) and t such that
for every \( t\in [0, T]\). Herein, \(\gamma _1=2\beta ^*\), \(\gamma _2=\min \{\beta ^*+2r_1-2s, 2r_2\}\).
Proof of Theorem 4
Using Theorem 3 and Lemma 5, we will prove the results of the theorem. Using the triangle inequality, we obtain
for any \(t \in [0, T]\). In addition, we note that
From the latter result, we can verify directly the main results (2.5), (2.6), (2.7) and (2.8) of the theorem. In fact, if \(s\in [0, \beta /2)\) then with k large enough, we have \(\beta _k/2 \ge s\). Hence, we can combine Lemma 3, Lemma 5 with (3.17) to obtain (2.5). We also use Lemmas 3 and 5 to deduce (2.6)-(2.7). Finally, combining Lemmas 3 with 5 (ii), we obtain (2.8). This completes the core of the proof. \(\square \)
3.6 The proof of Lemma 5
By a direct computation, we have
where
and the function Q is defined in (3.6). We will estimate \(I_k (k=1, 2, 3)\) one by one.
Estimate for \(I_1\). To give an estimation for \(I_1\), we separate the sum \(I_1\) into two sums as follows:
where
For convenience in estimating for \(I_{11}(N), I_{12}(N)\), let us assume \(N> \max \{e, \theta \}\).
Estimate for \(I_{11}(N)\). By Lemma 2, there exist two constants
\( C=C(\alpha _*, \alpha ^*, \beta _*, \beta ^*, T)>0 \) , \( C_0=C_0(\alpha _*, \alpha ^*, \beta _*, \beta ^*, \theta , T)>0 \) such that
where \(C_N=C_0 {\left| \!\left| \zeta \right| \!\right| }^2_s N^{2\beta ^*} \ln ^2 N\).
Estimate for \(I_{12}(N)\). We note that \(0\le E_\alpha (-x) \le 1\) for \(x>0\). This gives
Substituting (3.20) and (3.21) into (3.19), we obtain
where \(C_N\) is defined in (3.20).
Estimate for \(I_2\). Similarly to the proof of Lemma 3, we get
Estimate for \(I_3\). Recall that Q is defined in (3.6) as follows:
By the Hölder inequality and direct computation, we have
where
We will estimate \( I_{31}(N)\) and \( I_{32}(N)\) one by one.
Estimate for \( I_{31}(N)\). By Lemma 2, we have
By the mean value theorem, for \(\lambda \le N\) with N large enough, we obtain
On the other hand, there exists \(C_3=C_3(\alpha _*, \alpha ^*, \beta _*)\) such that
Plugging (3.26) and (3.27) into (3.25), we obtain
for N large enough and \(C_4=2C_2C_3\). Furthermore, thanks to the condition (2.2), we get that
where \(C_5=C_5(\alpha _*, \alpha ^*, \beta _*, { M})\). Combining the inequality (3.28) with (3.29), we obtain
where \(D_N=C_4C_5 N^{\beta ^*+2s} \ln N\).
Estimate for \(I_{32}(N)\). Thanks to (3.8), one has
Consequently,
where \(C_6=C_6(\beta _*, \beta ^*, \theta )\), and that
From (3.22), (3.24), (3.30) and (3.31), for \(|\alpha - \widetilde{\alpha } | + | \beta - \widetilde{\beta } |\le 1\), we obtain
where \(E_N=2C_N+6D_N\) with \(C_N\) defined in (3.20) and \(D_N\) defined in (3.30).
Let us mention (3.29) that
and \(\zeta \in D(A^s)\). This leads to the fact that there exists \(N=N(\epsilon )\) independent of \(\alpha , \widetilde{\alpha }\) and \(\beta , \widetilde{\beta }\) such that
Combining (3.32) and the latter inequality, one gets
Substituting (3.23) and (3.33) into (3.18), we obtain
Since \(u_{\zeta , \widetilde{\alpha }, \widetilde{\beta }}\) and \(u_{\zeta , \alpha , \beta }\) are the solution of the equations \({ F}_{\zeta , \widetilde{\alpha }, \widetilde{\beta }, A}(v)=v\) and \({ F}_{\zeta , \alpha , \beta , A}(u)=u\), respectively. We conclude from Lemma 3 that
where \(P_0=4 \varGamma (1-2\nu )\). This completes the proof of the first part of Theorem 5.
Now we consider the proof of the second part of the theorem. From (3.32) in the proof of Theorem 5, we obtain
where
In fact, using the assumption \(\zeta \in D(A^{\beta ^*/2+r_1})\), \(f(.,u(.))\in C([0,T],D(A^{r_2}))\) yields
and \(J_2 \le CN^{-2r_2}.\) Hence, putting \(\gamma _1=2\beta ^*\), \(\gamma _2=\min \{\beta ^*+2r_1-2s, 2r_2\}\), we get
Hence, we can use Lemma 3 to prove that
where \(Q_0\) is independent of \(N, \, \alpha , \, \widetilde{\alpha }, \, \beta , \, \widetilde{\beta }\).
Since \(\ln N<N\), we obtain
Let us suppose that \(|\alpha - \widetilde{\alpha }| + |\beta - \widetilde{\beta } | \le 1\), and we can choose
It is easy to see that
Hence, by (3.34), we obtain
This completed the proof of Lemma 5.
3.7 Proof of Theorem 5
Proof
Before proving the theorem, we set up some notations. We will use Theorem 1 to prove Part (i). For \(M>0\), we put
Verifying directly, we can prove that the function \(f_M\) is global Lipschitz with respect to the variable v, i.e.,
where \(\kappa _M>0\) depends on M. We consider the problem of finding \(U\in C([0,T],D(A^s))\) satisfying
From Theorem 1, for any \(T>0\), the equation (3.35) has a unique solution
-
(i)
For any \(m>0\), we put \( { M}= 2 \Vert \zeta \Vert + m\). Since \(U_T(0)=\zeta \), we can use the continuity of \(U_T\) to find a constant \(T_{ M}\in (0,T]\) such that \(\sup _{0\le t\le T_{ M}}\Vert U_{{ M},T}(t)\Vert _s\le { M}\). In this case \(f_{ M}(t,U_{{ M},T}(t))=f(t,U_{{ M},T}(t))\) for all \(t\in [0,T_{ M}]\) and \(U_{{ M},T}(t)\) satisfies (1.3) for \(t\in [0,T_{ M}]\).
-
(ii)
If \(V,W\in C([0,T];D(A^s))\) are solutions of (1.3), we denote
$$\begin{aligned} \mu =1+\max \left\{ \sup _{0\le t\le T}\Vert V(t)\Vert _s, \sup _{0\le t\le T}\Vert W(t)\Vert _s\right\} \end{aligned}$$and consider the equation
$$\begin{aligned} U(t) = E_\alpha \big (- t^\alpha A^\beta \big ) \zeta + \int _0^t { E}_{\alpha , \beta }(A, t, \tau ) f_\mu (\tau , U(\tau )) \mathrm {d}\tau . \end{aligned}$$(3.36)From Theorem 1, the equation (3.36) has a unique solution
$$\begin{aligned} U_{\mu ,T}\in C([0,T]; D(A^s)). \end{aligned}$$Since \(\Vert V(t)\Vert _s,\Vert W(t)\Vert _s\le \mu \) for \(t\in [0,T]\), we have
$$\begin{aligned} f(t,V(t))=f_\mu (t,V(t)),\ f(t,W(t))=f_\mu (t,W(t)). \end{aligned}$$Hence, V, W satisfy (3.36). By Theorem 1, we have \(V=U_{\mu ,T}=W\).
-
(iii)
For every \(T\in (0,T_{\zeta ,\alpha ,\beta })\), the equation (1.3) has a unique solution \(U_T\in C([0,T];D(A^s)\). From Part (ii), for \(T_1,T_2\in (0,T_{\zeta ,\alpha ,\beta })\), \(T_1<T_2\), we have \(U_{T_1}(t)=U_{T_2}(t)\) for \(t\in [0,T_1]\). Hence, we can put \(u_{\zeta ,\alpha ,\beta }(t)=U_T(t)\) for all \(t\in [0,T], T\in (0,T_{\zeta ,\alpha ,\beta })\). The function \(u_{\zeta ,\alpha ,\beta }\) is the unique solution of (1.3) on \([0,T_{\zeta ,\alpha ,\beta })\). We prove the second result of Part (iii). Assume by contradiction that \(T_{\zeta ,\alpha ,\beta }<\infty \) and \(\Vert u_{\zeta ,\alpha ,\beta }(t)\Vert _s\le M\) for every \(t\in [0,T_{\zeta ,\alpha ,\beta })\). We consider the equation
$$\begin{aligned} U(t) = E_\alpha \big (- t^\alpha A^\beta \big ) \zeta + \int _0^t { E}_{\alpha , \beta }(A, t, \tau ) f_M(\tau , U(\tau )) \mathrm {d}\tau . \end{aligned}$$(3.37)From Theorem 1, the equation (3.37) has a unique solution \(U_{M,\delta +T_{\zeta ,\alpha ,\beta }}\) with \(\delta >0\). From Part (ii) we have \(u_{\zeta ,\alpha ,\beta }(t)=U_{M,\delta +T_{\zeta ,\alpha ,\beta }}(t)\) for every \(t\in [0,T_{\zeta ,\alpha ,\beta })\). Since \(U_{M,\delta +T_{\zeta ,\alpha ,\beta }} \in C([0,\delta +T_{\zeta ,\alpha ,\beta }]; D(A^s))\), we can find a constant \(\delta '\in (0,\delta )\) such that \(\Vert U_{M,\delta +T_{\zeta ,\alpha ,\beta }}(t)\Vert _s\le M\) for \(t\in [0,\delta '+T_{\zeta ,\alpha ,\beta }]\). Hence the equation (1.3) has a unique solution on \([0,T_{\zeta ,\alpha ,\beta }+\delta ']\). It follows that \(T_{\zeta ,\alpha ,\beta }+\delta '\le T_{\zeta ,\alpha ,\beta }\), which is a contradiction. Finally, the proof of the last inequality of the theorem is similar to the inequality (2.3). Hence we omit it.
-
(iv)
Choose \(T\in (0,T_{\zeta ,\alpha ,\beta })\) and \(M=1+|u_{\zeta ,\alpha ,\beta }|_{s,T}\) and consider the problem (3.37) and
$$\begin{aligned} U(t) = E_{\alpha _k} \big (- t^{\alpha _k} A^{\beta _k} \big ) \zeta _k + \int _0^t { E}_{\alpha _k, \beta _k}(A, t, \tau ) f_M(\tau , U(\tau )) \mathrm {d}\tau . \end{aligned}$$(3.38)Denote the solution of (3.37), (3.38) by \(U_{\zeta ,\alpha ,\beta }\) and \(U_{\zeta _k,\alpha _k,\beta _k}\) respectively. From the stability result, we obtain
$$\begin{aligned} U_{\zeta _k,\alpha _k,\beta _k}\rightarrow U_{\zeta ,\alpha ,\beta }~\ \mathrm{in}\ ~C([0,T];D(A^s))~\ \mathrm{as}\ ~k\rightarrow \infty . \end{aligned}$$(3.39)Since \(|u_{\zeta ,\alpha ,\beta }|_{s,T}<M\), we have \(u_{\zeta ,\alpha ,\beta }=U_{\zeta ,\alpha ,\beta }\). From (3.39), there is a \(k_0\in \mathbb {N}\) such that \(| U_{\zeta _k,\alpha _k,\beta _k}|_{s,T}<M\) which gives \(U_{\zeta _k,\alpha _k,\beta _k}=u_{\zeta _k,\alpha _k,\beta _k}\) is the solution of Problem \(P_{\zeta _k,\alpha _k,\beta _k}\). It follows that \(T<T_{\zeta _k,\alpha _k,\beta _k}\) for \(k>k_0\) which implies \(\liminf _{k\rightarrow \infty }T_{\zeta _k,\alpha _k,\beta _k}\ge T_{\zeta ,\alpha ,\beta }\). Using (3.39) yields
$$\begin{aligned} \lim _{k\rightarrow \infty } |u_{\zeta _k,\alpha _k,\beta _k}-u_{\zeta ,\alpha ,\beta }|_{s,T}=0. \end{aligned}$$This completes the proof of the theorem.
\(\square \)
3.8 Proof of Theorem 6
Proof
\(\omega '\le \omega (\eta +1)-\ell \), \(\omega '<1/2-\nu \), \(\omega -\nu -\omega '<0\). We have \(\omega -\nu <\omega '\le \omega (\eta +1)-\ell \) which gives \(-\nu <\omega \eta -\ell \) or \(\omega >\frac{\ell -\nu }{\eta }\). We also need \(\omega -\nu <1/2-\nu \) which gives \(\omega <1/2\). Hence \(\frac{\ell -\nu }{\eta }<1/2\) Since \(\eta >\max \left\{ 2(\ell -\nu ),0\right\} \) we can choose \(\omega \) such that
which gives
Choosing \(\omega '\) such that
we obtain
The maximal solution \(u=u_{\zeta ,\alpha ,\beta }\in C([0,T_{\zeta ,\alpha ,\beta }),D(A^s))\) satisfies
We claim that \(T_{\zeta ,\alpha ,\beta }=\infty \) and \(\sup _{t>0}(1+t)^\omega \Vert u(t)\Vert _s<\infty \). We note that
As in the proof of Theorem 1, choosing \(r=0\) in Lemma 4 gives
and we have
Putting \(w(t)=(1+t)^\omega u(t)\) and, for \(\varLambda >\Vert \zeta \Vert _s\), denoting
We claim that there are \(\varLambda>0, \delta _0>0\) such that \(T_\mathrm{bound}=T_{\zeta ,\alpha ,\beta }\) for every \(\Vert \zeta \Vert ^2_s+m^2_{\infty ,\alpha ,\omega }<\delta _0\). Assume by contradiction that \(T_\mathrm{bound}<T_{\zeta ,\alpha ,\beta }\). Then, we obtain \(\Vert w(T_\mathrm{bound})\Vert _s=\varLambda \). The inequality (3.41) yields
where
From (3.4) we obtain
Next, we consider \(K_2\). For \(t\ge 1\), we obtain in view of Lemma 2 and (3.40) that
where \(\omega '\le \omega (\eta +1)-\ell \), \(\omega '<1/2-\nu \), \(\omega -\nu -\omega '<0\). We estimate for the case \(0<t<1\). We have
Combining the two cases gives
From the estimates for \(K_1, K_2\) we obtain
Since \(2(\eta +1)>2\) we can choose \(\varLambda , \zeta _0>0\) such that \( 1 > C\varLambda ^{2\eta }, \zeta ^2_0 < \varLambda ^2-C\varLambda ^{2(\eta +1)}\) and
In this case we get \(\varLambda ^2 = \Vert w(T_\mathrm{bound})\Vert _s^2 \le \zeta _0^2 + C\varLambda ^{2(\eta +1)}<\varLambda ^2 \) which is a contradiction. Hence, we have to obtain \(T_\mathrm{bound}=T_{\zeta ,\alpha ,\beta }\). So we have
Using the continuation result in Theorem 5 leads to \(T_{\zeta ,\alpha ,\beta }=\infty \). Finally, we consider the case \(\frac{\ell -\nu }{\eta }<\alpha <1/2\). In this case, we can choose \(\omega =\alpha \) and obtain the desired result by a similar argument as in Theorem 2. This completes the proof of the theorem. \(\square \)
References
Aldoghaither, A., Liu, D., Laleg-Kirati, T.: Modulating functions based algorithm for the estimation of the coefficients and differentiation order for a space-fractional advection-dispersion equation. SIAM J. Sci. Comput. 37(6), A2813–A2839 (2015)
Andrade, B., Au, V. V., O’Regan, D., Tuan, N. H.: Well-posedness results for a class of semilinear time-fractional diffusion equations. Z. Angew. Math. Phys. 71(5), Paper No. 161, 24 pp. (2020)
Asogwa, A. Foodun, M., Mijena, J., Nane, E.: Critical parameters for reaction-diffusion equations involving space-time fractional derivatives. Nonlinear Differ. Equ. Appl. 27(3) Paper No. 30, 22 pp. (2020)
Cao, J., Song, G., Wang, J., Shi, Q., Sun, S.: Blow-up and global solutions for a class of time fractional nonlinear reaction diffusion equation with weakly spatial source. Appl. Math. Lett. 91, 201–206 (2019)
Cazenave, T., Haraux, A.: An Introduction to Semilinear Evolution Equations, vol. 13. Oxford University Press, on Demand (1998)
Clement, P., Londen, S.-O., Simonett, G.: Quasilinear evolutionary equations and continuous interpolation spaces. newblock J. Differential Equations 196, 418–447 (2004)
Cheng, J., Nakagawa, J., Yamamoto, M., Yamazaki, T.: Uniqueness in an inverse problem for a one-dimension fractional diffusion equation. Inverse Problems 25(11), Art. 115002 (2009)
Dang, D.T., Nane, E., Nguyen, D.M., Tuan, N.H.: Continuity of solutions of a class of fractional equations. Potential Anal. 49(3), 423–478 (2018)
Dien, N.M., Trong, D.D.: Stability of solutions of a class of nonlinear fractional Laplacian parabolic problems. Math. Meth. Appl. Sci. 42(8), 2513–2532 (2019)
Duan, J. S.: Time-and space–fractional partial differential equations. J. Math. Phys. 46, Art. 013504 (2005)
Fabry, C.: Nagumo conditions for systems of second-order differential equations. J. Math. Anal. Appl. 107, 132–143 (1985)
Foias, C., Temam, R.: Gevrey class regularity for the solutions of the Navier-Stokes equations. J. Funct. Anal. 87, 359–369 (1989)
Gorenflo, R.: Abel Integral Equations. Springer-Verlag, Berlin-Heidelberg (1991)
Gorenflo, R., Kilbas, A. A., Mainardi, F., Rogosin, S. V.: Mittag-Leffler Function, Related Topics, and Applications. Springer-Verlag, Berlin-Heidelberg (2014), 2nd Ed. (2020)
Kateregga, M., Mataramvura, S., Taylor, D.: Parameter estimation for stable distributions with application to commodity futures log-returns. Cogent Economics & Finance 5, Art. 131 (2017)
Li, G., Zhang, D., Jia, X., Yamamoto, M.: Simultaneous inversion for the space–dependent diffusion coefficient and the fractional order in the time-fractional diffusion equation. Inverse Problems 29(6), Art. 065 (2013)
Schdmügen, K.:, Unbounded Self-adjoint Operators on Hilbert Space. Springer Science+Business Media, Dordrecht (2012)
Sakamoto, K., Yamamoto, M.: Initial value/boundary value problems for fractional diffusion-wave equations and applications to some inverse problems. J. Math. Anal. Appl. 382(1), 426–447 (2011)
Trong, D.D., Dien, N.M., Viet, T.Q.: Global solution of space-fractional diffusion equations with nonlinear reaction source terms. Appl. Anal. 355, 1–31 (2019)
Thach, T.N., Tuan, N.H.: Stochastic pseudo-parabolic equations with fractional derivative and fractional Brownian motion. Stoch. Anal. Appl. (2021). https://doi.org/10.1080/07362994.2021.1906274
Tuan, N.H., Au, V.V., Xu, R.: Semilinear Caputo time-fractional pseudo-parabolic equations. Commun. Pure Appl. Anal. 20(2), 583–621 (2021)
Viet, T.Q., Dien, N.M., Trong, D.D.: Stability of solutions of a class of nonlinear fractional Laplacian parabolic problems. J. Comput. Appl. Math. 355, 51–76 (2019)
Wang, W., Cheng, S., Guo, Z., Yan, X.: A note on the continuity for Caputo fractional stochastic differential equations. Chaos 30, Art. 073106 (2020)
Cheng, X., Li, Z., Yamamoto, M.: Asymptotic behavior of solutions to space-time fractional diffusion- reaction equations. Math. Methods Appl. Sci. 40, 1019–1031 (2017)
Yakubov, S., Yakubov, Y.: Differential-Operator Equation. Ordinary and Partial Differential Equations. Chapman and Hall/CRC, Boca Raton, FL (2000)
Zhang, Q., Li, Y.: Global well-posedness and blow-up solution of the Cauchy problem for a time-fractional superdiffusion equation. J. Evol. Equ. 19, 271–303 (2019)
Zhang, Q.-G., Sun, H.-R.: The Blow-up and global existence of solutions of Cauchy problems for a time fractional diffusion equation. Topol. Methods Nonlinear Anal. 46, 69–92 (2015)
Acknowledgements
The authors thank an anonymous referee and the editor for helpful comments that improved the quality and presentation of the paper. The research was supported by Vietnam National University of Hochiminh City [Grant No. B2021-18-02]. The research of Erkan Nane was partially supported by the Simons Foundation Collaboration Grants for Mathematicians.
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
The authors declare that they have no conflict of interest.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
About this article
Cite this article
Dien, N.M., Nane, E., Minh, N.D. et al. Global solutions of nonlinear fractional diffusion equations with time-singular sources and perturbed orders. Fract Calc Appl Anal 25, 1166–1198 (2022). https://doi.org/10.1007/s13540-022-00056-w
Received:
Revised:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s13540-022-00056-w
Keywords
- Fractional diffusion (primary)
- Caputo derivative
- Initial value problem
- Maximal solution
- Global solution
- Decay rate