Research article

On asymptotics of solutions for superdiffusion and subdiffusion equations with the Riemann-Liouville fractional derivative

  • Received: 19 March 2023 Revised: 25 April 2023 Accepted: 26 May 2023 Published: 07 June 2023
  • MSC : 26A33, 35R11, 35B40

  • In the present paper, we focus on the study of the asymptotic behaviors of solutions for the Cauchy problem of time-space fractional superdiffusion and subdiffusion equations with integral initial conditions, where the Riemann-Liouville derivative is used in the temporal direction and the integral fractional Laplacian is applied in the spatial variables. The fundamental solutions of the considered equations, which can be represented in terms of the Fox H-function, are constructed and investigated by using asymptotic expansions of the Fox H-function. Then, we obtain the asymptotic behaviors of solutions in the sense of Lp(Rd) and Lp,(Rd) norms, where Young's inequality for convolution plays a very important role. Finally, gradient estimates and large time behaviors of solutions are also provided. In particular, we derive the optimal L2- decay estimate for the subdiffusion equation.

    Citation: Zhiqiang Li, Yanzhe Fan. On asymptotics of solutions for superdiffusion and subdiffusion equations with the Riemann-Liouville fractional derivative[J]. AIMS Mathematics, 2023, 8(8): 19210-19239. doi: 10.3934/math.2023980

    Related Papers:

    [1] Saad Ihsan Butt, Artion Kashuri, Muhammad Umar, Adnan Aslam, Wei Gao . Hermite-Jensen-Mercer type inequalities via Ψ-Riemann-Liouville k-fractional integrals. AIMS Mathematics, 2020, 5(5): 5193-5220. doi: 10.3934/math.2020334
    [2] Iman Ben Othmane, Lamine Nisse, Thabet Abdeljawad . On Cauchy-type problems with weighted R-L fractional derivatives of a function with respect to another function and comparison theorems. AIMS Mathematics, 2024, 9(6): 14106-14129. doi: 10.3934/math.2024686
    [3] Ndolane Sene . Fractional input stability for electrical circuits described by the Riemann-Liouville and the Caputo fractional derivatives. AIMS Mathematics, 2019, 4(1): 147-165. doi: 10.3934/Math.2019.1.147
    [4] Ravi Agarwal, Snezhana Hristova, Donal O'Regan . Integral presentations of the solution of a boundary value problem for impulsive fractional integro-differential equations with Riemann-Liouville derivatives. AIMS Mathematics, 2022, 7(2): 2973-2988. doi: 10.3934/math.2022164
    [5] Yu-Pei Lv, Ghulam Farid, Hafsa Yasmeen, Waqas Nazeer, Chahn Yong Jung . Generalization of some fractional versions of Hadamard inequalities via exponentially (α,hm)-convex functions. AIMS Mathematics, 2021, 6(8): 8978-8999. doi: 10.3934/math.2021521
    [6] Snezhana Hristova, Antonia Dobreva . Existence, continuous dependence and finite time stability for Riemann-Liouville fractional differential equations with a constant delay. AIMS Mathematics, 2020, 5(4): 3809-3824. doi: 10.3934/math.2020247
    [7] Zhiying Li, Wangdong Jiang, Yuehong Zhang . Dynamic analysis of fractional-order neural networks with inertia. AIMS Mathematics, 2022, 7(9): 16889-16906. doi: 10.3934/math.2022927
    [8] Muhammad Tariq, Sotiris K. Ntouyas, Hijaz Ahmad, Asif Ali Shaikh, Bandar Almohsen, Evren Hincal . A comprehensive review of Grüss-type fractional integral inequality. AIMS Mathematics, 2024, 9(1): 2244-2281. doi: 10.3934/math.2024112
    [9] Dumitru Baleanu, Muhammad Samraiz, Zahida Perveen, Sajid Iqbal, Kottakkaran Sooppy Nisar, Gauhar Rahman . Hermite-Hadamard-Fejer type inequalities via fractional integral of a function concerning another function. AIMS Mathematics, 2021, 6(5): 4280-4295. doi: 10.3934/math.2021253
    [10] Muhammad Amer Latif, Humaira Kalsoom, Zareen A. Khan . Hermite-Hadamard-Fejér type fractional inequalities relating to a convex harmonic function and a positive symmetric increasing function. AIMS Mathematics, 2022, 7(3): 4176-4198. doi: 10.3934/math.2022232
  • In the present paper, we focus on the study of the asymptotic behaviors of solutions for the Cauchy problem of time-space fractional superdiffusion and subdiffusion equations with integral initial conditions, where the Riemann-Liouville derivative is used in the temporal direction and the integral fractional Laplacian is applied in the spatial variables. The fundamental solutions of the considered equations, which can be represented in terms of the Fox H-function, are constructed and investigated by using asymptotic expansions of the Fox H-function. Then, we obtain the asymptotic behaviors of solutions in the sense of Lp(Rd) and Lp,(Rd) norms, where Young's inequality for convolution plays a very important role. Finally, gradient estimates and large time behaviors of solutions are also provided. In particular, we derive the optimal L2- decay estimate for the subdiffusion equation.



    The aim of this paper is to consider asymptotic behaviors of solutions for the following time-space fractional superdiffusion equation with integral initial conditions and α(1,2):

    {RLDα0,tu(x,t)+(Δ)su(x,t)=f(x,t),xRd,t>0,RLDα20,tu(x,0)=φ(x),xRd,RLDα10,tu(x,0)=ψ(x),xRd, (1.1)

    and subdiffusion equation with integral initial condition and α(0,1):

    {RLDα0,tu(x,t)+(Δ)su(x,t)=g(x,t),xRd,t>0,RLDα10,tu(x,0)=ϕ(x),xRd, (1.2)

    where RLDα0,tu is the Riemann-Liouville derivative of u, (Δ)s denotes the integral fractional Laplace operator with s(0,1), and φ(x), ψ(x), ϕ(x), f(x,t), and g(x,t) are given functions. Moreover, the symbols RLDα20,t in Eq (1.1) and RLDα10,t in Eq (1.2) are Riemann-Liouville integral operators, and the symbol RLDα10,t in Eq (1.1) is Riemann-Liouville derivative operator.

    It is well known [14,19,28] that the Riemann-Liouville fractional integral of a function f(t)L1[a,b](<a<b<+) can be defined by

    RLDαa,tf(t)=1Γ(α)ta(tτ)α1f(τ)dτ,α>0,a<t<b, (1.3)

    and the Riemann-Liouville fractional derivative may be represented in the form

    RLDαa,tf(t)=dndtn(RLD(nα)a,tf(t))=1Γ(nα)dndtnta(tτ)nα1f(τ)dτ,a<t<b, (1.4)

    where n1<α<nN and f(t)ACn[a,b], here ACn[a,b] denotes the set of functions with an absolutely continuous (n1)st derivative.

    For a function v(x)H2s(Rd)={vS(Rd)|(Δ)svL2(Rd),s(0,1)} with S(Rd) being the Schwartz space, the integral fractional Laplacian of the function v(x) is given by [8]

    (Δ)sv(x)=C(d,s) P.V.Rdv(x)v(y)|xy|d+2sdy,xRd, (1.5)

    where P.V. denotes the Cauchy principal value and C(d,s) is a dimensional constant

    C(d,s)=(Rd1cosy1|y|d+2sdy)1,y=(y1,y2,,yd)Rd.

    Superdiffusion and subdiffusion equations in the forms of Eqs (1.1) and (1.2) have drawn much interest in developing existence, uniqueness, stability as well as asymptotics of the solutions, due to their excellent modelling capability for various applications such as theory of viscoelasticity [23], signal and image processing [31], anomalous diffusion [24], control theory [26], epidemic phenomena in biology [1], economics [4], etc. For more widespread applications on fractional differential equations we refer the reader to other works [3,9,10,12,14,19,27,28,33] and the references cited therein.

    As far as the asymptotic behaviors of solutions of fractional partial differential equations is concerned, we review some results on this topic in the current literatures. For the fractional superdiffusion (or call diffusion-wave) equation, the authors in [25] first studied the asymptotics of solutions in the sense of L norm, where time derivatives are the Riemann-Liouville and Caputo ones respectively and spatial derivative is the standard Laplace operator. After that, the article [7] investigated the asymptotic estimates of solution under Lp norm with 1p, where the Riemann-Liouville derivative replaced by Caputo derivative in Eq (1.1) and the initial conditions are written as u(x,0)=u0(x) and ut(x,0)=u1(x). Recently, Li and Li [21,22] discussed the same problem as above and derived similar asymptotic behaviors, in which the temporal derivative are taken as Caputo-Hadamard and ψ-Caputo fractional ones.

    On the other hand, concerning the fractional subdiffusion equation, Ma et al. [25] considered the asymptotic properties of such equation in the cases of the Riemann-Liouville and Caputo derivatives for Eq (1.2) when the force term is equal to zero, where the spatial direction is the standard Laplacian and the initial value is u(x,0)=u0(x). Subsequently, the paper [17] generalized these conclusions of [25] and they established the asymptotic analysis of solution in terms of Lp norm in which the Caputo derivative is used as temporal one. Shortly after, the results in [17] are further extended to time-space fractional subdiffusion equation [18] with the Caputo derivative and integral fractional Laplacian. Very recently, Li et al. [20,22] devoted to asymptotic properties of solution of Eq (1.2), where the Caputo-Hadamard and ψ-Caputo derivatives substituted for the Riemann-Liouville one. For other related studies we refer the reader to [16,32]. However, to the best of our knowledge, the asymptotic behaviors of solutions for Eqs (1.1) and (1.2) with the Riemann-Liouville derivative have been less studied and the literature [25] only considered very special cases for which the results obtained there can also be further improved.

    Based on the above reasons and existing research works, the goal of this paper is to study the asymptotic behaviors of solutions of Eqs (1.1) and (1.2) in the sense of more general Lp or weak Lp norms. Specifically, we first investigate asymptotic estimates of the solution to Eq (1.1). Using the technique of integral transforms the solution of convolutional form of Eq (1.1) is constructed and the fundamental solutions are also explicitly expressed by the Fox H-function. Then we estimate the fundamental solutions by means of asymptotic expansions of the H-function and further obtain the asymptotics of solution with the help of Young's inequality for convolution. By applying similar argument we can derive gradient estimates and large time behaviors of solution to Eq (1.1). For the subdiffusion Eq (1.2), we likewise discuss the asymptotic properties, gradient estimates and large time behaviors of solution. In particular we obtain the optimal decay rate in the sense of L2 norm. We find that these results with the Riemann-Liouville derivative in time are different from the Caputo case, for example, see Theorem 3.1 of this paper and Proposition 5.7 in [18].

    The remaining part of this article is organized as follows. In Section 2, the asymptotic behaviors of solution of the fractional superdiffusion Eq (1.1) are studied by means of Young's inequality for convolution. Further, the gradient estimates and large time behaviors of the solution are also presented. By using the almost same methods, Section 3 discusses decay estimates of the solution for the fractional subdiffusion Eq (1.2) and the optimal L2-decay rate is particularly derived. Some conclusions and remarks are presented in Section 4. At last, the Appendix recalls several integral transforms and concept of the Fox H-function. Throughout the paper we denote by C a generic positive constant whose value may vary from line to line.

    In this section, we shall study asymptotic analysis of the solution to Eq (1.1). First, the solution of convolutional form for Eq (1.1) is constructed in terms of Fourier and Laplace transforms, where the fundamental solutions are written via the Fox H-functions. We subsequently investigate asymptotic behaviors and estimations of Lp-norm for the fundamental solutions. Then, the asymptotic estimates of solution of Eq (1.1) are established by means of Young's inequality for convolution. Finally, we present gradient estimates and large time behaviors of the solution to Eq (1.1) by using the almost same argument.

    We first deduce the fundamental solutions and solution of Eq (1.1) by using integral transforms. Making use of the standard Laplace transform for temporal variable t and the Fourier transform for spatial variable x, and taking the formulas (A2) and (A4) into account, it follows that

    λα¯ˆu(ω,λ)λˆφ(ω)ˆψ(ω)+|ω|2s¯ˆu(ω,λ)=¯ˆf(ω,λ). (2.1)

    Furthermore,

    ¯ˆu(ω,λ)=¯ˆGφ(ω,λ)ˆφ(ω)+¯ˆGψ(ω,λ)ˆψ(ω)+¯ˆGf(ω,λ)¯ˆf(ω,λ), (2.2)

    where ¯ˆGφ(ω,λ)=λλα+|ω|2s and ¯ˆGψ(ω,λ)=¯ˆGf(ω,λ)=1λα+|ω|2s.

    Applying the inverse Fourier transform and inverse Laplace transform to the identity (2.2) we obtain

    u(x,t)=Gφ(x,t)φ(x)+Gψ(x,t)ψ(x)+Gf(x,t)f(x,t)=RdGφ(xy,t)φ(y)dy+RdGψ(xy,t)ψ(y)dy+t0RdGf(xy,tτ)f(y,τ)dydτ, (2.3)

    where the character denotes the standard convolution with respect to spatial variable, and the symbol is used as a convolution in time and space directions.

    In the following part, we present the explicit expressions of fundamental solutions Gφ(x,t), Gψ(x,t), and Gf(x,t) in (2.3). In terms of the relation (A3) for Laplace and Mellin transforms, one has

    ˜ˆGφ(ω,ξ)=M[ˆGφ(ω,t),ξ]=1Γ(1ξ)M[L[ˆGφ(ω,t),λ],1ξ]=1Γ(1ξ)M[¯ˆGφ(ω,λ),1ξ]=1Γ(1ξ)M[λλα+|ω|2s,1ξ]=1αΓ(1ξ)(|ω|2s)2ξα1Γ(2ξα)Γ(12ξα).

    The inverse Fourier transform of the above equality yields

    ˜Gφ(x,ξ)=1(2π)dRd˜ˆGφ(ω,ξ)eiωxdω=1(2π)d1αΓ(1ξ)Γ(2ξα)Γ(12ξα)Rd(|ω|2s)2ξα1eiωxdω=1(2π)d1αΓ(1ξ)Γ(2ξα)Γ(12ξα)(2π)d2|x|d22×0(ρ2s)2ξα1ρd2Jd21(ρ|x|)dρ,

    where Jd21(ρ|x|) is the first kind of Bessel function, see [14] for related definition and property. Observing that the formula (2.6.4) in [15], one has

    0(ρ2s)2ξα1ρd2Jd21(ρ|x|)dρ=0ρd2+(2ξα1)2sJd21(ρ|x|)dρ=|x|d2(2ξα1)2s12d2+(2ξα1)2sΓ(d2+(2ξα1)s)Γ((2ξα1)s).

    Therefore, we arrive at

    ˜Gφ(x,ξ)=|x|(12ξα)2sα|x|dπd22(12ξα)2sΓ(2ξα)Γ(12ξα)Γ(d2s(12ξα))Γ((2ξα1)s)Γ(1ξ).

    Finally, it follows from the inverse Mellin transform that

    Gφ(x,t)=12πic+ici˜Gφ(x,ξ)tξdξ=1|x|dπd212πic+iciΓ(2ξα)Γ(12ξα)Γ(d2s(12ξα))Γ((2ξα1)s)Γ(1ξ)×(|x|2s22s)12ξαtξd(12ξα)=tα2|x|dπd212πic+iciΓ(d2+s(2ξα1))Γ(1+(2ξα1))Γ(11(2ξα1))Γ(α1+α(2ξα1))Γ(11s(2ξα1))×(|x|2s22stα)(2ξα1)d(2ξα1)=tα2|x|dπd2H2123(|x|2s22stα|(1,1);(α1,α)(1,1),(d2,s);(1,s)),

    that is,

    Gφ(x,t)=tα2|x|dπd2H2123(|x|2s22stα|(1,1);(α1,α)(1,1),(d2,s);(1,s)). (2.4)

    Similarly, we can derive the expression of fundamental solutions Gψ(x,t)=Gf(x,t) in the form

    Gψ(x,t)=Gf(x,t)=tα1|x|dπd2H2123(|x|2s22stα|(1,1);(α,α)(1,1),(d2,s);(1,s)). (2.5)

    By using asymptotic expansions of the Fox H-function at infinity and zero [15], we can prove the following lemma on the fundamental solutions Gφ(x,t) in (2.4) and Gψ(x,t)=Gf(x,t) in (2.5).

    Lemma 2.1. Let dN, 1<α<2 and 0<s<1. Suppose R=tα|x|2s. For the fundamental solutions Gφ(x,t) in (2.4) and Gψ(x,t)=Gf(x,t) in (2.5), the following asymptotic behaviors hold.

    (1) If R>1, then

    |Gφ(x,t)|Ct2α2|x|d2s, (2.6)

    and if R1, then

    |Gφ(x,t)|{Ctα2|x|d+4s,d>4s,Ctα2(1+|log((|x|/2)2stα)|),d=4s,Ctα2αd2s,d<4s. (2.7)

    (2) If R>1, then

    |Gψ(x,t)|=|Gf(x,t)|Ct2α1|x|d2s, (2.8)

    and if R1, then

    |Gψ(x,t)|=|Gf(x,t)|{Ctα1|x|d+4s,d>4s,Ctα1(1+|log((|x|/2)2stα)|),d=4s,Ctα1αd2s,d<4s. (2.9)

    Proof. (1) For the fundamental solution Gφ(x,t) given by (2.4), we need to estimate asymptotic expansions of the H-function H2123(|x|2s22stα) which is the most important step. Noting that

    H2123(|x|2s22stα)=H2123(|x|2s22stα|(1,1);(α1,α)(1,1),(d2,s);(1,s)),x0,

    then we find that a=2α>0.

    We first prove (2.6) for R>1. Using Theorems 1.4 and 1.7 in [15] gives

    H2123(|x|2s22stα)=1l=1k=0hlk(|x|2s22stα)11k1=k=0h1k(|x|2s22stα)k,

    where

    h10=Γ(1)Γ(d2)Γ(α1)Γ(0)=0,  h11=Γ(2)Γ(d2+s)Γ(2α1)Γ(s)>0.

    Hence, one get

    H2123(|x|2s22stα)=h11(|x|2s22stα)1+o[(|x|2s22stα)1],|x|2s22stα.

    Furthermore, there holds

    |Gφ(x,t)|Cπd2|x|dtα2h11(|x|2s22stα)1Ct2α2|x|d2s,R.

    This illustrates that there is a positive constant M satisfying

    |Gφ(x,t)|Ct2α2|x|d2s,R>M. (2.10)

    In light of analyticity of the H-function H2123(|x|2s22stα), we find that it is bounded for 1<RM. Hence,

    |Gφ(x,t)|Cπd2|x|dtα2=C(|x|2stα)t2α2|x|d2sCMt2α2|x|d2sCt2α2|x|d2s,1<RM. (2.11)

    Combining (2.10) and (2.11) we obtain

    |Gφ(x,t)|Ct2α2|x|d2s.

    Next, we show (2.7) with R1. If d>4s, then b1σ=1+σ1=(σ+1) and b2k=d/2+ks for σ,k=0,1,2,. Therefore, b10 is a simple pole and Theorems 1.3 and 1.11 in [15] implies

    H2123(|x|2s22stα)=l=0h1l(|x|2s22stα)1+l.

    Since

    h10=Γ(d2s)Γ(1)Γ(1)Γ(s)=0,  h11=Γ(d22s)Γ(2)Γ(α1)Γ(2s)>0,

    then it follows that

    H2123(|x|2s22stα)=h11(|x|2s22stα)2+o[(|x|2s22stα)2],|x|2s22stα0,

    which indicates

    |Gφ(x,t)|Cπd2|x|dtα2h11(|x|2s22stα)2Ctα2|x|d+4s,R0.

    Consequently, there exists a positive constant δ such that

    |Gφ(x,t)|Ctα2|x|d+4s,R<δ. (2.12)

    Exploiting again analyticity of the H-function H2123(|x|2s22stα) we get

    |Gφ(x,t)|Cπd2|x|dtα2=C(|x|2stα)2tα2|x|d+4s=CR2tα2|x|d+4sCtα2|x|d+4s,δR1. (2.13)

    Using (2.12) and (2.13) yields

    |Gφ(x,t)|Ctα2|x|d+4s

    for d>4s, and which is the first inequality in (2.7).

    If d=4s, then the poles b10 is simple and the poles b11=b20=2 are coincided. In view of Theorems 1.5 and 1.12 in [15] we have

    H2123(|x|2s22stα)=H201(|x|2s22stα)2log(|x|2s22stα)+o[(|x|2s22stα)2log(|x|2s22stα)],  |x|2s22stα0,

    where H201=Γ(2)sΓ(α1)Γ(2s)0. As a result,

    |Gφ(x,t)|Cπd2|x|dtα2|H201|(|x|2s22stα)2|log(|x|2s22stα)|Ctα2|log(|x|2s22stα)|,R0.

    That is to say that there exists a positive constant δ1 such that

    |Gφ(x,t)|Ctα2|log(|x|2s22stα)|,R<δ1.

    We further derive

    |Gφ(x,t)|Ctα2(1+|log(|x|2s22stα)|)

    for d=4s, and the second inequality in (2.7) is proved.

    Finally, we show that the third inequality in (2.7) holds when d<4s. To do this, we consider three cases respectively. If d=2s, then the poles b10=1 and b20=d2s are coincide, but the coefficients H100=H101=0 by a direct calculation in terms of Theorems 1.5 and 1.12 in [15]. If d>2s, then b10=1 is a simple pole, but we find h10=0 in this case. If d<2s, then b20=d2s is a simple pole, by using Theorem 1.11 in [15] one has

    h2=Γ(1d2s)Γ(d2s)sΓ(α1αd2s)Γ(d2)>0.

    In either case, we can obtain

    H2123(|x|2s22stα)=h2(|x|2s22stα)d2s+o[(|x|2s22stα)d2s],|x|2s22stα0.

    Consequently,

    |Gφ(x,t)|Cπd2|x|dtα2|h2|(|x|2s22stα)d2sCtα2αd2s,R0.

    Furthermore it holds that

    |Gφ(x,t)|Ctα2αd2s

    for d<4s, and the third inequality holds.

    Similarly, we can prove (2.8) and (2.9) by using the same technique as the above (1) and omit them. The proof is now completed.

    In our further consideration, ||||p and ||||p, are used to simplify ||||Lp(Rd) and ||||Lp,(Rd) respectively, where Lp,(Rd) means weak Lp(Rd) space on Rd, for example, see [11]. We can also introduce

    κ(d,s)={dd4s,d>4s,,d4s,

    and

    κ(d,s)={dd+14s,d+2>4s,,d+24s.

    The following estimates of the fundamental solution Gφ(x,t) and Gψ(x,t)=Gf(x,t) in Lp(Rd) and Lp,(Rd) norms are crucial in proving the asymptotic behaviors of the solution of Eq (1.1).

    Lemma 2.2. Let dN, 1<α<2 and 0<s<1. Then for any t>0, it holds that Gφ(x,t)Lp(Rd) and

    ||Gφ(x,t)||pCtα2αd2s(11p), (2.14)

    for every 1p<κ(d,s). Moreover, if p=dd4s for d>4s, we have Gφ(x,t)Ldd4s,(Rd) and

    ||Gφ(x,t)||dd4s,Ctα2. (2.15)

    Proof. Firstly, we prove (2.14). Note that

    ||Gφ(x,t)||pp=R>1|Gφ(x,t)|pdx+R1|Gφ(x,t)|pdx.

    From (2.6) one can get

    R>1|Gφ(x,t)|pdxCR>1t(2α2)p|x|dp2spdxCt(2α2)ptα2sρdp2spρd1dρCtαp2pαd2s(p1),

    namely,

    (R>1|Gφ(x,t)|pdx)1pCtα2αd2s(11p),1p<. (2.16)

    On the other hand, when d>4s and 1p<κ(d,s), it follows from the first inequality in (2.7) that

    R1|Gφ(x,t)|pdxCR1t(α2)p|x|dp+4spdxCt(α2)ptα2s0ρ(4sd)pρd1dρCtαp2pαd2s(p1),

    i.e.,

    (R1|Gφ(x,t)|pdx)1pCtα2αd2s(11p),1p<κ(d,s). (2.17)

    If d=4s, applying the second inequality in (2.7) we obtain

    R1|Gφ(x,t)|pdxCR1t(α2)p(1+|log(|x|/2)2stα|)pdxCt(α2)p+2α122s0η(1+|logη|)pdηCt(α2)p+2α

    for 1p<. Consequently,

    (R1|Gφ(x,t)|pdx)1pCtα2+2αpCtα2αd2s(11p),1p<. (2.18)

    For d<4s, we use the third inequality in (2.7) to derive

    R1|Gφ(x,t)|pdxCR1tαp2pαd2spdxCtα2s0tαp2pαd2spdxCtαp2pαd2sp+αd2s

    for 1p<, which leads to

    (R1|Gφ(x,t)|pdx)1pCtα2αd2s(11p),1p<. (2.19)

    Collecting the above estimates (2.16)–(2.19), it follows that

    ||Gφ(x,t)||p(R>1|Gφ(x,t)|pdx)1p+(R1|Gφ(x,t)|pdx)1pCtα2αd2s(11p)

    for 1p<κ(d,s) with d1 and 0<s<1.

    We next show (2.15). Let R=tα|x|2s and p=dd4s for d>4s. Due to the fact

    ||Gφ(x,t)||p,=(||Gφ(x,t)χ{R>1}(t)+Gφ(x,t)χ{R1}(t)||p,)2(||Gφ(x,t)χ{R>1}(t)||p,+||Gφ(x,t)χ{R1}(t)||p,),

    where χ{E}(t) means the characteristic function of the set E. In terms of (2.16), there holds

    ||Gφ(x,t)χ{R>1}(t)||p,||Gφ(x,t)χ{R>1}(t)||pCtα2αd2s(11p)=Ctα2. (2.20)

    To estimate ||Gφ({\rm{x}},t)χ{R1}(t)||p,, we may use the first inequality in (2.7) to obtain

    dGφ(x,t)χ{R1}(t)(γ)=ϱ({xRd:|Gφ(x,t)|>γ and R1})ϱ({xRd:γ<Ctα2|x|4sd})C(tα2γ1)p,

    where ϱ stands for the measure on Rd. Thus we have

    γ(dGφ({\rm{x}},t)χ{R1}(t)(γ))1pCtα2.

    That is

    ||Gφ(x,t)χ{R1}(t)||p,Ctα2. (2.21)

    Therefore the required result follows by using (2.20) and (2.21) and the proof is thus completed.

    Remark 2.1. If d<4s, we infer from the third inequality of (2.7) in Lemma 2.1 that Gφ(,t)L(Rd) and ||Gφ(x,t)||Ctα2αd2s for all t>0.

    Lemma 2.3. Let dN, 1<α<2 and 0<s<1. If 1p<κ(d,s), then Gψ(x,t)=Gf(x,t)Lp(Rd) for any t>0 and

    ||Gψ(x,t)||p=||Gf(x,t)||pCtα1αd2s(11p),t>0. (2.22)

    Moreover, if p=dd4s and d>4s, then Gψ(x,t)=Gf(x,t)Ldd4s,(Rd) for any t>0 and

    ||Gψ(x,t)||dd4s,=||Gf(x,t)||dd4s,Ctα1,t>0. (2.23)

    Proof. The proof is similar to that of Lemma 2.2 above.

    Remark 2.2. For the case d<4s, it follows from the third inequality of (2.9) in Lemma 2.1 that Gψ(,t)=Gf(,t)L(Rd) and ||Gψ(x,t)||=||Gf(x,t)||Ctα1αd2s for any t>0.

    Let us now turn our attention to the asymptotic estimates of the solution to Eq (1.1) when the force term f0 and the initial values φ=ψ=0, respectively.

    Theorem 2.1. Let dN, 1<α<2 and 0<s<1. Suppose f0. Then the solution u(x,t)=Gφ(x,t)φ(x)+Gψ(x,t)ψ(x) to Eq (1.1), where φ,ψLq(Rd) for 1q, has the following asymptotic estimates:

    (1) If q=, then

    ||u(x,t)||Ctα2||φ(x)||+Ctα1||ψ(x)||,t>0. (2.24)

    (2) If 1q<, then

    ||u(x,t)||rCtα2αd2s(1q1r)||φ(x)||q+Ctα1αd2s(1q1r)||ψ(x)||q,t>0, (2.25)

    for any

    {r[q,dqd4sq),  if d>4sq,r[q,),            if d=4sq,r[q,],            if d<4sq.

    Moreover, it holds that

    ||u(x,t)||dqd4sq,Ctα2||φ(x)||q+Ctα1||ψ(x)||q,t>0, (2.26)

    if d>4sq.

    Proof. Let 1p,q,r satisfy the relation

    1+1r=1p+1q. (2.27)

    In view of Young's inequality for convolution, see (57) in [21], we get

    ||u(x,t)||r||Gφ(x,t)||p||φ(x)||q+||Gψ(x,t)||p||ψ(x)||q. (2.28)

    (1) If q=, then r= and p=1 by (2.27). Observe that (2.14) in Lemma 2.2 and (2.22) in Lemma 2.3 for p=1, one has

    ||Gφ(x,t)||1Ctα2,t>0,

    and

    ||Gψ(x,t)||1Ctα1,t>0,

    which together with (2.28) yields

    ||u(x,t)||Ctα2||φ(x)||+Ctα1||ψ(x)||,t>0.

    (2) If 1q<, then for r[q,dqd4sq) when d>4sq, we have 1p<dd4s when d>4s. Therefore, substituting (2.14) and (2.22) into (2.28) there holds

    ||u(x,t)||rCtα2αd2s(1q1r)||φ(x)||q+Ctα1αd2s(1q1r)||ψ(x)||q,t>0,

    as required (2.25). Similarly, we can prove (2.25) for r[q,) if d=4sq and r[q,] if d<4sq.

    Finally we show (2.26) for d>4sq. Recalling that Young's inequality for weak Lp-norm, see (58) in [21], combining (2.15), (2.23) and (2.27), it follows that

    ||u(x,t)||dqd4sq,Ctα2||φ(x)||q+Ctα1||ψ(x)||q,t>0,

    which is expected inequality (2.26) and the proof is thus complete.

    Theorem 2.2. Let dN, 1<α<2 and 0<s<1. And let 1q<. Assume that f(,t)Lq(Rd) for any t>0 and there exists some γ>0 such that

    ||f(x,t)||qC(1+t)γ,t>0. (2.29)

    Then for every

    {r[q,dqd2sq),  for 1q< and d>2sq,r[q,),            for 1<q< and d2sq,

    the solution u(x,t)=Gf(x,t)f(x,t) has the following estimates:

    ||u(x,t)||rCtαmin{1,γ}αd2s(1q1r),t>0,

    if γ1, and

    ||u(x,t)||rCtα1αd2s(1q1r)log(1+t),t>0,

    if γ=1.

    Proof. The proof of this theorem can be referred to that of Proposition 5.15 in [18] or Theorem 3.9 in [20].

    This subsection will develop gradient estimates and large time behaviors of the solution for Eq (1.1). Let us start with estimates of the derivatives for the fundamental solutions.

    Lemma 2.4. Let dN, 1<α<2 and 0<s<1. Suppose R=tα|x|2s. Then the spatial derivatives of Gφ(x,t) and Gψ(x,t) and the temporal derivatives of Gf(x,t) have the following estimates:

    (1) If R>1, then

    |Gφ(x,t)|Ct2α2|x|(d+1)2s, (2.30)

    and if R1, then

    |Gφ(x,t)|{Ctα2|x|(d+1)+4s,d+2>4s,Ctα2|x|(1+|log((|x|/2)2stα)|),d+2=4s,Ctα2α(d+2)2s|x|,d+2<4s. (2.31)

    (2) If R>1, then

    |Gψ(x,t)|Ct2α1|x|(d+1)2s, (2.32)

    and if R1, then

    |Gψ(x,t)|{Ctα1|x|(d+1)+4s,d+2>4s,Ctα1|x|(1+|log((|x|/2)2stα)|),d+2=4s,Ctα1α(d+2)2s|x|,d+2<4s. (2.33)

    (3) If R>1, then

    |tGf(x,t)|Ct2α2|x|d2s, (2.34)

    and if R1, then

    |tGf(x,t)|{Ctα2|x|d+4s,d>4s,Ctα2|x|(1+|log((|x|/2)2stα)|),d=4s,Ctα2αd2s,d<4s. (2.35)

    Proof. We only give the proof of (1), while (2) and (3) can be handled by similar method. Recalling that Property 2.8 in [15] results in

    Gφ(x,t)=tα2|x|d+1πd2H3134(|x|2s22stα|(1,1);(α1,α),(d,2s)(d+1,2s),(1,1),(d2,s);(1,s))(x1|x|,x2|x|,,xd|x|)=tα2|x|d+1πd2H3134(|x|2s22stα|(1,1);(α1,α),(d,2s)(1,1),(d2,s),(d+1,2s);(1,s))(x1|x|,x2|x|,,xd|x|),x0.

    Thus the modulus of Gφ(x,t) is

    |Gφ(x,t)|=tα2|x|d+1πd2|H3134(|x|2s22stα|(1,1);(α1,α),(d,2s)(1,1),(d2,s),(d+1,2s);(1,s))|,x0. (2.36)

    First of all, we prove (2.30). By means of Theorems 1.4 and 1.7 in [15] we find that

    H3134(|x|2s22stα)=k=0h1k(|x|2s22stα)k,

    where

    h10=Γ(1)Γ(d+1)Γ(d2)Γ(α1)Γ(d)Γ(0)=0,  h11=Γ(2)Γ(d2+s)Γ(d+1+2s)Γ(2α1)Γ(d+2s)Γ(s)>0.

    Hence, it follows that

    H3134(|x|2s22stα)=h11(|x|2s22stα)1+o[(|x|2s22stα)1],|x|2s22stα,

    from which one has

    |Gφ(x,t)|Cπd2|x|d1tα2h11(|x|2s22stα)1Ct2α2|x|(d+1)2s,R.

    Using this inequality and the analyticity of the H-function H3134(|x|2s22stα), we conclude that there exists a positive constant C such that

    |Gφ(x,t)|Ct2α2|x|(d+1)2s

    for R>1 and which gives (2.30).

    Next we show (2.31). According to Theorems 1.3 and 1.11 in [15], we find h10=h20=h30=0. Hence, b11=2 is a simple pole when d+2>4s and

    H3134(|x|2s22stα)=h11(|x|2s22stα)2+o[(|x|2s22stα)2],|x|2s22stα0,

    where

    h11=Γ(d22s)Γ(d+14s)Γ(2)Γ(α1)Γ(d4s)Γ(2s).

    Thus there is a positive constant C such that

    |Gφ(x,t)|Cπd2|x|d1tα2|h11|(|x|2s22stα)2Ctα2|x|(d+1)+4s,R0.

    We further obtain

    |Gφ(x,t)|Ctα2|x|(d+1)+4s (2.37)

    for R1 and d+2>4s.

    If d+2=4s, we see that the poles b11=b21=b31=2 are coincided, then

    H3134(|x|2s22stα)=H111(|x|2s22stα)2log(|x|2s22stα)+o[(|x|2s22stα)2log(|x|2s22stα)],  |x|2s22stα0,

    where H1110. Then one gets

    |Gφ(x,t)|Ctα2|x|d+1πd2|H111|(|x|2s22stα)2|log(|x|2s22stα)|Ctα2|x||log(|x|2s22stα)|,R0.

    Furthermore, it follows that

    |Gφ(x,t)|Ctα2|x|(1+|log((|x|/2)2stα)|) (2.38)

    for R1 and d+2=4s.

    It remains to show the case d+2<4s. Since the poles b21=b31=d+22s are coincided, then

    H3134(|x|2s22stα)=H210(|x|2s22stα)d+22s+o[(|x|2s22stα)d+22s],|x|2s22stα0,

    with H2100, which leads to

    |Gφ(x,t)|Ctα2|x|d+1πd2|H210|(|x|2s22stα)d+22sCtα2α(d+2)2s|x|,R0.

    So we have

    |Gφ(x,t)|Ctα2α(d+2)2s|x| (2.39)

    for R1 and d+2<4s.

    Based on (2.37)–(2.39), the desired assertion (2.31) is obtained and the proof is now completed.

    According to Lemma 2.4, we can establish estimates of ||Gφ(x,t)||p, ||Gψ(x,t)||p, and ||Gf(x,t)||p and further derive asymptotic properties of u(x,t) to Eq (1.1), whose proofs are very similar to those of counterparts in the previous subsection and omitted.

    Lemma 2.5. Let dN, 1<α<2 and 0<s<1. Assume that 1p<κ(d,s). Then it holds that Gφ(x,t)Lp(Rd;Rd) for any t>0 and

    ||Gφ(x,t)||pCtα2α2sαd2s(11p),t>0. (2.40)

    Moreover, if p=dd+14s for d+2>4s, then Gφ(x,t)Ldd+14s,(Rd;Rd) for all t>0 and

    ||Gφ(x,t)||dd+14s,Ctα2,t>0. (2.41)

    Lemma 2.6. Let dN, 1<α<2 and 0<s<1. Assume that 1p<κ(d,s). Then it holds that Gψ(x,t)=Gf(x,t)Lp(Rd;Rd) for all t>0 and

    ||Gψ(x,t)||p=||Gf(x,t)||pCtα1α2sαd2s(11p),t>0. (2.42)

    Moreover, if p=dd+14s for d+2>4s, then Gψ(x,t)=Gf(x,t)Ldd+14s,(Rd;Rd) for all t>0 and

    ||Gψ(x,t)||dd+14s,=||Gf(x,t)||dd+14s,Ctα1,t>0. (2.43)

    Theorem 2.3. Let dN, 1<α<2 and 0<s<1. Suppose that 1q and f0. Then the following estimates hold on the gradient of solution u(x,t)=Gφ(x,t)φ(x)+Gψ(x,t)ψ(x) with φ,ψLq(Rd).

    (1) If q=, then

    ||u(x,t)||Ctα2α2s||φ(x)||+Ctα1α2s||ψ(x)||,t>0.

    (2) If 1q<, then for any

    {r[q,dqd(4s1)q),  if d>(4s1)q,r[q,),                     if d(4s1)q,

    one has

    ||u(x,t)||rCtα2α2sαd2s(1q1r)||φ(x)||q+Ctα1α2sαd2s(1q1r)||ψ(x)||q,t>0.

    Moreover, if d>(4s1)q, then

    ||u(x,t)||dqd(4s1)q,Ctα2||φ(x)||q+Ctα1||ψ(x)||q,t>0.

    Theorem 2.4. Let dN, 1<α<2 and 0<s<1. Let 1q< and the condition (2.29) be satisfied. Then for every

    {r[q,dqd(4s1)q),  for 1q< and d>(4s1)q,r[q,),                     for 1<q< and d(4s1)q,

    the gradient of solution u(x,t)=Gf(x,t)f(x,t) has the following relations:

    ||u(x,t)||rCtαmin{1,γ}α2sαd2s(1q1r),t>0, (2.44)

    if γ1, and

    ||u(x,t)||rCtα1α2sαd2s(1q1r)log(1+t),t>0, (2.45)

    if γ=1.

    The last two theorems present the large time behavior of the solution u(x,t) for Eq (1.1).

    Theorem 2.5. Let dN, 1<α<2 and 0<s<1. Denote Mφ=Rdφ(x)dx and Mψ=Rdψ(x)dx with φ,ψL1(Rd). Assume that f0 and 1p<κ(d,s). Then we have the following results.

    (1) If |||x|φ(x)||1< and |||x|ψ(x)||1<, then

    tαd2s(11p)+1α||u(x,t)MφGφ(x,t)MψGψ(x,t)||pCtα2s1+Ctα2s (2.46)

    for any t>0. Moreover, when p=dd+14s, one gets

    tα(4s1)2s+1α||u(x,t)MφGφ(x,t)MψGψ(x,t)||dd+14s,Ctα2s1+Ctα2s (2.47)

    for any t>0.

    (2) It follows that

    tαd2s(11p)+1α||u(x,t)MφGφ(x,t)MψGψ(x,t)||p0 (2.48)

    as t.

    Proof. (1) Note that the conditions φ,ψL1(Rd) and |||x|φ(x)||1< and |||x|ψ(x)||1<. It follows from the decomposition lemma (see Lemma 8.4 [18]) that there exists functions Φ,ΨL1(Rd;Rd) such that

    φ=Mφδφ+divΦ,   ψ=Mψδψ+divΨ,

    where ||Φ||1C|||x|φ(x)||1 and ||Ψ||1C|||x|ψ(x)||1. Therefore we find that

    u(x,t)=Gφ(x,t)(Mφδφ+divΦ)+Gψ(x,t)(Mψδψ+divΨ)=MφGφ(x,t)+Gφ(x,t) Φ(x)+MψGψ(x,t)+Gψ(x,t) Ψ(x),

    where means the convolution between two vector functions. Further there holds

    u(x,t)MφGφ(x,t)MψGψ(x,t)=Gφ(x,t) Φ(x)+Gψ(x,t) Ψ(x). (2.49)

    By using Young's inequality for convolution (57) in [21], and taking (2.40) and (2.42) into account, we obtain

    ||u(x,t)MφGφ(x,t)MψGψ(x,t)||p||Gφ(x,t)||p||Φ(x)||1+||Gψ(x,t)||p||Ψ(x)||1Ctα2α2sαd2s(11p)+Ctα1α2sαd2s(11p)

    for 1p<κ(d,s), which gives

    tαd2s(11p)+1α||u(x,t)MφGφ(x,t)MψGψ(x,t)||pCtα2s1+Ctα2s,

    and the claim (2.46) holds. For the limit case p=dd+14s, applying Young's inequality for convolution (58) in [21], (2.41) and (2.43) to (2.49) we immediately know that (2.47) is true.

    (2) Let a sequence {ηm(x)}C0(Rd) satisfy Rdηm(x)dx=Mφ for all m and ηm(x)φ(x) as m in L1(Rd). Likewise, set a sequence {ζn(x)}C0(Rd) satisfy Rdζn(x)dx=Mψ for all n and ζn(x)ψ(x) as n in L1(Rd). Now we use Young's inequality for convolution (57) in [21], (2.14), (2.22) and the conclusion of (1) to derive for any m,n

    ||u(x,t)MφGφ(x,t)MψGψ(x,t)||p||Gφ(x,t)φ(x)MφGφ(x,t)||p+||Gψ(x,t)ψ(x)MψGψ(x,t)||p||Gφ(x,t)(φ(x)ηm(x))||p+||Gφ(x,t)ηm(x)MφGφ(x,t)||p+||Gψ(x,t)(ψ(x)ζn(x))||p+||Gψ(x,t)ζn(x)MψGψ(x,t)||pCtα2αd2s(11p)||φηm||1+Cmtα2α2sαd2s(11p)+Ctα1αd2s(11p)||ψζn||1+Cntα1α2sαd2s(11p).

    Consequently,

    tαd2s(11p)+1α||u(x,t)MφGφ(x,t)MψGψ(x,t)||pCt1||φηm||1+C||ψζn||1+Cmtα2s1+Cntα2s,

    from which we have

    lim supttαd2s(11p)+1α||u(x,t)MφGφ(x,t)MψGψ(x,t)||pC||ψζn||1,

    and the claimed result (2.48) can be achieved by letting n. The proof of this theorem is now ended.

    Theorem 2.6. Let dN, 1<α<2 and 0<s<1. Let φψ0 and denote Mf=0Rdf(x,t)dxdt. Moreover, let us assume that f(x,t)L1(Rd×(0,)) and there exists some γ>1 such that

    ||f(x,t)||1C(1+t)γ,t>0.

    Then it holds that

    t1α+αd2s(11p)||u(x,t)MfGf(x,t)||p0

    as t for any

    {1p,        if d<4s,1p<κ(d,s),  if d4s.

    Proof. The proof can be completed by using the methods provided by Theorem 2.21 in [18] or Theorem 3.16 in [20] and the details are omitted here.

    In this section we shall deal with decay behaviors of the solution for Eq (1.2). Similar to the previous section, we present the asymptotics of the fundamental solution, decay estimates of the solution, gradient estimates and large time behaviors. In particular the optimal L2-decay estimates are provided by virtue of Plancherel's theorem and the boundedness of Mittag-Leffler function. For the most of theorems and lemmas in the section, we directly give results without proofs since their proof techniques are very similar to ones of corresponding conclusions in the previous section.

    We first construct the solution of Eq (1.2) by integral transforms. Applying Fourier and Laplace transforms to Eq (1.2), and noticing that formula (2.248) in [28] and equality (A4), there holds

    λα¯ˆu(ω,λ)ˆϕ(ω)+|ω|2s¯ˆu(ω,λ)=¯ˆg(ω,λ). (3.1)

    Then we get

    ¯ˆu(ω,λ)=1λα+|ω|2sˆϕ(ω)+1λα+|ω|2s¯ˆg(ω,λ):=¯ˆG(ω,λ)ˆϕ(ω)+¯ˆG(ω,λ)¯ˆg(ω,λ). (3.2)

    Performing the inverse Fourier transform and inverse Laplace transform on both sides of (3.2), the solution of Eq (1.2) reads as

    u(x,t)=G(x,t)ϕ(x)+G(x,t)g(x,t)=RdG(xy,t)ϕ(y)dy+t0RdG(xy,tτ)g(y,τ)dydτ, (3.3)

    where the fundamental solution [18]

    G(x,t)=tα1|x|dπd2H2123(|x|2s22stα|(1,1);(α,α)(1,1),(d2,s);(1,s)). (3.4)

    Lemma 3.1. [18] Let dN, 0<α<1 and 0<s<1. Suppose R=tα|x|2s. Then for the fundamental solution G(x,t) in (3.4) we have

    |G(x,t)|Ct2α1|x|d2s (3.5)

    for R>1, and

    |G(x,t)|{Ctα1|x|d+4s,d>4s,Ctα1(1+|log((|x|/2)2stα)|),d=4s,Ctα1αd2s,d<4s (3.6)

    for R1.

    Lemma 3.2. [18] Let dN, 0<α<1 and 0<s<1. If 1p<κ(d,s), then we have G(x,t)Lp(Rd) for all t>0 and

    ||G(x,t)||pCtα1αd2s(11p),t>0. (3.7)

    And if moreover p=dd4s for d>4s, then it follows that G(x,t)Ldd4s,(Rd) for all t>0 and

    ||G(x,t)||dd4s,Ctα1,t>0. (3.8)

    Remark 3.1. If d<4s, by the third inequality of (3.6) in Lemma 3.1, it is clear that G(,t)L(Rd) and ||G(x,t)||Ctα1αd2s hold for all t>0.

    Theorem 3.1. Let dN, 0<α<1 and 0<s<1. Let 1q and g0. Then the solution u(x,t)=G(x,t)ϕ(x) to Eq (1.2) with ϕLq(Rd) has the following decay estimates:

    (1) If q=, then

    ||u(x,t)||Ctα1||ϕ(x)||,t>0.

    (2) If 1q<, then

    ||u(x,t)||rCtα1αd2s(1q1r)||ϕ(x)||q,t>0

    holds for any

    {r[q,dqd4sq),  if d>4sq,r[q,),            if d=4sq,r[q,],            if d<4sq.

    If moreover d>4sq, then

    ||u(x,t)||dqd4sq,Ctα1||ϕ(x)||q,t>0.

    Theorem 3.2. [18] Let dN, 0<α<1 and 0<s<1. Let 1q< and ϕ0. Assume that g(,t)Lq(Rd) for all t>0 and there is some γ>0 such that

    ||g(x,t)||qC(1+t)γ,t>0. (3.9)

    Then the solution u(x,t)=G(x,t)g(x,t) satisfies the following

    ||u(x,t)||rCtαmin{1,γ}αd2s(1q1r),γ1,

    and

    ||u(x,t)||rCtα1αd2s(1q1r)log(1+t),γ=1,

    for any t>0 and

    {r[q,dqd2sq),  for 1q< and d>2sq,r[q,),            for 1<q< and d2sq.

    The gradient estimates and large time behaviors of the solution for Eq (1.2) are provided in the subsection.

    Lemma 3.3. [18] Let dN, 0<α<1 and 0<s<1. Suppose R=tα|x|2s. Then the spatial and time derivatives of the fundamental solution G(x,t) in (3.4) have the following decay behaviors:

    (1) If R>1, then

    |G(x,t)|Ct2α1|x|(d+1)2s,d1,0<s<1, (3.10)

    and if R1, then

    |G(x,t)|{Ctα1|x|(d+1)+4s,d+2>4s,Ctα1|x|(1+|log((|x|/2)2stα)|),d+2=4s,Ctα1α(d+2)2s|x|,d+2<4s. (3.11)

    (2) If R>1, then

    |tG(x,t)|Ct2α2|x|d2s,d1,0<s<1, (3.12)

    and if R1, then

    |tG(x,t)|{Ctα2|x|d+4s,d>4s,Ctα2|x|(1+|log((|x|/2)2stα)|),d=4s,Ctα2αd2s,d<4s. (3.13)

    Lemma 3.4. Let dN, 0<α<1 and 0<s<1. For 1p<κ(d,s), it holds that G(x,t)Lp(Rd;Rd) for any t>0 and

    ||G(x,t)||pCtα1α2sαd2s(11p),t>0. (3.14)

    Moreover, if p=dd+14s for d+2>4s, we have G(x,t)Ldd+14s,(Rd;Rd) for any t>0 and

    ||G(x,t)||dd+14s,Ctα1,t>0. (3.15)

    Theorem 3.3. Let dN, 0<α<1 and 0<s<1. Let 1q and g0. Then for u(x,t)=G(x,t)ϕ(x) with ϕLq(Rd), we have:

    (1) If q=, then

    ||u(x,t)||Ctα1α2s||ϕ(x)||,t>0.

    (2) If 1q<, then

    ||u(x,t)||rCtα1α2sαd2s(1q1r)||ϕ(x)||q,t>0

    for any

    {r[q,dqd(4s1)q),  if d>(4s1)q,r[q,),                     if d(4s1)q.

    Moreover if d>(4s1)q, then

    ||u(x,t)||dqd(4s1)q,Ctα1||ϕ(x)||q,t>0.

    Theorem 3.4. Let dN, 0<α<1 and 0<s<1. Let 1q< and ϕ0. Assume that g(,t)Lq(Rd) for any t>0 and the assumption (3.9) holds. Then u(x,t)=G(x,t)g(x,t) has the estimates

    ||u(x,t)||rCtαmin{1,γ}α2sαd2s(1q1r),γ1,

    and

    ||u(x,t)||rCtα1α2sαd2s(1q1r)log(1+t),γ=1,

    for all t>0 and for every

    {r[q,dqd(4s1)q),  if 1q< and d>(4s1)q,r[q,),                     if 1<q< and d(4s1)q.

    Theorem 3.5. Let dN, 0<α<1 and 0<s<1. Let 1p<κ(d,s) and g0. Assume ϕL1(Rd) and denote Mϕ=Rdϕ(x)dx. Then the following results hold.

    (1) If |||x|ϕ(x)||1<, then

    t1α+αd2s(11p)||u(x,t)MϕG(x,t)||pCtα2s,t>0.

    When p=dd+14s, one gets

    t1α+α(4s1)2s||u(x,t)MϕG(x,t)||dd+14s,Ctα2s,t>0.

    (2) It follows that

    t1α+αd2s(11p)||u(x,t)MϕG(x,t)||p0

    as t.

    Theorem 3.6. Let dN, 0<α<1 and 0<s<1. Let ϕ0 and denote Mg=0Rdg(x,t)dxdt. Let us assume g(x,t)L1(Rd×(0,)) and there exists some γ>1 such that

    ||g(x,t)||1C(1+t)γ,t>0.

    Then for all

    {1p,         if d<4s,1p<κ(d,s),   if d4s,

    it holds that

    t1α+αd2s(11p)||u(x,t)MgG(x,t)||p0

    as t.

    In this subsection our attention will be restricted to decay estimate of the solution for Eq (1.2) in the sense of L2-norm when g0. To do this, let us consider the solution of Eq (1.2) under the Fourier transform. By using the inverse Laplace transform on both sides of (3.1) we obtain

    ˆu(ω,t)=ˆG(ω,t)ˆϕ(ω)+t0ˆG(ω,tτ)ˆg(ω,τ)dτ,

    where

    ˆG(ω,t)=tα1Eα,α(|ω|2stα).

    To derive optimal L2 decay rate, we need investigate some properties of the Mittag-Leffler function Eα,α(η).

    Lemma 3.5. [13] Let 0<α<1. Then the Mittag-Leffler function Eα,α(η)>0 for η(0,).

    Lemma 3.6. [13] Let 0α1 and βα. Then the Mittag-Leffler function Eα,β(η) is completely monotone for η(0,).

    Lemma 3.7. Let 0<α<1. Then there are positive constants C1 and C2 such that

    C11+η2Eα,α(η)C21+η2,η0.

    Proof. First of all, we know that Eα,α(η)>0 for 0<α<1 and η0. In fact, one has Eα,α(0)=1/Γ(α)>0. For η>0, it is evident that Eα,α(η)>0 by Lemma 3.5.

    In view of asymptotic expansions for the Mittag-Leffler function (see Theorem 1.4 in [28]) we obtain

    Eα,α(η)=1Γ(α)1η2+O(η3),η.

    This implies the function Eα,α(η) behaves as C0/η2 when η with some positive constant C0, i.e., there exist a positive real number X and two positive constants C3 and C4 such that

    C31+η2Eα,α(η)C41+η2,η>X. (3.16)

    Since the Mittag-Leffler function Eα,α(η) is analytic for any ηR, then it is bound on the interval [0,X]. Moreover, it follows from Lemma 3.6 that the Mittag-Leffler function Eα,α(η) is monotonically decreasing for η(0,). Therefore, there are positive constants C5 and C6 satisfying C5Eα,α(η)C6 with 0ηX. We further obtain

    C71+η2Eα,α(η)C81+η2,η[0,X] (3.17)

    with positive constants C7 and C8.

    The desired result follows from (3.16) and (3.17) and the proof is complete.

    We first present the estimates of lower bound for the solution u(x,t)=G(x,t)ϕ(x) to Eq (1.2) with g0 in the sense of L2-norm.

    Theorem 3.7. Let dN, 0<α<1, 0<s<1, and d8s. Let g0 in Eq (1.2). If ϕL1(Rd)L2(Rd) and Rdϕ(x)dx0, then the solution u(x,t)=G(x,t)ϕ(x) to Eq (1.2) has the lower bound estimate

    ||u(x,t)||2Ctmin{α+1,1α+αd4s},t>t0>0.

    Proof. Let ρ=ρ(t)(0,ρ0] with t>0 and ρ0>0. According to Plancherel's theorem and the estimate for the Mittag-Leffler function in Lemma 3.7, it follows that

    ||u(x,t)||22=||ˆu(ω,t)||22=Rd|ˆG(ω,t)|2|ˆϕ(ω)|2dωBρ(0)|tα1Eα,α(|ω|2stα)|2|ˆϕ(ω)|2dωCt2α2(1+ρ4st2α)2Bρ(0)|ˆϕ(ω)|2dω=Ct2α2(1+ρ4st2α)2ρd(ρdBρ(0)|ˆϕ(ω)|2dω). (3.18)

    From the Plancherel's theorem and Riemann-Lebesgue lemma, it is easy to verify that ˆϕC0(Rd)L2(Rd) holds. By making use of Lebesgue differentiation theorem we have for a sufficient small ρ0

    ρdBρ(0)|ˆϕ(ω)|2dω|ˆϕ(0)|22,ρ(0,ρ0].

    Substituting this into (3.18) leads to

    ||u(x,t)||22Ct2α2|ˆϕ(0)|2ρd2(1+ρ4st2α)2. (3.19)

    Now letting ρ=ρ0 in (3.19) one gets

    ||u(x,t)||22Ct2α2,t>t0>0. (3.20)

    On the other hand, we may take ρ=ρ0(1+t2α)1/4s in (3.19) to derive

    ||u(x,t)||22Ct2α2αd2s,t>t0>0. (3.21)

    Together with (3.20) and (3.21), the desired lower bound is established and the proof is thus completed.

    The upper bound for the solution u(x,t)=G(x,t)ϕ(x) to Eq (1.2) is estimated in the following theorem when g0.

    Theorem 3.8. Let dN, 0<α<1, 0<s<1, and d8s. Let g0 in Eq (1.2). If ϕL1(Rd)L2(Rd) and Rdϕ(x)dx0, then the solution u(x,t)=G(x,t)ϕ(x) to Eq (1.2) satisfies the upper bound estimate

    ||u(x,t)||2Ctmin{α+1,1α+αd4s},t>0. (3.22)

    For d=8s one has

    ||u(x,t)||2,Ctα1,t>0. (3.23)

    Proof. We divide into two cases to prove the assertion (3.22). When d<8s, applying Plancherel's theorem and Lemma 3.7 one gets

    ||u(x,t)||22=||ˆu(ω,t)||22=Rd|ˆG(ω,t)|2|ˆϕ(ω)|2dω||ˆϕ||2Rd|ˆG(ω,t)|2dωC||ϕ||21Rdt2α2(1+|ω|4st2α)2dω=C||ϕ||21t2α20ρd1dρ(1+ρ4st2α)2=Ct2α2αd2s||ϕ||210ρd11dρ1(1+ρ4s1)2,

    i.e.,

    ||u(x,t)||2Ctα1αd4s. (3.24)

    If d>8s, then ϕL1(Rd)L2(Rd) implies ϕL2dd+8s(Rd) by interpolation. Furthermore, Theorem 8.5 in [18] gives

    ||(Δ)2sϕ||2C||ϕ||2dd+8s<. (3.25)

    It follows from Plancherel's theorem and Lemma 3.7 that

    ||u(x,t)||22=||ˆu(ω,t)||22=Rd|ˆG(ω,t)|2|ˆϕ(ω)|2dωCRdt2α2(1+|ω|4st2α)2|ˆϕ(ω)|2dω=Ct2α2Rd|ω|8st4α(1+|ω|4st2α)2||ω|4sˆϕ(ω)|2dωCt2α2Rd||ω|4sˆϕ(ω)|2dω=Ct2α2||(Δ)2sϕ||22. (3.26)

    Substituting (3.25) into (3.26) we have

    ||u(x,t)||2Ctα1. (3.27)

    Based on (3.24) with (3.27), the expected inequality (3.22) is proved.

    If d=8s, then by using Young's inequality for convolution and (2.15) in Lemma 2.2 it is evident that

    ||u(x,t)||2,=||G(x,t)ϕ(x)||2,C||G(x,t)||2,||ϕ||1Ctα1,t>0,

    which is the required result (3.23). This finishes the proof of theorem.

    Remark 3.2. For 1<α<2, we have not give the optimal L2-norm estimates for the solution of Eq (1.1). The reason is as follows. We can apply the inverse Laplace transform to equality (2.1) to obtain the solution of Eq (1.1) in the Fourier domain

    ˆu(ω,t)=ˆGφ(ω,t)ˆφ(ω)+ˆGψ(ω,t)ˆψ(ω),

    where

    ˆGφ(ω,t)=tα2Eα,α1(|ω|2stα)

    and

    ˆGψ(ω,t)=tα1Eα,α(|ω|2stα).

    However, we can not derive positive lower bounds for the Mittag-Leffler functions Eα,α1(η) and Eα,α(η) with η0 and 1<α<2 as Lemma 3.7 since these two functions have zeros on the real axis. In fact, using the result of Theorem 2 in [29] we immediately see that the Mittag-Leffler function Eα,α1(η) exists real zeros. For the Mittag-Leffler function Eα,α(η), it is obvious that Eα,α(0)=1/Γ(α)>0. Additionally, In view of asymptotic expansion of the Mittag-Leffler function, see Theorem 1.4 in [28], we have

    Eα,α(η)=pk=1(η)kΓ(ααk)+O(η1p)=1Γ(α)η2+O(η3)

    when η. Since 1<α<2, then one gets Γ(α)>0. Hence the Mittag-Leffler function Eα,α(η) behaves like

    Eα,α(η)1η2,η,

    which implies Eα,α(η)<0 for sufficiently large η. Combining the above analysis and taking the analyticity of the Mittag-Leffler function Eα,α(η) into account, we find that Eα,α(η) has one real zeros at least. In Figure 1, we depict graphs of the Mittag-Leffler functions Eα,α1(η) and Eα,α(η) with η[0,50], where the parameter α takes α=1.2,1.5,1.9, respectively.

    Figure 1.  Plots of the Mittag-Leffler functions Eα,α1(η) and Eα,α(η).

    This work investigates asymptotic behaviors of solutions of Cauchy problems for superdiffusion equation (1.1) and subdiffusion equation (1.2) with integral initial conditions in the sense of Lp(Rd) and Lp,(Rd). For these two kinds of equations, we construct their fundamental solutions and solutions, analyze asymptotic behaviors of solutions, and study gradient estimates and large time behaviors. In particular, the optimal L2 decay estimate of solution is derived for Eq (1.2). Compared with the cases of Caputo derivative in the time direction [7,18] in Eqs (1.1) and (1.2), it is not difficult to see that the asymptotic rates are faster in the cases of Riemann-Liouville derivative.

    The authors declare they have not used Artificial Intelligence (AI) tools in the creation of this article.

    The work was supported in part by the Scientific and Technological Innovation Programs of Higher Education Institutions in Shanxi (No. 2021L573), the Key Research and Development Project of Lvliang city (No. 2022RC11), the Fundamental Research Program of Shanxi Province (No. 202103021224317), the Innovation and Entrepreneurship Training Program of College Students for Higher Education Institutions in Shanxi Province (No. 20221257).

    The authors declare no conflicts of interest.

    In the appendix we recall several concepts of integral transforms [5] such as Laplace transform, Mellin transform and Fourier transform, which play an important role on the process of deriving the solutions of Eqs (1.1) and (1.2). Moreover, we introduce the definition of the Fox H-function too.

    The standard Laplace transform of a function f(t) is defined by

    ¯f(λ)=L[f(t),λ]:=0eλtf(t)dt,λC.

    Correspondingly, the inverse Laplace transform is given by

    f(t)=L1[¯f(λ),t]:=12πic+icieλt¯f(λ)dλ,c=Re(λ),t>0.

    In particular, the Laplace transforms of the Riemann-Liouville fractional integral (1.3) and derivative (1.4) with the starting point a=0 can be written as [28]

    L[RLDα0,tf(t),λ]=λαL[f(t),λ] (A1)

    and

    L[RLDα0,tf(t),λ]=λαL[f(t),λ]n1k=0λk[RLDαk10,tf(t)]|t=0 (A2)

    for n1<α<nN.

    The standard Mellin transform of a function f(t) is defined by

    ˜f(ξ)=M[f(t),ξ]:=0tξ1f(t)dt,ξC.

    The inverse Mellin transform is correspondingly represented as

    f(t)=M1[˜f(ξ),t]:=12πic+icitξ˜f(ξ)dξ,c=Re(ξ),t>0.

    The relation connecting the Laplace transform and the Mellin transform has the following form [6]:

    M[f(t),ξ]=1Γ(1ξ)M[L[f(t),λ],1ξ]. (A3)

    The Fourier transform of a function f(x) is defined by

    ˆf(ω)=F[f(x),ω]:=Rdeiωxf(x)dx,ωRd,

    while the corresponding inverse Fourier transform can be written as

    f(x)=F1[ˆf(ω),x]:=1(2π)dRdeixωˆf(ω)dω,xRd.

    Then the Fourier transform of the integral fractional Laplacian (1.5) is given by [8]

    F[(Δ)sv(x),ω]=|ω|2sF[v(x),ω],xRd,s(0,1). (A4)

    Next, let us briefly recall the definition of the Fox H-function. The Fox H-function are special functions, where the fundamental solutions of Eqs (1.1) and (1.2) are written by means of these functions, and they paly a basic role in asymptotic analysis of the solutions.

    According to the contour integral of Mellin-Barnes type, the Fox H-function can be represented as

    Hmnμν(z)Hmnμν(z|(a1,α1),,(an,αn);(an+1,αn+1),,(aμ,αμ)(b1,β1),,(bm,βm);(bm+1,βm+1),,(bν,βν)):=12πiLHmnμν(τ)zτdτ, (A5)

    where

    Hmnμν(τ):=mj=1Γ(bj+βjτ)nl=1Γ(1alαlτ)μl=n+1Γ(al+αlτ)νj=m+1Γ(1bjβjτ), (A6)

    and m,n,μ,ν are nonnegative integers with 0mν and 0nμ, and αl,βj are positive real numbers and al,bj are complex numbers for l=1,,μ; j=1,,ν. All the poles

    bjσ=bj+σβj,j=1,,m;σ=0,1,2,

    of the gamma functions Γ(bj+βjτ) and

    alk=1al+kαl,l=1,,n; k=0,1,2,

    of the gamma functions Γ(1alαlτ) are not equal, i.e.,

    αl(bj+σ)βj(alk1),j=1,,m; l=1,,n; σ,k=0,1,2,

    The contour L is an infinite contour in the complex plane which separates all the poles bjσ from all the poles alk, and it may take L=L or L=L+ or L=Liγ with γR and i2=1. Besides, we denote

    a=nl=1αlμl=n+1αl+mj=1βjνj=m+1βj.

    A comprehensive and detailed description for the Fox H-function can be available from [2,14,15,30].



    [1] E. Ahmed, A. M. A. El-Sayed, H. A. A. El-Saka, Equilibrium points, stability and numerical solutions of fractional-order predator-prey and rabies models, J. Math. Anal. Appl., 325 (2007), 542–553. https://doi.org/10.1016/j.jmaa.2006.01.087 doi: 10.1016/j.jmaa.2006.01.087
    [2] B. L. J. Braaksma, Asymptotic expansions and analytical continuations for a class of Barnes integrals, Compos. Math., 15 (1962), 239–341.
    [3] D. Baleanu, G. C. Wu, Some further results of the Laplace transform for variable-order fractional difference equations, Fract. Calc. Appl. Anal., 22 (2019), 1641–1654. https://doi.org/10.1515/fca-2019-0084 doi: 10.1515/fca-2019-0084
    [4] W. C. Chen, Nonlinear dynamics and chaos in a fractional-order financial system, Chaos Solitons Fract., 36 (2008), 1305–1314. https://doi.org/10.1016/j.chaos.2006.07.051 doi: 10.1016/j.chaos.2006.07.051
    [5] B. Davies, Integral transform and their applications, New York: Springer, 2002. https://doi.org/10.1007/978-1-4684-9283-5
    [6] J. S. Duan, Time- and space-fractional partial differential equations, J. Math. Phys., 46 (2005), 13504–13511. https://doi.org/10.1063/1.1819524 doi: 10.1063/1.1819524
    [7] J. D. Djida, A. Fernandez, I. Area, Well-posedness results for fractional semi-linear wave equations, Discrete Contin. Dyn. Syst. B, 25 (2020), 569–597. https://doi.org/10.3934/dcdsb.2019255 doi: 10.3934/dcdsb.2019255
    [8] E. Di Nezza, G. Palatucci, E. Valdinoci, Hitchhiker's guide to the fractional Sobolev spaces, Bull. Sci. Math., 136 (2012), 521–573. https://doi.org/10.1016/j.bulsci.2011.12.004 doi: 10.1016/j.bulsci.2011.12.004
    [9] K. Diethelm, V. Kiryakova, Y. Luchko, J. A. T. Machado, V. E. Tarasov, Trends, directions for further research, and some open problems of fractional calculus, Nonlinear Dyn., 107 (2022), 3245–3270. https://doi.org/10.1007/s11071-021-07158-9 doi: 10.1007/s11071-021-07158-9
    [10] S. D. Eidelman, A. N. Kochubei, Cauchy problem for fractional diffusion equations, J. Differ. Equ., 199 (2004), 221–255. https://doi.org/10.1016/j.jde.2003.12.002 doi: 10.1016/j.jde.2003.12.002
    [11] L. Grafakos, Classical and modern Fourier analysis, Pearson Education, 2004.
    [12] R. Hilfer, Applications of fractional calculus in physics, Singapore: World Scientific, 2000. https://doi.org/10.1142/3779
    [13] B. T. Jin, Fractional differential equations, Cham: Springer, 2021. https://doi.org/10.1007/978-3-030-76043-4
    [14] A. A. Kilbas, H. M. Srivastava, J. J. Trujillo, Theory and applications of fractional differential equations, Amsterdam: Elsevier Science, 2006.
    [15] A. A. Kilbas, M. Saigo, H-transforms: theory and applications, Boca Raton: CRC Press, 2004. https://doi.org/10.1201/9780203487372
    [16] K. H. Kim, S. Lim, Asymptotic behaviors of fundamental solution and its derivatives to fractional diffusion-wave equations, J. Korean Math. Soc., 53 (2016), 929–967. https://doi.org/10.4134/JKMS.j150343 doi: 10.4134/JKMS.j150343
    [17] J. Kemppainen, J. Siljander, V. Vergara, R. Zacher, Decay estimates for time-fractional and other non-local in time subdiffusion equations in Rd, Math. Ann., 366 (2016), 941–979. https://doi.org/10.1007/s00208-015-1356-z doi: 10.1007/s00208-015-1356-z
    [18] J. Kemppainen, J. Siljander, R. Zacher, Representation of solutions and large-time behavior for fully nonlocal diffusion equations, J. Differ. Equ., 263 (2017), 149–201. http://dx.doi.org/10.1016/j.jde.2017.02.030 doi: 10.1016/j.jde.2017.02.030
    [19] C. P. Li, M. Cai, Theory and numerical approximations of fractional integrals and derivatives, Philadelphia: SIAM, 2019. https://doi.org/10.1137/1.9781611975888
    [20] C. P. Li, Z. Q. Li, Asymptotic behaviors of solution to Caputo-Hadamard fractional partial differential equation with fractional Laplacian, Int. J. Comput. Math., 98 (2021), 305–339. https://doi.org/10.1080/00207160.2020.1744574 doi: 10.1080/00207160.2020.1744574
    [21] C. P. Li, Z. Q. Li, Asymptotic behaviors of solution to partial differential equation with Caputo-Hadamard derivative and fractional Laplacian: Hyperbolic case, Discrete Contin. Dyn. Syst. S, 14 (2021), 3659–3683. https://doi.org/10.3934/dcdss.2021023 doi: 10.3934/dcdss.2021023
    [22] Z. Q. Li, Asymptotics and large time behaviors of fractional evolution equations with temporal ψ-Caputo derivative, Math. Comput. Simul., 196 (2022), 210–231. https://doi.org/10.1016/j.matcom.2022.01.023 doi: 10.1016/j.matcom.2022.01.023
    [23] F. Mainardi, Fractional calculus and waves in linear viscoelasticity: an introduction to mathematical models, Singapore: Imperial College Press, 2010. https://doi.org/10.1142/p614
    [24] R. Metzler, J. Klafter, The random walk's guide to anomalous diffusion: a fractional dynamics approach, Phys. Rep., 339 (2000), 1–77. https://doi.org/10.1016/S0370-1573(00)00070-3 doi: 10.1016/S0370-1573(00)00070-3
    [25] Y. T. Ma, F. R. Zhang, C. P. Li, The asymptotics of the solutions to the anomalous diffusion equations, Comput. Math. Appl., 66 (2013), 682–692. https://doi.org/10.1016/j.camwa.2013.01.032 doi: 10.1016/j.camwa.2013.01.032
    [26] C. A. Monje, Y. Q. Chen, B. M. Vinagre, D. Y. Xue, V. Feliu, Fractional-order systems and controls: fundamentals and applications, London: Springer, 2010. https://doi.org/10.1007/978-1-84996-335-0
    [27] L. Ma, B. W. Wu, Finite-time stability of Hadamard fractional differential equations in weighted Banach spaces, Nonlinear Dyn., 107 (2022), 3749–3766. https://doi.org/10.1007/s11071-021-07138-z doi: 10.1007/s11071-021-07138-z
    [28] I. Podlubny, Fractional differential equations, Academic Press, 1999.
    [29] A. V. Pskhu, On the real zeros of functions of Mittag-Leffler type, Math. Notes, 77 (2005), 546–552. https://doi.org/10.1007/s11006-005-0054-7 doi: 10.1007/s11006-005-0054-7
    [30] H. M. Srivastava, K. C. Gupta, S. P. Goyal, The H-functions of one and two variables with applications, New Delhi: South Asian Publishers, 1982.
    [31] J. Sabatier, O. P. Agrawal, J. A. T. Machado, Advances in fractional calculus, Dordrecht: Springer, 2007. https://doi.org/10.1007/978-1-4020-6042-7
    [32] V. Vergara, R. Zacher, Optimal decay estimates for time-fractional and other nonlocal subdiffusion equations via energy methods, SIAM J. Math. Anal., 47 (2015), 210–239. https://doi.org/10.1137/130941900 doi: 10.1137/130941900
    [33] A. Yacine, L. Ma, On criteria of existence for nonlinear Katugampola fractional differential equations with p-Laplacian operator, Fract. Differ. Calc., 11 (2021), 51–68. https://doi.org/10.7153/fdc-2021-11-04 doi: 10.7153/fdc-2021-11-04
  • This article has been cited by:

    1. Zhiqiang Li, Asymptotic analysis of solutions to fractional diffusion equations with the Hilfer derivative, 2025, 44, 2238-3603, 10.1007/s40314-025-03105-1
  • Reader Comments
  • © 2023 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(1447) PDF downloads(55) Cited by(1)

Figures and Tables

Figures(1)

Other Articles By Authors

/

DownLoad:  Full-Size Img  PowerPoint
Return
Return

Catalog