Interpolating the stochastic heat and wave equations with time-independent noise: solvability and exact asymptotics (Q6078573)

From MaRDI portal
scientific article; zbMATH DE number 7742938
Language Label Description Also known as
English
Interpolating the stochastic heat and wave equations with time-independent noise: solvability and exact asymptotics
scientific article; zbMATH DE number 7742938

    Statements

    Interpolating the stochastic heat and wave equations with time-independent noise: solvability and exact asymptotics (English)
    0 references
    0 references
    0 references
    27 September 2023
    0 references
    This paper treats a class of stochastic partial differential equations (SPDEs) with fractional differential operators subject to some time-independent multiplicative Gaussian noise. The authors succeed in deriving sharp conditions, under which a unique global \(L^p(\Omega)\)-solution exists for all \(p \geq 2\), and also in obtaining exact moment asymptotics. Especially when there exists only a local solution, they determine the precise deterministic time \(T_2\), before which a unique \(L^2(\Omega)\)-solution exists, but after which the series corresponding to the \(L^2(\Omega)\) moment of the solution blows up. By properly choosing the parameters, results in this paper interpolate the known results for both stochastic heat and wave equations. More precisely, the authors study the following SPDE with fractional differential operators: \[ \left\{ \begin{aligned} & \left( \partial_t^b + \frac{\nu}{2} ( - \varDelta )^{ a/2} \right) u(t,x) = I_t^r [ \sqrt{ \theta} u(t,x) \dot{W}( x) ], \qquad x \in \mathbb{R}^d, \quad t > 0, \\ & u(0, \cdot) = 1, \qquad \qquad \qquad \qquad \qquad \qquad \qquad b\in (0, 1], \\ & u(0, \cdot) = 1, \qquad \partial_t u(0, \cdot) = 0, \qquad \qquad \qquad b \in (1, 2), \end{aligned} \right.\tag{1} \] where \(a \in (0, 2]\), \(b \in (0,2)\), \(r \geq 0\), \(\nu > 0\) and \(\theta > 0\). Here \(W = \{ W(\phi): \phi \in \mathcal{D} ( \mathbb{R}^d ) \}\) is a centered and time-independent Gaussian noise, with mean zero and covariance \[ \mathbb{E} [ W( \phi ) W (\psi ) ] = \int_{ \mathbb{R}^d } \mathcal{F} \phi(\xi) \overline{ \mathcal{F} \psi(\xi)} \mu( d \xi) =: \langle \phi, \psi \rangle_{ \mathcal{H} }, \tag{2} \] where the spectral measure \(\mu\) is a nonnegative and nonnegative definite tempered measure on \(\mathbb{R}^d\), the Fourier transform of a test function \(\phi\) is given by \[ \mathcal{F} \phi(\xi) = \int_{ \mathbb{R}^d } \exp ( - i x \xi ) \phi(x) dx, \] \(( - \varDelta)^{a/2}\) is the fractional Laplacian of order \(a\), \(\partial_t^b\) denotes the Caputo fractional differential operator, and the Riemann-Liouville fractional integral of order \(r > 0\) is defined by \[ I_t^r f(t) := \frac{1}{ \Gamma(r) } \int_0^t (t - s)^{r-1} f(s) ds, \qquad \text{for} \quad t > 0.\tag{3} \] The fundamental solution to (1) is expressed explicitly in terms of the Fox \(H\)-function \(H_{r, q}^{m,n} (z)\). When we denote the fundamental solution as \(G(t,x)\) \(:=\) \(G_{a,b,r,\nu,d} (t,x)\), then in fact it is given by \[ G(t,x) = \pi^{- d/2} \vert x \vert^{-d} t^{b +r-1} \cdot H_{2,3}^{2,1} \left( \frac{ \vert x \vert^a }{ 2^{ a-1} \nu t^b } \left\vert \begin{aligned} & (1,1), (b + r, b) \\ & (d / 2, a/2), (1,1), (1, a/2) \end{aligned} \right. \right),\tag{4} \] (see [\textit{L. Chen} et al., Stochastic Processes Appl. 129, No. 12, 5073--5112 (2019; Zbl 1427.60119)] for more details). So that, (1) can be written as the following stochastic integral equation: \[ u(t,x) = 1 + \sqrt{ \theta} \int_0^t \left( \int_{ \mathbb{R}^d } G(t-s, x-y) u(s, y) W( \delta y) \right) ds, \tag{5} \] where the stochastic integral is in the Skorohod sense. We need the following assumptions. Assumption 1. (Nonnegativity) The fundamental solution \(G(t,x)\) is assumed to be nonnegative for all \(t > 0\) and \(x \in \mathbb{R}^d\). Assumption 2. (Noise) Let \(k \in \{ 1, \dots , d \}\) and partition the \(d\)-coordinates of \(x\) \(=\) \(( x_1\), \(\dots\), \(x_d)\) into \(k\) distinct groups of size \(d_i\) so that \(d_1 +\) \(\cdots\) \(+ d_n\) \(=\) \(d\). Denote \(x_{(i)} =\) \(( x_{ i_1}\), \(\dots\), \(x_{ i_{d_i} } )\) to be the coordinates in the \(i^th\) partition. Assume that the correlation function of the Gaussian noise is given by \[ \gamma(x) = \prod_{i=1}^k \left\vert x_{ ( i) } \right\vert^{- \alpha_i} \qquad \text{with} \quad \alpha_i \in ( 0, d_i).\tag{6} \] Define \(\alpha := \sum_{i=1}^k \alpha_i\). \par Here is the first main theorem. The first result deals with the well-posedness of the SPDE (1) ( or (5) ) as stated in the following theorem. Then we need to introduce the vatiational constant: \[ \mathcal{M}_{a,d} (t) := \sup_{ g \in \mathcal{F}_a } \left\{ \langle g^2 * g^2, f \rangle_{ L^2( \mathbb{R}^d)}^{1/2} - \frac{1}{2} \mathcal{E}_a( g, g) \right\}, \tag{7} \] with \[ \mathcal{E}_a (g, g) := ( 2 \pi)^{-d} \int_{ \mathbb{R}^d } \vert \xi \vert^a \vert \mathcal{F} g(\xi ) \vert^2 d \xi\tag{8} \] and \[ \mathcal{F}_a := \left\{ f \in L^2( \mathbb{R}^d ) : \,\, \Vert f \Vert_{ L^2(\mathbb{R}^d )} = 1, \mathcal{E}_a(f,f) < \infty \right\}.\tag{9} \] Note that we use the simple notation \(\mathcal{M}_a(f) := \mathcal{M}_{a,d}(f)\) when the dimension \(d\) is clear from the context, and \(\mathcal{M}_a := \mathcal{M}_a( \gamma )\) with \(\gamma\) defined in the above Assumption 2. Theorem 1. (Solvability) (a) Eq.(1) has a unique global solution \(u(t,x)\) in \(L^p(\Omega)\) for all \(p \geq 2\), \(t > 0\), and \(x \in \mathbb{R}^d\) provided that \[ 0 < \alpha < \min \left( \frac{a}{b} [ 2 (b+r) -1 ], 2a, d \right). \tag{10} \] (b) Otherwise, if \(r \in [0, 1/2)\) and \[ 0 < \alpha = \frac{a}{b} [ 2(b+r) -1 ] \leq d,\tag{11} \] then (1) has a local solution in the sense that (b-i) For any \(p \geq 2\), Eq.(1) has a unique solution \(u(t,x)\) in \(L^p(\Omega)\) for all \(p \geq 2\) and \(x \in \mathbb{R}^d\), but only for \(t \in (0, T_p)\), where \[ T_p := \frac{ \nu^{\alpha/a} }{ 2 \theta (p-1) \mathcal{M}_a^{ (2a - \alpha)/\alpha } }. \tag{12} \] (b-ii) For any \(t > T_2\), the series \[ \mathbb{E} [ u(t,x)^2 ] = \sum_{ h \geq 0} \theta^n n ! \Vert \tilde{f}_n (\cdot, x,t) \Vert_{ \mathcal{H}^{ \otimes n} }^2 \qquad \text{for all} \quad (t,x) \in (0,T) \times \mathbb{R}^d \tag{13} \] diverges, that is, the \(L^2(\Omega)\)-solution \(u(t,x)\) to (1) does not exist whenever \(t > T_2\). Here \(\tilde{f}_n\) is the symmetrization of \(f_n\) \(=\) \(f_n( \cdot, x, t)\) in the Wiener chaos expansion, cf. [\textit{R. M. Balan} and \textit{J. Song}, ALEA, Lat. Am. J. Probab. Math. Stat. 14, No. 2, 799--849 (2017; Zbl 1386.60210)]. The second main result of the paper is about the moment asymptotics. Theorem 2. We have \[ \begin{aligned} \lim_{ t_p \to \infty} t_p^{- \beta} \log \Vert u(t,x) \Vert_p &= \left( \frac{1}{2} \right) \left( \frac{2a}{ 2a ( b+r) - b \alpha } \right)^{\beta} \left( \theta \nu^{- \alpha/a} \mathcal{M}_a^{ (2a - \alpha)/a} \right)^{ \Xi(a,b,r)} \cdot \\ & \qquad \qquad \qquad \qquad \qquad \times \left( 2 (b + r) - \frac{b \alpha}{a} - 1 \right), \end{aligned} \tag{14} \] where \[ \beta := \frac{ 2 ( b + r ) - \frac{b \alpha}{a} }{ 2 ( b + r ) - \frac{b \alpha}{a} - 1}, \qquad \quad t_p := (p -1)^{1 - 1/ \beta} t, \qquad \qquad \text{and} \] \[ \Xi(a,b,r) := \frac{a}{ 2a (b + r) - b \alpha - a }. \] \par Lastly the authors list several examples for the asymptotics when global solution exist. In particular, they show that the asymptotics in (14) interpolates the corresponding results for both stochastic wave and heat equations. For other related works, see e.g. [\textit{L. Chen}, Trans. Am. Math. Soc. 369, No. 12, 8497--8535 (2017; Zbl 1406.60093)] for nonlinear stochastic time-fractional diffusion equations: moments and regularity, [\textit{X. Chen}, Ann. Inst. Henri Poincaré, Probab. Stat. 53, No. 2, 819--841 (2017; Zbl 1386.60214)] for moment asymptotics for parabolic Anderson equation with fractional time-space noise, and [\textit{X. Chen} et al., Ann. Inst. Henri Poincaré, Probab. Stat. 51, No. 4, 1529--1561 (2015; Zbl 1337.60201)] for exponential asymptotics for time-space Hamiltonians.
    0 references
    0 references
    stochastic partial differential equation
    0 references
    Caputo derivative
    0 references
    fractional integral
    0 references
    fractional Laplacian
    0 references
    Skorohod integral
    0 references
    Malliavin calculus
    0 references
    time-independent Gaussian noise
    0 references
    stochastic heat equations
    0 references
    stochastic wave equations
    0 references
    0 references

    Identifiers

    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references