$L^{p}$-strong convergence orders of fully discrete schemes for the SPDE driven by Lévy noise

Stochastic differential equations (SDEs) with Lévy noise are widely used to model sudden events and irregular changes in stochastic phenomenon, such as financial market crashes, abrupt phase transitions, and neural spiking patterns; see e.g. [3, 9, 12] and references therein. In numerical studies of SDEs, the $L^{p}$-strong convergence analysis of numerical methods has received much attention, as it assesses the accuracy of simulating the paths of the underlying solution process. The $L^{p}$-strong convergence analysis for SDEs with Lévy noise is different from that of the Gaussian noise case (see e.g. [5, 10]). For instance, there is an extra term appearing in the Burkholder–Davis–Gundy inequality of the stochastic integral with respect to the compensated Poisson measure compared with that in the Gaussian noise case. As a result, the temporal Hölder continuity in $L^{p}$ sense of the exact solution to an SDE with Lévy noise does not exceed $1/p$, resulting in the deterioration of the $L^{p}$-strong convergence order of a numerical scheme for large $p$ if one uses the temporal Hölder continuity directly. A natural question arises:

• Q:

  Can one obtain the $L^{p}$-strong convergence order of a numerical scheme for the SDE with Lévy noise that does not depend on $p$?

A remarkable progress for this question has been made recently for the case of the stochastic ordinary differential equation (SODE) with Lévy noise: authors in [1] studied the $L^{p}$-strong convergence order of the Euler–Maruyama scheme for a multidimensional SODE with irregular Hölder drift, driven by additive Lévy process with exponent $\alpha\in(0,2]$, and the obtained convergence order does not depend on $p$. For the case of SPDEs, to the best of our knowledge, this question is still open so far. The aim of this paper is to provide a positive answer for fully discrete schemes of the SPDE driven by Lévy noise.

We consider the following SPDE driven by Lévy noise:

where $T>0$ and $A:=\Delta:\mbox{Dom}(A)\subset H\to H$ is the Laplacian with homogeneous Dirichlet boundary condition. Here $H:=L^{2}(0,1)$ with usual inner product $\langle\cdot,\cdot\rangle$ and norm $\|\cdot\|$, and $\chi:=H\backslash\{0\}$ is the mark set. The process $\{W(t)\}_{t\in[0,T]}$ is an $H$-valued Gaussian space-time white noise defined on a complete filtered probability space $(\Omega,\mathcal{F},(\mathcal{F}_{t})_{t\geq 0},\mathbb{P})$ that satisfies usual conditions. Let $\tilde{N}(dz,dt):=N(dz,dt)-\nu(dz)dt$ stand for a compensated Poisson random measure, where $\{N(\cdot,t)\}_{t\in[0,T]}$ is the Poisson random measure and $\nu$ is a Lévy measure on Borel $\sigma$-algebra $\mathcal{B}(\chi)$ satisfying $\nu(\{0\})=0$ and $\int_{\chi}\min\{1,\|z\|^{2}\}\nu(dz)<\infty.$ Here $\{W(t)\}_{t\in[0,T]}$ and $\{\tilde{N}(\cdot,t)\}_{t\in[0,T]}$ are supposed to be independent. Precise assumptions on coefficients $F$ and $G$ will be given in Section 2.1. For the convergence analysis of numerical methods for  (1), we refer interested readers to [2] for the $L^{2}$-strong convergence and to [6] for the $L^{p}$-strong convergence with $p\leq 2$.

We apply the spectral Galerkin method to approximate (1) in space, and further use the Euler-type methods in time to obtain the fully discrete schemes. The corresponding mapping for the one-step approximation is given as follows:

where $n_{T}$ is the subscript corresponding to the time node $T$ (i.e., $t_{n_{T}}=T$), and we mention that the time step size $\Delta t_{i+1}:=t_{i+1}-t_{i}$ may be path-dependent. The main contribution of our work is that, we show for the first time that proposed schemes converge in $L^{p}$ sense with orders almost $1/2$ in both space and time for all $p\geq 2$. Indeed, the $L^{p}$-strong convergence analysis in time is more technical compared with that in space. This is related to the aforementioned issue yielding poor order for large $p$. To overcome this order barrier, we present two different strategies for the cases of the linear multiplicative Poisson noise with $\nu(\chi)<\infty$ and the additive Poisson noise with $\nu(\chi)\leq\infty$, respectively. More precisely,

• (i)

  Case of the linear multiplicative Poisson noise with $\nu(\chi)<\infty$. In this case, there are a.s. finitely many jumps. We introduce the jump-adapted time discretization by a superposition of finitely many jump times to a deterministic equidistant grid. Then the jump-adapted fully discrete scheme is constructed as follows:

  $\displaystyle X^{N}_{i+1-}=\Phi(X^{N}_{i},\Delta t_{i+1}),\quad X^{N}_{i+1}=X^{N}_{i+1-}+G_{N}(X^{N}_{i+1-},p_{t_{i+1}})\mathbf{1}_{\{p_{t_{i+1}}\neq 0\}}.$

  (2)

  Here $\{p_{t_{i}}\}_{i=0,1,...,n_{T}}$ is an $\mathcal{F}_{t_{i}}$-adapted Poisson point process with the intensity measure $\nu(dz)dt$. In the convergence analysis of this scheme, there are two critical steps to overcome the order barrier. First, this scheme evolves without jump between time grid points, which implies that the stochastic integral with respect to the compensated Poisson random measure in (2) can be transformed into the integral with respect to the intensity measure. Second, it requires establishing the $p$th moment estimates on the terms involving the multiplicative noise or the jump process when accumulating the local errors into the global one.

• (ii)

  Case of the additive Poisson noise with $\nu(\chi)\leq\infty$. In this case, there may be a.s. infinitely many jumps, in which the jump-adapted strategy is invalid. We consider the fully discrete scheme $X^{N}_{i+1}=\Phi(X^{N}_{i},\Delta t)$ with constant step size $\Delta t$. In the $L^{p}$-strong convergence analysis of this scheme, new strategy is required to deal with the terms involving the temporal Hölder continuity of the stochastic convolution with respect to the compensated Poisson random measure of (1). Our idea is to bound the $L^{p}$-norm of these terms by moment estimates of the nested conditional “$L^{1}$-norm” based on the recently obtained Lê’s quantitative John–Nirenberg inequality. To this end, we carefully analyze some essential a priori estimates including the nested conditional “$L^{2}$-norms” of both the stochastic convolutions and the solution of the perturbed stochastic equation, which help us to obtain the rigorous bound of the nested conditional “$L^{1}$-norm” and then the desired convergence orders.

The rest of the paper is organized as follows. The preliminaries, including notations, assumptions, and the well-posedness of the exact solution are given in Section 2.1. Section 2.2 is devoted to the introduction of the fully discrete schemes and their $L^{p}$-strong convergence order results. In Section 3, we present the $L^{p}$-strong convergence analysis of the fully discrete scheme for (1) in the case of the linear multiplicative Poisson noise with $\nu(\chi)<\infty$. In Section 4, we present the corresponding $L^{p}$-strong convergence analysis for the case of the additive Poisson noise with $\nu(\chi)\leq\infty$. Proofs of some essential propositions used in the convergence analysis are given in Section 5.

In this section, we first give some preliminaries, including notations, assumptions, the well-posedness of the exact solution and some useful inequalities. Second, we present the fully discrete schemes of (1) for two different cases of noises, and respectively show their $L^{p}$-strong convergence orders, as the main results of this paper.

In this section, we present the $L^{p}$-strong convergence analysis of fully discrete scheme (7) for (1) driven by the linear multiplicative Poisson noise with finite Lévy measure. Note that the stochastic integral with respect to the compensated Poisson random measure in (7) can be transformed into the integral with respect to the intensity measure, as shown in (2.2). This transformation is crucial to overcome the order barrier caused by the $p$th moment estimates of the stochastic integral with respect to the compensated Poisson random measure.