Higher order numerical schemes for SPDEs with additive Noise

Abhishek Chaudhary, Andreas Prohl · 2025

Read Higher order numerical schemes for SPDEs with additive Noise by Abhishek Chaudhary, Andreas Prohl (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2510-23210.txt.

Introduction

Abhishek Chaudhary and Andreas Prohl present higher-order numerical schemes for stochastic partial differential equations with additive noise. Standard Euler schemes converge slowly for such equations because noise is rough in time, limiting rates to between one half and one. They study linear stochastic heat and wave equations with fixed boundary values.

For the heat equation, a modified Crank-Nicolson scheme with a proper quadrature for the noise term reaches a strong rate of three halves. A matching approach gives order two for the wave equation. Faster schemes let engineers simulate noisy heat and vibration with far fewer time steps.

Catalog record

CREATORAbhishek Chaudhary, Andreas Prohl
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE50 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2510-23210

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Higher order numerical schemes for SPDEs with additive Noise Abhishek Chaudhary, Andreas Prohl Abstract We present high-order numerical schemes for linear stochastic heat and wave equations with Dirichlet boundary conditions, driven by additive noise. Standard Euler schemes for SPDEs are limited to an order convergence between 1 / 2 1/2 and 1 1 due to the low temporal regularity of noise. For the stochastic heat equation, a modified Crank-Nicolson scheme with proper numerical quadrature rule for the noise term in its reformulation as random PDE achieves a strong convergence rate of 3 / 2 3/2 . For the stochastic wave equation with additive noise a corresponding approach leads to a scheme which is of order 2. Mathematisches Institut Universität Tübingen, Auf der Morgenstelle 10, 72076 Tübingen, Germany [email protected], [email protected] Keywords: stochastic wave equation; stochastic heat equation; additive noise; Wiener process; Crank-Nicolson scheme; Euler-Maruyama scheme. 1 Introduction Stochastic partial differential equations (SPDEs) provide robust mathematical frameworks for modeling complex systems influenced by randomness across fields such as physics, engineering, finance, and biology. These equations describe phenomena like heat diffusion under random forcing, wave propagation with stochastic perturbations, financial models with spatial correlations, and biological processes affected by environmental noise. In this paper, we focus on two fundamental SPDEs: the linear parabolic stochastic heat equation and the stochastic wave equation, both driven by additive noise and defined on a smooth bounded domain D\subset\mathbb{R}^{d} with Dirichlet boundary conditions. A fundamental challenge in the numerical analysis of SPDEs stems from the limited temporal regularity of the driving Wiener process which is only Hölder continuous of order at most 1/2 in time. This inherent roughness directly constrains the strong convergence rates attainable by a standard numerical analysis of temporal discretizations — as is shortly evidenced in Remark 1.1 . For instance, typical error analyses for the classical Euler–Maruyama scheme for parabolic SPDEs typically lead to a strong order of convergence of at most 1/2 . An improvement is obtained in [ 1 , 2 ] , where an order of convergence 1 is established for an implicit Euler discretization of SPDEs with colored additive noise. This result is achieved by employing a transformation technique that reformulates the original SPDE into a random PDE. In this work, we use this idea as an algorithmic tool in first line to develop high-order numerical schemes for both, the stochastic heat and wave equations driven by additive noise. 1.1 Stochastic Heat Equation Let D\subset\mathbb{R}^{d} be smooth, bounded domain, and T>0 . We consider the stochastic heat equation driven by additive noise: \begin{cases}\mathrm{d}X(t)+AX(t)\,\mathrm{d}t=\Phi\,\mathrm{d}W(t),&\text{in}\,(0,T]\times D,\\[4.0pt] X=0,&\text{on}\,(0,T]\times\partial D,\\[4.0pt] X(0)=X_{0},&\text{in}\,D,\end{cases} (1.1) where W is an \mathbb{R}^{m} -valued Wiener process, the initial datum X_{0}:D\to\mathbb{R} , \Phi:D\to\mathbb{R}^{m} is the additive noise coefficient, and A:\mathbb{H}^{2}(D)\cap\mathbb{H}_{0}^{1}(D)\to\mathbb{L}^{2}(D) is a second-order linear uniformly elliptic operator of the form \displaystyle A=\sum_{i,j=1}^{d}a_{i,j}(x)\partial_{x_{i},x_{j}}+\sum_{i=1}^{d}b_{i}(x)\partial_{x_{i}}+c(x), (1.2) with smooth enough coefficients a_{i,j},b_{i},c for all i,j=1,\dots,d ; see, e.g. [ 9 , Sec. 3.1] . A standard approach to time-discretization of SPDE ( 1.1 ) is the implicit Euler-Maruyama scheme: \begin{cases}\displaystyle X_{j+1}-X_{j}+\tau AX_{j+1}=\Phi\Delta_{j+1}W,&j=0,\dots,N-1,\\[4.0pt] X(0)=X_{0},\\[4.0pt] X_{j}\big|_{\partial D}=0,\qquad j=1,\dots,N,\end{cases} (1.3) with a uniform time grid t_{j}=j\tau , j=0,1,\dots,N , \tau=\frac{T}{N} and the Wiener increment \Delta_{j+1}W=W(t_{j+1})-W(t_{j}) . Due to the low temporal regularity of the noise, typical proofs of strong convergence for ( 1.3 ) show a rate of 1/2 , see e.g. [ 11 ] , a limited order result that is shortly evidenced in the following remark. Remark 1.1 . Let e_{j}=X(t_{j})-X_{j} , where \{X_{j};1\leq j\leq N\} solves the scheme ( 1.3 ). In integrated form on [t_{j},t_{j+1}] , the first equation in ( 1.1 ) reads as X(t_{j+1})-X(t_{j})+\int_{t_{j}}^{t_{j+1}}AX(s)\,{\rm d}s=\Phi\Delta_{j+1}W. Taking differences of this equation and ( 1.3 ) amounts to \displaystyle e_{j+1}-e_{j}+\int_{t_{j}}^{t_{j+1}}A\bigl[X(s)-X_{j+1}\bigr]\,{\rm d}s=0. (1.4) The order limiting term is now the integrand, since the solution \{X(t);t\in[0,T]\} to ( 1.1 ) is known to be Hölder continuous in time up to 1/2 . Different ansatzes to ours given below were considered in the last decades to construct a scheme of (improved) strong order 1 , which employs the explicit knowledge of the related semigroup for ( 1.2 ) to approximately solve ( 1.1 ). A particular example where eigenvalues and -functions are explicitly known is A=-\Delta on the torus to use the related semigroup for an algorithm; see e.g. [ 5 ] and the cited references there, as well as how to herewith set up the exponential Euler method in this context to settle strong order 1 . Despite its elegant analysis, such algorithms are of limited use in practice since their related characterizations are not available for general A in ( 1.2 ), acting on general domains D . We therefore here aim to construct implementable higher-order methods that may be used to approximate the solution of ( 1.1 ) for general data settings — i.e. , general A from ( 1.2 ), as well as general smoothly bounded domains D . To accomplish this goal, we take motivation from the works [ 1 , 2 ] , and consider instead of the solution X from ( 1.1 ) the transformation \displaystyle u=X-\Phi W, (1.5) and therefore convert the SPDE ( 1.1 ) into a random PDE: \begin{cases}u_{t}(t)+Au(t)=-A\bigl[\Phi W(t)\bigr],&\text{in}\,(0,T]\times D,\\[4.0pt] u=0,&\text{on}\,(0,T]\times\partial D,\\[4.0pt] u(0)=X_{0},&\text{in}\,D,\end{cases} (1.6) where u_{t}:=\partial_{t}u . It is known that the solution u of ( 1.6 ) now does possess temporal derivatives; in fact, it can be shown that u_{t}\in\mathbb{L}_{\mathbb{F}}^{2}(\Omega;\mathbb{L}^{2}([0,T];\mathbb{L}^{2}(D)) , provided X_{0}\in\mathbb{H}_{0}^{1}(D) and \Phi\in(\mathbb{H}^{2}(D)\cap\mathbb{H}_{0}^{1}(D))^{m} ; see ( 3.3 ). This is the guiding motivation to verify order 1 for ( 1.3 ) in the context of more regular data X_{0}\in\mathbb{H}_{0}^{1}(D)\cap\mathbb{H}^{2}(D) and \Phi\in(\mathbb{H}_{0}^{1}(D)\cap\mathbb{H}^{2}(D))^{m} with compatibility condition A\Phi\in(\mathbb{H}_{0}^{1}(D))^{m} . To verify this result in [ 1 , 2 ] — and instead of starting with ( 1.1 ) directly — we consider first implicit Euler scheme for ( 1.6 ) which reads as \begin{cases}\displaystyle U_{j+1}-U_{j}+\tau AU_{j+1}=-\tau A\bigl[\Phi W(t_{j+1})\bigr],&j=0,\dots,N-1,\\[8.0pt] U^{0}=X_{0},\\[4.0pt] U_{j}\big|_{\partial D}=0,\qquad j=1,\dots,N,\end{cases} (1.7) and exploit improved the temporal regularity for the solution u of ( 1.6 ) to verify \displaystyle\sup_{1\leq j\leq N}\mathbb{E}\left[\|u(t_{j})-U_{j}\|_{\mathbb{L}^{2}(D)}^{2}\right]\leq C\tau^{2}. (1.8) Afterwards, we revert the transformation procedure on a discrete level, by defining X^{j}:=U_{j}+\Phi W(t_{j}) as approximate for X(t_{j}) from ( 1.1 ). In fact, it can be immediately seen that X^{j}=X_{j} , such that we have \displaystyle\sup_{1\leq j\leq N}\mathbb{E}\left[\|X(t_{j})-X_{j}\|_{\mathbb{L}^{2}(D)}^{2}\right]=\sup_{1\leq j\leq N}\mathbb{E}\bigl[\|u(t_{j})-U_{j}\|^{2}_{\mathbb{L}^{2}(D)}\bigr]\leq C\tau^{2}. (1.9) This convergence proof of rate 1 for implicit scheme ( 1.3 ) via ( 1.6 ) –and ( 1.7 ) – in the presence of regular data X_{0} and \Phi nourishes the hope to construct even higher order methods since the time regularity of the solution u to random PDE ( 1.6 ) is even higher, i.e. , \displaystyle u\in C^{1,{1}/{2}}\big([0,T];\mathbb{L}^{2}(\Omega;\mathbb{H}_{0}^{1}(D))\big), (1.10) if data X_{0}\in\mathbb{H}^{4}(D)\cap\mathbb{H}_{0}^{1}(D) and \Phi\in\mathbb{H}^{3}(D)\cap\mathbb{H}_{0}^{1}(D) with compatibility conditions AX_{0}\in\mathbb{H}_{0}^{1}(D) and A\Phi\in(\mathbb{H}_{0}^{1}(D))^{m} ; see ( 3.4 ). This result for the solution u of ( 1.6 ) again rests on regularity theory for deterministic PDEs with temporal Hölder regularity of the right-hand side; see [ 9 , Thm. 4.3.1] . For this purpose, we propose the new modified Crank–Nicolson Scheme 3.1 for SPDE ( 1.1 ) which is a strong convergence rate of 3/2 ; see Theorem 3.1 . 1.2 Stochastic wave equation Let D\subset\mathbb{R}^{d} be smooth, bounded domain, and T>0 . We consider the stochastic wave equation driven by additive noise: \begin{cases}\displaystyle{\rm d}X_{t}(t)+AX(t)\,{\rm d}t=\Phi\,{\rm d}W(t),&\text{in}\,(0,T]\times D,\\[6.0pt] X(0)=X_{0},\quad

Read the complete text