Lower bounds for weak approximation errors for spatial spectral Galerkin approximations of stochastic wave equations

Ladislas Jacobe de Naurois, Arnulf Jentzen, Timo Welti

Abstract

While existence and uniqueness results for semilinear stochastic wave equations (SWEs) are well-established, explicit solutions are rarely available. Consequently, numerical approximation methods are essential for practical applications. This necessitates a rigorous analysis of convergence rates. While previous research established upper bounds for weak errors in spatial spectral Galerkin approximations, this work provides the corresponding lower bounds. These findings demonstrate that the established upper bounds are essentially sharp, indicating that the current theoretical convergence rates cannot be significantly improved within the general framework of spectral Galerkin methods.

Introduction

Stochastic wave equations (SWEs) serve as a fundamental mathematical framework for modeling physical systems subject to random perturbations, such as the vibration of strings in a turbulent environment or the propagation of electromagnetic waves through fluctuating media. Given the complexity of these equations, closed-form solutions are seldom attainable, making numerical approximation schemes—specifically spatial spectral Galerkin methods—indispensable for researchers and engineers.

In the field of numerical analysis, evaluating the efficiency of an approximation involves two primary metrics:

  1. Strong Convergence: Measures the proximity of a specific numerical path to the actual stochastic path.
  2. Weak Convergence: Measures how accurately the numerical method captures the expected value (the average behavior) of the solution.

While recent literature has successfully established upper bounds for weak convergence errors, a fundamental question remained: are these bounds tight? If the upper bound suggests a specific error rate, but the actual error is significantly lower, the analysis is considered "loose." This paper closes that gap by deriving lower bounds, proving that the established rates are the best possible for this class of problems.

Methodology

The research focuses on semilinear stochastic wave equations, employing a rigorous mathematical approach to determine the fundamental limits of approximation accuracy. To prove that the error cannot be reduced beyond a certain point, the authors construct "worst-case" scenarios using stochastic wave equations characterized by additive noise and a lack of drift nonlinearity.

Spatial Spectral Galerkin Approximation

The core method analyzed is the spectral Galerkin approximation, which involves projecting an infinite-dimensional stochastic process onto a finite-dimensional subspace. This subspace is defined by the eigenvalues ($\lambda_n$) of a linear operator, typically the Laplacian. By reducing the dimensionality, the equation becomes computationally tractable.

Analytical Tools

To derive the lower bounds, the authors utilized several key mathematical techniques:

  • Itô’s Isometry: This was instrumental in calculating the variance and covariance of the stochastic integrals that constitute the solution.
  • Gaussian Analysis: In the simplified models featuring additive noise, the solutions behave as Gaussian random variables. This allows for the use of exact formulas when calculating the expectations of test functions.
  • Test Function Evaluation: The error was assessed using specific classes of regular test functions, such as squared norms and exponential functions, to determine how effectively the Galerkin scheme captures the underlying distribution of the solution.

Key Insights: The Sharpness of Convergence Rates

The most significant contribution of this work is the proof that convergence rates for these approximations are "essentially sharp." This implies that the mathematical community has reached a fundamental limit regarding the accuracy of spatial spectral Galerkin methods for SWEs.

The Role of Additive Noise

The authors demonstrate that even in the most straightforward cases—linear equations with additive noise—the approximation error hits a theoretical floor. Because semilinear (more complex) equations encompass these linear cases, the general error for the complex cases cannot be lower than the floor established for the simple ones.

Spectral Decay and Error

The error is intrinsically linked to the eigenvalues ($\lambda_n$) of the spatial operator. The rate at which these eigenvalues grow dictates the speed of convergence. This paper provides a precise quantification of this relationship, showing that the spatial resolution of the Galerkin scheme is the primary bottleneck for weak convergence.

Main Results

The findings are crystallized in Theorem 1.1, which effectively "sandwiches" the approximation error between a lower and an upper bound:

  • Lower Bound: $c \cdot (\lambda_N)^{-\eta}$
  • Upper Bound: $C_\epsilon \cdot (\lambda_N)^{\epsilon-\eta}$

In these expressions, $\lambda_N$ represents the $N$-th eigenvalue and $\eta$ is a parameter determined by the regularity of the noise. Because these bounds differ only by an arbitrarily small $\epsilon$, the convergence rate is considered tight. For a 1D stochastic wave equation using a Laplacian operator, these results confirm that the weak convergence rate is strictly limited by the spatial resolution of the chosen numerical scheme.

Limitations and Future Directions

While this paper provides a definitive answer for spectral Galerkin methods, it also highlights areas for further exploration:

  • Method Specificity: The results are tailored to spectral Galerkin approximations. While the logic likely extends to finite element or finite difference methods, explicit lower bounds for those techniques remain to be formally proven.
  • Test Function Sensitivity: The lower bounds were established using "regular" test functions. It remains possible that non-smooth test functions might exhibit different error characteristics.
  • Noise Complexity: Future research could extend these techniques to equations involving more complex noise profiles, such as Lévy noise, or investigate whether temporal discretization (time-stepping) introduces similar fundamental limits.

Summary updated

Abstract

Although for a number of semilinear stochastic wave equations existence and uniqueness results for corresponding solution processes are known from the literature, these solution processes are typically not explicitly known and numerical approximation methods are needed in order for mathematical modelling with stochastic wave equations to become relevant for real world applications. This, in turn, requires the numerical analysis of convergence rates for such numerical approximation processes. A recent article by the authors proves upper bounds for weak errors for spatial spectral Galerkin approximations of a class of semilinear stochastic wave equations. The findings there are complemented by the main result of this work, that provides lower bounds for weak errors which show that in the general framework considered the established upper bounds can essentially not be improved.
BibTeX
@misc{naurois2017lower,
  author = {Ladislas Jacobe de Naurois and Arnulf Jentzen and Timo Welti},
  title = {{Lower bounds for weak approximation errors for spatial spectral Galerkin approximations of stochastic wave equations}},
  year = {2017},
  eprint = {1701.04351},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1701.04351},
}
Figure