Convergence of an operator splitting scheme for abstract stochastic evolution equations

Joshua Lee Padgett, Quan Z. ShengarXiv 2019 · DOI · arXiv

TL;DR

This paper provides a rigorous mathematical proof for the convergence of the Lie-Trotter operator splitting scheme when applied to stochastic semi-linear evolution equations. The authors demonstrate that the scheme achieves a maximum strong convergence rate of half-order ($1/2$), and crucially, they show that this rate is directly tied to the regularity (or "smoothness") of the noise in the system.

What Problem Does This Paper Address?

In the world of partial differential equations (PDEs), "operator splitting" is a favorite tool for researchers. Instead of solving one massive, complex equation, you break it into smaller, simpler pieces (like separating the linear parts from the nonlinear parts) and solve them sequentially. While this works beautifully for deterministic equations, things get messy when you add randomness—specifically, stochastic PDEs (SPDEs).

The challenge is that stochastic terms, especially "multiplicative noise" (where the noise level depends on the state of the system), introduce significant analytical hurdles. Previous studies often looked at specific cases or linear versions. This paper seeks to provide a generalized convergence analysis for a common splitting method—the Lie-Trotter scheme—within an abstract Hilbert space framework. This allows the results to apply to a wide variety of real-world physical systems modeled by SPDEs.

Methods

The authors analyze the Lie-Trotter splitting scheme applied to the equation: $$du = [Au + f(u)]dt + g(u)dW$$

The scheme breaks the evolution into three distinct steps over a small time increment $h$:

  1. The Linear Step: Solving the part governed by the operator $A$ (often representing diffusion).
  2. The Deterministic Nonlinear Step: Solving the part governed by the function $f$ (drift).
  3. The Stochastic Step: Solving the part governed by the function $g$ and the Wiener process $W$ (noise).

The key technical contribution lies in the use of Ito’s isometry and a stochastic version of Taylor’s theorem to perform a detailed consistency analysis. By comparing the "split" solution to the true "mild" solution of the SPDE, the authors bound the error across multiple time steps.

Datasets

Not discussed in the paper. As a theoretical study in numerical analysis, the work relies on mathematical proofs and error estimates rather than experimental datasets or benchmarks.

Key Insights and Ideas

  • Noise Regularity is the Bottleneck: The most important insight is that the convergence rate is not a fixed constant. It depends on a parameter $\beta$, which measures the regularity of the noise's covariance operator.
  • The Half-Order Ceiling: The paper confirms that for these types of stochastic problems, you generally cannot do better than a convergence rate of $1/2$ (in the mean square sense) using this splitting method. This matches the "optimal" order known for simpler stochastic ordinary differential equations.
  • Unified Framework: By working in an abstract Hilbert space, the authors ensure their proof covers both the original continuous problem and the versions that have been "discretized" (turned into grids) for computer simulation.
  • Handling Nonlinearity: Unlike previous works that focused on additive noise (where noise is independent of the solution), this paper successfully tackles multiplicative noise, which is much more common in complex systems like fluid dynamics or finance.

Key Results

The main result is summarized in Theorem 2, which establishes the error bound: $$\mathbb{E} \|(S^n - T^n)(u_0)\|^2 \le Ch^\beta$$

  • Convergence Rate: The mean square convergence rate is $\beta/2$.
  • Trace Class Noise: If the noise is "smooth" (trace class noise, where $\beta = 1$), the scheme achieves the classic half-order convergence ($h^{1/2}$).
  • Space-Time White Noise: If the noise is "rougher" (like space-time white noise in one dimension, where $\beta < 1/2$), the convergence rate drops accordingly.

Limitations and Future Work

  • Initial Condition Restrictions: The proof assumes the initial state ($u_0$) is sufficiently "nice" (belonging to the domain of the operator $A$). In some physical applications, the starting state might be much rougher.
  • First-Order Splitting Only: This paper focuses on the Lie-Trotter scheme, which is a first-order splitting method. The authors do not explore higher-order methods like Strang splitting, which are popular in deterministic settings but much harder to analyze stochastically.
  • Computational Efficiency: While the paper proves that the method converges, it doesn't provide a numerical comparison of its speed or resource usage compared to non-splitting methods (like Euler-Maruyama).

Summary updated

Abstract

In this paper we study the convergence of a Lie-Trotter operator splitting for stochastic semi-linear evolution equations in a Hilbert space. The abstract Hilbert space setting allows for the consideration of convergence of the approximation for both the original and spatially discretized problems. It is known that the strong convergence of this scheme is classically of half-order, at best. We demonstrate that this is in fact the optimal order of convergence in the proposed setting, with the actual order being dependent upon the regularity of noise collected from applications.
BibTeX
@misc{padgett2019convergence,
  author = {Joshua Lee Padgett and Quan Z. Sheng},
  title = {{Convergence of an operator splitting scheme for abstract stochastic evolution equations}},
  year = {2019},
  eprint = {1901.06371},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1901.06371},
}
Figure