Extrapolation-based implicit-explicit general linear methods

Hong Zhang, Angelamaria Cardone, Zdzislaw Jackiewicz, Adrian Sandu, Hong Zhang 0006Numer. Algorithms 2014 · DOI

TL;DR

This paper introduces a novel extrapolation-based approach to construct high-order implicit-explicit general linear methods (IMEX GLMs) for solving stiff systems of ordinary differential equations. These new methods offer improved stability properties, particularly for the stiff components, making them attractive for a wide range of scientific and engineering problems.

What Problem Does This Paper Address?

Many real-world problems in science and engineering, such as those arising from the discretization of partial differential equations, lead to systems of ordinary differential equations (ODEs) with both stiff and non-stiff components. Explicit numerical methods struggle with stiff components due to stability constraints, while fully implicit methods can be computationally expensive. Implicit-explicit (IMEX) methods offer a compromise by treating the stiff part implicitly and the non-stiff part explicitly. While existing IMEX GLMs have shown promise, this paper aims to develop practical IMEX GLM pairs of high order with large absolute stability regions, specifically focusing on methods where the implicit part is A- or L-stable.

Methods

The core contribution of this paper is the development of an extrapolation-based approach to construct IMEX GLMs. This method builds upon existing general linear methods (GLMs) by combining an explicit scheme for the non-stiff part with an implicit scheme for the stiff part. The key innovation lies in using an extrapolation formula to construct the IMEX GLM pair. This involves:

  • Review of GLMs and IMEX GLMs: The paper begins by reviewing the theoretical framework of general linear methods and their implicit-explicit variants.
  • Extrapolation Formula: A new extrapolation formula is introduced, which relates stage values at two consecutive time steps.
  • Construction of IMEX GLMs: This extrapolation formula is then substituted into the general form of IMEX GLMs to derive a new family of methods. The paper details the derivation of the coefficient matrices for these extrapolation-based IMEX GLMs.
  • Stability Analysis: A significant portion of the paper is dedicated to analyzing the stability properties of these new schemes. This involves defining stability regions and investigating conditions for A(a)-stability and A-stability for the implicit part.
  • Construction of Specific Methods: The authors construct several IMEX GLM pairs of orders p=1 to p=4, focusing on those with desirable stability properties, particularly large stability regions for the explicit part and A- or L-stability for the implicit part. This involves extensive numerical searches for optimal parameter values.

Datasets

Not discussed in the paper. The paper focuses on theoretical development and numerical analysis of methods, not on training or evaluating on specific datasets.

Key Insights and Ideas

  • Extrapolation as a Unifying Framework: The paper demonstrates that extrapolation can be a powerful tool for constructing high-order IMEX GLMs, offering a systematic way to achieve desired stability and accuracy properties.
  • Bridging Explicit and Implicit Stability: A key challenge in IMEX methods is ensuring good stability for the combined scheme, even when the implicit part is stable. The extrapolation approach allows for the construction of methods where the stability region of the explicit part is significantly enlarged, contributing to the overall stability of the IMEX scheme.
  • High Stage Order Advantage: The paper reiterates the benefit of high stage orders in GLMs, which, in the context of IMEX GLMs, simplifies order conditions and avoids order reduction.
  • Trade-offs in Stability Regions: The analysis reveals trade-offs between the stability regions of the explicit and implicit parts. While aiming for A- or L-stability for the implicit part, the extrapolation method allows for significant expansion of the explicit part's stability region, which is crucial for practical applications.

Key Results

  • Development of a New Class of IMEX GLMs: The paper successfully derives and analyzes a new family of extrapolation-based IMEX GLMs.
  • Construction of High-Order Methods: Practical IMEX GLM pairs of orders p=1, 2, 3, and 4 with high stage orders are constructed.
  • Enhanced Stability Properties: The developed methods exhibit improved stability properties, particularly for the explicit component, with stability regions that are often larger than those of standard explicit methods.
  • Numerical Verification: The theoretical findings are confirmed through numerical experiments, including application to a two-dimensional shallow-water equations test problem, where the observed orders match theoretical predictions.
  • Prothero-Robinson Convergence: The paper shows that these extrapolation-based IMEX GLMs do not suffer from order reduction phenomenon when applied to stiff systems, as demonstrated by their Prothero-Robinson convergence.

Limitations and Future Work

  • Computational Cost: While IMEX methods reduce implicitness compared to fully implicit methods, high-order schemes can still involve significant computational cost, especially for very large systems.
  • Parameter Optimization: The construction of optimal methods often involves extensive computer searches for parameter values that maximize stability regions.
  • Future Work: The authors suggest extending the extrapolation idea to construct other types of partitioned GLMs, including parallel time integrators and asynchronous pairs of methods that do not share the same abscissae. Further exploration of different implicit components and their impact on the overall stability is also a promising direction.

Summary updated

BibTeX
@article{zhang2014extrapolation,
  author = {Hong Zhang and Angelamaria Cardone and Zdzislaw Jackiewicz and Adrian Sandu and Hong Zhang},
  title = {{Extrapolation-based implicit-explicit general linear methods}},
  journal = {Numer. Algorithms},
  year = {2014},
  doi = {10.1007/s11075-013-9759-y},
}
Figure