Moving Mesh Methods with Upwinding Schemes for Time-Dependent PDEs

Shengtai Li, Linda PetzoldJournal of Computational Physics 1997 · DOI

TL;DR

This paper explores combining moving mesh methods with upwinding schemes to improve the accuracy and robustness of solving time-dependent partial differential equations (PDEs), particularly those with shock or steep wave fronts. The authors demonstrate that this combination, especially with the local piecewise hyperbolic method (PHM), significantly enhances results compared to traditional central difference methods, particularly for nonconvex flux functions.

What Problem Does This Paper Address?

Many real-world phenomena, like fluid dynamics and wave propagation, are modeled by partial differential equations (PDEs) that can develop sharp features such as shocks or steep layers. Accurately capturing these features is crucial for reliable simulations. Traditional numerical methods often struggle with these sharp gradients, either requiring an excessive number of grid points (leading to inefficiency) or producing spurious oscillations. Moving mesh methods address this by dynamically adapting the grid to concentrate points where the solution changes rapidly. However, even with moving meshes, standard discretization schemes can still exhibit oscillations, especially when dealing with nonconvex flux functions. This paper aims to bridge this gap by investigating how different upwinding schemes, designed to handle sharp gradients, can be effectively integrated with moving mesh strategies to achieve more accurate and stable solutions.

Methods

The paper proposes a method of lines (MOL) approach, where the spatial derivatives are discretized, leaving a system of ordinary differential equations (ODEs) or differential-algebraic equations (DAEs) to be solved. The core of the paper lies in the investigation of various upwinding schemes for discretizing the convective term on nonuniform meshes, which arise from the moving mesh strategy.

Key methods discussed include:

  • Moving Mesh Strategy: The authors employ an equidistribution principle with an arclength monitor function to adapt the mesh. This strategy aims to place more grid points in regions of high spatial error. They also discuss regularization techniques to ensure mesh quality and prevent issues like mesh point crossing.
  • Upwinding Schemes:
    • First-Order Roe Scheme: A basic upwinding scheme that uses the "Roe speed" to determine the direction of information flow.
  • Second-Order ENO Scheme: An extension of the Roe scheme that uses divided differences to achieve second-order accuracy.
  • Third-Order ENO Scheme: A more complex scheme that uses higher-order divided differences to achieve third-order accuracy, aiming for non-oscillatory solutions.
  • Flux-Limited Schemes (e.g., Van Leer): Schemes that use flux limiters to prevent oscillations while maintaining accuracy.
  • Local Piecewise Hyperbolic Method (PHM): A scheme that constructs a piecewise hyperbolic approximation of the flux. The authors simplify this for nontransition cells in the context of moving meshes.
  • Numerical Solution: The resulting systems of ODEs/DAEs are solved using the stiff DAE solver DASSL, employing backward differentiation formulas (BDF) for time integration.

Datasets

Not discussed in the paper. The paper focuses on numerical methods and their application to well-known benchmark problems.

Key Insights and Ideas

  • Synergy of Moving Meshes and Upwinding: The combination of moving mesh methods and upwinding schemes is crucial for accurately resolving shocks and steep gradients, especially in problems with nonconvex flux functions where central differencing fails.
  • PHM's Superiority: The local piecewise hyperbolic method (PHM) emerged as a particularly effective scheme, offering excellent accuracy and robustness, often outperforming other upwinding schemes and even central differencing on moving meshes.
  • Trade-offs in Upwinding: While ENO schemes can achieve high order, they can be computationally expensive and sometimes unstable with adaptive time integration. Flux-limited schemes can be good but may not always yield optimal results. PHM offers a good balance of accuracy and efficiency.
  • Importance of Mesh Quality: The quality of the mesh generated by the moving mesh strategy significantly impacts the accuracy of the spatial discretization, the convergence of numerical solvers, and the overall stability of the simulation. Regularization and smoothing techniques are vital for maintaining good mesh quality.
  • Nonconvex Fluxes are Challenging: Problems with nonconvex flux functions pose a significant challenge, often leading to spurious oscillations even with moving meshes if appropriate upwinding schemes are not employed.

Key Results

  • Improved Accuracy for Shocks: Numerical experiments demonstrated that moving mesh methods combined with upwinding schemes (especially PHM) significantly reduced spurious oscillations and provided more accurate solutions for problems with shocks and steep fronts compared to central differencing.
  • Efficiency Gains: While some upwinding schemes (like third-order ENO) can be computationally intensive, the PHM scheme, when integrated with the moving mesh, offered good efficiency, as evidenced by the number of time steps and Jacobian evaluations in the provided tables.
  • Robustness for Difficult Problems: The combined approach proved robust for challenging problems, including Riemann problems with nonconvex flux functions, where central differencing failed to produce acceptable results.

Limitations and Future Work

  • Computational Cost: While PHM is efficient, some higher-order upwinding schemes can be computationally expensive. The paper notes that higher-order differencing in moving mesh computations has proven unstable in numerical experiments.
  • Generalization to Higher Dimensions: The paper primarily focuses on 1D problems. Extending these methods to higher dimensions would be a natural next step.
  • Explicit Methods for Non-Stiff Systems: The authors mention that for non-stiff systems, explicit methods can be more efficient, and moving mesh methods resulting in ODEs that can be solved by explicit solvers have been developed. Further exploration of these combinations could be beneficial.
  • Further Experimental Validation: The authors suggest that a wide range of experiments should be conducted to fully explore the capabilities of these combined methods.

Summary updated

BibTeX
@article{li1997moving,
  author = {Shengtai Li and Linda Petzold},
  title = {{Moving Mesh Methods with Upwinding Schemes for Time-Dependent PDEs}},
  journal = {Journal of Computational Physics},
  year = {1997},
  doi = {10.1006/jcph.1996.5611},
}
Figure