Iterated Numerical Homogenization for MultiScale Elliptic Equations with Monotone Nonlinearity

Xinliang Liu, Eric Chung, Lei Zhang

TL;DR

This paper introduces an "iterated numerical homogenization" method to tackle complex multi-scale nonlinear partial differential equations (PDEs) common in materials science and biology. By combining techniques for linear and nonlinear problems, they achieve efficient and accurate simulations for challenging scenarios where traditional methods struggle.

What Problem Does This Paper Address?

Many real-world phenomena, from material properties to biological processes, involve interactions across vastly different scales (e.g., microscopic material structures influencing macroscopic behavior). When these phenomena are described by nonlinear PDEs, simulating them becomes incredibly difficult. The "monotone nonlinearity" aspect, particularly in problems like the $p$-Laplacian equation, presents a significant hurdle because the nonlinear behavior cannot be easily simplified or parameterized. This makes standard numerical homogenization techniques, which work well for linear problems, insufficient. The core challenge is to develop a robust numerical method that can handle both the multi-scale nature and the complex nonlinearities without prohibitive computational cost.

Methods

The authors propose an "iterated numerical homogenization" scheme. This approach iteratively refines an approximate solution by:

  • Leveraging Linear Homogenization: They adapt existing numerical homogenization techniques (specifically mentioning generalized rough polyharmonic splines - GRPS) developed for linear problems to construct coarse-scale basis functions.
  • Employing a "Quasi-Norm" Approach: For the nonlinear part, they utilize an iterative method based on a "quasi-norm" that offers exponential energy convergence, independent of the material's heterogeneity.
  • Introducing Residual Regularization: To ensure convergence, especially when the residual (a measure of how well the current solution satisfies the equation) becomes problematic, they introduce a residual regularized line search. This helps stabilize the iterative process.
  • Developing Sparse Updating: To reduce computational cost, they propose a "sparse updating" strategy for the coarse-scale basis functions. Instead of recalculating all basis functions at each iteration, they only update a fraction, assuming that as the solution converges, the underlying operator doesn't change drastically.

Datasets

Not discussed in the paper. The paper focuses on developing and validating numerical methods using synthetic examples and a benchmark problem.

Key Insights and Ideas

  • Bridging Linear and Nonlinear Homogenization: The core idea is to adapt successful linear homogenization techniques to the nonlinear setting by applying them iteratively within a nonlinear solver. This is a powerful conceptual leap.
  • Exponential Convergence for Nonlinear Problems: The "quasi-norm" based approach offers a theoretical guarantee of exponential energy error decay, which is a significant achievement for nonlinear multi-scale problems.
  • Residual Regularization as a Stabilizer: The paper highlights that standard iterative methods can fail due to residual issues in nonlinear problems. Their residual regularization technique provides a practical way to ensure convergence.
  • Sparse Updating for Efficiency: The insight that coarse-space basis functions don't need to be fully recomputed at every iteration, leading to significant computational savings, is crucial for practical applications.

Key Results

  • The proposed iterated numerical homogenization method demonstrates exponential convergence in energy error for nonlinear multi-scale elliptic problems.
  • Numerical experiments show that the method is effective for both a synthetic multi-scale trigonometric example and a challenging SPE10 benchmark problem with high contrast coefficients.
  • The residual regularization technique successfully stabilizes the iterative process, enabling convergence where standard methods might fail.
  • The sparse updating strategy significantly improves computational efficiency, especially for more nonlinear problems. For instance, for the SPE10 example with $p=20$, full updating achieved 100% efficiency, while sparse updating with a threshold $\delta_t = 10^{-8}$ achieved 55.75% efficiency with a comparable accuracy.

Limitations and Future Work

  • Regularization Parameters: The choice and decay of regularization parameters ($\epsilon_-, \epsilon_+$) are not fully analyzed in terms of their impact on the error, though they are shown to be effective in practice.
  • Computational Cost of Basis Construction: While sparse updating helps, constructing and updating the coarse-space basis functions can still be computationally intensive.
  • Future Work: The authors suggest extending their methods to more complex scenarios, including problems with nonconvex flux, nonlinear heterogeneous elasticity, and exploring nonlocal upscaling techniques. They also aim to study more complicated nonlinear multiscale problems.

Summary updated

Abstract

Nonlinear multiscale problems are ubiquitous in materials science and biology. Complicated interactions between nonlinearities and (nonseparable) multiple scales pose a major challenge for analysis and simulation. In this paper, we study the numerical homogenization for multiscale elliptic PDEs with monotone nonlinearity, in particular the Leray--Lions problem (a prototypical example is the $p$-Laplacian equation), where the nonlinearity cannot be parameterized with low dimensional parameters, and the linearization error is nonnegligible. We develop the iterated numerical homogenization scheme by combining numerical homogenization methods for linear equations and the so-called quasi-norm based iterative approach for monotone nonlinear equation. We propose a residual regularized nonlinear iterative method and, in addition, develop the sparse updating method for the efficient update of coarse spaces. A number of numerical results are presented to complement the analysis and validate the numerical method.
BibTeX
@inproceedings{liu2021iterated,
  author = {Xinliang Liu and Eric Chung and Lei Zhang},
  title = {{Iterated Numerical Homogenization for MultiScale Elliptic Equations with Monotone Nonlinearity}},
  booktitle = {Multiscale Model. Simul},
  year = {2021},
  url = {https://openreview.net/forum?id=V8fES-l1YC},
}
Figure