On the Application of Stable Generalized Finite Element Method for Quasilinear Elliptic Two-Point BVP

Tilsa Aryeni, Quanling Deng, Victor Ginting

TL;DR

This paper introduces a Stable Generalized Finite Element Method (SGFEM) designed to solve complex nonlinear (quasilinear) equations where material properties jump abruptly at internal interfaces. By using specialized "enrichment functions" and a Lagrange multiplier technique, the authors achieve optimal convergence rates and enforce local mass conservation without needing to painstakingly align the computational mesh with every material boundary.

What Problem Does This Paper Address?

In fields like hydrology or petroleum engineering, researchers model fluid flow through layered soils (e.g., sand meeting clay). These systems are described by quasilinear elliptic equations where the "conductivity" coefficient changes suddenly at the interface between materials.

Standard Finite Element Methods (FEM) face a major hurdle here: if the mesh triangles (or line segments in 1D) don't perfectly align with these interfaces, the accuracy of the simulation plummets. While you could simply refine the mesh, doing so is computationally expensive and impractical for "moving interface" problems or Monte Carlo simulations where interface locations change constantly. Furthermore, standard FEM often fails to satisfy local conservation—the physical requirement that the amount of "stuff" entering a small volume must equal the amount leaving it (minus what's stored).

Methods

The authors employ the Stable Generalized Finite Element Method (SGFEM). Here’s the breakdown of their approach:

  1. Enrichment Functions: Instead of just using standard polynomial basis functions, they "enrich" the space with auxiliary functions that specifically mimic the behavior of a solution near a discontinuity.
  2. The "Stable" Modification: A known issue with Generalized FEM is that it can lead to "ill-conditioned" matrices (mathematical instability where tiny errors explode). The authors use a specific construction—subtracting the linear interpolant from the enrichment function—to ensure the method remains as stable as standard FEM.
  3. Newton’s Method: Since the problem is quasilinear (the coefficients depend on the unknown solution itself), they use an iterative Newton's solver to handle the nonlinearity.
  4. Lagrange Multipliers for Conservation: To fix the local conservation problem, they treat conservation as a constraint. They use Lagrange multipliers to "force" the solution to obey mass balance over specific control volumes.

Datasets

As a theoretical numerical analysis paper, the authors do not use external real-world datasets. Instead, they evaluate their method using synthetic benchmarks where analytical (exact) solutions are mathematically derived for comparison:

  • Example 1: A 1D problem with two interfaces and an exponential conductivity coefficient ($\kappa(x, u) = e^{au}$).
  • Example 2: A more complex 1D problem with three interfaces and a high "contrast ratio" (where conductivity differs by orders of magnitude across layers).

Key Insights and Ideas

  • Mesh Independence: The "aha moment" is that SGFEM allows the interface to cut right through the middle of an element without losing the "optimal" rate of convergence. You don't have to move your mesh nodes to match the soil layers.
  • Stability is Key: The paper reinforces that GFEM is only useful if it's stable. By carefully constructing the enrichment functions, they prove that the conditioning of the system doesn't deteriorate compared to standard FEM.
  • Conservation vs. Accuracy: A critical insight is that imposing local conservation via Lagrange multipliers preserves the optimal convergence in the $H^1$ norm (energy-like norm), but can slightly degrade the $L^2$ norm (standard error) unless a "corrector" term is applied.

Key Results

  • Optimal Convergence: The numerical tests confirmed the theoretical analysis. For a polynomial degree $p$, the error decreased at a rate of $h^p$ in the $H^1$ semi-norm and $h^{p+1}$ in the $L^2$ norm.
  • High Contrast Handling: The method remained robust even when the material properties differed by a factor of over 2,600 (the contrast ratio).
  • Zero Conservation Error: While standard SGFEM showed fluctuating local conservation errors, the version using Lagrange multipliers reduced these errors to near-zero (machine precision, $\sim 10^{-12}$), effectively "fixing" the physics of the simulation.

Limitations and Future Work

  • Dimensionality: This study is strictly limited to one-dimensional (1D) problems. While 1D provides a proof-of-concept, real-world porous media flow is 3D. The authors note that modifying enrichment functions for higher dimensions is significantly more complex.
  • Time Dependence: The paper focuses on steady-state (elliptic) problems. The authors aim to extend this to the Richards equation, which is time-dependent (parabolic) and describes how water actually soaks into the ground over time.
  • Computational Overhead: While SGFEM saves time on mesh generation, the use of Lagrange multipliers increases the size of the algebraic system to be solved. Future work may need to address the efficiency of these larger solvers.

Summary updated

Abstract

In this paper, we discuss an application of the generalized finite element method (GFEM) to approximate the solutions of quasilinear elliptic equations with multiple interfaces in one dimensional space. The problem is characterized by spatial discontinuity of the elliptic coefficient that depends on the unknown solution. It is known that unless the partition of the domain matches the discontinuity configuration, accuracy of standard finite element techniques significantly deteriorates and standard refinement of the partition may not suffice. The GFEM is a viable alternative to overcome this predicament. It is based on the construction of certain enrichment functions supplied to the standard finite element space that capture effects of the discontinuity. This approach is called stable (SGFEM) if it maintains an optimal rate of convergence and the conditioning of GFEM is not worse than that of the standard FEM. A convergence analysis is derived and performance of the method is illustrated by several numerical examples. Furthermore, it is known that typical global formulations such as FEMs do not enjoy the numerical local conservation property that is crucial in many conservation law-based applications. To remedy this issue, a Lagrange multiplier technique is adopted to enforce the local conservation. A numerical example is given to demonstrate the performance of proposed technique.
BibTeX
@article{aryeni2022application,
  author = {Tilsa Aryeni and Quanling Deng and Victor Ginting},
  title = {{On the Application of Stable Generalized Finite Element Method for Quasilinear Elliptic Two-Point BVP}},
  journal = {J. Sci. Comput.},
  year = {2022},
  url = {https://openreview.net/forum?id=oifkFHscYDsR},
}

Loading related content...

Figure