A Discontinuous Ritz Method for a Class of Calculus of Variations Problems

Xiaobing Feng, Stefan Schnake

Abstract

This paper develops an analogue to discontinuous Galerkin (DG) methods for approximating a general class of calculus of variations problems. The proposed method, called the discontinuous Ritz (DR) method, constructs a numerical solution by minimizing a discrete energy over DG function spaces. The discrete energy includes standard penalization terms as well as the DG finite element (DG-FE) numerical derivatives developed recently. It is proved that the proposed DR method converges and that the DG-FE numerical derivatives exhibit a compactness property which is crucial for applying the method to complex energy functionals. Numerical tests on the classical $p$-Laplace problem gauge the performance of the proposed DR method.

Introduction

In the calculus of variations, the objective is typically to find a function that minimizes a given energy functional, such as the $p$-Laplace energy. Traditionally, researchers use an indirect approach: they derive the Euler-Lagrange equations (a PDE) and solve those. However, the direct approach—minimizing the energy functional itself—is often more desirable because it naturally preserves physical properties like conservation laws.

The challenge arises when using Discontinuous Galerkin (DG) methods. While DG methods are flexible and handle complex geometries well, the "broken" nature of the functions at element boundaries means the standard gradient operator ($\nabla$) is not well-defined globally. A "naive" approach—simply taking the gradient of each piece separately—often fails to converge to the correct solution. This paper introduces the Discontinuous Ritz (DR) method to bridge this gap, creating a discrete energy functional that guarantees convergence while maintaining the flexibility of discontinuous spaces.

Methodology

The DR method relies on several core components to ensure stability and consistency:

  1. DG-FE Numerical Derivatives: Instead of standard piecewise gradients, the method uses numerical partial derivatives that incorporate "trace operators." These operators account for jumps across element edges, effectively capturing the distributional derivative that a naive gradient misses.
  2. Discrete Energy Minimization: The method constructs a discrete energy functional $J_h$ consisting of the energy density evaluated with these numerical derivatives, plus standard interior penalty terms to weakly enforce continuity and boundary conditions.
  3. Mathematical Equivalence: A key theoretical contribution is proving that this DR method is equivalent to the "variational DGFEM" proposed by Buffa and Ortner. The DR method provides a more concrete "numerical calculus" framework that is easier to implement.
  4. Compactness Proof: The authors prove a compactness property for their numerical gradients, which is the mathematical engine required to guarantee that a sequence of discrete minimizers actually converges to the true solution as the mesh is refined.

Numerical Experiments

The method was evaluated using several benchmark problems for the $p$-Laplace problem:

  • Test 1 (1D, $p=2.5$): A smooth cubic solution to test optimal convergence.
  • Test 2 (1D, $p=1.5$): A degenerate case where the solution is a sine wave, testing robustness when derivatives vanish.
  • Test 3 (1D, $p=8.3$): A case with an unknown exact solution, compared against a highly refined reference solution.
  • Test 4 (2D, $p=2.5$): Extending the test to two dimensions with an exponential solution.

Key Results and Insights

The DR method successfully achieved convergence for a general class of convex and coercive energy densities. In numerical tests, the method achieved optimal convergence rates for the gradient error (the $W^{1,p}$ semi-norm).

A critical insight of the paper is the demonstration of the "Naive Trap." As shown in the convergence data below, simple piecewise gradients can lead to inconsistent results where the error plateaus regardless of mesh refinement, highlighting the necessity of the specialized DG-FE derivatives.

Convergence of the $H^1$ error for varying penalty parameters $\gamma_e$ and mesh sizes $h$. The plateauing of the error illustrates the consistency challenges that the DR method aims to overcome.

While the gradients converged optimally, the $L^p$ norm of the function values sometimes showed sub-optimal rates, which is a known characteristic of these types of non-linear variational problems. Nonetheless, the method proved robust across both $p > 2$ and $1 < p < 2$ regimes, handling cases with singularities or degeneracies effectively.

Conclusion

The Discontinuous Ritz method provides a robust framework for the direct minimization of energy functionals using discontinuous spaces. By framing the solution as a "numerical calculus" problem, the authors have made the method highly implementable, even providing a Matlab toolbox to make these derivatives accessible. Future work will involve applying the DR method to more complex physical problems, such as the Erickson energy for nematic liquid crystals or minimal surface problems.

Summary updated

Abstract

This paper develops an analogue (or counterpart) to discontinuous Galerkin (DG) methods for approximating a general class of calculus of variations problems. The proposed method, called the discontinuous Ritz (DR) method, constructs a numerical solution by minimizing a discrete energy over DG function spaces. The discrete energy includes standard penalization terms as well as the DG finite element (DG-FE) numerical derivatives developed recently by Feng, Lewis, and Neilan in [Feng2013]. It is proved that the proposed DR method converges and that the DG-FE numerical derivatives exhibit a compactness property which is desirable and crucial for applying the proposed DR method to problems with more complex energy functionals. Numerical tests are provided on the classical $p$-Laplace problem to gauge the performance of the proposed DR method.
BibTeX
@misc{feng2017discontinuous,
  author = {Xiaobing Feng and Stefan Schnake},
  title = {{A Discontinuous Ritz Method for a Class of Calculus of Variations Problems}},
  year = {2017},
  eprint = {1709.04297},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1709.04297},
}

Loading related content...

Figure