💻 computer science

Convergence of Discontinuous Galerkin Methods for Quasiconvex and Relaxed Variational Problems

This paper establishes that discontinuous Galerkin methods provide reliable and convergent approximations for nonlinear variational problems in elasticity, proving their effectiveness for both quasiconvex energies and non-convex cases where discrete minimizers converge to solutions of the relaxed problem defined by the quasiconvex envelope.

Original authors: Georgios Grekas, Konstantinos Koumatos, Charalambos Makridakis, Andreas Vikelis

Published 2026-07-29

📖 3 min read☕ Coffee break read

CC BY 4.0

Original authors: Georgios Grekas, Konstantinos Koumatos, Charalambos Makridakis, Andreas Vikelis

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to build a perfect model of a rubber band stretching, a piece of metal bending, or a crystal changing shape. In the world of physics and engineering, this is called "elasticity." Scientists use complex math to predict exactly how these materials will behave under stress. Usually, they rely on a rule called "convexity," which is a fancy way of saying that if you mix two good solutions, the result is also a good solution. It's like baking: if you mix two batches of perfect cookie dough, you get more perfect dough.

However, real life is messy. Some materials, especially those that change phase (like ice turning to water, or metals changing their internal structure), don't follow this simple rule. Their energy landscapes are like a mountain range with many deep valleys. The "best" solution might not be a smooth, continuous path but a jagged, chaotic mix of different states. This is where things get tricky. Standard computer methods, which try to force the material to be smooth and connected everywhere, often get stuck. They might find a "good enough" solution that looks smooth but is actually wrong, missing the true, complex behavior of the material. This is a major headache for scientists trying to design new materials or understand why things break.

This paper tackles that headache by introducing a smarter way to use computers to solve these tricky math problems. The authors, a team of mathematicians, propose using a technique called "Discontinuous Galerkin" methods. Think of this as a construction crew that isn't afraid to leave tiny gaps between their bricks. Instead of forcing every piece of the material to be perfectly glued to its neighbor (which causes the computer to miss the complex, jagged solutions), they allow the pieces to slide and jump slightly.

The paper proves two big things. First, when the material behaves in a "quasiconvex" way (a slightly more flexible version of the smooth rule), these gap-allowing methods work perfectly and find the right answer. Second, and even more impressively, when the material is truly chaotic and non-convex, the method doesn't fail; instead, it naturally converges to the "relaxed" solution. This is the mathematical equivalent of finding the average of all the possible chaotic states, which is exactly what nature does when it forms complex patterns like the stripes on a zebra or the layers in a crystal.

The authors didn't just guess this would work; they provided a rigorous mathematical proof showing that as the computer grid gets finer and finer, the answers get closer and closer to the true physical reality. They also ran simulations to back it up. In one test, they looked at a material under compression and showed that their method found the correct energy level much faster and more accurately than older methods, even with small settings. In another test, they simulated a material that wants to split into two different phases. As they made the computer grid smaller, the simulation naturally formed intricate, microscopic patterns (like a digital mosaic) that matched the theoretical prediction of how the material would behave.

Essentially, this paper gives scientists a new, more flexible tool. It shows that by letting the computer model be a little "discontinuous"—by allowing the pieces to be slightly out of sync—we can actually capture the wild, complex, and beautiful ways real materials behave, solving problems that older, stricter methods simply couldn't crack.

Technical Summary: Convergence of Discontinuous Galerkin Methods for Quasiconvex and Relaxed Variational Problems

Problem Statement
The paper addresses a fundamental challenge in the vectorial calculus of variations and materials science: the reliable numerical approximation of nonlinear energy minimization problems where the energy density $W$ is non-convex. Specifically, the authors focus on problems where minimizers may possess singularities or where the "Lavrentiev phenomenon" occurs, leading standard finite element methods (which enforce inter-element continuity) to converge to energy values strictly greater than the exact minimum. In the absence of convexity, the classical minimization problem may lack solutions; instead, one must consider the "relaxed" problem, where the energy density is replaced by its quasiconvex envelope, $W^{qc}$ . The central difficulty lies in designing numerical schemes that remain stable and convergent when $W$ is only quasiconvex, a condition weaker than convexity but essential for elasticity.

Methodology
The authors propose and analyze Discontinuous Galerkin (DG) methods, which utilize discontinuous piecewise polynomial spaces ( $V_h^q(\Omega)$ ) to eliminate the inter-element continuity constraints that limit standard finite elements. The study covers two specific DG formulations:

The method of Ten Eyck and Lew (2006): Based on a functional involving a "discrete gradient" $G_h(u_h) = \nabla_h u_h - R_h(u_h)$ , where $R_h$ is a lifting operator.
A modified interior penalty method: Based on the functional introduced in prior work [27], utilizing the element-wise gradient $\nabla_h u_h$ .

A critical component of the methodology is the introduction of a modified penalty term, $\text{Pen}(u_h)$ , designed to ensure discrete stability for non-convex energies. Unlike previous penalty terms tailored for convex energies, this new term is constructed to handle the lack of a true gradient structure in the discrete setting (where $\text{curl}(\nabla_h u_h) \neq 0$ ).

The theoretical analysis relies on $\Gamma$ -convergence. The authors establish that the discrete energy functionals $\Gamma$ -converge to the relaxed continuous energy functional $E_{qc}(u) = \int_\Omega W^{qc}(\nabla u) dx$ . Key technical tools include:

Reconstruction Operators: Extending results from Karakashian and Pascal [28] to $L^p$ spaces to construct continuous approximations $w_h$ from discontinuous $u_h$ .
Error Estimates: Proving that the error between the discontinuous function and its continuous reconstruction is controlled by the jump terms across element interfaces.
Lower Semicontinuity: Overcoming the lack of lower semicontinuity for quasiconvex integrands in DG spaces by transitioning to the limit of the reconstructed continuous functions, where standard weak lower semicontinuity arguments apply.

Key Contributions and Results
The paper establishes two primary convergence theorems (Theorems 3.1 and 3.2):

Convergence for Quasiconvex Energies: If the energy density $W$ is quasiconvex, the discrete minimizers of the DG schemes converge (up to a subsequence) to a minimizer of the continuous quasiconvex energy.
Convergence for Non-Convex Energies: If $W$ is non-convex, the discrete minimizers converge to a minimizer of the relaxed problem defined by the quasiconvex envelope $W^{qc}$ .

The authors demonstrate that their approach covers all known DG formulations that converge for convex energies, extending their validity to the quasiconvex and non-convex regimes. The analysis proves that the discrete minimizers do not get trapped in energy values higher than the relaxed minimum, a failure mode often observed in standard conforming methods for problems exhibiting the Lavrentiev phenomenon.

Numerical Evidence
The theoretical results are supported by numerical experiments:

Polyconvex Energy: For a polyconvex strain energy $W(F) = |\det F|^2$ , the authors compare their new penalty term against a standard penalty used for convex energies. Results show that the new penalty achieves convergence to the exact homogeneous minimizer with significantly smaller stabilization parameters ( $\alpha$ ), whereas the standard penalty requires large $\alpha$ to suppress oscillations.
Non-Convex Two-Well Potential: The authors simulate a frame-indifferent two-well energy function modeling phase transitions. As the mesh size $h$ decreases, the discrete minimizers develop increasingly fine microstructures (laminates) that approach the infimum of the total potential energy. The numerical solutions strongly converge in $L^2$ to the minimizer of the relaxed problem, capturing the formation of a "macroscopic" deformation equal to the relaxed solution, despite the presence of microscopic oscillations.

Significance
The paper claims to resolve a significant technical obstacle in the design of approximation methods for elasticity and variational problems: the lack of lower semicontinuity for quasiconvex energies in discontinuous discrete spaces. By rigorously justifying that DG methods can produce convergent approximations for quasiconvex and relaxed problems, the work provides a robust numerical framework for materials science applications where singular solutions and phase mixtures are common. The authors position this work as a step toward solving the open challenge of reliably approximating nonlinear energy minimization problems with potentially singular solutions, addressing the limitation of standard finite element methods noted by Ball [2] and Carstensen [12].

More like this