Finite element discretizations of nonlocal minimal graphs: Convergence

Juan Pablo Borthagaray, Wenbo Li, Ricardo H. Nochetto

Abstract

This paper presents the first rigorous numerical analysis and finite element framework for computing fractional minimal graphs—the nonlocal counterparts to classical minimal surfaces like soap films. The authors establish a convergence theory for piecewise linear discretizations and introduce a novel geometric error metric that bridges the gap between nonlocal and classical surface geometry. Numerical experiments validate the theoretical findings, specifically capturing the characteristic "stickiness" phenomenon inherent to fractional minimal surfaces.

Introduction

Classical minimal surfaces are defined by local properties: the mean curvature vanishes at every point, a concept exemplified by soap films stretched across a wire frame. However, many complex physical phenomena involving surface tension or phase transitions exhibit "nonlocal" behavior, where the energy depends on interactions between distant points rather than just immediate neighbors.

Mathematically, these are described by fractional minimal surfaces. While the theoretical understanding of these surfaces has matured since the late 2000s, numerical methods to compute them have lagged behind. The challenge lies in the nature of the governing operators: they are nonlinear, nonlocal, and often degenerate. Furthermore, nonlocal surfaces exhibit a unique "stickiness" phenomenon where the surface may refuse to meet the boundary data continuously, creating jumps that violate standard numerical assumptions.

Numerical Methodology

The authors propose a numerical scheme based on the Finite Element Method (FEM), focusing on the fractional order $s \in (0, 1/2)$.

Discretization and Boundary Handling

The framework utilizes standard continuous, piecewise linear Lagrangian finite elements on shape-regular meshes. Because the problem is nonlocal, "boundary conditions" must be imposed on the entire complement of the domain ($\Omega^c$), not just the boundary edge. To account for this exterior data, the authors employ a specialized Clément-type interpolation operator, ensuring the discretization remains consistent with the nonlocal energy functional.

Nonlinear Solvers

To solve the resulting discrete system, the paper implements and compares two primary iterative approaches:

  1. $H^\alpha$-gradient flow: A stable iterative descent method that ensures convergence but typically requires more iterations.
  2. Damped Newton Method: A significantly faster approach utilizing the first variation of the energy functional. In practice, the Newton method proved highly efficient, requiring as few as 4 iterations compared to 26 for gradient flow in identical test cases.

A New Geometric Metric

A major contribution of this work is the definition of a new error functional, $e_s(u, u_h)$. This metric measures the discrepancy between the "nonlocal normals" of the exact and approximate surfaces. Crucially, as $s \to 1/2$, both the nonlocal energy and this $e_s$ metric recover their classical, local counterparts, providing a unified geometric framework for both regimes.

Key Insights and Theoretical Results

Rigorous Convergence

The authors provide a mathematical proof that the numerical scheme converges even under low-regularity conditions. Specifically, the discrete solutions $u_h$ converge to the true solution $u$ in the Sobolev space $W^{2r}_1(\Omega)$ for all $r < s$ as the mesh size $h \to 0$. This is a vital result given that nonlocal minimal surfaces often exhibit discontinuities at the boundaries.

The Stickiness Phenomenon

Numerical experiments on a symmetric annulus domain ($B_1 \setminus B_{1/2}$) visually confirm the "stickiness" phenomenon. For small values of $s$, the computed graph shows a visible gap at the inner boundary of the annulus. This detachment from the Dirichlet data is a known theoretical property of nonlocal minimal surfaces that distinguishes them from classical soap films.

Convergence Rates

In the limit where the nonlocal surface approaches a classical one ($s \approx 0.5$), numerical tests suggest a second-order convergence rate ($O(h^2)$) in the $L^1$ norm. This matches the established rates for classical minimal surface approximations, suggesting the method is optimally efficient in the local limit.

Limitations and Future Directions

While this paper establishes a foundational framework, several challenges remain:

  • Exterior Data: The current analysis assumes the exterior data $g$ has compact support. Handling truly unbounded data is reserved for future work.
  • Sharp Error Estimates: While convergence is proven, deriving sharp error rates remains difficult because nonlocal minimal surfaces can be discontinuous at the boundary.
  • Computational Complexity: Nonlocal operators result in dense matrices, unlike the sparse matrices found in classical FEM. Scaling these methods to large-scale 3D problems will require advanced compression techniques or fast multipole methods to manage the increased computational cost.

Conclusion

By bridging the gap between nonlocal theory and numerical practice, this work provides a robust toolset for studying fractional minimal surfaces. The introduction of the $e_s$ metric and the proof of convergence for piecewise linear elements represent significant steps forward in the computational geometry of nonlocal phenomena.

Summary updated

Abstract

In this paper, we propose and analyze a finite element discretization for the computation of fractional minimal graphs of order~$s \in (0,1/2)$ on a bounded domain $Ω$. Such a Plateau problem of order $s$ can be reinterpreted as a Dirichlet problem for a nonlocal, nonlinear, degenerate operator of order $s + 1/2$. We prove that our numerical scheme converges in $W^{2r}_1(Ω)$ for all $r<s$, where $W^{2s}_1(Ω)$ is closely related to the natural energy space. Moreover, we introduce a geometric notion of error that, for any pair of $H^1$ functions, in the limit $s \to 1/2$ recovers a weighted $L^2$-discrepancy between the normal vectors to their graphs. We derive error bounds with respect to this novel geometric quantity as well. In spite of performing approximations with continuous, piecewise linear, Lagrangian finite elements, the so-called {\em stickiness} phenomenon becomes apparent in the numerical experiments we present.
BibTeX
@inproceedings{borthagaray2019finite,
  author = {Juan Pablo Borthagaray and Wenbo Li and Ricardo H. Nochetto},
  title = {{Finite element discretizations of nonlocal minimal graphs: Convergence}},
  booktitle = {Nonlinear},
  year = {2019},
  doi = {10.1016/j.na.2019.06.025},
}
Figure