Numerical Methods for Non-Smooth Interfaces and Boundaries

92 concepts76 papers1 proposal

Most classical numerical methods for partial differential equations (PDEs) are built on an implicit assumption of smoothness. Whether using Taylor expansions in finite differences or polynomial basis functions in finite elements, these solvers expect the underlying physical fields and the geometries they inhabit to be differentiable.

However, the physical world is rarely smooth. Material interfaces—such as the boundary between a protein and water or the junction of two different metals—introduce sharp jumps in coefficients like dielectric permittivity or conductivity. Furthermore, engineering geometries are often defined by sharp corners, edges, and cusps. When these "singularities" meet standard numerical stencils, the mathematical machinery breaks down: convergence rates plummet, oscillations appear, and solvers may fail to converge entirely.

Research in this area focuses on redesigning numerical frameworks to treat these discontinuities not as errors to be smoothed away, but as fundamental features to be explicitly matched.

The Challenge of the Jump

When a physical property changes abruptly, the solution to the governing PDE typically exhibits a "jump condition." For example, in elliptic interface problems, the gradient of the solution might be discontinuous across the boundary. Beams et al. note that these conditions often require localized mesh refinement or specialized stencils to prevent errors from propagating throughout the entire domain.

The intuition behind modern solutions is to decouple the representation of the solution on either side of the interface. Instead of forcing a single continuous function to approximate a sharp kink, methods like the Random Feature Method (RFM) proposed by Chi et al. assign distinct sets of basis functions to each side of the boundary. By coupling these functions through the interface conditions, the model can represent sharp jumps in physical properties without needing the mesh to perfectly align with the boundary.

Matched Interfaces and Correction Functions

To maintain high-order accuracy on simple grids, researchers have developed techniques to "correct" standard stencils near an interface. The Matched Interface and Boundary (MIB) method, introduced by Wei et al., uses fictitious points near the interface. By iteratively enforcing jump conditions to determine the values at these points, the authors allow standard finite difference schemes to maintain high-order accuracy even when coefficients change abruptly.

A similar philosophy underpins the Correction Function Method (CFM). As Law et al. demonstrate for Maxwell’s equations, adding a correction term to the standard discretization can effectively handle discontinuities in material properties and complex geometries. This prevents the spurious oscillations that typically plague simulations of electromagnetic wave propagation across dielectric boundaries.

Handling Geometric Singularities

While material interfaces are one source of trouble, the shape of the domain itself is another. Many theoretical guarantees for surface-conforming meshes or spectral solvers rely on the assumption of smooth (often $C^2$) boundaries. Abdelkader et al. point out that these proofs often fail to extend to geometries with sharp features like corners or edges, which are ubiquitous in mechanical parts.

In 3D elasticity, these non-smooth features—such as tips and cusps—are particularly problematic because they violate the local smoothness required for high-order convergence. Wang et al. argue that standard solvers lose accuracy in these regions because the surface normal is discontinuous. To combat this, "immersed" methods allow the geometry to cut through a background mesh. However, as Xiao et al. observe, even high-order immersed boundary formulations can revert to lower-order linear approximations when encountering sharp knife-edges, representing a significant ongoing challenge in the field.

Evolution and Viscosity Solutions

The problem becomes even more complex when the boundaries move. In models of cell motility or crystal growth, the interface evolves over time, often developing sharp corners or "kinks" even if it started smooth. Mazurenko shows that using standard radial functions to describe these evolving sets leads to equations where classical derivatives are undefined.

To handle this, researchers turn to viscosity solutions for Hamilton-Jacobi equations. This framework provides a mathematically rigorous way to define how a surface moves even at points where it is not differentiable. Prados et al. applied this to the "Shape from Shading" problem, demonstrating that viscosity solutions can accurately reconstruct surfaces with sharp ridges where traditional smooth-surface assumptions fail. Similarly, Chambolle et al. have extended these ideas to nonlocal curvature flows, allowing for the analysis of "crystalline" perimeters that favor flat facets over smooth curves.

Beyond Traditional Solvers

The rise of machine learning has introduced new perspectives on these singularities. Standard Physics-Informed Neural Networks (PINNs) often struggle with the sharp transitions found in interface problems. To address this, Xu et al. proposed a Deep Domain Decomposition Method (DeepDDM), which uses neural networks as mesh-free local solvers. By breaking the domain into pieces, the networks can more easily capture the discontinuous coefficients and complex interface geometries that would baffle a single global model.

Whether through fictitious points, specialized basis functions, or viscosity solutions, the goal remains the same: to build numerical tools that are as robust as the non-smooth physical reality they aim to simulate.

Key Foundational Work

Quelques résultats de régularité pour un système elliptique avec conditions aux limites couplées (1980) — contributed 1 concept elements to this direction.
The Wulff shape as the asymptotic limit of a growing crystalline interface (1997) — contributed 1 concept elements to this direction.
History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_geometric_singularity_handling_d85e4c1e,
  title = {geometric singularity handling},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/geometric-singularity-handling/rdr_d85e4c1ed2954cfbb851b4c5c7e9a37f},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-22},
}

This exact version (version 1, frozen text):

@misc{lacuna_geometric_singularity_handling_d85e4c1e_v1,
  title = {geometric singularity handling},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/geometric-singularity-handling/rdr_d85e4c1ed2954cfbb851b4c5c7e9a37f/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-22},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake