[Paper Review] The Parabolic Two-Phase Membrane Problem: Regularity in Higher Dimensions

Henrik Shahgholian, N. N. Uraltseva+1 more|ArXiv.org|Dec 20, 2007

Geotechnical Engineering and Underground Structures10 references3 citations

TL;DR

This paper establishes the regularity of the free boundary in the parabolic two-phase membrane problem, proving that near branch points—where the solution vanishes and its gradient is zero—the interfaces $Σ_+ = \{u>0\}$ and $\Sigma_- = \{u<0\}$ are locally the graphs of Lipschitz functions that are continuously differentiable in space. The result extends prior elliptic regularity to the parabolic setting and shows the bound is optimal, as $C^1$ regularity does not hold in general.

ABSTRACT

For the parabolic obstacle-problem-like equation $$Δu - \partial_t u = λ_+ χ_{\{u&gt;0\}} - λ_- χ_{\{u&lt;0\}} ,$$ where $λ_+$ and $λ_-$ are positive Lipschitz functions, we prove in arbitrary finite dimension that the free boundary $\partial\{u&gt;0\} \cup\partial\{u&lt;0\}$ is in a neighborhood of each ``branch point'' the union of two Lipschitz graphs that are continuously differentiable with respect to the space variables. The result extends the elliptic paper \cite{imrn} to the parabolic case. The result is optimal in the sense that the graphs are in general not better than Lipschitz, as shown by a counter-example.

Motivation & Objective

• To establish the regularity of the free boundary in the parabolic two-phase membrane problem, where the solution satisfies $\Delta u - \partial_t u = \lambda_+ \chi_{\{u>0\}} - \lambda_- \chi_{\{u<0\}}$.
• To extend the known $C^1$ regularity result from the elliptic case to the parabolic setting in arbitrary finite dimensions.
• To analyze the behavior of the free boundary near branch points—points where $u=0$, $\nabla u=0$, and both phases meet.
• To show that the regularity is optimal, i.e., Lipschitz regularity is the best possible, by constructing a counterexample to $C^1$ regularity.

Proposed method

• Adapts the method of [10] from the elliptic to the parabolic case, using a two-stage proof of directional monotonicity to handle the lack of time derivative continuity.
• Employs a blow-up analysis and rescaling techniques to study the local structure of the free boundary near branch points.
• Uses a supremum-mean-value estimate and non-degeneracy properties to control the growth of the solution near zero sets.
• Applies a comparison principle via a two-stage argument to circumvent the lack of continuity in $\partial_t u$, which prevents direct use of standard comparison tools.
• Establishes equicontinuity of the spatial normal vectors to the free boundary by bounding the $C^{1/2,1}$ norm of the solution and using gradient estimates.
• Constructs a counterexample based on a one-phase solution with non-tangential free boundary contact, reflected to produce a two-phase solution with only Lipschitz regularity.

Experimental results

Research questions

• RQ1 Can the free boundary regularity result from the elliptic two-phase membrane problem be extended to the parabolic setting in higher dimensions?
• RQ2 What is the optimal regularity of the free boundary near branch points in the parabolic two-phase problem?
• RQ3 Is the time derivative $\partial_t u$ continuous at branch points, and how does this affect the regularity of the free boundary?
• RQ4 Can the regularity of the free boundary be quantified in terms of the data, such as the infimum of $\lambda_\pm$ and their Lipschitz norms?
• RQ5 Is $C^1$ regularity of the free boundary achievable, or is Lipschitz the best possible?

Key findings

• The free boundary $\partial\{u>0\} \cup \partial\{u<0\}$ is locally the union of two Lipschitz graphs in space near each branch point, with spatial gradients continuously differentiable.
• The Lipschitz constants and modulus of continuity of the spatial normal vectors depend only on $\inf \min(\lambda_+, \lambda_-)$, the Lipschitz norms of $\lambda_\pm$, the $L^\infty$ norm of $u$, and the space dimension $n$.
• The regularity result is optimal: there exists a counterexample showing that the free boundary is not generally $C^1$, even when $\lambda_+ = \lambda_- = 1$.
• The proof establishes directional monotonicity in two stages to overcome the lack of continuity in $\partial_t u$, a key obstacle in the parabolic setting.
• The equicontinuity of the spatial normal vectors is proven via rescaling and gradient estimates, ensuring the free boundary is locally a $C^{1,\alpha}$ graph in space.
• The counterexample is constructed by reflecting a one-phase solution with non-tangential free boundary contact at the origin, resulting in a two-phase solution with only Lipschitz regularity at the origin.