← Back to arXiv
arXivAnalysis / PDEsarXiv:2609.13215

Ball Rigidity of Local Minimizing Domains for the Best Fractional Sobolev Constant: The Subquadratic Case

The paper addresses a geometric optimization problem: among all regions (domains) of a fixed volume in space, which shape minimizes a certain mathematical quantity called the "best fractional Sobolev constant"? This constant measures how efficiently a region can control the behavior of functions defined on it, using a nonlocal notion of smoothness that involves interactions between points at a distance, governed by a parameter s between 0 and 1. A previous paper had shown that for the case where p is at least 2, the answer is a ball, and the authors of that paper conjectured the same should be true when p is between 1 and 2. The present paper confirms this conjecture: any locally optimal domain for this problem must be a ball.

The central challenge is that when p is between 1 and 2, a key function in the problem behaves badly near the boundary of the region, specifically it fails to be Lipschitz continuous, which breaks a classical geometric technique called the moving-plane method. This method works by reflecting the domain across hyperplanes to prove symmetry, but it requires comparing values of the function near the boundary in a controlled way. To restore this control, the authors develop a new technical tool, a weighted maximum principle that works even in thin regions near the boundary, where the absorption rate is carefully calibrated to the geometry of the problem.

A second major difficulty arises at corners, meaning points where different boundary reflections intersect. Near such points, the function has a complicated layered structure involving powers of the distance to the boundary with logarithmic corrections appearing at a critical resonance. The authors carefully expand the function in terms of these layers and show that when comparing the function on two sides of a corner, all the lower-order contributions cancel out, leaving just enough information to continue the moving-plane argument. The result holds without assuming the domain is convex and covers the full range of both parameters, completing the picture for this class of shape optimization problems.

Read original →