Nonlocal basis pursuit: Nonlocal optimal design of conductive domains in the vanishing material limit

TL;DR

This paper investigates the limiting behavior of nonlocal optimal design problems for conductive materials as the available material vanishes, revealing new solution spaces and a simplified approach to the local limit.

Contribution

It extends the understanding of nonlocal optimal design problems in the vanishing material limit, showing solutions in Lebesgue spaces and simplifying the transition to local problems by disregarding antisymmetry.

Findings

Nonlocal problems admit solutions in Lebesgue spaces with mixed exponents.

Disregarding antisymmetry transforms one-sided estimates into true limits.

The approach generalizes Sobolev space characterizations to mixed Lebesgue exponents.

Abstract

We consider the problem of optimal distribution of a limited amount of conductive material in systems governed by local and non-local scalar diffusion laws. Of particular interest for these problems is the study of the limiting case, which appears when the amount of available material is driven to zero. Such a limiting process is of both theoretical and practical interest and continues to be a subject of active study. In the local case, the limiting optimization problem is convex and has a well understood basis pursuit structure. Still this local problem is quite challenging both analytically and numerically because it is posed in the space of vector-valued Radon measures. With this in mind we focus on identifying the vanishing material limit for the corresponding nonlocal optimal design problem. Similarly to the local case, the resulting nonlocal problem is convex and has the basis pursuit structure in terms of nonlocal antisymmetric two-point fluxes. In stark contrast with the local case, the nonlocal problem admits solutions in Lebesgue spaces with mixed exponents. When the nonlocal interaction horizon is driven to zero, the "vanishing material limit" nonlocal problems provide a one-sided estimate for the corresponding local measure-valued optimal design problem. The surprising fact is that in order to transform the one-sided estimate into a true limiting process it is sufficient to disregard the antisymmetry requirement on the two point fluxes. This result relies on duality and requires generalizing some of the well known nonlocal characterizations of Sobolev spaces to the case of mixed Lebesgue exponents.