Optimization under second order constraints: are the finite element discretizations consistent ?

TL;DR

This paper examines the consistency of finite element discretizations for second order constraints like convexity, showing non-convergence issues related to mesh richness and exploring the convergence of alternative discretization methods.

Contribution

It improves previous non-convergence results by linking them to mesh richness and investigates the consistency of various finite element discretizations for second order constraints.

Findings

Non-convergence is due to mesh richness limitations.

Some discretizations of second order constraints are consistent.

Numerical results support the convergence of a proposed method.

Abstract

It is proved in Chon{\'e} and Le Meur (2001) that the problem of minimizing a Dirichlet-like functional of the function $u\_h$ discretized with $P\_1$ Finite Elements, under the constraint that $u\_h$ be convex, cannot converge. Here, we first improve this result by proving that non-convergence is due to the mesh refinment lack of richness, remains local and is true even for any mesh. Then, we investigate the consistency of various natural discretizations ($P\_1$ and $P\_2$) of second order constraints (subharmonicity and convexity) without discussing the convergence. We also numerically illustrate convergence of a method proposed in the literature that is simpler than existing methods.