A generalization of L\"owner-John's ellipsoid theorem

TL;DR

This paper generalizes the Lowner-John ellipsoid theorem to higher-degree homogeneous polynomials, allowing for non-convex sets and providing a convex optimization framework with uniqueness and characterization results.

Contribution

It introduces a convex optimization approach for finding minimal volume polynomial sublevel sets containing a given set, extending classical quadratic results to higher degrees without convexity assumptions.

Findings

The problem is convex even without convexity of the set.

Unique solutions exist for the polynomial case.

A numerical scheme using convex optimization and LMI constraints is proposed.

Abstract

We address the following generalization $P$ of the Lowner-John ellipsoid problem. Given a (non necessarily convex) compact set $K\subset R^n$ and an even integer $d$, find an homogeneous polynomial $g$ of degree $d$ such that $K\subset G:=\{x:g(x)\leq1\}$ and $G$ has minimum volume among all such sets. We show that $P$ is a convex optimization problem even if neither $K$ nor $G$ are convex! We next show that $P$ has a unique optimal solution and a characterization with at most ${n+d-1\choose d}$ contacts points in $K\cap G$ is also provided. This is the analogue for $d\textgreater{}2$ of the Lowner-John's theorem in the quadratic case $d=2$, but importantly, we neither require the set $K$ nor the sublevel set $G$ to be convex. More generally, there is also an homogeneous polynomial $g$ of even degree $d$ and a point $a\in R^n$ such that $K\subset G\_a:=\{x:g(x-a)\leq1\}$ and $G\_a$ has minimum volume among all such sets (but uniqueness is not guaranteed). Finally, we also outline a numerical scheme to approximate as closely as desired the optimal value and an optimal solution. It consists of solving a hierarchy of convex optimization problems with strictly convex objective function and Linear Matrix Inequality (LMI) constraints.