Take the 2-minute tour ×
MathOverflow is a question and answer site for professional mathematicians. It's 100% free, no registration required.

Let $K\subseteq \mathbb{R}^n$ be a full-dumensional convex body. The Löwner ellipsoid of $K$ is the unique ellipsoid of smallest volume containing $K$. My question is about a related object: the ellipsoid containing $K$ that minimizes the sum of squared axis lengths (the square root of this sum is what I called the Hilbert-Schmidt norm of the ellipsoid).

My question is a reference request: has the circumscribed ellipsoid of minimum Hilbert-Schmidt norm been studied?

Let me, for context, say briefly what I know about this ellipsoid. It is not hard to use convex duality to derive a characterization of this ellipsoid which is similar to John's decomposition of the identity. I can show that the minimizing ellipsoid is $E = FB_2^n$, for a linear map $F$ and $B_2^n$ the Euclidean $n$-dimensional ball, if and only if there exist contact points $v_1, \ldots, v_m \in K \cap E$ and positive reals $\mu_1, \ldots, \mu_m$, such that $$ \sum_{i=1}^m{\mu_i v_i} = 0 \ \text{ and }\ \sum_{i=1}^m{\mu_i v_i \otimes v_i} = (FF^T)^2. $$ The minimizer is unique because the objective is strictly convex. The only difference with John's theorem is that $(FF^T)^2$ has replaced $FF^T$.

It follows that the minimizer $E$ is a ball if and only if the Löwner ellipsoid of $K$ is a ball, although this fact has a more direct proof using the arithmetic mean-geometric mean inequality.

share|improve this question
add comment

1 Answer 1

up vote 2 down vote accepted

One reference that I could locate is the following Minimum norm ellipsoids as a measure in high-cycle fatigue criteria by Nestor Zouain, presented at a conference in 2005 (see $\S5$ of that pdf). However, I believe this question must have been considered earlier---if I find something older I'll update my answer.

share|improve this answer
    
Thanks Suvrit! I also think it must've been considered. –  Sasho Nikolov Nov 27 '13 at 18:38
add comment

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.