The Löwner-John ellipsoid of a convex set $C$ is the minimum-volume ellipsoid (MVE) that encloses it. The ellipsoid can be computed using Khachiyan's method, and there are a number of approximations available if $C$ is (the convex hull of) a set of points.

Are there fast (i.e non-ellipsoid-method based) approximations to the MVE of a bounded polyhedron presented only in terms of the halfplanes whose intersections define it ? In particular, I'd be interested in methods that run in time polynomial in the dimension and the inverse error $1/\varepsilon$.

  • 2
    $\begingroup$ do we even know how to compute/approximate the radius of the polyhedron in this regime? because without that I don't even see a separation oracle for the ellipsoid algorithm $\endgroup$ – Sasho Nikolov Dec 22 '12 at 12:14

According to Boyd, it's NP-hard: http://youtu.be/mNzu42FrlHo?t=41m3s

| cite | improve this answer | |
  • $\begingroup$ Interesting. and he also suggests that the opposite is true for max volume inscribed ellipsoid (easy for the H-rep, but hard for the V-rep). I'm still hoping there are some good approiximations, so I'll hold off on accepting the answer, but I'll revisit in a day or two. $\endgroup$ – Suresh Venkat Dec 22 '12 at 22:54

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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