isoperimetric: Nonlinear Function
Created:
Modified:

isoperimetric

This page is from my personal notes, and has not been specifically reviewed for public consumption. It might be incomplete, wrong, outdated, or stupid. Caveat lector.

The isoperimetric problem: among all closed curves in the plane with equal perimeter, which encloses the largest volume? It's well-known that this is the circle.

A differentiable manifold MM satisfies a dd-dimensional isoperimetric inequality if for any open set DMD \subset M,

area(D)Cvolume(D)d1d\text{area}(\partial D) \ge C \text{volume}(D)^\frac{d-1}{d}

Here dd is called the 'isoperimetric dimension' of MM. In general, an isoperimetric inequality is a lower bound on a surface-area-to-volume ratio. In Euclidean space, this inequality corresponds to the fact that any body with the same volume as the unit ball must have larger surface area.In this case CC is known precisely since it follows from the formulae for the surface area and volume of a ball.

Probability measures: generalizing volume to an arbitrary measure μ\mu, surface area is defined in terms of the infinitesimal change in volume as the set DD is expanded to a new set DϵD^*_\epsilon that includes all points within an infinitesimal distance ϵ\epsilon of its previous boundary. The generalized isoperimetric problem is then: of all sets DD having a given measure μ(D)\mu(D), which has the smallest surface area μ(Dϵ)ϵ\frac{\partial \mu(D^*_\epsilon)}{\partial \epsilon}?

  • For Gaussian measure, the answer is not balls but half-spaces.

Concentration:

References: