Epi-convergence

From HandWiki

In mathematical analysis, epi-convergence is a type of convergence for real-valued and extended real-valued functions. Epi-convergence is important because it is the appropriate notion of convergence with which to approximate minimization problems in the field of mathematical optimization. The symmetric notion of hypo-convergence is appropriate for maximization problems. Mosco convergence is a generalization of epi-convergence to infinite dimensional spaces.

Definition

Let X be a metric space, and fn:X a real-valued function for each natural number n. We say that the sequence (fn) epi-converges to a function f:X if for each xX

lim infnfn(xn)f(x) for every xnx and lim supnfn(xn)f(x) for some xnx.

Extended real-valued extension

The following extension allows epi-convergence to be applied to a sequence of functions with non-constant domain.

Denote by ={±} the extended real numbers. Let fn be a function fn:X for each n. The sequence (fn) epi-converges to f:X if for each xX

lim infnfn(xn)f(x) for every xnx and lim supnfn(xn)f(x) for some xnx.

In fact, epi-convergence coincides with the Γ-convergence in first countable spaces.

Hypo-convergence

Epi-convergence is the appropriate topology with which to approximate minimization problems. For maximization problems one uses the symmetric notion of hypo-convergence. (fn) hypo-converges to f if

lim supnfn(xn)f(x) for every xnx

and

lim infnfn(xn)f(x) for some xnx.

Relationship to minimization problems

Assume we have a difficult minimization problem

infxCg(x)

where g:X and CX. We can attempt to approximate this problem by a sequence of easier problems

infxCngn(x)

for functions gn and sets Cn.

Epi-convergence provides an answer to the question: In what sense should the approximations converge to the original problem in order to guarantee that approximate solutions converge to a solution of the original?

We can embed these optimization problems into the epi-convergence framework by defining extended real-valued functions

f(x)={g(x),xC,,x∉C,fn(x)={gn(x),xCn,,x∉Cn.

So that the problems infxXf(x) and infxXfn(x) are equivalent to the original and approximate problems, respectively.

If (fn) epi-converges to f, then lim supn[inffn]inff. Furthermore, if x is a limit point of minimizers of fn, then x is a minimizer of f. In this sense,

limnargminfnargminf.

Epi-convergence is the weakest notion of convergence for which this result holds.

Properties

  • (fn) epi-converges to f if and only if (fn) hypo-converges to f.
  • (fn) epi-converges to f if and only if (epifn) converges to epif as sets, in the Painlevé–Kuratowski sense of set convergence. Here, epif is the epigraph of the function f.
  • If fn epi-converges to f, then f is lower semi-continuous.
  • If fn is convex for each n and (fn) epi-converges to f, then f is convex.
  • If fn1fnfn2 and both (fn1) and (fn2) epi-converge to f, then (fn) epi-converges to f.
  • If (fn) converges uniformly to f on each compact set of n and (fn) are continuous, then (fn) epi-converges and hypo-converges to f.
  • In general, epi-convergence neither implies nor is implied by pointwise convergence. Additional assumptions can be placed on an pointwise convergent family of functions to guarantee epi-convergence.

References