0
$\begingroup$

Suppose $X$ is a Hilbert space with norm $||.||$ and $K$ is a weak compact and convex subset of $X$.

The supporting functional: $$h(x^*)=\sup_{x\in K} \langle x^*, x \rangle$$

The indicator function: $$\delta (x|K)=0 \phantom{0}if \phantom{0} x\in K \phantom{0} and \phantom{0}+\infty\phantom{0} if \phantom{0}x\notin K$$

What is the relation between the following two duality theorems? Are they equivalent and 2 implies 1?

  1. $$\delta (x|K)=\sup_{x^*\in X} \langle x^*, x \rangle - h(x^*)$$

  2. $$ \inf_{x'\in K}||x- x' ||= \max_{||x^*||\leq 1} \langle x^*, x \rangle - h(x^*)$$

  3. Will the following claim be true: Fix any $1\leq M< +\infty$,

$x\in K$ if and only if for all $x^*$ satisfying $||x^*||\leq M$, we have

$$ \sup_{||x^*||\leq M} \langle x^*, x \rangle - h(x^*)=0.$$

$\endgroup$
0
$\begingroup$

The statement (1) is the observation that $\delta(\cdot|K)$ is the convex conjugate of $h$ (and vice-versa). So (1) is always true under your assumptions. The function $h$ is usually called the support functional of $K$.

The statement (2) states that there is no duality gap between the primal problem $$ \min_{x'\in X} \|x-x'\| + \delta(x'|K) $$ and its dual $$ \max_{x^*\in X}\ \delta(x^*|B) + \langle x^*,x\rangle - h(x^*) $$ with $B$ being the unit ball. Observe that $f^*(x^*)=\delta(x^*|B) + \langle x^*,x\rangle$ is the conjugate of $f(x')=\|x-x'\|$.

Question (3) is related to the primal problem above by a scaling with $M$, $$ \min_{x'\in X}\ M^{-1} \|x-x'\| + \delta(x'|K) $$

$\endgroup$
  • $\begingroup$ Thanks for the answer. So (3) is also a true statement? $\endgroup$ – user91360 Jun 8 '15 at 11:38

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.