There is a quite short proof of the following variant of separation theorem:

Let $X$ be a normed space, $V$ convex and closed and $x\notin V$. Then there is a $x^*\in X^*$ so that $x^*(x) < \inf_{v\in V}x^*(v)$.

The proof I know relies on Hahn-Banach. Now I want to proof it in another way using following famous theorem:

Let $1<p<\infty$, $K$ be a closed convex set in $L^p(\mu)$ and $f\in L^p(\mu)$. Then there is a unique $g \in K$ such that $\vert\vert f-g\vert\vert_{L^p(\mu)}=dist(f,K)$.

Is it true that this works?


Yes, this works.

Hint: you know that $$\|f - g\|_p^p \le \|f - k\|_p^p$$ for all $k \in K$. Now, use $k = (1-t) \, g + t \, v$ for $v \in K$, $t \in (0,1]$ and let $t \searrow 0$.

In the case $p = 2$, you have $$0\ge\|f-g\|_2^2 - \|f - g + t \, (g-v)\|_2^2 \\= -2 \, t \, (f-g,g-v) - t^2 \, \|g-v\|^2 $$ Now, dividing by $t$ and $t \searrow 0$ implies $$ 0 \le 2 \, (f - g,g - v) $$ for all $ v \in K$. Hence, $(f-g,f) > (f-g,g) \ge (f-g, v)$.

  • $\begingroup$ Fist, thanks a lot for your fast answer, but I do not understand why your hint would help me. Letting $t$ goes to $0$ just gives me equality in your inequality one line above, isn't it? $\endgroup$ – tubmaster Jan 26 '17 at 18:29
  • $\begingroup$ Yes, of course. You had first divide by an appropriate power of $t$. It might be easier to first consider the Hilbert space case $p = 2$. $\endgroup$ – gerw Jan 26 '17 at 18:36
  • $\begingroup$ sorry, what do you mean with "divide by an appropriate power to $t$"? $\endgroup$ – tubmaster Jan 26 '17 at 18:39
  • $\begingroup$ You divide the inequality by $t^\alpha$ for some $\alpha > 0$ (depending on $p$). Then, you let $t \to 0$. $\endgroup$ – gerw Jan 26 '17 at 20:16
  • $\begingroup$ ah I understand, but I don't see to what limit it converge... $\endgroup$ – tubmaster Jan 26 '17 at 20:22

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.