2
$\begingroup$

Let $X$ be a Banach space. For $\epsilon \in (0,2]$, define:

$$\delta_X(\epsilon) = \inf_{x,y \in X}\{1 - \|\frac{1}{2}(x + y)\| : \|x\| = \|y\| = 1, \|x-y\| \ge \epsilon\}.$$

Then we say that $X$ is uniformly convex if $\delta_X(\epsilon) > 0$ for any$\epsilon \in (0,2]$. Furthermore, we say that $X$ is convex of power type $p$ (or $p$-convex) if there exists some $C>0$ so that $\delta_X(\epsilon) \ge C\epsilon^p$ for any $\epsilon \in (0,2]$.

I would like the show that, if $(\mathcal{H}, (\cdot,\cdot))$ is a Hilbert space, then $\mathcal{H}$ is of power type $2$.

I do know that every Hilbert space is uniformly convex, and that this proof uses the parallelogram law. For if we have $x, y \in \mathcal{H}$ and $\epsilon \in (0, 2]$ with $\|x\| = \|y\| = 1$ and $\|x-y\| \ge \epsilon$, we get:

$$\|x+y\|^2 = \|x+y\|^2 +\|x-y\|^2 - \|x-y\|^2 = 2\|x\|^2 + 2\|y\|^2 - \|x - y\|^2 \le 4- \epsilon^2.$$

We can then set $\eta = 1 - \frac{1}{2} \sqrt{4 - \epsilon^2}>0$ and perform a few manipulations to get $\eta \le 1 - \|\frac{1}{2}(x+y)\| \implies \delta_{\mathcal{H}}(\epsilon) \ge \eta.$

I'm wondering if the parallelogram law is also enough to prove $2$-convexity, or if I need some more powerful machinery?

Hints or solutions are greatly appreciated.

$\endgroup$
2
$\begingroup$

No, you don't need another tool. Parallelogram law is good enough:

One has $$4=\|x+y\|^2+\|x-y\|^2\ge \|x+y\|^2+\epsilon^2.$$ So $$\epsilon^2\le 4-\|x+y\|^2=(2-\|x+y\|)(2+\|x+y\|).$$ Can you find the constant $C$ so that $1-\|\frac12(x+y)\|\ge C\epsilon^2$ from here?

$\endgroup$
  • 1
    $\begingroup$ To finish, we use the fact that $\|x\| = \|y\| = 1$. So this gives $(2- \|x+y\|)(2 + \|x + y\|) \le 2(1- \frac{1}{2}\|x+y\|)(2 + \|x\| + \|y\|) \le 8(1 - \frac{1}{2} \|x + y\|)$. So we can use $C = \frac{1}{8}$. $\endgroup$ – JZS Aug 13 '14 at 23:27

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.