1
$\begingroup$

Let $H$ be a Hilbert space with ONB $\{b_1, b_2, \cdots \}$. Let $x_n = \frac{1}{n}\sum_{i=1}^{n^2} b_i$.

Let $A=\overline{\operatorname{co}(\{x_1, x_2, x_3, \cdots \})}$, the closure of the convex hull of the $x$'s.

I want to prove that $0\in A$.

So I need to find a sequence $y_n\in \operatorname{co}(\{x_1, x_2,\cdots \})$ such that $\| y_n \|\to 0$. I've tried all that I could think of, e.g. $y_n = 1/n x_n + (1-1/n) x_{n-1}$, but in each case I found $\|y_n\| \to 1$.

Any hint would be MUCH appreciated.

$\endgroup$
6
  • $\begingroup$ Are you sure it's supposed to be the norm closure and not maybe the weak closure? $\endgroup$
    – David C. Ullrich
    Jan 11, 2018 at 15:27
  • $\begingroup$ Yes the norm closure, @DavidC.Ullrich $\endgroup$
    – user519975
    Jan 11, 2018 at 15:29
  • 1
    $\begingroup$ Assume $0 \notin A$. Reread Hahn-Banach. $\endgroup$
    – Daniel Fischer
    Jan 11, 2018 at 15:37
  • $\begingroup$ @DanielFischer Touche... $\endgroup$
    – David C. Ullrich
    Jan 11, 2018 at 15:55
  • $\begingroup$ Which version of Hahn Banach do you refer to @DanielFischer ? $\endgroup$
    – user519975
    Jan 11, 2018 at 16:10

1 Answer 1

Reset to default
2
$\begingroup$

Hint: Say $f,g$ are in some $L^2$ space, $||f||_2=||g||_2=1$, and $f$ and $g$ have disjoint support. Then $||(f+g)/2||_2=1/\sqrt 2$. Similarly for $(f_1+\dots+f_n)/n$.

Now if $n$ is much much larger than $m$ then your $x_n$ and $x_m$ "almost" have disjoint support. Show that $$\lim_{n\to\infty}\langle x_n,x_m\rangle=0.$$

$\endgroup$
2
  • $\begingroup$ No! The point is you need to take $n_2$ muuch larger than $n_1$, then $n_3$ mmuch much larger than $n_2$, and consider $y=(x_{n_1}+\dots x_{n_N})/N$. If the $x_{n_j}$ were orthogonal what would $||y||$ be? You can make them as close to orthogonal as you want... $\endgroup$
    – David C. Ullrich
    Jan 11, 2018 at 17:04
  • $\begingroup$ It's a norm-closed, hence weakly closed, subset of the closed unit ball, so it's weakly compact. Not norm compact, since $(x_n)$ has no norm-convergent subsequence (since $x_n$ and $x_m$ are close to orthogonal if $m$ is much larger than $n$, so that $||x_n-x_m||$ is close to $\sqrt 2$.) If this is in regard to Daniel Fischer's comments you might note that $\{0\}$ is compact. $\endgroup$
    – David C. Ullrich
    Jan 12, 2018 at 16:12

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.