Would anyone be so kind to tell me if my answer to a problem is correct? Thank you! Happy new year!
The problem appeared on the UW-Madison Analysis Qual in January 2016.
Give an example of a non-empty closed subset of $L^2([0,1])$ that does not contain a vector of smallest norm.
I know an example that works. It's $\frac{n+1}{n}e_n$ where $e_n$ is an orthonormal basis for the $L^2([0,1])$ space. You can choose $e_n=\sqrt{2}\sin(n\pi x)$.
What I would like to know is the following:
My initial guess was $\{\sqrt{n+1}\chi_{[0,1/n]} : n\in \mathbb{N}\}$ where $\chi_{[0,1/n]}$ is the characteristic function of $[0,1/n]$. I thought this example worked because the $L^2$ norm of each term is $\sqrt{\frac{n+1}{n}}$, which is decreasing in $n$, so the smallest norm is never achieved. Does this example work? Is the set closed?
I have done some computation and realized that the distances between arbitrary terms does go to zero, which means we can't use the same argument as the orthonormal basis example.
Similar questions have been asked on this website:
Find a example in $L^2$, nonempty closed but contains no element of smallest norm.
Minimum principle in Hilbert space
counter example of minimum principle in incomplete inner product space
Best Answer
Well, it is closed. Assume a sequence from $\{\sqrt{n+1}\chi_{[0,1/n]}\}$ were convergent in $L^2$ and denote the limit by $g$. The sequence must have a subsequence that converges to $g$ a.e. Suppose $g$ is not in the set $\{\sqrt{n+1}\chi_{[0,1/n]}\}$. Since $\sqrt{n+1}\chi_{[0,1/n]}$ converges to zero function a.e. as $n\to\infty$, any non-constant convergent sequence must converge to the zero function a.e. So $g = 0$ a.e. However, $||\sqrt{n+1}\chi_{[0,1/n]}||_2>1$ and $||g||_2=0$, so there can't be an $L^2$ convergence. Contradiction. $g$ is in the set and the set is closed.