1
$\begingroup$

This question is more or less a cross post of https://math.stackexchange.com/questions/1218660/orthogonal-functions-with-shrinking-support.

Let $X$ be a metric space (compact, if it helps) and let $Y$ be a closed subset of $X$. Fix a Borel measure $\mu$ on $X$. My question:

Is there an orthogonal set $\{u_n\}$ of functions in $L^2(X,\mu)$ such that the support of $u_n$ lies in a $2^{-n}$-neighborhood of $Y$?

I can construct such a set in the case $X = \mathbb{R}$ and $Y = \{0\}$ as follows. Partition the interval $[-2^{-n},2^{-n}]$ into subintervals $I_1, \ldots, I_{2^n}$ of equal length and define

$$u_n(t) = \begin{cases} 1 & t \in I_{2k+1} \\ -1 & t \in I_{2k} \\ 0 & t \notin [-2^{-n},2^{-n}] \end{cases}$$

This construction does not obviously generalize to much more general metric spaces. My only other idea was to apply Gram-Schmidt to an arbitrary sequence of functions whose supports converge to $Y$, but it is not clear how to do this while maintaining control of the supports.

$\endgroup$
1
  • $\begingroup$ Do you want an orthonormal set $\{u_n\}$? Otherwise, $u_n\equiv0$ for all $n$ would be an obvious solution. But then, what about the situation that $X=X_0\sqcup \{\operatorname{pt}\}$, $\operatorname{dist}(\operatorname{pt},X_0)>0$, and $Y=\{\operatorname{pt}\}$? $\endgroup$
    – ifw
    Apr 6, 2015 at 11:02

1 Answer 1

2
$\begingroup$

As was pointed out by ifw, we should forbid $u_n$ from being identically $0$ (a.e.), in order to make the question interesting. Also, addressing ifw's additional observation, if we let $Y_n$ be the closed $2^{-n}$-neighborhood of $Y$, then we need to assume $L^2(Y_n,\mu|_{Y_n})$ is infinite-dimensional. (If $X$ is compact, or even just complete and separable, and $\mu$ is $\sigma$-finite, this simply means that the restriction of $\mu$ to $Y_n$ is not the sum of finitely many atoms.)

Under the above assumption on the dimension, it is easy to construct $\{u_n\}$ by induction. For each $k < n$, the inner product with $u_k$ defines a linear functional on $L^2(Y_n,\mu|_{Y_n})$. Since there are only finitely many linear functionals, and $L^2(Y_n,\mu|_{Y_n})$ is infinite-dimensional, there is some nonzero $u_n$ that is in the kernel of all of them.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.