Consider $L_2[0,1]$ with the usual inner product $\langle f, g \rangle = \int_{0}^1 f(t)g(t) \, dt$ and define a new norm

$$ \| f \|^2_{\star} = \sum_{i=1}^\infty \langle f, \phi_i\rangle^2 \lambda_i $$

where $\phi_1, \phi_2, \ldots$ are an orthonormal basis for $L_2[0,1]$ and $\lambda_n$ is a sequence of non increasing postitive numbers such that $\sum_{i=1}^\infty \lambda_i < \infty$.

For some $0<c<1$ the indicator function $I_{[0,c]}(t)$ and define the operator $T : L_2 \to L_2$ such that $T(f) = f I_{[0,c]}$

Is T a bounded operator under $\| \|_{\star}$? (Does there exists $M >0$ such that $ \| T(f) \|_{\star} \leq M \|f\|_{\star}$.

I have tried bounding $\langle f I_{[0,c]} , \phi_i \rangle^2 = (\int_{0}^c f(t) \phi_i(t) \, dt )^2$ for each $i$ but failed.


The answer depends on the basis $\phi_n$ and its relation to the operator $I_{[0,c]}$. Lets construct an example where it doesn't work.

First note that $I_{[0,c]}$ is a self-adjoint projection with infinite dimensional image and kernel. Let $e_k$ be an ONB of the image and $\widetilde e_k$ an ONB of the kernel.

$\sigma:\Bbb N\to \Bbb N$ will be a bijection that we will describe more later. Let $\phi_{2k}=\frac{e_{\sigma(k)}+\widetilde{e_{\sigma(k)}}}{\sqrt 2}$ and $\phi_{2k+1}=\frac{e_k-\widetilde{e_{k}}}{\sqrt 2}$, $\lambda_{2k}=2^{-k}=\lambda_{2k+1}$.

Now let $f=\frac{e_k+\widetilde{e_k}}{\sqrt 2}$. You have $$\|f\|_{\star}^2=2^{-\sigma^{-1}(k)},\quad I_{[0,c]}f = \frac{e_k}{\sqrt 2}=\frac{\phi_{2\sigma^{-1}(k)}+\phi_{2k+1}}2,\quad \|I_{[0,c]}f\|_\star^2=2^{-\sigma^{-1}(k)-1}(1+2^{\sigma^{-1}(k)-k}).$$

If there exists some permutation of $\Bbb N$ so that the difference $\sigma^{-1}(k)-k$ can get arbitrarily large we are done. It is elementary to define such a permutation.

If on the other hand the $\phi_n$ all either lie in the image or in the kernel of $I_{[0,c]}$, it is simple to check that $I_{[0,c]}$ is a bounded operator of norm $1$.

  • $\begingroup$ This is great, but I am not sure how to check that $I_{[o,c]}$ is bounded of norm $1$ in the case all $\phi_n$ lie in the image. Could you expand a bit please. $\endgroup$ – Manuel Feb 25 at 20:43
  • $\begingroup$ also I belive itshould be a $\sqrt{2}$ instead of $2$ in the second equality for $I_{[0,c]}f$. $\endgroup$ – Manuel Feb 25 at 20:48
  • $\begingroup$ The factor $\frac12$ is correct, note that $\phi$ already has a $\sqrt2$ factor. For the norm of $I_{[0,c]}$, note that $\langle I_{[0,c]}f, \phi_k\rangle = \langle f, I_{[0,c]}\phi_k\rangle$ by symmetry of $I_{[0,c]}$. The term on the right is however either equal to $\langle f, \phi_k\rangle$ or $0$, depending on whether $\phi_k$ is in the image or in the kernel of $I_{[0,c]}$. What you get is that either terms in the sum $\|I_{[0,c]}f\|_\star^2$ drop out or remain the same, and thus $\| I_{[0,c]}f\|_\star^2≤\|f\|_\star^2$. $\endgroup$ – s.harp Feb 25 at 23:07
  • $\begingroup$ Many thanks. I failed to notice that $e_k$ where 0 after $c$. $\endgroup$ – Manuel Feb 26 at 17:12
  • $\begingroup$ Do you think that it is posible to obtain boundeness imposing other type of conditions over the base $\phi_n$? $\endgroup$ – Manuel Feb 26 at 17:14

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.