2
$\begingroup$

Let $(T_1,..., T_n)\in \mathcal{L}(E)^n$ be a tuple of commuting normal operators (i.e. each $T_k$ is normal and $T_iT_j=T_jT_i$ for all $i,j$), where $E$ is a complex Hilbert space.

I want to show that $$\displaystyle\sup_{\|x\|=1}\bigg(\displaystyle\sum_{i=1}^n|\langle T_ix,x\rangle|^2\bigg)\geq\displaystyle\sup_{\|x\|=1}\bigg(\displaystyle\sum_{i=1}^n\|T_ix\|^2\bigg)\;.$$

$\endgroup$
6
  • $\begingroup$ Do you also assume $T_i^*T_j=T_jT_i^*$ for all $i,j$ or not? (you need it to get the representation with multiplication operators) $\endgroup$ Feb 17, 2018 at 20:17
  • $\begingroup$ @PietroMajer thank you but we don't need it. $\endgroup$
    – Student
    Feb 17, 2018 at 20:41
  • $\begingroup$ I'm just saying that for the representation with multiplication operators to hold, you need all the $2n$ operators $A_1,\dots, A_n,A_1^*,\dots, A_n^*$ to commute, that is, that they generate a commutative normal algebra. However, this is not required for your inequality to hold. $\endgroup$ Feb 18, 2018 at 13:35
  • $\begingroup$ @PietroMajer For the representation with multiplication operators to hold, we need only that the operators $A_k$ are normal and commuting. $\endgroup$
    – Student
    Feb 18, 2018 at 14:40
  • $\begingroup$ Yes, right: if they are normal and commuting, then $A_i^*A_j=A_jA_i^*$ follows as a consequence. $\endgroup$ Feb 18, 2018 at 15:44

2 Answers 2

4
+50
$\begingroup$

Consider the case of finite dimensional Hilber space. Then may suppose that operators $T_1,\dots,T_d$ are diagonal in the same orthonormal basis $(e_1,e_2,\dots)$, denote the diagonal of $T_i$ by $(p_{i1},p_{i2},\dots)$. The square of RHS is nothing but $\sup_j |p_{1j}|^2+|p_{2j}|^2+\dots+|p_{dj}|^2$. For fixed $j$, taking $x=e_j$ we get the same thing $|p_{1j}|^2+|p_{2j}|^2+\dots+|p_{dj}|^2$ for LHS.

Now do the same in general situation. Note that $$ \sup_{\|f\|_{L^2(\mu)}=1}\bigg[\displaystyle\sum_{i=1}^n\left(\int_Y|\varphi_i|^2|f|^2d\mu \right)\bigg]={\rm essup}\,\sum_i |\varphi_i|^2=:A. $$ Note that (by separability of a complex plane and by the definition of essential supremum) we may find discs $D_i$ such that $\mu(\Omega)>0$, where $\Omega=\{x:\varphi_i(x)\in D_i,i=1,\dots,n\}>0$ and $\sum_i |\theta_i|^2>A-\varepsilon$ whenever $\theta_i\in D_i$. Choose any function $f$ with $\|f\|=1$ concentrated in $\Omega$. Then $\int \varphi_i |f|^2d\mu\in D_i$ (indeed, $D_i$ are convex, thus a weighted mean of elements of $D_i$ belongs to $D_i$). It implies that LHS is not less than $A-\varepsilon$.

$\endgroup$
10
  • 1
    $\begingroup$ This is only the case $\dim E<\infty$ (or $T_j$'s with pure point spectrum), though a version of this should work in general (using the spectral theorem for commuting operators). $\endgroup$ Oct 30, 2017 at 16:46
  • 1
    $\begingroup$ Saying "we may suppose" I mean a certain limit argument involved. $\endgroup$ Oct 30, 2017 at 16:59
  • $\begingroup$ Where you have used that the operators are commuting? Thank you $\endgroup$
    – Student
    Nov 3, 2017 at 17:11
  • $\begingroup$ When I reduce the problem to the case of simultaneously diagonal operators. $\endgroup$ Nov 3, 2017 at 18:22
  • $\begingroup$ @Student yes, exactly. What is not rigorous in this answer? $\endgroup$ Feb 16, 2018 at 18:12
4
$\begingroup$

Remark: Note that this is an answer to the original version of the question, for general (normal) $T_j$'s; the OP has now added the assumption that the $T_j$ commute.


This is false, unless $d=1$, when it has an easy proof using the spectral theorem. For a counterexample, consider $$ T_1 = \begin{pmatrix} 1 & 0 \\ 0 & 0 \end{pmatrix}, \quad T_2 = \begin{pmatrix} 0 & -1 \\ 1 & 0 \end{pmatrix} . $$ The RHS of your inequality equals $\sqrt{2}$ (take $x=e_1$); on the LHS, $|\langle x, T_j x\rangle |\le \|T_j\|= 1$ for each summand individually for any $x$, so the only chance would be to make them both $=1$, but that's clearly not working since $\langle e_1, T_2 e_1\rangle = 0$.

$\endgroup$
1
  • 2
    $\begingroup$ they do not commute $\endgroup$ Oct 24, 2017 at 12:44

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.