Let $e_n$ a orthonormal basis for a Hilbert space and $T$ a bounded linear operator. Is the following correct?

$$\lVert T \lVert^2 \leq \sup_{n \in \mathbb{N}} \sum_{k \in \mathbb{N}} |\langle e_n,Te_k\rangle|^2$$

  • $\begingroup$ Your inequality is not homogeneous in $T$ (a square is missing in the LHS). The $\max$ should be a $\sup$ and note that by Bessel's inequality, the RHS is $\sup_n\lVert T^*e_n\rVert^2$. $\endgroup$ – Davide Giraudo Jun 15 '12 at 10:51

Since by Bessel's inequality $$\sum_{k\in\Bbb N}|\langle e_n,Te_k\rangle|^2=\sum_{k\in\Bbb N}|\langle T^*e_n,e_k\rangle|^2=\lVert T^*e_n\rVert^2$$ and $\lVert T\rVert=\lVert T^*\rVert$, the question is: given a linear bounded operator $T\colon H\to H$, do we have $\lVert T\rVert^2=\sup_{n\in\Bbb N}\lVert Te_n\rVert^2$? (the inequality $\geqslant$ is quite obvious)

The answer is no. Indeed, consider $T\colon \ell^2(\Bbb N)\to \ell^2(\Bbb N)$ given by $T(x)(n)=0$ for $n\geq 1$ and $T(x)(0)=\sum_{j=1}^{+\infty}\frac{x(j)}j$ (it is well-defined by Cauchy-Schwarz inequality). Such an operator is linear an bounded. Its norm is reached only for a multiple of the sequence $\{\frac 1j\}_{j\geq 1}$. Denote $e^{(n)}$ the sequence whose $n$-th term is $1$, the other are $0$ (this gives an orthonormal basis). We have $\lVert T(e^{(n)})\rVert=\frac 1n\leq 1<\lVert T\rVert=\frac{\pi}{\sqrt 6}$.

| cite | improve this answer | |
  • 2
    $\begingroup$ Thank you for your answer. Do you know if there is a different way to estimate the norm in terms of $<e_n, T e_k>$? $\endgroup$ – Rejnik Jun 15 '12 at 18:28
  • $\begingroup$ @Davide Could you explain the notation $T(x)(n)$ and $T(x)(0)$ please, thank you! $\endgroup$ – Xiao Aug 7 '14 at 10:41
  • $\begingroup$ $T(x)(n)$ is the $n$-th term of the sequence $T(x)$. $\endgroup$ – Davide Giraudo Aug 7 '14 at 11:32

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.