1

Let $H$ be a separable Hilbert space. Let $\{e_n\}$ be an orthonormal basis for $H$. Let $(y_n)$ be a bounded sequence in $H$. Suppose if I prescribe $T(e_i)=y_i$, can I extend $T$ as a bounded linear operator from $H \to H$. We know span$(e_i)$ is dense in $H$. We can extend linearly to $span(e_i)$. But I am not able to show it is bounded as a linear map from $span(e_i) \to H$. If we an prove this, then clearly it can be extended to a bounded operator to the closure of $span(e_i)$ which is full of $H$. I started thinking of this by the definition of the shift operator which is defined everywhere as follows. Let $H$ be a separable Hilbert space. Let $\{e_n\}$ be an orthonormal basis for $H$. Define $T(e_n)=e_{n+1}$ and extend this to $H$. I am confused how can we extend this to $H$ as a bounded operator.

budi
  • 1,920
  • 1
    In general, you can’t. For example, if $y_n = e_1$ for all $n$, then you can easily see the linear extension to $\text{span}(e_i)$ is not bounded. The shift operator works because $y_n = e_{n+1}$ is orthonormal, from which you can show that the linear extension to $\text{span}(e_i)$ is bounded, in fact, even isometric, so it can be extended to the entirety of $H$. – David Gao Mar 19 '25 at 09:59
  • 1
    You can do this if $\sum |y_n|^{2}<\infty$. – Kavi Rama Murthy Mar 19 '25 at 10:02
  • On computing I am getting $||Tx||=||x||$ for $x$ in the span of $(e_i)$. If we have a bounded operator on a dense set we can xtend it to the full space preserving the norm. Instead of a bounded sequence $(y_n)$, if I take an orthonormal sequence $y_n$, can I get the same? – budi Mar 19 '25 at 10:05
  • @budi I’ve essentially said as much in my comment: yes, as long as $y_n$ is orthonormal, then $T$ is isometric on $\text{span}(e_i)$. – David Gao Mar 19 '25 at 10:07
  • @geetha290krm Can you elaborate little more? – budi Mar 19 '25 at 10:07
  • @budi geetha290krm‘s comment is quite clear. Insofar as $\sum |y_n|^2 < \infty$, you can prove $T$ is bounded on $\text{span}(e_i)$ and so can be extended to the entirety of $H$. To prove this, use the Cauchy-Schwarz inequality. – David Gao Mar 19 '25 at 10:10
  • 1
    Without any orthogonality condition on $(y_n)$ you can define $T (\sum a_n e_n)=\sum a_n y_n$ and use Cauchy-Schwartz inequality to prove that $T$ is well-defined and bounded as long as $\sum |y_n|^{2}<\infty$. – Kavi Rama Murthy Mar 19 '25 at 10:11
  • Let $x=\sum_{k=1}^n<x,e_k>e_k$ in span of $e_i$. so that $Tx=\sum_{k=1}^n<x,e_k>y_k$. So $||Tx||\leq \sum_{k=1}^n|<x,e_k>|||y_k||$ and I am getting stuck in applying the Cauchy Schwarz inequality – budi Mar 19 '25 at 10:25
  • @budi By C-S, $|Tx| \leq \sum_k |\langle x,e_k\rangle||y_k| \leq (\sum_k |\langle x,e_k\rangle|^2)^{1/2}(\sum_k |y_k|^2)^{1/2}$. The first term is just $|x|$ as $(e_k)$ is an orthonormal basis, and the second term is finite by assumption. – David Gao Mar 19 '25 at 10:31
  • Thank you so much. I was confused with Cauchy Schwarz inequality involving inner product. – budi Mar 19 '25 at 10:34
  • In case $\sum |y_n|^2<\infty $ we obtain a Hilbert-Schmidt operator, a compact operator in particular. – Ryszard Szwarc Mar 19 '25 at 21:38

1 Answers1

1

As David correctly argued in this comment, such a bounded extension is not always possible: choosing $y_n:=e_1$ for all $n$ would mean that $T$ maps $\sum_j\frac{e_j}{j}\in\mathcal H$ to $(\sum_j\frac1j)e_j$, but $\sum_j\frac1j=\infty$, a contradiction. However, cases where this works are

Proposition. Let any orthonormal basis $(e_i)_{i \in I}$ of $\mathcal{H}$ and a family of pairwise orthogonal vectors $(y_i)_{i \in I} \subseteq \mathcal{H}$ (i.e., $\langle y_i,y_j\rangle=0$ for all $i\neq j$) with $\sup_{i \in I} \| y_i \| < \infty$ be given. Then there exists unique $T \in \mathcal{B}(\mathcal{H})$ with $T e_i = y_i$ for all $i \in I$. In this case $\| T \| = \sup_{i \in I} \| y_i \|$.

Proof. Define $ \mathcal{H}_0 := \operatorname{span} \{ e_i : i \in I \} $ and a linear map $ T_0:\mathcal{H}_0\to\mathcal{H} $ via $ T_0 e_i := y_i$ for all $i \in I$. Now for every $ x \in \mathcal{H}_0 $ there exist $ i_1, \dots, i_m \in I $ such that $ x = \sum_{j=1}^{m} \langle x,e_{i_j} \rangle e_{i_j} $. Using orthogonality, the Pythagorean theorem then implies $ \|x\|^2 = \sum_{j=1}^{m} |\langle x,e_{i_j} \rangle|^2 $ as well as \begin{align*} \|T_0 x\|^2 = \Big\| \sum_{j=1}^{m} \langle x,e_{i_j} \rangle y_{i_j} \Big\|^2 &=\sum_{j=1}^{m} |\langle x,e_{i_j} \rangle|^2 \|y_{i_j}\|^2 \\ &\leq \left( \sup_{i \in I} \|y_i\|^2 \right) \sum_{j=1}^{m} |\langle x,e_{i_j} \rangle|^2 = \left( \sup_{i \in I} \|y_i\|^2 \right) \|x\|^2. \end{align*} This shows $ \|T_0\| \leq \sup_{i \in I} \|y_i\| < \infty $ so by the BLT theorem there exists a bounded extension $T\in\mathcal B(\mathcal H)$ of $T_0$ with the same norm. Finally, uniqueness is evident from continuity. $\square$

This proposition works for the shift operator you mentioned and, more importantly, this shows that a bounded operator is unitary if and only if it maps one (every) orthonormal basis to an orthonormal basis again. Hence for every two orthonormal bases there exists a unitary mapping one to the other.

Frederik vom Ende
  • 5,187
  • 1
  • 11
  • 39
  • 1
    Another possibility: $|y_n|1\le C$ and $\sum{k=1}^\infty |\langle y_n,e_k\rangle |\le D.$ Then the operator is bounded by the Schur test. – Ryszard Szwarc Mar 20 '25 at 09:33