2
$\begingroup$

$(X, \|•\|) $and $(Y, \|•\|') $ be two normed space and $\begin{align} {\scr{B}}{(X, Y) }&=\{T\in {\scr{L}}{(X,Y)} : T \text { is bounded } \}\end{align}$

$\|T\|_{op}=\sup_{\|x\|\le1}\|Tx\|' $

I want to understand the convergence of sequences in the space ${(\scr{B}}{(X, Y) }, \|•\|_{op}) $

$T_n\to T$ pointwise on $X$ i.e $\forall x\in X, T_n x \to Tx$$ \text { in the space }\space \space$$ (Y, \|•\|') $.

Question : Does this implies $T_n\to T $ in the space ${(\scr{B}}{(X, Y) }, \|•\|_{op}) $?

  1. $T_n\to T $ pointwise implies $T\in {\scr{L}}{(X, Y) }$

Proof: $\forall$ $x, y\in X$ and $\lambda \in \mathbb{K}$

$\begin{align}T(x+\lambda y) &=\lim_{n\to\infty} T_n(x+\lambda y)\\&=\lim_{n\to\infty} T_nx +\lambda T_ny\\&=\lim_{n\to \infty} T_nx+\lambda \lim_{n\to\infty}T_ny\\&=Tx+\lambda Ty \end{align}$

  1. $T$ may not be continuous.

For an example, choose $ (\ell_1, \|•\|_{\infty}) $ and

$(f_n) \in (\ell_1,\|•\|_{\infty})^{*}$ defined by $$f_n(x) =\sum_{k=1}^{n} x_k$$

Where $x=(x_1, x_2,...) \in\ell_1$

Then, $f_n\to f$ pointwise where$$f(x) =\sum_{k\in \Bbb{N}} x_k$$

But $f\notin (\ell_1,\|•\|_{\infty})^{*}$

Here we need $(X, \|•\|) $ to be a Banach space to ensure that $T\in {\scr{B}}{(X, Y) }$ [Banach-Steinhaus Theorem]

  1. Suppose in addition, $(X, \|•\|) $ is a Banach space and then $T\in {\scr{B}}{(X, Y) }$ but still $(T_n)$ may not converge to $T$ in the space ${(\scr{B}}{(X, Y) }, \|•\|_{op}) $

For an example, choose $(c_0, \|•\|_{\infty})$ and $(f_n) $ defined by $$f_n(x) =x_n$$ where $x=(x_1, x_2,...) \in c_0$ i.e converges to $0$.

Then, $f_n\to 0 $ pointwise but $f_n \nrightarrow 0$ in the Dual space.

My question : What will be the next step to ensure that the pointwise convergence implies convergence in the operator norm?

$\endgroup$
1
  • 2
    $\begingroup$ There is no useful sufficient condition for that. $\endgroup$ Apr 27 at 8:24

1 Answer 1

4
$\begingroup$

As Kavi mentioned, there are no good answers to your question. Pointwise convergence is strictly weaker than convergence in $\|\cdot\|_{op}$, so you will need to add fairly strong hypotheses to the former in order to get the latter.

The following are examples of some of these additional hypotheses:

  1. The domain $X$ is finite-dimensional.

  2. All of the $T_n$ lie in some fixed finite-dimensional subspace of ${\scr B}(X, Y)$.

  3. $\{T_n\}_n$ is a Cauchy sequence for $\|\cdot\|_{op}$.

  4. When verifying that $T_n(x)→T(x)$, that is, that for every $ε>0$ there is some $n_0$ such that $n≥n_0 \Rightarrow \|T_n(x)-T(x)\|<\epsilon$, you are able to exhibit $n_0$ depending only on $\epsilon$ and $\|x\|$ (rather than $\epsilon$ and $x$).

$\endgroup$

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.