I'm working my way through the book Introduction to Hilbert Spaces with Applications and trying to do the exercise of proving a theorem (1.5.11) (its proof is left as an exercise).
It states that if $L: E_1 \rightarrow E_2$ is a continuous linear mapping, then:
- The nullspace $\mathcal{N}(L)$ is a closed subspace of $E_1$
- If the domain $\mathcal{D}(L)$ is closed then the graph $\mathcal{G}(L)$ is a closed subspace of $E_1 \times E_2$
The first statement was pretty straight forward, but I'm having a few concerns about my solution (bottom) to the second and would like to ask for help with:
- How does the choice of norm on $\mathcal{G}(L)$ affects the answer? The book has yet to suggest one for that space and I used an alternative I found elsewhere. But is this statement true regardless of the norm chosen and if so how can I show that?
- Does my solution look reasonable?
Thanks!
Since $E_1$ and $E_2$ were not specified at all I assumed they're both normed spaces for continuity to be definable. Furthermore, given $(x, y) \in E_1 \times E_2$, I take the norm of $E_1 \times E_2$ to be $||(x, y)||_{1, 2} = ||x||_1 + ||y||_2$.
To prove the statement I took the path of showing that $\mathcal{G}(L)$ is closed by asserting that every sequence $(z_n)$ in $\mathcal{G}(L)$ with $z_n \rightarrow z$, implies $z \in \mathcal{G}(L)$. Take any such sequence. By the definition of the graph, for every $n$ there exists $x_n \in \mathcal{D}(L)$ such that
\begin{equation} z_n = (x_n, Lx_n) \end{equation}
Taking $z = (x, \zeta)$ we have $(x_n, Lx_n) \rightarrow (x, \zeta)$, or \begin{equation} \forall \varepsilon > 0, \, \exists N \in \mathbb{N} \text{ such that } ||(x_n, Lx_n) - (x, \zeta)||_{1, 2} = ||x_n - x||_1 + ||Lx_n - \zeta||_2 < \varepsilon \ \forall \, n \ge N \,. \end{equation}
This implies $x_n \rightarrow x$ and $Lx_n \rightarrow \zeta$. Since $\mathcal{D}(L)$ closed, $x \in \mathcal{D}(L)$. Also, because $L$ is continuous, $\zeta = Lx$. Therefore, $z \in \mathcal{G}(L)$. $$\tag*{$\Box$}$$
Your solution looks good. In regards to the choice of norm on the graph I believe that unless the operator is bounded then the only appropriate (in the sense of convergence) choice of norm is the graph norm itself. Notice that, \begin{align} \|x\|_{E_{1}}\leq\|x\|_{E_{1}}+\|Tx\|_{E_{2}}=:\|x\|_{T}, \end{align} and if $T$ is not bounded then the graph norm is stronger than the norm on $E_{1}$. Now clearly you have that $T$ is bounded, but if $T$ is not bounded then convergence in $\|\cdot\|_{E_{1}}$ implies nothing about convergence in $\|\cdot\|_{T}$.
As you may have worked out however, is if $T$ is bounded then $\|\cdot\|_{E_{1}}$ and $\|\cdot\|_{T}$ are equivalent norms. In which case if, \begin{align} \|x_{n}-x\|_{E_{1}}\to0, \end{align} then, \begin{align} \|Tx_{n}-\zeta\|_{E_{2}}\to0. \end{align}
The point here is that what you have shown ($T$ continuous $\implies$ graph$(T)$ closed) is in fact an if and only if statement (please do not take what I have written to be a proof, it is not).