I came across the following technical question, to which I could not - after some time of thinking - find an answer:

Let $\mathcal{U},\mathcal{H}$ be two real (in general infinite dimensional) separable Hilbert spaces. For some linear subspace $\bar{U} \subseteq \mathcal{U}$, let $\Pi_U$ denote the orthogonal projection on this subspace. The question is:

Is the mapping $T \mapsto \Pi_{\text{ker}T}$ measurable from $L(\mathcal{U},\mathcal{H})$ to $L(\mathcal{U})$ when both spaces are endowed with the strong operator topology (and measurability is meant w.r.t. the Borel-$\sigma$-algebra of the SOT on both sides)?

Any hints and thoughts on this are more than appreciated!

  • $\begingroup$ can you write the projection as $lim 1 - (T^*T)^{\frac 1 n}$, and therefore the limit of continuous funtions ? $\endgroup$
    – user83457
    Mar 13, 2018 at 15:34
  • $\begingroup$ Could you please elaborate on this? I can't work your argument out. Would appreciate it! $\endgroup$ Mar 15, 2018 at 9:40
  • $\begingroup$ It looks like the answer below does that $\endgroup$
    – user83457
    Mar 15, 2018 at 11:09
  • $\begingroup$ I (partially) agree. However, how do you deal with the two questions I added below? $\endgroup$ Mar 15, 2018 at 12:35

1 Answer 1


It is true if you consider the unit ball. (Which should be sufficient because $\Pi_{\ker(\lambda T)} = \Pi_{\ker(T)}$ for all $\lambda\neq 0$.)

Also note that $\ker(T) = \ker(T^\ast T)$. I will therefore restrict my attention to normal operators $T$. That may or may not be sufficient for the application you have in mind.

Then we have $\Pi_{\ker(T)} = 1_{\{0\}}(T)$, where we use the measurable functional calculus.

If we define $f_n(x) = \begin{cases} (1-|x|)^n & |x|\leq 1 \\ 0 & |x|\geq 1\end{cases}$ we have a sequence continuous, uniformly bounded functions on $\mathbb{C}$ with $f_n\to 1_{\{0\}}$ pointwise. Therefore $f_n(T) \xrightarrow[n\to\infty]{s.o.t} 1_{\{0\}}(T)$.

It is therefore sufficient to prove that $B_1(L(U))_{sot} \to L(U)_{sot}, T\mapsto f(T)$ is continuous for all continuous $f:\mathbb{R}\to[0,1]$. If $T_i\xrightarrow{s.o.t} T$ and $x\in U$ arbitrary, then approximate $f$ uniformly on the unit disk $\mathbb{D}=\{z\in\mathbb{C} : |z|\leq1\}$ by some polynomial $p$. Now remember that $p(X)-p(Y) = \tilde{p}(X,Y)(X-Y)$ for some polynomial $\tilde{p}$ of degree $\deg(p)-1$ and that $\mathbb{D}$ contains $\sigma(T)$ for all $T\in B_1(L(U))$ so that:

$\begin{align*} \|f(T)x-f(S)x\| &\leq \|p(T)x-p(S)x\| + 2\epsilon\|x\| \\ &\leq \|\tilde{p}(T,S)\cdot (T-S)x\| + \epsilon\|x\| \\ &\leq const\|(T-S)x\| + 2\epsilon\|x\| \end{align*}$

Now if $T\xrightarrow{s.o.t}S$, then this proves $\forall\epsilon>0: \limsup_{T\to S} \|f(T)x-f(S)x\|\leq 2\epsilon\|x\|$ to that $f(T)\xrightarrow{s.o.t}f(S)$.

  • 2
    $\begingroup$ How does non-measurability follow from that? $\endgroup$ Mar 13, 2018 at 14:56
  • $\begingroup$ It doesn't. I thought you asked about continuity. $\endgroup$ Mar 13, 2018 at 15:22
  • $\begingroup$ I have edited the answer. $\endgroup$ Mar 13, 2018 at 16:04
  • $\begingroup$ Thank you very much for your answer! I have no experience with measurable functional calculus, so right now I cannot follow all of your arguments. However, I see your general line of argument on how to obtain the measurability. But intuitively: Concerning the reduction to the unit ball, isn't it a problem that the constant $\lambda$ can certainly not be chosen independent of $T$? Wouldn't you need to be able to push an arbitrary open set of $L(U,H)$ into the unit ball? $\endgroup$ Mar 14, 2018 at 12:19
  • $\begingroup$ Why do the weak and strong operator topology coincide on bounded sets of $L(U,H)$, $L(H,U)$ respectively? I could neither prove this myself nor find a source, which states this statement. Can you help me out on this? $\endgroup$ Mar 15, 2018 at 9:07

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.