2
$\begingroup$

I am studying operator calculus of complex-valued bounded Borel measurable function. In our textbook it is induced by Gelfand representation over $A_N$, which is the smallest C* algebra generated by normal operator $N$. It turns out that $$\phi (\psi(N))=(\phi \circ \psi) (N), \forall \phi, \psi \text{ continuous.}$$ which can be shown using Gelfand representation. Does it still correct in the case $\phi, \psi$ is bounded Borel measurable? I have seen the spectrum decomposition in this case, but don’t know if it can helps. Thanks in advance.

$\endgroup$
1
$\begingroup$

I know at least one variation that is true: Consider the following theorem from Murphy's excellent text "$C^*$-algebras and operator theory":

Theorem 2.5.7 (p73): Let $u$ be a normal operator on the Hilbert space $H$ and let $g: \mathbb{C}\to \mathbb{C}$ be a continuous function. Then $(g\circ f)(u) = g(f(u))$ for all $f \in B_\infty(\sigma(u)).$

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ It helps! Thank you so much! $\endgroup$ – milky sausage Oct 30 at 10:10
  • 1
    $\begingroup$ First I thought that one could get a more general result by approximating $g\in B_\infty(\sigma(f(u)))$ with continuous functions. When this is possible in the $\Vert\cdot \Vert_{L^\infty}$-norm, then the equality should persist. However, continuous functions are unfortunately not dense in Borel functions. $\endgroup$ – Jan Bohr Oct 30 at 10:13
0
$\begingroup$

The answer is affirmative for all Borel functions $f$ and $g$. Here is the reason.

First of all let me say that the most concrete form of the Spectral Theorem I know asserts that, given any normal operator $T$ on a separable Hilbert space $H$, there exists a $\sigma $-finite measure space $(X, \mathscr A, \mu )$, and a unitary operator $U:H\to L^2(X)$, such that $$ T = U^{-1}M_\varphi U, \tag{1} $$ for a certain $\varphi \in L^\infty (X)$, where $M_\varphi $ refers to the pointwise multiplication operator $$ \xi \in L^2(X)\mapsto \varphi \xi \in L^2(X). $$ In other words, every normal operator is unitarily equivalent to a multiplication operator.

This result is perhaps not so popular because its uniqueness part is a bit messy (see section (3.5) in Sunder, V. S., Functional analysis: spectral theory, Birkhäuser Advanced Texts. Basel: Birkhäuser. ix, 241 p. (1997). ZBL0919.46002.) but it is pretty useful, e.g. here.

Next let $B(\sigma (T))$ denote the algebra of all bounded Borel functions on $\sigma (T)$ and consider the *-homomorphism$^{\dagger}$ $$ f\in B(\sigma (T)) \mapsto U^{-1}M_{f\circ \varphi }U \in \mathscr B(H). \tag{2} $$

It is easy to prove that this satisfies all of the properties of the Borel functional calculus, and hence this is the Borel calculus!

Given any $f$ in $B(\sigma (T))$, we thus have that $$ f(T) = U^{-1}M_{f\circ \varphi }U. \tag{3} $$

Observe that (3) is precisely the expression of (1) for $T'=f(T)$ and $\varphi '=f\circ \varphi $, so the same reasoning above implies that the Borel functional calculus for $f(T)$ is $$ g\in B(\sigma (f(T))) \mapsto U^{-1}M_{g\circ f\circ \varphi }U \in \mathscr B(H). $$

We then conclude that $$ g(f(T)) = (g\circ f)(T), $$ for all $g$, as desired.


$^{(\dagger)}$ It should be noted that the spectrum of $T$ coincides with the essential range of the above function $\varphi $. Moreover, it is well known that $\varphi (x)$ lies in its essential range for almost all $x$, so the composition $f\circ \varphi $ in (1) is defined a.e. on $X$, and hence the multiplication operator $M_{f\circ \varphi }$ is well defined.

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ Thanks for your inspiring answer! But H here is just a Hilbert space, not supposed to be separable, sorry for didn’t mentioned. Does it still hold true? $\endgroup$ – milky sausage Oct 31 at 15:02
  • $\begingroup$ It is a bit awkward, and often also unnecessary from the point of view of applications, to mix measure theory with nonseparable spaces, but yes, I think the above also holds in general, except that the measure space $X$ will no longer be $\sigma$-finite. It will still be decomposable, so the main Theorems of analysis would still hold. $\endgroup$ – Ruy Oct 31 at 15:14
  • $\begingroup$ thanks for your patience! You give me a new perspective to see the problem! $\endgroup$ – milky sausage Oct 31 at 15:17

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.