I am trying to understand how to develop the spectral measure of a bounded self-adjoint operator on a Hilbert space. For every continuous function on its spectrum, $f: C(\sigma(A)) \to \mathbb{C}$, and for every $w \in H$, by Riesz-Markov, we have:

$\langle w, f(A)w \rangle = \int_{\sigma(A)} f \mathrm{d}\mu_w$, for a finite regular Borel measure of total mass $||w||^2$.

Evidently, the left side makes sense for measurable $f$ (with respect to the Borel algebra). Since the values of $\langle w,Aw\rangle$ as $w$ spans the whole space uniquely determines the operator $A$, we define $\phi: M(\sigma(A)) \to B(H)$, $g \mapsto g(A)$ satisfying the above condition. Now, I want to check that $\phi$ is a unital *-homomorphism, but I can't seem to prove multiplicativity - that is,

$\langle w, f(A)g(A)w\rangle=\langle w, (fg)(A)w\rangle$

I tried using the polarization identity, writing the LHS as

$$ \langle w, f(A)g(A)w\rangle = \frac{1}{4} \int_{\sigma(A)} f \mathrm{d}\mu_{w+g(A)w} - \frac{1}{4}\int_{\sigma(A)} f \mathrm{d}\mu_{w-g(A)w} + \frac{i}{4}\int_{\sigma(A)} f d\mu_{w+ig(A)w} -\frac{i}{4}\int_{\sigma(A)} f \mathrm{d}\mu_{w-ig(A)w} $$

But then I got completely lost, and I have no clue how to make the $g$ appear in the integrand. Any help would be greatly appreciated.

EDIT: Yes, sorry, I am assuming a self-adjoint operator. I added that above now.


I prefer the Mathematician's inner product which is linear in the first coordinate. For each $w \in H$, there is a unique regular Borel measure $\mu_{w}$ on $\sigma(A)$ such that $$ (f(A)w,w) = \int_{\sigma} f\,d\mu_{w},\;\;\; f \in C(\sigma). $$ Because of this, $\|w\|^{2}=(Iw,w)=\mu_{w}(\sigma)$. Then $$ \begin{align} \int_{\sigma}fd(\mu_{w+w'}+\mu_{w-w'}) & = (f(A)(w+w'),(w+w'))+(f(A)(w-w'),(w-w')) \\ & = 2(f(A)w,w)+2(f(A)w',w')\\ & = 2\int_{\sigma}fd(\mu_{w}+\mu_{w'}). \end{align} $$ By uniqueness, the first identity below follows, and the second follows using a similar argument: $$ \mu_{w+w'}+\mu_{w-w'}=2\mu_{w}+2\mu_{w'}\\ \mu_{\alpha w}=|\alpha|^{2}\mu_{w}. $$ For any bounded Borel function $f$ on $\mathbb{R}$, define $$ q_{f}(w) = \int f d\mu_{w}. $$ Then $q_{f}$ is a bounded quadratic form because $$ q_{f}(w+w')+q_{f}(w-w')= 2q_{f}(w)+2q_{f}(w'),\\ q_{f}(\alpha w)=|\alpha|^{2}q_{f}(w),\\ |q_{f}(w)| \le \|f\|_{\infty}\|w\|^{2}. $$ That's enough to guarantee the existence of a unique operator $T_{f} \in \mathcal{L}(H)$ such that $$ q_{f}(w) = (T_{f}w,w). $$ By uniqueness, $T_{f}=f(A)$ if $f \in C(\sigma)$. So this extends the continuous functional calculus to bounded Borel functions. Of course--as you noted--there is no obvious reason that $T_{fg}=T_{f}T_{g}$ will continue to hold for bounded Borel functions $f$, $g$.

Consider the set of Borel subsets $S$ of $\mathbb{R}$ for which $$ T_{f}T_{\chi_{S}} = T_{\chi_{S}}T_{f} = T_{\chi_{S}f}. $$ Start by showing that this holds for open intervals $S=(a,b)$. Define $g_{\epsilon} \in C(\sigma)$ such that $g_{\epsilon}$ is $0$ outside $[a,b]$, is $1$ on $(a-\epsilon,b+\epsilon)$ and is linear and continuous on the remaining intervals $(a,a+\epsilon)$ and $(b-\epsilon,b)$. As $\epsilon\downarrow 0$, the function $g_{\epsilon}\uparrow \chi_{(a,b)}$. Because of that, $g_{\epsilon}(A)$ converges in the strong operator topology up to an operator you can show to be $T_{\chi_{(a,b)}}$. After a little thought, you see that the above identity holds for $S=(a,b)$. The above also holds for the complement of $S$ if it holds for $S$. After a little work, you show that the above holds for a sigma algebra of sets containing the open intervals, which gives the identity for all $S$. Then another approximation argument will give you $T_{f}T_{g}=T_{g}T_{f}=T_{fg}$ for all bounded Borel functions $f$, $g$.


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.