Let $A$ be a self-adjoint (unbounded) linear operator in a Hilbert space $H$. Let $E$ be the unique regular resolution of the identity on the Borel subsets of $\mathbb{R}$ such that $$A= \int_\mathbb{R} t dE(t).$$ We then write $$f(A)= \int_\mathbb{R} f dE$$ for a measurable function $f: \mathbb{R}\to \mathbb{C}$ and we have $$f(A)+ g(A)\subseteq (f+g)(A), \quad f(A)g(A)\subseteq (fg)(A).$$ Note that the right hand sides $(f+g)(A)$ and $(fg)(A)$ are closed operators, so this begs the questions:
Is $(f+g)(A)$ the closure of $f(A)+g(A)$? Similarly, is $(fg)(A)$ the closure of $f(A)g(A)$? How to prove this?
I learnt this stuff from chapter 13 in Rudin's functional analysis book, so an answer that uses tools from this chapter is highly appreciated! Of course, any input is welcome.
Yes, this is true. In each case, the argument works by localizing the functions to sets where they are bounded. I will only prove it for sums here.
Let $M_n=\{\lambda\in\mathbb R: \lvert f(\lambda)\rvert+\lvert g(\lambda)\rvert\leq n\}$. Clearly, $E(M_n)H\subset D(f(A))\cap D(f(B))$. For $\xi\in D((f+g)(A))$ let $\xi_n=E(M_n)\xi$. As $M_n\nearrow \mathbb R$, we have $\xi_n\to \xi$ in $H$. Moreover, $$ \lVert (f+g)(A)(\xi-\xi_n)\rVert^2=\lVert (f+g)(A)E(M_n^c)\xi\rVert^2=\int_{\mathbb R}1_{M_n^c}\lvert f(\lambda)+g(\lambda)\rvert^2\,d\langle\xi,E(\lambda)\xi\rangle. $$ This last integral converges to zero by the dominated convergence theorem. Thus $\xi_n\to \xi$ with respect to the graph norm.