Confusion regarding spectral decomposition of normal operator

41 Views Asked by At

I'm reading out of Operator theoretic aspects of ergodic theory, and specifically looking at pages 380 and 381, and specifically at Remark 18.14. The authors have already proven the existence of a maximal spectral type for normal operators on a separable Hilbert space. This means that given an operator $T$ on a separable Hilbert space $\mathscr{H}$, there exists a positive bounded Borel measure $\mu_\mathrm{max}$ on $K = \sigma(T)$ such that (1) $\mu_y \ll \mu_\mathrm{max}$ for all $y \in \mathscr{H}$, and (2) if $0 \leq \mu \ll \mu_\mathrm{max}$, then there exists $y \in \mathscr{H}$ such that $\mu_y = \mu$. Further, this $\mu_\mathrm{max}$ is unique up to equivalence, earning it the name maximal spectral type.

Now, in Remark 18.14, they mention that this result can be sharpened. Specifically, they say that there exists a (possibly finite) sequence $(x_n)$ in $\mathscr{H}$ such that $\mu_{x_1} \gg \mu_{x_2} \gg \cdots$, and $\mathscr{H} = \bigoplus_n Z(x_n)$, where $Z(x) : = \overline{\operatorname{span}} \left\{ T^k x : k \geq 0 \right\} \subseteq \mathscr{H}$. One can then recover the maximal spectral type as $\mu_{x_1}$.

If I'm understanding it correctly, what I want to do is take $\mu_{x_1}$ to be the maximal spectral type of $T$ on $\mathscr{H}$, then find $\mu_{x_2}$ as the maximal spectral type of $T \vert_{\mathscr{H} \ominus Z(x_1)}$, and continue by finding $x_n$ so that $\mu_{x_n}$ is the MST of $T$ restricted to $\mathscr{H} \ominus \left( Z(x_1) \oplus \cdots \oplus Z(x_{n - 1}) \right)$.

My hang-up here is that it's not clear to me why $\mathscr{H} = \bigoplus_n Z(x_n)$. I understand that $\mathscr{H}$ is separable, so it can't hold an uncountable number of nonzero orthogonal subspaces. But it seems to me possible that perhaps $\bigoplus_n Z(x_n)$ is only a proper subspace of $\mathscr{H}$, and I need to do some kind of transfinite recursion method so that I can choose $x_\omega$ by finding the MST of $T \vert_{\mathscr{H} \ominus \bigoplus_{n = 1}^\infty Z(x_n)}$, and continue this way. How do I know that $\mathscr{H} = \bigoplus_n Z(x_n)$?

EDIT: I've resolved my issue. Looking at another book, their discussion of the MST mentioned that given a fixed vector $z \in \mathscr{H}$, one could specifically choose a vector $x$ such that $\mu_x = \mu_\mathrm{max}$ and $z \in Z(x)$. To answer my question, we can fix an orthonormal basis $\{e_n\}_{n = 1}^\infty$ and just make sure that we always choose $x_n$ so that $e_n \in \bigoplus_{i = 1}^n Z(x_i)$. If we keep this schedule, then we'll have that $\mathscr{H} = \bigoplus_{n = 1}^\infty Z(x_n)$.