Canonical: https://knowlpedia.com/ergodic-theory/koopman-spectral-measure/

# Spectral measure of an observable

The positive measure whose Fourier coefficients are the autocorrelations of an observable.

Let \(T\) be invertible and probability preserving, and let \(E\) be the spectral resolution of the forward [Koopman operator](https://knowlpedia.com/ergodic-theory/koopman-operator/) \(U_T\). The **spectral measure of \(f\in L^2\)** is the finite positive Borel measure
\[
\sigma_f(B)=\langle E(B)f,f\rangle=\|E(B)f\|_2^2
\qquad(B\subseteq\mathbb S^1\text{ Borel}).
\]
Its total mass is \(\|f\|_2^2\), and its Fourier coefficients are
\[
\int_{\mathbb S^1}z^n\,d\sigma_f(z)=\langle U_T^nf,f\rangle.
\]

## Atoms and fixed vectors

An atom at \(\lambda\) records a nonzero component of \(f\) in the \(\lambda\)-eigenspace. In particular, mass at \(1\) is the squared norm of its invariant component. A unit-norm eigenfunction has spectral measure \(\delta_\lambda\).

## Mixing

Weak mixing makes \(\sigma_f\) atomless for every mean-zero \(f\). Strong mixing requires its nonzero-time Fourier coefficients to tend to zero, which is a stronger condition than atomlessness.
