Canonical: https://knowlpedia.com/ergodic-theory/correlation-function/

# Correlation of dynamical observables

Inner products between an evolved observable and a fixed observable.

For a [probability-preserving transformation](https://knowlpedia.com/ergodic-theory/measure-preserving-transformation/) \(T\) and \(f,h\in L^2\), their **correlation** at time \(n\geq0\) is
\[
c_n(f,h)=\langle U_T^nf,h\rangle
=\int_X f(T^nx)\overline{h(x)}\,d\mu(x).
\]
For invertible \(T\), negative times are also defined. The **centered correlation** subtracts \((\int f\,d\mu)\overline{\int h\,d\mu}\); the **autocorrelation** takes \(h=f\).

## Events and representations

For \(f=1_A,h=1_B\), correlation equals \(\mu(T^{-n}A\cap B)\). It is a [matrix coefficient](https://knowlpedia.com/harmonic-analysis/coefficient-function/) of the Koopman representation, with the sign of \(n\) adjusted for the inverse-pullback convention.
