First, suppose we have calculate the left eigenvectors and eigenvalues:
Suppose that \(\pi\) is the equilibrium population. Then, we can normalize the eigenvectors such that:
Above, we denote \(\pi^{-1}\) to be a diagonal matrix with elements \(\pi_i^{-1}\).
The autocorrelation function of the observable \(f_i\) can be denoted:
We know that
Thus,
Where
Finally, note that \(\lambda_i(\infty) = \delta_{i0}\), so the long-timescale behavior is simply:
For most applications, one is interested in the zero-centered ACF, so we simply skip the \(k = 0\) term in the summation.