- Predicting variance of variance by direct calculation – still need to crunch some math for the expected convergence.
- Still, the approach should be able to do more than describe single points as unexpected deviates.
- Still need to address gradual change vs change point analysis.
- Essentially the same as the phyolgenentic problem – one rate vs two rates. Model selection approaches?
- So far theory is essentially built on a model selection between linear models.
- Calculate the eigenvalue directly rather than ratio of eigenvalue to noise:
- Estimate the eigenvalue from the correlation function and from power spectrum, rather than the lag-1 autocorrelation, or variance.
- Proper signal processing techniques for detecting bifurcations?
Coding Progress
- Added a proper autocorrelation function calculation, log transform and linear regression gives the eigenvalue and the variance.
- Tested using the Langevin model

Whose correlation function is given by
