Solved an issue with the passing of ancestral states to the likelihood function. It now solves the overestimation of the sigma2 under MCMC...

Functional version that estimates the ancestral states instead of using the contrast approach. MCMC implementation also available

Initial test by setting the internal nodes as variables in the estimation of BM. Seems to work so far.

There are mistakes in the formula from Novembre and Slatkin 2008. I changed the modification of S and the weighted average. Now it gives perfectly similar values than with the multinormal density

Working now when assuming that the internal nodes are weighted averages from the two descendants. Crazy speed-ups vs fitBM...

adding some function to compute the likelihood of a continuous character under BM using the pruning algorithm