Figure 1
A dynamic programming algorithm for branch length factorization. Values f u (τ u ) are calculated for all possible choices of τ u . τ u is u's divergence time counted from the leaves and it takes values corresponding to the equidistant grid. f u (τ u ) is a product of rate priors for the edges leading to u, u's contribution to the time prior, and the corresponding probabilities for u's children v and w: f v (τ u ) and f w (τ u ) respectively.