arXiv:2611.15784 · v1

Small Maximum Likelihood on Fixed Cavender-Farris-Neyman Trees: An Approximation Scheme and Limits to Fully Polynomial Approximation

首次提交 2026-11-12 01:59

版本更新 2026-11-12 01:59

雷达观测 2026-11-13 14:16

摘要

Small Maximum Likelihood fits the edge substitution probabilities of a given phylogenetic tree to an observed alignment, summing over the unobserved ancestral states. For the binary Cavender--Farris--Neyman model, we assume a uniform root and an edge-specific substitution probability $p_e\in[0,\tfrac12]$ for each edge $e$, shared across all alignment sites. We give a deterministic polynomial-time approximation scheme. The algorithm achieves a factor of $1+\varepsilon$ for both the negative log likelihood and the loss after subtracting the uniform-root contribution. Its main step fixes ancestral sequences at one periodically spaced class of depths while summing over all remaining ancestors. A posterior entropy bound controls the cost of this restriction, and dynamic programming optimizes the resulting components over a finite parameter grid. We also give a randomized reduction from Clique showing that an FPTAS returning feasible rational edge probabilities would imply $RP=NP$. These results separate fixed-accuracy approximation from fully polynomial approximation for the same fixed-tree model.