Fig. 4.

Importance sampling. The data likelihood $\mathrm{Pr}(\mathbf{A}=A|\mathbf{D}=D,\mathbf{F}={\pi }_i°\overline{F},\omega )$ is the same for all permutations π of $\overline{F}$ for this tree T. However, ${\pi }_1°\overline{F}$ satisfies the Sum Condition, and ${\pi }_2°\overline{F}$ does not. Thus ${\pi }_2°\overline{F}$ has a probability $\mathrm{Pr}(\mathbf{F}={\pi }_2°\overline{F}|\mathbf{T}=T,\mu )=0$