Figure 2

Accuracy results. Accuracy has been tested using JC and HKY models. For each run, the normalized deviation is calculated: $\left(\widehat{\Sigma }\left(C;a,b,t\right)-\Sigma \left(C;a,b,t\right)\right)∕\Sigma \left(C;a,b,t\right)$ where Σ is the correct value and $\widehat{\Sigma }$ is the calculated one. Each plotted point represents the average over a, b and 5 different randomly generated matrices C as described in the main text.