Figure 1.

Nested sampling. Nested sampling: On each iteration, the worst sample (likelihood = L*) in the set of n active points is replaced by a new sample chosen uniformly at random within the likelihood contour defined by L*. For n=2, the enclosed prior mass X shrinks by a factor of 2 on average as illustrated. The evidence integral is computed from the resulting estimates of ΔX_i and the likelihood of each eliminated sample. Note that the likelihood contours in parameter space (left) are shown schematically and are not assumed to be uni-modal or aligned to the parameter axes.