MCMC guarantees that the sampled genetic states X have positive
probability under 
, but we still need to find an
admissible state to start from. This may be difficult if the space of
admissible X is severely restricted by the data Y. Ideally, we
would also like to start near a major mode of the distribution, to
avoid that the sampler spends a long time in low probability regions
at the beginning of the simulation. The most recent approaches to
finding initial configurations use reverse peeling on parts of the
pedigree where it is simple to do, and simulation from approximate
genotype distributions elsewhere (Heath [3]).