19 September 2006
Andrew Fernandes, a fellow Canadian expat and PhD student at NC State, responded to my earlier request for advice on exploring a Dirichlet-type simplex.
Among other places, the idea is presented in the Wikipedia entry for Simplex. He suggests perturbing the cumulative sums, then putting the perturbed sums back in order to draw a time-reversible proposal. This has the advantage of not sending too many parameters below zero - a maximum of one - as opposed to an equal perturbation of each parameter, and not pinning a high-valued parameter in place with a standard Dirichlet proposal.