Outline:

Markov chain Monte Carlo algorithms are not handling wll multimodal problems where the target density includes several significant modes that are separated by low probability regions. Looking at foldings of the original space that superpose several parts of the density function over a compact set may produce considerable improvement in this regard at a negligible additional cost. Determining such foldings in an automated or semi-automated way and assessing the gains in convergence properties is thus of direct interest for the field of MCMC algorithms.

 

Themes:

Markov chain theory, simulation, Monte Carlo approximations, projective geometry