← The Peter Green
← Every Peter Green
← Peter Green (Statistician)

Reversible-jump MCMC, explained

Peter Green's 1995 contribution to Bayesian statistics

Reversible-jump MCMC, explained (placeholder)

Reversible-jump Markov chain Monte Carlo (RJMCMC), introduced by statistician Peter Green in 1995, lets a Bayesian sampler jump between models that have different numbers of parameters — making it a cornerstone method for model selection.

Most Markov chain Monte Carlo methods explore a single, fixed-size model. Peter Green's reversible-jump MCMC — set out in a 1995 paper in Biometrika — removed that limit, letting the sampler move between models of different dimension (say, deciding how many components a mixture should have) while keeping the mathematics valid.

That “trans-dimensional” move turned out to be enormously useful, and RJMCMC is now a standard tool across statistics, genetics and signal processing. Green, a Fellow of the Royal Society, is emeritus professor at the University of Bristol.

Sources: Peter Green (statistician), Wikipedia  ·  Biometrika (1995)