Difference between revisions of "Heated chains"

Revision as of 20:04, 30 December 2005

In Metropolis-coupled Markov chain Monte Carlo, heated chains have the following characteristics:

• The most sophisticated way to avoid getting stuck in local optima involves heated chains.
• There is a main, true chain, and several heated chains, all running in parallel.
• Every so often, the true chain picks as its proposed tree the location of a heated chain.
• If the new tree is accepted, the heated chain and main chain swap and continue.
• The heated chains experience a more “flattened” landscape and so can move through deep valleys more easily.

Note: Also, multiple heated chains.