Consider a parallel-tempering Markov-Chain Monte Carlo (MCMC) simulation of a system, with inverse temperatures $\beta_1 < \cdots < \beta_N$ and energies $E_1, \cdots, E_N$.

If there are two temperatures, $N=2$, the temperature-swap procedure is clear: First, let each replica evolve with the MCMC dynamics, and once in a while exchange $\beta_1 \leftrightarrow \beta_2$ with probability $p_{1,2} =\min \{ 1, \exp[-(\beta_1-\beta_2)(E_1-E_2)]\}$.

However, by looking in the literature, I could not find any detailed information on how to swap temperatures in the more general case where there are $N>2$ temperatures. One possible choice would be to do, once in a while in the simulation,

for $n=1,...N-1$

$\,\;$Exchange $\beta_n \leftrightarrow \beta_{n+1}\,$ with probability $p_{n,n+1}$


Is this choice correct? Does it satisfy ergodicity and detailed balance?

I am worried by the fact that this choice does not treat the temperatures $\beta_1$ and $\beta_N$ at the boundaries in the same way as those in the bulk. In fact, it attempts to swap a temperature in the bulk twice as often as those at the boundary.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.