simulated tempering (re)weighting

[Daniel Konstantinovsky]

Hi OpenMM!

I am wondering what is the right way to reweight observables calculated from simulated tempering trajectories with the default weighting algorithm? Is reweighting even necessary? I am new to this technique so I apologize if the question is naive.

Thank you!
Dan

[Peter Eastman]

On every output, it reports the current temperature. Every conformation visited by the simulation reflects a Boltzmann distribution at the current temperature, so the probability is proportional to exp(-E/kT) where E is the total energy (potential+kinetic). The most common use of simulated tempering is that you want to collect a distribution at one temperature (say 300K) but use higher temperatures to let it cross barriers more easily. In that case you either could ignore all the samples from higher temperatures, or you could reweight them by the ratio of temperature factors, exp(-E/kT1)/exp(-E/kT2).

[Daniel Konstantinovsky]

So the E in the numerator and denominator is the same (it's the current total energy at whatever temperature?), just the temperatures are different?

[Peter Eastman]

Exactly. Different samples are drawn from different temperature ensembles. At higher temperatures, most samples will have energies that would be very improbable at the lower temperature, so the weighting factor will be very small.

[Daniel Konstantinovsky]

Got it, thank you Peter!