Simulated Tempering Temperature Weights

[Kyle Kihn]

Hi All,
We have been attempting to implement the enhanced sampling method of simulated tempering using openMM. We had a question regarding if we are implementing the weights correctly. We let the script determine the weights for us, but provide a min and max temp as well as a number of temperatures to sample. We this was used on a WT system we see fairly even temperature sampling distribution (Temperature_HistogramPhuS_ST_520K_WT_500ns.png). However when we run a single residue mutant (generated using the CHARMM PDB generator) and run it under the exact same script we get a very uneven sampling of the temperature space (Temperature_Histogram_PhuS_ST_520K_H212R_500ns.png. Is this expected? or are we implementing something incorrectly?
Thank you,
Kyle

[Peter Eastman]

That looks to me like it's getting stuck in a set of bad weights that don't produce an even distribution. You can also see that it's gradually fixing the problem as it adjusts the weights. With time, it starts to access higher and higher temperatures. But it's happening very slowly, which isn't what you want.

The first thing I would try is increasing the number of temperatures a bit, so as to get better overlap between them. Does that give better behavior?