The "inf" values appearing in the python output are the values of the MBAR/UWHAM objective function (the MBAR/UWHAM solutions are obtained via nonlinear optimization of the objective function). My interpretation is that it's taking the natural log 0.0. Therefore, it is unable to solve the MBAR equations (however, it still produces reasonable TI values ddG = 2.0 +- 0.8 kcal/mol). The numerical issue is likely caused by unusual energy values in the "MBAR analysis" section appearing in the mdout files. I do not have these outputs so I cannot track it down futher; however, there are two things I would immediately try:
1. Rerun edgembar-amber2dats.py with the "--exclude" option, and try running edgembar again. This option was originally added because there were a few cases were the MBAR analysis on pmemd.cuda seemed to overflow the numerical representation when distant lambda states produced very different energies.
2. If the above doesn't work, then try running edgembar with the "--mode=BAR" option. If the numerical issue is caused by very large energy differences between distant lambda states, then this should avoid it by limiting the analysis to neighboring lambda states.
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Mon Dec 09 2024 - 08:00:03 PST
