nan (Not a number) error in meta-eABF simulation

From: Dhiman Ray (dray1_at_uci.edu)
Date: Sat Aug 31 2019 - 22:00:39 CDT

Dear NAMD users,

I am recently facing an issue with the meta-eABF calculations in the GPU
version of NAMD 2.12. For long simulations (>20 ns), the NAMD starts
printing -nan (Not a number) in its output (.log, .colvars.traj, .xsc etc.
files). *The simulation does not stop or it does not print any error
message*. I have tried with changing the parameters of eABF and
metadynamics, (and also timestep of dynamics) but that did not make any
difference. Also NAMD starts printing -nan at different times for different
runs. I have carefully checked that there is no rapid increase in energy or
force prior to the point when nan is being printed.

I have performed much longer metadynamics and eABF simulations with the
same system without any issues. Surprisingly two of my meta-eABF
simulations went on till 200 ns and gave converged results while the rest
printed nan. This has become completely unpredictable. Also I cannot
restart the trajectories because the simulation does not stop and keeps on
replacing the .restart.xsc, .restart.vel, .restart.coor files with nan. So
I do not have any good restart file.

I am following this paper for meta-eABF implementation in NAMD:
https://pubs.acs.org/doi/10.1021/acs.jpclett.8b01994

Please suggest me how to overcome this problem.

Thank you,

Dhiman.
************************************
Dhiman Ray
Graduate Student
Prof. Ioan Andricioaei group
University of California, Irvine
Department of Chemistry
Irvine, CA

This archive was generated by hypermail 2.1.6 : Sat Dec 07 2019 - 23:20:49 CST