From: Michael Joseph Robertson (mjrober_at_stanford.edu)
Date: Fri Feb 26 2021 - 04:22:32 CST
Hello,
I am currently trying to test NAMD 3.0 alpha 8 GPU FEPs on a membrane protein-ligand system I have run with NAMD 2.13 successfully and obtained good results. The unbound leg ran fine, however for the bound leg when I perform my separate minimization and heating jobs I get an error as below very rapidly at the start of heating in the 0K phase:
ERROR: velocity is -700074 0 0 (limit is 6000, atom 6187 of 49671 pe 1)
ERROR: velocity is 58883.4 -24764.6 -4433.74 (limit is 6000, atom 6393 of 49671 pe 1)
ERROR: Atoms moving too fast at timestep 17; simulation has become unstable (2 atoms on pe 1).
FATAL ERROR: SequencerCUDA: Atoms moving too fast
Although the atom(s) with the excess velocity is inconsistent from run to run. Curiously, if I set the tilmestep to 1 fs, I am able to heat the system, however when I go back to a 2 fs tilmestep after heating I get the error:
ERROR: Atoms moving too fast at timestep 33; simulation has become unstable (0 atoms on pe 1).
FATAL ERROR: SequencerCUDA: Atoms moving too fast
With no atom listed. Also despite being a ‘fatal error’ the NAMD instance doesn’t exit but instead hangs.
Other than CUDASOAintegrate on, are there any other necessary options for NAMD3.0 FEPs? Is the old fep.tcl file still applicable?
Any suggestions would be greatly appreciated!
Thanks,
-Michael
This archive was generated by hypermail 2.1.6 : Fri Dec 31 2021 - 23:17:10 CST