CHARMM Development Project Forums User Discussion & Questions Performance and Benchmarking heating using CHARMM/OpenMM implementation

Dear all,
I am having trouble to set up heating using charmm/openmm implementation and need some help. For a fixed temperature, I can run NVE/NVT/NPT with charmm/openmm implementation without any problem. However, when temperature changes (for heating process), the temperature control seems not working and it dies in the first several steps (with error Energy is NaN). Below you can find my script. I would be very thankful if anyone can instruct me what I did wrong or share with me a working script for heating using charmm/openmm.

Thank you very much.
James

You may need to use the standard MD code for the heating and early equilibration, before changing to the OpenMM code.

I always heat without using the GPU.
For production runs using OpenMM
ECHECK -1.0 turns off the energy check.
You should probably use the langevin thermostat with OpenMM, and the mixed precision floating point option. I/O is fairly expensive so don't print or output anything more often than you write frames to the trajectory file; set ilbfrq to 0!
It is quite pointless to save the restart file during the trajectory. You should not run longer pieces than you can afford to loose.

This is what I use for production runs (with OPENMM_PRECISION=mixed):

dynamics restart timestep 0.002 nstep 125000000 -
firstt 100 finalt @T - ! firstt is not actually used here
ewald pme kappa 0.34 order 6 fftx @fx ffty @fy fftz @fz -
vswitch cutnb 12.0 cutim 12.0 ctofnb 9.0 ctonnb 8.0
echeck -1.0 imgfrq -1 inbfrq -1 nprint 50000 iprfrq 50000 - ntrfrq 250000 -
iunread 101 iunwrite 201 iuncrd 202 nsavc 50000 ilbfrq 0 -
omm langevin gamma 5

Another performance issue are the NBFIXES - they are expensive, so if you know what you are doing and don't really need them you can remove them from the parameter file; this in particular applies to water-ion nbfixes at low ion concentrations.

thank you all very much. very helpful.