GROMACS version:2018.3
GROMACS modification: No
Hello
I have plotted the energy for one of my systems (solution of 9 protein chains, 168 aminoacids each, waters, 150mM NaCl in Martini3 FF, simulated with Gromacs 2018.3). The conserved energy, which supposed to be total energy exchange between the barostat and the system, linear drift due to num.rounding errors and approximations has huge drifts which apparently depend on the length of simulated interval (first 5 intervals were 3.5ns and last one 8ns). Can anyone advise why I’m generating these drifts and what might be wrong with my protocol (link to files attached)? Thanks.
You files are not accessible.
But some linear drift is normal and usually not problematic. There are many sources of drift. One is the pair-list buffering. This drift is controlled by the verlet-buffer-tolerance parameter. The default value is 0.005 kJ/mol/ps per atom. So for a 200 000 atom system you get max 1000 kJ/mol drift in a ps, or a million kJ/mol in a ns. This might seem like a lot, but per atom per ps this is negligble and does not cause artifacts.