GROMACS version: 2024 or higher
GROMACS modification: No
Hi,
I am a little bit confused by the recommandations provided here: Getting good performance from mdrun - GROMACS 2026.0 documentation . If I set mass-repartition-factor=3, I can increase dt to 4fs with constraints=h-bonds. If, simultaneously, I set mts=yes with default parameters, it means that the electrostatic forces will be computed every 8fs. Is it what is recommanded? Do you have the reference for an article showing that this is stable?
Also, can you please clarify When running on CPUs only or with GPUs but integration on the CPU? Which mdrun flag should be used (or not used) to have integration on the CPU?
Hi! I’m not familiar with combining HMR and MTS, and I’m also not aware of any systematic study of the effects of the two in conjunction. This paper has a nice discussion of HMR by itself. Note that the integration error of the integrator depends (quadratically) on the time step, so that by itself is a limiting factor. Given that, I don’t think that the GROMACS manual necessarily recommends using the two algorithms together like that (or even in isolation), but I think it should be possible. You would always need to assess stability of the simulation and the energy drift for your specific system.
To perform integration on the CPU, you should pass the -update cpu flag to mdrun. To do everything on the CPU, it should be enough to specify -update cpu -nb cpu, but check your log file.