Unexpected results for MSD, for various -dt
Dear Devs,
I am trying to calculate diffusion coefficients from mean squared displacements (MSD) from water. However, I am getting some puzzling results for the MSD, if I change the -dt argument of gmx msd. If I change -dt, I find kinks in the MSD plot, which I would not expect to be there (see figure below)
I am aware of the known bug, that -dt may not be chosen smaller than 1 ps (#4694 ), but I wonder if there are other issues with the choice of -dt. Might that be possible?
I am not sure, if this is really a bug at all, or if this has something to do with my system setup. (See below)
Disclaimer: Currently, I am doing this analysis with GMX-2023.3. I will compile the newest version tomorrow and verify if the results are still the same. UPDATE Tested it with GMX-2024.1 and I get the same result.
Of course, I will provide input and results of these simulations, to faciliate eventual debugging.
Best regards, Patrick
Additional info: I perform vanilla MD simulations: I simulate pure water (no ions) with the md integration scheme. I perform simulations at T=300 K with the velocity rescale thermostat. I use a cubic box with PBC. I use pme for long-range charged interactions and a cut-off for long range LJ interactions.
So far I investigated the following parameters:
- NVT vs. NPT simulation (equivalent results), for eventual pressure-coupling I am using the cell-rescale barostat.
- simulation with different framerates (I tried
nstxtcout=1000andnstxtcout=250with a timestep of 2 fs) - instead of simulating 216 water molecules, I simulated a larger system, N=432, as well (I plan to look into even larger systems, since the different in box length from N=216 to N=432 is not big.
The following figure was obtained with different values for -dt, analyzing a trajectory with frames written every 2 ps.
