Accumulating error when running long simulations through checkpoint files

GROMACS version: GROMACS 2021.4
GROMACS modification: No

I am wanting to run a simulation in 4 hour increments (on a cluster with 4 hour time limit). When creating checkpoint files, are coordinates and other values rounded? And if so, will value error accumulate when running a simulation that takes several days in 4 hour increments? Will this make the final trajectory incomparable to a simulation run in full length without checkpoint files?


Checkpoint restarts are exact and use full precision for all values stored,so there are no rounding errors. But even there would be rounding errors, these would not accumulate. You would only me adding an extremely minor thermalization very infrequently.