GROMACS version: 2025.3
GROMACS modification: No
In GROMACS tutorials for umbrella sampling simulations:
https://pubs.acs.org/doi/full/10.1021/acs.jpcb.4c04901
the pulling simulation is performed and the resulting trajectories are saved to .xtc files, which contain only positions and not velocities. Then, the selected “windows” act as starting structures for a series of biased-MD simulations. For each of them, velocties are generated again at the beginning of the corresponding NVT/NPT equilibration.
I wonder whether a slightly modified approach would be valid from the statistical thermodynamics point of view. If I saved pulling .trr files, containing velocities, am I allowed to reuse these velocities in a simulation for a corresponding “window”?
Even though you can do it, this is not what I would recommend. I have found that most of the time, to achieve good convergence of umbrella sampling simulations, it is better to perform several short simulations for each window instead of a single long one. For example, 10 simulations of 2ns instead of one of 20ns. And to do this, you need different initial velocities for each umbrella sampling simulations, thus you should not use the velocities from the trr file. To obtain your starting points for each window, performing different independent steered-MD is also a wise choice, to have in addition different structures.
To know how much replicas in each window, you should check that two subsets provide the same results. For example, with 10 replicas at each window, you can compute the free energy profile with the first 5 windows, then the last 5, then the windows with an odd index, and the windows with an even index. The four free energy profiles should be the same (within the error bars). If that’s not the case, you must increase the number of replicas in each window