Which precision does umbrella sampling to calculate PMF require? Double or single?

GROMACS version: 2019.4
GROMACS modification: No

Dear all,
We are getting too much error for the PMF (Potential of Mean Force) calculations. We think that the precision might be the factor. We currently have single/mixed precision in order to have faster simulations.

Is it necessary to have double precision trajectories for umbrella sampling simulations, or the single precision is sufficiently enough?

Please let us know if you have any idea.
Much appreciated,
Anh Mai

Single precision is suitable for almost all MD simulations. Large errors in PMF calculations are almost always due to either lack of sampling/insufficient time, suboptimal window spacing, or poorly chosen reaction coordinate.

Thank you.