[gmx-users] different PMF results using MPI

[Srinivasa Ramisetti]

Dear GMX-USERS,


I have a doubt trusting the PMF curve of a single molecule pulled from an immobile calcite surface using pull code with umbrella sampling in GROMACS. I extracted two PMF’s for the same system (all inputs are the same) by running simulations on 36 and 72 cores with MPI to check the consistency of the results. I find a significant change in the magnitude of the two curves (please see attached the image). The figure also shows the error of PMF obtained using bootstrap method. The histogram for both the cases are well overlapped.


Is it okay to rely on this results? Can anyone suggest how I can reproduce similar PMF results by running simulations on any number of cores?


These results were produced using gromacs 2016.4 version with the following pull code:


pull = yes

pull_ngroups = 2

pull_ncoords = 1

pull_group1_name = SUR ; calcite surface

pull_group2_name = MDX ; organic compound

pull_coord1_type = umbrella ; harmonic biasing force

pull_coord1_geometry = distance ; simple distance increase

pull_coord1_groups = 1 2

pull_coord1_dim = N N Y

pull_coord1_k = 500 ; kJ mol^-1 nm^-2

pull_coord1_rate = 0.0

pull_coord1_init = XXX ; XXX is the COM distance between 2 groups

pull_coord1_start = no


Thank you,

Srinivasa