Convergence for awh

GROMACS version:2020
GROMACS modification: No


I am running a multiple walker awh simulation on a 1-D reaction coordinate. Upon comparing the PMFs at different times the PMF is converged. I also checked that the coordinate distribution ≈ reference target distribution and the reference target distribution ≈ target distribution. The target distribution was chosen to be uniform. Based on these observations I considered the PMF to be converged. However, the simulation never exited the initial stage. There were several coverings but never exited the initial stage. Can we still have a reliable PMF if the convergence criteria is satisfied but never exited the initial stage of AWH?

The PMF is usually not reliable in the initial stage.
So either you simulation is too short and/or your reaction coordinate is problematic.

I am now wondering if it could be that you chose some AWH parameter sub-optimal such that your sampling is actually converged, but it is not detected. Could you post you AWH and pull parameters?

Yes,sure. Here are the AWH and pull parameters I used:

;AWH calculations
pull = yes
pull_ncoords = 1 ; only one reaction coordinate
pull_ngroups = 2 ; two groups defining one reaction coordinate
pull_group1_name = LIG
pull_group2_name = r1180
pull_coord1_geometry = direction ; simple distance increase

pull_coord1_dim = N N Y ; pull along z
pull_coord1_groups = 1 2 ; groups 1 and 2 define the reaction coordinate
pull_coord1_start = yes ; define initial COM distance > 0
pull_coord1_rate = 0 ; 0.0005 nm per ps = 0.5 nm per ns
pull_coord1_k = 0 ; kJ mol^-1 nm^-2
pull_coord1_vec = 0 0 1

pull-coord1-type = external-potential
pull-coord1-potential-provider = awh

awh = yes
awh-potential = convolved
awh-share-multisim = yes
awh1-share-group = 1
awh-nbias = 1
awh-nstout = 50000
awh1-ndim = 1
awh1-dim1-start = 0.4
awh1-dim1-end = 4.0
awh1-dim1-diffusion = 5e-5
awh1-dim1-force-constant = 12800
awh1-equilibrate-histogram = yes
awh1-error-init = 20