The AWH friction metric scale
Summary
Currently, the AWH friction metric is too large by a factor of (kb T)^2, so the diffusion coefficient from gmx awh -fric is too small by the same factor, 6.22 at 300 K in kJ/mol.
Existing runs do not need redoing and the correction can be applied in post-processing by multiplying/dividing the diffusion/friction by (kb*T)^2.
Exact steps to reproduce
Run any AWH run with a pull dimension and compare 1/friction against D from the MSD of the same trajectory.
Measured on a Martini3 solute in water, D from the MSD of the same trajectory, D in 1e-4, k = 500:
| T (K) | D_MSD | D_AWH (2026.1) | ratio | (k_BT)^2 |
|---|---|---|---|---|
| 285 | 25.3 | 4.66 | 5.44 | 5.62 |
| 300 | 30.9 | 5.13 | 6.03 | 6.22 |
| 350 | 49.7 | 6.06 | 8.21 | 8.47 |
D_AWH hardly moves with temperature while the MSD nearly doubles. I also ran a 4 force constant x 3 temperature grid, ratio/(kb*T)^2 = 1.02 +/- 0.10.
Possible fixes
Fix could just loop over forceFromNeighbor and scaling by beta in updateForceCorrelationGrid(), after the calcUmbrellaForceAndPotential() call and before addData().