I am currently trying to run a AWH multiwalker simulation, using 4 walkers for a single bias, to transition an amino acid side chain from one side to the other. I’ve run the simulation now for approx. 300 ns per walker, and the PMF is now barely changing anymore. I’ve checked that the initial stage finished and that there have been transitions happening. However, the ref. value distribution is still not very close to the target one (uniform). Am I correct that this means I should continue to simulate? Or are there some AWH-parameters that I should adjust, like the force constant? Up to now I have been using defaults/tutorial ones:
awh1-dim1-diffusion = 5e-05 ; Default diffusion value
awh1-dim1-force-constant = 5000 ; Force constant should be > 0
awh-nstout = 25000 ; Step interval for writing awh*.xvg files.
awh-share-multisim = yes ; Share bias across simulations
awh1-share-group = 1 ; Non-zero share group index
awh1-equilibrate-histogram = yes ; Ensure histogram is close to target
It can take a very long time before you have sampled very close to the target distribution, especially if your reaction coordinate is large. How large is the difference in general? If it’s less than a factor 2 or 3 and the PMF is stable, I would consider it converged. But in some cases and even larger difference can still produce reliable results, especially if you are still sampling the whole reaction coordinate. I think the settings look reasonable.
Thanks for the answer :) Indeed, the ref.value distribution is in the range of 2 or 3. I would consider my complete reaction coordinate to be rather small; it’s a 1D-RC covering approx. 1.6 nm (though since I am quite a beginner, I have no real reference or experience what could be considered a ‘small’ or ‘large’ reaction coordinate).
So, I guess the next step would be to run replicas and see if I can reproduce the PMF. My initial coordinates, which I used to start the first multi-walker simulation, did not cover the whole RC. Is it a good idea to use snapshots from my first awh-run, which are more evenly distributed along the reaction coordinate, to start my replicas? In theory, this should shorten the equilibration of the histogram, right?