[gmx-users] some thoughts on gmx wham

[Justin Lemkul]

On 4/30/18 6:04 PM, Alex wrote:
> This is probably for Justin once again...
> My understanding is that you plan to work on a new tutorial for PMF
> calculations, so may be this could be useful. I recall asking you about
> selection of k and the corresponding spacing between configurations. We
> agreed that there is no strict way of setting this up to achieve decent
> overlapping between histograms. And I've been having a pretty hard time
> getting proper convergence while guessing e.g. spacing.
>
> For "heavy" objects this problem remains, but for e.g. ions it seems there
> is a way to get a semi-quantitative handle, which helped me quite a bit.
> For a given reaction coordinate, the position "spread" is of order x =
> sqrt(k_boltz*T/k), where k is the bias spring constant, i.e. this is the
> order of position's sigma. What worked for me is spacing between configs
> equal to x. For example, with k = 1000 kJ/mol/nm2, T = 300 K, we get x =
> 0.5 A, and with spacing close to x, we get good convergence. Prior to that,
> I used spacing of 1 A and would sometimes get "gaps" in the PMF curve after
> wham's attempts to converge for something like half a million iterations.

Thanks, I'll keep this in mind. The updates to the tutorials won't 
necessarily involve anything "new" but rather necessary revisions and a 
corresponding paper that delves more deeply into the theory and 
algorithms behind everything, not just a "do this, get this" kind of thing.

-Justin

-- 
==================================================

Justin A. Lemkul, Ph.D.
Assistant Professor
Virginia Tech Department of Biochemistry

303 Engel Hall
340 West Campus Dr.
Blacksburg, VA 24061

jalemkul at vt.edu | (540) 231-3129
http://www.thelemkullab.com

==================================================