[gmx-developers] jarzynski/slow growth

[Berk Hess]

David Mobley wrote:

> All,
>
> So I am getting ready to run some calculations where I restrain a
> protein to a reference structure, and then gradually release the
> restraints in a time-dependent way using the free energy code and
> GROMACS' facility to have different restraints for the A and B states.
>
> I need to be able to get out the work done in releasing the
> restraints, at every step (for Jarzynski, in case anyone is curious).
> Can anyone think of an easy way to do this?
>
> Also, the catch is that I need to be able to do it without writing the
> potential energy to an .ene file every step, because of bandwidth
> restrictions on the size of the .ene files (need them to be smaller
> than a few MB). Here are a couple ideas I had:
>
> (1) Modify the code to take the difference in potential energy from
> each step to the next, which will be W; keep a running total of the
> integrated W, and write this out to the .ene file every time the .ene
> file is dumped, as with the rest of the energies.
> (2) If dH/dlambda is correct for time-dependent position restraints,
> then dH/dlambda is probably the work value I'm interested in. Do the
> current dH/dlambda averages printed out in the log file include *all*
> of the steps? If not, is the integrated dH/dlambda tracked anywhere in
> a way that includes all of the steps?
>
> If (1) turns out to be the best option, would the developers be
> interested in including this in CVS, etc., after I get it working, so
> that it will persist in future versions?

For any change between topology A and B Gromacs writes
the work. It always writes the file dgdl.xvg which contains
dG/dlambda for every step. And in the energy file there
are the exact averages and fluctuations and the instant
values at the energy file writing times.
You will not need the energy file though, since you have
all the values in dgdl.xvg.

Berk.