[gmx-developers] doing energy evaluations at alternate lambda values (i.e. for bennett acceptance ratio)

[hessb at mpip-mainz.mpg.de]

> Berk,
>
>>  From your mail it seems as if your bonded interactions do not influence
>> dvdl,
>> then you can also not calculate bonded interactions, this also makes
>> sure the bonded
>> forces are not double counted.
>
> I am not sure what you mean by the bonded interactions not influencing
> dvdl? And actually, what I am interested in here is the total
> potential energy, not dvdl: I want the correct total potential energy
> for the timestep I am at, at the lambda value of interest. I don't
> want to have to assume I am not changing any bonds as a function of
> lambda: This isn't something I want to use just for one particular
> application; rather, a bunch of folks will be doing free energy
> calculations with GROMACS, and will switch to using this approach for
> calculating the free energy differences, so it needs to be fully
> general.
>
>> The only thing I am not sure of is if the virial is not overwritten.
>>
>> For efficiency you also want to call only the free-energy energy loops
>> in nonbonded.c
>> for the other lambda values.
>
>
> What do you think of the approach Erik discussed in his mail?

Sorry, I did not read your mail carefully enough.
I realized later you want the potential, not dvdl.

If Erik hsa it working, you should use of course that.
Still most of my comments apply, but Erik probably has
already thought of most of them.
The most important thing, in terms of performance,
is to store the energy of the free energy loops only
and only recompute those bonded interactions for the other
lambda points.

Berk.