[gmx-users] hydration free energy

[Wes Barnett]

On Fri, Jan 5, 2018 at 2:29 PM, <asaffarhi at post.tau.ac.il> wrote:

> Thanks for discussion. I am waiting already few days for anyfing reply.
> The technique of calculation in the file job.sh. The value of coul-λ is
> uniformely increasing from 0 to 1 with the step 0.02 for 50 first λ-steps
> and in the same way for vdw-λ last 50 λ-steps. The problem is in the last
> trajectory with λ=1 there are no coulomb and vdw intermolecular
> interactions with the estimate of the entropy according BAR method
> S=62204236256674537472.00 kT. The neigbouring state used in BAR with λ=0.99
> contains no coulomb and some remaining vdw force. Therefore I consider that
> overlapping of atoms can't produce this error. It happens only then
> opposite coulomb charges overcome vdw in nearly decoupled state. Therefore
> there is no need in soft-core interactions because all coulomb is decupling
> previously to vdw.
> I made the search in archive and this problem was discussed sometimes, but
> didn't found clear answer for my case. Now I decided to run the cycle in
> the reversed order from decoupled state to real interactions.
>
>
> ---------------------------
>
> You're welcome.
>
> If there are no vdW and Coul it still calculates the average energy, which
> without soft-core technique can be very large when atoms overlap.
> You can read
> http://www.sciencedirect.com/science/article/pii/S0010465516303411
>
> Eq. (2) and the following text etc.
>
> You can try a soft-core technique to see if it is better.
>
> Best,
> Asaf
> Quoting dgfd dgdfg <roinato at mail.ru>:
>
> Hi,
>>
>>>
>>> Do you know which values of lambdas are used?
>>> Are you using a soft-core technique?
>>>
>>> Asaf
>>>
>> Thanks for discussion. I am waiting already few days for anyfing reply.
>> The technique of calculation in the file job.sh. The value of coul- λ is
>> uniformely increasing from 0 to 1 with the step 0.02 for 50 first  λ- steps
>> and in the same way for vdw- λ last 50  λ- steps. The problem is in the
>> last trajectory with  λ=1 there are no coulomb and vdw intermolecular
>> interactions with the estimate of the entropy according BAR method
>> S=62204236256674537472.00 kT. The neigbouring state used in BAR with
>> λ=0.99 contains no coulomb and some remaining vdw force. Therefore I
>> consider that overlapping of atoms can't produce this error. It happens
>> only then opposite coulomb charges overcome vdw in nearly decoupled state.
>> Therefore there is no need in soft-core interactions because all coulomb is
>> decupling previously to vdw.
>> I made the search in archive and this problem was discussed sometimes,
>> but didn't found clear answer for my case. Now I decided to run the cycle
>> in the reversed order from decoupled state to real interactions.
>>
>>
Have you tried other FE estimate algorithms (i.e. thermodynamic
integration, MBAR, etc.). Look into the project "alchemical-analysis". It's
a python script that will give you some good visualizations on what is
happening at that last step (it's compatible with GROMACS output files).
There's an associated paper with it as well.

Also double check that you are somehow not mistakingly leaving on some kind
of interaction that is blowing up after everything else is decoupled.

James "Wes" Barnett
Postdoctoral Research Scientist
Department of Chemical Engineering
Kumar Research Group <http://www.columbia.edu/cu/kumargroup/>
Columbia University
w.barnett at columbia.edu
http://wbarnett.us