[gmx-users] Re: 1-4 interactions free energy calculations

[Justin Lemkul]

On 6/25/13 4:36 PM, Sonia Aguilera wrote:
> Hi Justin,
>
> Thank you for your answer. I’m performing several tests to see what is the
> best for my system. The reason I decided not to couple the intramolecular
> interactions is because I think that the annihilation of the molecule will
> lead to very extreme configurations and that will affect my phase-space
> overlap. I know I can just increase the number of intermediate states, but I
> wanted to first test my theory. Also, I have limited resources to run.
>
> There is a gap between the indirect and direct calculation methods  for my
> system. If I set up couple-intramol=no, then I can´t use domain
> decomposition but particle decomposition in all simulations (charge groups
> are too big, and there is no domain decomposition… ) and I got the 1-4
> interactions warning. If I restrain the h-bonds, then the minimization
> converges but not to Fmax less than 100 in only 499 steps and 12 minutes (8
> cores machines). If I continue and run the NVT stabilization, then it takes
> 17h20min to complete only 100 ps, and I still got the 1-4 interactions
> warning. I still have to run the NPT and 3 ns MD, which will take too much
> time. I think it worth it  because I think it will help with the phase-space
> overlap, and I will only have to perform 20 series of simulations (vdw and
> coulomb separately) to get the free energy change.
>
> On the other hand, if I set up couple-intramol=yes, then I have to run 20
> extra simulations in vacuo to counter the annihilation effect. I think that
> it will also affect the phase-space overlap. The good thing is that I can
> use domain decomposition and the minimization converges to machine precision
> (takes 7000 steps and around 40 minutes in a machine with the same
> characteristics that the previous one). Also, the same 100-ps NVT takes only
> 3h20min. If I think of the overall picture, it seems that the total time I
> will spend doing dd (coupleintramol=yes) rather than pd (coupleintramol=no)
> will be less. However I’m worried about the phase-space overlap and the
> available resources I have for running since it makes difficult to just
> increase the intermediate states.
>

You may be able to tweak the DD setup using mdrun command line options.  You 
haven't posted specific messages, which would have been useful.  It sounds like 
you may have a problem with your charge groups, and if that's coming from grompp 
then you have a topology problem.  If it's coming from mdrun, then again it's 
probably just a matter of tweaking some things.  Normally, one does not have to 
adjust the DD setup, but if you have long-range interactions that are being used 
by the free energy code, that may be necessary.

> Also, I think I can try doing the same but with the 4.6.1 version. It looks
> like I can now couple both vdw and coulomb in the same simulation
> (http://www.alchemistry.org/wiki/GROMACS_4.6_example:_Direct_ethanol_solvation_free_energy).
> I have also wanted to try expanded ensemble to improve my sampling, but I
> have not found that much documentation about the implementation on gromacs.
> It would be great if you can provide an example.
>

Michael provided some good commentary here; there's nothing I can add.

> So, do you know another way to improve my sampling and the phase-space
> overlap that can help me to solve my problem?  Also, I’m working with a
> charged molecule. Would it help in any aspect if I neutralize the system?

PME should compensate for the net charge, IIRC.

-Justin

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

Justin A. Lemkul, Ph.D.
Research Scientist
Department of Biochemistry
Virginia Tech
Blacksburg, VA
jalemkul[at]vt.edu | (540) 231-9080
http://www.bevanlab.biochem.vt.edu/Pages/Personal/justin

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