[gmx-users] Domain decomposition

[Alexander Alexander]

On Tue, Jul 26, 2016 at 7:54 PM, Justin Lemkul <jalemkul at vt.edu> wrote:

>
>
> On 7/26/16 1:16 PM, Alexander Alexander wrote:
>
>> On Tue, Jul 26, 2016 at 6:07 PM, Justin Lemkul <jalemkul at vt.edu> wrote:
>>
>>
>>>
>>> On 7/26/16 11:27 AM, Alexander Alexander wrote:
>>>
>>> Thanks.
>>>>
>>>> Yes indeed it is a free energy calculation in which no problem showed up
>>>> in
>>>> the first 6 windows where the harmonic restrains were applying on my
>>>> amino
>>>> acid but the DD problem came up immediately in the first windows of the
>>>> removing charge. Below please find the mdp file.
>>>> And If I use -ntmpi = 1 then it takes ages to finish. Although my
>>>> gromcas
>>>> need to be compiled again with thread-MPI .
>>>>
>>>>
>>>> I suspect you have inconsistent usage of couple-intramol.  Your
>>> long-distance LJC pairs should be a result of "couple-intramol = no" in
>>> which you get explicit intramolecular exclusions and pair interactions
>>> that
>>> occur at longer distance than normal 1-4 interactions.  If you ran other
>>> systems without getting any problem, you probably had "couple-intramol =
>>> yes" in which all nonbonded interactions are treated the same way and the
>>> bonded topology is the same.
>>>
>>>
>> Actually I always have had "couple-intramol = no" in all my other
>> calculation(a single amino acid in water solution), and not problem has
>> shown up. But FEP calculations of the charged amino acid where I have also
>> an Ion for neutralization of the system and "ion+amino acid" is used as
>> "couple-moltype", this problem emerges. And if you noticed the Ion here CL
>> is always one of the atom involving in the problem. I hope
>> "couple-intramol
>> = yes"can sove the problem in charged amino acid.
>>
>>
> Well, there are implications for the results.  Consider what it says in
> the manual.  But yes, this is your problem.  You've got physically separate
> molecules that you call one [moleculetype] for the purpose of
> transformation, and you're running into a problem that isn't really
> physically meaningful in any way.
>

Actually yes, Ion and amino acid both as  [moleculetype]are really far away
from each other. But I usually use the final .gro file of the last step as
input file in the new step, and this separation is what that gro file has.
I hope " couple-intramol = yes"  can help.

>
>
>>> Another question is that if really this amount of pull restrain is
>>>
>>>> necessary to be applied on my molecules (singke amino acid) before
>>>> removing
>>>> the charge and vdW?
>>>>
>>>>
>>>> You're decoupling a single amino acid?  What purpose do the pull
>>> restraints even serve?  CA-HA, etc. should be bonded in a single amino
>>> acid, so why are you applying a pull restraint to them?  I really don't
>>> understand.
>>>
>>>
>> I want to make sure sudden conformational changes of amino acid do not
>> occur during the perturbation. In particular, when the charge is turned
>> off.  Applying a harmonic restraint to keep the geometry the same during
>> FEP is a well-established procedure, e.g. Deng, Y.; Roux, B. J Chem Theory
>> Comput 2006, 2 (5), 1255. I might reduce the number of restraints to only
>> between 1 or 2 pairs.
>>
>>
> Preserving the A-state in the bonded topology (and using couple-intramol =
> no) will prevent any weirdness from happening without needing any of these
> restraints.  As in my previous message, restraining CA-HA with a harmonic
> potential makes no sense at all.  Those atoms have a bond between them.
> The pull code is not doing anything useful.
>

Then, If the " couple-intramol = yes" hopefully solves the problem
discussed above, then, maybe applying restrain in the presence of "
couple-intramol = yes" is not avoidable.


>
> The whole task is to calculate the binding free energy of amino acid to a
>> metal surface, although here I am still dealing with the amino acid in
>> only
>> water without surface yet.
>>
>
> I believe I've mentioned this before, but in case it got lost along the
> way - using the free energy decoupling technique is a very ineffective way
> of calculating this binding free energy.  Do a PMF.  It's extremely
> straightforward and you don't deal with any of these algorithmic problems.
> It will also likely converge a lot faster than try to do complex decoupling.
>

Actually I should have known this in beginning, but, now is a bit late for
me to switch to PMF.

Best regard,
Alex