[gmx-developers] Exclude specific nonbonded interactions during md rerun for Verlet

[Bernd Doser]

Hi,

it is not only the force between two atoms but between two larger groups
of atoms. The topology exclusions are typically within one molecule, but
the do_force()-function get a gmx_localtop_t, which is a topology struct
for one MPI node, if I understand it correctly. If I extent these arrays
over all atoms, no nonbonded interactions will be calculated, also for
the water dimer. Therefore, I was thinking this would be a way for doing
the exclusions. But I don't know if this works also for more complex
structures or only for my tiny example.

Cheers,
Bernd

On 12.09.2014 21:35, Berk Hess wrote:
> Hi,
> 
> But topology exclusions can only be applied within, not between molecules.
> If you really want the force between two atoms only, you can calculate 
> the distance and manually calculate the force.
> You can also convert the trajectory to two atoms only and make a 
> topology for the two atoms and rerun with that.
> 
> Cheers,
> 
> Berk
> 
> On 09/12/2014 09:13 PM, Bernd Doser wrote:
>> Hi Berk, Hi Mark,
>>
>> thanks for the interesting suggestion. I will try it. However, what do
>> you think about my first idea, to extent the topology list, which will
>> be also excluded as nonbonded interactions?
>>
>> Cheers,
>> Bernd
>>
>>
>> On 12.09.2014 19:30, Berk Hess wrote:
>>> Hi,
>>>
>>> You can use the group scheme for the rerun only. If you use the same
>>> potential modifiers and add a buffer to rlist, you should get exactly
>>> the same answer as the Verlet scheme.
>>>
>>> We should come up with a way to do this with the Verlet scheme when we
>>> remove the group scheme.
>>>
>>> Cheers,
>>>
>>> Berk
>>>
>>> On 09/12/2014 05:20 PM, Bernd Doser wrote:
>>>> Hi Berk,
>>>>
>>>> thanks for your answer. We store force contributions between two atoms
>>>> to analyze them. This storage will be performed during a rerun. We want
>>>> to exclude the unneeded interactions to speed-up the rerun, since we are
>>>> only interested in forces between two atom groups.
>>>>
>>>> Cheers,
>>>> Bernd
>>>>
>>>> On 09/12/2014 04:21 PM, Berk Hess wrote:
>>>>> Hi,
>>>>>
>>>>> Why do you need to exclude energy group interactions during rerun?
>>>>> Since you will have all the energy group pair contributions you can
>>>>> manually subtract the terms you want to exclude. You wouldn't even need
>>>>> a rerun for that if you used the energy groups for the production run.
>>>>>
>>>>> Cheers,
>>>>>
>>>>> Berk
>>>>>
>>>>> On 09/12/2014 03:26 PM, Bernd Doser wrote:
>>>>>> Dear GROMACS developers,
>>>>>>
>>>>>> As the energy group exclusions are not implemented for the Verlet cutoff
>>>>>> scheme, I am search for a alternative possibility to exclude specific
>>>>>> nonbonded interactions during an md rerun.
>>>>>>
>>>>>> One idea is to use the topology exclusion list. The top->excl arrays
>>>>>> could be modified for example at the beginning of do_force() in
>>>>>> sim_util.c. I have tested it for a small protein and a water dimer and
>>>>>> it works. Is there any problem doing this for more complex structures or
>>>>>> have anyone a better idea to exclude nonbonded interactions to reduce
>>>>>> computation time?
>>>>>>
>>>>>> Thank you very much in advance!
>>>>>>
>>>>>> Best regards,
>>>>>> Bernd Doser
>>>>>>
>