[gmx-developers] User defined potential between specific atoms

[hessb at mpip-mainz.mpg.de]

>> I am using the CVS version (downloaded yesterday).
>>
>> I am trying to include interactions between specific atoms (i.e. atom 1
>> and
>> 40 are interacting and atom 30 and 80 are interacting).  I am using
>> "pairs"
>> to do this.  If I look at  large systems, where two bodies can
>> dissociated,
>> I run into a problem with dynamics load balancing.  Since "pairs" is
>> considered  a bonded interaction I need to set "-rdd" to a large value.
>> If
>> i do this, then I can not run on many nodes because the domains are
>> forced
>> to be too large.  Is there an alternate way to include these
>> interactions
>> that won't run into this issue with DD?  I have looked at the user
>> defined
>> potentials, but it appears you can only set the form of the potential,
>> which
>> is then used in "pairs".
>
> Possibly a bond potential type 6 might work, but I can't think of a reason
> this wouldn't produce the same kind of problem.

It would.

There are two ways of solving this problem.
You could make these atoms separate energy groups
and use the energygrp_table option to supply a tabulated
non-bonded interaction.

You could also simply remove the missing interactions check in the code.
Since somebody else also asked about this, we might consider
adding an option to mdrun to turn off this check, or to turn it off
for interactions other than normal bond, angle and dihedral potentials.

Note that your problem is not specific for dynamic load balancing.
Setting -rdd much larger than your non-bonded cutoff distance
will kill your performance.

Berk.