[gmx-developers] scaling 1-4 electrostatic interactions separately

[Berk Hess]

chris.neale at utoronto.ca wrote:

> Quoting hessb at mpip-mainz.mpg.de:
>
>>> Quoting David van der Spoel <spoel at xray.bmc.uu.se>:
>>>
>>>> chris.neale at utoronto.ca wrote:
>>>>
>>>>> Hello,
>>>>>
>>>>> I am wondering why there is a [ pairtypes ] section only for LJ and
>>>>>  not coulombic. I have been able to trick gromacs into properly
>>>>> combining two forcefields that have a different fudgeQQ, but that
>>>>> trick is fairly specific:
>>>>> http://www.gromacs.org/pipermail/gmx-users/2006-September/023761.html
>>>>
>>>> Without looking things up you could use multiple pair interactions for
>>>> each pair, but that will only help if one fudgeQQ_A = n * fudgeQQ_B
>>>> with n integer.
>>>
>>>
>>> Thank you for your comments. This is exactly what I have done. Where
>>> fudgeQQ_A = 0.5 (opls) and fudgeQQ_B = 1.0 (lipids). I then specify [
>>> pairtypes ] for group B as half of what they should be and include the
>>> 1-4 pairs twice for group B, while setting fudgeQQ = 0.5. At first
>>> glance it is simple enough to warrant only a single sentence in a
>>> methods section, but I had to test it since one can imagine that a
>>> second pair might overwrite the first or something else unexpected
>>> since I don't think that gromacs was coded with this usage in mind. In
>>> fact it works perfectly with the only difference being rounding errors
>>> from slightly different mathematical form (two calculations instead of
>>> one).
>>>
>>>>
>>>>
>>>>>
>>>>> I am in the process of writing up the results of my protein-lipid
>>>>> simulations and I am currently trying to decide how to split the
>>>>> paper. I would like to split the method plus proof-of-concept
>>>>> section out into a simple one page methods letter but in that case
>>>>> I would need to convince the reader that the method that I have
>>>>> used is necessary and could not easily be replaced by a slight code
>>>>>  modification to gromacs. In fact this may be the case and is the
>>>>> reason why I am writing to this list.
>>>>>
>>>>> First: is there some important reason why a [ coulombic-pairtypes ]
>>>>>  section could not exist, or is it just that nobody has dedicated a
>>>>>  day of coding to make this happen? I figure that PME could make it
>>>>>  more difficult, but don't really know.
>>>>>
>>>>> Second: is gromacs 4.x likely to allow users to arbitrarily set
>>>>> coulombic 1-4 interactions in the same way that this can be done
>>>>> for LJ 1-4 interactions by [ pairtypes ] ?
>>>>
>>>>
>>>> Section VIII of the gromacs 4 paper desribes very briefly the new
>>>> tabulated bonded potentials. It could however be that there is not
>>>> something like a Coulomb interaction, that multiplies a function by 
>>>> the
>>>> charges (like there is for non-bonded potentials). Maybe Berk can
>>>> comment.
>>>
>>>
>>> Thanks for pointing this out. I did take a look at this section, but
>>> didn't find any particular mention here for exactly what I am
>>> describing. For other readers, the section referred to reads:
>>>
>>> "User defined nonbonded interactions that can be assigned
>>> independently for each pair of charge groups were already supported,
>>> and we have now additionally implemented user defined bonds, angles,
>>> and dihedrals functions."
>>>
>>> It would be great if you could comment here Berk.
>>>
>>> Thank you again,
>>> Chris.
>>
>>
>> User defined bond potentials can do what you want.
>> But it might be very tedious, depending on the complexity of your 
>> system.
>> Everything is described in the CVS pdf manual.
>> You have c*f(r) at your disposal, where one could choose
>> c and f different for every atom pair you defined.
>> So you could set fudgeQQ to zero and define all Coulomb-14 through 
>> tables.
>> But if you need many different atom pairs with different charges,
>> you need to define the proper c for every one.
>>
>> But it would be much better to be able to set fudgeQQ per molecule,
>> which by itself would not be difficult, but we would need to
>> extend the top file format.
>>
>> Berk.
>
>
> Thank you for the explanation. Are you likely to include the  
> per-molecule fudgeQQ in gromacs 4.x? I think that it would be a good  
> thing to include but am mostly just interested in knowing if you are  
> likely to include it.
>
> Thanks,
> Chris.

I have just implemented a [ pairs ] type 2, which takes as parameters:
fudgeQQ, q_i, q_j, V, W
You should be able to supply only fudgeQQ and the charges and LJ parameters
should be looked up on atom/atomtype.

I haven't tried it yet (I just use these for free-energy decoupling, 
where grompp
sets things automatically).

Please report back when you have them.

Note that it would be better to be able to set fudgeQQ per molecule,
but currently the interactions in Gromacs are not linked to molecules.

Berk.