[gmx-users] Shift functions

[Mark Abraham]

On 11/02/2012 1:19 AM, Elisabeth wrote:
>
>>     Hello all,
>>
>>     Does the shift function use group based truncation?
>
>     See the discussion of charge groups in manual section 3.4.2.
>
>
> Thanks Mark.
>
> -1- First of all if I am right  charge groups in gromacs language in 
> identical to "group based truncations"?

Using charge groups as the indivisible entity upon which neighbour list 
construction is based is using "group based truncations". The usual 
alternative is using atoms as the, well, *atomic* unit. The former can 
be equivalent to the latter if there's one atom per charge group.

>
>
>  Manual 342: "This reduces the cut-off effects from the charge-charge 
> level to the dipole-dipole level, which decay much faster"
>
> 2- I am not able to realize why we go from charge-charge the 
> dipole-dipole changes?

Charge groups are constructed to have neutral charge (or integer charge 
where necessary). To first order, the difference between any such group 
being in the neighbour list of another such group or not (according to 
the cut-off radius) is equivalent to a point dipole being in the cut-off 
sphere or not. Charge groups with arbitrary charge (or partially-charged 
atoms, when using atom-based truncation) do not have this quality, and 
the distance at which a charge-charge interaction is significant is much 
larger than that of a dipole-dipole.

>
>
>
>>     In the manual I see: by using shifted forces there is no need for
>>     charge groups (=group based?!) in the neighbor list?
>>
>>     Can anyone shed some light on calculation of shifted forces?
>
>     What's not clear from the above and 4.1.5?
>
>
> 3- I understand that use of shift makes the potentials have continuous 
> derivatives at cutoffs but that how this makes use of charge groups 
> unnecessary, I dont see!

Remember that the neighbour list is constructed to permit the 
computation of a finite number of interactions with the central 
atom/group. You want to stop computing them when they're close enough to 
zero that you don't care. If they actually go to zero, then you don't 
care. If they decay as 1/r (charge-charge) then at typical r_c values 
you should care. If they decay as 1/r^2 (dipole-dipole, IIRC) then at 
typical r_c values things are OK.

If the value of the force is non-zero at the cut-off, then there is an 
interaction at that distance and not one just past that distance. This 
generates artefacts that are serious for non-zero charges at the kinds 
of cut-off distances for which force fields are parametrized, but much 
less serious if computed over neutral charge groups.

If the value of the force is zero at the cut-off (i.e. shift potential), 
then no atom or charge group has any interaction with the central 
group/atom at that distance, so you don't need to care about whether the 
truncation is based on atoms or groups. You do have to care about the 
effect of the modified Coulomb potential, however.

>
> 4- and based on 3,  shift forces dont neglect tail corrections for LJ 
> as cutoffs do? Am I correct?

There's a cut-off used with shifted forces (indeed, in GROMACS it 
*defines* the shift), so I don't understand your question.

Mark

>
> Thank you
>
>
>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-users/attachments/20120211/6e123c00/attachment.html>