[gmx-users] Again: Non-integer charges in tpr-files

[Erik Lindahl]

K.A. Feenstra wrote:
> Erik Lindahl wrote:
> 
>>I actually tried to "fix" this a long time ago, but it isn't trivial; we
>>are summing a lot of charges, and due to the way they are represented in
>>hardware they aren't exact. One option would of course be to only warn
>>when the net charge is more than 0.01% of the system charge RMSD or
>>something, but that might hide things like having one charged lipid
>>among 1024 uncharged ones... I'll see if we attack it again.
> 
> 
> I've though about this a bit a while ago. We possibly could re-set the
> total charge after every charge-group, providing the total charge at 
> that point is 'integer enough'? I.e. if we end up with 0.000001, we
> re-set to zero, if we end up with 9.999999e-1 we re-set to 1. That 
> would reduce the number charges summed at one time.
>  

Won't work, since we don't request that all charge groups are strictly 
neutral. It's only with cutoffs that the neutrality is important, but 
the concept of charge groups still speeds up neighborsearching quite a lot.

Cheers,

Erik