[gmx-developers] WARNING: Found more than 12 triclinic correction vectors, ignoring some.

[Berk Hess]

Tsjerk Wassenaar wrote:

> Hi Berk,
>
> No, I'm not using CVS code. But I was not aware of this requirement,
> which is not mentioned in the manual. There it's stated (equation 3.3)
> that neither |box[1][x]| nor |box[2][x]| should be larger than
> box[0][x]/2. It doesn't mention the combination. Also, although it is
> always possible to transform the box to meet the requirements
> mentioned in the manual, I believe this is not so for the requirement
> you mention. That would mean that certain lattice arrangements are
> thus forbidden in Gromacs. Or am I missing something?
>
> What is the consequence of having more than 12 correction vectors? I
> presume they are used to short cut determination of minimal distances.
> So, with more correction vectors, this may be slower in some cases? It
> seems not to affect the simulation in any case.
>
Ah, your are right, the box requirements only work on the vector components
separately.
So I guess this means I have made a mistake when I determined that
one would never need more than 12 correction vectors,
or there is a mistake in the algorithm for determining these vectors.

These vectors are used to try to obtain shorter distances after a simple
pbc operation where the 3 components of the distance vector between
atoms are made smaller than half the box lengths.

In most cases this error is not a problem as you are usually only interested
in distances that are smaller than half the shortest periodic image 
distance.
But there could be problems for longer distances.

I currently don't have time to look into this.
Could you sumbit a bugzilla entry?

Berk.