Negative dielectric, it's a feature, not a bug!

Dear all,

I am using LAMMPS for an astrophysical study on molecular clouds. In
order to have a gravitational potential solver, I am "tweaking" the
Coulomb potential as follows:

dielectric = -1
charge = m_i*sqrt(G)
with G the gravitational constant.

This is not particularly pretty, but works fine. For a recently
submitted article, I used version (1 Nov 2013).

There has been a change in newer versions (I'm using (15 Aug 2014)),
where LAMMPS stalls when setting the dielectric to -1. The problem ist
the variable q2, in the old version it is divided by the dielectric
where in the new version it is not.

I was able to get a running gravitational solver by changing in the
function KSpace::qsum_qsq the line kspace.cpp:276 into:

Stan can look into this.