Cartesian vs. spherical Gaussians in Quickstep

[Ondrej Marsalek]

Hi,

I would like to understand under what circumstances Quickstep uses
spherical Gaussians as opposed to Cartesian Gaussians. From reading
the 2005 Quickstep paper, I was under the impression that it only ever
uses Cartesians. However, from some recent confusion in the output of
some jobs a brief look at the code (did not go too deep, yet) it seems
that spherical Gaussians come into play as well. Before I bring
specific input file which triggered this, could someone please give me
some general information on this topic?

Thanks a lot,
Ondrej