Comment (by Peter Diener):

I suggest adding the following to the end of section 3.4:

In both cases, the derivatives are not calculated in ghost zones of
inter-processor or symmetry boundaries, i.e. the returned derivatives are
undefined in ghost zones and will require synchronization of either the
derivatives or the evolved fields depending on the scheme used.

It would be wrong to switch to use one sided derivatives in ghost zones as,
for most operators, the region, where one sided derivatives would be
calculated, is larger than the ghost region. Hence some cells where centered
finite differences could be calculated would contain one sided derivatives and
the result would depend on the number of processors used.

--
Ticket URL: https://bitbucket.org/einsteintoolkit/tickets/issues/2633/summationbyparts-diff_gv-aliased-function