Brian - I just ran your in.Test and it gave no problems on my box, with
either the g++ or Intel compiler. I can run on 1,2,4 procs and get
essentially the same answers after 20 timesteps (no crash during setup).
This is with the long 19.5 cutoff.
I also ran it with valgrind on 1 and 2 procs and there were no issues.
I think you will need to do the same on your box. Note that if
you do a mpirun -np 2 valgrind lmp_g++ < in.Test that valgrind
tends to dump out lots of "errors" from MPI that are not true errors.
It takes a practiced eye to pick out the ones that are actual code
problems.
This testing was all with the current (patched thru Nov 6) version.
The aliasing compile flags shouldn't matter for this pair style. Or
at all really, in this current version, since Axel sent some better code
for a few potentials that doesn't have the aliasing issue anymore.
Steve