Please post to the list, not to me.
b) running any minimization to 1.0e-12 accuracy is a calculation
sensitive to small numerical differences. Do you get similar
differences if you run on different numbers of processors?
Yes, I tried 16 cpus and 8 cpus.
Let me clarify my question. You say you get “big”
differences when you run with balance vs no-balance.
I’m asking if you also get similar “big” differences
when you run with no-balance on different
#s of procs. If so, then I’m guessing it has
nothing to do with balancing, but with simple round-off
issues in a minimization performed to extremely high
tolerance, which is a not a stable numerical process
for many kinds of problems. I.e. small differences
can lead to divergent trajectories in the minimization
sense.
Steve