logging at key points shows:
msbdf_apply of order 4
finish msbdf_apply
step accepted
msbdf_apply of order 4
Stability enhancement heuristic for msbdf - decrease order to 3
finish msbdf_apply
step rejected
msbdf_apply of order 3
finish msbdf_apply
step rejected
msbdf_apply of order 3
msbdf_eval_order /* Consider and execute order change for next step */
decreases order to 2
finish msbdf_apply
step rejected
msbdf_apply of order 2
4-2>1 => sanity check kills the simulation
It seems like reproducibility of this behaviour depends on cpu
architecture, fpu settings, libm, floating point related compiler
flags, maybe even weather or moon phase. I can provide a core dump or
longer and more detailed logs in case of good weather. Feel free to ask
for specific details.
Fero
