Closed pjuda closed 6 years ago
might be or might not. if the execution is not threaded then the MC should pick from the same PRNG stream as everyhting else and so results with the same seed should be reproducible.
Fixed. Was using wrong PRNG source.
To reproduce the issue, checkout fix-examples branch, go to examples/regtest/lammps and execute regtest to create a reference run:
../../../tools/py/regtest.py --create-reference --tests-folder . --folder-run run
Next, run the regtest:
../../../tools/py/regtest.py --tests-folder . --folder-run run
You will notice that h2o-alchemical fails the regtest. Is it expected behaviour due to the fact that Monte Carlo procedure uses a different seed for each run?