Aidan Thompson

For our applications, QEq cost (relative to the force calcualtion) grows roughly logarithmically with atom count. For 1M atoms, it is 30%, but for 100M atoms it is 60%. I...

That might create new problems, now or in the future. If you don't want the intramolecular energy to be included, you can remove it using the gcmc keyword intra_energy: "When...

I looked at the code and I see that the real issue is in attempt_molecule_deletion_full(). When a molecule is initially "deleted," the atoms are not removed, but simply turned off....

I have changed USER-REAXC and KOKKOS packages, so that numbonds > 0 check is off by default, and is controled by option enobonds. It is checked in to upstream now....

I see. Taking numbers from your test scripts: -248.00334 - (-245.49798) = -2.50536 If "enobonds no" solves the problem, we could just document that in fix gcmc, but this is...

Can you post a minimal reproducer of the problem, with all input files, possibly output too?

Thanks! Your example uses neigh_modify exclude molecule/intra, as suggested in the discussion. That creates its own problems, as you demonstrate. Going back to your original problem, I think this is...

Thanks! I understand the problem better now. With full_energy molecule deletions, the bonded contributions to the intramolecular energy are turned off term-by-term in toggle_intramolecular(). However, there are also nonbonded intramolecular...

I discussed with originator of this proposed change (Pieter in 't Veld) and he agrees that current documentation is consistent with current code (i.e. p-SLLOD algorithm), hence no change is...

I have been following this discussion with interest, because it is relevant to other ML potentials that are starting to provide similar functionality. I avoided commenting until now, because I...