QM/MM Interaction When Using MLCG Force Field via PLUMED

[Somayeh Ahmadkhani]

Dear CP2K users and developers,

I am using PLUMED to connect a machine learning force field (MLCG) trained on an ab initio dataset.

I would like to clarify the following point: In a QM/MM calculation in CP2K, if I apply this MLCG force field (via PLUMED) to the MM region, what happens to the QM region and the QM/MM interaction term?

Specifically, I have a coarse-grained (CG) force field for the MM region that is applied through PLUMED. Since this ML-based force field does not include explicit charge information, I am wondering:

• Does it have any effect on the QM region itself?

• How is the QM/MM interaction energy affected?

• Are the Hellmann–Feynman forces still governing the QM region, or does the MLCG force field influence those calculations?

• How is the interaction term treated when the MM region is described by a CG machine-learning potential?

I would appreciate any clarification on how CP2K handles the interaction term in this setup.

Best regards,
Somayeh Ahmadkhani

[Marcella Iannuzzi]

Hello Somayeh Ahmadkhani

If the MM part does not have charges, there is no electrostatics between QM and MM.

Whether other interactions are calculated or not it depends on your MM (MLCG) force files. Does it consider interactions between the MM and QM part?

Regards

Marcella