Our active learning protocol for accelerating surface hopping dynamics with machine learning is now available in MLatom 3.10! You can check out the video for more information.

It is efficient and robust: often, you can do surface-hopping dynamics from start to finish within a couple of days on a single GPU!

We have come up with multiple innovations to make it work:

• multi-state learning model (MS-ANI) that has unrivaled accuracy for excited state properties (accuracy is often better than for models targeting only ground state!). We demonstrate that this model can be used for trajectory-surface hopping of multiple molecules (not just for a single molecule!)
• gapMD for efficient sampling of the vicinity of conical intersection
• samplings based on uncertainty in hopping probabilities
• and many other technical solutions making life easier (such as automatic uncertainty thresholds identification).

Many of these innovations can be handy in other than surface hopping applications.

How to run the simulations?

Please update your MLatom:

pip install --upgrade mlatom

The open-source code and tutorials on how to use this protocol are available as described at https://github.com/dralgroup/al-namd. Please let us know if you are interested in running the protocol on our XACS cloud computing platform (this would need lots of extra work, so this is to survey whether there is enough interest).

Theory and applications of the protocol

The details of the protocol and examples of its applications are given in our preprint:

1.      Mikołaj Martyka, Lina Zhang, Fuchun Ge, Yi-Fan Hou, Joanna Jankowska, Mario Barbatti, Pavlo O. Dral. Charting electronic-state manifolds across molecules with multi-state learning and gap-driven dynamics via efficient and robust active learning. 2024. ChemRxiv: https://doi.org/10.26434/chemrxiv-2024-dtc1w.