Abstract
The main purpose of this work is to develop a robust approach to accelerate QM
photodynamics simulation. DFT calculation is conducted to provided a standard of the predicted properties for each molecular geometry. CNNs are applied to the descriptors to predict energies and forces for all states rapidly. Active learning is coordinated with the simulation itself, which guarantees that the underexplored regions are calculated with QM, and the model covers the region where the trajectories are reaching.
Conclusions:
* With the QM-ML workflow,QM photodynamics is accelerated 14.5 times
* Ensemble LOL + SchNet predict energy and forces more accurately and is
more reliable on querying undersampled points
* Incorporating active learning with the simulation enhances the robustness
of the workflow
* Results of QM-ML + ZNSH simulation agree well with QM + FSSH
* More trajectories with longer time lengths can be simulated
photodynamics simulation. DFT calculation is conducted to provided a standard of the predicted properties for each molecular geometry. CNNs are applied to the descriptors to predict energies and forces for all states rapidly. Active learning is coordinated with the simulation itself, which guarantees that the underexplored regions are calculated with QM, and the model covers the region where the trajectories are reaching.
Conclusions:
* With the QM-ML workflow,QM photodynamics is accelerated 14.5 times
* Ensemble LOL + SchNet predict energy and forces more accurately and is
more reliable on querying undersampled points
* Incorporating active learning with the simulation enhances the robustness
of the workflow
* Results of QM-ML + ZNSH simulation agree well with QM + FSSH
* More trajectories with longer time lengths can be simulated
| Original language | English |
|---|---|
| Number of pages | 1 |
| Publication status | Published - 21-Jun-2022 |
| Event | 12th Congress on Electronic Structure Principles and Applications (ESPA-2022) - Valencia, Spain Duration: 21-Jun-2022 → 24-Jun-2022 Conference number: 12 https://espa2022.webs.uvigo.es/welcome/ |
Conference
| Conference | 12th Congress on Electronic Structure Principles and Applications (ESPA-2022) |
|---|---|
| Abbreviated title | ESPA2022 |
| Country/Territory | Spain |
| City | Valencia |
| Period | 21/06/2022 → 24/06/2022 |
| Internet address |