Pavlo O. Dral in collaboration with Mario Barbatti from Aix-Marseille-University (France) have described state-of-the-art, provided insights and highlighted challenges concerning application of machine learning for excited-states simulations and description. This Review was published online in Nat. Rev. Chem. 2021, with the title of "Molecular excited states through a machine learning lens".
Machine learning emerges as a breakthrough technique for improving accuracy and speed of quantum chemical simulations of excited-state simulations. It can also completely replace quantum chemical methods in simulations, e.g., by learning directly on the experimental data. Analysis of experimental observations with machine learning can enable efficient structure determination and discover rules for materials design.
Optoelectronic materials design is enjoying the developments in machine learning as it not just provides rules for design, but allows for accelerated high-throughput screening of existing databases and can guide the automatic search in the chemical space. Such approaches already resulted in discover of many novel materials and in the future, we expect to see more examples of intelligent materials discovery, particularly when machine learning is merged with robotic laboratories.
The Review also described Dral group’s contributions to the field of machine learning simulations of excited states, such as development of methods for nonadiabatic dynamics (J. Phys. Chem. Lett. 2018, 9, 5660; 2018, 9, 6702) and UV/vis absorption spectra simulations (J. Phys. Chem. A 2020, 124, 7199).
The research was supported by the National Natural Science Foundation of China (22003051). The Review is dedicated to 100th anniversary of Xiamen University.
Link to the paper: https://www.nature.com/articles/s41570-021-00278-1
