Significance of the Chemical Environment of an Element in Nonadiabatic Molecular Dynamics: Feature Selection and Dimensionality Reduction with Machine Learning

Abstract

Using supervised and unsupervised machine learning (ML) on features generated from nonadiabatic (NA) molecular dynamics (MD) trajectories under the classical path approximation, we demonstrate that mutual information with the NA Hamiltonian can be used for feature selection and model simplification. Focusing on CsPbI 3 , a popular metal halide perovskite, we observe that the chemical environment of a single element is sufficient for predicting the NA Hamiltonian. The conclusion applies even to Cs, although Cs does not contribute to the relevant wave functions. Interatomic distances between Cs and I or Pb and the octahedral tilt angle are the most important features. We reduce a typical 360-parameter ML force-field model to just a 12-parameter NA Hamiltonian model, while maintaining a high NA-MD simulation quality. Because NA-MD is a valuable tool for studying excited state processes, overcoming its high computational cost through simple ML models will streamline NA-MD simulations and expand the ranges of accessible system size and simulation time.

Quantum machine learning for chemistry and physics

“…For instance, a model of metal halide perovskites (MHPs) based on unsupervised MI unveiled the importance of geometric features as compared to the atomic velocities while predicting non-adiabatic Coupling (NAC). 596 In a recent study by How et al, 600 supervised and unsupervised ML techniques have been used for feature selection, prediction of non-adiabatic couplings, and excitation energies of NA MD simulations of CsPbI 3 metal halide perovskites (MHPs). MHPs have high optical absorption, low cost of manufacturing and long carrier diffusion [601][602][603] which make them an ideal candidate for their use in optoelectronics and solar energy harvesting materials.…”

“…In order to improve the design of MHPs it is important to develop a computationally efficient and a systematic NA MD which utilizes the theory as well as simulations to predict the physical properties of MHPs. How et al 600 fill this knowledge gap by employing MI on the NA MD trajectory data set of CsPbI 3 perovskite and extracting the most important features that determine the NA Hamiltonian. The ML model is then validated by checking the performance of the extracted important features to predict the band gap and NAC.…”

“…In a recent study by How et al , 600 supervised and unsupervised ML techniques have been used for feature selection, prediction of non-adiabatic couplings, and excitation energies of NA MD simulations of CsPbI 3 metal halide perovskites (MHPs). MHPs have high optical absorption, low cost of manufacturing and long carrier diffusion 601–603 which make them an ideal candidate for their use in optoelectronics and solar energy harvesting materials.…”

The Principal Component Analysis of the Ring Deformation in the Nonadiabatic Surface Hopping Dynamics

“…In addition, some efforts were also made to employ the unsupervised machine learning approaches to analyse the nonadiabatic dynamics of the solid state systems. 42,43 Recently, the unsupervised machine learning approaches were also applied by Choi et al to understand and propagate the dynamical evolution of open quantum systems. 44 The deformation of aromatic ring widely exists in many nonadiabatic dynamics processes, including the photostability of the DNA bases, [45][46][47][48][49][50] the internal conversion of the sunscreen molecules, 51,52 the "channel-III" nonradiative process of benzene and its simple derivatives, 53,54 etc.…”

The principal component analysis of the ring deformation in the nonadiabatic surface hopping dynamics

“…23 In addition, some efforts were also made to employ the unsupervised machine learning algorithms to analyse the nonadiabatic dynamics of solid state systems. 46,47 Recently, the unsupervised machine learning algorithms were also applied by Choi et al to understand and propagate the dynamics evolution of open quantum systems. 48…”