Interpolating Nonadiabatic Molecular Dynamics Hamiltonian with Artificial Neural Networks

Abstract

Nonadiabatic (NA) molecular dynamics (MD) allows one to study farfrom-equilibrium processes involving excited electronic states coupled to atomic motions. While NAMD involves expensive calculations of excitation energies and NA couplings (NACs), ground-state properties require much less effort and can be obtained with machine learning (ML) at a fraction of the ab initio cost. Application of ML to excited states and NACs is more challenging, due to costly reference methods, many states, and complex geometry dependence. We developed a NAMD methodology that avoids time extrapolation of excitation energies and NACs. Instead, under the classical path approximation that employs a precomputed ground-state trajectory, we use a small fraction (2%) of the geometries to train neural networks and obtain excited-state energies and NACs for the remaining 98% of the geometries by interpolation. Demonstrated with metal halide perovskites that exhibit complex MD, the method provides nearly two orders of computational savings while generating accurate NAMD results.

Common Defects Accelerate Charge Carrier Recombination in CsSnI3 without Creating Mid-Gap States

“…After that, 9 ps adiabatic MD trajectories are obtained in a microcanonical ensemble with a 1 fs atomic time step. The nonadiabatic couplings (NACs) are calculated using the CA-NAC package, , considering the overlap of two wave functions at adjacent timesteps …”

Dependence between Structural and Electronic Properties of CsPbI3: Unsupervised Machine Learning of Nonadiabatic Molecular Dynamics

“…In supervised learning a labeled output is used to learn or approximate relationships between input and observed output. 25 Unsupervised learning has no labeled output and is used to infer hidden structures and similarities in a data set. As demonstrated in previous work, 24 insights into the charge carrier dynamics of halide perovskites.…”

“…ML can be either supervised or unsupervised. In supervised learning a labeled output is used to learn or approximate relationships between input and observed output . Unsupervised learning has no labeled output and is used to infer hidden structures and similarities in a data set.…”

Modeling Non-adiabatic Dynamics in Nanoscale and Condensed Matter Systems

“…Unsupervised learning can be employed to analyze NA-MD Hamiltonians and simulation results, 65 while supervised learning can help construct NA Hamiltonians and accelerate simulation. 4,66,67 tron−hole recombination, which limits efficiencies of perovskite solar cells, 65 Figure 7. Surprisingly, geometries rather than motions govern the NA coupling, although it explicitly depends on atomic velocity, eq 1.…”