Interpolating Nonadiabatic Molecular Dynamics Hamiltonian with Bidirectional Long Short-Term Memory Networks

Abstract

Essential for understanding far-from-equilibrium processes, nonadiabatic (NA) molecular dynamics (MD) requires expensive calculations of the excitation energies and NA couplings. Machine learning (ML) can simplify computation; however, the NA Hamiltonian requires complex ML models due to its intricate relationship to atomic geometry. Working directly in the time domain, we employ bidirectional long short-term memory networks (Bi-LSTM) to interpolate the Hamiltonian. Applying this multiscale approach to three metal-halide perovskite systems, we achieve two orders of magnitude computational savings compared to direct ab initio calculation. Reasonable charge trapping and recombination times are obtained with NA Hamiltonian sampling every half a picosecond. The Bi-LSTM-NAMD method outperforms earlier models and captures both slow and fast time scales. In combination with ML force fields, the methodology extends NAMD simulation times from picoseconds to nanoseconds, comparable to charge carrier lifetimes in many materials. Nanosecond sampling is particularly important in systems containing defects, boundaries, interfaces, etc. that can undergo slow rearrangements.

Photocatalytic activity of dual defect modified graphitic carbon nitride is robust to tautomerism: machine learning assisted ab initio quantum dynamics

“…Our group has recently demonstrated that the computational cost of NAMD can be reduced by interpolating the NAMD Hamiltonian along an MLFF trajectory. [40][41][42][43] In this work, we report a multiscale methodology and study the coupled structural evolution and quantum dynamics of charge carriers in dual defect ON-GCN over a nanosecond timescale by combining NAMD and real-time time-dependent density functional theory (RT-TDDFT) with supervised ML learning. We train an MLFF to investigate structural changes in dual defect ON-GCN over nanosecond MD trajectories, revealing hydrogen hopping involving four tautomeric structures.…”

“…Our group has recently demonstrated that the computational cost of NAMD can be reduced by interpolating the NAMD Hamiltonian along an MLFF trajectory. 40–43…”

Estimating Nonradiative Excited-State Lifetimes in Photoactive Semiconducting Nanostructures

“…The trajectory analysis requires ab initio molecular dynamics (AIMD) or, alternatively, interatomic potentials or tight-binding DFT (TBDFT) methods. Fortunately, recent machine learning (ML) techniques have been implemented in NAMD simulations. − ML models may overcome the computational time scale limitation, especially in large systems composed by thousands of atoms. Furthermore, the analysis of excitation energies requires TDDFT calculations over each one of the systems collected in the trajectory and also contributes to the computational increase.…”

Nonadiabatic Dynamics in Two-Dimensional Perovskites Assisted by Machine Learned Force Fields

“…Ab initio molecular dynamics (AIMD) are used to determine how atoms move over time, but AIMD is computational demanding which imposes limits on the size of models studied. − One potential way to address this problem is by combining AIMD simulations and machine learning to generate machine learned force fields (MLFFs). , Large and accurate data sets are used to construct traditional MLFFs; therefore, there is a lot of trial and error involved in parameter optimization and data selection . Generating on-the-fly MLFF through active learning schemes gives the advantage that no prior training is required and the on-the-fly force fields are generated automatically during the AIMD simulation. , Using this approach, the results can be achieved with the accuracy of AIMD while also speeding up the simulation .…”