Extending the Time Scales of Nonadiabatic Molecular Dynamics via Machine Learning in the Time Domain

Abstract

A novel methodology for direct modeling of long-time scale nonadiabatic dynamics in extended nanoscale and solid-state systems is developed. The presented approach enables forecasting the vibronic Hamiltonians as a direct function of time via machine-learning models trained directly in the time domain. The use of periodic and aperiodic functions that transform time into effective input modes of the artificial neural network is demonstrated to be essential for such an approach to work for both abstract and atomistic models. The best strategies and possible limitations pertaining to the new methodology are explored and discussed. An exemplary direct simulation of unprecedentedly long 20 picosecond trajectories is conducted for a divacancy-containing monolayer black phosphorus system, and the importance of conducting such extended simulations is demonstrated. New insights into the excited states photophysics in this system are presented, including the role of decoherence and model definition.

Simulations of molecular photodynamics in long timescales

“…Nevertheless, most photoinduced processes occur in much longer timescales, several orders of magnitude over our current research capabilities. A few groups, including ours, are already probing dynamics simulations into these long timescales [4][5][6][7][8][9][10][11][12][13].…”

Strain modulation of the exciton anisotropy and carrier lifetime in black phosphorene

“…46 For comparison, the carrier lifetimes of the other materials 47–58 are listed in Table 2. Although the calculated carrier lifetimes of monolayer BP are influenced by the setups, the methodologies 59 and the band number 60 used, the results show that the strain and doping can extend the carrier lifetime of the system. The measured carrier lifetimes are influenced by the pump power, the thickness of the sample, and so on, but the experimental results suggest that the carriers in the heterostructures possess a longer lifetime.…”

“…46 For comparison, the carrier lifetimes of the other materials [47][48][49][50][51][52][53][54][55][56][57][58] are listed in Table 2. Although the calculated carrier lifetimes of monolayer BP are influenced by the setups, the methodologies 59 and the band number 60 used, the results show that the strain and doping can extend the carrier lifetime of the Table 2 Carrier lifetimes of 2D materials (in units of ps), including monolayer BP (1L), 1L with a phosphorus divacancy (1L-DV), 1L with a phosphorus vacancy (1L-P v ), 1L with interstitial phosphorus (1L-P int ), 1L with a phosphorus adatom (1L-P ad ), 1L with 1% strain (1L-1%), 1L with 1.5% strain (1L-1.5%), 1L with 2% strain (1L-2%), bilayer phosphorus (2L), 2L with 3% strain (2L-3%), a single O atom adsorbed on BP (BP-O), two O atoms adsorbed separately on the top and bottom surface (BP-2O), a phosphorus atom substituted by an O atom (BP-O sub ), the BP-O structure interacting with an H 2 O molecule (BP-O-H 2 O), the BP-2O structure interacting with an H 2 O molecule (BP-2O-H 2 O), BP doped with a N/As/Sb/Bi atom (BP-N, BP-As, BP-Sb, BP-Bi), BaZrS 3 (BZS), BZS with 12.5 or 25% Ti/Hf doping (BZTS, BZHS), BZS with 50% sulfur substitution by oxygen (BZSO) or selenium (BZSSe), phosphorus nanosheets/ multilayer/flakes, heterostructures (BP/polymer, SiH/a-TiO 2 , SiH/r-TiO 2 , GeH/a-TiO 2 ), and crystalline TiO 2 system. The measured carrier lifetimes are influenced by the pump power, the thickness of the sample, and so on, but the experimental results suggest that the carriers in the heterostructures possess a longer lifetime.…”

A comparative study of different machine learning methods for dissipative quantum dynamics

“…Recently, many ML models have been applied to simulate the dynamics of quantum systems. [32][33][34][35][117][118][119][120][121][122][123][124][125][126] We note that ML can also be applied to quantum dynamics in a different context-namely as surrogate models for quantum chemical properties such as potential energies and forces in different electronic states as well as cou-plings between states eliminating the need for expensive (excited-state) electronic structure calculations. [127][128][129] Here we apply ML to propagate a quantum system assuming potential energies are readily available.…”

“…We note that ML methods based on FFNNs have also been recently applied to model quantum dynamics. 117,118 The recent upsurge of applications of ML methods to dissipative quantum dynamics calls for a systematic benchmark of such methods.…”