Machine learning accelerated nonadiabatic dynamics simulations of materials with excitonic effects

The Journal of Chemical Physics, Journal Year: 2025, Volume and Issue: 162(2)

Published: Jan. 8, 2025

This study presents an efficient methodology for simulating nonadiabatic dynamics of complex materials with excitonic effects by integrating machine learning (ML) models simplified Tamm–Dancoff approximation (sTDA) calculations. By leveraging ML models, we accurately predict ground-state wavefunctions using unconverged Kohn–Sham (KS) Hamiltonians. These ML-predicted KS Hamiltonians are then employed sTDA-based excited-state calculations (sTDA/ML). The results demonstrate that energies, time-derivative couplings, and absorption spectra from sTDA/ML accurate enough compared those conventional density functional theory based sTDA (sTDA/DFT) Furthermore, sTDA/ML-based molecular simulations on two different systems, namely chloro-substituted silicon quantum dot monolayer black phosphorus, achieve more than 100 times speedup the linear response time-dependent DFT simulations. work highlights potential ML-accelerated studying complicated photoinduced large offering significant computational savings without compromising accuracy.

Language: Английский

---

Advancing nonadiabatic molecular dynamics simulations in solids with E(3) equivariant deep neural hamiltonians

Nature Communications, Journal Year: 2025, Volume and Issue: 16(1)

Published: Feb. 27, 2025

Abstract Non-adiabatic molecular dynamics (NAMD) simulations have become an indispensable tool for investigating excited-state in solids. In this work, we propose a general framework, N 2 AMD (Neural-Network Non-Adiabatic Molecular Dynamics), which employs E(3)-equivariant deep neural Hamiltonian to boost the accuracy and efficiency of NAMD simulations. Distinct from conventional machine learning methods that predict key quantities NAMD, computes these directly with Hamiltonian, ensuring excellent accuracy, efficiency, consistency. not only achieves impressive performing at hybrid functional level within framework classical path approximation (CPA), but also demonstrates great potential predicting non-adiabatic coupling vectors suggests method go beyond CPA. Furthermore, generalizability enables seamless integration advanced techniques infrastructures. Taking several extensively investigated semiconductors as prototypical system, successfully simulate carrier recombination both pristine defective systems large scales where often significantly underestimates or even qualitatively incorrectly predicts lifetimes. This offers reliable efficient approach conducting accurate across various condensed materials.

Language: Английский

---

Improvement of Fourteen Coupled Global Potential Energy Surfaces of ³A′ States of O + O₂

The Journal of Physical Chemistry A, Journal Year: 2025, Volume and Issue: unknown

Published: March 20, 2025

We improved the potential energy surfaces for 14 coupled 3A' states of O3 by using parametrically managed diabatization deep neural network (PM-DDNN) with three improvements: (1) used a new functional form activation function, which ensures continuity coordinates in parametric management. (2) higher weighting low-lying to achieve smoother surfaces. (3) The asymptotic behavior was further refined utilizing better low-dimensional potential. As result these improvements, we obtained significantly potentials that are suited dynamics calculations. For version surfaces, entire set 532,560 adiabatic energies fit mean unsigned error (MUE) 45 meV, is only 0.7% data set, 6.24 eV.

Language: Английский

---

Charting electronic-state manifolds across molecules with multi-state learning and gap-driven dynamics via efficient and robust active learning

npj Computational Materials, Journal Year: 2025, Volume and Issue: 11(1)

Published: May 13, 2025

Language: Английский

---

Notizen aus der Forschung

Nachrichten aus der Chemie, Journal Year: 2024, Volume and Issue: 72(12), P. 46 - 49

Published: Dec. 1, 2024

---