Machine Learning Accelerated Non-Adiabatic Molecular Dynamics Elucidates Local Polarization Effects on Non-radiative...

Area of Science:

• Materials Science
• Computational Chemistry
• Solid-State Physics

Background:

• Non-radiative recombination significantly limits the optoelectronic performance of halide perovskites.
• The role of local polarization in charge redistribution and its impact on recombination is not well understood.
• Conventional non-adiabatic molecular dynamics (NAMD) methods are computationally expensive for studying these processes.

Purpose of the Study:

• To develop a computationally efficient machine learning framework for NAMD simulations.
• To investigate the influence of local polarization on non-radiative recombination in halide perovskites.
• To identify strategies for suppressing non-radiative recombination and improving perovskite optoelectronic properties.

Main Methods:

• Developed Hefei-NAMD-S, a machine learning framework using stacked models for NAMD simulations.
• Validated the ML framework by comparing predicted non-adiabatic coupling and pure-dephasing times with first-principles calculations.
• Performed NAMD simulations to analyze the role of B-site local polarization in recombination dynamics.

Main Results:

• The Hefei-NAMD-S framework achieved high accuracy (relative errors < 1.10%) and reduced computational time by ~78%.
• B-site local polarization was found to play a crucial role in regulating non-radiative recombination.
• Rubidium-substitution (FA_Rb) and cesium interstitial doping extended recombination times by ~2.8 times compared to the pristine system.

Conclusions:

• Local polarization at the B-site is a significant factor in suppressing non-radiative recombination in halide perovskites.
• The developed ML framework offers an accurate and efficient tool for studying complex dynamics in materials.
• These findings provide a theoretical basis for designing high-performance perovskite materials via local polarization modulation.