Advancing Nonadiabatic Molecular Dynamics Simulations for Solids: Achieving Supreme Accuracy and Efficiency with Machine Learning

Changwei Zhang,Yang Zhong,Zhi-Guo Tao,Xinming Qing,Honghui Shang,Zhenggang Lan,Oleg V. Prezhdo,Xin-Gao Gong,Weibin Chu,Hongjun Xiang

2024-08-13

Abstract:Non-adiabatic molecular dynamics (NAMD) simulations have become an indispensable tool for investigating excited-state dynamics in solids. In this work, we propose a general framework, N$^2$AMD which employs an E(3)-equivariant deep neural Hamiltonian to boost the accuracy and efficiency of NAMD simulations. The preservation of Euclidean symmetry of Hamiltonian enables N$^2$AMD to achieve state-of-the-art performance. Distinct from conventional machine learning methods that predict key quantities in NAMD, N$^2$AMD computes these quantities directly with a deep neural Hamiltonian, ensuring supreme accuracy, efficiency, and consistency. Furthermore, N$^2$AMD demonstrates excellent generalizability and enables seamless integration with advanced NAMD techniques and infrastructures. Taking several extensively investigated semiconductors as the prototypical system, we successfully simulate carrier recombination in both pristine and defective systems at large scales where conventional NAMD often significantly underestimates or even qualitatively incorrectly predicts lifetimes. This framework not only boosts the efficiency and precision of NAMD simulations but also opens new avenues to advance materials research.

Computational Physics,Materials Science

What problem does this paper attempt to address?

### Problems the Paper Aims to Solve

This paper aims to address the inefficiencies and inaccuracies encountered when applying non-adiabatic molecular dynamics (NAMD) simulations to solid materials. Specifically:

1. **High computational cost**: Traditional NAMD simulations are computationally intensive, especially for solid-state materials, requiring repeated solutions of electronic structure equations. This makes the use of hybrid functionals for NAMD simulations impractical.
2. **Accuracy dependence on exchange-correlation functionals**: The accuracy of NAMD simulations heavily depends on the chosen exchange-correlation functional. Commonly used local density approximation (LDA) or generalized gradient approximation (GGA) often lead to severe self-interaction errors, resulting in qualitatively incorrect outcomes.
3. **Limitations of existing machine learning methods**: Although many machine learning-based methods exist to accelerate NAMD simulations, their accuracy and transferability in excited-state dynamics are still far below their performance in ground-state dynamics, especially in solid-state systems.

### Solution

To address the above issues, the authors propose a new framework—N2AMD (Neural-Network Non-Adiabatic Molecular Dynamics), which utilizes E(3)-equivariant graph neural networks (GNN) to efficiently and accurately compute the key quantities required for NAMD. Specific features include:

1. **E(3)-equivariant deep neural Hamiltonian**: By preserving the Euclidean symmetry of the Hamiltonian, N2AMD achieves state-of-the-art performance.
2. **Direct computation of key quantities**: Unlike traditional machine learning methods, N2AMD directly uses the deep neural Hamiltonian to compute the key quantities required for NAMD, ensuring high accuracy, efficiency, and consistency.
3. **Excellent generalization ability**: N2AMD not only performs well within the training set but also seamlessly integrates advanced NAMD techniques and infrastructure, making it suitable for various materials and complex systems.
4. **Large-scale simulations**: N2AMD significantly reduces computational costs, making large-scale NAMD simulations at the hybrid functional level feasible.

### Application Examples

The authors validate the effectiveness of N2AMD through several extensively studied semiconductor materials (such as titanium dioxide, gallium arsenide, molybdenum disulfide, and silicon). The results show that N2AMD can accurately predict carrier recombination processes in both defect-free and defective systems, whereas traditional NAMD simulations often severely underestimate or qualitatively mispredict lifetimes. Additionally, N2AMD excels in predicting non-adiabatic coupling vectors, further advancing the application of NAMD in solid-state materials.

In summary, by proposing the N2AMD framework, this paper not only improves the efficiency and accuracy of NAMD simulations but also opens new avenues for materials science research.