Transferable empirical pseudopotenials from machine learning

Rokyeon Kim,Young-Woo Son

DOI: https://doi.org/10.1103/PhysRevB.109.045153

2024-02-07

Abstract:Machine learning is used to generate empirical pseudopotentials that characterize the local screened interactions in the Kohn-Sham Hamiltonian. Our approach incorporates momentum-range-separated rotation-covariant descriptors to capture crystal symmetries as well as crucial directional information of bonds, thus realizing accurate descriptions of anisotropic solids. Trained empirical potentials are shown to be versatile and transferable such that the calculated energy bands and wave functions without cumbersome self-consistency reproduce conventional ab initio results even for semiconductors with defects, thus fostering faster and faithful data-driven materials researches.

Materials Science,Computational Physics

What problem does this paper attempt to address?

### Problems the Paper Aims to Solve This paper aims to generate empirical pseudopotentials that describe the local screening interactions in solids through machine learning, enabling fast and accurate electronic structure calculations. Specifically: 1. **Improving Computational Efficiency**: - Existing Density Functional Theory (DFT) methods, while powerful, are computationally expensive when dealing with large-scale material systems. The method proposed in this paper generates empirical pseudopotentials using machine learning techniques, significantly speeding up electronic structure calculations without sacrificing accuracy. 2. **Addressing Limitations of Traditional Empirical Pseudopotential Methods**: - Traditional Empirical Pseudopotential Methods (EPM) are efficient but lack accuracy in wavefunctions and transferability of pseudopotentials. The proposed method introduces rotationally covariant descriptors to capture crystal symmetry and bond direction information, thereby improving the accuracy and transferability of pseudopotentials. 3. **Applicability to Various Material Systems**: - This method can accurately describe defect-free materials and is also well-suited for semiconductor materials with defects. The trained empirical pseudopotentials can reproduce traditional ab initio results, including quantum properties such as energy bands and wavefunctions. 4. **Extension to Non-local Correlation Effects**: - Beyond basic pseudopotential calculations, the paper also demonstrates how to incorporate non-local correlation effects within the existing framework, further enhancing the accuracy of material property predictions. In summary, this paper addresses the high computational resource consumption of traditional DFT calculations and improves the accuracy and applicability of empirical pseudopotential methods through machine learning, providing a reliable platform for data-driven materials research.

Transferable empirical pseudopotenials from machine learning

By-passing the Kohn-Sham equations with machine learning

Self-consistent Coulomb interactions for machine learning interatomic potentials

Transferable screened range-separated hybrid functionals for electronic and optical properties of van der Waals materials

A Fourth-Generation High-Dimensional Neural Network Potential with Accurate Electrostatics Including Non-local Charge Transfer

Active and Transfer Learning of High-Dimensional Neural Network Potentials for Transition Metals

Training Transferable Interatomic Neural Network Potentials for Reactive Chemistry: Improved Chemical Space Sampling

Phase transitions of hybrid perovskites simulated by machine-learning force fields trained on-the-fly with Bayesian inference

Parametrization protocol and refinement strategies for accurate and transferable analytic bond-order potentials: Application to Re

Developments and Further Applications of Ephemeral Data Derived Potentials

Transferable Water Potentials Using Equivariant Neural Networks

End-to-end Symmetry Preserving Inter-atomic Potential Energy Model for Finite and Extended Systems

Transfer-Learned Potential Energy Surfaces: Towards Microsecond-Scale Molecular Dynamics Simulations in the Gas Phase at CCSD(T) Quality

Transferable Performance of Machine Learning Potentials Across Graphene-Water Systems of Different Sizes: Insights from Numerical Metrics and Physical Characteristics.

Prediction of many-electron wavefunctions using atomic potentials: refinements and extensions to transition metals and large systems

An Accurate and Transferable Machine-Learning Interatomic Potential for Silicon

Evaluating the Transferability of Machine-Learned Force Fields for Material Property Modeling

Training Data Selection for Accuracy and Transferability of Interatomic Potentials

De novo exploration and self-guided learning of potential-energy surfaces

Universal Machine Learning Kohn-Sham Hamiltonian for Materials