Abstract:The application of first-principles calculations for predicting lattice thermal conductivity (LTC) in crystalline materials, in conjunction with the linearized phonon Boltzmann equation, has gained increasing popularity. In this calculation, the determination of force constants through first-principles calculations is critical for accurate LTC predictions. For material exploration, performing first-principles LTC calculations in a high-throughput manner is now expected, although it requires significant computational resources. To reduce computational demands, we integrated polynomial machine learning potentials on-the-fly during the first-principles LTC calculations. This paper presents a systematic approach to first-principles LTC calculations. We designed and optimized an efficient workflow that integrates multiple modular software packages. We applied this approach to calculate LTCs for 103 compounds of the wurtzite, zincblende, and rocksalt types to evaluate the performance of the polynomial machine learning potentials in LTC calculations. We demonstrate a significant reduction in the computational resources required for the LTC predictions.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to reduce the computational requirements of first - principles calculations in lattice thermal conductivity (LTC) calculations in order to achieve high - throughput LTC calculations. Specifically, the authors propose a method of integrating polynomial machine - learning potentials (MLP) during the first - principles LTC calculation process to significantly reduce the need for computational resources. Through this method, the LTC values of materials can be predicted using fewer computational resources, which is of great significance for material exploration and discovery. ### Main research content 1. **Method introduction**: - **First - principles calculations and the linearized phonon Boltzmann equation**: Currently, LTC calculations based on first - principles calculations and the linearized phonon Boltzmann equation have become increasingly popular. However, these calculations still require a large amount of computational resources. - **Application of polynomial machine - learning potentials (MLP)**: To reduce the computational requirements, the authors introduce polynomial MLP and integrate it into the intermediate stage of LTC calculations. These MLPs are trained with a small data set and then generate a large displacement - force data set for calculating supercell force constants. 2. **Calculation process**: - **Data set preparation**: First, generate a set of supercells containing random atomic displacements, and then use first - principles calculations to calculate the energies and forces of these supercells to form an initial data set. - **MLP training**: Use the above - mentioned data set to train the polynomial MLP. - **Force calculation**: Generate another set of supercells containing random atomic displacements, and use the trained MLP to calculate the atomic forces in these supercells. - **Supercell force constant calculation**: Calculate the supercell force constants from the displacement - force data set generated by the MLP. - **LTC calculation**: Finally, use the calculated supercell force constants to calculate the LTC value. 3. **Performance evaluation**: - The authors performed LTC calculations on 103 compounds (including hexagonal nitrides, zinc blende, and rock - salt - type structures) to evaluate the performance of polynomial MLP in LTC calculations. - The results show that a data set of 20 supercells can be used to estimate the LTC value relatively well, especially for rock - salt - type compounds. As the data set increases, the convergence of the LTC value is further improved. ### Key formulas - **Taylor expansion of potential energy**: \[ V=\Phi_0+\sum_{l\kappa\alpha}\Phi_{l\kappa\alpha}u_{l\kappa\alpha}+\frac{1}{2}\sum_{l\kappa\alpha,l'\kappa'\alpha'}\Phi_{l\kappa\alpha,l'\kappa'\alpha'}u_{l\kappa\alpha}u_{l'\kappa'\alpha'}+\frac{1}{3!}\sum_{l\kappa\alpha,l'\kappa'\alpha',l''\kappa''\alpha''}\Phi_{l\kappa\alpha,l'\kappa'\alpha',l''\kappa''\alpha''}u_{l\kappa\alpha}u_{l'\kappa'\alpha'}u_{l''\kappa''\alpha''}+\cdots \] where \(l\), \(\kappa\) and \(\alpha\) represent the unit cell, atomic index and Cartesian coordinate respectively. - **Expression of force**: \[ -f_{l\kappa\alpha}=\Phi_{l\kappa\alpha}+\sum_{l'\kappa'\alpha'}\Phi_{l\kappa\alpha,l'\kappa'\alpha'}u_{l'\kappa'\alpha'}+\frac{1}{2}\sum_{l'\kappa'\alpha',l''\kappa''\alpha''}\Phi_{l\kappa\alpha,l'

On-the-fly training of polynomial machine learning potentials in computing lattice thermal conductivity

Lattice Dynamics and Thermal Conductivity of Lithium Fluoride Via First-Principles Calculations

Efficient Calculation of the Lattice Thermal Conductivity by Atomistic Simulations with Ab Initio Accuracy

Lattice Thermal Conductivity: An Accelerated Discovery Guided by Machine Learning

A High-Throughput Framework for Lattice Dynamics

High-throughput computational discovery of 3218 ultralow thermal conductivity and dynamically stable materials by dual machine learning models

Lattice thermal conductivity of solid LiF based on machine learning force fields and the Green–Kubo approach

Lattice dynamics simulation using machine learning interatomic potentials

Discovery of low thermal conductivity compounds with first-principles anharmonic lattice dynamics calculations and Bayesian optimization

Correcting force error-induced underestimation of lattice thermal conductivity in machine learning molecular dynamics

Machine-Learning Interatomic Potentials Enable First-Principles Multiscale Modeling of Lattice Thermal Conductivity in Graphene/Borophene Heterostructures

Accelerated molecular dynamics force evaluation on graphics processing units for thermal conductivity calculations.

Fast and accurate machine learning prediction of phonon scattering rates and lattice thermal conductivity

Accelerating Phonon Thermal Conductivity Prediction by an Order of Magnitude Through Machine Learning-Assisted Extraction of Anharmonic Force Constants

Predicting lattice thermal conductivity from fundamental material properties using machine learning techniques

GPU_PBTE: an Efficient Solver for Three and Four Phonon Scattering Rates on Graphics Processing Units

Lattice Anharmonicity And Thermal Conductivity From Compressive Sensing Of First-Principles Calculations

Machine Learning Models for the Lattice Thermal Conductivity Prediction of Inorganic Materials

Accelerating first-principles estimation of thermal conductivity by machine-learning interatomic potentials: A MTP/ShengBTE solution

Accelerating Evaluation of Converged Lattice Thermal Conductivity