Abstract:Phonons play a critical role in determining various material properties, but conventional methods for phonon calculations are computationally intensive, limiting their broad applicability. In this study, we present an approach to accelerate high-throughput harmonic phonon calculations using machine learning universal potentials. We train a state-of-the-art machine learning interatomic potential, based on multi-atomic cluster expansion (MACE), on a comprehensive dataset of 2,738 crystal structures with 77 elements, totaling 15,670 supercell structures, computed using high-fidelity density functional theory (DFT) calculations. Our approach significantly reduces the number of required supercells for phonon calculations while maintaining high accuracy in predicting harmonic phonon properties across diverse materials. The trained model is validated against phonon calculations for a held-out subset of 384 materials, achieving a mean absolute error (MAE) of 0.18 THz for vibrational frequencies from full phonon dispersions, 2.19 meV/atom for Helmholtz vibrational free energies at 300K, as well as a classification accuracy of 86.2% for dynamical stability of materials. A thermodynamic analysis of polymorphic stability in 126 systems demonstrates good agreement with DFT results at 300 K and 1000 K. In addition, the diverse and extensive high-quality DFT dataset curated in this study serves as a valuable resource for researchers to train and improve other machine learning interatomic potential models.

What problem does this paper attempt to address?

The paper attempts to address the issues of high computational cost and low efficiency in traditional methods for calculating the phonon properties of materials. Specifically, traditional phonon calculation methods based on Density Functional Theory (DFT) require a large number of supercell calculations to capture short-range to long-range interactions, leading to significant consumption of computational resources and limiting the application of these methods in large-scale material screening. To solve this problem, the paper proposes a method to accelerate high-throughput harmonic phonon calculations through machine learning universal potentials. The main contributions of the paper include: 1. **Dataset Construction**: Trained an advanced machine learning interatomic potential model based on the Multi-Atom Cluster Expansion (MACE), which is based on a high-quality DFT calculation dataset containing 2,738 crystal structures, 77 elements, and a total of 15,670 supercell structures. 2. **Improved Computational Efficiency**: By randomly perturbing the positions of all atoms and using the Compressed Sensing Lattice Dynamics method to generate a small number of supercell structures, the number of required DFT self-consistent calculations is significantly reduced while maintaining high accuracy in predicting harmonic phonon properties. 3. **Model Validation**: Extensive validation of the model was conducted, including comparison of phonon calculation results for 384 materials. The results show that the model has low prediction errors in terms of vibrational frequencies, Helmholtz vibrational free energy, and the dynamical stability of materials. 4. **Thermodynamic Analysis**: Thermodynamic analysis of the polymorphic stability of 126 systems was performed, showing good consistency with DFT calculation results at 300K and 1000K. In summary, this study significantly improves the efficiency of phonon calculations through machine learning methods, providing strong support for high-throughput calculations and new material discovery in the field of materials science.

Accelerating High-Throughput Phonon Calculations via Machine Learning Universal Potentials

Machine learning a universal harmonic interatomic potential for predicting phonons in crystalline solids

Universal Machine Learning Interatomic Potentials are Ready for Phonons

Machine Learned Potential for High-Throughput Phonon Calculations of Metal-Organic Frameworks

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

Exploring Phononic Properties of Two-Dimensional Materials using Machine Learning Interatomic Potentials

Benchmarking phonon anharmonicity in machine learning interatomic potentials

Lattice dynamics simulation using machine learning interatomic potentials

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

Accelerating the calculation of electron-phonon coupling strength with machine learning

Accelerating thermal conductivity prediction through machine-learning: Two orders of magnitude reduction in phonon-phonon scattering rates calculation

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

Developing machine learning potential for classical molecular dynamics simulation with superior phonon properties

Million-scale data integrated deep neural network for phonon properties of heuslers spanning the periodic table

Accurate description of high-order phonon anharmonicity and lattice thermal conductivity from molecular dynamics simulations with machine learning potential

Virtual node graph neural network for full phonon prediction

Universal Machine Learning Kohn-Sham Hamiltonian for Materials

Understanding Phonon Transport Properties Using Classical Molecular Dynamics Simulations

High-Throughput Estimation of Phonon Thermal Conductivity from First-Principles Calculations of Elasticity.