An official website of the United States government
Machine learning interatomic potentials (MLPs) are promising for accelerating the simulation of ion transport in all‐solid‐state battery materials, but their accuracy across diverse material compositions and symmetries remains unquantified. Here, we systematically benchmark six state‐of‐the‐art MLPs, namely CHGNet, EquiformerV2 in two training variants, MatterSim, SevenNet, and MACE, on representative Li‐ and Na‐based superionic conductors. By comparing predicted atomic forces, diffusion coefficients (DCs) from MLP‐driven molecular dynamics simulations, and second‐ and third‐order interatomic force constants (IFCs) against density functional theory (DFT) and ab initio molecular dynamics (AIMD), we assess the predictive reliability of each model across these properties. EquiformerV2 models, especially the variant trained only on the OMAT dataset, exhibit the lowest force prediction errors and yield DCs in the closest agreement with AIMD. Continually, MLPs with more accurate force predictions produce more reliable diffusion metrics. However, while second‐order IFCs are reasonably captured, all models struggle to reproduce third‐order (anharmonic) IFCs, highlighting significant challenges in modeling anharmonic lattice dynamics with current MLPs. This benchmark study highlights the importance of accurately capturing atomic forces in reasonably producing ion transport properties in all‐solid‐state battery materials and provides practical guidance for universal MLP selection for ion transport simulations and future improvement and development.
more »
« less
This content will become publicly available on October 15, 2026
A Comprehensive Assessment and Benchmark Study of Large Atomistic Foundation Models for Phonons
The rapid development of universal machine learning potentials (uMLPs) has enabled efficient, accurate predictions of diverse material properties across broad chemical spaces. While their capability for modeling phonon properties is emerging, systematic benchmarking across chemically diverse systems remains limited. We evaluate six recent uMLPs—EquiformerV2, MatterSim, MACE, and CHGNet—on 2429 crystalline materials from the Open Quantum Materials Database. Models were used to compute atomic forces in displaced supercells, derive interatomic force constants (IFCs), and predict phonon properties including lattice thermal conductivity (LTC), compared with density functional theory and experimental data. The EquiformerV2 pretrained model trained on the OMat24 dataset exhibits strong performance in predicting atomic forces and third‐order IFCs, while its fine‐tuned counterpart consistently outperforms other models in predicting second‐order IFCs, LTC, and other phonon properties. Although MACE and CHGNet demonstrated comparable force prediction accuracy to EquiformerV2, notable discrepancies in IFC fitting led to poor LTC predictions. Conversely, MatterSim, despite lower force accuracy, achieved intermediate IFC predictions, suggesting error cancellation and complex relationships between force accuracy and phonon predictions. This benchmark guides the evaluation and selection of uMLPs for high‐throughput screening of materials with targeted thermal transport properties.
more »
« less
- PAR ID:
- 10655943
- Publisher / Repository:
- Wiley
- Date Published:
- Journal Name:
- Advanced Intelligent Discovery
- ISSN:
- 2943-9981
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Abstract Metal–organic frameworks (MOFs) are highly porous and versatile materials studied extensively for applications such as carbon capture and water harvesting. However, computing phonon-mediated properties in MOFs, like thermal expansion and mechanical stability, remains challenging due to the large number of atoms per unit cell, making traditional Density Functional Theory (DFT) methods impractical for high-throughput screening. Recent advances in machine learning potentials have led to foundation atomistic models, such as MACE-MP-0, that accurately predict equilibrium structures but struggle with phonon properties of MOFs. In this work, we developed a workflow for computing phonons in MOFs within the quasi-harmonic approximation with a fine-tuned MACE model, MACE-MP-MOF0. The model was trained on a curated dataset of 127 representative and diverse MOFs. The fine-tuned MACE-MP-MOF0 improves the accuracy of phonon density of states and corrects the imaginary phonon modes of MACE-MP-0, enabling high-throughput phonon calculations with state-of-the-art precision. The model successfully predicts thermal expansion and bulk moduli in agreement with DFT and experimental data for several well-known MOFs. These results highlight the potential of MACE-MP-MOF0 in guiding MOF design for applications in energy storage and thermoelectrics.more » « less
-
In this Letter, we report that the fourth-order interatomic force constants (4th-IFCs) are significantly sensitive to the energy surface roughness of exchange-correlation (XC) functionals in density functional theory calculations. This sensitivity, which is insignificant for the second- (2nd-) and third-order (3rd-) IFCs, varies for different functionals in different materials and can cause misprediction of thermal conductivity by several times of magnitude. As a result, when calculating the 4th-IFCs using the finite difference method, the atomic displacement needs to be taken large enough to overcome the energy surface roughness, in order to accurately predict phonon lifetime and thermal conductivity. We demonstrate this phenomenon on a benchmark material (Si), a high-thermal conductivity material (BAs), and a low thermal conductivity material (NaCl). For Si, we find that the LDA, PBE, and PBEsol XC functionals are all smooth to the 2nd- and 3rd-IFCs but all rough to the 4th-IFCs. This roughness can lead to a prediction of nearly one order of magnitude lower thermal conductivity. For BAs, all three functionals are smooth to the 2nd- and 3rd-IFCs, and only the PBEsol XC functional is rough for the 4th-IFCs, which leads to a 40% underestimation of thermal conductivity. For NaCl, all functionals are smooth to the 2nd- and 3rd-IFCs but rough to the 4th-IFCs, leading to a 70% underprediction of thermal conductivity at room temperature. With these observations, we provide general guidance on the calculation of 4th-IFCs for an accurate thermal conductivity prediction.more » « less
-
ABSTRACT Double perovskites (ABC2D6) are versatile materials with applications in photovoltaics, optoelectronics, and thermoelectrics, where phonon‐mediated thermal transport is critical. However, high‐throughput phonon calculations by density functional theory (DFT) are computationally prohibitive due to the large supercells required. We develop a deep learning interatomic potential, Elemental‐SDNNFF, trained directly on DFT‐calculated forces within an active learning framework, enabling efficient prediction of phonon properties across thousands of double perovskites. Using this model, we screened 9709 cubic double perovskite structures, identifying 1597 dynamically stable candidates. Their lattice thermal conductivities (LTCs) were predicted by coupling Elemental‐SDNNFF with the Boltzmann Transport Equation, including off‐diagonal contributions. For the most promising compounds, DFT validation and four‐phonon scattering calculations revealed ultralow LTCs (<0.1 Wm−1K−1). Remarkably, Cs2HgPtCl6was found to possess a bandgap of 0.35 eV and an LTC of 0.071 Wm−1K−1at room temperature—the lowest ever reported for isotropic bulk materials, comparable to air. The result was independently confirmed by molecular dynamics simulations with a DeePMD potential and phonon lifetime extraction using DynaPhoPy. This work establishes an efficient machine learning‐assisted framework for fast screening of dynamic stability and accurate prediction of phonon transport in complex materials, highlighting double perovskites as Double Perovskites, High‐Throughput DFT, Machine Learning, Phonon Boltzmann Transport, Record‐Low Lattice Thermal Conductivity promising candidates for thermoelectric and thermal insulation applications.more » « less
-
Phonons, quantized vibrations of the atomic lattice, are central to thermal transport, structural stability, and phase behavior in crystalline solids. However, most large‐scale materials databases rely on the harmonic approximation and neglect important temperature‐dependent anharmonic effects. Here, we present a scalable framework combining machine learning interatomic potentials, anharmonic lattice dynamics, and high‐throughput calculations to predict finite‐temperature phonons across thousands of materials. By fine‐tuning the universal M3GNet potential with high‐quality phonon data, we improve phonon prediction accuracy fourfold while retaining computational efficiency. We integrate this refined model with a high‐throughput implementation of the stochastic self‐consistent harmonic approximation to compute temperature‐dependent phonons for 4669 inorganic compounds. The resulting dataset reveals systematic elemental and structural trends in anharmonic phonon renormalization, especially in alkali metals, perovskite‐derived frameworks, and related systems. Machine learning analysis identifies weak bonding, large atomic radii, and specific coordination motifs as key drivers of strong anharmonicity. First‐principles validation further shows that anharmonic effects can change lattice thermal conductivity by factors of two to four. This work provides an efficient data‐driven platform for predicting finite‐temperature phonon behavior and guiding the discovery of materials with tailored thermal and vibrational properties.more » « less
