Evaluation of Foundational Machine Learned Interatomic Potentials for Migration Barrier Predictions

Achinthya Krishna Bheemaguli, Penghao Xiao, Gopalakrishnan Sai Gautam · 2025

Read Evaluation of Foundational Machine Learned Interatomic Potentials for Migration Barrier Predictions by Achinthya Krishna Bheemaguli, Penghao Xiao, Gopalakrishnan Sai Gautam (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2512-03642.txt.

Introduction

Achinthya Bheemaguli, Penghao Xiao, and Gopalakrishnan Gautam benchmark five foundation machine-learned potentials at predicting how hard ions hop through battery materials. That hop barrier decides charging speed, and computing it properly with quantum methods is slow. Their 2025 study races the learned models against density functional references.

MACE-MP-0 and Orb-v3 posted the lowest barrier errors across diverse battery chemistries. Two models sorted good from bad ionic conductors correctly more than four times in five. Their intermediate pathway snapshots also beat standard interpolation as starting guesses for the expensive quantum refinement in most structures. Fast screening of candidate battery materials is the intended payoff.

Catalog record

CREATORAchinthya Krishna Bheemaguli, Penghao Xiao, Gopalakrishnan Sai Gautam
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE42 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2512-03642

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Evaluation of Foundational Machine Learned Interatomic Potentials for Migration Barrier Predictions Achinthya Krishna Bheemaguli, Penghao Xiao, Gopalakrishnan Sai Gautam Abstract Fast, and accurate prediction of ionic migration barriers ( E m E_{m} ) is crucial for designing next-generation battery materials that combine high energy density with facile ion transport. Given the computational costs associated with estimating E m E_{m} using conventional density functional theory (DFT) based nudged elastic band (NEB) calculations, we benchmark the accuracy in E m E_{m} and geometry predictions of five foundational machine learned interatomic potentials (MLIPs), which can potentially accelerate predictions of ionic transport. Specifically, we assess the accuracy of MACE-MP-0, Orb-v3, SevenNet, CHGNet, and M3GNet models, coupled with the NEB framework, against DFT-NEB-calculated E m E_{m} across a diverse set of battery-relevant chemistries and structures. Notably, MACE-MP-0 and Orb-v3 exhibit the lowest mean absolute errors in E m E_{m} predictions across the entire dataset and over data points that are not outliers, respectively. Importantly, Orb-v3 and SevenNet classify ‘good’ versus ‘bad’ ionic conductors with an accuracy of > > 82%, based on a threshold E m E_{m} of 500 meV, indicating their utility in high-throughput screening approaches. Notably, intermediate images generated by MACE-MP-0 and SevenNet provide better initial guesses relative to conventional interpolation techniques in > > 71% of structures, offering a practical route to accelerate subsequent DFT-NEB relaxations. Finally, we observe that accurate E m E_{m} predictions by MLIPs are not correlated with accurate (local) geometry predictions. Our work establishes the use-cases, accuracies, and limitations of foundational MLIPs in estimating E m E_{m} and should serve as a base for accelerating the discovery of novel ionic conductors for batteries and beyond. 1 Introduction Developing next-generation batteries is essential for our transition into sustainable energy usage, given that the state-of-the-art lithium-ion batteries (LIBs), while already delivering excellent performance, are approaching their fundamental limits [ 1 , 2 ] , necessitating the discovery of novel beyond-LIB materials and chemistries. One key materials property that governs battery performance is the ionic diffusivity ( D ) of the electroactive ion in electrodes and (solid) electrolytes. D is exponentially influenced by the energy barrier each ion must overcome, commonly referred to as the migration barrier ( E_{m} ), to hop from its initial lattice site to a symmetrically equivalent final site [ 3 , 4 , 5 ] . Each ionic hop is often mediated through vacancies in the lattice with the ion overcoming transition state(s) along the hop. D and E_{m} are thus related via the Arrhenius expression, D=D_{0}\exp(-E_{m}/{k_{B}T}) [ 6 ] . Accordingly, materials with a low E_{m} – in both electrodes and (solid) electrolytes exhibit higher ionic conductivity and enable faster charge/discharge rates [ 7 ] . In particular, emerging multivalent battery chemistries, such as Mg or Ca-based systems that promise higher volumetric energy densities [ 8 ] , often suffer from poor rate performance [ 9 , 10 , 11 , 12 , 13 , 14 ] . Some Na-ion cathodes such as marcite-Na(Mn/Fe)PO 4 , phosphate alluaudite-Na x MnFe 2 (PO 4 ) 3 and sulfate sodium superionic conductors (NaSICONs) that offer lower costs compared to LIB cathodes also suffer from poor rate performance [ 15 , 16 ] . Therefore, understanding and minimizing the E_{m} in candidate materials is crucial for advancing the next-generation of high-performance batteries. Experimental techniques such as quasi-elastic neutron scattering [ 17 ] , electrochemical impedance spectroscopy [ 18 ] , nuclear magnetic resonance measurements [ 19 ] , and galvanostatic intermittent titration techniques [ 20 ] are commonly employed to study ion dynamics in solids [ 21 ] . However, these methods often require access to large-scale facilities, and can exhibit chemistry or material-specific constraints/requirements, limiting their accessibility. As a result, computational approaches, particularly density functional theory (DFT [ 22 , 23 ] )-based nudged elastic band (NEB [ 24 ] ) calculations are commonly used for computationally estimating E_{m} with reasonable precision [ 25 ] . While ab initio molecular dynamics (AIMD) simulations can also be used for estimating E_{m} , such simulations are computationally expensive since they require the sampling over large length and long time scales across different temperatures to provide reasonable E_{m} [ 26 , 27 ] . Moreover, AIMD simulations can be unreliable for systems exhibiting high E_{m} (i.e., the ‘true negatives’ among materials that can conduct ions) due to insufficient sampling of ion dynamics, resulting in DFT-NEB being the usual technique deployed for E_{m} predictions. Calculating E_{m} using DFT-NEB requires an initial guess for the minimum energy path (MEP), which is typically constructed by linearly interpolating the coordinates of the initial and final configurations of the moving ion. Each ‘image’ generated by linear interpolation is subsequently connected via an auxiliary spring force. Note that the initial interpolated guess is often far from the true MEP, increasing the computational expense of DFT-NEB calculations and making them prone to convergence difficulties [ 25 ] . Alternative approaches such as ‘ApproxNEB’ [ 28 ] , have been proposed to reduce computational intensity, with limited efficacy. Recently, foundational machine learned interatomic potentials (MLIPs [ 29 , 30 , 31 ] ), also referred to as universal potentials, have emerged as a new paradigm in computational materials science. The foundational MLIPs, pre-trained on large and diverse datasets, can generalize to a wide range of downstream tasks [ 31 ] and are transferable across different materials and property prediction tasks [ 32 , 33 , 34 , 35 ] , unlike classical MLIPs or force-fields that are constrained by a specific chemistry or property. Thus, foundational MLIPs are attractive candidates for accelerating atomistic simulations, including NEB calculations, by potentially improving initial MEP guesses and reducing the need for extensive DFT-based refinement or optimization, which can enable high-throughput screening of materials based on their E_{m} . Indeed, a recent work by Kang et al. [ 36 ] proposed an alternate to traditional DFT-based NEB calculations for E_{m} estimations by using an MLIP for generating the potential energy surface on a spatial grid and extracting the MEP without the need for pre-defined NEB-images. Several studies have benchmarked the performance of foundational MLIPs on diverse material properties [ 37 , 38 , 39 , 40 ] , but not on predicting E_{m} in solids. For instance, the ‘Matbench discovery’ platform [ 41 ] provides a standardized framework for ranking universal potentials, but does not yet evaluate their integration with NEB workflows for E_{m} predictions. Other MLIP benchmarking studies include the work by Zhao et al. [ 42 ] that evaluated the MLIPs on transition state search for chemical reactions involving molecules. Bihani et al. [ 43 ] benchmarked the performance of equivariant MLIPs on their generalizability to higher temperature simulations and unseen compositions, while Mannan et al. evaluated the performance of universal potentials against experimental measurements of elastic properties and structural accuracy among minerals [ 39 ] . So far, there has been no benchmark of the performance of state-of-the-art universal potentials in predicting E_{m} across a wide range of (battery) chemistries and materials, especially by integrating them with NEB workflows. Here, we assess the performance of foundational MLIPs, namely, MACE-MP-0 [ 44 , 45 ] , SevenNet [ 46 ,

Read the complete text