[Paper Review] Are Universal Potentials Ready for Alkali-Ion Battery Kinetics?

Xingyu Guo, Cheng Gui+1 more|arXiv (Cornell University)|Jan 16, 2026

Machine Learning in Materials Science0 citations

TL;DR

The paper benchmarks state-of-the-art universal machine learning interatomic potentials (uMLIPs) for alkali-ion battery kinetics, finding Orb-v3 best for static migration barriers and GRACE (OMat24-trained) best for dynamic ion transport, showing data diversity is a key driver.

ABSTRACT

Accelerating alkali-ion battery discovery requires accurate modeling of atomic-scale kinetics, yet the reliability of universal machine learning interatomic potentials (uMLIPs) in capturing these high-energy landscapes remains uncertain. Here, we systematically benchmark state-of-the-art uMLIPs, including M3GNet, CHGNet, MACE, SevenNet, GRACE, and Orb, against DFT baselines for cathodes and solid electrolytes. We find that the Orb-v3 family excels in static migration barrier predictions (MAE $\approx$ 75--111 meV), driven primarily by architectural refinements. Conversely, for dynamic transport, the GRACE model trained on the OMat24 dataset demonstrates superior fidelity in reproducing ion diffusivities and structural correlations. Our results reveal that while architectural sophistication (e.g., equivariance) is beneficial, the inclusion of high-temperature, non-equilibrium training data is the dominant driver of kinetic accuracy. These findings establish that modern uMLIPs are sufficiently robust to serve as zero-shot surrogates for high-throughput kinetic screening of next-generation energy storage materials.

Motivation & Objective

• Assess the reliability of state-of-the-art uMLIPs for kinetic properties in alkali-ion battery materials.
• Benchmark migration barriers against DFT using NEB across diverse cathodes and solid electrolytes.
• Identify how architecture vs. training data impacts kinetic accuracy and robustness in high-energy regions.

Proposed method

• Evaluate a suite of uMLIPs (M3GNet, CHGNet, MACE, SevenNet, GRACE, Orb) against DFT baselines for Li/Na migration barriers.
• Use CI-NEB to compute migration barriers for symmetric Li/Na pathways in layered, olivine, and maricite cathodes; and for solid electrolytes.
• Perform MD simulations (NVT, 800 K and 1200 K) to compare MSDs, diffusivities, and RDFs against AIMD benchmarks.
• Train models on diverse datasets (MPF, MPTrj, MPA, Alexandria, sAlex, OMat24) to assess data diversity effects on kinetic predictions.
• Assess robustness of NEB results by counting failures (Type I divergences, Type II unphysical barriers) across models.

Figure 1: Overview of the uMLIP development ecosystem. The schematic illustrates the progression from training datasets, categorized by scale and thermodynamic state, through a range of training strategies to distinct model architectures. Architectures are grouped by their geometric symmetry propert

Experimental results

Research questions

• RQ1 Can current uMLIPs reproduce DFT-calibrated migration barriers for Li/Na in representative cathodes and solid electrolytes?
• RQ2 How do architectural features (invariant vs equivariant, direct vs conservative) and training data composition affect kinetic predictions (barriers and diffusivities)?
• RQ3 Is high-temperature, non-equilibrium training data essential for capturing dynamic ion transport compared to static barrier accuracy?
• RQ4 Do uMLIPs provide robust zero-shot surrogates for high-throughput kinetic screening without material-specific fine-tuning?

Key findings

• Orb-v3 achieves the lowest static migration barrier MAEs (75–111 meV) among tested models, powered mainly by architectural refinements.
• For dynamic transport, GRACE trained on OMat24 yields the best MD fidelity, with MSD and diffusivity errors notably reduced (e.g., GRACE OA) compared to others.
• Models trained on diverse, non-equilibrium data (OMat24, MPA) outperform MPF/MPTrj-only variants in barrier predictions; adding high-temperature data reduces MAE substantially for several models.
• MatterSim with active learning achieves strong barrier and diffusion predictions, illustrating data scale and diversity can surpass architectural differences in some cases.
• For MD properties, equivariant architectures and diverse training data dramatically improve predictions of MSD, diffusivity, and RDFs, but ranking (R^2) for diffusivity remains challenging across models.
• A key insight is the robustness of barrier and diffusion predictions despite force deviations at transition states, suggesting energy-topology accuracy can suffice for kinetic screening.

Figure 2: Number of NEB calculation failures across various uMLIP models. Failures are categorized as (top) Type I: computational divergence and (bottom) Type II: unphysical barriers (where $E_{\mathrm{TS}}<E_{\mathrm{initial/final}}$ ).