How AI‑Powered FastTrack Accelerates Ion Diffusion Modeling by Tenfold

Background

Ion diffusion governs the rate at which charge carriers move through solid‑state energy‑storage materials. The migration barrier—the energy required for an ion to hop between adjacent lattice sites—directly influences power density, cycle life, and safety of batteries.

Limitations of Conventional DFT‑NEB Calculations

Standard workflows combine density‑functional theory (DFT) with the nudged elastic band (NEB) method to locate transition states. A single diffusion pathway can require several hours to days of CPU time, making high‑throughput screening of hundreds of compounds impractical.

FastTrack Framework

FastTrack is a machine‑learning‑augmented pipeline that replaces the expensive DFT‑NEB step with a three‑dimensional potential‑energy‑surface (PES) reconstruction based on a machine‑learning force field (MLFF). By first training an MLFF on a modest DFT dataset, the method can generate the full PES of a crystal within minutes and automatically extract the lowest‑energy migration routes.

Workflow

DFT data generation : Compute energies and forces for a representative set of atomic configurations (typically 100–500 structures) using a chosen exchange‑correlation functional (e.g., PBE or PBE+U).

MLFF training : Fit an MLFF (e.g., GPT‑FF, CHGNet, or MACE) to the DFT dataset. The training script accepts standard ase or dpdata formats and outputs a model file ( .pth or .ckpt).

PES sampling : The trained MLFF evaluates the energy of the ion at a dense grid (e.g., 0.2 Å spacing) throughout the unit cell, producing a 3D energy map.

Interpolation & path finding : FastTrack interpolates the grid to a continuous surface and applies a Dijkstra‑type algorithm to locate the minimum‑energy path between symmetry‑equivalent sites without requiring user‑specified intermediate images.

Barrier extraction : The maximum energy along the identified path minus the energy of the initial site yields the migration barrier.

Visualization : The package FastTrace visualizes the PES iso‑surfaces and the migration trajectory for qualitative analysis.

Performance and Benchmark Results

FastTrack reproduces DFT‑level barriers with a typical error < 50 meV while delivering a speedup of roughly tenfold compared with direct NEB calculations. Representative case studies:

Layered LiCoO₂ : Single‑vacancy diffusion barrier ≈ 600 meV; double‑vacancy barrier ≈ 250 meV, matching previously reported values.

Olivine LiFePO₄ : One‑dimensional channel along [010] with a barrier ≈ 300 meV, reflecting the rigidity of the phosphate framework.

Model Evaluation and Transferability

FastTrack is force‑field agnostic. The authors benchmarked three state‑of‑the‑art MLFFs:

GPT‑FF

CHGNet

MACE

All three delivered consistent barrier predictions across diverse chemistries. Fine‑tuning the MLFF on material‑specific PBE or PBE+U datasets further reduced prediction errors, underscoring the importance of high‑quality training data.

Software Availability

The open‑source implementation, named FastTrace , is hosted on GitHub (https://github.com/fasttrack‑mlff/fasttrace). It provides command‑line tools for:

Training MLFFs from DFT datasets

Generating 3D PES grids

Automatic path finding and barrier extraction

Visualization of iso‑energy surfaces and migration pathways

Installation follows standard pip install fasttrace procedures, and example input files are included in the examples/ directory.

Implications

By decoupling accurate barrier estimation from expensive DFT‑NEB calculations, FastTrack enables high‑throughput screening of ion‑conducting materials, facilitates systematic exploration of defect chemistry, and opens the door to AI‑augmented computational workflows where MLFF‑driven PES mapping becomes a routine step in materials design.

Reference: https://iopscience.iop.org/article/10.1088/3050-287X/ae0808