Screening novel cathode materials from the Energy-GNoME database using foundation machine learning force fields and DFT

Abstract

The development of new cathodes is essential for enabling next generation energy storage technologies. Here, we introduce a multi-fidelity workflow that bridges AI-based candidate identification and density functional theory refinement for accelerated discovery of cathode materials. Starting from the Energy-GNoME database, we use foundational machine-learning force fields (MLFFs) as an intermediate-fidelity layer to screen Li-, Na-, K-, Mg-, and Ca-ion intercalation cathodes. We assess dynamical stability and simulate intercalation voltage profiles to determine the average operating voltage and specific energy. The surviving candidates are further prioritized using experimentally motivated filters based on space-group prevalence, elemental availability, and cost, followed by DFT+U refinement of the most promising structures. Finally, combined MLFF and DFT nudged elastic band calculations are used to assess ion-migration kinetics and identify candidates with the potential for good rate capability. The workflow reduces 615 high-confidence Energy-GNoME cathodes to 10 prioritized candidates with favorable electrochemical properties and manageable volume changes. In addition to delivering a shortlist of promising candidates, this work benchmarks the consistency of Energy-GNoME, MACE, and DFT predictions and demonstrates how MLFFs can transform AI-prioritized materials databases into physics-grounded discovery pipelines.