InterOptimus: An AI-assisted robust workflow for screening ground-state heterogeneous interface structures in lithium batteries

The formation of interphase layers, including the cathode-electrolyte interphase (CEI) and solid-electrolyte interphase (SEI), exhibits significant chemical complexity and plays a pivotal role in determining the performance of lithium batteries. Despite considerable advances in simulating the bulk phase properties of battery materials, the understanding of interfaces, including crystalline interfaces that represent the simplest case, remains limited. This is primarily due to challenges in performing ground-state searches for interface microstructures and the high computational costs associated with first-principles methods. Herein, we introduce InterOptimus, an automated workflow designed to efficiently search for ground-state heterogeneous interfaces. InterOptimus incorporates a rigorous, symmetry-aware equivalence analysis for lattice matching and termination scanning. Additionally, it introduces stereographic projection as an intuitive and comprehensive framework for visualizing and classifying interface structures. By integrating universal machine learning interatomic potentials (MLIPs), InterOptimus enables rapid predictions of interface energy and stability, significantly reducing the necessary computational cost in density functional theory (DFT) by over 90%. We benchmarked several MLIPs at three critical lithium battery interfaces, Li₂S|Ni₃S₂, LiF|NCM, and Li₃PS₄|Li, and demonstrated that the MLIPs achieve accuracy comparable to DFT in modeling potential energy surfaces and ranking interface stabilities. Thus, InterOptimus facilitates the efficient determination of ground-state heterogeneous interface structures and subsequent studies of structure-property relationships, accelerating the interface engineering of novel battery materials.