Integrating machine learning interatomic potentials with batched optimization for crystal structure prediction

Chengxi Zhao, Zhaojia Ma, Dingrui Fan, Siyu Hu, Leping Wang, Weile Jia, En Shao, Guangming Tan, Jun Jiang, Linjiang Chen

Abstract: Molecular crystal structure prediction (CSP) faces a persistent computational bottleneck: it requires exhaustive sampling of vast packing landscapes while resolving energy differences of only a few kJ·mol-1. We introduce BOMLIP-CSP, an open-source Python framework that integrates machine learning interatomic potentials (MLIPs) with a tailored batched optimization strategy, enabling rapid, unbiased structure prediction across the full crystal density range. By introducing tailored parallelism into modern MLIPs, BOMLIP-CSP achieves a ~2.1–2.3× acceleration in large-scale CSP searches without compromising accuracy. In benchmarks covering 34 experimental structures from six CSP blind tests, over 50% of experimental crystals are recovered with foundational MLIPs (namely, MACE-OFF-small and SevenNet-0-D3), rising above 70% with judicious MLIP selection. Importantly, we show that MLIPs with comparable equilibrium energy accuracy can yield strikingly different CSP outcomes, underscoring that not only local energy fidelity but also the global topology of the crystal lattice energy landscape governs predictive success. Together, these results establish BOMLIP-CSP as a broadly accessible platform for accelerated CSP and provide new insight into the interplay between MLIP characteristics and crystal structure discovery.

External IDs:doi:10.26434/chemrxiv-2025-fqs7n