HULU: A unified Monte Carlo framework for adsorption simulations with machine learning potentials

Adsorption in nanoporous materials is pivotal for addressing global challenges in gas storage, separation, sensing, catalysis, and atmospheric water harvesting. Consequently, molecular simulations are essential for understanding adsorption mechanisms and accelerating material discovery. Key thermodynamic descriptors, such as adsorption isotherms, density distributions, and Henry constants, are particularly valuable for high-throughput screening and predicting separation performance. Recently, machine learning potentials (MLPs) have emerged as a powerful tool, offering near-ab-initio accuracy with high computational efficiency. While MLPs have been extensively applied in molecular dynamics simulations, their integration into Monte Carlo (MC) simulations for adsorption remains largely untapped. This limitation arises primarily because mainstream MC simulation codes are designed for empirical force fields and lacks native support for MLPs. In this work, we developed a flexible Python package, HULU (High-throughput Universal Learning-enabled Utility for Adsorption), to bridge this gap. We present the first demonstration of calculating full adsorption isotherms using state-of-the-art foundation MLPs (MACE-MATPES-PBE-0, NEP89, and ORB v3). Furthermore, we systematically benchmark these models against standard baselines, such as available experimental or density-functional-theory calculation data, and elucidate the microscopic origins of deviations in the simulation results. Ultimately, HULU paves the way for incorporating high-fidelity MLPs into high-throughput screening workflows, significantly enhancing the predictive design of nanoporous materials for energy and environmental applications.