High-throughput computational screening of metal–organic frameworks for 3He sorption pumps in dilution refrigerators

Abstract

Sub-Kelvin dilution refrigerators driven by ³He sorption pumps offer advantages in low vibration and long lifespan over those using mechanical pumps. While activated carbon remains the most widely used adsorbent in such sorption pumps, its limited adsorption capacity results in a small ³He circulation flow rate, determined by the difference in adsorption amounts between the high- and low-loading state points per cycle. Here, we proposed metal–organic frameworks (MOFs) as alternative adsorbents and developed a machine learning (ML)-accelerated high-throughput screening framework to identify optimal candidates. Structural and energetic features of MOFs were represented as descriptors and used as inputs to ML models. Models based on energetic descriptors achieve high prediction accuracy for ³He adsorption at both state points. Feature importance analysis and adsorption visualization reveals a transition from layered adsorption at the high-loading point to a site-preferential distribution at the low-loading point. Structure–performance and energy–performance analyses indicate that at the high-loading point, the adsorption amount increases approximately linearly with the accessible surface area, while at the low-loading point, it correlates with the fraction and potential energy of specific adsorption sites. Notably, the majority of the screened MOFs outperform the specific activated carbon investigated in this study, and the strong agreement between the ML-predicted and grand canonical Monte Carlo-simulated rankings of high-performance MOFs validated the screening method. This work not only identifies superior adsorbents for cryogenic sorption applications but also establishes structure–performance and energy–performance relationships that serve as preliminary screening criteria and guide the rational design of next-generation porous materials.