Seeking metal–organic frameworks for hydrogen storage using classical and quantum active learning

Abstract

Metal–organic frameworks (MOFs) are porous materials with applications from chemical sensing to gas storage and separation. The development of MOFs for hydrogen storage is highly desired. Hydrogen is a clean source of energy and storing it within MOFs depends on their structure, stability, synthesis design and adsorption characteristics. The design space to obtain MOFs with suitable target properties can be aided by artificial intelligence methods that use few data for new discovery, such as active learning (AL) and the recent quantum AL (QAL) method. In this work, AL and QAL methods for MOF design were developed and tested in the search for MOFs and experimental conditions (temperature and pressure) that have enhanced hydrogen storage capability. Our aim is to explore the performance of these techniques in finding this optimum material within an experimental MOF dataset obtained from the literature. The AL methods investigated in this work use artificial neural networks, support vector regression, and classical and quantum Gaussian processes (GP and QGP) as regression models for inference. Different uncertainty quantification methods for the regression models as well as different acquisition functions for decision making were considered to select the next MOF to be measured by further “experiments”. The QAL performance in finding the optimum material and experimental conditions is reported. QAL uses QGP with a projected quantum kernel with a feature map with entanglement. A network graph method was developed to analyze the AL and QAL performance for MOF search. The results obtained from applying AL and QAL in this known data set indicate that it is possible to search for MOFs with enhanced properties for hydrogen storage with very few data, with the ability to distinguish the optimum MOF and conditions from similar ones. This finding highlights the potential of classical and quantum “machines” (i.e. AL and QAL methods) to indicate new MOFs to be synthesized with enhanced adsorption properties.