No abstract

Natural gas remains an essential energy source for the industrial and residential sectors. However, selective valorization of methane (the main component of natural gas) into more mobile liquid energy carriers such as methanol remains challenging. Inspired by pMMO enzymes, many recent studies have examined Cu-exchanged zeolites as promising catalysts, specifically through [CuOCu]2+ sites. These efforts, in part, have been motivated by the possibility of finding an elusive “Goldilocks” active site or topology that can outperform known catalysts while also maintaining selectivity towards methanol. As large-scale experiments with 1000s of material variations are impossible, theory will likely play an important role. Although computational screening studies are now routine for metals and alloys, similar studies for zeolites are not as straightforward due to the diversity of local chemical environments, and the aforementioned studies are not trivial using the traditional density functional theory (DFT)-based approach. Therefore, the overarching goal of this study is to leverage large-scale DFT calculations to develop a reactive machine learning-based potential (rMLP) capable of systematically sampling the stability and reactivity of all [CuOCu]2+ sites within a representative set of zeolites. Specifically, using methane activation as a prototypical example of an industrially relevant zeolite-catalyzed reaction, we have developed a novel multistage active learning algorithm that preferentially samples the potential energy surface of the system near the transition state of methane activation. We show that the resulting rMLP replaces the expensive DFT-based NEB calculations without any appreciable loss in accuracy (within 0.07 eV of the DFT computed energy barriers) – we evaluate C-H bond activation energies for 5,400 distinct sites across 52 zeolites and obtain 3,356 valid sites suitable for methane activation. By replacing the expensive DFT-based NEB calculations with rMLPs, we now report an exhaustive high-throughput screening study of thousands of [CuOCu]2+ sites in zeolites, comparing the maximum rates of methane activation across 52 zeolite topologies and more than 3,000 sites. To the best of our knowledge, this work represents the first example of using reactive MLPs to identify the transition state geometries and screen the catalytic performance of thousands of zeolite-based active sites at DFT accuracies.

Supported Ni catalysts were synthesized using the beta-zeolite framework, with and without the framework Al, as a platform for dispersing Ni. The silanol nest sites of dealuminated zeolite beta provide isolated cationic Ni sites that can be reduced under relatively mild conditions to create highly dispersed metal clusters. Compared to the Ni sites present in Ni-[Al]-beta-19, Ni-[DeAl]-beta exhibit a 20-fold increase in the apparent reaction rate for C2H4 hydrogenation and is stable, with little deactivation over 16 h of catalysis. Ni K-edge X-ray absorption spectroscopy (XAS), as well as CO adsorption monitored with Fourier transform infrared spectroscopy, shows that in the oxidized Ni-[DeAl]-beta catalyst Ni reoccupies vacant silanol nests produced from dealumination. After reductive treatment, XAS shows that approximately 50% of Ni is reduced to metallic Ni, forming clusters that are approximately 1 nm in size. Scanning transmission electron microscopy images are consistent with the absence of large (>1 nm) metallic Ni clusters. These results indicate that [DeAl]-beta can be used to synthesize isolated cationic Ni sites as well as stabilize highly dispersed metal clusters that can be used as a highly active and stable C2H4 hydrogenation catalyst.

Copper-based zeolites have been widely explored as promising catalysts for the methane valorization reaction to form methanol. These studies are motivated by the hope of finding an elusive ‘Goldilocks’ topology or an active site that shows high methanol selectivity at reasonable methane conversions. As large-scale screening studies with density functional theory (DFT) remain challenging for zeolite catalysts, we now show that a reactive and interpretable machine learning-based potential (rMLP), developed using multistage active learning algorithm and a curriculum-based training (CBT) approach can be used to overcome this bottleneck. Our rMLP approach replaces expensive DFT-based NEB calculations without appreciable accuracy loss. We calculate methane activation barriers for all possible [CuOCu]2+ sites across 52 zeolites with an MAE of 0.07 eV versus DFT. By comparing with known experimental measurements, our approach establishes the limits of methane activation performance across 52 zeolite topologies. Finally, we show that our curriculum-based training (CBT) approach, which relies on several different types of calculations, gradually “teaches” the model about different relevant parts of the PES. This progressive training approach has important implications for the interpretability of emerging machine learning-based approaches.

No abstract