Machine-learning driven high-throughput DFT calculations for catalyst screening

Electrocatalysis provides a promising solution towards removing nitrate through electrochemical reduction reactions. Electrochemical reduction reactions transform nitrate into benign by-products such as nitrogen or ammonia that stem from the available reaction pathways rather than storing it in excess brine solutions or biological wastes, reducing the cost of subsequent water treatments. However, the materials for high-performance electrocatalysts in nitrate reduction are costly, hindering commercialization. To respond to this challenge, we develop a high-throughput computational platform for identifying novel materials that integrate machine learning and high fidelity first-principles simulations. We conducted the most extensive search to date for electrocatalysts that can facilitate nitrate reduction reaction, evaluating 59,390 bimetallic alloys from the Materials Project and Automatic Flow (AFLOW) databases. We screened our candidates based on corrosion resistance, catalytic activity, selectivity, cost, and synthesizeability using a joint machine learning and density functional theory (DFT) -based screening strategy. We firstly found that 25 materials will satisfy all criteria according to our machine learning screening strategy. We then verified the adsorption energies of materials from our final viable candidates using a high-throughput DFT workflow programmed in the software package atomate. After comparing the calculation results with the activity volcano plot generated from nitrate reduction reactions, three materials stand out as the potential catalysts that merit further experimental validations. By modifying the workflow according to other reaction situations, we expect a high-throughput platform to screen the catalyst materials to remove other oxyanions in wastewater purification, such as selenium or boron.