Transfer learning aided high-throughput computational design of oxygen evolution reaction catalysts in acid conditions

Sluggish oxygen evolution reaction(OER)in acid conditions is one of the bottlenecks that prevent the wide adoption of proton exchange membrane water electrolyzer for green hydrogen production.Despite recent advancements in developing high-performance catalysts for acid OER,the current electrocatalysts still rely on iridium-and ruthenium-based materials,urging continuous efforts to discover better performance catalysts as well as reduce the usage of noble metals.Pyrochlore structured oxide is a family of potential high-performance acid OER catalysts with a flexible compositional space to tune the electrochemical capabilities.However,exploring the large composition space of pyrochlore compounds demands an imperative approach to enable efficient screening.Here we present a highthroughput screening pipeline that integrates density functional theory calculations and a transfer learning approach to predict the critical properties of pyrochlore compounds.The high-throughput screening recommends three sets of candidates for potential acid OER applications,totaling 61 candidates from 6912 pyrochlore compounds.In addition to 3d-transition metals,p-block metals are identified as promising dopants to improve the catalytic activity of pyrochlore oxides.This work demonstrates not only an efficient approach for finding suitable pyrochlores towards acid OER but also suggests the great compositional flexibility of pyrochlore compounds to be considered as a new materials platform for a variety of applications.

An insight to calculate soil conservation service

Quantifying soil conservation service is critical for sustainable land management. However, the traditional algo-rithm includes a gap between barren land and real soil erosion (BSERef), and its paradox is that a mass of barrenland rarely exists in ecologically adaptable areas with fertile soils, which might overestimate soil conservationservice. This study suggests a practical algorithm to effectively evaluate soil conservation service of differentecosystems. We propose a new algorithm based on the difference between cropland without control practicesand real soil erosion (CSERef), which is compared with the BSERef algorithm. Our results show that: (1) a nearlyfive times smaller mean soil conservation amount of cropland occurred in the CSERef algorithm (3.76×106 t·yr–1)than in the BSERef algorithm (20.04×106 t·yr–1);(2) land use has higher explanatory power for the spatial dif-ferentiation of soil conservation rate in the CSERef algorithm (15.93% – 46.34%) than in the BSERef algorithm(5.95% – 44.49%). Our results demonstrate that the BSERef algorithm overestimates the soil conservation serviceof cropland in ecologically adaptable areas, whereas the CSERef algorithm can effectively assess the influence ofland use change induced by anthropogenic activities on soil conservation service. Furthermore, we develop anassessment framework in terms of land use classification system following the Chinese Academy of Sciences. Theframework considers that for water bodies, there is no soil conservation service;for non-ecologically adaptable ar-eas (unused land) with vast barren lands and vegetation cover below 5%, the BSERef algorithm is recommended;the CSERef algorithm is suggested to use in ecologically adaptable areas (including woodland, grassland, crop-land and construction land). This assessment framework can provide scientific assistance for decision-makers toformulate strategies for sustainable land management.