Predicting the oxidation of carbon monoxide on nanoporous gold by a deep-learning method

Nanoporous golds (NPG) exhibit excellent performance on CO oxidation due to unique pore structures with huge surface areas. However, it is a challenge for the prediction-assessing of catalytic performance without any experimental data. Herein, deep-learning methods are used to establish the structure-property relationship on NPGs. By comparing different learning algorithms, the deep recursive neural network (DRNN) can be assessed as the best network up to more than 98% accuracy. With several structural parameters, our trained model can be employed as an accurate descriptor to predict the transformation capability of CO molecules on gold catalysts and further to estimate the optimal zone in this system combined with nonlinear programming map. These meaningful findings greatly expand a way to optimize catalysts combining interdisciplinary research means, exhibiting a great application potential in the design of future catalysts.

Automated summary

Deep-learning methods were used to establish the structure-property relationship of NPGs, accurately predicting the transformation capability of CO molecules on gold catalysts and demonstrating great application potential in future catalyst design.