Graph neural network-accelerated multitasking genetic algorithm for optimizing PdxTi1–xHy surface under various CO2 reduction reaction conditions

Palladium (Pd) hydride-based catalysts have been reported to have excellent performance for CO2 reduction (CO2RR) and hydrogen evolution reactions (HER). Our previous doped-PdH and alloy hydride works showed that Ti-doped and Ti-alloyed Pd hydride could improve the performance of the CO2 reduction reaction compared to pure Pd hydride. Compositions and chemical orderings of the surfaces with only one adsorbate under certain reaction condition are linked to their stablity, activity, and selectivity towards CO2RR and HER in our previous work. In fact, various coverages, types, and mixture of the adsorbates as well as state variable such as, temperature, pressure, applied potential, and chemical potential could have an impact on their stability, activity, and selectivity. However, these factors are usually fixed at common values in order to reduce the complexity of structures and complexity of reaction conditions in most theoretical work. To address the above complexities and thus huge search space, a deep learning-assisted multitasking genetic algorithm is used to screen for PdxTi1-xHy surfaces containing multiple adsorbates for CO2RR under different reaction conditions in this work. The ensemble deep learning model can greatly speed up the structure relaxations and keep a high accuracy and low uncertainty of energy and forces. The multitasking genetic algorithm is used to simultaneously globally find stable surface structures at each reaction condition. Finally, 23 stable structures are screened out under different reaction conditions. Among them, Pd0.56Ti0.44H1.06+25%CO, Pd0.31Ti0.69H1.25+50%CO, Pd0.31Ti0.69H1.25+25%CO, and Pd0.88Ti0.12H1.06+25%CO are found to be very active for CO2RR and suitable to generate syngas consisting of CO2 and H2.