High-throughput screening of MXenes for hydrogen storage via graph neural network

To better understand the recent experiment on hydrogen storage in MXene multilayers [Nature Nanotechnol. 2021, 16, 331], we propose a multiscale workflow to computationally screen 23,857 compounds of MXene for hydrogen storage in near ambient condition. By using density functional theory simulation to produce the dataset, we trained physics-informed atomistic line graph neural networks to predict hydrogen's adsorption performance on MXenes, which is further validated through grand canonical Monte Carlo simulation. As a result, ScYC is identified to exhibit a hydrogen storage capacity of 5.7 wt% at 230 K and 100 bar, showing the promise for hydrogen storage.

Automated summary

In this study, a multiscale workflow was proposed to computationally screen a large number of MXene compounds for hydrogen storage under near ambient conditions. By training neural networks and validating through simulations, it was found that ScYC exhibits a high hydrogen storage capacity.