Nanotube Derived Ordered Carbons Predicted by Neural Network Potential

Searching for novel carbon allotropes with excellent mechanical and interesting electronic properties is valuable, but such a large structural search remains a challenge if purely based on the traditional density functional theory (DFT) combined with Monte-Carlo (MC) methods. Herein, the neural network potential is utilized to accelerate the sampling of the stochastic surface walking algorithm for a global structural search of ordered carbons from carbon nanotubes (CNTs) under pressure. A variety of unreported ordered carbons are obtained, among which CNTs with diameters smaller than 0.7 nm are more sensitive to pressure than bigger tubes. Most ordered carbons obtained show great thermodynamical and kinetic stability, verified by ab initio molecular dynamics simulations and phonon spectra. The ordered carbons demonstrate direct or indirect band gaps in the range of 0 to 4.4 eV, including 13 superhard (H_v > 40 GPa) structures and 1 ductile (Pugh's Ratio G/B < 0.57) structure, in which the modulus of ordered carbons exhibits a linear correlation with the density. Our study provides a pathway to create new carbons from nanotubes and a deeper understanding of the structural evolution of CNTs under pressure.