Formation of amorphous carbon multi-walled nanotubes from random initial configurations

Amorphous carbon nanotubes (a-CNT) with up to four walls and sizes ranging from 200 to 3200 atoms have been simulated, starting from initial random configurations and using the Gaussian Approximation Potential [Phys. Rev. B 95, 094203 (2017)]. The important variables (like density, height, and diameter) required to successfully simulate a-CNTs, were predicted with a machine learning random forest technique. The models were validated $\textit{ex post facto}$ using density functional codes. The a-CNT models ranged from 0.55 nm - 2 nm wide with an average inter-wall spacing of 0.31 nm. The topological defects in a-CNTs were discussed and new defect configurations were observed. The electronic density of states and localization in these phases were discussed and delocalized electrons in the $\pi$ subspace were identified as an important factor for inter-layer cohesion. Spatial projection of the electronic conductivity favors axial transport along connecting hexagons, while non-hexagonal parts of the network either hinder or bifurcate the electronic transport. A vibrational density of states was calculated and is potentially an experimentally testable fingerprint of the material and the appearance of a low-frequency radial breathing mode was discussed. The thermal conductivity at 300 K was calculated using the Green-Kubo formula.