Towards fully ab initio modeling of soot formation in a nanoreactor

A neural network (NN)-based model is proposed to construct the potential energy surface of soot formation. Our NN-based model is proven to possess good scalability of O(N) and retain the ab initio accuracy, which allows the investigation of the entire evolution of soot particles with tens of nm from an atomic perspective. A series of NN-based molecular dynamics (NNMD) simulations are performed using a nanoreactor scheme to investigate critical processes in soot formation, acetylene polymerization, and inception of PAH radicals. This shows that NNMD can capture the dynamic process of acetylene polymerization into PAH precursors. The simulation of PAH radicals reveals that physical interaction enhances chemical nucleation, and such enhancement is observed for clusters of π- and σ-radicals, which is distinct from the dimer. We also observed that PAH radicals of ~ 400 Da can produce core-shell soot particles at a flame temperature, with a disordered core and outer shell of stacked PAHs, suggesting a potential physically stabilized soot inception mechanism.