Nanoscale structure of ordered mesoporous carbon formed by nanocasting within a hard template: MD simulation with ML potentials

Abstract

Molecular Dynamics simulations with machine learned (ML) potentials are utilized to develop models for the structure of ordered mesoporous carbons (OMC) formed by nanocasting in hard templates with linear cylindrical pores, e.g., SBA-15. Potentials are initially trained on PBE DFT, but subsequently enhanced by transfer-learning to suitably incorporate vdW energetics. Rather than analyze the larger-scale mesoporous structure, we focus on the disordered partially graphitized structure of a carbon nanorod formed within a single pore of the template. In benchmark simulations, carbon is confined to a cylindrical pore by surrounding inert immobile atoms, and in more realistic simulations by utilizing an atomistic model for amorphous mesoporous silica. We assess the dependence of OMC structure on the description of the hard template, and on the selection of the carbon precursor. Depending on the template materials and synthesis conditions, structures ranging from multi-wall carbon nanotubes or nanoscrolls to more disordered and complex structures involving carbon nanoribbon and nanodisc motifs, are found in simulations with the more realistic templates. At lower temperatures, where dehydrogenation processes occur at a similar or slower time scale than the growth of the carbon material, the choice of carbon feedstocks also affects the structures of the synthesized OMC.