Amorphous
carbon systems are emerging to have unparalleled properties
at multiple length scales, making them the preferred choice for creating
advanced materials in many sectors, but the lack of long-range order
makes it difficult to establish structure/property relationships.
We propose an original computational approach to predict the morphology
of carbonaceous materials for arbitrary densities that we apply here
to graphitic phases at low densities from 1.15 to 0.16 g/cm3, including glassy carbon. This approach, dynamic reactive massaging
of the potential energy surface (DynReaxMas), uses the ReaxFF reactive
force field in a simulation protocol that combines potential energy
surface (PES) transformations with global optimization within a multidescriptor
representation. DynReaxMas enables the simulation of materials synthesis
at temperatures close to experiment to correctly capture the interplay
of activated vs entropic processes and the resulting
phase morphology. We then show that DynReaxMas efficiently and semiautomatically
produces atomistic configurations that span wide relevant regions
of the PES at modest computational costs. Indeed, we find a variety
of distinct phases at the same density, and we illustrate the evolution
of competing phases as a function of density ranging from uniform vs bimodal distributions of pore sizes at higher and intermediate
density (1.15 g/cm3 and 0.50 g/cm3) to agglomerated vs sparse morphologies, further partitioned into boxed vs hollow fibrillar morphologies, at lower density (0.16
g/cm3). Our observations of diverse phases at the same
density agree with experiment. Some of our identified phases provide
descriptors consistent with available experimental data on local density,
pore sizes, and HRTEM images, showing that DynReaxMas provides a systematic
classification of the complex field of amorphous carbonaceous materials
that can provide 3D structures to interpret experimental observations.