Skip to content

Generation and characterization of an improved carbon fiber model by molecular dynamics

Summary

Carbon fiber properties emerge from nanoscale graphitic ordering, cross-linking, and porosity, but building faithful atomistic models that span density, texture, and mechanical observables remains nontrivial. This Carbon article generates atomistic fiber models by coupling kinetic Monte Carlo (kMC) cross-linking with large-scale molecular dynamics, producing two families: a fiber-core motif representing an interior slice of a thick fiber and a thin fiber with explicit surfaces. Initial densities span 1.2–2.0 g/cm\(^3\). The authors characterize models using shape, mass density, pore-size distributions, and hybridization statistics, then validate against virtual X-ray diffraction patterns compared to experimental references. Additional variants remove graphitic sheets stochastically along the fiber axis; partial healing occurs during equilibration, and uniaxial tensile MD yields Young’s moduli within experimental spreads reported for carbon fibers. Industrial carbon fiber modeling frequently jumps directly to continuum or mesoscale representations; this work instead asks what atomistic texture is minimally required before modulus and diffraction signatures simultaneously look credible against experiments, which motivates the dual kMC + MD construction. The authors thereby aim for models that are falsifiable against two independent experimental observables, not just one tuned mechanical number. See the Carbon paper for numerical benchmarks.

Methods

1 — MD application. Atomistic fiber models are built with kinetic Monte Carlo cross-linking of graphitic layers followed by LAMMPS molecular dynamics equilibration and mechanical testing. Fiber-core and thin-fiber morphologies span 1.2–2.0 g cm⁻³ target density; stochastic removal of layers explores defect variants with partial healing during relaxation. MD uses PBC supercells with thousands of atoms (exact counts in Carbon), NVT or NPT segments as appropriate for density control (see article), thermostat and timestep (fs) per Methods, and equilibration then production runs totaling ns-scale sampling where reported. Uniaxial strain MD yields Young’s modulus; strain rate and temperature (K) are specified in the paper. N/A — external electric field; N/A — umbrella / metadynamics in the generation workflow.

2 — Force field. Classical empirical carbon potential (non-ReaxFF) with fixed bond topology after kMC—suitable for elastic and structural property evaluation without reactive bond breaking in the reported tests.

3 — Static QM. N/A.

4 — Experiments (laboratory). N/A — virtual XRD is compared to experimental diffractograms from the literature.

Findings

Fiber-core and thin-fiber models reproduce the densities and structural signatures reported in the benchmarks shown. Virtual XRD matches experimental profiles in the comparisons presented. Defective variants with random layer removals partially heal upon relaxation, and predicted moduli fall within the experimental spread for the models tested—supporting the authors’ claim that the workflow yields improved fiber-like models for downstream property studies compared to simpler graphitic constructs. The practical implication for multiscale workflows is that kMC + MD can jointly control pore statistics and graphitic alignment, producing microstructures that are not forced to match a single ideal crystal assumption yet still reproduce XRD and modulus checks used in the carbon fiber literature.

Comparisons, sensitivity, limitations (corpus). Moduli and XRD are compared to experiment; density and defect density are the main levers tuned. Full numerical tables remain in the version-of-record PDF.

Limitations

Nanometer-scale models omit full fiber diameter, industrial processing heterogeneity, and misorientation statistics beyond those encoded in the generation rules.

Relevance to group

Classical carbon microstructure generation benchmark in the corpus (non-ReaxFF).

Citations and evidence anchors