Simulation of the elastic and ultimate tensile properties of diamond, graphene, carbon nanotubes, and amorphous carbon using a revised ReaxFF parametrization
Scope
Large-strain reactive MD benchmarks of ReaxFFC‑2013 (carbon parametrization including DFT mechanical training data) against DFT and experiment for diamond, graphene, amorphous carbon, and nanotubes, using VASP reference calculations and LAMMPS simulations.
Summary¶
The study evaluates the ReaxFFC‑2013 reactive force field for elastic response and fracture of several carbon allotropes. DFT reference data use VASP with PAW–PBE and Grimme DFT-D2 (PBE-D2) van der Waals corrections. Reactive MD uses ReaxFF in LAMMPS, comparing ReaxFFC‑2013 to the older ReaxFF CHO parametrization for diamond and graphene, then testing transferability to amorphous carbon and carbon nanotubes.
Methods¶
Static QM (DFT reference). Reference stresses and energies use VASP with PAW–PBE and Grimme DFT-D2 (PBE-D2), virial stresses at deformed geometries, and k-meshes 15×15 (800 eV cutoff) for graphene and 16×16×16 (700 eV cutoff) for diamond primitive cells (convergence-checked as reported).
Reactive MD (LAMMPS + ReaxFF). Large-strain tests use LAMMPS ReaxFF with ReaxFFC‑2013 everywhere; ReaxFF CHO appears only for the diamond/graphene comparison against the older parametrization. Systems include periodic diamond supercells (~8000 atoms, ⟨100⟩ and ⟨112⟩ orientations), single-layer graphene (~6696 atoms, 3D PBC with vacuum normal to the sheet, 0.34 nm effective thickness for stress), plus amorphous carbon and CNT arrays detailed in the article and SI (structures visualized with Ovito).
Equilibration before loading (diamond/graphene). Each allotrope model is heated 0 → 300 K over 5 ps (Langevin), held at 300 K and ~0 pressure for 16 ps (Berendsen thermostat + barostat), then 2 ps at 300 K after removing the barostat. Tensile production segments use Δt = 0.2 fs as stated.
CNT and amorphous carbon paths. CNT arrays are heated 0 → 300 K over 100 ps (Langevin + Berendsen barostat), then run 50 ps with Berendsen thermostat only; array mass densities are tabulated in the paper. Amorphous carbon cells are built by heating disordered configurations to 3000 K under NVT and quenching—densities and cycle counts are in Computational Details.
Findings¶
For diamond and graphene—allotropes represented in the ReaxFFC‑2013 training set—simulated elastic and large-strain responses track DFT (PBE-D2) and available experiment more closely than the older ReaxFF CHO parametrization in the article’s comparisons. Amorphous carbon and CNT arrays, which were not in the mechanical fitting set, still show largely consistent trends with experiment for single- and multi-walled tubes and for amorphous carbon across densities, with specific deviations the authors flag for future parametrization. The discussion stresses GGA+D2 limitations, k-mesh and cutoff choices, and strain-rate / model-size effects common to large-deformation MD. Numerical moduli, strengths, and residuals are figure/table-specific; cite the JPCA PDF for any number used downstream.
Limitations¶
DFT uses a specific GGA+D2 setup; mechanical observables are sensitive to k-sampling, strain rate, and model size. Transferability targets (amorphous carbon, CNTs) lie outside the original mechanical training data for the parametrization.
Relevance to group¶
Benchmarks ReaxFF carbon parametrizations for large-deformation carbon mechanics—adjacent to reactive carbon/hydrocarbon simulation practice in the broader ReaxFF literature.
Citations and evidence anchors¶
Prefer the publisher version via DOI 10.1021/acs.jpca.5b05889 (see pdf_path for the local PDF used here).