Skip to content

The Effect of Time Step, Thermostat, and Strain Rate on ReaxFF Simulations of Mechanical Failure in Diamond, Graphene, and Carbon Nanotube

Summary

Reactive simulations of mechanical failure are sensitive to numerical choices: integration timestep, thermostat algorithm and coupling strength, and loading rate can shift predicted strength and failure mode even when the same ReaxFF parameterization is used. This study systematically varies these controls for ReaxFF simulations of tensile fracture in diamond, single-layer graphene, and a carbon nanotube using the Chenoweth C/H/O reactive force field. Stress–strain curves are compared across thermostat choices, timestep sweeps, and strain-rate decades, and results are discussed relative to literature mechanical benchmarks for carbon allotropes.

The paper’s practical message is controls-first: before debating force-field physics, simulations of failure should document timestep, thermostat coupling, and strain rate because those knobs can move peak stress and ductility as much as small parameter tweaks.

Methods

Potential (force-field training context). All fracture sensitivity tests use the Chenoweth et al. ReaxFF C/H/O parameterization—the combustion-oriented fit whose mechanical performance is being stress-tested.

MD application — systems, code, and stress model. ReaxFF tensile tests are run in LAMMPS on three periodic carbon models used throughout the parameter sweeps: bulk diamond (864 atoms, [111] tensile axis), single-layer graphene (1008 atoms, in-plane periodicity, 0.34 nm effective thickness for stress reporting), and a (12,0) CNT (1152 atoms, axial periodicity, cylindrical shell volume definition for stress). True strain and virial (true) stress definitions follow the Computational Details section.

Ensembles, barostat, temperature. Samples are equilibrated at 300 K for 10–40 ps (multiple seeds reported) before deformation. Diamond and graphene runs use a Berendsen barostat to keep zero lateral pressure while axially loading, allowing Poisson contraction; the CNT uses axial periodicity with the lateral treatment described in the article.

Timestep sensitivity. The timestep study strains each system with Δt from 1.0 fs down to 0.1 fs at the fixed high strain rate used in Fig. 5 (~2.2×10⁹ s⁻¹; the publisher PDF line-breaks the exponent) to probe stability near fracture.

Thermostat sensitivity. The authors compare Berendsen, Nosé–Hoover, and Langevin thermostats, including a wide sweep of coupling constants (2, 25, 250, 2500 fs) for the Berendsen case (other thermostats discussed with the same representative damping where plotted).

Strain-rate sensitivity. Additional sweeps vary applied strain rate to identify a rate-independent regime and a high-rate regime where inertial effects appear.

Loading ensemble. Stress–strain production runs are thermostatted constant-volume tensile tests at 300 K; diamond and graphene use a Berendsen barostat on lateral directions to maintain ~zero lateral pressure while axially loading (Computational Details), rather than NPT hydrostatic control of the full cell.

Force-field reparameterization / static QM. N/A — this paper is a controls study on an existing ReaxFF table, not a new QM training manuscript.

Findings

Thermostat and coupling. Stress–strain curves are largely insensitive to thermostat algorithm if coupling is not too weak; very small damping (~2 fs in their Berendsen sweep) can artificially quench distant segments after fracture, whereas very large damping (~2500 fs) fails to hold 300 K faithfully during rapid exothermic bond rupture—see Figs. 9–11 for the detailed argument.

Timestep. 1.0 fs can shift predicted failure strain (example discussed for graphene) and even show energy drift precursors in diamond; ≤0.25 fs behaves more consistently in the tests shown.

Strain rate. Below a material-dependent maximum strain rate, responses are approximately rate-independent; above it, inertial effects contaminate the measured strength.

Comparisons to references. Benchmarked elastic and failure-related quantities for Chenoweth carbon models deviate from selected experimental/QM literature values in ways the authors tabulate—highlighting validation needs when using this parameter set for mechanics rather than combustion chemistry alone.

Limitations

Finite simulation cells cannot capture long-wavelength crack instabilities; thermostats alter heat transport near crack tips and may bias ductile versus brittle appearance.

Relevance to group

Practical sensitivity-analysis guidance for ReaxFF mechanical failure workflows distinct from reaction-focused training.

Citations and evidence anchors