Revisiting silica with ReaxFF: Towards improved predictions of glass structure and properties via reactive molecular dynamics

Summary¶

The authors compare ReaxFF and a conventional classical potential for a model silica glass, assessing how well each reproduces structural disorder and selected bulk properties. The focus is whether reactive potentials can match or exceed classical descriptions for glassy silicates while remaining affordable relative to ab initio methods, including implications for later surface reactivity studies. The abstract notes that applicability of reactive potentials to glasses “remains poorly understood,” motivating a side-by-side assessment because many ReaxFF fits target crystalline oxides even though glasses are formed by quenching liquids and therefore demand accurate liquid and supercooled-liquid behavior.

Methods¶

MD application (LAMMPS). Glassy SiO₂ is built along two parallel routes in §2.2 (papers/ReaxFF_others/reaxff_bauchy.pdf): user-reaxc ReaxFF and a modified BKS potential with 5.5 Å van der Waals and 10.0 Å Coulomb cutoffs. Each route starts from 4536 atoms in a cubic cell with 3D PBC, melted 4000 K for 100 ps, then cooled 4000 K → 300 K at 1 K/ps under NPT at zero pressure, followed by 1 ns at 300 K to relax the glass. A hybrid path quenches with BKS only, then runs 1 ns at 300 K with ReaxFF without manual reconnectivity edits. Supercooled-liquid analysis (§3.5) uses equilibrated liquid trajectories above T_g for MSD and oxygen self-diffusion. Integration timestep and explicit thermostat/barostat damping values are not in the indexed p1–2 extract—take them from pdf_path before reproducing inputs. Electric fields and replica / enhanced sampling are not part of the quench protocol described in the excerpt.

Force-field training: N/A — the paper benchmarks published ReaxFF and BKS parameterizations; it does not report a new fit.

Static QM / DFT: N/A — no standalone DFT production study; QM appears through the ReaxFF training literature cited in §2.1.

Findings¶

ReaxFF matches neutron S(Q)—including the FSDP window—better than BKS, indicating more realistic short- and medium-range disorder in the modeled glass. Young’s and shear moduli from ReaxFF track experiment far more closely than BKS; hybrid glasses inherit BKS-like medium-range errors and degraded moduli, so the hybrid route is a poor compromise. Oxygen self-diffusion in the supercooled liquid follows Arrhenius trends, with ReaxFF activation behavior closer to experimental extrapolations than BKS. Under identical quench schedules, ReaxFF therefore outperforms BKS on both structure and the bulk properties highlighted in the article, while BKS remains computationally cheaper. The 1 K/ps cooling rate is far above laboratory quenches; sensitivity to thermal history follows the glass-MD literature and limits direct mapping to experiment.

Limitations¶

Cooling rates (1 K/ps) are far above laboratory quenches. Timestep and thermostat damping values are not recovered from the short extract used when drafting this page—confirm against pdf_path for exact reproducibility.

Relevance to group¶

Benchmark-style silica-glass study using ReaxFF in the broader PARISlab/ReaxFF silica literature context (UCLA-led).

Citations and evidence anchors¶

DOI: 10.1016/j.jnoncrysol.2016.03.026.