Skip to content

ReaxFF study of surface chemical reactions between α-Al₂O₃ substrates and H₂O/H₂ gas-phase molecules

An Al/O/H ReaxFF is trained to α-Al₂O₃ surface energies (A, C, R, M facets; Al- and O-terminated models), hydrolysis and hydrogen-diffusion targets on (0001), and step–terrace dehydration data, then used in large-scale MD of H₂O and H₂ on α-Al₂O₃(0001) and related terminations with mixed Berendsen thermostats mimicking vacuum MOCVD-like conditions.

Summary

The authors develop an Al/O/H reactive potential starting from prior Al₂O₃ bulk data and DFT surface energies for stoichiometric and non-stoichiometric α-Al₂O₃ slabs (Sun/Kurita/Hüttner-style references in the paper), then refine Al–O, Al–H, and Al–O–H parameters against hydroxylation, diffusion, and step–terrace dehydration energies from literature DFT. Section 4 documents MD protocols: relaxation and heating with Berendsen thermostat/barostat in NPT then NVT segments; mixed thermostats (100 fs substrate vs 10³ fs gas) for H₂O + α-Al₂O₃(0001) with ordered Al terminations (50% vs 100% Al, C_Al_I / C_Al_II) and random Al distributions; timestep 0.15 fs; isothermal holds (e.g., 0.5 ns) at 350–1500 K depending on case; H₂ exposure protocols including heating to 1275 K, optional modified H–H σ bond energy to accelerate H₂ dissociation on O-terminated surfaces (~19.4 kcal/mol barrier reduction stated in abstract), and ~1.5 ns segments to remove O while tracking hydroxyl remnants.

Methods

  • Force field: ReaxFF energy decomposition as in standard formulation; training via successive single-parameter search against QC datasets (formation energy, surface slabs, hydrolysis 1–2 and 1–4 pathways on 50% Al (0001), step–terrace dehydrations, bulk lattice/aH).
  • DFT sources: Surface energies and reaction energies from cited PBE/surface studies; comparisons to bond-scan artificial forces in Fig. 3 for pathway training.
  • MD: H₂O/α-Al₂O₃(0001) grids in Tables 3–4 (varying H₂O/Al ratios); random surfaces with fixed H₂O load; H₂ runs on multiple terminations (A, C, M, R families) per Table 5; Berendsen thermostats/barostats as quoted in §4.1–4.2; σ(H–H) parameter perturbation case for O removal kinetics.

1 — MD application (ReaxFF, vacuum/MOCVD-mimic). Engine: LAMMPS-class ReaxFF molecular dynamics; NVT H₂O and H₂ exposure stages at isothermal K in ~350–1500 K (temperature) windows per Table; 0.15 fs time step; ~0.5 ns holds / ~1.5 ns H₂ segments as cited; mixed Berendsen thermostat on gas vs substrate (e.g. 100 fs substrate vs 10³ fs vapor); earlier segments also use NPT relax with Berendsen thermostat and barostat (near-ambient pressure in bar/GPa units as listed in §4) in article text. PBC slab; E-field and rare-event methods — N/A in the stated H₂O/H₂ protocols on this summary (use VOR/SI).

2 — Force-field training (Al/O/H). Iterative ReaxFF fit to DFT α-Al₂O₃ (0001) A/C/R/M surface energies, hydrolysis/diffusion/dehydration paths, H₂O /H₂ chemistry with PBE-style literature targets (see force-field bullet in Summary and SI for parrex-style search). 3 — Static QM — PBE slab and reaction energies in the fit; separate QM “application-only” manuscript N/A beyond training and select plots.

Findings

  1. The ReaxFF reproduces DFT surface energetics trends and hydrolysis/dehydration training targets for flat and step–terrace models within the figures reported.
  2. Water autocatalysis accelerates hydroxylation: at moderate vapor loads, proton transfer via hydrogen-bond networks and oxohydroxide clustering (O\(_x\)H\(_y\)) competes with full dissociation; 100% Al-terminated (0001) hydroxylates more readily at 350 K than 50% termination and desorbs more water above ~500 K, yet full dehydroxylation is not achieved and a gibbsite-like overlayer is deemed unlikely under the explored sprays.
  3. Random vacancy-like Al distributions on (0001) deviate from ideal 1–2 and 1–4 pathways and appear more reactive than ordered surfaces at comparable Al coverage.
  4. H₂ on O-rich surfaces shows enhanced dissociation kinetics when oxygen coverage is high and surface stability is low; lowering the H–H σ energy (or equivalent barrier scaling) in a controlled parameter test accelerates O removal toward Al-terminated-like states while leaving some hydroxyls after ~1.5 ns.

Corpus honesty — galley PDF; duplicate ingest 2024zhang-venue-manuscript-2 may differ by file hash only.

Limitations

Publisher galley PDF; kinetic rates are classical-ReaxFF estimates; parameter modification for H–H is a diagnostic tool rather than a general gas-phase H₂ model.

Relevance to group

Group-authored ReaxFF for sapphire/2D-growth substrate chemistry with explicit MD protocols for H₂O and H₂ environments.

Citations and evidence anchors

Reader notes (navigation)