ReaxFF parameter set for boron clusters and icosahedral boron crystals: comparison with density functional theory and machine-learning potentials
Summary¶
The authors reoptimize ReaxFF parameters for elemental boron (clusters and condensed phases) using VASP PAW GGA-PBE training data for B/O/Ni-related sets (iteratively adding B\(_{80}\) cluster motifs, including a low-energy core–shell structure and a high-energy “Pouch” structure from early ReaxFF runs). The new parameter set is compared against DFT (also GPAW PBE for a 58-cluster test set) and several machine-learning interatomic potentials (MACE, PFP, SevenNet, etc.) on relative cluster stabilities, icosahedral character in melt/crystal simulations, and B solubility in molten Ni.
Methods¶
- Training DFT (VASP): PAW GGA-PBE, 520 eV cutoff, Γ-centered k-meshes at 0.03 Å\(^{-1}\)-type density for periodic systems; Γ-only for 20 Å cubic boxes of isolated clusters; spin-polarized treatment for Ni-containing cells; electronic/ionic convergence thresholds as stated (10\(^{-6}\) eV SCF, −0.02 eV/Å ionic for bulk).
- Test-set DFT (GPAW): PBE, 520 eV, Γ sampling, 0.2 eV Fermi smearing, BFGS relaxation (0.02 eV/Å force threshold) for 58 clusters spanning 8–103 atoms.
- ReaxFF optimization: Iterative Versions 1–3 expand training with B\(_{80}\) data from Li et al. and internal ReaxFF-generated isomers; energies are compared after ReaxFF-side geometry relaxation (RMSD bounds and bond-network checks reported).
- MD (LAMMPS, Jun 2022): ReaxFF with GPU kernels; 0.25 fs timestep; NVT (Langevin, 200 ps damping) for clusters; NPT (Nosé–Hoover, 100 fs time constants) for melts/crystals; CG minimization to 1×10\(^{-8}\) kcal mol\(^{-1}\) Å\(^{-1}\) force tolerance; analysis/visualization in VMD with Tcl scripts (Zenodo).
N/A (add-on protocol notes for this reparameterization paper): External electric field; umbrella / metadynamics (not used in the training workflow summarized above). MD validation temperatures follow the K schedules in the melt and NPT blocks (e.g. 1300–3600 K anneal / quench windows in the PDF), with Langevin or Nose-Hoover thermostat selection as written for each stage.
Findings¶
- Among prior ReaxFF sets and the MLIPs tested for ranking B\(_{80}\) isomers, only the new ReaxFF and PFP v7.0.0 reproduce the DFT ordering for the two B\(_{80}\) reference morphologies highlighted.
- On the 58-cluster test set, the refined parameters improve energy agreement versus earlier boron ReaxFF and compare competitively with several MLIPs in aggregate metrics reported in the paper.
- Larger supercooled liquid boron simulations show higher icosahedral content with the new field than the older boron parameterization.
- Solid B | molten Ni interface simulations yield boron solubility closer to experimental expectation, whereas the previous boron parameter set underestimated solubility strongly.
Limitations¶
- Transferability to boron-rich compounds beyond the training/test sets (e.g., complex borides/hetero-systems) still requires separate validation.
- Cluster DFT references mix VASP and GPAW pipelines; users should follow SI for exact cluster-by-cluster handling.
Relevance to group¶
Adri C. T. van Duin is a co-author; the paper is a core ReaxFF parameterization reference for icosahedral boron and Ni–B processing contexts aligned with other boron/hBN work in the corpus.