Skip to content

Optimization of the Reax force field for the lithium-oxygen system using a high fidelity charge model

ACKS2-enabled ReaxFF for Li-O is trained on DFT, MRCI, and MRCI+Q bond-breaking data with genetic optimization, then used in LAMMPS (ACKS2-enabled branch) for Li2O melt and fracture tests vs the 2016 QEq ReaxFF parameterization.

Summary

The paper targets unphysical partial charges during ionic bond cleavage in reactive MD by replacing EEM/QEq-style equilibration with atom-condensed Kohn-Sham DFT to second order (ACKS2) charge coupling inside ReaxFF. The new fit improves charge evolution along Li-O stretch dissociation and yields more realistic mechanical response than the 2016 QEq Li-O parameter set, yet still does not reproduce full brittle fracture in the slab tests shown.

Methods

  • Reference data: Training/validation sets include DFT and multireference MRCI / MRCI+Q dissociation curves for Li-O-containing motifs (detailed in the article).
  • Optimization: Genetic algorithm (OGOLEM workflow as described) to optimize ReaxFF parameters against the quantum references.
  • MD engine: LAMMPS with the authors' ACKS2 branch for charge updates during MD.
  • Integration checks: Bulk Li2O cubes relaxed in NPT at 0 pressure, 300 K (40 ps reported in one protocol block); NVE tests establish 0.25 fs as the stable timestep (used consistently thereafter).
  • Fracture protocol: Notched Li2O slabs equilibrated with NPT, then strained under NVT plane-strain loading at 300 K (geometry and strain rate in article).

1 — MD application (atomistic dynamics). Engine / code: LAMMPS with the authors’ ACKS2-enabled ReaxFF branch. System size & composition: Li₂O bulk supercells (melt tests) and notched fracture slab supercells; atom counts and dimensions in the PDF (pdf_path). Boundaries / periodicity: PBC in 3D for the bulk melt; the fracture setup uses the in-plane PBC and free surfaces normal to the tensile axis as in the article’s figures. Ensemble: NVE for 0.25 fs timestep stability checks; NPT for 0-pressure equilibration of Li₂O at 300 K (~40 ps in one protocol line); NVT for plane-strain fracture at 300 K. Thermostat: the published protocol specifies integrators/ensembles for 300 K NPT and NVT stages; copy thermostat type and damping from pdf_path when reproducing. Barostat: N/A — during the NVT fracture segment; NPT applies only in melt preequilibration. Pressure: N/A — explicit stress control in NVE; 0 GPa target in NPT melt preequilibration. Electric field: N/A — not used. Shear / shock: N/A as a driver beyond the prescribed strain protocol. Replica / enhanced sampling: N/A — not used.

2 — Force-field training (ReaxFF + ACKS2). Parent FF / elements: ReaxFF for Li–O, compared head-to-head to a 2016 QEq-based Li–O parameter set. QM reference: DFT and MRCI / MRCI+Q Li–O dissociation and reaction data (code and basis / active-space details in the article and SI). Training set: Bond-breaking and charge-response targets along Li–O coordinates and related entries in the fit (per article). Optimization: Genetic algorithm in the OGOLEM-style workflow described in the text. Reference data after fitting: Li₂O melt and fracture benchmarks in LAMMPS to stress-test QEq vs ACKS2 on the same cells.

Findings

  • Outcomes & mechanisms: ACKS2 ReaxFF reproduces training-set charges and bond-breaking trends substantially better than 2016 QEq for the Li–O stretches highlighted; NPT melt of crystalline Li2O shows qualitatively different amorphization/volume response between ACKS2 vs 2016 QEq (including RDF signatures discussed in the paper).
  • Comparisons: Head-to-head comparison against the 2016 QEq parameterization on the same Li2O melt and fracture test cells frames where charge fidelity changes observables.
  • Sensitivity & design levers: Temperature is fixed at 300 K in the showcased melt/fracture benchmarks; strain rate and slab notch geometry enter the fracture phenomenology as reported in the article.
  • Limitations & outlook (as authored): Fracture simulations still exhibit residual ductile character vs expected brittleness; charge fixes help but do not fully solve fracture phenomenology in the showcased setups—the article frames remaining gaps as motivation for further reactive-model development.
  • Corpus / KB honesty: This page condenses version-of-record-style claims from pdf_path with good extraction_quality; line-level thermostat/barostat labels should be confirmed against the PDF if reproducing protocols exactly.

Limitations

Accepted-manuscript PDF in corpus may differ slightly from final typesetting; fracture outcomes remain sensitive to classical reactive FF limits and simulation size/rate. ACKS2 improves charge response along dissociation coordinates, but brittle vs ductile fracture remains a stringent test that can expose remaining classical limitations beyond charge models alone.

Relevance to group

van Duin as co-author (parameterization lineage); MSU/Brown collaboration on charge fidelity for ionic ceramics. Pair with other Li–O / solid electrolyte parameterization notes when comparing QEq vs ACKS2 charge pathways in reactive simulations.

Reader notes (navigation)