Skip to content

The quantum mechanics-based polarizable force field for water simulations

Summary

RexPoN is a polarizable molecular-mechanics water potential whose parameters are fit entirely to quantum data rather than empirical liquid properties. Naserifar and Goddard III assemble coupled-cluster CCSD(T) energies for the water dimer across distances and orientations, X3LYP energies for 19 larger clusters capturing many-body polarization, fluctuating charges via a polarizable charge-equilibration formalism, and DFT-D3 dispersion references anchored to H\(_2\) and O\(_2\) crystals. The JCP article situates RexPoN within Goddard-school FF development where QM training data are preferred over empirical liquid fits, trading lower computational cost than explicit QM solvent for higher fidelity than fixed-charge water models in hybrid pipelines. The resulting field reproduces melting point, 298 K density, heat of vaporization, entropy, dielectric constant, and self-diffusivity near experiment without post hoc scaling to experimental densities.

Methods

1 — MD application (RexPoN validation). After integrating RexPoN into LAMMPS, the authors run liquid water benchmarks (e.g., 216 molecules/cell at 298 K) and ice melting studies. Protocol excerpts in J. Chem. Phys. include NPT equilibration segments (~1 ns), NVT segments (150 ps blocks), and NPT relaxations (120 ps) feeding two-phase thermodynamics analysis, followed by additional NVT cycles for analysis windows (100 ps total split into 5 × 20 ps in the excerpt). Thermostat: Nosé–Hoover with 100 fs damping (rigid water unless noted otherwise in their rigid-molecule discussion). Barostat: relaxation time ~1 ps for pressure control in NPT melting studies. Timestep: 1.0 fs for the reported RexPoN LAMMPS validation and melting workflows (the article notes 2.0 fs would likely be acceptable but was not used for the quoted benchmarks). PBC: three-dimensional PBC for bulk water cells. Electric fields / enhanced sampling: N/A — not used in the benchmark suite summarized here.

2 — Force-field training (RexPoN). Parent / functional form: new RexPoN polarizable water model (not ReaxFF). QM reference: CCSD(T) dimer surfaces across orientations/distances; X3LYP DFT on 19 larger clusters; DFT-D3 dispersion coefficients anchored to H₂ and O₂ molecular crystals. Training set: dimer + cluster binding/polarization manifolds described in Section II. Optimization: parameters determined to reproduce QM many-body energies while retaining efficient MD (see energy decomposition and PQEq coupling in the article). Reference data: compared against experimental melting (273.15 K target), 298 K density, ΔH_vap, entropy, dielectric constant, and ln D_s tabulated in the abstract.

3 — Reactive ReaxFF MD. N/A — this paper is not a ReaxFF reactive study.

Findings

Outcomes / mechanisms: RexPoN reproduces ice melting near 273.3 K (cf. 273.15 K experiment) and 298 K liquid benchmarks (density, ΔH_vap, entropy, dielectric constant, ln D_s) close to experiment without empirical liquid refitting.

Comparisons: metrics are compared explicitly to experimental values listed in the abstract.

Sensitivity / outlook: accuracy comes at higher cost than TIP3P-class models; electrolyte and reactive interface extensions are flagged as needing future validation.

Limitations: rigid-water approximations in parts of the workflow and 2PT-based analysis windows depend on the detailed Methods sections.

Corpus honesty: benchmark numbers are taken from the JCP abstract on papers/Others/JCP-RexPoN_forcefield.pdf; this page is not ReaxFF despite the wiki’s optional Related topics link.

Limitations

Not a ReaxFF model; transferability to electrolyte or reactive interfaces requires separate validation. Computational cost exceeds fixed-charge water models.

Relevance to group

Complementary polarizable water FF literature adjacent to ReaxFF/eReaxFF aqueous workflows.

Citations and evidence anchors