Skip to content

Development of a reactive force field for CaCl2·nH2O, and the application to thermochemical energy storage

Summary

The authors re-parameterize a ReaxFF description of CaCl\(_2\)·\(n\)H\(_2\)O (emphasis on \(n=0,2\)) building on Pathak et al.’s prior fit, which showed unphysical disintegration of higher hydrates in MD. A Metropolis Monte Carlo (MMC) optimizer with DFT (and MD-trajectory snapshots added iteratively) stabilizes bulk crystals and reproduces EOS, surface energies, reaction enthalpies, and RDFs. Applications include NEMD thermal conductivity (anisotropic x–y–z), vacuum-slab dehydration (300–500 K, 0.25 fs, NPT equilibration at 1 bar with H\(_2\)O removal in the vacuum), and cracked / porous morphologies; reported \(\kappa\) for anhydrous and dihydrate agree with measurements to the level quoted in the paper.

Methods

1 — MD application (atomistic dynamics)

  • Engine / code: Reactive MD in the workflow described in Comput. Mater. Sci. 197 (2021) 110595; text references NEMD for \(\kappa\) with Berendsen-style region thermostats (hot 330 K / cold 300 K source/sink with τ = 100 fs; interior weak coupling τ = 100 ps to avoid spurious heat leaks as stated in §2.4.2).
  • System & composition: CaCl\(_2\) and CaCl\(_2\)·2H\(_2\)O slabs and defective crystals; 6×6 in-plane slabs with vacuum (e.g. 1000 Å in one direction for dehydration); grain boundaries and crack models as in Figs. / sections on imperfect crystals. H\(_2\)O that moves >30 Å from the surface is removed every 0.125 ps in the dehydration protocol to preserve driving force.
  • Boundaries / periodicity: 3D PBC in supercell equilibration; slabs with vacuum in one direction for dehydration; NEMD \(\kappa\) uses in-cell thermostat zones as in the article.
  • Ensemble: NPT at 1 bar for 3×10\(^5\) iterations in slab preequilibration (0.25 fs); NEMD uses NVT-like heating control in slices (as described in NEMD); NPT / NVT choices follow each subsection in Section 2.4 of the PDF.
  • Timestep: 0.25 fs in the equilibration example in §2.4.1; NEMD \(\Delta t\) per reactive MD standard in the article.
  • Duration / stages: 3×10\(^5\) steps equilibration before slab extraction; 300–500 K dehydration runs; \(\kappa\) at various system lengths to mitigate finite-size (see NEMD discussion).
  • Thermostat: Berendsen (τ = 100 fs) in NEMD hot/cold blocks; weaker coupling (τ = 100 ps) in the transit region as quoted.
  • Barostat / pressure: NPT 1 bar isotropic (Berendsen-style) in slab preequilibration; N/A for pure NEMD \(\kappa\) strips (those fix T zones rather than bulk hydrostat in the NEMD paragraph).
  • Temperature: 300–500 K dehydration; 330/300 K NEMD \(\Delta T\); heating of crack models as in relevant §.
  • Electric field: N/A.
  • Enhanced sampling: N/A.

Corpus note: The extracted text may use "MD" without a dedicated LAMMPS string; reactive dynamics are run with ReaxFF in the software stack cited in the article (confirm in the VOR PDF for version-specific keywords).

2 — Force-field training

  • Parent / scope: Pathak et al. CaCl\(_2\) / hydrate ReaxFF as starting point; goal: fix crystal stability for higher hydrates and condensed phases used in TCM modeling. QM reference: ADF GGA-PW92 on molecular / non-periodic training geometries; VASP PBE + PAW + DFT-D3(BJ) on periodic training crystals; force convergence <0.026 eV/Å as in §2.1. Training set: reactions, condensed phases, and (critically) recomputed QM on “bad” MD frames from unstable intermediate ReaxFF fits, iteratively appended (Fig. 2 loop). Optimization: MMC (Metropolis Monte Carlo) reoptimization of ReaxFF parameters against the augmenting QM set. External validation: \(\kappa\) vs experiment; (de)hydration trends vs T / vapor pressure assumptions in the main text, Figs. 1–3 et seq.

3 — Static QM (training reference)

Covered under §2.1 as the source of Energies/structures for ReaxFF; N/A as a separate conclusion — the manuscript is centered on reactive FF + MD.

4 — Review

N/A.

Findings

Force-field quality: The refined ReaxFF eliminates unphysical melting of dihydrate / bulk samples in the cases shown while matching reaction enthalpies (Fig. 1), EOS (Fig. 6), and elastic/thermal data as tabulated/figured in the article. \(\kappa\): ~1.1 W m\(^{-1}\) K\(^{-1}\) (anhydrous CaCl\(_2\)) and ~0.5 W m\(^{-1}\) K\(^{-1}\) (dihydrate) align with measurements in the quoted comparison. Anisotropy / microstructure: Layered stacks and grain boundaries reduce \(\kappa\) and slow z-direction dehydration (initial rates ~1.9–2.5× slower along z in the cited result). Pores / cracks open fast H\(_2\)O egress pathways (Figure / section on heating + cracks). Comparisons: Experiment (thermal conductivity), prior ReaxFF / literature DFT (training). Limitations (authored + empirical): ReaxFF is still empirical; NEMD \(\kappa\) suffers finite-size effects (mitigation described). Corpus / KB honesty : numeric detail from pdf_path and SI citation in the VOR file.

Limitations

ReaxFF may extrapolate poorly outside the calibrated (de)hydration and T window; NEMD and reactive kinetics remain models, not reactor-scale continua.

Relevance to group

Salt-hydrate ReaxFF development (not PSU author list); relevant to reactive FF practice in energy storage materials.

Citations and evidence anchors

Sections 2.1–2.4 and main figures on training, NEMD \(\kappa\), dehydration / cracks, Comput. Mater. Sci. 197 (2021) 110595.