ReaxFF Reactive Force Field for Exploring Electronically Switchable Polarization in Zn\(_{1-x}\)Mg\(_x\)O Ferroelectric Semiconductors
Summary¶
Sepehrinezhad and colleagues develop a ReaxFF parameterization for wurtzite Zn\(_{1-x}\)Mg\(_x\)O to study ferroelectric switching in Mg-alloyed ZnO, training against Quantum ESPRESSO DFT data for equations of state, phase pathways, and domain-wall and intrinsic switching paths (including nudged elastic band (NEB) images). Large-scale MD with sinusoidal electric fields explores coercive fields vs Mg content, temperature, and Mg spatial distribution (random, uniform/spaced, clustered), reporting P–E hysteresis characteristics such as ~10 nm critical thickness for observable switching and ~100 μC/cm\(^2\) residual polarization in the parameterized scenarios described in the abstract.
Methods¶
DFT reference. Quantum ESPRESSO with norm-conserving PseudoDojo potentials; PBE GGA; k-spacing 0.05 Å\(^{-1}\); 80 Ry wavefunction cutoff / 320 Ry density cutoff (with stated SCF thresholds). EOS scans use 2×2×2 wurtzite/rocksalt supercells across x = 0, 25, 50, 75, 100%; intrinsic switching uses 3×3×2 supercells at x = 5.6–27.8% with 75-image NEB; domain-wall migration uses 8×1×2 cells at 12.5% Mg with 25-image paths for Mg behind/in front of the wall.
ReaxFF training. Starting from merged ZnO / MgO ReaxFF priors (literature sets cited in §2.2), authors refine Zn–Mg–O interactions against DFT, including bond/angle/off-diagonal terms enumerated in §3.2.
MD protocol. After minimization, NPT equilibration at 100 K with Δt = 0.25 fs; Berendsen thermostat (100 fs coupling) and barostat (2500 fs coupling). Sinusoidal E-field sweeps at 100–1000 K for Mg fractions 0–37.5%; five repeats per condition. Fixed charges (post-equilibration averages by species) are used during field runs to avoid spurious bulk conduction from EEM-style long-range charge transfer in wide-gap slabs (§3.4). Polarization computed via Born effective charges and atomic positions (eq. 2).
Mg distributions. Random substitution, uniform spacing via farthest-point sampling, and clustered arrangements.
1 — MD application (field-driven ReaxFF). Engine: ReaxFF in LAMMPS-class workflow (article). System: wurtzite Zn\(_{1-x}\)Mg\(_x\)O slabs and supercells; sizes and PBC in §3. Ensemble and relaxation: after minimization, NPT equilibration at 100 K; Berendsen thermostat (100 fs coupling) and barostat (2500 fs coupling); Δt = 0.25 fs. Production: sinusoidal electric field sweeps at 100–1000 K; fixed post-equilibration species charges during field runs to limit spurious long-range charge sloshing (§3.4). Polarization: via Born effective charges and atomic positions (eq. 2). Electric field — time-dependent E-field (sinusoidal sweep) as in §3. N/A — umbrella sampling, metadynamics, and replica exchange (not used as the main enhanced-sampling method in the summarized protocol).
2 — Force-field training. Parent: merged ZnO / MgO ReaxFF priors (cited in §2.2). QM (training reference): Quantum ESPRESSO, PBE GGA, norm-conserving PseudoDojo potentials, 0.05 Å\(^{-1}\) k-spacing, 80 Ry wavefunction / 320 Ry density cutoffs, stated SCF tolerances. Targets: EOS for wurtzite/rocksalt 2×2×2 supercells at x = 0, 25, 50, 75, 100%; NEB (75 images, 3×3×2 intrinsic switching at 5.6–27.8% Mg); domain wall paths (25 images, 8×1×2 at 12.5% Mg). Optimization: Zn–Mg–O bond/angle/off-diagonal refits against DFT in §3.2.
3 — Static QM — the DFT reference block is embedded in training above; no separate “QM-only” application section is the focus.
Findings¶
Hysteresis and coercivity (trends with composition and arrangement). Higher Mg (alloying) and T both trend toward lower coercive E in the simulations; clustered Mg can raise E_c vs uniform spacing, and random Mg tends to lower it vs uniform at similar x (abstract/§3).
Polarization, thickness, and levers. Simulations point to residual P on the order of ~100 μC/cm\(^2\) and a ~10 nm-scale critical thickness for switching in the parameterized scenarios (abstract; verify numerics in VOR if pagination differs from galley). Sinusoidal E-field sweeps over T and x are the main sensitivity probes.
Corpus honesty — see ## Limitations (galley, fixed-charge post-processing).
Limitations¶
Galley PDF may differ slightly from final JPCC pagination; fixed-charge approximation addresses EEM artifacts but is not a full polarizable electronic treatment. DFT and ReaxFF remain approximations for defective, disordered alloys.
Relevance to group¶
Direct ReaxFF parameterization output from the van Duin group on ferroelectric oxide alloys with explicit DFT training sets and field-driven MD.
Citations and evidence anchors¶
- Methods: §3 (J. Phys. Chem. C, DOI 10.1021/acs.jpcc.4c02233).