Defect Dynamics in 2-D MoS2 Probed by Using Machine Learning, Atomistic Simulations, and High-Resolution Microscopy
Genetic-algorithm searches with ReaxFF energies, large-scale ReaxFF molecular dynamics in LAMMPS, and in situ HRTEM are combined to map how sulfur vacancies organize in monolayer MoS2 and how those defects participate in the semiconducting 2H to metallic 1T transition.
Summary¶
The study couples supervised learning style genetic algorithms (GA) with ReaxFF and microscopy to describe sulfur-vacancy ordering, dynamics, and their connection to the 2H→1T transition in monolayer MoS2. GA searches evolve binary “genomes” for sulfur occupancy on a triangular lattice using ReaxFF energies in LAMMPS (population 32, ~100-generation convergence) on ~5.2 nm × 5.8 nm sheets. MD uses the MoS2 ReaxFF parametrization in the NPT ensemble (300–1500 K, ambient pressure) on large defective sheets (up to ~21.8 nm × 24.9 nm; 17,140 atoms) with Nosé–Hoover thermostat and barostat. Experiments use electro-ablated monolayer MoS2 on grids for AC-S/TEM imaging.
Abstract-level context notes that sulfur vacancies are common in MoS\(_2\) and can reorganize under beam or thermal stimuli, so connecting GA-selected low-energy vacancy motifs with in situ microscopy and large-cell reactive MD targets the 2H semiconducting versus 1T metallic phase competition in defective monolayers.
Methods¶
1 — MD application (defective MoS₂). ReaxFF (Mo–S parametrization cited from JPCL 2017) is run in LAMMPS on periodic monolayer MoS₂ supercells up to ~21.8 nm × 24.9 nm (17 140 atoms in the large-cell example) prepared with GA-derived vacancy line patterns. MD uses the isotropic NPT ensemble from 300–1500 K at ambient pressure with Nosé–Hoover thermostat and barostat as stated in ACS Nano. Timestep: N/A — the accessible PDF text for this pass does not spell out Δt; confirm in the article/SI before reproducing runs. Duration: reactive MD segments span the picosecond–nanosecond range reported for 2H/1T nucleation studies (exact lengths in PDF/figures). PBC: in-plane periodic monolayer cells. Electric fields / enhanced sampling: N/A — not used in the MD workflow summarized here.
2 — Genetic algorithm search. GA evolves binary S-occupancy genomes on a triangular lattice (population 32, ~100 generations) using ReaxFF energies evaluated in LAMMPS.
3 — Microscopy. Electro-ablated monolayer MoS₂ on TEM grids is imaged with aberration-corrected STEM/TEM under controlled dose.
4 — Force-field training / DFT-only blocks. N/A — this work applies a published ReaxFF and cites DFT for select energetic checks rather than presenting a new FF fit or standalone static-QM study.
Findings¶
Outcomes / mechanisms: GA, HRTEM, and MD consistently indicate sulfur vacancies organize into extended lines as favored motifs. Electron-beam exposure localizes 2H → 1T transformations near these line defects; MD shows how local bond rearrangements nucleate 1T domains, with temperature and defect density influencing finite 1T fractions.
Comparisons: simulation motifs are compared to in situ micrographs and to DFT energy ordering cited in the article.
Sensitivity: defect density and temperature (up to 1500 K in MD) modulate 1T stabilization.
Limitations / outlook: ReaxFF remains approximate versus DFT for phase energetics; long-time defect diffusion may need additional sampling strategies beyond the reported MD segments.
Corpus honesty: large-cell dimensions and GA parameters are taken from the ACS Nano main text; timestep must be read from SI if not visible in your local PDF export.
Limitations¶
ReaxFF accuracy for MoS2 phase energetics is approximate relative to DFT; long-time defect evolution may require enhanced sampling beyond the MD segments reported.
Relevance to group¶
Uses the MoS2 ReaxFF line developed in work connected to van Duin-group parametrization (cited ReaxFF MoS2 reference).