Large‐Scale Cooperative Sulfur Vacancy Dynamics in Two‐Dimensional Mos2 From Machine Learning Interatomic Potentials

Why It Matters

Understanding and controlling vacancy dynamics unlocks faster design of MoS2‑based catalysts and next‑generation memory devices, giving firms a competitive edge in advanced materials markets.

Key Takeaways

• MLIPs enable nanosecond MD of MoS2 vacancies.
• Two approaches: GAP on‑the‑fly and equivariant model.
• Simulations reproduce experimental line defect formation.
• Vacancy clustering drives catalytic and memristive properties.
• Cooperative transport accelerates defect growth beyond diffusion limits.

Pulse Analysis

Machine‑learning interatomic potentials (MLIPs) are reshaping computational materials science by bridging the accuracy of quantum‑mechanical methods with the speed required for large‑scale simulations. Techniques such as Gaussian approximation potentials and equivariant neural networks learn from first‑principles data, allowing researchers to run nanosecond molecular dynamics on systems containing millions of atoms—orders of magnitude beyond traditional density functional theory. This scalability opens new avenues for exploring defect behaviour in two‑dimensional crystals where experimental observation is challenging.

In MoS2 monolayers, sulfur vacancies act as active sites for catalysis and underpin memristive switching. The study reveals that vacancies do not migrate independently; instead, they exhibit cooperative transport, readily joining clusters of any size and forming extended line defects. These line defects, observed experimentally after electron‑beam irradiation, emerge naturally in the simulations, providing a coherent atomistic explanation for the patterns that enhance catalytic performance and enable resistive memory functions. By capturing both the energetics and kinetics of vacancy aggregation, the MLIP models deliver insights that were previously inaccessible.

For industry, the ability to predict defect evolution in real time accelerates the development cycle of functional 2D materials. Companies can now screen MoS2‑based catalysts for hydrogen evolution or design memristive devices with tailored switching characteristics without costly trial‑and‑error labs. Investment in ML‑driven simulation platforms thus translates into shorter time‑to‑market, reduced R&D expenditure, and stronger intellectual property portfolios in the rapidly expanding quantum‑materials sector.