Speaker
Description
Experiments show that when a layer of hexagonal boron nitride (hBN) is grown on a curved platinum substrate, it induces a faceting process that creates stable facets, where hBN exhibits distinct electronic properties. Thus, the precise control of the curvature geometry in 2D materials can redefine this material family. However, the underlying mechanisms for faceting are not yet fully understood. To understand the curvature-induced effect, we first performed Density Functional Theory (DFT) calculations, and to capture the dynamics of the faceting mechanism, we employed an Atomic Cluster Expansion (MACE) interatomic potential. hBN adsorption is studied across nine Pt facets — Pt(111), Pt(443), Pt(553), Pt(221), Pt(331), Pt(441), Pt(881), Pt(991), and Pt(110) — using low-mismatch structural models constructed for each surface to minimize artificial strain effects. The DFT results reveal a consistent and physically meaningful trend. Experimentally stable facets bind hBN more strongly than their unstable counterparts. Pt(110) exhibits the strongest adsorption at 0.625 eV/BN, followed by Pt(221) at 0.373 eV/BN, Pt(881) at 0.285 eV/BN, and Pt(441) at 0.232 eV/BN. In contrast, the unstable facets bind more weakly: Pt(331) at 0.183 eV/BN, Pt(443) at 0.181 eV/BN, and Pt(553) at 0.160 eV/BN. Pt(991), at 0.216 eV/BN, represents a borderline case between the two groups. Structurally, hBN on unstable facets shows more pronounced bending near step-edge regions, indicating poorer registry with the underlying Pt surface. To move beyond static energetics, a MACE interatomic potential was trained on DFT energies and forces using an active-learning pipeline. Force root-mean-square error (RMSE) improved from 50.1 meV/Å in the baseline model to 29.3 meV/Å in the final model, while the energy RMSE converged to 6.6 meV/atom. The resulting potential performs reliably across near equilibrium hBN/Pt configurations. Overall, the results establish a consistent picture in which stronger hBN–Pt interaction correlates with experimental facet stability. Future work will extend the training dataset through an AiiDA-based active-learning workflow, incorporating height scans, lateral registry shifts, and atomic displacements across all facets. This will enable large-scale molecular dynamics (MD) simulations to directly capture the dynamic faceting mechanism.