Descriptor‐driven cross‐scale prediction of nanosheet network evolution in multicomponent layered double hydroxides

Abstract

The mesoscopic morphology of multicomponent layered double hydroxides (LDHs) is closely related to local structural disorder, interlayer coupling, and nanosheet assembly. However, how atomic‐scale disorder is transferred into mesoscale nanosheet‐network formation remains unclear. Here, we develop a descriptor‐driven cross‐scale framework assisted by a pretrained machine‐learning potential to predict mesoscopic LDH morphology. The framework integrates CHGNet‐based atomistic relaxation, structural descriptor extraction, descriptor‐to‐parameter mapping, and a substrate‐rooted, facet‐resolved growth model. The extracted descriptors quantify bond‐length distortion, mixing thermodynamics, interlayer sliding/exfoliation, interlayer stiffness, and edge migration. Increasing compositional complexity enhances local structural disorder and modifies interlayer and edge‐growth dynamics, leading to a relative transition from ordered two‐dimensional nanosheet stacking to open three‐dimensional interlocked networks. The normalized SEM edge‐length density increases from 1.000 for Control to 1.262 for HE5, indicating higher nanosheet boundary complexity and stronger network interweaving.