Reconstruction of the interatomic forces from dynamic scanning transmission electron microscopy data

Abstract

We explore the possibility for reconstruction of the generative physical models describing interactions between atomic units in solids from observational electron microscopy data. Here, scanning transmission electron microscopy (STEM) is used to observe the dynamic motion of Si atoms at the edge of monolayer graphene under continuous electron beam illumination. The resulting time-lapsed STEM images represent the snapshots of observed chemical states of the system. We use two approaches: potential of mean force calculation using a radial distribution function and a direct fitting of the graphene–Si interatomic pairwise potentials with force matching, to reconstruct the force fields in the materials. These studies lay the foundation for quantitative analysis of materials energetics from STEM data through the sampling of the metastable states in the chemical space of the system.

Roadmapping Report: Focused Research Organizations in Atomistic Simulation

“…At the other extreme, some advanced spectroscopy techniques can potentially resolve atomic motions at short timescales, but these methods are relatively new and challenging to implement [55][56][57][58] . Instead, there are two techniques that strike a balance between providing detailed data on atomic dynamics while being relatively costeffective.…”

Machine learning in scanning transmission electron microscopy

“…325,328,329 Beyond statistical analyses, observations of the multiple metastable configurations have been used to reconstruct the force fields acting between atoms. 330,331 Learning the generative model from atomically-resolved data, incorporating prior knowledge, and yielding corresponding uncertainties as posterior parameter distributions is therefore a clear opportunity for characterizing the intrinsic properties of material systems.…”

Machines for Materials and Materials for Machines: Metal-Insulator Transitions and Artificial Intelligence

“…Therefore, in some cases it might be difficult to identify the most relevant information from a given dataset and to efficiently extract it, particularly when studying an unknown compound. In this respect, ML has unlocked the ability to rapidly and automatically extract physically relevant features from STEM images [29], for example defects and dopants in graphene lattices [30][31][32][33][34][35][36], allowing access to a treasure-trove of information on the atomic scale physics and chemistry of materials.…”