Beam induced heating in electron microscopy modeled with machine learning interatomic potentials.

Area of Science:

• Materials Science
• Computational Physics
• Nanotechnology

Background:

• Electron microscopy is crucial for nanoparticle characterization.
• Understanding nanoparticle heating during imaging is essential for accurate analysis.
• Existing methods for estimating beam-induced heating are limited.

Purpose of the Study:

• To develop a combined theoretical and experimental method for estimating heating in metallic nanoparticles during electron microscopy.
• To model thermal transport using molecular dynamics and neural network potentials.
• To validate the method using experimental data from electron energy loss spectroscopy.

Main Methods:

• Molecular dynamics simulations coupled with equivariant neural network potentials trained on Density Functional Theory (DFT) calculations.
• Utilizing an ensemble of neural network potentials to estimate prediction errors.
• Employing electron energy loss spectroscopy (EELS) to measure electron beam energy deposition.

Main Results:

• A robust method for predicting nanoparticle heating as a function of size, shape, support material, and beam parameters.
• Demonstrated the utility of neural network ensembles for error estimation in simulations.
• Successfully combined theoretical modeling with experimental measurements for accurate heating quantification.

Conclusions:

• The developed method provides a powerful tool for understanding and predicting beam-induced heating in nanoparticles.
• This work advances the accuracy of electron microscopy analysis for nanomaterials.
• The approach is applicable to various nanoparticle systems and imaging conditions.