Learning the diffusion of nanoparticles in liquid phase TEM via physics-informed generative AI.

Area of Science:

• Nanoscience and nanotechnology
• Soft matter physics
• Machine learning applications in science

Background:

• Studying nanoparticle motion in liquids offers insights into molecular-level interactions.
• Liquid phase transmission electron microscopy (LPTEM) captures real-time nanoparticle dynamics.
• Modeling nanoparticle motion is complex, especially without established physical equations.

Purpose of the Study:

• To present LEONARDO, a novel deep generative model for analyzing nanoparticle motion in LPTEM.
• To overcome challenges in linking observed motion to underlying molecular interactions.
• To interpret interactive forces without relying on closed-form Langevin equations.

Main Methods:

• Utilized a deep generative model with a physics-informed loss function.
• Employed an attention-based transformer architecture to learn stochastic motion.
• Applied the model to nanoparticle trajectories obtained from LPTEM.

Main Results:

• LEONARDO successfully learned the statistical properties of nanoparticle motion.
• The model captured signatures of environmental heterogeneity and viscoelasticity.
• Demonstrated the capability to model complex stochastic processes.

Conclusions:

• LEONARDO provides a powerful new approach for analyzing nanoparticle dynamics in complex fluids.
• The model facilitates a deeper understanding of nanoscale interactions and material properties.
• Physics-informed machine learning can address limitations in traditional modeling techniques.