Exploring Causal Physical Mechanisms via Non-Gaussian Linear Models and Deep Kernel Learning: Applications for...

Area of Science:

• Materials Science
• Physics
• Data Science

Background:

• Multimodal imaging in microscopy generates complex datasets.
• Understanding correlations and causal mechanisms in these datasets is challenging.
• Piezoresponse Force Microscopy (PFM) is a key technique for nanoscale material characterization.

Purpose of the Study:

• To develop an analysis framework for PFM data.
• To explore causal physical mechanisms underlying nanoscale material properties.
• To predict non-observed properties using experimental data and prior knowledge.

Main Methods:

• Linear causal analysis applied to PFM observables.
• Deep Kernel Learning (DKL) model to ascertain causal link strengths.
• Utilizing prior knowledge of material structure and domain morphology.

Main Results:

• Demonstrated linear causal analysis for PFM observables.
• Identified correlative relationships between morphology, piezoresponse, and elastic properties at the nanoscale.
• Successfully predicted material morphology and physical parameters using DKL.

Conclusions:

• The DKL framework enables high-fidelity reconstruction of functionalities and physical mechanisms.
• The analysis reveals mutual interactions between surface conditions and physical properties in ferroelectric materials.
• This universal framework can be extended to other multichannel datasets.