Bayesian Conavigation: Dynamic Designing of the Material Digital Twins via Active Learning

Area of Science:

• Materials Science
• Computational Science
• Physical Chemistry

Background:

• Scientific progress relies on theory and experiment, but this loop is often slow, especially for complex, high-dimensional systems.
• Integrating theoretical insights directly into automated experimental workflows is crucial for accelerating discovery.

Purpose of the Study:

• To develop a method for integrating theory into automated experiments in real-time, enabling faster scientific advancement.
• To minimize epistemic uncertainty in experimental object spaces through adaptive theoretical model updates.

Main Methods:

• A Bayesian conavigation approach is used to explore theoretical model space and guide experimentation.
• Concurrent development of surrogate models for both simulation and experimental domains.
• Adjustment of theoretical model control parameters to reduce uncertainty.

Main Results:

• Demonstrated a methodology for creating digital twins of material structures, combining surrogate and theoretical models.
• Successfully applied the approach to study functional responses in ferroelectric materials.
• The method enables on-the-fly theory updates within automated experimental setups.

Conclusions:

• The proposed theory-in-the-loop method significantly accelerates the scientific discovery process.
• This approach facilitates the creation of accurate digital twins for complex material systems.
• The methodology shows broad applicability across various scientific domains, including molecular and nanocluster systems.