Designing strongly coupled polaritonic structures via statistical machine learning.

Area of Science:

• Photonics and Materials Science
• Quantum Optics
• Machine Learning Applications

Background:

• Strong coupling photonics is crucial for energy conversion and information processing.
• Designing strongly coupled polaritonic structures is challenging due to difficulties in measuring decay factors.
• Traditional methods struggle with scalable and effective design of these complex systems.

Purpose of the Study:

• To develop a novel hybrid machine learning (ML) framework for precise determination of strong-weak coupling transition boundaries.
• To enable efficient and large-scale design of strongly coupled polaritonic structures using sparse data.
• To overcome limitations of traditional methods in predicting coupling regimes.

Main Methods:

• Integration of physics-informed modeling with uncertainty quantification in a hybrid ML framework.
• Accurate determination of coupling transition boundaries even with inaccessible decay factors.
• Leveraging sparse data for efficient and scalable design of polaritonic structures.

Main Results:

• Achieved a computational speedup of approximately 10^4 times compared to conventional simulations.
• Successfully designed and experimentally validated a hexagonal boron nitride (hBN) polariton coupling structure exhibiting strong coupling.
• Demonstrated accurate determination of strong-weak coupling transition boundaries.

Conclusions:

• The proposed hybrid ML framework offers a generalizable optimization methodology for strongly coupled photonic devices.
• This approach facilitates efficient design of polaritonic structures for advanced applications.
• Opens new avenues for polariton-enhanced energy conversion and optical information modulation.