An inversion problem for optical spectrum data via physics-guided machine learning.

Area of Science:

• Computational Physics
• Materials Science
• Machine Learning

Background:

• Deriving the pairing glue function is crucial for understanding material properties.
• Experimental optical spectra present challenges for accurate data analysis.
• Inverse problems, like this one, are often ill-posed and require robust methods.

Purpose of the Study:

• To introduce a novel machine learning approach, the regularized recurrent inference machine (rRIM).
• To address the challenge of deriving pairing glue functions from optical spectra.
• To develop a method robust to noise and flexible with data variations.

Main Methods:

• The regularized recurrent inference machine (rRIM) was developed as a machine learning model.
• Physical principles were integrated into both the training and inference stages of the rRIM.
• The rRIM was trained and tested using measured optical spectra data.

Main Results:

• The rRIM successfully derived reliable pairing glue functions from experimental optical spectra.
• The method demonstrated robustness against noise in the spectral data.
• The rRIM showed flexibility in handling out-of-distribution data and reduced data requirements.

Conclusions:

• The regularized recurrent inference machine (rRIM) offers a powerful new tool for analyzing optical spectra.
• This approach provides promising solutions for inverse problems involving Fredholm integral equations of the first kind.
• The rRIM's ability to incorporate physical principles enhances its applicability in scientific discovery.