Predicting optical spectra for optoelectronic polymers using coarse-grained models and recurrent neural networks

Area of Science:

• Computational materials science
• Organic optoelectronics
• Spectroscopy

Background:

• Coarse-grained modeling is vital for studying organic optoelectronic materials.
• Ultraviolet-visible (UV-Vis) spectroscopy is a key experimental technique.
• A direct link between coarse-grained simulations and spectroscopy is currently lacking.

Purpose of the Study:

• To develop a method for predicting conjugated polymer spectra directly from coarse-grained models.
• To avoid computationally expensive back-mapping to atomistic representations.
• To establish a rigorous connection between simulation and experimental spectroscopy.

Main Methods:

• Utilized a generative deep-learning model, specifically a long-short-term memory recurrent neural network (LSTM-RNN).
• Leveraged the mathematical similarities between natural language processing and perturbative expansions of excited-state energies.
• Investigated the sensitivity of spectral predictions to the coarse-grained representation.

Main Results:

• Successfully predicted spectra directly from coarse-grained polymer structures.
• Demonstrated a method that bypasses the need for atomistic back-mapping and quantum chemistry calculations.
• Quantified the impact of coarse-grained representation choices on spectral predictions.

Conclusions:

• The developed LSTM-RNN approach provides a direct bridge between coarse-grained simulations and UV-Vis spectroscopy for conjugated polymers.
• This tool enhances post-simulation analysis and offers potential for refining coarse-grained potentials using spectral data.
• The method streamlines the investigation of organic optoelectronic material physics.