Learning radical excited states from sparse data

Area of Science:

• Materials Science
• Computational Chemistry
• Organic Electronics

Background:

• Emissive organic radicals are promising for advanced organic light-emitting diode (OLED) devices and molecular qubits.
• Simulating their optoelectronic properties is difficult due to spin-contamination and multiconfigurational excited states.

Purpose of the Study:

• To develop a data-driven approach for accurately learning the excited electronic states of organic radicals directly from experimental data.
• To overcome the challenges in simulating optoelectronic properties of organic radicals.

Main Methods:

• A data-driven approach using experimental excited state data to train a surrogate physical model (ExROPPP).
• Compilation of the largest known database of organic radical geometries and UV-vis data for model training.
• Utilizing a fast, spin-pure semiempirical method (ExROPPP) as the base for parameter optimization.

Main Results:

• The trained model achieved root mean square error of 0.24 eV and mean absolute error of 0.16 eV for excited state energies, significantly outperforming standard ExROPPP.
• The model demonstrated high accuracy on newly synthesized organic radicals, with even lower errors.
• The approach requires substantially less data than traditional Machine Learning methods.

Conclusions:

• This data-driven method enables accurate and efficient simulation of organic radical optoelectronic properties.
• It paves the way for high-throughput discovery of novel radical-based materials for next-generation optoelectronics.
• The findings offer a significant advancement in the computational study of organic radicals.