Machine learning the frontier orbital energies of SubPc based triads

Area of Science:

• Materials Science
• Computational Chemistry
• Renewable Energy

Background:

• Organic photovoltaic (OPV) devices offer efficient solar energy conversion.
• Designing novel dye molecules for OPVs is computationally intensive.
• Molecular orbital energies are critical for OPV material properties.

Purpose of the Study:

• Develop an efficient screening method for dye molecules in OPVs.
• Utilize machine learning to predict molecular orbital energies.
• Identify new boron-Subphthalocyanine dyes with improved electronic properties.

Main Methods:

• Generated a database of 12,102 PM6 optimized structures.
• Performed time-dependent density functional theory calculations.
• Employed machine learning algorithms (SVM, RF, NN, SLR) with Electrotopological-state index representation.

Main Results:

• Machine learning models accurately predicted frontier orbital energies (RMSE ~0.05 eV).
• Successfully predicted energies for over 40,000 new structures.
• Identified 237 (ESI) and 132 (OneHot) dyes with upshifted orbital energies compared to the parent structure.
• Acetamide and hydroxyl ligands were found to increase frontier molecular orbital energy.

Conclusions:

• Machine learning provides an effective strategy for rapid screening of OPV dye candidates.
• The developed method significantly reduces the computational cost of identifying high-performance dyes.
• This work paves the way for accelerated discovery of advanced organic solar cell materials.