Equivariant Graph Neural Networks Improve Optical Spectra Prediction for Materials Screening
Researchers have developed equivariant graph neural networks (EGNNs) that significantly improve the prediction of optical spectra for materials screening. By adapting the GotenNet architecture, these EGNNs offer greater geometric expressiveness than previous models, which were limited by rotation-invariant scalar features. The new approach demonstrates superior performance, particularly in predicting spectra within the 0-8 eV range and static real permittivity, crucial for optoelectronic applications like solar cells. AI
IMPACT This research advances AI capabilities in materials science, potentially accelerating the discovery of new optoelectronic materials.