Equivariant Graph Neural Networks Improve Optical Spectra Prediction for Materials Screening
For materials scientists screening optoelectronic materials, this provides more accurate surrogate models for optical spectra, enabling higher-throughput discovery.
Equivariant graph neural networks (GotenNet) were adapted to predict optical spectra from 10,533 structures at the RPA level, outperforming state-of-the-art models with largest gains in the 0-8 eV range and static real permittivity prediction.
Scalable prediction of optical spectra is a critical component of high-throughput materials screening for optoelectronic applications such as solar cells. Existing surrogate models are trained on spectra computed from lower levels of theory or rely on rotation-invariant scalar features, limiting their geometric expressiveness. We explore the use of equivariant graph neural networks for optical spectra prediction, adapting GotenNet to this task and evaluating it on multiple datasets including a recently published collection of 10,533 structures with spectra computed at the level of the random phase approximation (RPA). The proposed model outperforms the current state of the art, with the largest gains in the 0-8 eV range and on predicting the static real permittivity, both of particular relevance for thin-film optics.