Data and Trained Model Weights of the "Accelerating dynamic polarizability calculations of organic molecules using equivariant graph neural networks" paper

Abstract

Data and trained model weights for "Accelerating dynamic polarizability calculations of organic molecules using equivariant graph neural networks". This record contains the datasets and the trained model weights used in the manuscript. The code is available on GitHub: https://github.com/aimat-lab/DynPolDetanet Contents: - QM9SPol.pt: a sampled subset of 5160 molecules from the QM9 dataset. Each molecule has TD-DFT dynamic polarizability tensors at 61 frequencies between 1.55 and 6.2 eV, with the corresponding UV-vis spectra.- HOPV.pt: 347 molecules from the Harvard Organic Photovoltaic dataset (HOPV15), with dynamic polarizability tensors and UV-vis spectra.- HOPV_241pol.pt: the HOPV15 data with the polarizability tensors interpolated to 241 frequencies over the same energy range, using the AAA algorithm.- QM9S_dynamic_polarizability.pth: trained weights of the QM9SPol model (61 frequencies, UV-vis input).- HOPV241_dynamic_polarizability.pth: trained weights of the HOPV15 model (241 frequencies, UV-vis input). Place the .pt files in the data/ folder and the .pth files in code/trained_param/dynamic-polarizability/ of the GitHub repository. The README there gives the full instructions. Please cite the manuscript when using these data or weights.