Tensor Channel Equivariant Graph Neural Networks for Molecular Polarizability Prediction

This paper improves uponn the previous paper published at the NeurIPS '25 ML4PhysicalScience-Workshop by using overcomplete generating sets rather than a basis to represent local frames and does some fine-tuning of the tensor-representation, which improves performance.

Data and Resources

Additional Info

Field	Value
Publication(s)	Publication 1 Title of related publication Tensor Channel Equivariant Graph Neural Networks for Molecular Polarizability Prediction Year of related publication 2026 DOI of related publication (not DOI of data resources)
Funding	DFG Project No. 233630050-TRR146
Subproject	Project B7: Automated model building and representation learning for multiscale simulations
Cooperation partner(s)	Johannes Gutenberg University Mainz
Responsible Person's Name (PI)	Prof. Dr. Michael Wand
Responsible Person's Email for further data requests	wandm@uni-mainz.de
Responsible Person's Affiliation	Institute of Computer Science, Johannes Gutenberg University Mainz, Germany