Periodic-TDL: Topological Deep Learning for Polymer Design
A new deep learning framework, Periodic-TDL, uses periodic Vietoris-Rips complexes and hierarchical simplicial message-passing to capture many-body interactions across spatial scales in polymers. It outperforms state-of-the-art models on electronic, optical, physical, and thermal property prediction tasks. The method addresses limitations of graph-based approaches that miss periodicity and higher-order interactions. The work is published on arXiv (2605.26833) and targets applications in energy, healthcare, and materials science.
Key facts
- Periodic-TDL uses periodic Vietoris-Rips complexes
- HSMP encoder propagates information from long-range interactions to covalent bonds
- Outperforms all state-of-the-art models on polymer property prediction
- Tasks include electronic, optical, physical, and thermal targets
- Addresses missing periodicity and many-body interactions in polymer graphs
- Published on arXiv with ID 2605.26833
- Targets applications in energy, healthcare, and materials science
- Ester-to-amide validation is mentioned but details are cut off
Entities
Institutions
- arXiv