{"paper":{"title":"Tensor Channel Equivariant Graph Neural Networks for Molecular Polarizability Prediction","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Propagating explicit tensor channels through message passing improves molecular polarizability tensor predictions over readout-only baselines.","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Daniel Franzen, Jean Philip Filling, Michael Wand","submitted_at":"2026-05-16T09:07:07Z","abstract_excerpt":"We introduce a tensor-channel equivariant graph neural network for direct prediction of molecular polarizability tensors. Building on the efficient PaiNN architecture, we augment the hidden representation with explicit symmetric rank-2 tensor channels aligned with the decomposition of polarizability into isotropic and anisotropic components. In contrast to approaches that construct tensor outputs only at readout, our model propagates tensor structure throughout message passing using geometrically motivated tensor bases. This yields a target-aligned architecture for tensor-valued molecular pred"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"On optimized QM7-X geometries the proposed model achieves lower full-tensor and anisotropic error than both a PaiNN-style readout baseline and a dielectric MACE baseline under matched training conditions and at nearly identical parameter count.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The performance gain is attributable to explicit tensor propagation and traceless target parameterization rather than to differences in optimization dynamics or data preprocessing that were not fully controlled in the reported experiments (abstract, paragraph on ablation studies).","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"A tensor-channel equivariant GNN based on PaiNN propagates symmetric rank-2 tensor features during message passing and achieves lower full-tensor and anisotropic error than readout-only and MACE baselines on QM7-X geometries.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Propagating explicit tensor channels through message passing improves molecular polarizability tensor predictions over readout-only baselines.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"bce641daf0837d9a2d986eb368a28deebd49c86c8ac271e4fce012d0cb0fae2c"},"source":{"id":"2605.16891","kind":"arxiv","version":1},"verdict":{"id":"2b6da77a-190b-42fa-9f48-e65cfd225063","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-19T20:30:07.786305Z","strongest_claim":"On optimized QM7-X geometries the proposed model achieves lower full-tensor and anisotropic error than both a PaiNN-style readout baseline and a dielectric MACE baseline under matched training conditions and at nearly identical parameter count.","one_line_summary":"A tensor-channel equivariant GNN based on PaiNN propagates symmetric rank-2 tensor features during message passing and achieves lower full-tensor and anisotropic error than readout-only and MACE baselines on QM7-X geometries.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The performance gain is attributable to explicit tensor propagation and traceless target parameterization rather than to differences in optimization dynamics or data preprocessing that were not fully controlled in the reported experiments (abstract, paragraph on ablation studies).","pith_extraction_headline":"Propagating explicit tensor channels through message passing improves molecular polarizability tensor predictions over readout-only baselines."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.16891/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"doi_title_agreement","ran_at":"2026-05-19T21:01:19.188630Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"cited_work_retraction","ran_at":"2026-05-19T20:52:41.720216Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T20:40:52.303979Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"claim_evidence","ran_at":"2026-05-19T18:41:56.283190Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T18:33:26.361586Z","status":"skipped","version":"1.0.0","findings_count":0}],"snapshot_sha256":"74274160bab0356ca0cc3ec84ae35286d142df4703fb834767262b7609174428"},"references":{"count":21,"sample":[{"doi":"","year":2026,"title":"Mace-polar-1: A polarisable electrostatic foun- dation model for molecular chemistry","work_id":"e7ea9246-2cfb-4414-85a7-ebc45c0caafa","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2022,"title":"Advances in neural information processing systems35, 11423–11436 (2022)","work_id":"cfe0cf7f-9fc2-4616-9fc1-03d8071274df","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2022,"title":"Nature communications13(1), 2453 (2022)","work_id":"de90175b-9784-4047-bd0a-3de6183872b6","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2019,"title":"04615 12 Filling et al","work_id":"1ad7a100-fc88-44d7-a387-e226279083e8","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2016,"title":"S., & Welling, M., 2016,Group Equivariant Convolu- tional Networks, Proc","work_id":"473fade8-b16b-45a0-b764-68a0f3957588","ref_index":5,"cited_arxiv_id":"1602.07576","is_internal_anchor":true}],"resolved_work":21,"snapshot_sha256":"5030f49e836b9af972931f00e2b0966394e452305ade0946682c38afb67f1697","internal_anchors":2},"formal_canon":{"evidence_count":2,"snapshot_sha256":"eb455a6f7e2861c158c0b27c43265269cbe901478c599a6daf3854ccb35f685a"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}