{"paper":{"title":"Accelerating 3D Non-LTE Synthesis with Graph Neural Networks","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Graph neural networks predict 3D non-LTE calcium populations in the solar atmosphere with correlations above 0.99 and million-fold speedups.","cross_cats":["astro-ph.IM"],"primary_cat":"astro-ph.SR","authors_text":"A. Asensio Ramos, A. Vicente Ar\\'evalo, C. J. D\\'iaz Baso","submitted_at":"2026-05-10T13:55:18Z","abstract_excerpt":"Spectropolarimetric interpretation of chromospheric lines requires solving the radiative transfer problem under non-local thermodynamic equilibrium (non-LTE) conditions. This means computing atomic-level populations self-consistently with the radiation field. While traditional inversion codes employ 1.5D approximations, they neglect horizontal radiative transfer, which can be significant near magnetic structures and in the chromosphere. We present a method to solve 3D atomic-level populations using Graph Neural Networks (GNNs), extending prior 1.5D work to the full 3D domain. By discretizing t"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"The trained GNN accurately predicts populations of the five-level Ca II atom plus continuum. Correlations exceed 0.99 in the photosphere and mid-chromosphere; errors in the upper chromosphere remain unbiased. Inference is ∼10^6 times faster than traditional iterative solvers. Spectral synthesis of the Ca II 8542 Å line yields intensity profiles with < 2 % mean residuals relative to the full 3D solution.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That a GNN trained on populations from a single Bifrost simulation snapshot will generalize accurately to other atmospheric conditions, and that the directed graph discretization with distance-based edges fully captures the essential radiative couplings without missing key non-local effects.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Graph neural networks can approximate full 3D non-LTE Ca II populations in solar models with correlations above 0.99 and extreme computational efficiency.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Graph neural networks predict 3D non-LTE calcium populations in the solar atmosphere with correlations above 0.99 and million-fold speedups.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"a2b0ec30d74c9f075adaa2ebe32f0bd8e7ee1625e22cf6c552fa3629a6306d31"},"source":{"id":"2605.09543","kind":"arxiv","version":1},"verdict":{"id":"29317321-505c-48c9-a1fd-fce8663e8c6e","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-12T04:28:33.243830Z","strongest_claim":"The trained GNN accurately predicts populations of the five-level Ca II atom plus continuum. Correlations exceed 0.99 in the photosphere and mid-chromosphere; errors in the upper chromosphere remain unbiased. Inference is ∼10^6 times faster than traditional iterative solvers. Spectral synthesis of the Ca II 8542 Å line yields intensity profiles with < 2 % mean residuals relative to the full 3D solution.","one_line_summary":"Graph neural networks can approximate full 3D non-LTE Ca II populations in solar models with correlations above 0.99 and extreme computational efficiency.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That a GNN trained on populations from a single Bifrost simulation snapshot will generalize accurately to other atmospheric conditions, and that the directed graph discretization with distance-based edges fully captures the essential radiative couplings without missing key non-local effects.","pith_extraction_headline":"Graph neural networks predict 3D non-LTE calcium populations in the solar atmosphere with correlations above 0.99 and million-fold speedups."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.09543/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"claim_evidence","ran_at":"2026-05-20T07:42:01.339362Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T16:40:24.243335Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-19T13:01:17.597884Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T10:09:19.429642Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"aa84e4293c351d97f7b4ba38e75dd25707de7e949baa0cf0cf4f5db95fdad519"},"references":{"count":16,"sample":[{"doi":"","year":2017,"title":"2017, A&A, 599, A133 Asensio Ramos, A., Trujillo Bueno, J., & Landi Degl’Innocenti, E","work_id":"674c9691-a167-4d2c-af75-add668cba306","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2018,"title":"Relational inductive biases, deep learning, and graph networks","work_id":"858410c0-7a66-4b27-b4e5-49aee9725be0","ref_index":2,"cited_arxiv_id":"1806.01261","is_internal_anchor":true},{"doi":"","year":2015,"title":"Carlsson, M., Hansteen, V . H., Gudiksen, B. V ., Leenaarts, J., & De Pontieu, B. 2015, Astronomy & Astrophysics, 585, A4","work_id":"8cdd8e73-cd0d-4e69-a093-068450a4ecf6","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2022,"title":"Chappell, B. A. & Pereira, T. M. D. 2022, A&A, 658, A182","work_id":"36f91005-f348-4460-a521-ad72636106c8","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2019,"title":"Cheung, M. C. M., Rempel, M., Chintzoglou, G., et al. 2019, Nature Astronomy, 3, 160 de la Cruz Rodríguez, J., Leenaarts, J., Danilovic, S., & Uitenbroek, H. 2019, A&A, 623, A74 Díaz Baso, C. J., Asen","work_id":"cf079b0d-171d-4a90-931e-a26c755420d3","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":16,"snapshot_sha256":"790a75f22b0016c0a4dff478a38df2056659534baf9361fbaf4aba0d6e233409","internal_anchors":2},"formal_canon":{"evidence_count":2,"snapshot_sha256":"c2dc8133b2cdfba588b4248a549f60abed55ba51f5843a3e3950bd4033f29697"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}