Chandrasekhar, Achuth, Farimani, Amir Barati · Nov 9, 2026
Agent-Guided Discovery of Exchange-Correlation Functionals with Physics-based Validation
Markdown · JSON · Original source
Achuth Chandrasekhar and Amir Barati Farimani report that language-model agents produced O159, a compact exchange-correlation functional, and tested it with self-consistent and external evaluation. O159 multiplies same-spin correlation by a smooth density window. Against a refitted no-window control, the authors report 29% lower heavy-atom total-energy errors on GSCDB137. On the held-out MGCDB84 test, self-consistent O159 lowers the error of ωB97M-V by 11.7%; across 36 GSCDB137 datasets, its mean normalized error ratio is comparable to ωB97M-V and below GAS22. The authors also report that Opus-5.5 agents passed the screening gate with about 40-fold fewer tokens than Codex (GPT-6-Astra) agents. In their comparison with evolutionary search, SyFES did not improve on its GAS22 seed, while the agents found forms with lower validation error by adding an explicit density scale. These results matter because the paper tests an agent-proposed functional beyond fitting: the reported improvement is evaluated after self-consistency and on held-out or external datasets. The evidence described in the abstract concerns particular functionals, controls and datasets; it does not establish performance beyond those evaluations.
Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.
Original source
cs.AI · Article
Source details
Original source date: Nov 9, 2026 (2026-11-09).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-12T01:47:00.062Z. First observed: 2026-11-11T05:06:26.469Z. Retrieved: 2026-11-12T01:46:04.522Z.
Referenced by
Recorded citations and public file attachments. A reference alone does not establish agreement or influence.
No public references recorded here yet.
Limits
The evidence is the paper's abstract; the full paper's methods and results are not included here. Only one candidate advanced to self-consistent evaluation: “only this candidate advanced to self-consistent evaluation.” The reported results concern the named functionals, controls and datasets; the abstract does not report evaluations beyond these.
Open question
How does O159 perform on self-consistent and external evaluations beyond the MGCDB84 and GSCDB137 results reported here?
Source-stated
The authors report that frontier LLM agents produced O159, a 13-parameter functional, and evaluated it self-consistently and externally.
We connect frontier LLM agents to a GPU evaluator through the Model Context Protocol, with training-only fitting, persistent memory, and self-consistent and external evaluation.
Source-stated
O159 multiplies same-spin correlation by a smooth density window.
The resulting 13-parameter functional, O159, multiplies same-spin correlation by a smooth density window.
Source-stated
Against a refitted no-window control, O159 lowers heavy-atom total-energy errors on GSCDB137 by 29%.
against a refitted no-window control, it lowers heavy-atom total-energy errors on GSCDB137 by 29\%.
Source-stated
Self-consistent O159 lowers the held-out MGCDB84 test error of ωB97M-V by 11.7%.
Self-consistent O159 lowers the held-out MGCDB84 test error of $\omega$B97M-V by 11.7\%
Source-stated
Across 36 GSCDB137 datasets, O159's mean normalized error ratio is comparable to ωB97M-V and below GAS22.
its mean normalized error ratio on 36 GSCDB137 datasets (1.139) is comparable to $\omega$B97M-V (1.144) and below GAS22 (1.190).
Source-stated
Opus-5.5 agents passed the screening gate with about 40-fold fewer tokens than Codex (GPT-6-Astra) agents.
Opus-5.5 agents passed the screening gate with about 40-fold fewer tokens than Codex (GPT-6-Astra) agents
Source-stated
In the reported comparison, SyFES evolution did not improve on its GAS22 seed, while agents found forms with lower validation error by adding an explicit density scale.
SyFES evolution did not improve on its GAS22 seed, whereas the agents found forms 0.10--0.13 \kcal\ lower in validation error by adding an explicit density scale.
Source and contributor context
Chandrasekhar, Achuth, Farimani, Amir Barati · Article author
These are credits stated by the source; they do not establish a profile match or authorship of every passage.
Source media
No suitable media was found by the supported page extraction.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Benchmark
A defined set of tests used to compare systems. Results depend on the tasks and measures selected.
arXiv
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Editorial provenance
Version 1. Publication ID: 4ccb0f96-4f1d-44be-807f-ba166a8592cf.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-12T01:47:00.062Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: 3ef1cd78d5dabe7d03e3cb07677096e1150c810ba065b3d988378b698c23566f. Source text hash: 257f95799e932b1860f956572ff8e1860f86eb674aa52ab7710ddb0d206245b5.
Revision and source-check history
Version 1 · 2026-11-12T01:47:00.062Z
Agent-Guided Discovery of Exchange-Correlation Functionals with Physics-based Validation
Achuth Chandrasekhar and Amir Barati Farimani report that language-model agents produced O159, a compact exchange-correlation functional, and tested it with self-consistent and external evaluation. O159 multiplies same-spin correlation by a smooth density window. Against a refitted no-window control, the authors report 29% lower heavy-atom total-energy errors on GSCDB137. On the held-out MGCDB84 test, self-consistent O159 lowers the error of ωB97M-V by 11.7%; across 36 GSCDB137 datasets, its mean normalized error ratio is comparable to ωB97M-V and below GAS22. The authors also report that Opus-5.5 agents passed the screening gate with about 40-fold fewer tokens than Codex (GPT-6-Astra) agents. In their comparison with evolutionary search, SyFES did not improve on its GAS22 seed, while the agents found forms with lower validation error by adding an explicit density scale. These results matter because the paper tests an agent-proposed functional beyond fitting: the reported improvement is evaluated after self-consistency and on held-out or external datasets. The evidence described in the abstract concerns particular functionals, controls and datasets; it does not establish performance beyond those evaluations.
Original source
cs.AI · Article
Source details
Original source date: Nov 9, 2026 (2026-11-09).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-12T01:47:00.062Z. First observed: 2026-11-11T05:06:26.469Z. Retrieved: 2026-11-12T01:46:04.522Z.
Limits
The evidence is the paper's abstract; the full paper's methods and results are not included here. Only one candidate advanced to self-consistent evaluation: “only this candidate advanced to self-consistent evaluation.” The reported results concern the named functionals, controls and datasets; the abstract does not report evaluations beyond these.
Open question
How does O159 perform on self-consistent and external evaluations beyond the MGCDB84 and GSCDB137 results reported here?
Source-stated
The authors report that frontier LLM agents produced O159, a 13-parameter functional, and evaluated it self-consistently and externally.
We connect frontier LLM agents to a GPU evaluator through the Model Context Protocol, with training-only fitting, persistent memory, and self-consistent and external evaluation.
Source-stated
O159 multiplies same-spin correlation by a smooth density window.
The resulting 13-parameter functional, O159, multiplies same-spin correlation by a smooth density window.
Source-stated
Against a refitted no-window control, O159 lowers heavy-atom total-energy errors on GSCDB137 by 29%.
against a refitted no-window control, it lowers heavy-atom total-energy errors on GSCDB137 by 29\%.
Source-stated
Self-consistent O159 lowers the held-out MGCDB84 test error of ωB97M-V by 11.7%.
Self-consistent O159 lowers the held-out MGCDB84 test error of $\omega$B97M-V by 11.7\%
Source-stated
Across 36 GSCDB137 datasets, O159's mean normalized error ratio is comparable to ωB97M-V and below GAS22.
its mean normalized error ratio on 36 GSCDB137 datasets (1.139) is comparable to $\omega$B97M-V (1.144) and below GAS22 (1.190).
Source-stated
Opus-5.5 agents passed the screening gate with about 40-fold fewer tokens than Codex (GPT-6-Astra) agents.
Opus-5.5 agents passed the screening gate with about 40-fold fewer tokens than Codex (GPT-6-Astra) agents
Source-stated
In the reported comparison, SyFES evolution did not improve on its GAS22 seed, while agents found forms with lower validation error by adding an explicit density scale.
SyFES evolution did not improve on its GAS22 seed, whereas the agents found forms 0.10--0.13 \kcal\ lower in validation error by adding an explicit density scale.
Source and contributor context
Chandrasekhar, Achuth, Farimani, Amir Barati · Article author
These are credits stated by the source; they do not establish a profile match or authorship of every passage.
Source media
No suitable media was found by the supported page extraction.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Benchmark
A defined set of tests used to compare systems. Results depend on the tasks and measures selected.
arXiv
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Editorial provenance
Version 1. Publication ID: 4ccb0f96-4f1d-44be-807f-ba166a8592cf.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-12T01:47:00.062Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: 3ef1cd78d5dabe7d03e3cb07677096e1150c810ba065b3d988378b698c23566f. Source text hash: 257f95799e932b1860f956572ff8e1860f86eb674aa52ab7710ddb0d206245b5.
uncertain · 2026-11-12T01:47:00.062Z · anthropic/claude-opus-5.5
Publication: 4ccb0f96-4f1d-44be-807f-ba166a8592cf
attribution — pass: The reading correctly attributes the work to Achuth Chandrasekhar and Amir Barati Farimani on arXiv. It frames the results as author reports, including the O159 results, the token comparison and the SyFES comparison.
Authors:Achuth Chandrasekhar, Amir Barati Farimani
summary — uncertain: The figures are faithful, but a reader without specialist background cannot tell what an exchange-correlation functional is or why it matters. The draft never says these are approximations at the heart of density functional theory, which the abstract names as their purpose. Central terms are also left opaque: self-consistent evaluation, same-spin correlation, density window, screening gate and tokens. Smallest repair: one plain sentence that these functionals are formulas used in density functional theory, a method for predicting molecular energies, plus brief glosses or tooltips for the other terms. The question and title are acceptable.
Accurate, compact exchange--correlation functionals are central to predictive density functional theory.
date — pass: The displayed original publication date of 2026-11-09 matches the arXiv v1 submission date. It is not the later feed date.
[Submitted on 9 Nov 2026]
support — pass: Each claim matches the abstract. The significance sentence correctly ties the improvement to self-consistent and held-out evaluation without overstating real-world reliability, and the limitations preserve scope. Minor nuance: the 29% figure is framed in the abstract around the window versus a refitted no-window control. The draft attributes it to O159, which is a reasonable reading of the abstract's wording and not a contradiction.
against a refitted no-window control, it lowers heavy-atom total-energy errors on GSCDB137 by 29\%.
media — pass: The selected media is the verified direct arXiv abstract page for this paper. This is the most direct access to the source.
Challenge or correct this reading
Send a correction or an opposing source with this reading’s URL. Disagreement and uncertainty can remain unresolved.
A defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Definition referencearXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivA defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Definition referenceA defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Definition referencearXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivA defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Definition reference