Wu, Di, Liang, Ling, Yang, Haizhao · Oct 29, 2026
arXiv cs.AI lists FARMS: A Verifiable Foundation for Automated Computational Science
Markdown · JSON · Original source
Di Wu, Ling Liang and Haizhao Yang present FARMS, a system intended to make mathematical knowledge usable by AI systems with explicit guarantees. They argue that automated computational science could speed discovery by selecting, adapting and developing numerical methods, but that this requires knowledge AI can interpret and reuse. FARMS links formal statements to their source context, algorithmic meaning and quantitative guarantees through source preparation, reusable representations and semantic review. In tests covering 171 targets in randomized numerical linear algebra and numerical partial differential equations, the pipeline accepted 157; the authors report mean estimated costs of US $0.85–2.29 per target across sources. They also describe -checked implications from explicit source premises connecting selected statements to computational skills, with a heat-flow solve and wave-response compression as examples. The authors present these results as support for an affordable, reusable knowledge foundation for trustworthy automated computational science.
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: Oct 29, 2026 (2026-10-29).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-11T07:39:25.164Z. First observed: 2026-10-30T06:41:15.316Z. Retrieved: 2026-10-31T07:17:50.107Z.
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 publisher page's abstract, which does not provide details of the pipeline's evaluation methods or how the estimated costs were calculated. The reported targets cover randomized numerical linear algebra and numerical partial differential equations; the abstract does not report performance on other areas of computational science. The abstract reports examples and selected statements, but does not specify how broadly the reported guarantees apply.
Open question
How does FARMS perform on kinds of computational science beyond the randomized numerical linear algebra and numerical partial differential equations targets described in the abstract?
Source-stated
The authors present FARMS as a foundation for making computational mathematics usable by AI systems with explicit guarantees.
FARMS, a faithful and efficient foundation that transforms computational mathematics into AI-ready computing knowledge.
Source-stated
FARMS connects formal statements to source context, algorithmic meaning and quantitative guarantees using three described stages.
through source preparation, reusable representations and semantic review.
Source-stated
The pipeline accepted 157 of 171 targets in two named research areas.
Across 171 targets in randomized numerical linear algebra and numerical partial differential equations, 157 were accepted by the pipeline
Source-stated
The authors report mean estimated costs of US $0.85–2.29 per target across sources.
with mean estimated costs of US \$0.85-2.29 per target across sources.
Source-stated
The authors say Lean-checked implications from explicit source premises connect selected statements to computational skills, illustrated with two examples.
Lean-checked implications from explicit source premises connect selected statements to computational skills, illustrated by a heat-flow solve and wave-response compression.
Source and contributor context
Wu, Di, Liang, Ling, Yang, Haizhao · 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.
arXiv
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Lean
Lean is a programming language and proof assistant used to express mathematical statements and check formal proofs.
Editorial provenance
Version 1. Publication ID: 3ee39139-5259-4d27-8c65-5884a3326010.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-11T07:39:25.164Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: 1af54cf8609267ce7ad8f3f96d1b74ac2d1db1a64c21a0c2927272336291c822. Source text hash: 2698c700be1ab7031936cd9515d030bba0d9c31f9de76bbb92ad33fa8fd606cb.
Revision and source-check history
Version 1 · 2026-11-11T07:39:25.164Z
arXiv cs.AI lists FARMS: A Verifiable Foundation for Automated Computational Science
Di Wu, Ling Liang and Haizhao Yang present FARMS, a system intended to make mathematical knowledge usable by AI systems with explicit guarantees. They argue that automated computational science could speed discovery by selecting, adapting and developing numerical methods, but that this requires knowledge AI can interpret and reuse. FARMS links formal statements to their source context, algorithmic meaning and quantitative guarantees through source preparation, reusable representations and semantic review. In tests covering 171 targets in randomized numerical linear algebra and numerical partial differential equations, the pipeline accepted 157; the authors report mean estimated costs of US $0.85–2.29 per target across sources. They also describe -checked implications from explicit source premises connecting selected statements to computational skills, with a heat-flow solve and wave-response compression as examples. The authors present these results as support for an affordable, reusable knowledge foundation for trustworthy automated computational science.
Original source
cs.AI · Article
Source details
Original source date: Oct 29, 2026 (2026-10-29).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-11T07:39:25.164Z. First observed: 2026-10-30T06:41:15.316Z. Retrieved: 2026-10-31T07:17:50.107Z.
Limits
The evidence is the publisher page's abstract, which does not provide details of the pipeline's evaluation methods or how the estimated costs were calculated. The reported targets cover randomized numerical linear algebra and numerical partial differential equations; the abstract does not report performance on other areas of computational science. The abstract reports examples and selected statements, but does not specify how broadly the reported guarantees apply.
Open question
How does FARMS perform on kinds of computational science beyond the randomized numerical linear algebra and numerical partial differential equations targets described in the abstract?
Source-stated
The authors present FARMS as a foundation for making computational mathematics usable by AI systems with explicit guarantees.
FARMS, a faithful and efficient foundation that transforms computational mathematics into AI-ready computing knowledge.
Source-stated
FARMS connects formal statements to source context, algorithmic meaning and quantitative guarantees using three described stages.
through source preparation, reusable representations and semantic review.
Source-stated
The pipeline accepted 157 of 171 targets in two named research areas.
Across 171 targets in randomized numerical linear algebra and numerical partial differential equations, 157 were accepted by the pipeline
Source-stated
The authors report mean estimated costs of US $0.85–2.29 per target across sources.
with mean estimated costs of US \$0.85-2.29 per target across sources.
Source-stated
The authors say Lean-checked implications from explicit source premises connect selected statements to computational skills, illustrated with two examples.
Lean-checked implications from explicit source premises connect selected statements to computational skills, illustrated by a heat-flow solve and wave-response compression.
Source and contributor context
Wu, Di, Liang, Ling, Yang, Haizhao · 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.
arXiv
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Lean
Lean is a programming language and proof assistant used to express mathematical statements and check formal proofs.
Editorial provenance
Version 1. Publication ID: 3ee39139-5259-4d27-8c65-5884a3326010.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-11T07:39:25.164Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: 1af54cf8609267ce7ad8f3f96d1b74ac2d1db1a64c21a0c2927272336291c822. Source text hash: 2698c700be1ab7031936cd9515d030bba0d9c31f9de76bbb92ad33fa8fd606cb.
uncertain · 2026-11-11T07:39:25.164Z · anthropic/claude-opus-5.5
Publication: 3ee39139-5259-4d27-8c65-5884a3326010
attribution — pass: The reading correctly attributes FARMS and its claims to Di Wu, Ling Liang and Haizhao Yang. It keeps the results and the conclusion framed as the authors' report, for example 'the authors report' and 'the authors present these results as support'.
Authors:Di Wu, Ling Liang, Haizhao Yang
summary — uncertain: The summary is faithful, but central terms are left unexplained for a non-specialist. It never says what a 'target' is, so the headline result of 157 accepted out of 171 is hard to interpret. It also does not say that 'accepted by the pipeline' means passing FARMS's own review rather than independent validation. 'Lean-checked' carries the verifiability claim, yet Lean (a proof-checking system) is not explained. Smallest repair: one clause defining a target, plus tooltips or an added term-library entry for Lean. The question is a fair, open issue.
Across 171 targets in randomized numerical linear algebra and numerical partial differential equations, 157 were accepted by the pipeline
date — pass: originalPublished is 2026-10-29, which matches the v1 submission date on the arXiv page. The 2026-10-31 feed date is correctly not used.
[Submitted on 29 Oct 2026]
support — pass: Every claim matches the abstract verbatim in substance: the 157 of 171 acceptance figure, the US $0.85–2.29 mean estimated cost, and the Lean-checked implications with two examples. The limitations correctly note that only the abstract was available and that scope is limited. No benefits or costs are invented.
with mean estimated costs of US \$0.85-2.29 per target across sources.
media — pass: The selected media is the verified direct arXiv abstract page for this paper, which 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.
Lean is a programming language and proof assistant used to express mathematical statements and check formal proofs.
Explore LeanarXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivLean is a programming language and proof assistant used to express mathematical statements and check formal proofs.
Explore LeanarXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXiv