Rectangular observations for finite matrix gradients: critical budgets and the weak endpoint
Citable version that has not completed AIRR screening. Not peer reviewed. Screening record
For a finite periodic lattice and arbitrary finite symmetric matrices acting on its actual forward gradient stack, we construct a coefficient-only partition into integer rectangles whose normalized component means restore coercivity with output dimension O_D(h(1+F_c(A))). The finite quadratic penalty may depend on the entire coefficient profile, including its negative spectral depth. Positive spectral replacement, a complete scalar balanced-ring BV and weighted-gap construction, and singleton refinement at affected starts and forward endpoints derive the result. A positive diagonal family shows worst-case product order hF_c(A) is necessary for common partitions retaining every component mean on the all-field domain. At the scalar inverse-D/2 weak endpoint, finite layer cake gives a logarithmic upper bound, and explicit actual flat coexact fields force matching worst-case logarithmic dimension for centered differentiable observations with locally quadratic upper penalties. All proofs are included. The mechanisms are classical; historical priority and exact non-equivalence to classical approximation theory remain unestablished. No interacting/covariant producer, variable-weight curl identity, continuum theorem or physical mass gap follows. This exact20-page working-paper candidate has one numerical AI assessment and complete original reports, objections and context.
Preview PDF on this page v14
Uses your browser's PDF viewer. Open the PDF directly if a preview is unavailable.
Cite this paper
Lluis Eriksson (2026). Rectangular observations for finite matrix gradients: critical budgets and the weak endpoint. AIRR.SCIENCE, ARR-2026-7EPM3V3DP796BBRR, v14. https://airr.science/papers/ARR-2026-7EPM3V3DP796BBRR/versions/v14/ Research preprint; ARR-2026-7EPM3V3DP796BBRR v14. Not peer reviewed by AIRR. Working paper; not admitted to the AIRR accepted collection.
Use this version link to cite the manuscript you read. A Working paper label means the version has not completed AIRR screening. AIRR screening is not journal peer review. Import BibTeX, RIS or CSL JSON into your reference manager to apply a citation style.
Version history
The paper identifier remains stable. Each version has its own immutable release, timestamp and version identifier.
- v15 · major revision · source snapshot available
- v14 · major revision · source snapshot available · viewing
- v13 · major revision · source snapshot available
- v12 · major revision · source snapshot available
- v11 · major revision · source snapshot available
- v10 · major revision · source snapshot available
- v9 · major revision · source snapshot available
- v8 · major revision · source snapshot available
- v7 · major revision · source snapshot available
- v6 · major revision · source snapshot available
- v5 · major revision · source snapshot available
- v4 · major revision · source snapshot available
- v3 · major revision · source snapshot available
- v2 · major revision · source snapshot available
- v1 · source snapshot available
Revision statement
Major revision. Complete matrix rectangular construction225, common-full-vector product-rank lower227, and weak-critical coexact logarithmic obstruction224, with full scalar proof appendix.
Verification record
- Frontier-model screening
- Not assessed
- Source integrity
- Pass
- Bibliographic integrity
- Not assessed
- Reproducibility
- Not assessed
- Lean 4
- Not assessed
Recorded under ERIKSSON-EXACT-ARTIFACT-20261031. AIRR verification and screening are not peer review.
AI assistance statement
Codex/OpenAI assisted drafting and source-role audits; exposed as based on GPT-6, exact backend/effort unknown. The first exact20-page226C numerical assessment gives scientific importance3.25/10 and3 overall stars; criteria4/4/2/3/3,minor_revision,no main-invalidating objection identified. Full original report/context and all minor objections are retained unchanged. Fresh review context requested without prior numerical reports; same account/system/filesystem and unverified memory/provider-history controls mean independence unknown, not blind/independent/human peer review. Complete226B source rejection for an optional negative-M error is preserved, correctedC separately source-audited. No numerical regrading, inherited score or scientific/formal replay by reviewer. Working-paper candidate only; no acceptance/publication asserted. Proposedcoexact228extension is separate and absent from this PDF.
Frontier-model screening
Status: not_assessed. Any listed reports correspond to this exact version under ARR-SCREEN-1.0; no absent assessment is represented as a pass.
Model assessments and review context
- Distinct models
- 1
- Range
- 3.25–3.25
- Method
- bounded capability-weighted mean · exact version only
Limited evidence · 0/1 benchmark matches. This is an evidence description, not a probability of correctness.
Only one distinct model; One provider; shared blind spots possible; Fresh blinded review context not documented for every model; Benchmark match incomplete; Adverse reports or unresolved objections require editorial disposition
OpenAI · Codex, exposed as based on GPT-6; exact backend and effort unknown · unknown · review context: unknown3.25
Recommendation: Minor Revision · Material objections: 0
The exact 20-page PDF gives coherent finite coefficient-selected rectangular observation bounds, a common-full-vector product obstruction, and a coexact weak-endpoint logarithmic lower family. Full proof reading found no main-invalidating defect. The value is a precise finite synthesis rather than a new general spectral principle. Minor revision should sharpen the comparison with classical partition algorithms, improve the dependency guide, and clean up presentation. Scientific importance is 3.25/10; overall_stars is its rounded value, not a separate quality score.
- Correctness confidence
- Full text proof reading and adversarial boundary checks found no main-invalidating gap; no formal or numerical replay was performed.
- Rigor
- Actual-start weights, complete endpoint correction, ring interfaces and lower-bound domains are explicit; scalar BV and tree proofs are dense but coherent.
- Novelty
- Classical ring/tree, spectral replacement and rank mechanisms dominate. Exact finite/coexact combinations are candidates; priority and non-equivalence remain unestablished.
- Significance
- Substantive finite observation bounds with matching restricted product and weak-endpoint orders; no continuum or Millennium transfer is established.
- Reproducibility
- Self-contained finite proofs and an identified exact PDF support inspection. Historical diagnostics, source identity and execution claims were not replayed.
Strengths
- Complete scalar BV, weighted ring-gap and heavy-chain proof, retaining actual original-start weights and intra-ring interfaces.
- Matrix replacement charges the entire signed stacked correction and isolates every affected forward endpoint.
- Actual compact coexact low fields establish the logarithmic obstruction for the explicitly centered differentiable/local-quadratic class.
- The common-partition product lower bound clearly distinguishes genuine full-domain row rank from its one-component low restriction.
Weaknesses
- Historical priority and exact non-equivalence to classical approximation and tree-partition theorems remain unestablished.
- Constants are very large and no practical implementation, sharp rank or uniform negative-depth penalty is supplied.
- Proof dependencies are dense; unexplained programme shorthand and an almost empty final reference page distract from the mathematics.
Potential errors
- No new actual counterexample was found. The corrected nonnegative spectral-depth choice avoids the auxiliary negative-M error described in the manuscript.
- Historical diagnostic and source-audit claims inside Section 5 were not independently checked and must not be treated as verified by this review.
Strong novelty candidates
- The precise finite coefficient-only rectangular synthesis for arbitrary local symmetric stacked-gradient matrices, with capped budget and full endpoint correction.
- The explicit actual-mean weak-critical family of flat coexact fields forcing logarithmic dimension for all observations in the stated local class.
Unresolved material objections
None reported.
arr:assessment:6554671e-ffcf-4ac9-bf5d-2dc754281df2 · prompt ARR-ASSESS-1.0 · response SHA-256 6a1151d67b8df056001b1243dcb13e8c4419564f1bccca57d0d7c9378aa3a609 · canonical PDF SHA-256 28cb2a7502e05cd8588aa6294f07a7eca5e0e452cce1bf5e4f09e9ec3c89e1a6
OpenAI · Codex, exposed as based on GPT-6; exact backend and effort unknown
Original AI-only score: 3.25 / 10 · Scientific importance 0-10; native rounded stars 1-10; five criteria 1-5
minor_revision
The exact 20-page PDF gives coherent finite coefficient-selected rectangular observation bounds, a common-full-vector product obstruction, and a coexact weak-endpoint logarithmic lower family. Full proof reading found no main-invalidating defect. The value is a precise finite synthesis rather than a new general spectral principle. Minor revision should sharpen the comparison with classical partition algorithms, improve the dependency guide, and clean up presentation. Scientific importance is 3.25/10; overall_stars is its rounded value, not a separate quality score.
Backend identity: unknown. First exact 20-page 226C assessment. Independence, exact served backend and effort unknown. No blind/independent/human review or scientific/formal replay. Complete original reports unchanged; bound-model-response adds only the documented version UUID. Native authority is evidenced separately.
Report captured · Exact PDF SHA-256 28cb2a7502e05cd8588aa6294f07a7eca5e0e452cce1bf5e4f09e9ec3c89e1a6
- report-hashes.json · SHA-256
d75ab088c5cdf6c3485cc355ab52734a18da9b16ced71d9988d0d5a6b0238153 - response.json · SHA-256
0846434a3d04a32fa0b125d2e0b99ac4b94808e27e56c080ec270324345767b1 - review-context.json · SHA-256
2d1fe32e616fe9528c798e5d2a8385295c90a0a7304a070635d9e1793907892a - scientific-review.txt · SHA-256
bd00e31013a1bdbf1f2766b25665af17ab7792aae8fcdff3c50670393a1d7272
A model assessment is not peer review or a correctness certificate. AIRR preserves disagreement, exact-version provenance and later reassessments.
Editorial disclosure
Working paper only, not editorial acceptance or human peer review. Actual eligible clean linked native receipt/readback and exact-artifact permission checked. First exact-PDF AI scientific importance 3.25/10, native 3 stars, criteria 4/4/2/3/3, minor_revision. Full objections, source-B rejection, corrected-C audit and author response retained; independence/backend/effort unknown. No physical or continuum theorem.