{
  "schemaVersion": "1.6",
  "slug": "column-pivoted-qr-orthonormal-worst-case",
  "title": "How badly can column-pivoted QR choose rows of an orthonormal basis? The sharp stretched-exponential worst case",
  "shortTitle": "The sharp worst case of column-pivoted QR on orthonormal rows",
  "url": "https://evidencepress.org/releases/column-pivoted-qr-orthonormal-worst-case/",
  "oneLine": "Greedy row selection loses at most exp(C(H)√k) against the best possible, with an exact constant, not 2^k.",
  "abstract": "Many numerical methods keep k of the N coordinates of an N×k matrix Q with orthonormal columns, for example the interpolation points of the discrete empirical interpolation method (DEIM). The quality of a choice I is γ = ‖Q(I,:)⁻¹‖₂, the factor by which interpolation from the chosen coordinates can be worse than the best approximation; some choice always has γ ≤ √(k(N−k+1)). The rule used in practice is column-pivoted QR (QRCP) on Q*, whose classical guarantee (Drmač and Gugercin) grows like 2ᵏ. Chen, Liu, He and Dong gave the first answer to Problem 4.3 of Amsel et al.: QRCP can be worse than every polynomial multiple of the existence scale, and their construction already gives growth of order exp(c√k). This candidate determines the worst case on the logarithmic scale. If the coherence μ(Q), the largest row energy divided by the average, is at most H, every legal QRCP selection satisfies γ ≤ √(e(N−k+1)) exp(C(H)√k) with an explicit constant C(H): C(1) = 2 and C(2) ≈ 2.25. Matching examples show that C(H) cannot be lowered (at H = 1 under the smallest-label tie rule). Since μ(Q) ≤ N/k, the bound needs no assumption; without one, the worst case is exp(Θ(√(k log(N/k)))) once N ≥ k^(2+ε). An a posteriori bound from the observed pivots has an optimal coefficient. Limits: exact arithmetic and worst cases only; sharpness is on the logarithmic scale, with a factor polynomial in k remaining; the worst examples need N of order k³, and the regime N = O(k) is open. AI-assisted, unrefereed candidate; no external peer review, independent reproduction or formal verification is claimed.",
  "datePublished": "2026-11-16",
  "dateModified": "2026-11-16",
  "version": "0.1.0-candidate",
  "doi": "10.5281/zenodo.24278831",
  "doiUrl": "https://doi.org/10.5281/zenodo.24278831",
  "conceptDoi": "10.5281/zenodo.24278830",
  "pdfUrl": "https://github.com/ipitchford/column-pivoted-qr-orthonormal-worst-case/releases/download/v0.1.0-candidate/column-pivoted-qr-orthonormal-worst-case-0.1.0-candidate.pdf",
  "altPdfUrl": null,
  "altPdfLabel": null,
  "zenodoUrl": "https://zenodo.org/records/24278831",
  "repoUrl": "https://github.com/ipitchford/column-pivoted-qr-orthonormal-worst-case",
  "releaseUrl": "https://github.com/ipitchford/column-pivoted-qr-orthonormal-worst-case/releases/tag/v0.1.0-candidate",
  "markdownUrl": "https://evidencepress.org/releases/column-pivoted-qr-orthonormal-worst-case/index.md",
  "bibtexUrl": "https://evidencepress.org/releases/column-pivoted-qr-orthonormal-worst-case/cite.bib",
  "audioUrl": "https://evidencepress.org/assets/audio/column-pivoted-qr-orthonormal-worst-case.mp3",
  "imageUrl": "https://evidencepress.org/assets/og/column-pivoted-qr-orthonormal-worst-case.png?v=5e69c7f4e2",
  "coverArtUrl": "https://evidencepress.org/assets/art/column-pivoted-qr-orthonormal-worst-case.svg?v=9582afadd1",
  "media": [
    {
      "type": "audio",
      "url": "https://evidencepress.org/assets/audio/column-pivoted-qr-orthonormal-worst-case.mp3",
      "name": "Plain-English audio briefing",
      "description": "Communication aid; not additional research evidence.",
      "transcriptUrl": "https://evidencepress.org/assets/audio/column-pivoted-qr-orthonormal-worst-case.txt"
    }
  ],
  "authors": [
    "Anonymous"
  ],
  "license": "CC0-1.0",
  "status": "unrefereed-candidate",
  "verification": {
    "peerReviewed": false,
    "independentlyReproduced": false,
    "formallyVerified": false,
    "internallyReplayed": true,
    "detail": "Unrefereed AI-assisted mathematics candidate with written proofs and finite computational checks. Internal AI-assisted review in five reviewer roles, an adversarial review by a second AI model, a clean replay, one revision round with extensions (the coherence-free bound and the law in N and k), then a re-review and a proof audit of the new proofs by Claude verification passes; a second-model audit of the new proofs was not run because of a usage limit. External specialist review, independent reproduction and formal verification have not been established."
  },
  "assurance": [
    {
      "dimension": "availability",
      "label": "Availability and archiving",
      "question": "Is the evidence package publicly retrievable from an archive under a persistent identifier?",
      "state": "passed",
      "evidenceUrl": "https://zenodo.org/records/24278831",
      "note": "The versioned research package and paper are publicly archived on GitHub and Zenodo. Provider asset parity was checked against frozen local files; availability does not establish correctness or peer review."
    },
    {
      "dimension": "internalReplay",
      "label": "Internal replay",
      "question": "Does the producer’s own pipeline reproduce the stated result from the archived package?",
      "state": "passed",
      "note": "verify.py, check_extensions.py and fingerprints.py (also under python -O) and the figure script were rerun from fresh copies under CPython 3.12 with pinned versions; every output matched the recorded file apart from timestamps and timings, and a deliberately wrong constant (3 changed to 1 in inequality (7.5)) made check_extensions.py exit with failure. GitHub CI on Linux reruns the scripts under -O and compares the outputs under a declared policy, with the negative control (blocking)."
    },
    {
      "dimension": "independentRerun",
      "label": "Independent rerun",
      "question": "Has someone else run the supplied implementation and obtained the stated result?",
      "state": "not-assessed",
      "note": "No unaffiliated rerun has been established."
    },
    {
      "dimension": "independentReimplementation",
      "label": "Independent reimplementation",
      "question": "Has someone else reached the result from an independent implementation?",
      "state": "not-assessed",
      "note": "No unaffiliated reimplementation; verify.py factorises the completed examples with two independent QRCP implementations."
    },
    {
      "dimension": "formalVerification",
      "label": "Formal verification",
      "question": "Is a formalised statement machine-checked, and over which trusted base?",
      "state": "not-assessed",
      "note": "No proof-assistant formalisation."
    },
    {
      "dimension": "specialistReview",
      "label": "Specialist review",
      "question": "Has a domain specialist assessed the argument?",
      "state": "not-assessed",
      "note": "AI-assisted internal review is not external specialist review."
    },
    {
      "dimension": "editorialPeerReview",
      "label": "Editorial peer review",
      "question": "Has a journal or venue run peer review to a decision?",
      "state": "not-assessed",
      "note": "No external editorial peer review."
    },
    {
      "dimension": "dataEnvironmentReproducibility",
      "label": "Data and environment reproducibility",
      "question": "Are data and computational environment pinned well enough to rebuild?",
      "state": "passed",
      "note": "Pinned NumPy, SciPy, SymPy, mpmath and Matplotlib; outputs reproduced on the release platform under CPython 3.12.11 and on Linux CI under the declared policy (exact files byte-identical apart from an interpreter-version field; floating-point fields within relative 1e-9; rounding diagnostics below the scripts' own thresholds). Not a locked container."
    }
  ],
  "provenance": {
    "aiGenerated": true,
    "aiAssisted": true,
    "generatedBy": [
      "AI-assisted mathematical research",
      "AI-assisted internal review (five reviewer roles), adversarial review by a second AI model, re-review, proof audit of the new proofs by Claude verification passes, clean replays",
      "Claude Code: publication packaging and release"
    ],
    "humanRole": "Publication authority only; no human research contribution inferred.",
    "reviewStatus": "Internal AI-assisted review, not external peer review."
  },
  "attributionBoundary": null,
  "problem": null,
  "orderedProductAverage": null,
  "corrections": [],
  "publicCorrections": [],
  "publicCorrectionIndexes": null,
  "pageStructureVersion": "2.0",
  "pageStructureVariant": "canonical-theorem",
  "pageStructureWaivers": [],
  "recordMaturity": "current",
  "metadataProvenance": "Prepared from the final paper (revision of 16 November 2026 with the release DOI on the title page, 18 pages), its scripts and recorded outputs, the claim-to-output binding and the frozen release archive. The versioned GitHub and Zenodo research archives were checked against the frozen release assets on 16 November 2026.",
  "grandfatheredAtSchemaVersion": null,
  "keywords": [
    "column-pivoted QR",
    "discrete empirical interpolation",
    "DEIM",
    "coherence",
    "Kahan matrix",
    "triangular inverse",
    "subset selection",
    "worst-case analysis"
  ],
  "keyResults": [
    "Theorem 2.3 (proof): if the coherence is at most H, every legal QRCP selection, with any tie-breaking, satisfies γ ≤ √(e(N−k+1)) exp(C(H)√k), where C(H) = √ρ + 1/√ρ and ρ − log ρ = 1 + log H; C(1) = 2, C(2) ≈ 2.25 and C(H)²/log H → 1.",
    "Theorem 2.3 (proof): the constant is sharp on the logarithmic scale, log G_k(H) = (C(H) + o(1))√k, with real examples of fewer than 3k³ rows and unique pivots for H > 1, and with the smallest-label tie rule at H = 1.",
    "Corollary 2.6 (proof): every orthonormal input satisfies γ ≤ √(e(N−k+1)) exp(C(N/k)√k), so no coherence assumption is needed.",
    "Theorem 2.7 and Corollary 2.8 (proof): without any assumption the worst case is exp(Θ(√(k log(N/k)))) once N ≥ k^(2+ε) and log N = o(k), and exp((1+o(1))√(k log N)) when log N/log k → ∞.",
    "Theorem 7.1 and Proposition 7.2 (proof): γ ≤ (√e/d_k) exp(√((k−1)L)) from the observed pivot range L = log(d₁²/d_k²), with an optimal coefficient.",
    "Table 2 (computation): the new bound is smaller than the classical one from k = 13 at H = 1; at k = 100 it allows about 10⁹ to 10¹⁰ where the classical bound allowed about 10³⁰."
  ],
  "reviews": [],
  "evidencePackage": "An 18-page paper with a plain introduction, written proofs, two tables, one figure, practical guidance and a section on checks and open problems; check scripts with recorded outputs (verify.py: exact rational identities, 25 orthonormal inputs, eleven completed examples at k = 16 to 128 factorised by two QRCP implementations, a rebuilt failed candidate that must be rejected, large-rank sums against 80-digit values; check_extensions.py: 50-digit checks of C(H), the comparison table, the Kahan family and an inequality on 779 grid points). A clean replay under Python 3.12, also under python -O, reproduced every output, Linux CI agreed under a declared comparison policy, and a deliberately wrong constant was caught.",
  "openProblems": [],
  "mathObjects": [],
  "relatedWorks": [
    {
      "citation": "Chen, Liu, He and Dong (2026), Iteris: Agentic Research Loops for Computational Mathematics. arXiv preprint. Gives the first, negative answer to Problem 4.3 of Amsel et al.; the construction in its Appendix B (Lemma 24) already gives growth of order exp(c√k) for every c below ½√(log H).",
      "url": "https://arxiv.org/abs/2606.02484",
      "doi": "10.48550/arXiv.2606.02484"
    },
    {
      "citation": "Amsel et al. (2026), Linear Systems and Eigenvalue Problems: Open Questions from a Simons Workshop. arXiv preprint (versions 1 and 3). Problem 4.3, posed by Damle and Kressner, asks whether column-pivoted QR meets the existence scale on orthonormal rows; version 3 records the answers.",
      "url": "https://arxiv.org/abs/2602.05394",
      "doi": "10.48550/arXiv.2602.05394"
    },
    {
      "citation": "Drmač and Gugercin (2016), A New Selection Operator for the Discrete Empirical Interpolation Method — Improved A Priori Error Bound and Extensions. SIAM Journal on Scientific Computing. Theorem 2.1 is the classical bound that grows like 2^k.",
      "url": "https://doi.org/10.1137/15m1019271",
      "doi": "10.1137/15m1019271"
    },
    {
      "citation": "Damle (2026), Computing Strong Rank-Revealing Factorizations for Matrices with Orthonormal Rows. arXiv preprint. Shows that Bischof–Stewart pivoting meets the existence scale on orthonormal rows.",
      "url": "https://arxiv.org/abs/2607.13532",
      "doi": "10.48550/arXiv.2607.13532"
    },
    {
      "citation": "Engler (1997), The behavior of the QR-factorization algorithm with column pivoting. Applied Mathematics Letters. The pivot floor used in Lemma 3.1, the average-determinant property and Hadamard completions.",
      "url": "https://doi.org/10.1016/s0893-9659(97)00098-0",
      "doi": "10.1016/s0893-9659(97)00098-0"
    },
    {
      "citation": "Kahan (1966), Numerical Linear Algebra. Canadian Mathematical Bulletin. The classical triangular matrix behind the matching examples.",
      "url": "https://doi.org/10.4153/cmb-1966-083-2",
      "doi": "10.4153/cmb-1966-083-2"
    },
    {
      "citation": "Sra (2026, as \"Count ex Machina\"), GPT, the Counterexample Machine. arXiv preprint. Records a small exact example with k = 3 and N = 8 (Appendix A.5).",
      "url": "https://arxiv.org/abs/2608.29595",
      "doi": "10.48550/arXiv.2608.29595"
    },
    {
      "citation": "Gu and Eisenstat (1996), Efficient Algorithms for Computing a Strong Rank-Revealing QR Factorization. SIAM Journal on Scientific Computing. Strong rank-revealing QR, a selection rule with a polynomial guarantee.",
      "url": "https://doi.org/10.1137/0917055",
      "doi": "10.1137/0917055"
    },
    {
      "citation": "Drmač and Veselić (2008), New Fast and Accurate Jacobi SVD Algorithm. I. SIAM Journal on Matrix Analysis and Applications. Related work on scaled triangular inverses (a different quantity).",
      "url": "https://doi.org/10.1137/050639193",
      "doi": "10.1137/050639193"
    },
    {
      "citation": "Gilles (2026), Convergence rates for pivoted QR and LU. arXiv preprint. Approximation error of pivoted QR under singular-value decay (a different quantity).",
      "url": "https://arxiv.org/abs/2607.26863",
      "doi": "10.48550/arXiv.2607.26863"
    },
    {
      "citation": "Rusu and Dobra (2007), Pseudo-random number generation for sketch-based estimations. ACM Transactions on Database Systems. The finite-field sign construction used in Appendix A.",
      "url": "https://doi.org/10.1145/1242524.1242528",
      "doi": "10.1145/1242524.1242528"
    }
  ],
  "claimAssurance": null,
  "operatingModel": {
    "version": "1.0",
    "workId": "ep-work:column-pivoted-qr-orthonormal-worst-case",
    "attemptIds": [
      "ep-attempt:column-pivoted-qr-orthonormal-worst-case-publication"
    ],
    "aims": [
      "science"
    ],
    "artifactRoles": [
      "research-output",
      "communication"
    ],
    "lineageId": null,
    "accelerationPrimitives": [
      "adversarial-controls",
      "assurance-vector",
      "agent-readable-research-object"
    ],
    "decisionObject": {
      "type": "other",
      "description": "Know how much worse than the best possible choice the column-pivoted QR selection of rows of an orthonormal basis can be, with the exact constant in the exponent.",
      "scope": "Exact arithmetic, worst cases; the bound exp(C(H)√k) with exact C(H) on the logarithmic scale; the coherence-free law in N and k; the a posteriori bound; H = 1 examples under a tie rule; N = O(k) open."
    },
    "bottleneckTargeted": [
      "assurance",
      "publication"
    ],
    "semanticBridge": {
      "state": "explicit",
      "description": "Section 1 separates what was known (the existence scale and Problem 4.3, the 2^k bound, Engler's pivot floor, Kahan's matrix, Chen et al.'s negative answer with exp(c√k) growth, Damle's modified rule) from what is new (the matching upper bound, the exact constant C(H), the endpoint H = 1, the coherence-free law and the a posteriori bound).",
      "remainingRisks": [
        "The bounded literature search is not exhaustive priority certification.",
        "Exact arithmetic only; floating-point behaviour is not analysed."
      ]
    },
    "humanJudgmentGates": [
      "Publication is authorised; scientific endorsement is not inferred.",
      "External specialist review and independent reproduction remain open."
    ],
    "parentLinks": [],
    "assuranceTarget": {
      "dimensions": [
        "specialistReview",
        "independentRerun",
        "formalVerification",
        "priorityAssessment"
      ],
      "nextAction": "Have a numerical linear algebra specialist check the weighted trailing-energy estimate (Lemma 3.2), the upper-bound optimisation of Section 4 and the completion argument of Section 5.",
      "claimCeiling": "Written proofs of Theorem 2.3, Corollary 2.6, Theorem 2.7, Corollary 2.8, Theorem 7.1 and Proposition 7.2 in exact arithmetic; sharpness on the logarithmic scale; internal verification is not external assurance."
    },
    "impactClaims": [
      {
        "id": "column-pivoted-qr-orthonormal-worst-case-no-impact",
        "aim": "science",
        "outcome": "An openly inspectable determination of the worst-case growth of column-pivoted QR on orthonormal rows, with replayable checks.",
        "setting": "AI-assisted numerical linear algebra research",
        "status": "NO_IMPACT_EVIDENCE",
        "designClass": "none",
        "comparator": "None registered.",
        "estimand": "No research-speed or scientific-impact effect is estimated.",
        "evidenceRefs": [],
        "registeredDesignRef": null,
        "independentAssessment": null
      }
    ]
  },
  "researchMetrics": {
    "schemaVersion": "1.0",
    "policyUrl": "https://evidencepress.org/api/research-metrics-policy.json",
    "outcomePolicyUrl": "https://evidencepress.org/api/v2/research-metrics-policy.json",
    "workLedgerUrl": "https://evidencepress.org/api/work-ledger.json",
    "claimCeiling": "These records describe scoped workflow inputs and outcomes. They support calibration and reference-class construction, but do not by themselves establish research acceleration, theorem correctness, novelty, independent assurance or impact.",
    "attempts": [
      {
        "attemptId": "ep-attempt:column-pivoted-qr-orthonormal-worst-case-publication",
        "status": "active",
        "resultClass": "pending",
        "metrics": {
          "schemaVersion": "1.0",
          "measurementScope": "publication-only",
          "scopeBoundary": "Only archive packaging, site integration, guarded delivery and canonical readback from registration are prospective. Earlier research, review, revision and extension work are left-censored.",
          "forecast": {
            "frozenAt": "2026-11-16T11:00:45Z",
            "procedureClass": "full-candidate-publication",
            "targetOutcome": "Accurately scoped GitHub, Zenodo and canonical Evidence Press release with explanatory media and machine readback, if all publication gates pass.",
            "expectedActiveMinutes": 90,
            "plausibleLowMinutes": 60,
            "plausibleHighMinutes": 150,
            "expectedUnattendedWaitMinutes": 30,
            "referenceClass": {
              "label": "Full mathematical candidate publication",
              "sampleSize": 0,
              "basis": "Route benchmark planning prior, not empirical calibration."
            },
            "fermiComponents": [
              {
                "component": "Research archive packaging, replay and identities",
                "count": 1,
                "lowMinutesPerUnit": 25,
                "centralMinutesPerUnit": 35,
                "highMinutesPerUnit": 60,
                "basis": "One frozen package with a rating-free public export, clean replay and DOI insertion."
              },
              {
                "component": "Page, metadata and media integration",
                "count": 1,
                "lowMinutesPerUnit": 20,
                "centralMinutesPerUnit": 30,
                "highMinutesPerUnit": 45,
                "basis": "One reader-first release, drafted in parallel with verification."
              },
              {
                "component": "Guarded deployment and canonical readback",
                "count": 1,
                "lowMinutesPerUnit": 15,
                "centralMinutesPerUnit": 25,
                "highMinutesPerUnit": 45,
                "basis": "Mandatory site tests, one merge cycle with ledger carry-forward, zero-cost deployment and public checks."
              }
            ],
            "probabilityPositiveSignal": 0.9,
            "probabilityTargetClosure": 0.85,
            "probabilityHorizonMinutes": 300,
            "assumptions": [
              "The approved scientific package remains valid and configured providers remain available."
            ],
            "reforecastTriggers": [
              "A publication gate reveals a scientific or integrity defect, or a provider outage blocks delivery."
            ],
            "stopRule": "Do not weaken assurance, integrity or cost gates to meet the forecast."
          },
          "outcome": null
        },
        "missingnessReason": null
      }
    ]
  }
}