Geometric Criteria for Benign Overfitting

Geometric Criteria for Benign Overfitting

Definitions and mathematical statement

Shared notation.

Standard number systems are denoted by $\mathbb N,\mathbb Z,\mathbb Q,\mathbb R,\mathbb C$; index ranges, computational models and specialized notation follow the cited problem source and the definitions below.

In the linear interpolation setting, data form a matrix $X$ and response vector $y$. For a specified norm on the parameter space, a minimum-norm interpolator chooses $\widehat\beta\in\arg\min\{\lVert\beta\rVert:X\beta=y\}$. Interpolation means zero training residual, possibly fitting noise. Prediction risk is expected loss on a fresh observation from the specified data law; excess risk subtracts the population benchmark. Benign overfitting means that excess risk tends to zero along the source asymptotic sequence even though the observations are interpolated. Geometry includes the norm ball and its relation to the design and noise distributions.

Statement recorded in the supplied source.

Identify necessary and sufficient geometric conditions under which, for an arbitrary norm, the associated minimum-norm interpolating estimator exhibits benign overfitting. [S2]

Scope qualification.

Sample-size/dimension scaling, noise law, loss and tie-breaking among minimizers are part of the estimator model. A universal criterion is requested, rather than a claim that arbitrary norms automatically yield consistent interpolation.

Source status.

The supplied catalogue labels this record partially solved. Its source-status metadata was checked against the supplied snapshot on 2026-10-11. This import does not independently establish current open status or mathematical correctness. The catalogue literature assessment is dated 2026-08-22.

Short English statement

Which geometric properties exactly determine whether minimum-norm interpolation generalizes despite fitting noisy data?

Sources

• S1 Ulam AI and the UnsolvedMath Contributors, UnsolvedMath: A Curated Collection of Open Mathematics Problems, record 30005628 (OWR-14297738-001). Catalogue curation and metadata: CC BY 4.0. The supplied source question is retained; definitions and the English summary are editorial additions. https://huggingface.co/datasets/ulamai/UnsolvedMath.
• S2 Mini-Workshop: Interpolation and Over-parameterization in Statistics and Machine Learning​ $2024$, Oberwolfach Reports, DOI 10.4171/owr/2023/41, pp. 2359--2376 https://doi.org/10.4171/owr/2023/41.
• S3 Bartlett--Long--Lugosi--Tsigler, benign overfitting https://pmc.ncbi.nlm.nih.gov/articles/PMC7720150/. Underlying Oberwolfach Reports material retains its source-specific licence, including CC BY-SA 4.0 where applicable; see THIRD_PARTY_NOTICES.md.

Preparation and review.

Original question attributed to the cited source and UnsolvedMath contributors. Codex (GPT-6; exact serving revision not exposed) assisted with definitions, English summaries, source inspection and formatting. No new human mathematical authorship or independent proof verification is asserted. Canonical ID 30005628 and source code OWR-14297738-001 checked against the complete supplied registry and existing repository statements; known flagged repetitions and obvious identical conjectures excluded. Full mathematical-equivalence/literature review has not been performed. Statement preparation and final review followed contribution prompts C and D; automated checks verify structure and identity, not open status.

Research attempt

Literature, source context and scope

The full original catalogue body and source qualifications are preserved above. The relevant primary workshop passages were read in the cached publisher report. Historical claims and subsequent attributed results are distinguished from the mathematical work below. A restricted example, standard lemma or known theorem is not counted as new progress on the entire original target. Missing assumptions and repeated or rhetorical extracts are not silently turned into different solved problems.

A quadratic-norm calculation and the first arbitrary-norm obstruction

The primary source specifically seeks geometric understanding beyond Euclidean norms and flags dependence between noise and design as a difficulty. Bartlett--Long--Lugosi--Tsigler give covariance-based results in a specified Hilbert-space model; their assumptions are not a theorem for arbitrary norms. We first isolate an exact conditional calculation. Let $X\in\mathbb R^{n\times p}$ have full row rank, let $Q$ be positive definite, and use $\|\beta\|_Q^2=\beta^TQ\beta$. Conditional on this fixed design, suppose $y=X\beta_*+\xi$, $\mathbb E\xi=0$ and $\mathbb E\xi\xi^T=\sigma^2I_n$. Let the second-moment matrix of an independent test covariate be $\Sigma\succeq0$. Along any sequence in this conditional model, both nonnegative terms must vanish for the averaged excess risk to vanish. This exact identity is a known linear-algebra baseline, not the sought interpretable arbitrary-norm criterion. Its derivation relies on the quadratic objective producing a linear map $y\mapsto By$ and on the stated conditional noise moments. An attempted extension that disregards tie-breaking fails already for $\ell_1$. With $X=(1,1)$ and response $1$, every vector $(t,1-t)$ for $0\le t\le1$ has minimal $\ell_1$ norm one: the triangle inequality gives a lower bound $|\beta_1+\beta_2|=1$, attained by all these vectors. For test second moment $\Sigma=\operatorname{diag}(1,4)$ and zero target, their prediction errors are $t^2+4(1-t)^2$, including values $4,1,$ and $4/5$ at $t=0,1,4/5$. Thus the norm and interpolation constraint alone do not determine risk without the selection rule. This is a fixed-design diagnostic, not an i.i.d. asymptotic counterexample to a properly specified estimator.

Verification and remaining gaps

Rational full-row-rank designs and positive definite $Q$ were used to check $XB=I$, the orthogonality identity and the exact noise average over all sign-noise vectors. The $\ell_1$ risks were checked exactly. No distribution-free geometry theorem for nonquadratic norms or design-dependent noise is obtained; the displayed trace identity does not bypass those source difficulties.

References and source attribution

Primary workshop report: https://doi.org/10.4171/owr/2023/41. Relevant passages read: 2023/41 p. 2363, arbitrary-norm problem, geometric tools and two stated obstacles.

https://pmc.ncbi.nlm.nih.gov/articles/PMC7720150/. Read primary abstract, Definitions1--2 and main-theorem hypotheses for Hilbert minimum-norm regression; no arbitrary-norm extension is attributed to that theorem.

All retrievals were checked on 11 October 2026. Original question attribution remains in the preserved Sources section. External theorems are attributed and not claimed as new discoveries or fully verified merely by citation.

Integration review.

Codex (GPT-6; exact serving revision not exposed) reviewed the written arguments and expanded layout and source-reference wrappers. Repository metadata and the production web renderer were checked using Python 3.13 and JavaScript with MathJax 3.2.2. This model-assisted review is not independent human proof verification; no PDF compilation or formal proof checking is asserted.

Completion Estimate

COMPLETION: 0% No substantiated new progress toward the entire original target is claimed. The worked results and failed route are delimited above; a known result, illustrative model, or auxiliary obstruction is not counted as original-target completion.

Final status and attribution

Attempt status: unresolved. This describes this research attempt, not a new ruling on the source catalogue's status or on external literature claims. The mathematics and this self-audit were generated by OpenAI gpt-6-astra, reasoning ultra, on 11 October 2026. Codex (GPT-6; exact serving revision not exposed) orchestrated the batch, read and reviewed its written mathematical arguments, and contributed formatting/repository review. No human mathematical authorship, independent human proof checking, or formal verification is claimed. The historical preparation paragraphs in the reproduced catalogue body concern that body only.

Contribution prompt A was applied to the complete definition and supplied source record, followed by the adversarial checks of prompt B. Local reproduction files are batch025/checks.py, computations.json, and the per-ID audit JSON. They were not public links at preparation time. Numerical checks are finite and do not prove unrestricted results. No LaTeX engine was found locally; compilation was not performed by this research agent.