Optimal Worst-Case Estimators for Signal Reconstruction

Research-attempt provenance

Definitions and mathematical statement

Shared notation.

Standard number systems are $\mathbb N,\mathbb Z,\mathbb Q,\mathbb R,\mathbb C$. Specialized notation and index ranges follow the cited source and the definitions below.

A minimax estimator minimizes the largest risk over a stated model class. For a predictor $f$, distributional risk is $R_P(f)=\mathbb E_{z\sim P}[\ell(f,z)]$ for the specified loss. A distribution shift replaces $P$ by $Q$.

Record scope and qualifications.

The catalogue title is retained as metadata. Named certificates, contracts, bounds or frameworks in that title are not asserted as results of the original source extract. The signal/noise models, loss, admissible estimators, regularizers, and shift class are not supplied; the three directions are kept distinct.

Source-record limitation: The signal and noise families and the quantifiers for the proposed risk monotonicity are unspecified; no definite minimax value or equivalence criterion can be recovered.

Statement recorded in the supplied source.

Characterize minimax-optimal estimators for broad signal and worst-case noise models. When are robust-optimization formulations equivalent to regularized estimation, and when is in-distribution risk $R_P(f)$ monotonically related to shifted-distribution risk $R_Q(f)$? [S2]

Source status.

The supplied catalogue labels this source record open; that metadata was checked against the supplied snapshot on 2026-10-11. The label is preserved for provenance and is not an independent verification of present open status or mathematical correctness. The catalogue literature assessment is dated 2026-08-22.

Short English statement

Characterize optimal worst-case estimators and relationships between robust, regularized, and shifted-data risk.

Sources

• S1 Ulam AI and the UnsolvedMath Contributors, UnsolvedMath: A Curated Collection of Open Mathematics Problems, record 30005250 (OWR-11695855-003). Catalogue curation and metadata: CC BY 4.0. The source extract is retained; definitions, English summaries and qualifications are editorial additions. https://huggingface.co/datasets/ulamai/UnsolvedMath.
• S2 Mini-Workshop: Mathematical Foundations of Robust and Generalizable Learning $2023$, Oberwolfach Reports, DOI 10.4171/owr/2022/46, pp. 2659--2686 https://doi.org/10.4171/owr/2022/46.
• S3 Provably robust estimators for inverse problems https://pmc.ncbi.nlm.nih.gov/articles/PMC12721873/.
• Source-specific Oberwolfach Reports licences, including CC BY-SA 4.0 where applicable, are retained; see THIRD_PARTY_NOTICES.md.

Preparation and review.

The original source and UnsolvedMath contributors retain attribution for the source extract. Codex (GPT-6; exact serving revision not exposed) assisted with definitions, English summaries, scope notes and formatting. No new human mathematical authorship or independently verified solution is asserted. Permanent source ID 30005250 and code OWR-11695855-003 checked against the complete supplied registry and current repository. Distinct upstream records, including repeated questions and broad or damaged extracts, are retained for complete coverage of the source CS category. A repeated record is not a new independent conjecture. Complete mathematical-equivalence and current-literature review has not been performed. Preparation and final review follow contribution prompts C and D, with source gaps disclosed. Structural and rendering checks do not establish a new theorem or open status.

Literature and context check

I read the source discussion on page 2667, which asks for broad minimax models, robustness/regularization equivalences and monotone risk relations. The models below are explicitly auxiliary and finite-dimensional. The original statement contains no universal quantifiers that would make every regularizer or shift subject to one theorem, so the positive identity and its limitations are kept separate.

Research attempt: Exact scalar minimax estimation and one robust-regularization identity

An exact bounded-signal, bounded-noise minimax value.

Let $R,\delta\geq0$, observe $y=\theta+e$ with $|\theta|\leq R$, $|e|\leq\delta$, and measure absolute error. For any admissible observation, its feasible signal interval is \[ I_y=[-R,R]\cap[y-\delta,y+\delta]. \] The midpoint of $I_y$ minimizes its maximum distance to a point in $I_y$: if the endpoints are $l\leq u$, any estimate $a$ has maximum distance at least $(u-l)/2$ by $u-l\leq|u-a|+|a-l|$, and the midpoint attains it. The interval length is at most $2\min(R,\delta)$. At $y=0$ it is exactly that length. Therefore the global deterministic minimax absolute error is $\min(R,\delta)$, and the squared-error value is its square. Values of the estimator at infeasible observations do not affect this criterion.

For a scalar linear rule $\widehat\theta=ay$, its worst absolute error is \[ \sup_{|\theta|\leq R,|e|\leq\delta}|(a-1)\theta+ae| =|a-1|R+|a|\delta. \] The triangle inequality is attained by choosing the two signs to align. For $a\in[0,1]$ this expression is $R+a(\delta-R)$; outside that interval moving toward its nearest endpoint cannot increase it. Thus $a=1$ is optimal if $\delta<R$, and $a=0$ is optimal if $R<\delta$; all $a\in[0,1]$ are optimal when $R=\delta$. For $R,\delta>0$ unequal, ridge shrinkage $a=1/(1+\lambda)$ with finite $\lambda>0$ is strictly suboptimal in this worst-case model, although it may be optimal under a different probabilistic model. This falsifies a blanket identification of bounded-noise minimax estimation with arbitrary ridge tuning.

A positive equivalence with an explicitly matched uncertainty set.

Let $X\in\mathbb R^{m\times d}$, $y\in\mathbb R^m$, $m\geq1$, and $\rho\geq0$. For every coefficient vector $\beta$, \[ \sup_{\|\Delta\|_F\leq\rho}\|y-(X+\Delta)\beta\|_2 =\|y-X\beta\|_2+\rho\|\beta\|_2. \] The upper bound follows from $\|\Delta\beta\|\leq\|\Delta\|_F\|\beta\|$. If $r=y-X\beta\ne0$ and $\beta\ne0$, equality is attained by $\Delta=-\rho(r/\|r\|)(\beta/\|\beta\|)^T$. If $r=0$ and $\beta\ne0$, use any unit output vector in place of $r/\|r\|$; if $\beta=0$, both sides equal $\|y\|$. Thus minimizing this robust residual is exactly minimizing a residual norm plus an $\ell_2$ penalty. Squaring the residual before the supremum gives the square of that sum, with cross term $2\rho\|y-X\beta\|\|\beta\|$, not generally the ridge objective $\|y-X\beta\|^2+\lambda\|\beta\|^2$ for a fixed universal $\lambda$.

No unrestricted risk ordering follows from either fact.

Take losses $(0,1)$ for one estimator and $(1,0)$ for another on two cases. With $P=(3/4,1/4)$ and $Q=(1/4,3/4)$ their risks are respectively $(1/4,3/4)$ and $(3/4,1/4)$, reversing the order. As a concrete squared-error realization, the target is zero, the measurement is $0$ or $1$, and the estimators are $f(y)=y$ and $g(y)=1-y$. This shares the elementary two-case obstruction with record 30005249; it is not claimed as a second independent result. Restricted positive shift theorems need assumptions tying together the estimator class and the pair $(P,Q)$.

Verification and adversarial audit

The interval argument includes $R=0$ or $\delta=0$. Strict ridge suboptimality is claimed only for positive unequal parameters and finite positive regularization. The robust-design identity checks zero residual, zero coefficient and zero uncertainty radius before divisions. The risk example is compatible with squared loss and does not claim to contradict the source's restricted covariate-shift result.

Audit corrections.

Matched each equivalence to its precise norm and uncertainty set, and kept squared robust residual distinct from ridge. Included degenerate intervals and explicitly declared the repeated two-case shift obstruction.

External computation.

None; symbolic arguments and boundary checks only.

Scope and remaining gaps

One minimax model and one exact uncertainty/penalty correspondence are solved as auxiliary known-style calculations. No broad classification of optimal estimators or all regularization equivalences is supplied, and the original source's general model families remain unspecified. This is not a full resolution of the bundled research agenda.

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% The calculations below the preserved source statement investigate a concrete route, but establish no substantiated new progress toward the entire original target. Known special cases and failed sufficient conditions are not credited as a solution.

References

The original source references and catalogue attribution are preserved in the Sources section above. The following URLs were actually consulted for this attempt; retention of an additional original reference does not claim fresh consultation of it.

• https://ems.press/content/serial-article-files/46984 (consulted 2026-10-11).

Model and contributor disclosure

Codex GPT-6 completed written-proof review of this attempt at its declared scope; this is model-assisted review, not independent human proof checking. OpenAI gpt-6-astra, requested reasoning configuration ultra, generated this research attempt on 2026-10-11. The exact serving revision is not exposed. Codex GPT-6 performed orchestration and repository-format review; its exact serving revision is also not exposed. The mathematical route, boundary checks and final adversarial self-audit followed contribution prompts A then B in docs/contribution_prompts.md, after reading CONTRIBUTING.md. Python 3.14.2 assembled the standalone documents; any mathematical computation is specified in Verification. No theorem prover was used. No human mathematical authorship, substantial human proof correction, or independent human proof checking is asserted. Model-assisted review, source retrieval, and structural checks do not establish correctness or novelty. Full private reasoning transcripts are unavailable as reproducibility artifacts; the complete arguments, inputs and outcomes actually used are stated here or in the companion batch audit.

Final status

Unresolved as a new research contribution. This is a concrete route investigation, not a claim of new complete resolution or a current-status certification. A known answer reproduced for context does not become a new contribution. The catalogue's source-status label is preserved; this attempt status does not assert that a source-labelled solved question is open.