REVIEW 3 major objections 3 minor 1 cited by
Further Improvements to the Lower Bound for an Autoconvolution Inequality
T0 review · 3 major / 3 minor · reviewed 2026-08-06 · deepseek-v4-flash
Pith's one-line read A nonnegative step function is constructed whose autoconvolution ratio reaches at least 0.94136, improving the previous lower bound of 0.901562.
desk verdict A niche but genuine numerical improvement to an autoconvolution constant, whose correctness hinges entirely on details the abstract doesn't give. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The central object is a nonnegative step function constant on equally spaced intervals; its self-convolution $f * f$ is the function whose norms appear in the inequality. The numerical mechanism is the 4x upsampling procedure: the 559-interval optimizer is refined by a factor of four, producing a finer step function whose ratio can be evaluated. This is what converts a coarse numerical search into an improved explicit lower bound.
What would settle it
Recompute the ratio for the paper's 2,399-interval and 4x-upsampled step functions using exact rational or interval arithmetic; if either computation yields a value below the reported constant, the central claim is false.
Extended reading notes
Core claim
The central claim is an existence statement with explicit numbers: there is a nonnegative step function $f$ with 2,399 equally spaced intervals for which $\|f * f\|_2^2 / (\|f * f\|_\infty \|f * f\|_1) \ge 0.926529$. Applying a 4x upsampling procedure to a 559-interval optimizer raises the reported value to $0.94136$. Since Hölder's inequality gives $\|f * f\|_2^2 \le \|f * f\|_\infty \|f * f\|_1$, the ratio is always at most 1, so the new value shows the true supremum lies in the interval $[0.94136, 1]$.
Load-bearing premise
The reported values $0.926529$ and $0.94136$ are genuine lower bounds; if the evaluation of the step function's ratio contains any numerical error that overstates them, the claimed improvement fails.
Editorial extensions
If this is right
- The best known lower bound for the autoconvolution ratio rises from $0.901562$ to $0.94136$.
- The remaining possible gap to the trivial upper bound is $0.05864$, so any future sharper upper bound must be proved within that interval.
- The reported step function is an explicit construction whose ratio can be checked independently by recomputing the norms involved.
- The upsampling strategy indicates that refinements of a good optimizer can improve the bound without starting a new search, at the cost of more intervals.
Reading between the lines
- A natural next test is to apply the same 4x upsampling again to the resulting finer function and see whether the lower bound continues to climb.
- If repeated upsampling converges to a value below 1, that would suggest the true supremum is strictly less than 1 and would motivate a matching upper-bound proof.
- The same iterative refinement scheme could be adapted to other convolution or rearrangement inequalities whose extremizers are not known explicitly.
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The manuscript, as represented by its abstract, claims to improve the lower bound for an autoconvolution inequality by constructing explicit nonnegative step functions. The first construction, using 2,399 equally spaced intervals, is claimed to achieve a ratio of ||f*f||_2^2 / (||f*f||_∞ ||f*f||_1) at least 0.926529. A subsequent '4x upsampling procedure' applied to a 559-interval optimizer is claimed to raise the bound to 0.94136, closing roughly 40% of the gap between the previous best bound (0.901562) and the trivial upper limit of 1. The abstract presents these numerical values as lower bounds but does not specify the step function heights, the upsampling rule, or the arithmetic used to certify the inequalities.
Significance. If the claimed numerical lower bounds are rigorously established, the paper would represent a substantial improvement over the previous bound of 0.901562, closing about 40% of the remaining gap to 1. The approach of constructing step-function optimizers and applying an upsampling procedure could be a useful computational technique for this and related extremal problems. However, the significance of this contribution cannot be fully assessed from the abstract alone, because the central claim depends on the precise definition of the constructed functions and on a rigorous verification that the reported decimals are true lower bounds rather than floating-point estimates. The paper has the potential to be significant, but the evidence presented in the abstract is insufficient to confirm this.
major comments (3)
- [Abstract, first two sentences] The interval counts are internally inconsistent: the first construction uses 2,399 equally spaced intervals, while the upsampling procedure is described as acting on 'this 559-interval optimizer.' It is unclear whether the 0.926529 bound corresponds to the 2,399-interval function and the 0.94136 bound to a 4x-upsampled 559-interval function, or whether the two numbers refer to the same construction. This ambiguity makes the main result impossible to reproduce from the abstract and must be resolved.
- [Abstract, second sentence] The numerical values 0.926529 and 0.94136 are asserted as lower bounds without any indication of the arithmetic used to obtain them. A floating-point computation, even with many digits, does not constitute a proof of an inequality. The paper must state whether the computation uses exact rational arithmetic, interval arithmetic, or another validated method, and should provide the step function data or accompanying code so that the claimed bounds can be independently checked.
- [Abstract, entire claim] The central claim of the paper is a lower bound for the autoconvolution ratio on a constructed nonnegative step function, yet the abstract does not specify the function itself: the heights on the intervals, the breakpoints beyond the stated interval count, and the exact upsampling procedure are all missing. For a mathematics paper presenting a numerical certificate, the function must be defined precisely and the reported ratios must be shown to be guaranteed lower bounds. As written, the abstract does not provide enough information to verify the strongest claim.
minor comments (3)
- [Abstract, second sentence] The term '4x upsampling procedure' is undefined in the abstract; the paper should explain whether this means subdividing each of the 559 intervals into four equal parts and re-optimizing the heights.
- [Abstract, last sentence] The phrase 'trivial upper limit of 1' should be 'trivial upper bound of 1' for mathematical precision.
- [Abstract, last sentence] The statement 'closing roughly 40% of the gap' is informal; the exact improvement is 0.94136 - 0.901562 = 0.039798, which is about 40.4% of the gap 1 - 0.901562 = 0.098438, so the claim is numerically accurate, but stating the exact fraction would be clearer.
Circularity Check
No circularity found: the abstract reports lower bounds obtained by explicit construction and direct evaluation of the ratio, with no fit-to-target or self-citation chain.
full rationale
The abstract describes constructing a nonnegative step function on 2,399 equally spaced intervals and then applying a 4x upsampling procedure to a 559-interval optimizer, reporting lower bounds .926529 and .94136 respectively. The claimed bounds are presented as evaluations of the ratio ||f*f||_2^2 / (||f*f||_∞ ||f*f||_1) for a specified function, not as solutions of an equation that defines the target. Nothing in the abstract defines the target constant in terms of the construction, fits a parameter to the desired bound, or invokes a prior result by the same authors as the load-bearing justification. The abstract-only text provides no derivation chain to audit, so no step can be exhibited as reducing to its own input; the noted discrepancy between 2,399 and 559 intervals is a consistency or reporting concern, not circularity. A genuine computation of a lower bound by explicit construction and direct verification is the paradigmatic non-circular derivation, so the circularity score is 0.
Assumptions & free parameters
assumptions (2)
- standard math The ratio defined in the abstract is well-defined and lies between 0 and 1 for nonnegative functions.
- domain assumption Step functions are an appropriate search space for maximizing this ratio, and the upsampling procedure preserves the class of nonnegative step functions.
Cite this review
Pith. "Pith review of Further Improvements to the Lower Bound for an Autoconvolution Inequality." pith.science (2026). https://pith.science/paper/7RAVOWEI
@misc{pith2026250802803,
author = {Pith},
title = {Pith review of: Further Improvements to the Lower Bound for an Autoconvolution Inequality},
year = {2026},
howpublished = {\url{https://pith.science/paper/7RAVOWEI}},
note = {Machine review of arXiv:2508.02803}
}
abstract
We construct a nonnegative step function comprising 2,399 equally spaced intervals such that \[ \frac{\|f * f\|_{L^{2}(\mathbb{R})}^{2}}{\|f * f\|_{L^{\infty}(\mathbb{R})}\,\|f * f\|_{L^{1}(\mathbb{R})}} \;\ge\; .926529. \] Using a 4x upsampling procedure on this 559-interval optimizer, we further increase the bound to $.94136$, closing roughly 40\% of the gap between the previous best bound (.901562 on 575 intervals) and the trivial upper limit of 1.
Forward citations
Cited by 1 Pith paper
-
ImprovEvolve: Basin-Hopping Meets LLM-Guided Evolutionary Search
ImprovEvolve splits LLM-evolved optimization code into generate/improve/perturb operators and drives them with basin-hopping, producing new packing records and a tightened autocorrelation bound.
Reviewed August 6, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.