XSci

An exponentially accurate cubic-partition approximation for two-row rectangle partitions

Chandan Kumar Shukla

Submitted by Chandan Kumar Shukla

Published November 14, 2026 · Version v1, November 14, 2026 · DOI 10.66977/xsci.2611.002b

Combinatorics

Abstract

Let r(n) count multisets of integer-sided rectangular pieces that tile a 2 x n rectangle, with rotations allowed. Let a(N) be the cubic partition function, with generating function prod_{j>=1} (1-q^j)^{-1} (1-q^{2j})^{-1}. We show that, for every eps > 0, 0 <= a(2n) - a(2n-2) - r(n) = O_eps( exp( (2*pi/sqrt(3) + eps) * sqrt(n) ) ). Combining this estimate with a known Rademacher-type formula for cubic partitions gives an asymptotic expansion to every order. We obtain eventual strict log-concavity and eventual hyperbolicity of every fixed-degree Jensen polynomial of r(n). This establishes eventual log-concavity in a direction raised by Gajdzica, Visser and Zakarczemny. No explicit threshold for eventuality is claimed.

View PDF

Review reports

xPeerd (AI review)

November 14, 2026

AI Review (xPeerd — DAReview)

Summary

This draft studies the rectangle partition counting function for a 2 × n rectangle, denoted r(n), and compares it to differences of the cubic partition function, U(n)=a(2n)−a(2n−2). The core claim is an exponentially accurate approximation “0 ≤ a(2n) − a(2n − 2) − r(n) = Oε exp((2π/√3 + ε)√n)” and, from this, an all-orders asymptotic expansion for r(n), eventual strict log-concavity, and eventual hyperbolicity of every fixed-degree Jensen polynomial. The manuscript is clear about what is new and what is not: it explicitly states that the leading equivalent for r(n) was already known and that the proposed contribution is the “quantitative combination” of prior ingredients to obtain sharper consequences.

The work appears mathematically coherent at a high level and is more careful than many drafts in delimiting its claims. However, the paper is still vulnerable on novelty positioning, proof exposition, reproducibility of computational checks, and publication readiness. Several arguments rely on importing substantial prior theorems, and the manuscript itself acknowledges that “independent mathematical and novelty review” is still required before submission. The draft looks serious rather than obviously fabricated, but it carries some paper-risk signals typical of preliminary math manuscripts: compressed proofs, limited contextualization of novelty, no external validation of computations, and no explicit threshold results despite “eventual” claims.

Potential Major Revisions

  1. Novelty and contribution need stronger defense.

The manuscript candidly states that “The leading equivalent … is already Theorem 1.5 of [2]; it is not a new claim here” and later that “The contribution proposed here is their quantitative combination, the resulting all-order rectangle expansion, and its log-concavity and Jensen consequences.” This is intellectually honest, but it also exposes the manuscript to a novelty objection. In its current form, the paper does not sufficiently explain why combining known decomposition bounds, known cubic-partition formulas, and known Jensen-polynomial theory rises to the level of a publishable theorem rather than a technically competent corollary chain.

The author should therefore add a dedicated subsection in the introduction that distinguishes:

  • what is inherited directly from Gajdzica–Visser–Zakarczemny,
  • what is inherited from Guadalupe’s cubic-partition exact formula,
  • what is inherited from Griffin–Ono–Rolen–Zagier and O’Sullivan,
  • and exactly where the nontrivial new bridge argument lies.

At present, the sentence “A literature search is not a proof of priority; this draft requires independent mathematical and novelty review before submission” is prudent but also indicates that the paper’s originality case is not yet fully stabilized. For submission, this cannot remain merely cautionary; it must be replaced by a proactive novelty argument.

  1. Several proof steps are too compressed for verification by a referee.

In Section 2, the proof of the main estimate reduces the problem to bounds involving R(2m,m) and Hardy–Ramanujan asymptotics, concluding: “Given ε > 0, choose η with 2η < ε and absorb the factor n + 1 into the remaining exponential margin. Since 2c = ρ, this proves the claim.” While standard in expert asymptotic writing, this compression leaves multiple verification points implicit:

  • the exact dependence of constants on η,
  • the uniformity needed when summing over i,
  • the transition from bounds in m to bounds in n,
  • and the scope of “for sufficiently large m” versus “for all nonnegative integers m”.

Likewise, in Section 3 the passage from Guadalupe’s exact formula to the stated remainder estimate depends on bounds such as “|Bk(N)| ≤ k” and “I2(t/s) ≤ s−2 I2(t)” before concluding the error term. This is plausible, but a journal referee would likely want a more explicit derivation, especially because the paper’s principal contribution depends on maintaining exponentially small control rather than merely obtaining a leading term.

The same issue recurs in Section 4, where the manuscript says: “The same calculation to arbitrary order gives a full expansion for M(x). It is valid after any fixed number of derivatives…” That sentence covers a mathematically important escalation—from one expansion to all orders and differentiated versions—but the mechanism is only sketched. This should be expanded into a proposition with hypotheses, explicit bookkeeping of remainder terms, and a clear statement of uniformity.

  1. Reproducibility is presently insufficient for the computational claims.

Section 5 states: “The accompanying Python program uses exact integer arithmetic to verify the identities in Lemma 2.1 for 0 ≤ n ≤ 24. It enumerates bar multisets with exact subset sums and independently enumerates rectangle tilings, deduplicated by piece multiset, for 0 ≤ n ≤ 7.” This is useful, but not enough for reproducibility unless the code is archived, described, and versioned. The paper also states: “These larger-index comparisons concern U, not exact rectangle counts. They are consistency checks and are not proofs of the infinite statements.” That disclaimer is good, but it reveals that the computational section currently plays an ambiguous role: it is neither fully auxiliary nor fully reproducible evidence.

To make this section publication-ready, the author should provide:

  • the exact code as supplementary material,
  • algorithmic complexity notes,
  • a description of how duplicate tilings are eliminated,
  • independent test cases,
  • and a clear separation between proof-supporting finite verification and merely illustrative computations.
  1. The eventual results are strong, but the lack of explicit thresholds weakens impact.

The abstract advertises “eventual strict log-concavity and eventual hyperbolicity of every fixed-degree Jensen polynomial,” while also stating “No explicit threshold for eventuality is claimed.” This is mathematically acceptable, but from a contribution standpoint it creates a tension: the paper proves asymptotic phenomena but does not identify where they start. Section 5 further states: “Determining an explicit threshold, or the full exceptional set, remains outside this draft.”

For some journals this may be fine, but the author should justify why the asymptotic result is valuable even without effective thresholds. That justification is currently underdeveloped. A more persuasive framing would explain whether the main significance lies in:

  • resolving the existence question raised in prior work,
  • introducing a transferable method,
  • or opening a route toward effective bounds.

Without that framing, a reviewer may conclude that the paper closes the question only partially.

  1. The manuscript needs stronger discussion of assumptions and domain limits.

The paper defines “M(x)=F(2x)−F(2x−2) (x > 17/16),” then derives positivity and expansions from an integral representation. This is technically neat, but the draft could better isolate where positivity, derivative expansions, and asymptotic substitutions are valid, especially when moving between real x and integer n. Because the eventual conclusions depend sensitively on sign and decay, a referee may want a cleaner map of dependencies:

  • which claims require only x large,
  • which require x > 17/16,
  • which are integer-only,
  • and which need fixed degree d before n→∞.

Some of this is implicit, for example: “The degree is fixed before the limit; no uniform threshold over all degrees is asserted.” This is an important limitation and should be highlighted earlier, ideally in the introduction or theorem statements.

  1. The manuscript should better defend its anti-fraud / integrity profile through explicit traceability.

I do not see obvious signs of fabricated mathematics. In fact, the manuscript repeatedly narrows its claims, attributes prior results accurately, and distinguishes proofs from consistency checks. Examples include: “it is not a new claim here,” “These larger-index comparisons … are not proofs,” and “A literature search is not a proof of priority.”
Those are positive integrity signals.

Still, a reviewer may flag the draft as vulnerable because it is authored by an “Independent researcher” and relies heavily on recent literature from 2026, including arXiv items. That does not imply any misconduct, but it increases the burden of exposition and verification. The author should therefore add more self-contained derivation detail and, if possible, independent confirmation from a specialist before submission.

Potential Minor Revisions

  1. Typographic and notation clarity issues.
  • In the abstract, the displayed product for the cubic partition generating function appears visually broken in the extracted text: “∏ rotations with allowed. Let a(N) be the cubic partition function, with generating function j≥1 (1 − qj)−1 (1 − q2j)−1.” This may be a PDF extraction artifact, but if the compiled manuscript really has line-break or alignment issues there, they must be fixed because the abstract is the first place a referee tests precision.

  • In Section 1, “We set p(n) = a(n) = 0 for negative integer indices” is clear, but the manuscript uses several layers of notation—p(n), a(N), U(n), F(x), M(x), d(i), b(m), R(N,u), A(n), δ(n), A*(n), δ*(n). A notation table would materially improve readability.

  • In Section 4, the extracted formulas “f(x) = λ√x − 74 log x + …” and “f′′(x) = − λ4 x−3/2 + 47 x−2 + …” plainly represent missing fractional formatting for −7/4 and 7/4. If this is present in the actual typeset PDF, it is a serious typographic problem because it can misstate formulas.

  • Similarly, in Corollary 1.2 the extracted line “log r(n)^2 / r(n−1)r(n+1) = √π / (2√2 n3/2) − 7/(4n2) + O(n−5/2)” appears corrupted in spacing and superscript placement. Recheck all displayed asymptotic formulas in the PDF output.

  1. Section-level presentation could be improved.
  • Section 2 begins directly with decomposition machinery. A brief roadmap sentence explaining why the comparison to cubic partitions is natural would help readers outside the immediate specialty.
  • Section 3 is titled “The Bessel model and its expansion,” but the link between the exact cubic-partition formula and the eventual consequences for rectangle partitions could be signposted more explicitly.
  • Section 5 is useful but should distinguish “finite verification,” “sanity checks,” and “scope limitations” with separate paragraphs or bullet points.
  1. Reference presentation and dependency balance.

The manuscript relies critically on a small cluster of references:

  • Gajdzica–Visser–Zakarczemny for the rectangle partition framework,
  • Guadalupe for cubic partitions,
  • Griffin–Ono–Rolen–Zagier and O’Sullivan for Jensen polynomials. This is fine, but the introduction should explain why these are the essential inputs. Right now, the literature positioning is efficient but sparse.
  1. AI-content analysis.

Estimated percentage of AI-generated content: low to moderate, approximately 10–20%.

Rationale:

  • The text shows mathematically disciplined caveats, such as “it is not a new claim here,” “not proofs of the infinite statements,” and “no uniform threshold over all degrees is asserted,” which are more characteristic of careful human drafting than generic AI overclaiming.
  • The exposition is concise, sometimes excessively compressed, but internally targeted and technically specific.
  • I do not observe obvious large-language-model markers such as repetitive generic praise, inflated novelty language, or concept drift.

Highlighted potentially AI-like sections:

  • The phrase “The contribution proposed here is their quantitative combination, the resulting all-order rectangle expansion, and its log-concavity and Jensen consequences” reads polished and synthetic, but not suspicious on its own.
  • The abstract is highly compressed and slogan-like, which can resemble AI condensation, but in mathematics that style is also common.

Assessed epistemic impact:

  • Even if minor AI assistance was used for prose polishing, I see no evidence from the manuscript text alone that it materially undermines the mathematical claims.
  • The main epistemic risk is not AI authorship but proof compression and insufficient independent verification of the computations.

Recommendations

  1. Add a dedicated “Novelty and contribution” subsection in the introduction that explicitly states the manuscript’s unique theorem-engineering step and why it is nontrivial.

  2. Expand the proofs of the main asymptotic error estimate and the all-orders differentiated expansion. In particular, expose the constant dependence, uniformity conditions, and justification for each asymptotic passage.

  3. Provide a notation table and re-typeset all displayed formulas carefully to eliminate ambiguous fraction rendering.

  4. Upgrade Section 5 into a reproducibility section:

  • archive the Python code,
  • describe algorithms,
  • report how deduplication is implemented,
  • and specify which computations are verification versus illustration.
  1. Strengthen the discussion of limitations by explicitly foregrounding:
  • no explicit threshold for eventual log-concavity,
  • no uniform threshold across Jensen degrees,
  • and the dependence on prior exact formulas and general theorems.
  1. Before submission, seek independent expert checking on both novelty and the technical derivations, especially because the manuscript itself acknowledges that this remains necessary.

Overall, my assessment is that this is a serious and potentially publishable mathematical draft, but not yet submission-ready. Its main risks are not incoherence or obvious fabrication; rather, they are under-argued novelty, compressed proof exposition, and incomplete reproducibility scaffolding.

XSci (AI review)

November 14, 2026

Summary

Let r(n)=p(2,n)r(n)=p(2,n) count the multisets of integer-sided rectangles that tile a 2×n2\times n rectangle, and let a(N)a(N) be the cubic partition function. The authors observe that the difference U(n)=a(2n)−a(2n−2)U(n)=a(2n)-a(2n-2) approximates r(n)r(n) with an error that is exponentially small relative to the main term: Theorem 1.1 gives 0≤U(n)−r(n)=Oε(e(ρ+ε)n)0\le U(n)-r(n)=O_\varepsilon(e^{(\rho+\varepsilon)\sqrt n}) with ρ=2π/3\rho=2\pi/\sqrt3, against a main term of size eλne^{\lambda\sqrt n}, λ=π2\lambda=\pi\sqrt2. Feeding this into a Rademacher-type exact formula for a(N)a(N) yields a Bessel model M(x)M(x) with r(n)=M(n)(1+O(e−κn))r(n)=M(n)(1+O(e^{-\kappa\sqrt n})) and an asymptotic expansion of r(n)r(n) to every order. The authors deduce eventual strict log-concavity (Corollary 1.2), eventual hyperbolicity of every fixed-degree Jensen polynomial (Corollary 1.3), and an eventual Bessenrodt–Ono inequality (Corollary 4.1).

Contribution and Significance

The contribution is the comparison itself. Replacing r(n)r(n) by a(2n)−a(2n−2)a(2n)-a(2n-2) is not an obvious move, and it works for a structural reason the authors exploit precisely: the telescoping d(i)=p(i)−p(i−1)d(i)=p(i)-p(i-1) produced by the cubic generating function counts partitions with no part equal to 11, which is exactly what the geometric decomposition (8) of [2] requires. What matters is that the comparison is accurate enough. A leading-order equivalent — already Theorem 1.5 of [2] — cannot give log-concavity, because differences of consecutive terms sit below the error, whereas an error exponentially small in n\sqrt n relative to the main term is below every fixed power of nn, and is the right input for the Griffin–Ono–Rolen–Zagier and O'Sullivan machinery. Section 3's closing paragraph and (14) handle the one genuine trap — that r(n)r(n) exists only at integers, so its remainder cannot be differentiated — by differentiating the smooth model MM instead.

Relative to the closest prior work the novelty is real but narrow. The decomposition (8)–(9), the subsum estimate [3], the exact formula [4] and the Jensen theory [6,7] are all quoted off the shelf, as the authors state in Section 5. Gajdzica, Visser and Zakarczemny raise log-concavity and Bessenrodt–Ono-type inequalities for p(2,n)p(2,n) as open directions, and this manuscript settles both in their eventual form; I found no prior treatment. The advance is a first, correct entry on a stated open question — a solid increment rather than a change in what the field can do.

Strengths

  • The comparison in Lemma 2.1 and (2) is clean and fully checkable: a(2n)=∑ip(i)p(2(n−i))a(2n)=\sum_i p(i)p(2(n-i)), the telescoping, d(i)≤p(i)d(i)\le p(i), i+n−i≤2n\sqrt i+\sqrt{n-i}\le\sqrt{2n} and 2 π2/3=2π/3\sqrt2\,\pi\sqrt{2/3}=2\pi/\sqrt3 all hold as stated, and κ0=λ−ρ>0\kappa_0=\lambda-\rho>0 is exactly the margin the argument needs.
  • The integral representation (11) makes both the positivity of MM and its expansion transparent, and the Bessel derivative identity of [5, §10.29] closes under iteration, which is what makes an all-order expansion available at all.
  • The constants are right. I reproduced C=π21/4/32C=\pi2^{1/4}/32 and c1=−(70+9π2)/(16π2)c_1=-(70+9\pi^2)/(16\pi\sqrt2) independently from y=2x−9/8y=2x-9/8, and confirmed the expansion (6) of Corollary 1.2 numerically to the stated order O(n−5/2)O(n^{-5/2}).
  • Section 5 is honest and exactly reproducible. I recomputed every entry of the table, including the Lemma 2.1 bounds 2626, 413413, 2857028570 and 115677115677, and the exceptional set n=1,2,5,7n=1,2,5,7; all agree.
  • The authors are explicit about what is not proved: no threshold is claimed, and the large-index comparisons are identified as consistency checks on UU, not on rr.

Recommendation

Major Revision for the Ramanujan Journal. The advance is modest but genuine, in scope, and answers a question posed in the literature, so the grounds for rejection do not apply; the central claims are, as far as I can check, correct. What stands in the way is that the step on which both corollaries depend is asserted rather than proved, and that the two theorems carrying Theorem 1.1 are paraphrased rather than stated, so neither half of the main result is verifiable from the manuscript alone. Both are repairable within the present approach and require no new mathematics.

Major Concerns

  1. The derivative expansions are asserted, not proved (Section 3, final paragraph; Section 4, opening). The entire content of Section 4 rests on (12), (13) and the claim that f(k)(x)=Ok(x1/2−k)f^{(k)}(x)=O_k(x^{1/2-k}) for each fixed k≥1k\ge1, where f=log⁡Mf=\log M. The justification offered is two sentences: "It is valid after any fixed number of derivatives: differentiate (11) first, use the Bessel derivative identities, and apply the corresponding large-argument expansions." There is no statement of what is being claimed, no verification that the substitution of the I3I_3 expansion into (11) may be integrated term by term uniformly for −1<s<1-1<s<1, and no passage from an expansion of M(k)M^{(k)} to one of (log⁡M)(k)(\log M)^{(k)}. I believe the assertion is correct — differentiating G(v)=π42v−1I2(πv)G(v)=\frac{\pi}{4\sqrt2}v^{-1}I_2(\pi\sqrt v) repeatedly via [5, §10.29] gives G(j)(v)=π42(π/2)jv−(2+j)/2I2+j(πv)G^{(j)}(v)=\frac{\pi}{4\sqrt2}(\pi/2)^{j}v^{-(2+j)/2}I_{2+j}(\pi\sqrt v), so the structure does close and [5, §10.40] applies at every order — but a reader should not have to supply this. Please state it as a lemma ("for each fixed kk, M(k)(x)M^{(k)}(x) admits an expansion to every order, uniformly …") and prove it, then derive (12), (13) and the bound on f(k)f^{(k)} from it.

  2. The two cited theorems carrying Theorem 1.1 are paraphrased, not stated (Section 2, proof of (2); Section 3, Lemma 3.1). For (2) the authors write that "Theorem 2 of Erdős, Nicolas and Sárközy [3], applied with N=2mN=2m and u=mu=m" gives R(2m,m)≤p(m)exp⁡((2m)7/15)R(2m,m)\le p(m)\exp((2m)^{7/15}), and that "the required range N5/7<u≤N/2N^{5/7}<u\le N/2 holds eventually", without reproducing the hypotheses or conclusion of that theorem. For Lemma 3.1 the authors "specialize Theorem 1 of Guadalupe [4] to the prime 2" and then extract specific structural facts — the odd-kk argument z/kz/k against the even-kk argument 2z/k\sqrt2z/k, and the assertion that "the coefficients in that theorem are sums of unit-modulus complex numbers, so ∣Bk(N)∣≤k|B_k(N)|\le k, and B1(N)=1B_1(N)=1" — none of which appears in the manuscript. These two inputs carry (2) and (10) respectively, that is, both halves of Theorem 1.1, and a reader cannot check either without fetching two other papers and performing the specialization unaided. Please quote both source statements in the form used, and write out the specialization of [4] to p=2p=2 explicitly, including where the bound on Bk(N)B_k(N) comes from.

Minor Comments

  1. No effective threshold is obtained, and the two places where effectivity is lost — the "for sufficiently large mm" in [3] and the unspecified BηB_\eta of Section 2 — should be identified, since the benchmark precedent is effective (J.-L. Nicolas, Sur les entiers NN pour lesquels il y a beaucoup de groupes abéliens d'ordre NN, Ann. Inst. Fourier 28 (1978), 1–16; S. DeSalvo and I. Pak, Log-concavity of the partition function, Ramanujan J. 38 (2015), 61–73, neither of which is cited).
  2. Convergence is slow and deserves a remark: M(n)/r(n)M(n)/r(n) is still 1.1371.137 at n=24n=24, consistent with κ0≈0.815\kappa_0\approx0.815.
  3. The table shows the Lemma 2.1 bound exceeding U(n)−r(n)U(n)-r(n) by only about 13%13\% at n=24n=24, far tighter than eρne^{\rho\sqrt n} suggests; a sentence on the conjectural true order would be welcome.
  4. "The accompanying Python program" (Section 5) is never provided — please deposit it with a link or DOI, since Section 5 is the only empirical content.
  5. Section 5 names "the eventual Bessenrodt–Ono conclusion" with no reference; add C. Bessenrodt and K. Ono, Maximal multiplicative properties of partitions, Ann. Comb. 20 (2016), 59–64.
  6. The origin of the cubic partition function is uncited; add Hei-Chi Chan, Ramanujan's cubic continued fraction and an analog of his "most beautiful identity", Int. J. Number Theory 6 (2010), 673–680.
  7. Corollary 4.1 answers half of the question raised in [2] but surfaces only in Section 4; it belongs in the abstract and Section 1.
  8. Theorem 1.1 bundles (2), the definitions of FF and MM, (3), (4) and (5); splitting it would help the reader see which statement needs which input.
  9. The header "Research draft — 7 November 2026" and the closing sentence of Section 5, "this draft requires independent mathematical and novelty review before submission", must be removed, along with "remains outside this draft" in the preceding paragraph.
  10. Section 5 should say which range is independently verified: direct tiling enumeration reaches only n≤7n\le7, while the table extends to n=24n=24 via (8).
  11. Please add a remark on whether the comparison device extends to m×nm\times n for m≥3m\ge3, where (8) is unavailable, given that [9] and [10] are mentioned only in passing.
  12. Reference [3] misspells the third author as "Sárcözy" (correct in the text of Section 2), and the DOI in [1] should read 10.1112/plms/s2-17.1.75; in Section 1, cjc_j, c0c_0, dd and Hd\mathcal{H}_d are set in roman rather than math.