{"id":"5205b23b-1944-4502-ad63-44a491d18587","arxiv_id":"2601.01905","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"This paper proves Δ(x)-xδ(x)=1/4+O(x^{-1/4}log x) explicitly, and unconditionally settles the positivity of ∫_x^∞ Δ(u)/u^2 du.","lead":"This paper proves an explicit formula for a smoothed Dirichlet divisor sum, with error O(x^{-1/4} log x), and uses it to settle a conjecture on the sign of an integral of the divisor error term. The result is fully unconditional and gives explicit constants for all x≥1.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 1's second branch constant 0.236 is not supported by the proof; the coefficient is about 0.237. No deeper flaw found.","rationale":"The central qualitative result—Δ(x) − xδ(x) = 1/4 + O(x^{-1/4} log x)—is well supported by the explicit lemmas, and the proof of Corollary 2 through Proposition 1 is a legitimate van der Corput estimate with no obvious structural error. The main concrete weakness is the constant mismatch in Theorem 1's second branch: the final displayed estimate in §2.4 yields a √x coefficient no smaller than about 0.237, not 0.236. This is exactly the kind of small but real error the reader's verdict flagged, and it justifies a conditional accept with a requested correction. I do not regard Corollary 2 as a load-bearing failure; the bound is weak and the argument is standard. The reader's structured 'weakest assumption' points at Corollary 2, but the actual obstacle to accepting Theorem 1 as printed is the constant. Hence partial agreement.","tokens_in":10598,"tokens_out":41513,"duration_ms":375576,"concrete_test":"Recompute §2.4's final coefficient directly from Lemmas 1, 2, and 10: take α = 1/8, β ≤ 0.033 (or the sharper c/3(2 + 5e^{-5}) ≈ 0.0326), and √300 f(300) ≈ 0.0465. If α + 2β + 0.047 > 0.236, the theorem's printed constant should be corrected. This settles whether the second branch is exactly as stated.","verdict_should_be":"UNCHANGED","load_bearing_attack":"In the final step of §2.4, the second bound is |r(x)| ≤ log x / x^{1/4} + (α + 2β + 0.047)/√x + α²/x, with α ≤ 1/8 and β ≤ 0.033 from Lemmas 1 and 2, and 0.047 from Lemma 10. Thus the coefficient is at most 0.238, and with the sharpest numerical values (β ≈ 0.0326, √300 f(300) ≈ 0.0465) it is about 0.2368. The printed value 0.236 is smaller than anything the displayed inequalities can justify. Consequently Theorem 1 as stated is not fully proven for that constant. This is a minor fix—replacing 0.236 by 0.238 (or 0.237) preserves the theorem and does not affect Corollary 1, which uses only the first, constant-order branch. The reliance on Corollary 2, the effective generalized Chowla–Walum bound, is standard and appears internally sound; the van der Corput argument checks out in structure and constants.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the difference between the Dirichlet divisor summatory error term Δ(x) and its logarithmic analogue xδ(x). The main result, Theorem 1, asserts the explicit bound Δ(x)−xδ(x)=1/4+r(x) with |r(x)|≤1/8+0.316/√x+1/(64x) for all x≥1 and |r(x)|≤log x/x^{1/4}+0.236/√x+1/(64x) for x≥300. From this the authors derive Corollary 1, namely ∫_x^∞ Δ(u)/u^2 du≥0 for all x≥1, settling a conjecture of Berkane, Bordellès, and Ramaré. The proof combines Euler–Maclaurin estimates for harmonic-type sums, an explicit version of a generalized Chowla–Walum sum, and a van der Corput argument; all constants are claimed explicitly.","tokens_in":10832,"tokens_out":49165,"duration_ms":430614,"significance":"If the proof is fully correct, this is a substantial contribution: it gives an unconditional, explicit O(x^{-1/4} log x) estimate for a smoothed divisor problem, resolves a conjecture in the literature, and provides a useful conversion between Δ and δ. The paper is careful in proving most auxiliary lemmas and in making constants explicit. The main structural idea — that the smoothing eliminates the difficult fractional-part sum — is elegant and likely correct. However, as discussed below, several load-bearing points in the present manuscript need to be repaired before the stated results are fully established.","major_comments":[{"comment":"The second branch of Theorem 1 states the coefficient 0.236, but the proof gives only 0.238. From the displayed inequality |r(x)|≤log x/x^{1/4}+(α+2β+0.047)/√x+α²/x and Lemmas 1–2 and 10, one has α≤0.125, β≤0.033, so α+2β+0.047≤0.238. The printed value 0.236 is smaller than anything justified by the displayed inequalities. Replacing 0.236 by 0.238 (or sharpening one of the constituent bounds) is necessary.","section":"Theorem 1, §2.4"},{"comment":"The claimed identity Δ(x)/x−δ(x)=∫_x^∞ Δ(u)/u^2 du is not a direct consequence of summation by parts. Writing A(x)=∑_{n≤x}τ(n), summation by parts gives ∑τ(n)/n = A(x)/x + ∫_1^x A(u)/u^2 du, which leads to Δ/x−δ = −∫_1^x Δ(u)/u^2 du + γ²−2γ−2γ1+1 before one knows ∫_1^∞ Δ(u)/u^2 du = γ²−2γ−2γ1+1 (equivalently δ(x)→0). This nontrivial evaluation must be supplied, or the proof of Corollary 1 is incomplete. The corollary is load-bearing for the paper's main application.","section":"Proof of Corollary 1, §2.4"},{"comment":"The passage from Lemma 8 to Eq. (8) is not correct as written. With λ2=mx/(4N^3) and c2=8, the bound supplied by Lemma 8 has an additional factor 1/√π, and the displayed intermediate term √(mx)/N should be √(mx)/√N in order to match the subsequent x^{1/2}N^{β−1/2} term. As written, the inequality chain is invalid, so Proposition 1 and Corollary 2 are not established. Since Lemma 10 and the second branch of Theorem 1 use Corollary 2, the authors need to reconcile Lemma 8 with its application and provide a correct explicit constant.","section":"§2.3.3, Eq. (8)"}],"minor_comments":[{"comment":"The second exponent in the displayed bound after reporting (8) is printed identical to the first; it should be β/α + 3/(2α) − 1/2, not β/α − 1/(2α) + 1/2.","section":"§2.3.3, after Eq. (8)"},{"comment":"The statement of Lemma 8 appears to contain a typo: the proof gives 4/√π, while the displayed statement reads 4√π in the typeset text. This ambiguity should be corrected, as it affects the verification of the subsequent estimates.","section":"Lemma 8 and its application"},{"comment":"The notation R_1(t) is reused for R_1(t)−ψ(t)/t, which is confusing. A separate symbol or a clear definition would improve readability.","section":"Lemma 4 proof"}],"recommendation":"major_revision","confidential_remarks":"The central idea and most of the structure seem sound, and the issues in Theorem 1's constant and Corollary 1's proof are local fixes. My main uncertainty is the van der Corput step: if the apparent mismatch between Lemma 8 and Eq. (8) is only a typesetting/constant-display problem, a careful rewrite should suffice; if not, the proof of Corollary 2 and hence the second branch of Theorem 1 need substantive repair. I would ask the authors to provide a complete, line-by-line derivation of Eq. (8) from a correctly stated van der Corput lemma."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: the paper is a legitimate explicit result in a crowded area, and it does what it says. The new contribution is the fully explicit, unconditional estimate Δ(x) − xδ(x) = 1/4 + O(x^{−1/4} log x), with constants valid for all x ≥ 1, and the consequent proof of the Berkane–Bordellès–Ramaré positivity conjecture for ∫_x^∞ Δ(u)/u^2 du. The asymptotic version was already claimed by Mahatab and Mukhopadhyay, but the authors cite that only through a personal communication [14]; the explicit treatment here is their own.\n\nStructurally the proof is clean. Lemmas 1 and 2 give Euler–Maclaurin bounds with explicit constants. Proposition 1 is a general explicit bound for generalized Chowla–Walum sums, proved with a dyadic van der Corput argument; Corollary 2 is the concrete effective bound. Lemma 10 converts that into the required O(x^{−1/4} log x) contribution. Lemma 3 is borrowed from [2], a published paper, so the self-citation is not circular. The final assembly in Theorem 1 is straightforward and the main identities check out. This is honest, careful work.\n\nThe one real flaw is the constant in the second branch of Theorem 1: the text prints 0.236, but the proof at that point gives coefficient α + 2β + 0.047 with α = 1/8, β ≈ 0.033, so about 0.2368–0.238. The printed 0.236 is slightly smaller than the displayed inequalities justify. That is a minor numerical correction; replace it with 0.237 or 0.238 and the theorem stands. Corollary 1 is unaffected.\n\nA smaller citation concern: the claim that Mahatab–Mukhopadhyay settled the conjecture asymptotically rests on [14], 'Personal Communication.' That is unverifiable. A referee should ask the authors to either locate a public version or soften the wording. It does not affect the mathematics of this paper.\n\nWho is this for: people working on explicit estimates for the divisor problem. It is not a breakthrough but it is a solid, complete result, and the constant-order branch is strong enough to be usable. I would send it to a serious referee and expect acceptance after the constant fix and a reference cleanup.","headline":"A genuinely explicit and mostly sound proof of a real conjecture, with a small constant typo in the second branch of Theorem 1 that is trivial to fix; worth refereeing.","tokens_in":11332,"tokens_out":2509,"would_cite":true,"duration_ms":28232,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["11A25","11L07"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper proves that the difference between the summatory function of the divisor function and its logarithmic version equals exactly 1/4 plus an error term that is O(log x / x^{1/4}) for x ≥ 300, with explicit constants, and derives a po","keywords":["Dirichlet divisor problem","hyperbola principle","exponential sums","van der Corput estimates","Chowla–Walum sums","Bernoulli polynomials","explicit bounds","fractional parts"],"falsifier":"Compute the sum ∑_{n≤√x} n B_2({x/n}) numerically for a few large values of x (e.g., 10^4, 10^6, 10^8) and verify it stays below x^{3/4} log x; a violation would falsify the stronger error bound. Alternatively, evaluate Δ(x) − xδ(x) − 1/4 at those x and check that it satisfies the claimed inequalities, or search numerically for an x ≥ 1 where ∫_x^∞ Δ(u)/u^2 du is negative.","tokens_in":10467,"feed_emoji":"📊","tokens_out":8393,"duration_ms":60993,"temperature":0.7,"pith_summary":"This paper establishes that the difference between the divisor summatory function and its logarithmic version is exactly 1/4 plus an error term that stays bounded for all real x ≥ 1 and shrinks like O(x^{-1/4} log x) once x ≥ 300. That is far smaller than the error term in the classical Dirichlet divisor problem, whose order is believed to be at least x^{1/4}. The result is proved with explicit constants, using the hyperbola principle, Euler–Maclaurin estimates, and new explicit bounds for weighted sums of second Bernoulli polynomials. As a consequence, the integral of the divisor error term from x to infinity is shown to be non-negative for every x ≥ 1, confirming a conjecture. The closeness of the two error terms means that results about one can be converted to results about the other with almost no loss.","feed_headline":"Divisor-sum gap pinned at 1/4 with explicit error bounds","feed_subtitle":"New explicit bounds settle a positivity conjecture for the divisor error integral.","key_machinery":"The identity Δ(x) − xδ(x) = 1/4 + x R_1(√x)^2 − 2√x R_1(√x) ψ(√x) − 2x (R_2 − R_1)(√x) − 2x ∑_{k≤√x} R_1(x/k)/k, obtained from the hyperbola principle, expresses the difference as a sum of small error terms. The decisive new estimate is an explicit bound for the generalized Chowla–Walum sum ∑_{n≤√x} n B_2({x/n}) < x^{3/4} log x for x ≥ 300, proved by splitting the sum into dyadic intervals and applying an explicit van der Corput bound to the resulting exponential sums. This converts an O(1) term into O(x^{-1/4} log x), which is the key to the sharper error term.","core_discovery":"The central claim is that Δ(x) − x δ(x) = 1/4 + r(x), where Δ(x) is the usual sum of τ(n) up to x and δ(x) is the corresponding sum of τ(n)/n with its main term removed. The paper proves fully explicit bounds on r(x): |r(x)| ≤ 1/8 + 0.316/√x + 1/(64x) for all x ≥ 1, and the sharper |r(x)| ≤ (log x)/x^{1/4} + 0.236/√x + 1/(64x) for x ≥ 300. The constant 1/4 is ζ(0)^2, the residue of the Dirichlet series of τ(s)=ζ(s)^2. The proof is elementary in structure and yields the corollary that ∫_x^∞ Δ(u)/u^2 du ≥ 0 for all x ≥ 1.","pith_inferences":["The appearance of ζ(0)^2 as the constant suggests a general principle: for arithmetic functions whose Dirichlet series has a pole of order k at s = 1, the difference between the summatory function and its logarithmic analogue may equal the square of the residue of the zeta factor, possibly with oscillations when nontrivial zeros exist, as seen in the Möbius analogue in the paper.","The explicit exponential-sum technique used here may generalize to higher Bernoulli polynomials or other weights, producing effective estimates for other smoothed divisor problems that are currently conditional.","A natural test is whether the logarithmic factor in the error term can be removed via a more refined exponential-sum estimate, which would sharpen the bound to O(x^{-1/4}) with explicit constants."],"forward_implications":["The integral ∫_x^∞ Δ(u)/u^2 du is non-negative for every real x ≥ 1, confirming a conjecture about the divisor error term.","The classical result that the divisor error term is not o(x^{1/4}) transfers to the logarithmic version, so its error term is also not o(x^{1/4}).","Bounds on one error term convert to bounds on the other with almost no loss; the paper shows a bound of order x^{1/2} on Δ implies a bound of order x^{-1/2} on δ, improving earlier explicit results by a factor of 2.","For large x, the two error terms differ from 1/4 by O(x^{-1/4} log x), so they are interchangeable in asymptotic formulas up to that precision."],"fun_headline_variants":["Smoothed divisor error constant: exactly 1/4","Explicit error bounds settle divisor integral conjecture","Divisor sum gap fixed at 1/4 with sharp estimates","New bound resolves divisor error positivity question"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The sharper error term O(x^{-1/4} log x) in the theorem depends on the explicit bound |∑_{n≤√x} n B_2({x/n})| < x^{3/4} log x for x ≥ 300, proved with van der Corput exponential-sum estimates; if that bound were to fail, the x^{-1/4} log x branch would collapse, although the constant-order bound for all x ≥ 1 would remain intact.","fun_headline_variants_meta":{"raw":{"variants":["Smoothed divisor error constant: exactly 1/4","Explicit error bounds settle divisor integral conjecture","Divisor sum gap fixed at 1/4 with sharp estimates","New bound resolves divisor error positivity question"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000188,"raw_usage":{"total_tokens":1162,"prompt_tokens":730,"completion_tokens":432,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":474,"completion_tokens_details":{"reasoning_tokens":369}},"tokens_in":474,"tokens_out":432,"duration_ms":4099,"temperature":1.0,"reasoning_tokens":369,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T12:40:55.675560+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the sum ∑_{n≤√x} n B_2({x/n}) numerically for a few large values of x (e.g., 10^4, 10^6, 10^8) and verify it stays below x^{3/4} log x; a violation would falsify the stronger error bound. Alternatively, evaluate Δ(x) − xδ(x) − 1/4 at those x and check that it satisfies the claimed inequalities, or search numerically for an x ≥ 1 where ∫_x^∞ Δ(u)/u^2 du is negative.","supporting_citations":[],"review_version":1}