{"id":"9f75ebe6-a92d-4e51-b3a6-5c05dfc72629","arxiv_id":"2607.21883","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"Buchstab's function ω(u) is shown to equal 2|Φ_i(u)| cos(arg Φ_i(u)) plus an explicitly bounded error, yielding easy-to-evaluate two-sided bounds for u ≥ 3.","lead":"This paper gives explicit upper and lower bounds for Buchstab's function, which counts integers with no small prime factors, so the function can be evaluated from a formula instead of solving a delay differential equation. The bounds carry controlled error terms, and the main formulas use standard special functions such as the incomplete gamma function and Lambert W.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The u≤5500 range of Theorems 1–3 rests on unaudited numerical checks (contour plots, graphing, Mathematica NIntegrate, DDE solver) with no code/data; this is a reproducibility gap, not an identified mathematical error.","rationale":"I read the proof as a hybrid derivation. Equations (5)–(10) are internally coherent: the saddle point is at s=-ζ, the Taylor expansion around τ=0 with the h(u,τ) remainder is standard, the bounds in Lemmas 4 and 5 are explicit, and the final constant 0.005/u^2 is obtained by elementary arithmetic from (10). I found no circularity and no hidden parameter fitting; the main term has one natural parameter (the Lambert W branch) but no fitted constants. The dependency that is least secure is the set of finite-range verifications. The paper does not use them merely as sanity checks; they are the proof for the closing intervals. Statements such as 'we graph the expression ... to verify this claim' (proof of Theorem 2, u≥860), 'we verify ... by graphing' (Lemmas 3, 5, 7, Corollary 1), and 'we verify ... by numerical integration in Mathematica' (Section 4) are explicit assertions of unstated numerical evidence. Since no code, notebooks, or data files are provided, an independent reader cannot check the decisive finite computations. This does not by itself falsify the theorems, but it justifies conditional acceptance pending release of reproducible, ideally interval-arithmetic, verification. My conclusion agrees with the Reader's weakest assumption; I did not find a stronger internal flaw that would change the verdict.","tokens_in":11780,"tokens_out":6035,"duration_ms":61229,"concrete_test":"Run an independent, rigorous numerical verification of the decisive finite-range inequalities using interval arithmetic: for Theorem 2, evaluate W(u) for u in [6,5500] with an interval Taylor solver for the DDE (u≤50) and with rigorous quadrature of (7)/S(u) for u≥50 with a posteriori error bounds, and check |θ2(u)|<420u^{-6} for 6≤u≤19 and |θ2(u)|<0.005u^{-2} for 16≤u≤5500. Also interval-check Lemma 5's rectangle 50≤u≤10^4, 4≤τ≤370 and Lemma 7's domains. If all enclosures satisfy the claimed inequalities with margin, the central concern is resolved; if a rigorous enclosure violates a bound, Theorems 1–3 need range restrictions.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The load-bearing concern is the completeness of the proof for all u claimed, not the saddle-point analysis itself. The analytic argument proves (9) only for u≥10^3, and the stated 0.005u^{-2} bound in Theorem 2 is derived only for u≥5500; Theorem 1's 1/(12u log u) bound is derived only for u≥10^3; Theorem 3's small-u ranges are numerical. Everything below those cutoffs is justified by finite-range checks that are described, not supplied: Lemma 3 verifies (2) in a rectangle by contour plot; Lemma 5 verifies two inequalities 'by graphing' on 50≤u≤10^4, 4≤τ≤370 and 50≤u≤10^3, 1≤|τ|≤4; Lemma 7 verifies |f(x+iy)|<1 for 5≤x≤10 by contour plot and the claim for 2≤u≤25 by graphing; Section 4 uses the Marsaglia–Zaman–Marsaglia algorithm for u≤50 and Mathematica numerical integration of S(u) for 50≤u≤5500, with no error bounds and no output. If any of these checks is wrong or merely imprecise, the corresponding range of the theorems is unsupported. This is a verification/reproducibility gap, not a detected inconsistency in the analytic core.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper establishes explicit, easily evaluated upper and lower bounds for Buchstab's function W(u)=ω(u)−e^{-γ}, using the saddle-point method with a carefully chosen complex valued saddle point ζ=ζ(u) satisfying e^ζ=-uζ. The main results are Theorems 1–3, which express W(u) as 2|Φ_j(u)|{cos(arg Φ_j(u))+θ_j(u)} with explicit numerical error bounds; Theorem 2 is the most refined two-term approximation, and Theorem 3 adds a further β(u) term. The claimed advantage is that the resulting bounds involve only the incomplete gamma function Γ(0,s) and the Lambert W function, so no numerical solution of the delay-differential equation is required. The analytic saddle-point derivation in §2 is detailed, with Taylor polynomials of order seven, explicit lemmas estimating the derivatives of J(s), and large-u error terms derived. However, the full range of the theorems depends substantially on finite-range numerical verifications described in the text as 'graphing', 'contour plots', and Mathematica numerical integration, without supplying code, data, or rigorous error bounds.","tokens_in":12082,"tokens_out":2785,"duration_ms":28625,"significance":"If fully supported, the paper would be a valuable contribution: it gives simple, explicit two-sided estimates for Buchstab's function that improve on Hildebrand's asymptotic formula and avoid numerical DDE solving. The main-term formulas are parameter-free, use standard special functions, and are easy to evaluate in Mathematica or PARI/GP. The explicit constants, e.g. |θ_2(u)|<0.005u^{-2} for u≥16, are strong enough for applications such as Maier-type prime gap estimates. The analytic core appears sound and is a nontrivial extension of the author's earlier Dickman-function work to the complex saddle-point setting. The main weakness is not the analytic derivation for large u, but the reproducibility and verifiability of the finite-range checks that close the gap between the large-u proof and the claimed full range u≥3.","major_comments":[{"comment":"The proof of Theorem 2 for 6≤u≤5500 relies entirely on numerical computation: the Marsaglia–Zaman–Marsaglia DDE solver for u≤50 and Mathematica numerical integration of S(u) for 50≤u≤5500. The analytic bound |θ_2(u)|<0.005u^{-2} is derived only for u≥5500, so the theorem as stated for all u≥16 (and particularly the 16≤u≤5500 range) is supported only by these unaudited computations. Please supply the actual code, output data, and rigorous error bounds (e.g. interval arithmetic or explicit quadrature error estimates) for these verifications. Without that, the full-range claim is not reproducible.","section":"§4, proof of Theorem 2"},{"comment":"The proof of (2) in the rectangle 0≤x≤200, 3≤y≤400 is asserted to follow from a contour plot, and this is used to establish the bound Re(J(−ζ))≥u for u≤80. Since this finite-range check is load-bearing for the small-u behavior of the later estimates, the contour plot should be replaced by a machine-checkable verification, or by a rigorous explicit bound valid in that rectangle.","section":"Lemma 3"},{"comment":"The inequalities Re(H(u,τ))/u≤−0.39 for 50≤u≤10^3, 1≤|τ|≤4, and Re(H(u,τ))<−8u/log^2 u for 50≤u≤10^4, 4≤τ≤370, are verified 'by graphing'. These estimates are not peripheral: they control the entire contribution of the integration ranges |τ|≥δ in the saddle-point argument and are used to derive (9) for u≥10^3. Please supply a rigorous or reproducible verification (code, data, or exact bounds) for these rectangles.","section":"Lemma 5"},{"comment":"The bound |f(ζ)|<1 is verified by a contour plot for 5≤x≤10, and the inequality in Lemma 7 is checked by graphing for 2≤u≤25. This lemma is used in the proof of Theorem 1 and in the estimate |1+α(u)|>1−1/(10u), which is needed for the coefficient 0.005 in Theorem 2. Similarly, Corollary 1's verification for 3≤u≤6 is done by graphing a quotient. These finite-range verifications should be made explicit and machine-checkable, since they are part of the logical proof of the stated theorems.","section":"Lemma 7 and Theorem 1/Corollary 1"}],"minor_comments":[{"comment":"The notation 'a: Thm. 1' and 'a: Thm. 2' is a little cryptic; please clarify that the tabulated entries are a×10^{-b} with a truncated/rounded as stated.","section":"§1, Table 1"},{"comment":"The quantity f(u) is introduced as u^3(E_2(u)+∑_{k=1}^6 E_k(u)), but the definition of E_k(u) for k=1,...,6 is spread over the proof; a short summary equation would improve readability.","section":"§2.2"},{"comment":"The ranges '18≤u≤10^3' and 'u≥10^3' overlap at u=10^3; since the two bounds differ, please clarify which bound is intended at u=10^3.","section":"Theorem 3"},{"comment":"Figure 1 is referenced in the text but no actual figure appears in the manuscript. Please include the plot or remove the reference.","section":"General"},{"comment":"The statement 'The proof of Theorem 2 shows that the last integrand approaches the standard normal density function' is informal; since the paper is about rigorous bounds, a precise quantified statement would be preferable.","section":"§4"}],"recommendation":"major_revision","confidential_remarks":"The reader's stress-test concern lands: the analytic saddle-point derivation is credible, but the universal quantification over u in Theorems 1–3 currently depends on a collection of graphical and numerical checks that are described but not supplied. This is a reproducibility gap rather than a detected mathematical error, and it is fixable within the manuscript's scope by including code/data or replacing the checks with rigorous interval-arithmetic certificates. I do not recommend rejection, but the revision must make these finite-range verifications auditable."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nBottom line: this paper gives the first numerically explicit, easy-to-evaluate two-sided bounds for W(u)=omega(u)-e^{-gamma}, with error terms of order 1/(u log u) and 1/u^2. The main term uses the Lambert W branch and incomplete gamma function, so it is straightforward to compute in Mathematica or PARI/GP. The saddle-point derivation is careful, with Taylor expansions to order seven, explicit lemmas, and no fitted parameters. The correction terms are explicit rational functions of zeta. That is a real step beyond Hildebrand's asymptotic O(1/u) and Cheer-Goldston's numerical extrema.\n\nWhat the paper does well: the large-u analysis (roughly u >= 10^3 for Theorem 1, u >= 5500 for Theorem 2) is self-contained and convincing. Lemma 4 gives sharp bounds for J^{(k)}; Lemma 5 controls the decay of the Laplace integral away from the saddle point; the Taylor expansion around the saddle is handled carefully. Corollary 1 gives a clean bound |W(u)| < 2|Phi(u)| for all u >= 3. I agree with the reader that there is no circularity — the parameters are not fitted to the target values.\n\nWhere it gets shaky: the proof of the theorems for the full stated range depends on a long list of finite-range verifications done \"by graphing\", by contour plots, or by Mathematica numerical integration — Lemmas 3, 5, 7, the proof of Theorem 2 for 6 <= u <= 5500, and Section 4 for u <= 50 via the Marsaglia-Zaman-Marsaglia DDE solver. None of these checks is accompanied by code, output, or error bounds. If any one of them is wrong or merely imprecise, the corresponding range of the theorems is unsupported. The author is transparent that these are numerical checks, not analytic proofs, but the reader cannot audit them. This is a reproducibility gap, not an identified mathematical error. For a paper whose selling point is \"no need to solve the DDE\", relying on an unprovided DDE solve for u <= 50 and unprovided Mathematica integration for the intermediate range is a little ironic.\n\nMy take: the analytic core is solid and the small-range checks are probably right — the author has a track record with the analogous Dickman paper. But the absence of code and data makes the full-range theorem less than fully verified. For a serious referee, I would ask the author to supply a notebook with the numerical verifications, or better, replace the contour-plot arguments with interval-arithmetic or polynomial-certificate checks.\n\nThe paper should go to peer review, but with a request for the verification artifacts. It is a useful, citable tool for sieve theory and Maier-type applications.\n\nBest,\n[Sign-off]","headline":"Explicit, easy-to-evaluate bounds for Buchstab's function with a genuine new error term — but the full-range theorems rest on numerical checks that are described, not supplied.","tokens_in":12608,"tokens_out":2400,"would_cite":true,"duration_ms":24263,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["11N25","11N05","11N35","41A60"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper establishes rigorous closed-form bounds for Buchstab's function ω(u): for every u≥3, the difference W(u)=ω(u)−e^{−γ} is captured by an explicit cosine formula whose error is bounded by simple powers of u, with no numerical soluti","keywords":["Buchstab's function","delay differential equation","saddle point method","rough numbers","incomplete gamma function","Lambert W function","explicit error bounds","primes in short intervals"],"falsifier":"Take u=1000, compute the interval from Theorem 2 (the main term plus the claimed |θ_2|<0.005·10^{−6}=5×10^{−6}) and independently solve the delay equation with a high-precision solver at u=1000; if the value falls outside the interval, the theorem's error bound is false. Equally, one of the finite-range inequalities—such as the claimed maximum of Re(H(u,τ))(log^2 u)/u over 50≤u≤10^4, 4≤τ≤370 being below −8—can be checked by an independent numerical integration; a single violation refutes the corresponding bound.","tokens_in":11599,"feed_emoji":"🔢","tokens_out":8267,"duration_ms":76496,"temperature":0.7,"pith_summary":"The paper sets out to make the classic asymptotic description of Buchstab's function ω(u)—the function that counts how common integers are that have no small prime factors—fully explicit. It proves several versions of the identity W(u)=ω(u)−e^{−γ}=2|Φ(u)|cos(arg Φ(u))+error, where Φ(u) is built from the incomplete gamma function and a saddle point ζ(u) that solves e^ζ=−uζ. Each version comes with a rigorous, easily computable error bound, the sharpest being |error|<420 u^{−6} on 6≤u≤19 and <0.005 u^{−2} on u≥16. Because ω(u) oscillates about the limiting value e^{−γ}, these bounds directly quantify the fluctuations that drive the known results on primes in short intervals. A sympathetic reader would care because this replaces the need to solve a delay differential equation numerically with a few evaluations of standard special functions.","feed_headline":"Explicit formula bounds Buchstab's function to 0.005/u^2","feed_subtitle":"Rigorous bounds on ω(u) from one special-function evaluation, replacing numerical delay-equation solving in sieve and prime-count applicatio","key_machinery":"The central object is the saddle point ζ(u), the unique solution in the strip μ≥1, 0<η<3π/2 of e^ζ=−uζ, equivalently ζ=−W_1(1/u). Near this point the Laplace integrand e^{J(s)+us} has a vanishing derivative, so the main term Φ(u)=exp{−uζ+J(−ζ)}/√(2πu(1−1/ζ)) captures the oscillatory structure of W(u). The proof couples this with the identity J(s)=Γ(0,s), the derivative formula J^{(k)}(s)=(−1)^k e^{−s}/s times an explicit polynomial, Taylor polynomials of orders three and seven, and tail bounds showing the integrand decays like e^{−c u τ^2}, e^{−c u}, or e^{−8u/log^2 u} away from the saddle point. The explicit correction terms α(u) and β(u) are rational functions of ζ that give the next terms","core_discovery":"The paper proves that W(u) equals 2|Φ_2(u)|cos(arg Φ_2(u)) plus an explicit error θ_2(u), where Φ_2(u)=Φ(u)(1+α(u)), Φ(u)=exp{−uζ+J(−ζ)} divided by the square root of 2πu(1−1/ζ), J(−ζ) is the incomplete gamma function Γ(0,−ζ), and ζ(u) is the unique solution of e^ζ=−uζ in a specified region, equivalently the 1-branch of the Lambert W function applied to 1/u. The error is bounded by 420 u^{−6} for 6≤u≤19 and by 0.005 u^{−2} for u≥16; a simpler corollary gives |W(u)|<2|Φ(u)| for u≥3. The same method yields a refinement with error below 0.01 u^{−3} on 18≤u≤1000 and below 8.4 u^{−3} for larger u. This is achieved by the saddle point method: the Laplace transform of ω is e^{J(s)}−1, the contour i","pith_inferences":["A likely consequence the author leaves implicit: the same saddle-point expansion should extend to the wider family of differential-difference equations of the form uf'(u)+af(u)+bf(u−1)=0, giving explicit bounds whenever the associated saddle point is simple; the complex-versus-real character of the saddle point is the main new ingredient to manage.","The finite-range checks that fill the gap where the analytic estimates do not reach are asserted graphically; replacing those plots with certified interval arithmetic would turn the result into a fully machine-checkable proof and would also fix the sharpest constant across the entire range.","One testable extension: the formulas can be used to compute rigorous confidence intervals for ω(u) at its local extrema, sharpening the published numerical tables beyond u=11 and thereby making the short-interval prime oscillation statement quantitative at larger ranges.","If the bounds are as tight as stated, they imply that W(u) oscillates in sign infinitely often with amplitude decaying very fast, reinforcing the heuristic that the distribution of rough numbers tracks its smooth main term to an exponentially small relative error."],"forward_implications":["For any u≥3, ω(u) can be bracketed rigorously by evaluating a closed-form expression, eliminating the need to solve (uω(u))'=ω(u−1) numerically for a rigorous bound.","The bounds plug directly into the prime-short-interval formulas: the limsup and liminf of normalized prime counts in short intervals are trapped between explicit numbers built from max and min of 2|Φ|cos(arg Φ) plus a known error.","The corollary |W(u)|<2|Φ(u)| supplies a clean, universally valid envelope that may simplify future arguments needing an unconditional bound on ω(u)−e^{−γ}.","The third refinement gives an O(u^{−3}) error uniformly for u≥1000, so high-precision values of ω(u) are obtainable from the formula alone.","Because Φ(u) is expressed through Γ(0,−ζ) and W_1(1/u), standard numerical libraries already provide the ingredients; no delay-equation solver is needed."],"fun_headline_variants":["Explicit bounds for Buchstab's function via special functions","Buchstab's function: tight bounds without numerical ODE solving","New explicit error bounds for ω(u) improve sieve applications","Saddle-point method yields rigorous Buchstab bounds","Easy-to-evaluate bounds for Buchstab's function from one formula"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The proof relies on a chain of finite-range inequalities verified by graphing, contour plots, and numerical integration rather than by written analytic proof, and if any one of those unchecked numerical checks is wrong, the corresponding range of the theorem is unsupported.","fun_headline_variants_meta":{"raw":{"variants":["Explicit bounds for Buchstab's function via special functions","Buchstab's function: tight bounds without numerical ODE solving","New explicit error bounds for ω(u) improve sieve applications","Saddle-point method yields rigorous Buchstab bounds","Easy-to-evaluate bounds for Buchstab's function from one formula"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000636,"raw_usage":{"total_tokens":2730,"prompt_tokens":664,"completion_tokens":2066,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":408,"completion_tokens_details":{"reasoning_tokens":1983}},"tokens_in":408,"tokens_out":2066,"duration_ms":16069,"temperature":1.0,"reasoning_tokens":1983,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T06:25:29.073561+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take u=1000, compute the interval from Theorem 2 (the main term plus the claimed |θ_2|<0.005·10^{−6}=5×10^{−6}) and independently solve the delay equation with a high-precision solver at u=1000; if the value falls outside the interval, the theorem's error bound is false. Equally, one of the finite-range inequalities—such as the claimed maximum of Re(H(u,τ))(log^2 u)/u over 50≤u≤10^4, 4≤τ≤370 being below −8—can be checked by an independent numerical integration; a single violation refutes the corresponding bound.","supporting_citations":[],"review_version":1}