{"id":"5f01d85a-95b7-4346-adbd-e5dc7a4edb49","arxiv_id":"1908.08816","paper_version":3,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"For infinitely many n, the largest prime factor of n^2+1 is greater than n^1.279.","lead":"This paper proves that infinitely many numbers of the form n^2+1 have a prime factor larger than n^1.279, improving the previous record of 1.2182. The proof relies on a new bilinear estimate and Harman's sieve method.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified; the argument is internally coherent, with the narrow numerical margin as the only residual verification point.","rationale":"I read the paper in good faith and attempted to find an internal gap. The hardest part to check is the Type II estimate, but the reduction via Poisson summation and Cauchy-Schwarz is standard, and the final bounds (3.3), (3.5), (3.7), and (3.9) all have the expected exponents. The application of the Deshouillers–Iwaniec bound is subtle because the coefficients depend on the level r, but the paper acknowledges this and uses a pointwise bound in r; the resulting N-range is consistent with the stated exponent (57−32α)/96. The sieve bookkeeping in Proposition 5 ties the Type II ranges correctly to the Buchstab decomposition, including the ℓ > 1 case. Lemma 7's replacement of ρ by 1 looks abrupt but is justified by average prime distribution and produces only o(1) error. Therefore I do not see a correctness risk that would change the ACCEPT verdict. The narrowest point is numerical rather than structural. The reader's weakest-assumption pick (Kim–Sarnak) is real but external and not an internal flaw; I partially agree with it, while my own residual check concerns the numerical evaluation. Verdict: UNCHANGED.","tokens_in":23785,"tokens_out":27420,"duration_ms":278864,"concrete_test":"Independently recompute all algorithmic bounds in Section 2.4 using interval arithmetic or a second high-precision implementation: G1–G4, G5, G6 and the final integral 4(1−2θ)∫_{153/128}^{1.279} α/(1−2θα)dα with θ = 7/64. Verify the stated inequalities 25/157 + G < 0.553361 and 0.553361 + final integral < 1 (reported 0.997…). If any of the numerical bounds is exceeded, recompute the implied exponent; the theorem as stated would then need adjustment.","verdict_should_be":"UNCHANGED","load_bearing_attack":"No significant objection identified. The central claim is supported by a clear structure: Proposition 3 (Type I) and Proposition 4 (Type II) feed the Harman-sieve decomposition, the Buchstab identities match the Type II ranges, and the final summation leaves a positive margin. The Kim–Sarnak input θ = 7/64 is external and standard, and the paper correctly propagates it through Propositions 3, 4 and 12. I checked the bookkeeping of the Type II ranges in Proposition 5 (ℓ = 1 versus ℓ > 1), the Cauchy-Schwarz steps in Section 3.5, and the replacement of ρ by 1 in Lemma 7; the averages are consistent with the stated asymptotic formulas. The only residual load-bearing component is numerical: the final inequality is 0.997… < 1, leaving a margin of about 3 × 10^{-3}. Everything below that margin rests on the Python-evaluated bounds for G1–G6 and the final linear-sieve integral. The paper supplies code links, but not independent interval-arithmetic or machine-checked verification. This is a verification gap, not a demonstrated flaw.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proves that the largest prime factor of n^2+1 is infinitely often greater than n^1.279 (Theorem 1), improving the previous exponent 1.2182 of de la Bretèche and Drappeau. It also proves Theorem 2: assuming Selberg's eigenvalue conjecture, the exponent can be raised to 1.312. The method follows Chebyshev and Hooley in relating the desired lower bound to upper bounds for sums over primes p∼P. The authors combine the Type I information of de la Bretèche–Drappeau (Proposition 3) with a new Type II estimate (Proposition 4) obtained from Deshouillers–Iwaniec bounds on linear forms of Kloosterman sums together with the Kim–Sarnak spectral gap θ=7/64. The sieve part uses Harman's method and Buchstab's identity to produce explicit integral-defined deficiencies G1–G6, which are then summed to obtain the final numerical inequality. The paper contains a detailed proof of the Type II estimate (Section 3), including the application of the Cauchy–Schwarz inequality, completion of sums, and the Deshouillers–Iwaniec bound.","tokens_in":23941,"tokens_out":15409,"duration_ms":137040,"significance":"Should the result be correct, it is a substantial improvement in a classical problem of number theory: the exponent 1.279 improves on de la Bretèche–Drappeau's 1.2182, and the conditional 1.312 improves on Deshouillers–Iwaniec's conditional 1.2247. The main innovation, a Type II estimate for the quadratic congruence n^2+1≡0 mod m with a power saving of x, is clearly presented and is of independent interest for other applications of Harman's sieve. The proof is modular and transparent: Propositions 3 and 4 are clearly isolated as arithmetic inputs, the Buchstab decompositions in Section 2.4 are explicit, and the error terms are tracked. The paper also makes the numerical code available (though in ephemeral form). The most significant residual risk is that the final numerical margin is only about 3×10^{-3} and the numerical quadratures are not certified; this is a verification gap rather than a detected flaw in the analytic argument.","major_comments":[{"comment":"The conclusion of both Theorem 1 and Theorem 2 hinges on numerical inequalities with a margin of roughly 3×10^{-3}: in §2.5, 0.553361 + 4(1−2θ)∫_{153/128}^{1.279} α/(1−2θα) dα = 0.997… < 1, and in §2.6, 0.679914 + 4∫_{5/4}^{1.312} α dα = 0.997… < 1. The constants feeding these inequalities (G1<0.01745, G2<0.11478, G3<0.093754, G4<0.0057, G5<0.17877, G6>0.016329) are obtained by Python 3.7 quadrature, and the manuscript directs the reader to codepad.org links at the end of Section 2.5. No error analysis of the numerical integration is given, and a relative error in the computed constants larger than about 3×10^{-3} would overturn the theorem. Because this numerical verification is load-bearing, the revision should provide the code in a permanent repository, state the numerical method and precision, and supply either an interval-arithmetic certification or an independent evaluation of the integrals G1–G6 and the two final integrals.","section":"§2.4–2.5, end of Section 2.5"}],"minor_comments":[{"comment":"The passage 'for (m,P(x^η))=1 we can replace ρ(m) by 1 with negligible error by equidistribution of primes in arithmetic progressions' is terse: ρ(p) equals 2 or 0 according to p≡1 or 3 mod 4, so the replacement needs an averaging argument over the ordered variables q_1,…,q_k and m. Please provide the precise statement and proof (or a reference) showing that the error is o(1) in (2.4)–(2.7) uniformly in the ranges used.","section":"§2.3, Lemma 7"},{"comment":"The parenthetical '(for α < 758/733 part (ii) of Proposition 4 is stronger than (i))' is imprecise: by Remark 8 part (ii) is stronger than (i) for all α < 2671/2496 ≈ 1.0701. Please state the correct threshold or rephrase the sentence.","section":"§2.4.1"},{"comment":"In the sentence 'Since t ≠ 0 ≠ h1 n2 − n2 h1', the second term should read h2 n1, i.e. h1n2 − h2n1 ≠ 0, consistent with the definition of n in the same paragraph.","section":"§3.8"},{"comment":"The title in the full text contains the spacing artifact 'P F ACTOR', and the arXiv abstract gives the de la Bretèche–Drappeau year as 2019 while the text and reference [1] give 2020; please unify the citations.","section":"Title and Abstract"},{"comment":"The codepad.org links at the end of Section 2.5 are ephemeral and should be replaced in the final version by a permanent repository (e.g., Zenodo or a journal-hosted supplementary file) with a versioned, runnable script.","section":"§2.5, code links"}],"recommendation":"major_revision","confidential_remarks":"I could not identify any internal inconsistency or circularity in the analytic arguments; the proof appears sound up to the numerical verification. Given the small final margin, I recommend that the numerical integrals G1–G6 and the final integrals be independently checked, for instance by a referee using interval-arithmetic tools, before acceptance. The paper is within the scope of the journal."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear [Colleague],\n\nThe paper is a genuine record-breaker, and the real news is the new Type II estimate, not the exponent itself. Merikoski shows the largest prime factor of n^2+1 exceeds n^1.279 infinitely often, up from de la Bretèche–Drappeau's 1.2182, and gets 1.312 conditionally on Selberg's eigenvalue conjecture. The proof combines de la Bretèche–Drappeau's Type I information with a stronger bilinear estimate (Proposition 4) and feeds both through Harman's sieve. That combination is new for this problem, and it is what buys the improvement.\n\nI read through the structure carefully. The Type I proposition is quoted correctly from the literature, the Buchstab decompositions match the Type II ranges in Propositions 5 and 6, and the Cauchy–Schwarz steps in Section 3 have consistent sizes. The final numerical inequality is 0.997... < 1; the margin is about 3×10^-3, which is thin but real. I don't see a circular step or a post-hoc fitted constant. The exponent comes from a genuine optimization problem, not from tuning to match 1.279. The reliance on Kim–Sarnak's θ=7/64 is external, standard, and fully disclosed; Remark 1 states plainly that without it the argument gives only 1.23. The citation practice is honest and precise, especially the distinction from [1, Théorème 5.2].\n\nThe soft spots are minor, in proportion. The numerical margin is narrow enough that a small error in the Python-evaluated integrals G1–G6 could change the theorem into a weaker one. The code links are welcome, but they are not interval arithmetic or machine-checked, so the 0.997 constant rests on trust. I would want a referee to rerun the numerics independently, but I would not expect a problem. The Type II estimate only works for P < x^153/128, and the paper says so; above that range it falls back to the linear sieve. The final \"open problems\" section is genuinely useful, not padding.\n\nWho benefits: analytic number theorists working on sieve methods and quadratic polynomials. I'd bring it to the attention of anyone working on the n^2+1 family. It deserves a serious referee and a good journal slot. My recommendation: send it out, with a referee asked to verify the numerical integrals.\n\nBest,\n[You]","headline":"A legitimate record-breaker whose real value is the new Type II/Harman's-sieve combination; the numerical margin is thin but the structure is honest and sound.","tokens_in":24529,"tokens_out":2795,"would_cite":true,"duration_ms":28519,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["11N35","11N36","11N32","11L05","11F72"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper establishes that infinitely many integers $n$ have the largest prime factor of $n^2+1$ greater than $n^{1.279}$, and that assuming the eigenvalue conjecture the exponent can be raised to $1.312$.","keywords":["largest prime factor","n^2+1","Type II estimates","sieve methods","Kloosterman sums","spectral gap","eigenvalue conjecture"],"falsifier":"Recompute the numerical constants that close the proof: the paper reports that at $\\alpha=1.279$ the final coefficient is $0.997\\cdots<1$ after summing $25/157+G_1+G_2+G_3+G_4+G_5-G_6$ and the linear-sieve integral. A recalculation reaching $1$ or more would destroy the contradiction that forces the theorem. A direct disproof would be infinitely many $n$ with largest prime factor $\\le n^{1.279}$, which the theorem says cannot happen.","tokens_in":23535,"feed_emoji":"🔢","tokens_out":11145,"duration_ms":108229,"temperature":0.7,"pith_summary":"This paper proves that the largest prime factor of $n^2+1$ is infinitely often larger than $n^{1.279}$, improving the previous record exponent $1.2182$. The advance comes from a new asymptotic formula for bilinear Type II sums attached to the congruence $\\nu^2+1\\equiv0\\pmod{mn}$, combined with a sieve decomposition that converts this arithmetic information into a saving over the classical linear sieve. If the proof is right, this is the strongest unconditional lower bound known for the size of prime factors of $n^2+1$, a natural waystation toward the open question whether $n^2+1$ is prime infinitely often. Under the eigenvalue conjecture for congruence subgroups, the same argument raises the exponent to $1.312$.","feed_headline":"Largest prime factor of n^2+1 beats n^1.279 infinitely often","feed_subtitle":"A sharper sieve-plus-bilinear estimate lifts the record from n^1.2182; the spectral conjecture would take it to n^1.312.","key_machinery":"The engine is a Type II (bilinear) estimate. Defined in Proposition 4, it gives an asymptotic formula, up to $O(x^{1-\\eta})$, for $\\sum_{m\\sim M}\\sum_{n\\sim N} a_m b_n \\sum_{\\nu^2+1\\equiv0\\pmod{mn}} \\psi_P(mn)\\log(mn)$, for divisor-bounded $a_m,b_n$ with $b_n$ square-free, provided $MN=P=x^\\alpha$ and $N$ lies in one of two ranges. The proof goes through Poisson summation, Cauchy-Schwarz, and completion of an incomplete exponential sum to a linear form of Kloosterman sums, bounded with the current spectral-gap parameter $\\theta=7/64$; the sieve applies this estimate to all but finitely many pieces of an iterated decomposition of $S(A(P),2\\sqrt{P})$, and the leftover pieces are bounded by numerical integrals arising from the sieve.","core_discovery":"On the paper's own terms, the central claim is Theorem 1: for infinitely many integers $n$, the largest prime factor of $n^2+1$ exceeds $n^{1.279}$. The paper's new ingredient is Proposition 4, an asymptotic formula for bilinear sums over $m\\sim M$ and $n\\sim N$ with $MN=P$, in which the counting term $\\sum_{\\nu^2+1\\equiv0\\pmod{mn}} 1$ is matched to its expected main term up to an error $O(x^{1-\\eta})$; this holds when $N$ lies in a stated range depending on the spectral parameter $\\theta=7/64$. The sieve argument decomposes the prime-counting sum $S(x,P)$ with the standard iterated identity for sifted sums, applies the bilinear formula to the pieces in range, and bounds the remaining pieces numerically; adding the ranges up to $x^{153/128}$ and the linear-sieve bound beyond it keeps the total below the required $X\\log x$, forcing the existence of infinitely many $n$ with a prime factor above $n^{1.279}$. Theorem 2 is the same argument with $\\theta=0$, yielding $1.312$ conditionally.","pith_inferences":["If the average over the auxiliary level variable $r$ could be exploited, as in the paper's Conjecture 1, the present method would already reach exponent $1.286$ with the current spectral parameter; this marks the nearest plausible next step.","Further progress beyond $x^{153/128}$ appears to require genuinely new higher-order arithmetic information, such as asymptotic formulas for Type $I_2$ or $I_3$ sums; the paper notes that even the simplest Type $I_2$ case remains open.","The same strategy should transfer to $n^2-d$ for non-square $d$, giving analogous records for other quadratic sequences; the paper states this expectation but does not prove it."],"forward_implications":["Infinitely many $n$ have $P^+(n^2+1)>n^{1.279}$, and the proof locates such $n$ through a contradiction count rather than by construction.","Assuming the spectral eigenvalue conjecture, the same method gives infinitely many $n$ with $P^+(n^2+1)>n^{1.312}$, improving the previous conditional $1.2247$.","The new Type II estimate extends the range in which the arithmetic of $n^2+1$ is understood from $P=x^{1+o(1)}$ to an explicit interval, and the sieve bound varies continuously from an asymptotic formula to the linear-sieve bound as $P$ grows.","The paper's numerical constants could be pushed slightly higher by more careful optimization, so the exact limit of the method is not exhausted by the stated $1.279$."],"supporting_citations":[{"why":"Provides the previous exponent 1.2182 and the Type I estimate used as Proposition 3.","marker":"[1]"},{"why":"Introduced the weighted counting framework for this problem and the linear-sieve bound the paper improves.","marker":"[2]"},{"why":"Supplies the linear-forms-of-Kloosterman-sums bound used as Lemma 12 in the Type II proof.","marker":"[3]"},{"why":"Gives the prime-supported bilinear-sum method behind part (ii) of Proposition 4.","marker":"[4]"},{"why":"Provides the prime-detecting sieve formalism that converts Type I and Type II information into the final exponent.","marker":"[6]"},{"why":"Originates the problem's framework and the earlier exponent 1.10014...","marker":"[7]"},{"why":"Fixes the spectral parameter theta=7/64 on which Propositions 3 and 4 depend.","marker":"[10]"}],"fun_headline_variants":["Largest prime factor of n^2+1 tops n^1.279 infinitely often","Prime factor record: n^2+1's largest factor > n^1.279 infinitely often","Sieve method boosts prime factor exponent for n^2+1 to 1.279","Infinitely many n: largest prime factor of n^2+1 exceeds n^1.279","Conditional on Selberg: n^2+1 prime factor exponent 1.312"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole gain beyond the old exponent $1.23$ rests on a single deep external theorem: a lower bound for the smallest Laplace eigenvalue of the congruence subgroups used in the argument, written as $\\lambda_1\\ge 1/4-(7/64)^2$. If that bound were weakened, the exponent in Theorem 1 would drop to $1.23$; the argument does not create this assumption itself.","fun_headline_variants_meta":{"raw":{"variants":["Largest prime factor of n^2+1 tops n^1.279 infinitely often","Prime factor record: n^2+1's largest factor > n^1.279 infinitely often","Sieve method boosts prime factor exponent for n^2+1 to 1.279","Infinitely many n: largest prime factor of n^2+1 exceeds n^1.279","Conditional on Selberg: n^2+1 prime factor exponent 1.312"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000771,"raw_usage":{"total_tokens":3410,"prompt_tokens":933,"completion_tokens":2477,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":549,"completion_tokens_details":{"reasoning_tokens":2357}},"tokens_in":549,"tokens_out":2477,"duration_ms":17333,"temperature":1.0,"reasoning_tokens":2357,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T11:28:48.647605+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Recompute the numerical constants that close the proof: the paper reports that at $\\alpha=1.279$ the final coefficient is $0.997\\cdots<1$ after summing $25/157+G_1+G_2+G_3+G_4+G_5-G_6$ and the linear-sieve integral. A recalculation reaching $1$ or more would destroy the contradiction that forces the theorem. A direct disproof would be infinitely many $n$ with largest prime factor $\\le n^{1.279}$, which the theorem says cannot happen.","supporting_citations":[{"cited_title":"de la Bretèche and S","cited_arxiv_id":null,"evidence_quote":"Provides the previous exponent 1.2182 and the Type I estimate used as Proposition 3."},{"cited_title":"Deshouillers and H","cited_arxiv_id":null,"evidence_quote":"Introduced the weighted counting framework for this problem and the linear-sieve bound the paper improves."},{"cited_title":"Deshouillers and H","cited_arxiv_id":null,"evidence_quote":"Supplies the linear-forms-of-Kloosterman-sums bound used as Lemma 12 in the Type II proof."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the prime-supported bilinear-sum method behind part (ii) of Proposition 4."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the prime-detecting sieve formalism that converts Type I and Type II information into the final exponent."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Originates the problem's framework and the earlier exponent 1.10014..."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Fixes the spectral parameter theta=7/64 on which Propositions 3 and 4 depend."}],"review_version":1}