{"id":"8f7569ff-9462-4e5c-94ff-a0afbcf91593","arxiv_id":"2608.11611","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"For non-Kummer cubic L-functions over F_q[t], the mixed shifted moments are at most |C_g| g^{k_1^2+...+k_m^2} times pairwise shift factors, matching the conjectured unitary order.","lead":"Over finite fields, the paper proves an upper bound for correlations of shifted values of cubic L-functions, matching the growth rate that random matrix theory predicts. The proof is unconditional because the Riemann hypothesis for function fields is already proved.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Lemma 4.1's error-term bound (4.9) is not justified: the sum over primes in I_i has size ~q^{(1/2+ε)y}/y, so the claimed q^{O(y_iγ_i)} bound fails, and the length–power condition needed to control off-diagonal terms is violated.","rationale":"The reader's conditional verdict is directionally right, but the weakest point is not merely the growth of the 2^{ω(f)} factor in Lemma 2.1; it is the subsequent summation of that error term in Lemma 4.1. The inequality in (4.9) is internally unjustified: S_i is exponentially large in g/log^2 g, while the bound it is claimed to satisfy is only exponential in g/√log g, so the displayed estimate cannot hold. Moreover, the standard mean-value restriction 6ℓ·(max deg)≤g is violated by the allowed truncation lengths. This is a proof-level gap, not a disagreement with the conjectured order of magnitude; the theorem may well be true, and the gap may be repairable by a different choice of parameters or a more careful off-diagonal estimate. The reader also noted the much stronger large-g condition imposed by the dyadic intervals; that is a real secondary concern, but the invalid error-term estimate is the more load-bearing issue. I therefore keep the conditional verdict rather than moving to accept or reject, and I agree only partially with the reader's stated weakest assumption.","tokens_in":24695,"tokens_out":28133,"duration_ms":284143,"concrete_test":"Re-derive the O-term contribution in Lemma 4.1 under the stated hypotheses by applying Lemma 3.2 with ℓ=e^{2γ_i} and polynomial length (g+2)t_i; if 6ℓ·(g+2)t_i≤g+2 fails (it does for t_i≤e^{−80K}, b=3/4, since the left side is about 6g e^{−80K} e^{2K e^{60K}}≫g), then the error is not controlled. For a concrete numerical check, take K=1, g=e^{10^4}, compute S_1=∑_{deg P≤(g+2)t_1}|P|^{−1/2}≈q^{(g+2)t_1/2}/((g+2)t_1 log q), and verify that exp(K S_1) exceeds the bound claimed in (4.9) by multiple exponential factors.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The proof of Theorem 1.1 rests on Lemma 4.1. After expanding the exponential and applying Lemmas 2.1–2.2, the error contribution is bounded in (4.9) by q^{(1/2+ε)g} ∏_i (∑_{0≤α≤e^{2γ_i}} K^α/α! S_i^α)^2 ≪ q^{(1/2+ε)g+4e^2(1−2ε)(g+2)t_Jγ_J} exp(2KJ), where S_i=∑_{deg P∈I_i}|P|^{−1/2+ε}. This inequality is not valid. For I_1, with upper degree y=(g+2)t_1≈g/(e(log g)^2), one has S_1≍q^{(1/2+ε)y}/y, which is exponentially large. Already the α=1 term K S_1 exceeds the claimed final bound by a factor exp(Ω(yγ_1)), and the full sum is ≍exp(K S_1)=exp(exp(Ω(g/log^2 g))), far larger than the claimed q^{O(yγ_1)}. The displayed estimate appears to mistake a factor q^{O(y_iγ_i)} for a uniform bound on S_i^α. A correct treatment would need a mean-value theorem such as Lemma 3.2, which requires 6ℓ·(max deg)≤g; with the proof's parameters ℓ≤e^{2K t_i^{−3/4}} and max deg=(g+2)t_i, the condition 6e^{2K t_i^{−3/4}}t_i≤1 fails for every i (for t_i=e^{−80K}, the left side is roughly exp(2K e^{60K}−80K)≫1). Thus the off-diagonal error in Lemma 4.1 is uncontrolled, and Theorem 1.1 is not established as written.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies mixed shifted moments of cubic L-functions over the rational function field F_q[t] in the non-Kummer case q ≡ 2 (mod 3). The main result, Theorem 1.1, claims an upper bound of the conjectured order of magnitude for the moment S_g(θ^(m), k^(m)) defined in (1.1), under the assumption α_j ≪ 1/g and g > exp(4K^2), where K = k_1+...+k_m. The proof combines a Soundararajan-type upper bound (Proposition 3.1) with Harper's dyadic-multiscale method. A corollary for derivatives of the L-functions at the central point is also stated. The overall architecture follows standard recent work on moments of L-functions, and the imported inputs (Weil's Riemann hypothesis, orthogonality estimates, Pólya–Vinogradov bounds) are natural for this problem.","tokens_in":25143,"tokens_out":11469,"duration_ms":130297,"significance":"If Theorem 1.1 were proved as stated, it would be a valuable unconditional function-field analogue of the shifted-moment results of Ng–Shen–Wong and would confirm the conjectural order of magnitude predicted by David–Florea–Lalín for the non-Kummer cubic family. The paper also spells out a plausible route to derivative moments. The strategy is appropriate and the preliminary Section 3 material is largely sound. However, the proof of Lemma 4.1, which is the load-bearing step for Theorem 1.1, contains an error-term estimate that is not justified and appears to be false as written; consequently the main theorem is not established by the present argument.","major_comments":[{"comment":"The bound (4.9) is not valid. The inner sums S_i = Σ_{deg P ∈ I_i} |P|^{-1/2+ε} are exponentially large: for I_1, whose upper degree is y = (g+2)t_1 ≈ g/(e(log g)^2), the prime polynomial theorem gives S_1 ≍ q^{(1/2+ε)y}/y = exp(Ω(g/(log g)^2)). The second inequality in (4.9) replaces (Σ_{α≤[e^{2γ_i}]} K^α/α! S_i^α)^2 by q^{O(t_i γ_i g)} times a harmless factorial sum. Already the α=1 term contributes ≫ q^{(1/2+ε)g} K S_1, which exceeds the claimed final bound by a factor exp(Ω(g/(log g)^2)); for α = e^{2γ_i} the contribution is doubly exponential in (log g)^{2b} and totally dwarfs the RHS. Thus the error term in Lemma 4.1 is uncontrolled, and Theorem 1.1, whose proof rests on Lemma 4.1, is not established as written.","section":"Lemma 4.1, eq. (4.9)"},{"comment":"The dyadic decomposition has an internal consistency problem. The intervals are defined by t_0 = 1/(log g)^2 and t_j = e^{j-1}/(log g)^2 for 1 ≤ j ≤ J, while later the proof uses t_J = e^{-80K}. Since t_j is increasing in j, one needs t_0 = 1/(log g)^2 ≤ t_J = e^{-80K}, i.e. (log g)^2 ≳ e^{80K}, equivalently g ≳ exp(e^{40K}). This is not implied by the theorem's stated condition g > exp(4K^2); for K ≥ 1 and g just above exp(4K^2), the displayed expression for J is negative and the partition (4.5) is not meaningful. The theorem should either state 'for g sufficiently large depending on K', or the proof must be modified to work with the weaker threshold.","section":"Section 4.0.2"},{"comment":"The same erroneous estimate (4.9) is reused in Lemma 4.2's derivation, including the claimed error term q^{(7/8+ε)g}(log g)^{2K}, and Lemma 4.3 explicitly says its proof is a modification of Lemma 4.1. Hence the failure of (4.9) propagates to the estimates for all subfamilies C_g(j) and C_g(J). A rescue via Lemma 3.2 is not automatic: that lemma requires 6ℓ·(max degree) ≤ g+2, while the truncated exponential expansion involves powers up to e^{2γ_i} on intervals of length (g+2)t_i, and for t_i ≤ t_J = e^{-80K} the condition is violated. A correct proof will need a genuinely different treatment of the error terms or a different choice of parameters.","section":"Lemmas 4.2 and 4.3"}],"minor_comments":[{"comment":"The title contains a typo: 'V ALUES' should be 'VALUES'.","section":"Title and abstract"},{"comment":"There are typographical errors: 'Lemmma 2.10' should be 'Lemma 2.10', and 'Brunching process' should be 'branching process'.","section":"Section 2.2.2 and Section 4.0.2"},{"comment":"The displayed lower bound (1.2) contains a typo: 'g^{k_1+···+k_2^m}' should presumably be 'g^{k_1^2+...+k_m^2}'.","section":"Equation (1.2)"},{"comment":"The notation is overloaded: α denotes both the shift vector (α_1,...,α_m) and the summation parameter in Lemma 4.3; also 'meas(C_g(0))' is used for the cardinality of a finite set of characters, which is better written as #C_g(0) or |C_g(0)|.","section":"Section 4, notation"},{"comment":"The exponent in the denominators of (4.1) and (4.3), e.g. |P|^{1/2+1/(N log q)}, is written inconsistently with Lemma 2.4; the role of the term 1/(N log q) should be clarified or removed if it is a typo.","section":"Equation (4.1)"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is within the journal's scope and the claimed result is plausible, but the main proof has a serious gap in the error-term analysis of Lemma 4.1. I do not recommend rejection at this stage, because the overall Harper–Soundararajan framework is standard and the gap, while substantial, might be repairable with a different treatment of the error terms and a corrected parameter range. The authors should also carefully restate the quantitative hypothesis on g."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Bottom line: the paper asks the right question and the main-term plan is recognizable, but the proof has a load-bearing gap. I would not accept it as is; I would send it to a referee who knows the Harper machinery.\n\nWhat's new: this is the first sharp upper bound for shifted mixed moments of non-Kummer cubic L-functions over function fields. Only first and second moments were known for this family, and the adaptation of Soundararajan–Harper to a family with less transparent orthogonality is a natural next step. The main-term analysis—cube-versus-square counting, factorial factors, the g^{Σ k_j^2} shape—is broadly sound, and the unconditional setting via Weil's RH is a genuine plus.\n\nSoft spots, in order of severity:\n\n1. Lemma 4.1, specifically line (4.9), is not justified. The error term after expansion contains S_i = Σ_{deg P ∈ I_i} |P|^{-1/2+ε}. For i=1, y ≈ (g+2)t_1 ≈ g/(e (log g)^2), and S_1 ≈ q^{(1/2+ε)y}/y, which is exponentially large. Already the α=1 term K S_1 swamps the claimed q^{O(g t_i γ_i)} bound; the full sum is exp(K S_1), doubly exponential in g/(log g)^2. The displayed inequality appears to confuse a factor q^{O(y γ_i)} with a uniform bound on S_i^α. The off-diagonal error in Lemma 4.1 is therefore uncontrolled. This is not cosmetic—Lemma 4.1 is the engine of the proof.\n\n2. The dyadic decomposition has a consistency problem. With t_1 = 1/(log g)^2 and t_J = e^{-80K}, the needed inequality t_1 ≤ t_J means (log g)^2 ≳ e^{80K}, which is much stronger than the theorem's stated g > exp(4K^2). For large fixed K the interval ordering fails inside the stated range.\n\n3. Smaller: the Kummer case is only sketched. That would be fine if the non-Kummer proof worked.\n\nThe stress-test note you passed along holds up; I checked the relevant lines. I would push back on the reader's moderate confidence: the confidence in the proof as written should be low, even though the underlying conjecture and method are reasonable.\n\nWho this is for: people working on moment conjectures for cubic L-functions, and anyone interested in the Harper machinery outside the quadratic family. I would not cite Theorem 1.1 as established until Lemma 4.1 and the range condition are repaired. It does deserve a serious referee—the problem is right and the strategy is salvageable.","headline":"New target, plausible strategy, but the proof as written fails at a key error-term bound and at the stated range of g.","tokens_in":25687,"tokens_out":5489,"would_cite":false,"duration_ms":56134,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["11T06","11M38","11L40"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper proves an upper bound for mixed shifted moments of cubic L-functions over function fields that matches the conjectured order of magnitude, unconditionally.","keywords":["cubic L-functions","function fields","shifted moments","Soundararajan method","Harper method","non-Kummer characters","moments of L-functions"],"falsifier":"Compute the two-shift moment S_g((theta_1, theta_2), (1,1)) over a fixed F_q for a sequence of genera with |theta_1 - theta_2| = c/g and c fixed; Theorem 1.1 says this moment is O_{q,c}(|C_g| $g^{4}$). Exhibiting any sequence where the normalized moment grows like $g^{{4+delta}}$ for a fixed delta > 0 would disprove the theorem and the conjectured order.","tokens_in":24480,"feed_emoji":"🧮","tokens_out":6599,"duration_ms":67385,"temperature":0.7,"pith_summary":"This paper studies how values of cubic Dirichlet L-functions over the polynomial ring F_q[t] correlate at different points along the critical line. It proves that the mixed shifted moments, averaged over all primitive cubic characters of conductor genus g, are bounded above by exactly the order conjectured from random matrix theory. In the non-Kummer case q ≡ 2 mod 3, the bound is unconditional, since the Riemann hypothesis over function fields is a theorem. The same method is claimed to work in the Kummer case q ≡ 1 mod 3. If correct, the result gives the conjectured magnitude of all shifted moments for this family and, as a corollary, moment bounds for derivatives at the central point.","feed_headline":"Cubic L-function shifted moments pinned down","feed_subtitle":"Mixed shifted moments of cubic L-functions over function fields match the conjectured order, unconditionally.","key_machinery":"The argument is carried by a short-Dirichlet-polynomial approximation to the logarithm of the L-function (Lemma 2.4), together with a dyadic decomposition of that polynomial into pieces supported on disjoint prime-degree intervals. The family C_g is partitioned according to the sizes of these pieces, and Harper's method replaces the exponential of a bounded Dirichlet polynomial by a product of truncated Taylor sums, using the combinatorial Lemma 5.2 from the literature. The decisive number-theoretic input is the orthogonality lemma (Lemma 2.1): the sum over chi in C_g of chi($f^{3}$) equals (#C_g) times a local product over even-degree primes dividing f, plus an error O($q^{{(1/2+epsilon)g}}$ $2^{{omega(f)}}$). This forces the dominant contributions to come only from cube and square pairings of primes, which produces the diagonal factors $g^{{k_1^2+...+k_m^2}}$ and the shift-dependent products.","core_discovery":"Theorem 1.1 establishes that for q ≡ 2 mod 3, with K = k_1 + ... + k_m and shifts satisfying alpha_j << 1/g, the mixed moment S_g($\\theta$^(m), k^(m)) = sum over chi in C_g of the product |L(e(theta_j)/$q^{{1/2+alpha_j}}$, chi)|^{2k_j} is bounded by |C_g| $g^{{k_1^2+...+k_m^2}}$ times the product over i<j of (min{1/|theta_i - theta_j|, g})^{2k_i k_j}, provided g > exp($4K^{2}$). This is the first upper bound of the conjectured order for shifted cubic L-function moments in the function-field setting, and it holds unconditionally because Weil's Riemann hypothesis locates all zeros on the circle |u| = $q^{{-1/2}}$. The proof adapts Soundararajan's moment method and Harper's multiscale dyadic decomposition. A direct corollary bounds moments of derivatives: sum over chi in C_g of |$L^{{(ell)}}$($q^{{-1/2}}$, chi)|^{2k} << |C_g| $g^{{k^2+2k ell}}$.","pith_inferences":["The Kummer case q ≡ 1 mod 3 is only sketched, but the paper's Section 2.4 supplies the analogous orthogonality relations; a full write-up would most likely yield the same upper bound with the family size #C_tilde_g in place of #C_g.","The dyadic decomposition reveals the same log-correlated structure seen in moments of the Riemann zeta function, so one could reasonably aim the machinery at the maximum size of log |L(1/2, chi)| in this family.","The matching lower bound is explicitly left open, and the paper notes that even the two-shift case would require a power-saving error term for the second moment of L(1/2, chi), where only a logarithmic saving is currently known; improving that error term is therefore the natural next step.","Because the bound separates into products corresponding to distinct shifts when |theta_i - theta_j| >> 1/g, the result is consistent with the expected asymptotic independence of the L-values at mesoscopic separation, though the upper bound alone does not establish that independence."],"forward_implications":["Setting all shifts equal recovers the conjectural upper bound for moments of |L(1/2, chi)| in the cubic character family, consistent with the unitary symmetry predicted for this family.","Corollary 1.2 gives moment bounds for derivatives at the central point, namely sum over chi in C_g of |L^{(ell)}(q^{-1/2}, chi)|^{2k} << |C_g| g^{k^2+2k ell}.","The bound is unconditional over function fields; the analogous shifted-moment result for the Riemann zeta function in the number-field setting is conditional on the Riemann Hypothesis.","The result holds for any fixed number m of shifts and any positive exponents k_j as long as the total K is fixed and the genus satisfies the mild lower bound g > exp(4K^2)."],"supporting_citations":[{"why":"Supplies Harper's multiscale dyadic-decomposition method for bounding moments, adapted here to the cubic family.","marker":"[20]"},{"why":"Supplies the Soundararajan method that yields the preliminary g^epsilon-loss upper bound used for the large-value pieces.","marker":"[30]"},{"why":"Provides the family size, orthogonality relations, and L-function bounds for cubic characters over function fields used throughout the proof.","marker":"[12]"},{"why":"Provides Lemma 6.2, the moment bound for Dirichlet polynomials over C_g, and the conjectural moment upper bound that Theorem 1.1 recovers.","marker":"[11]"},{"why":"Provides the combinatorial Lemma 5.2 used to replace the exponential of a Dirichlet polynomial by truncated Taylor sums.","marker":"[23]"},{"why":"The source cited for Lemma 2.4, which bounds the logarithm of the cubic L-function by a short Dirichlet polynomial.","marker":"[7]"},{"why":"Gives Weil's Riemann hypothesis for curves over finite fields, which places the zeros on |u| = q^{-1/2} and makes the bound unconditional.","marker":"[31]"},{"why":"The current best asymptotic for the second moment of the cubic family, whose logarithmic error term the paper cites as the obstacle to a matching lower bound.","marker":"[19]"}],"fun_headline_variants":["Upper bound proven for cubic L-moments","Unconditional bound for shifted cubic L-moments","Large genus limit: cubic L-moments pinned","First tight bound for shifted cubic L-moments","Cubic L-moments over function fields: upper bound"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument depends on the character orthogonality Lemma 2.1 giving a main term with an error that stays under control even for polynomials with many prime factors; if that error grew faster with the number of prime factors, the cube-versus-square counting that produces the $g^{{k_1^2+...+k_m^2}}$ factor would break.","fun_headline_variants_meta":{"raw":{"variants":["Upper bound proven for cubic L-moments","Unconditional bound for shifted cubic L-moments","Large genus limit: cubic L-moments pinned","First tight bound for shifted cubic L-moments","Cubic L-moments over function fields: upper bound"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001544,"raw_usage":{"total_tokens":6137,"prompt_tokens":869,"completion_tokens":5268,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":485,"completion_tokens_details":{"reasoning_tokens":5191}},"tokens_in":485,"tokens_out":5268,"duration_ms":41204,"temperature":1.0,"reasoning_tokens":5191,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T00:34:40.559610+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the two-shift moment S_g((theta_1, theta_2), (1,1)) over a fixed F_q for a sequence of genera with |theta_1 - theta_2| = c/g and c fixed; Theorem 1.1 says this moment is O_{q,c}(|C_g| $g^{4}$). Exhibiting any sequence where the normalized moment grows like $g^{{4+delta}}$ for a fixed delta > 0 would disprove the theorem and the conjectured order.","supporting_citations":[{"cited_title":"Soundararajan, Moments of the Riemann zeta function, Ann","cited_arxiv_id":null,"evidence_quote":"Supplies the Soundararajan method that yields the preliminary g^epsilon-loss upper bound used for the large-value pieces."},{"cited_title":"David, A","cited_arxiv_id":null,"evidence_quote":"Provides the family size, orthogonality relations, and L-function bounds for cubic characters over function fields used throughout the proof."},{"cited_title":"David, A","cited_arxiv_id":null,"evidence_quote":"Provides Lemma 6.2, the moment bound for Dirichlet polynomials over C_g, and the conjectural moment upper bound that Theorem 1.1 recovers."},{"cited_title":"Kirila, An upper bound for discrete moments of the derivative of the Riemann zeta-function, Mathematika, 66 (2020), no","cited_arxiv_id":null,"evidence_quote":"Provides the combinatorial Lemma 5.2 used to replace the exponential of a Dirichlet polynomial by truncated Taylor sums."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"The source cited for Lemma 2.4, which bounds the logarithm of the cubic L-function by a short Dirichlet polynomial."},{"cited_title":"Weil, Sur les courbes alg´ ebriques et les vari´ et´ es qui s’en d´ eduisent,Actualit´ es Sci","cited_arxiv_id":null,"evidence_quote":"Gives Weil's Riemann hypothesis for curves over finite fields, which places the zeros on |u| = q^{-1/2} and makes the bound unconditional."}],"review_version":1}