{"id":"95db2ca9-a350-428c-b903-4f751387a8a9","arxiv_id":"2412.20099","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Under RH and three zero-correlation conjectures, the third moment of the real part of log zeta(1/2+it) is shown to equal c_P + c_Z + O(1/log T), while the third moment of the imaginary part is O(1/log T).","lead":"This paper proves a conditional formula for the third moment of the logarithm of the Riemann zeta function, assuming RH and three zero-correlation conjectures. The result matches random matrix predictions, and it introduces a new 'twisted pair correlation' conjecture to make the calculation work.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The cancellation that fixes c_Z depends on the sharp twisted pair correlation in the ramp region, but the paper proves only a smoothed version and flags a √n obstruction to removing the smoothing.","rationale":"The reader's weakest assumption was the exact shape of m_n(α) and the identity Λ(n)/logT m_n(α) = H_*(α, Λ(n)/logT) for primes. I agree that the shape of m_n is the load-bearing hinge, but I locate the concern more precisely in the sharp-to-smooth transition. The paper's proof of Theorem 1.1 itself is a valid conditional derivation assuming Conjecture 1.5; the unresolved issue is whether Conjecture 1.5 is a credible statement about the zeta zeros in the ramp region. The paper's own supporting Theorem 1.9 is only a smoothed result and the authors note the √n obstruction in removing the smoothing. This makes the cancellation that yields c_Z depend on an unverified sharp form of the conjecture. The conditional theorem is not falsified by this, so the verdict remains CONDITIONAL; no change is recommended. The omitted proof of Proposition 3.2 is a secondary gap, also worth addressing, but the sharp conjecture issue is the more consequential concern.","tokens_in":37592,"tokens_out":22828,"duration_ms":222536,"concrete_test":"Numerically evaluate the left side of Conjecture 1.5 for n=4 (so Λ(n)=log 2) at T=10^6, using a test function r whose Fourier transform \\hat{r} is a smooth bump supported in the ramp interval [1 - log4/logT, 1 - (log4 - log2)/logT], normalized so the predicted integral is O(1). Compute the double sum over the zeros near height T with the standard normalization. If the empirical value deviates from the conjectured ramp integral by more than 10(En + 1/logT), the sharp twisted pair conjecture fails in the regime that drives the cancellation; agreement would support its truth in the tested range.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Theorem 1.1's constant c_Z = -π^2/4 emerges from the exact cancellation of the β-dependent integral in Proposition 5.1 with the corresponding integral in Proposition 6.1. This cancellation requires the precise piecewise-linear shape of m_n(α) in Conjecture 1.5 on the intervals [-1-Λ(n)/logT, -1] and [1-logn/logT, 1-(logn-Λ(n))/logT]. This is the range where the paper's own supporting evidence is weakest. In the introduction the authors state: 'In our setup for the twisted pair correlation, the transition from Fn(α) to Fn(α;ψU) seems to incur an extra factor of size √n in the error term. This appears to cause trouble as soon as n is moderately large.' Theorem 1.9 proves the smoothed version only under Conjecture 1.8, with error O(log logT/Λ(n)); for high powers such as n=2^a, Λ(n)=log 2, so the error is not small. Conjecture 1.5, however, is the sharp unweighted statement used in the proof of Proposition 5.1. Thus the load-bearing part of the conjecture—the ramp that cancels the triple-correlation integral—has only smoothed, conditional support, and the authors explicitly identify a genuine obstruction to removing the smoothing. This does not refute the conditional theorem, but it makes the central constant hostage to an unverified sharp form of the conjecture.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proves conditional estimates for the third moment of the real and imaginary parts of log ζ(1/2+it), refining Selberg's central limit theorem. Theorem 1.1 states that, assuming RH, Montgomery's pair correlation conjecture, Hejhal's triple correlation conjecture, and a new \"twisted pair correlation\" conjecture (Conjecture 1.5), one has M_3^Re(T) = c_P + c_Z + O(1/log T) and M_3^Im(T) = O(1/log T), with c_P an explicit prime sum and c_Z = -π^2/4. The proof decomposes the third moment into prime and zero contributions: P^3, P^2 Z, P Z^2, and Z^3. The Z^3 term is evaluated using the triple correlation conjecture, and the P Z^2 term using the twisted pair correlation conjecture; the two contributions cancel in a way that produces c_Z. The paper also gives partial support for the twisted pair conjecture: Proposition 1.7 proves part of it unconditionally on RH in a restricted α-range, and Theorem 1.9 proves a smoothed version on a larger range conditional on a uniform Hardy-Littlewood conjecture.","tokens_in":37919,"tokens_out":4718,"duration_ms":51492,"significance":"If the conditional theorem is correct, it is a genuine refinement of Selberg's central limit theorem and provides the first third-moment analogue matching Keating-Snaith predictions, with the imaginary part vanishing at the predicted order. The paper contains a substantial amount of careful analytic-number-theoretic calculation: the decomposition into prime and zero contributions is carried out in detail, Proposition 1.7 gives real unconditional support for the twisted pair correlation in the small-α range, and Theorem 1.9 is a nontrivial smoothed conditional result. The main caveat is that the central constant depends on the exact shape of a new conjecture that is supported only weakly in the range that matters for the cancellation; this limits the strength of the claim that the paper fully explains the third moment.","major_comments":[{"comment":"The value of c_Z in Theorem 1.1 is fixed by the exact cancellation, in Propositions 5.1 and 6.1, of the β-dependent integrals, and that cancellation requires the precise piecewise-linear ramp of m_n(α) on the intervals [-1-Λ(n)/log T, -1] and [1-logn/log T, 1-(logn-Λ(n))/log T]. The paper's own evidence for Conjecture 1.5 is weakest precisely there: Proposition 1.7 covers only |α| < 1 - logn/logT - δ_T, and Theorem 1.9 is a smoothed statement whose error term O(log log T / Λ(n)) is not small for high prime powers such as n = 2^a, since then Λ(n) = log 2. The introduction explicitly states that the transition from F_n(α) to F_n(α;ψ_U) seems to incur an extra factor of size √n in the error term, which obstructs removing the smoothing. Since the identity Λ(n)/log T · m_n(α) = H_*(α, Λ(n)/log T) is described only as \"what is needed\" for the cancellation, the main constant is currently hostage to an unverified sharp form of the twisted pair conjecture. Please either supply a proof (or a much sharper verification) of the ramp region of Conjecture 1.5, or explicitly state in Theorem 1.1 that the constant is conditional on this sharp form and discuss how the constant would change under plausible deviations of m_n.","section":"Section 1, Conjecture 1.5 and following discussion; Section 5, Proposition 5.1; Section 9, Theorem 1.9"},{"comment":"The proof of Proposition 3.2, which establishes the O(1/log x) bound for the P(t)^3 contribution to the imaginary part, is omitted entirely with the text \"We will not give the proof details\". This is one of the four terms needed for the imaginary-part estimate in Theorem 1.1. Even if the adaptation from Proposition 3.1 is straightforward, a published proof should include the details or at least display the cancellation of the six terms explicitly; as written, this component of the main theorem is unsupported.","section":"Section 3, Proposition 3.2"},{"comment":"The application of Conjecture 1.5 in Proposition 5.1 is made with a specific test function r(u) = k_hat(2πβu), and the error term E_n from the conjecture is asserted to be absorbed \"on average over n\" without a displayed calculation. Because the sums over n run up to x ≤ T^{1/3}, the average of E_n is not immediate from its definition for n = q^a with a > 1; please provide the short calculation that justifies this absorption, or replace the assertion by a bound with the relevant averaging over prime powers displayed.","section":"Section 5, around Eq. (5.5)"}],"minor_comments":[{"comment":"In the displayed statement of Conjecture 1.5, the summation range \"T ≤ γ,γ′,2T\" is missing the condition γ′ ≤ 2T; it should read \"T ≤ γ,γ′ ≤ 2T\".","section":"Section 1, Conjecture 1.5"},{"comment":"The text refers to \"Theorem 1.7\" when it should refer to \"Proposition 1.7\" in the sentence \"Due to the symmetry of F_n(α) discussed above, Theorem 1.7 provides...\".","section":"Section 1, after Proposition 1.7"},{"comment":"The reference [Odl] is incomplete: it lacks a year and publication venue. Please provide a complete bibliographic entry for the preprint.","section":"References"},{"comment":"The proof of Lemma 2.6 is presented as a sketch; in particular, the contour-integral bounds leading to Eq. (2.26) are stated without all intermediate steps. Given that this lemma is used later in Lemma 2.8 and in Section 9, a slightly fuller derivation would improve readability.","section":"Section 2, Lemma 2.6"}],"recommendation":"major_revision","confidential_remarks":"The conditional theorem is defensible, but the paper's central constant depends on a new conjecture whose shape is chosen to produce the desired cancellation and for which the ramp region—where the cancellation occurs—has only smoothed, conditional support. This is a correctness-risk concern, not merely a matter of presentation. I would encourage the editor to ask for either stronger support for Conjecture 1.5 in the ramp region or a clear restatement of the theorem as conditional on an explicitly labeled conjecture, in addition to completing the omitted proof of Proposition 3.2."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a serious, carefully written paper with a genuinely new conditional result, but the load-bearing conjecture is shaped to produce the answer, and the evidence for it is weakest exactly where it matters.\n\nWhat is new: Theorem 1.1 gives the third moment of the real and imaginary parts of log zeta under RH and three correlation conjectures, refining Selberg's CLT and matching Keating-Snaith. Prior work stopped at the second moment. The twisted pair correlation function F_n(alpha) and Conjecture 1.5 are new objects of independent interest. The paper does real work to justify them: Proposition 1.7 proves asymptotic formulas unconditionally for small alpha, and Theorem 1.9 proves the smoothed version in the ramp range under a uniform Hardy-Littlewood conjecture. The proof of the conditional theorem is detailed, and the authors are transparent about the obstacles, including the sqrt(n) obstruction to removing the smoothing.\n\nThe soft spot is the status of Conjecture 1.5. The piecewise-linear m_n is not derived from first principles; the paper explicitly says the identity with H_* is 'what is needed' for the cancellation that produces c_Z. That is a circularity flag. The cancellation between the integrals in Proposition 5.1 and 6.1 depends on the exact shape of m_n on the ramp intervals. Proposition 1.7 does not cover that range, and Theorem 1.9 only handles a smoothed version with error O(log log T / Lambda(n)). For high prime powers Lambda(n) is as small as log 2, so the error is not small. The sharp unweighted version is assumed. So the theorem is a consistency check: if you assume RH, pair, triple, and a twisted pair conjecture chosen so the cancellation happens, you get the KS constant. That is still worth knowing, but it does not independently predict the constant.\n\nMinor: Proposition 3.2's proof is omitted with 'straightforward adaptation'; I believe it, but a referee will want it written out.\n\nWho this is for: analytic number theorists working on Selberg's CLT, Keating-Snaith, and correlation conjectures. It deserves serious refereeing. The referee should push on Conjecture 1.5 and ask for either a derivation from a more natural conjecture or a statement that isolates how much of the ramp is really needed.","headline":"The third moment result is new and the proof is careful, but the twisted pair correlation conjecture is visibly designed to make the constant come out, and the ramp region only has smoothed conditional support.","tokens_in":38420,"tokens_out":2734,"would_cite":true,"duration_ms":29334,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["11M06","11M26"],"pacs":[],"model":"deepseek-v4-flash","headline":"Assuming RH and three correlation conjectures, the third moment of the real part of log zeta is an explicit constant plus O(1/log T), matching random-matrix predictions.","keywords":["Riemann zeta function","logarithm of zeta","third moment","Selberg central limit theorem","pair correlation","triple correlation","twisted pair correlation","random matrix theory"],"falsifier":"Numerically evaluate the twisted pair correlation sum $F_n(\\alpha)$ for a small prime power, say $n=2$, at a height where millions of zeros are known, and compare with Conjecture 1.5's piecewise-linear $m_n(\\alpha)$ across the interval $[1-\\log n/\\log T,\\,1]$; a systematic deviation would falsify the cancellation mechanism. Alternatively, compute $M^\\Re_3(T)$ and $M^\\Im_3(T)$ on the same zero data and check that the real part approaches $c_P+c_Z$ at the $1/\\log T$ scale while the imaginary part tends to zero.","tokens_in":37388,"feed_emoji":"🎲","tokens_out":9992,"duration_ms":87269,"temperature":0.7,"pith_summary":"The paper establishes conditional estimates for the third moment of the logarithm of the Riemann zeta function on the critical line, refining what the Selberg central limit theorem alone predicts. Assuming the Riemann Hypothesis, the pair correlation conjecture, the triple correlation conjecture, and a new 'twisted' pair correlation conjecture, it proves that the average of $(\\Re\\log\\zeta(1/2+it))^3$ over $t\\in[T,2T]$ equals an explicit constant $c_P+c_Z$ with an error $O(1/\\log T)$, while the corresponding average for the imaginary part is $O(1/\\log T)$. This matches the finer random-matrix prediction [KS00], which high-precision numerical data had suggested. The interest is that the third moment is the first quantity that distinguishes the real and imaginary parts of $\\log\\zeta$ beyond their shared leading Gaussian behaviour.","feed_headline":"Third moment of log zeta pinned down to a constant","feed_subtitle":"Conditional on RH, the real part's third moment is c_P + c_Z, refining Selberg's law and matching random-matrix predictions.","key_machinery":"The load-bearing object is the twisted pair correlation function $F_n(\\alpha)$, a normalized sum over pairs of zeros $\\gamma,\\gamma'$ of $n^{i\\gamma}T^{i\\alpha(\\gamma-\\gamma')}\\omega(\\gamma-\\gamma')$, together with the conjectured piecewise-linear density $m_n(\\alpha)$ of Conjecture 1.5. For primes the paper notes the identity $\\frac{\\Lambda(n)}{\\log T}m_n(\\alpha)=H_*\\bigl(\\alpha,\\frac{\\Lambda(n)}{\\log T}\\bigr)$, where $H_*$ is the non-discrete part of the triple correlation kernel; this identity is what makes the $P(t)Z(t)^2$ contribution cancel against the $Z(t)^3$ contribution in the third moment. The rest of the machinery is the decomposition of $\\log\\zeta(1/2+it)$ as a sum $P(t)+Z(t)$ of a prime-power sum and a zero sum (Proposition 2.4), which converts the third moment into integrals of $P^3$, $P^2Z$, $PZ^2$, and $Z^3$ that are then evaluated with the correlation conjectures.","core_discovery":"The central discovery is Theorem 1.1: under RH, the pair correlation conjecture (Conjecture 1.2), the triple correlation conjecture (Conjecture 1.3), and the twisted pair correlation conjecture (Conjecture 1.5), the real-part third moment satisfies $M^\\Re_3(T)=c_P+c_Z+O(1/\\log T)$ with $c_Z=-\\pi^2/4$ and $c_P=\\frac{3}{4}\\sum_{p}\\sum_{m\\ge2}\\frac{1}{m\\,p^m}\\sum_{k+\\ell=m}\\frac{1}{k\\ell}$, while the imaginary-part third moment is $O(1/\\log T)$. The constant splits naturally: $c_Z$ comes from the triple correlation of zeros and $c_P$ from prime-power sums, and the twisted pair correlation conjecture is exactly the mechanism that forces the cross term between one prime power and two zeros to cancel against the pure zero contribution, leaving $c_Z$ intact. The paper argues that this twisted conjecture is of independent interest and supports it by proving it in a bounded range of $\\alpha$ under RH (Proposition 1.7) and in a larger smoothed range under a uniform Hardy-Littlewood conjecture (Theorem 1.9).","pith_inferences":["The identity linking $m_n(\\alpha)$ to $H_*$ likely reflects a deeper reciprocity between prime-twisted pair correlations and triple zero correlations; if made structural, it could supply a new route to the triple correlation conjecture.","One can test the twisted pair correlation numerically for small $n$ (for example $n=2$) with existing high-zero data; deviations from the piecewise-linear $m_n$ would change the third-moment constant and show up long before any failure of RH.","The dependence on the uniform Hardy-Littlewood conjecture may be loose: a weaker error term might still prove the twisted conjecture on a shorter $\\alpha$-range, leaving the final constant unchanged.","The same twisted device should transfer to other $L$-functions with the same random-matrix statistics, giving the third-moment constant for $\\log L(1/2+it)$."],"forward_implications":["The real part of $\\log\\zeta$ has a non-Gaussian constant-order third moment, so its logarithmic moments are not pure Gaussian at the constant scale.","The imaginary part's third moment vanishes at order $1/\\log T$, matching the random-matrix prediction that odd moments of the imaginary part are negligible.","The twisted pair correlation conjecture becomes a new structural statement linking prime powers to zero-pair statistics, and the paper proves it in restricted ranges.","A uniform Hardy-Littlewood conjecture implies the full smoothed twisted pair correlation on a large range, connecting a prime-counting heuristic to zero-correlation statistics.","The constants $c_P$ and $c_Z$ are explicit and computable, so the third-moment prediction can be verified numerically at finite $T$."],"supporting_citations":[{"why":"Supplies the random matrix prediction for logarithmic moments that the paper's third-moment constant is designed to match.","marker":"[KS00]"},{"why":"Gives the pair correlation conjecture (Conjecture 1.2) and the baseline F(alpha) machinery that the twisted version generalizes.","marker":"[Mon73]"},{"why":"Provides the triple correlation conjecture (Conjecture 1.3) and its explicit kernel H(a,b).","marker":"[Hej94]"},{"why":"Gives the decomposition of log zeta into P(t)+Z(t) (Proposition 2.4) and the conditional second-moment estimate for the real part.","marker":"[LMQH23]"},{"why":"Provides the approximation of the imaginary part by P(t)+Z(t) and the conditional second-moment estimate used for the imaginary part.","marker":"[Gol87]"},{"why":"Supplies the mean value theorems for long Dirichlet polynomials that prove the smoothed twisted pair correlation in Theorem 1.9.","marker":"[GG98]"},{"why":"Gives the Landau-Gonek explicit formula used to evaluate sums of n^{i gamma} over zeros.","marker":"[Gon85]"},{"why":"Provides the singular series average estimates used to control the arithmetic factors in the twisted correlation range.","marker":"[FG95]"}],"fun_headline_variants":["Twisted pair conjecture nails log zeta's third moment","RH plus new conjecture pins down zeta log third moment","Third moment of log zeta: constant from primes and zeros","Log zeta third moment matches random-matrix predictions"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is the exact claimed shape of the twisted pair correlation function, and in particular the identity for primes that makes the prime-zero cross term cancel the pure zero term; if the true twisted correlation differs, the constant $c_Z=-\\pi^2/4$ would change.","fun_headline_variants_meta":{"raw":{"variants":["Twisted pair conjecture nails log zeta's third moment","RH plus new conjecture pins down zeta log third moment","Third moment of log zeta: constant from primes and zeros","Log zeta third moment matches random-matrix predictions"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000342,"raw_usage":{"total_tokens":1876,"prompt_tokens":936,"completion_tokens":940,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":552,"completion_tokens_details":{"reasoning_tokens":873}},"tokens_in":552,"tokens_out":940,"duration_ms":10298,"temperature":1.0,"reasoning_tokens":873,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T23:32:58.326864+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Numerically evaluate the twisted pair correlation sum $F_n(\\alpha)$ for a small prime power, say $n=2$, at a height where millions of zeros are known, and compare with Conjecture 1.5's piecewise-linear $m_n(\\alpha)$ across the interval $[1-\\log n/\\log T,\\,1]$; a systematic deviation would falsify the cancellation mechanism. Alternatively, compute $M^\\Re_3(T)$ and $M^\\Im_3(T)$ on the same zero data and check that the real part approaches $c_P+c_Z$ at the $1/\\log T$ scale while the imaginary part tends to zero.","supporting_citations":[],"review_version":1}