{"id":"e3ddb9b0-1968-4fae-9301-dc079cb1aaca","arxiv_id":"2601.10459","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Prime-time return-times and Wiener–Wintner averages converge almost surely for every measure-preserving system and L∞ function.","lead":"This paper proves that the classical return-times and Wiener–Wintner ergodic theorems still hold when the averaging times are the prime numbers, for L∞ functions. It is the first claimed extension of these theorems to a nontrivial arithmetic sequence and builds on new higher-order Fourier estimates for the Heath–Brown model of the von Mangoldt function.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Lemma 2.13's f(5)=1 unresolved case is load-bearing: if the claimed case analysis fails, the fixed-complexity U^3 bound drops to Q^{-1/4}, which the paper itself says is insufficient for both main theorems.","rationale":"The reader's weakest-assumption analysis correctly identifies Lemma 2.13 as the most load-bearing point. I read the proof chain in the opposite direction from the main theorems: Theorem 1.4 depends on the Return Times/Wiener–Wintner transference arguments, which in turn depend on controlling the large-Q Heath–Brown pieces in Lemma 4.1 via (1.16) and the fixed-complexity U^3 bound. The fixed-complexity bound Proposition 2.1 depends on the counting inequality (2.8)/(2.9), and (2.9) depends on Lemma 2.13. The manuscript's own sentence 'The case with f(5)=1 is still unresolved' is a direct admission that the lemma is not proved as written. This is not a demonstrated falsehood, so the correct verdict remains conditional rather than reject: the burden of proof is unmet until the finite combinatorial case analysis is completed or replaced. The proposed exhaustive 256-case check is cheap and would settle the question definitively.","tokens_in":29057,"tokens_out":4264,"duration_ms":42903,"concrete_test":"Exhaustively verify Lemma 2.13 by enumerating all subsets S of the cube {0,1}^3 (for size 5 there are 56 cases; for completeness do sizes 4–8 as well). For each 5-element S, test whether there exists a singleton T⊂S such that the face-propagation algorithm in Section 2 colours all vertices of S; if no such T exists, Lemma 2.13 is false and Proposition 2.1 collapses to the Q^{-1/4} bound. The same script should also check that the propagated values uniquely determine a_ωp modulo each prime p, as required for (2.9). A positive exhaustive check would remove the 'unresolved' caveat; a counterexample would invalidate the central U^3 estimate.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Proposition 2.1's Q^{-3/8+o(1)} U^3 bound for the fixed-complexity Heath–Brown terms is the quantitative engine of the paper. Its proof reduces to counting tuples (a_ω,q_ω) satisfying the integrality conditions (2.6)–(2.7); the essential saving beyond the elementary Q^{-1/4} bound comes exclusively from inequality (2.9) for Δ(q_ω). That inequality is derived from Lemma 2.13. The proof of Lemma 2.13 contains the explicit sentence 'The case with f(5)=1 is still unresolved' and then refers to a figure-based case analysis rather than supplying the missing argument. If even one 5-vertex configuration of the cube {0,1}^3 lacks a singleton T that drives the colour-propagation algorithm to completion, the bound (2.9) is not established, Proposition 2.1 is reduced to the easy Q^{-1/4} estimate, and the introduction states that improvement beyond Q^{-1/4} is essential for the main proof. Lemma 4.1 invokes the fixed-complexity estimate (2.2) precisely in the large-Q regime where the Q^{-3/8} saving is needed, and the Return Times argument in Section 5 is built on the same U^3 control. Thus both conclusions of Theorem 1.4 rest on this finite combinatorial lemma, which the manuscript itself flags as unresolved. This is an internal gap, not a disagreement with consensus; a secondary omitted transference lemma (Lemma 5.1) is also deferred, but Lemma 2.13 is the more fundamental unresolved assertion.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper claims two results: a Wiener–Wintner theorem along the prime times (for f∈L^p, p>1, with a full-measure set of x making the exponential prime averages converge for every frequency), and a Return Times theorem along the primes (for f,g∈L∞, convergence for all second systems). The strategy is to approximate the von Mangoldt function by a Heath–Brown model, prove a fixed-complexity U^3 bound (Proposition 2.1) and a U^3 approximation theorem (Proposition 3.1), and then transfer uniform Wiener–Wintner/return-times information via Gowers-norm inequalities (Proposition 1.14) and transference arguments (Lemmas 4.1 and 5.1). The central claimed contribution is the U^3 control of Heath–Brown models.","tokens_in":29508,"tokens_out":17673,"duration_ms":172795,"significance":"If correct, the results would be significant: they give the first Wiener–Wintner and Return Times theorems along an arithmetic sequence, namely the primes. The proof architecture is credible and non-circular: it imports the integer Wiener–Wintner/return-times theorems, the uniform Wiener–Wintner input, Wierdl's prime ergodic theorem, and quantitative U^3 inverse theorems, with no fitted parameters. The fixed-complexity U^3 estimate for Heath–Brown models is a plausible independent contribution. However, the manuscript currently leaves several load-bearing proofs unfinished (Lemma 2.13, Lemma 3.23, Lemma 5.1) and contains a false periodicity bound in the proof of Lemma 4.1. I am not convinced the central claims are established as written.","major_comments":[{"comment":"Lemma 2.13 is the combinatorial engine behind inequality (2.9), which in turn yields the Q^{-3/8} bound in Proposition 2.1; the introduction states that improvement beyond Q^{-1/4} is essential. The proof asserts f(1)=f(2)=f(3)=0 and f(4)=1, but for |S|=5 it contains the sentence 'The case with f(5)=1 is still unresolved' and then refers to Figure 2 instead of giving the case analysis. Since |S|=5 requires |T|=1, an unresolved case invalidates (2.9). Also, with T empty, the stated algorithm cannot color a one-vertex S, so f(1)=0 is not justified as stated. This is a load-bearing gap, not a presentation issue.","section":"Section 2, proof of Lemma 2.13"},{"comment":"The proof asserts that Λ_Q(n) is periodic with period P_Q := lcm(q≈Q) and 'recall the bounds 2Q ≤ P_Q ≤ 3Q'. This is false: for Q=8, the relevant squarefree q in (4,8] are 5,6,7, whose lcm is 210, while 3Q=24; in general the lcm of the dyadic interval is far larger than Q. The bound P_Q ≤ 3Q is used to make P_Q/N0 small in the first case (3Q≤K); without it, the block decomposition of L^θ_{Q,N} does not establish the claimed estimate. Lemma 4.1 therefore has a concrete error at a load-bearing point.","section":"Section 4, Lemma 4.1 (period bound for Λ_Q)"},{"comment":"Lemma 5.1 is the return-times analogue of Lemma 4.1 and the central transference estimate for the prime Return Times theorem. The proof is omitted with the phrase 'very close to that of Lemma 4.1, and so we omit the details.' Lemma 4.1's proof is long and relies on specific U^3 and periodicity estimates; the extra dependence on g_x(y-kn) and L^2(Y) norms is not a routine modification. This omission prevents verification of Section 5.","section":"Section 5, Lemma 5.1"},{"comment":"Lemma 3.23 provides the required bound on products of Ramanujan sums (3.24) and is used in the proof of Lemma 3.19 to control the Siegel-zero correction (3.18). Its proof is reduced to a counting argument 'very similar to Lemma 2.13' and then left to the reader. Given that Lemma 2.13 is itself incomplete, this deferred proof is especially problematic. Proposition 3.1 is not fully demonstrated without it.","section":"Section 3, Lemma 3.23"}],"minor_comments":[{"comment":"The abstract in the paper text highlights the Return Times theorem while the arXiv metadata highlights the Wiener–Wintner theorem; make these consistent.","section":"Title/abstract"},{"comment":"In the final parameterization, the condition 's|r, s|r^4' should be 'rad(s)|r, s|r^4' (and the surrounding text should be adjusted accordingly).","section":"Section 2, proof of Proposition 2.1"},{"comment":"References [14] and [15] are the same book with inconsistent bibliographic data; they should be merged or distinguished clearly.","section":"References"},{"comment":"Remark 1.18 contains an unproved inequality with 'we omit the details'; since the remark is not used in the main proof, either move it to future work or provide the proof.","section":"Remark 1.18"},{"comment":"The notation X≪Y is introduced with an 'extremely large' implicit constant, which is nonstandard; use the standard Vinogradov convention or define it precisely.","section":"Notation"}],"recommendation":"major_revision","confidential_remarks":"For the editor: the paper's strategy is promising and the target results are important, but the current version has multiple load-bearing gaps, including one explicitly labeled unresolved (Lemma 2.13) and one concrete false periodicity estimate (Section 4). I would not accept at this stage. A major revision with full proofs of Lemmas 2.13, 3.23, and 5.1 and a corrected treatment of the periodicity in Lemma 4.1 would be needed."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The key thing to know: this paper is worth taking seriously, but it is not ready to accept. The main theorem — Return Times along prime times, with Wiener–Wintner as a corollary — is genuinely new, and the proof architecture is credible. The quantitative engine, the U^3 bound Q^{-3/8+o(1)} for the fixed-complexity Heath–Brown model, is a real improvement over the easy Q^{-1/4}, and the U^3 approximation of von Mangoldt by the Heath–Brown model is a solid piece of Fourier analysis. If the proof can be completed, this is a major within-field result.\n\nThe soft spots are exactly where the reader and stress-test put them. Lemma 2.13 is explicitly flagged: the sentence 'The case with f(5)=1 is still unresolved' appears inside its proof, and the argument then leans on a figure-based case analysis. This lemma supplies inequality (2.9), which is the only thing pushing Proposition 2.1 beyond Q^{-1/4}. The paper itself says improvement beyond 1/4 is essential for both main theorems. So this is not a cosmetic gap. Lemma 5.1, the transference estimate needed specifically for the Return Times Theorem, is stated and then 'we omit the details.' That is the heart of the return-times argument. Lemma 3.23 also defers details to the reader. These are specific, addressable omissions rather than demonstrated errors, but they are load-bearing omissions.\n\nThere is also a small but real inconsistency: the arXiv abstract claims the first extension of the Wiener–Wintner Theorem to arithmetic sequences, while the full-text abstract claims the first extension of the Return Times Theorem. The second statement implies the first, so the full-text framing is the stronger and more accurate one. The abstract should be fixed.\n\nThe mathematics that is actually present looks careful and correct. The reduction of the fixed-complexity estimate to counting tuples on the cube is clean, and the use of U^3 control in the Wiener–Wintner proof is standard in the best sense. No fitted parameters, no suspicious circularity. The gaps are internal and combinatorial, not disagreements with established results.\n\nWho this is for: anyone working on pointwise ergodic theorems for arithmetic sequences, or on quantitative bounds for the Heath–Brown model. It deserves a serious referee and, ultimately, publication if the missing proofs are supplied. My recommendation: send it to peer review with a request for major revision, and insist that Lemma 2.13's unresolved case be worked out and Lemma 5.1's proof be written out before acceptance. The authors have shown they can do the hard parts; the remaining work is finite but essential.","headline":"A serious, novel extension of return-times/Wiener–Wintner ergodic theorems along the primes, but the current manuscript has load-bearing gaps (Lemma 2.13's unresolved case and Lemma 5.1's omitted proof) that block acceptance as-is.","tokens_in":29938,"tokens_out":1758,"would_cite":true,"duration_ms":21277,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["37A30","37A45","11N05","11B30"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper proves a Wiener–Wintner theorem along the primes, with return-times convergence for every second system.","keywords":["Wiener-Wintner theorem","Return times theorem","prime numbers","von Mangoldt function","Gowers norms","Heath-Brown model","pointwise ergodic theorem","higher order Fourier analysis"],"falsifier":"Give a concrete configuration of 5 marked vertices on the cube {0,1}^3 with no face containing 4 marked vertices for which no single green vertex, run through the described algorithm, colours all 5 vertices green; that would falsify Lemma 2.13 and remove the Q^{-3/8} bound. Alternatively, compute directly the U^3 norm of Λ_Q for large Q and find a lower bound exceeding Q^{-3/8+o(1)}.","tokens_in":28943,"feed_emoji":"🔢","tokens_out":4309,"duration_ms":44763,"temperature":0.7,"pith_summary":"The paper's goal is to prove the Return Times Theorem with averages taken over the prime numbers, and the Wiener–Wintner theorem as a corollary. It claims that for any measure-preserving system and any bounded function f, there is a full-measure set of starting points x such that for every second system (Y,S) and every bounded g, the averages (1/N)∑_{n≤N} f(T^{p_n}x)g(S^{p_n}y) converge almost everywhere in y. For the Wiener–Wintner part, the same convergence holds for all weights e^{2πi p_n θ} and for f in L^p with p>1. This matters because it is the first extension of these classical pointwise ergodic theorems to a sparse arithmetic sequence, and it requires blending the Heath–Brown decomposition of the von Mangoldt function with higher-order Fourier analysis and uniform Wiener–Wintner estimates. The full text states the main theorem as a Return Times Theorem; the abstract emphasizes the Wiener–Wintner corollary.","feed_headline":"Prime averages converge for every return-time system","feed_subtitle":"First extension of the Wiener–Wintner and Return Times theorems to prime times, for all bounded functions.","key_machinery":"The central object is the Heath–Brown model Λ_Q(n)=∑_{Q/2<q≤Q} μ(q)/φ(q)c_q(n), a rational-complexity approximation to the von Mangoldt function. The authors prove a fixed-complexity U^3 bound ∥Λ_Q∥_{U^3[N]} ≲ Q^{-3/8+o(1)} — an improvement beyond the easy Q^{-1/4} bound that is essential to the argument — and a U^3 approximation of the full von Mangoldt function by the model truncated at Q=exp((log N)^{1/10}). These bounds feed a transference argument in which U^3-small weights are plugged into Gowers–Cauchy–Schwarz inequalities, reducing the prime averages to uniform Wiener–Wintner estimates on weakly mixing functions.","core_discovery":"The central claim is Theorem 1.4: along the sequence of primes p_n, the return-times averages converge pointwise for all second systems, and the Wiener–Wintner averages converge for each fixed frequency θ. The mechanism is to write the prime indicator through the von Mangoldt function, approximate it by a truncated Heath–Brown model, control the U^3 norm of the part of the model with fixed rational complexity Q by the bound Q^{-3/8+o(1)}, and then transfer Gowers-norm control of the weight to almost-everywhere convergence of weighted ergodic averages via uniform Wiener–Wintner and weak-mixing reductions.","pith_inferences":["If the fixed-complexity U^3 estimate generalizes to U^s norms — the paper states such a generalization is forthcoming — the same strategy may yield analogues for averages over primes in polynomial or multicorrelation settings.","The unresolved f(5)=1 case of Lemma 2.13 is the most checkable point of the proof; verifying or refuting it decides whether the Q^{-3/8} improvement, and with it the present proof of both main theorems, stands.","The method suggests a testable extension: apply the same Heath–Brown-plus-U^3 decomposition to other sparse sequences such as primes plus squares, for which direct computation of the relevant U^3 norm could be compared against the conjectured bounds.","A reading that separates the two stated results is useful: the Wiener–Wintner corollary is presented as depending on the same U^3 machinery but with a weaker uniformity requirement, so a purely Wiener–Wintner proof might survive even if the full return-times statement needs adjustment."],"forward_implications":["For any bounded function on any measure-preserving system, the prime Wiener–Wintner averages converge for every frequency on a full-measure set; the L^p version holds for p>1.","The return-times statement holds: for a full-measure set of starting points, the sequence f(T^{p_n}x) is a universally good weight for pointwise L^2 ergodic theorems along the primes.","The fixed-complexity U^3 estimate gives quantitative control on how close the Heath–Brown model is to a prime-weight sequence, making it reusable in other prime-weighted ergodic averaging problems.","The argument shows that a weight with sufficiently small U^3 norm is enough to force almost-everywhere convergence in both the Wiener–Wintner and the return-times form, not just L^2 convergence."],"fun_headline_variants":["Prime-time averages converge for every system","Wiener–Wintner theorem extended to prime times","Return times along primes: pointwise convergence","Ergodic averages along primes converge pointwise"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The proof stands on the combinatorial counting lemma that bounds the number of solutions to the integrality system derived from the cube {0,1}^3, and the text reports that the case f(5)=1 of that lemma is still unresolved; if that counting bound fails, the Q^{-3/8} U^3 estimate collapses, and with it the proof of the main theorems.","fun_headline_variants_meta":{"raw":{"variants":["Prime-time averages converge for every system","Wiener–Wintner theorem extended to prime times","Return times along primes: pointwise convergence","Ergodic averages along primes converge pointwise"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000686,"raw_usage":{"total_tokens":2944,"prompt_tokens":739,"completion_tokens":2205,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":483,"completion_tokens_details":{"reasoning_tokens":2156}},"tokens_in":483,"tokens_out":2205,"duration_ms":17534,"temperature":1.0,"reasoning_tokens":2156,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T10:17:24.675500+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Give a concrete configuration of 5 marked vertices on the cube {0,1}^3 with no face containing 4 marked vertices for which no single green vertex, run through the described algorithm, colours all 5 vertices green; that would falsify Lemma 2.13 and remove the Q^{-3/8} bound. Alternatively, compute directly the U^3 norm of Λ_Q for large Q and find a lower bound exceeding Q^{-3/8+o(1)}.","supporting_citations":[],"review_version":1}