{"id":"c3380593-a9a9-4d59-8828-de1e3ab534cc","arxiv_id":"2608.04948","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"high","formal_verification":"none","parameter_count":6,"one_line_summary":"Claims an integral mean value estimate for Weyl sums over broken arcs with arbitrary moments, but the derivation contains critical gaps and self-referential optimization.","lead":"This preprint claims a new bound for the average m-th power size of a Weyl sum (an exponential sum) over short intervals and 'broken arcs', working for any power m instead of only even powers. The supporting proof is a sketch with multiple unsupported steps, so the claim is not established.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Lemma 4.2's independence of S_j(C_j α) is false; uncorrelatedness is not independence, and the Section 4 factorization for p_l collapses.","rationale":"I agree with the reader's identification of Lemma 4.2 as the weakest load-bearing point. This is more fundamental than the unproved Lemma 2.2 or the self-referential optimization in Lemma 6.1: even if the pointwise bounds were granted, the measure-theoretic heart of the argument would still fail because the probabilities for the differencing process do not factor. The counterexample above is elementary and settles the issue. I am not claiming Theorem 1.1 is false, only that this proof does not support it. Since the reader's verdict is already REJECT and my analysis confirms that verdict, no adjustment is needed.","tokens_in":15793,"tokens_out":9673,"duration_ms":97820,"concrete_test":"Compute the covariance E[S_1 \\overline{S_2}] for the simplest instance realizing Lemma 4.2: take interval {1,2}, S_1(α)=∑_{u=1}^2 e(αu^2), and S_2(α)=∑_{u=1}^2 e(αu). The integral equals the number of integer solutions to u^2=v, namely 1, while E[S_1]=E[\\overline{S_2}]=0. A nonzero value directly falsifies Lemma 4.2's assertion that the relevant mixed moments all vanish, and shows the claimed independence cannot justify the product formula for p_l used in Section 4.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The claimed exponent in Section 6 ultimately depends on the measure of 'integral II', computed in Section 4 as p_l, the probability that the Weyl differencing process stops at step l. That computation multiplies the measures of the events {α: |S_j(C_jα)| ≤ N^h} and {α: |S_{l+1}(C_{l+1}α)| > N^h} for A ≤ j ≤ l, factorizing the joint probability. The sole justification is Lemma 4.2, which asserts that S_j(C_jα) are independent random variables in the single uniformly distributed variable α. The proof only checks that certain mixed moments vanish, and claims they are all 0. This is a category error: pairwise uncorrelatedness does not imply independence, and deterministic functions of one random variable are independent only in degenerate cases. The vanishing claim is also numerically false: with interval {1,2}, S_1(α)=e(α)+e(4α) and S_2(α)=e(α)+e(2α) give E[S_1 \\overline{S_2}] = 1 while E[S_1]=E[\\overline{S_2}]=0. Hence the product law for p_l has no basis. Without it, the bound p_l ≪ N^{θ(2lλ_l−φ(m))} and the β_2 term in Lemma 6.1 are unsupported. The central mean-value estimate is therefore not established by this manuscript.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper claims a mean value theorem for Weyl sums over a 'broken arc' m* contained in a minor arc set m, with the exponent N^{θ m((2k+1)/(2k+2)+m^{-1/2-δ})} for arbitrarily large moment m subject to m^{1/2-δ-ε} > k > log m. The proposed proof combines a pointwise estimate from Vinogradov's main conjecture, a refined Weyl-differencing procedure splitting the second differences into 'large' and 'small' contributions, and a probabilistic calculation of the measure of the set where the differencing process terminates at a given step. The final section reduces the bound to a two-parameter optimization problem in which the saving parameter φ(m) is chosen self-referentially.","tokens_in":16240,"tokens_out":8442,"duration_ms":86566,"significance":"If the claimed estimate were correct, it would improve the trivial bound N^{θm} by a factor N^{θm/(2k+2)} on any positive-measure subset of the minor arcs and would remove the even-m restriction typical of Vinogradov-type mean value theorems. The proposed method of 'efficient partition of Weyl differences' combined with measure estimates is an interesting idea, and the paper makes a genuine effort to replace even moments by a probabilistic argument. However, the argument as written contains several load-bearing gaps and outright false statements: a central pointwise lemma is never proved, the independence of the relevant Weyl sums is asserted on incorrect grounds, a measure lemma has the wrong inequality direction, and the final optimization is circular. Because these issues affect the derivation of the main exponent, the manuscript does not establish its central claim.","major_comments":[{"comment":"Lemma 2.2, the pointwise bound |∑_{x∈H} e(α f(x))| ≪ N^{θ(1−1/(4(k+1)))+ε} for all α∈m, is asserted but never proved. The appendix is headed '[The proof of lemma 2.2]' but the argument actually proves Lemma 3.1, the counting formula for T(K); the final line of the appendix states 'then we prove lemma 3.1'. Lemma 2.2 is used at a critical juncture in Section 3 to control the small-difference contribution G1 and again in Section 4 to bound the residual Weyl sum after n differencing steps. Without a proof of Lemma 2.2, neither Theorem 1.2 nor the central Theorem 1.1 is supported.","section":"§2 and Appendix"},{"comment":"Lemma 4.2 asserts that the sums S_j(C_j α) = ∑_{N1<u<N2} e(α(C_j u^j + T_j)) are independent random variables in the single uniformly distributed variable α. This is false: all S_j are deterministic functions of one random variable, and non-degenerate functions of the same random variable cannot be independent. The supplied proof only checks that certain mixed moments vanish, and even that claim is wrong in general (for example, with S_1(α)=e(α)+e(4α) and S_2(α)=e(α)+e(2α), E[S_1 \\overline{S_2}] = 1, not 0). The factorization of the probability p_l in Section 4 as a product of individual probabilities for the events {|S_j| ≤ N^h} and {|S_{l+1}| > N^h} therefore has no basis, and the β_2 term in Lemma 6.1 is unsupported.","section":"§4, Lemma 4.2"},{"comment":"Lemma 4.1 has the wrong inequality direction and an incorrect scaling. From ∫|S|^m ≤ N^{θ(m−φ(m))} one can upper-bound the measure of the set where |S| is LARGE (by Markov's inequality), not the set where |S| is small. The proof writes x^m L(...) ≤ ∫|S|^m, but with x = |S|/N^{θ/2} the left side should be x^m N^{θm/2} L(...); the displayed inequality omits the factor N^{θm/2} and consequently cannot hold as stated. Moreover, the set {|S| ≤ N^{h+θ/2}} can have full Lebesgue measure, contradicting the claimed upper bound for many choices of h and φ(m). The subsequent use of this lemma to estimate both P(|S_j| ≤ N^h) and P(|S_{l+1}| > N^h) by the same quantity in the computation of p_l is unjustified.","section":"§4, Lemma 4.1"},{"comment":"The definition of G* is self-referential and therefore circular: G* is defined as G* = inf ... inf_{φ(m) ≤ m−G*} max{β1, β2}, so G* appears in the constraint of its own definition. In the proof the author then sets φ(m)=G*, and in case 2 sets φ(m)=θm−G*−ε, which is inconsistent with the prior identification φ(m)=G*. The final step chooses A, y, n specifically to force G*/m = 1−1/(2k+2), so the claimed exponent is fitted by construction rather than derived from the optimization problem. Because of this circularity, the main bound of Theorem 1.1 is not established by the preceding estimates.","section":"§6, Lemma 6.1"},{"comment":"Lemma 3.2 is not proved by a valid argument. The proof invokes a 'stochastic process' G_α(x)=x+α and the orbit {T_α^n(x)}, but an irrational rotation is deterministic and not a stochastic process in this setting. The displayed chain 'Σ_{q<(log N)^A} P(qα ∉ m) = Σ_{j<N^{k/2}} P(||T^j_α(α)|| ∉ m) ≤ ...' is not a meaningful derivation, and the quantity T_α^{-n}((0,N^{-1/2}]) is undefined in the context. This lemma is used to define the Bad set and to justify |Bad(j)| ≍ N^{-j/2}; without it, the exclusion of the 'bad' differences in Sections 3–5 has no support.","section":"§3, Lemma 3.2"}],"minor_comments":[{"comment":"The notation is confusing: the moment m and the minor-arc set m are both denoted by 'm', and the paper switches between m, 𝔪, and m* without clear distinction. This makes the statement of Theorem 1.1 and the proof difficult to follow.","section":"Throughout"},{"comment":"The appendix uses the phrase 'By Pascal theorem' where the binomial theorem is intended. The interchange of an infinite sum and an integral in the derivation of T(K) is not justified; a dominated convergence argument is mentioned but no dominated function is identified.","section":"§2 and Appendix"},{"comment":"The relation between the error term m^{1/2+c_1 δ} in Lemma 6.1 and the error term m^{1/2−δ} in Theorem 1.1 is not explained. The symbols δ, c_1, c_2 are introduced with overlapping roles, and the final conclusion does not follow from the displayed choices of y_0 and n_0 without further argument.","section":"§6"},{"comment":"There are many grammatical and typographical errors, for example 'In this article, we research the mean value', 'm is be sufficiently large', 'By the version of probability', and 'we can overcomes the parity'. The paper would benefit from careful editing.","section":"Throughout"},{"comment":"Theorem 1.2 is stated in the introduction but its proof is never clearly identified. If it is intended to follow from Lemma 2.2, then the proof is incomplete because Lemma 2.2 is unproved; if it follows from other results, that dependence should be stated explicitly.","section":"§1, Theorem 1.2"}],"recommendation":"reject","confidential_remarks":"The manuscript has several fundamental mathematical errors that cannot be repaired by local edits: Lemma 4.2 is false as stated, Lemma 4.1 has the wrong inequality direction, Lemma 2.2 is unproved despite being the cornerstone of the pointwise bound, and Lemma 6.1 is self-referential. The probabilistic framework appears to be seriously flawed, and the final exponent is obtained by construction rather than by a valid derivation. I would not encourage resubmission in the current form."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this paper takes on a genuinely interesting open problem—mean values of a single Weyl sum over broken minor arcs with arbitrary moment m—and the result it announces would be a real advance. But the proof is not close to supporting it. The main probabilistic lemma is false, a key measure inequality is reversed, the appendix does not prove the lemma it claims to prove, and the final optimization is circular. I would not spend much time on the current version.\n\nWhat is genuinely new: the statement of Theorem 1.1 does not appear in the cited literature. Known mean value theorems for the single exponential sum e(αu^k) require even s, and the broken-arc mean value theorems for the full Vinogradov system are a different object. The idea of partitioning Weyl differences into large and small ranges, then using measure estimates to control the stopping time, is a reasonable high-level strategy. The author also correctly builds on Vinogradov's main conjecture. Credit where due.\n\nThe soft spots are not minor. First, Lemma 2.2 is asserted without proof; the appendix titled as its proof derives the counting function T(K) from Lemma 3.1, not the pointwise Weyl sum bound. That leaves case II without its main tool. Second, Lemma 4.1 has the inequality backwards. A mean value bound bounds the measure of the set where |S| is large, not where it is small; the proof's line 'x^m L(...) ≤ ∫|S|^m' is wrong in direction. Third, Lemma 4.2 claims that S_j(C_j α) are independent random variables in a single uniform α. They are deterministic functions of the same variable; uncorrelatedness never gives independence, and the specific moments are not all zero. The factorization of p_l, the probability that the differencing stops at step l, collapses, and with it the β2 term in Section 6. Fourth, Lemma 6.1 defines G* using a constraint that involves G*, then sets φ(m)=G* and chooses A,y,n to force the ratio to 1−1/(2k+2). That is fitting the answer, not deriving it.\n\nThe theorem might be true, and the broken-arc mean value problem is worth working on. But this manuscript does not establish it. The flaws are load-bearing.\n\nWho would get value from this? An expert in the area might read it for the formulation of the problem and the broad strategy; it could be a starting point for a real proof. It is not suitable for publication in anything like its current form.\n\nMy recommendation: if this lands on your desk, reject it, but a substantive referee report listing the four issues above would be more useful to the author than a summary desk rejection. The problem is significant enough that a serious referee should document the gaps; if the editor prefers to desk reject, I would not fight it.","headline":"Genuinely new problem and promising strategy, but the proof fails at multiple load-bearing points—false independence, reversed inequality, circular optimization—so the main theorem is not established.","tokens_in":16707,"tokens_out":5773,"would_cite":false,"duration_ms":61059,"reading_group":"maybe","serious_thinker":"no","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["11L15"],"pacs":[],"model":"deepseek-v4-flash","headline":"For degree-$k$ Weyl sums over short intervals, this paper proves an $m$-th moment bound on broken minor arcs saving $1/(2k+2)$ over the trivial estimate, for all sufficiently large $m$.","keywords":["Weyl sums","mean value theorem","broken arcs","minor arcs","short intervals","efficient partition of Weyl difference","Diophantine approximation","Vinogradov mean value theorem"],"falsifier":"Pick a concrete $k$, $N$, and $\\theta$, choose two of the differenced sums $S_i(C_i\\alpha)$ and $S_j(C_j\\alpha)$ from Lemma 4.2, and choose a level $h$ so that each individual set $\\{\\alpha: |S_i(C_i\\alpha)|\\le N^{h/2}\\}$ has measure about a small power of $N^{-1}$. Compute the measure of the intersection of the two sets by direct quadrature. If this joint measure differs from the product of the two individual measures by more than the allowed error, the factorization used to compute the probability in Section 4 fails, and the final exponent would need a different argument.","tokens_in":15557,"feed_emoji":"🧮","tokens_out":14104,"duration_ms":138119,"temperature":0.7,"pith_summary":"This paper studies the $m$-th moment integral of a Weyl sum $S(\\alpha)=\\sum_{N_1<u<N_2}e(\\alpha u^k)$ over a 'broken arc' $\\mathfrak{m}^*\\subset \\mathfrak{m}$, where the interval $[N_1,N_2]$ is short, of length about $2N^\\theta$ with $\\theta<1$. The central claim, Theorem 1.1, is that for every sufficiently large $m$ with $k>\\log m$ and $k<m^{1/2-\\delta-\\epsilon}$, any subset $\\mathfrak{m}^*$ of fixed positive measure satisfies $\\int_{\\mathfrak{m}^*}|S(\\alpha)|^m\\,d\\alpha\\ll_c N^{\\theta m((2k+1)/(2k+2)+m^{-1/2-\\delta})}$. The bound beats the trivial $N^{\\theta m}$ by the factor $N^{-\\theta m/(2k+2)}$ and applies to all large $m$, not only even $m$ as in classical single-variable mean value bounds. It supplies a new regime of mean value estimates for scalar exponential sums over incomplete arcs, with a proposed mechanism, an 'Efficient Partition of Weyl Difference', that separates large and small differencing steps and weights them by a measure-theoretic count.","feed_headline":"Broken-arc moment bound saves 1/(2k+2) over trivial Weyl estimate","feed_subtitle":"For all large m, no even-m restriction is needed for Weyl-sum moments over short intervals.","key_machinery":"The central machinery is the 'Efficient Partition of Weyl Difference'. After a standard Weyl differencing step, the remaining sum is split according to whether the difference variable is large (case I), small with good Diophantine quality (case II), or small with bad Diophantine quality (case III). Case III is shown to be negligible, while the relative size of cases I and II decides whether to continue differencing; the analysis is driven by a pointwise bound (Lemma 2.2) obtained from Diophantine approximation and the main conjecture for Vinogradov's mean value theorem, and by a probabilistic bookkeeping (Lemmas 4.1 and 4.2) that counts how many differencing paths contribute to each case. The final exponent comes from a constrained optimization problem (Lemma 6.1) in which the saving function $\\phi(m)$ appears on both sides, a self-reference the paper resolves by its choice of $A$, $y$, and $n$.","core_discovery":"On the paper's own terms, the discovery is Theorem 1.1: the $m$-th moment of a degree-$k$ Weyl sum over a short interval, integrated over any fixed-measure subarc of the minor-arc set $\\mathfrak{m}=\\bigcap_{j\\le k-1}\\{\\alpha: |\\alpha-h/q|>q^{-1}N^{-(k-j-1/2)\\theta}\\text{ for all }q<(\\log N)^A\\}$, obeys $\\int_{\\mathfrak{m}^*}|S(\\alpha)|^m\\,d\\alpha\\ll_c N^{\\theta m((2k+1)/(2k+2)+m^{-1/2-\\delta})}$. The exponent contains an explicit saving of $1/(2k+2)$ over the trivial $N^{\\theta m}$, and the hypothesis on $m$ is size alone ($m^{1/2-\\delta-\\epsilon}>k>\\log m$), with no parity condition. The paper also records Theorem 1.2, the same type of bound in the form $N^{\\theta(m-m/(4(k+1)))+\\epsilon'}$, and argues that the argument improves the pointwise Vinogradov-style estimate on almost all of $\\mathfrak{m}$ by using different bounds for different $\\alpha$. The method is claimed to remove the 'even-odd restriction of powers' that appears in the classical second form of the mean value theorem.","pith_inferences":["Editorial inference: a natural control problem suggested by the argument is to quantify the correlation between the differenced sums; the proof uses exact factorization, and any explicit correlation bound of size $o(1)$ in the exponent would make the same optimization scheme more robust.","Editorial inference: the efficient-partition scheme could be adapted to simultaneous Weyl sums over a $k$-dimensional torus, where the broken arc becomes a broken box; the obstacle would be a multidimensional analogue of the Diophantine approximation lemma.","Editorial inference: a testable consequence of the method is that the saving factor should be uniform in the choice of subset $\\mathfrak{m}^*$ as long as its measure is bounded below by a constant; this uniformity is stronger than what pointwise bounds would give and could be checked in small-parameter experiments."],"forward_implications":["For every fixed-measure broken arc, the $m$-th moment obeys $\\ll_c N^{\\theta m((2k+1)/(2k+2)+m^{-1/2-\\delta})}$, a saving of $N^{-\\theta m/(2k+2)}$ over the trivial estimate.","The parity restriction is removed: the bound holds for all sufficiently large $m$ with $k>\\log m$ and $k<m^{1/2-\\delta-\\epsilon}$, not just even $m$.","Theorem 1.2 gives the concrete incomplete-interval bound $N^{\\theta(m-m/(4(k+1)))+\\epsilon'}$ for moments of short-interval Weyl sums.","On almost all of the minor-arc set, the moment estimate beats the exponent obtained from pointwise Vinogradov-type bounds, because different $\\alpha$'s are treated with different estimates.","The three-case differencing decomposition, if correct, supplies a template for separating large and small difference ranges in other mean value problems."],"supporting_citations":[{"why":"Proves the main conjecture in Vinogradov's mean value theorem for degrees higher than three, giving the bound on the Vinogradov integral used in Lemma 2.1.","marker":"[3]"},{"why":"Supplies the cubic case of the main conjecture, so the pointwise bound covers all degrees k at least 3.","marker":"[15]"},{"why":"Provides the classical reduction of a Weyl sum to products of linear exponential sums and a Vinogradov-type integral, used in the proof of Lemma 2.1.","marker":"[12]"},{"why":"Gives the classical even-s bound for the single-variable mean value theorem that the new result removes the parity restriction from.","marker":"[17]"},{"why":"Provides the broken-arc mean value theorem that motivates the setting and serves as the comparison point for the incomplete-arc results.","marker":"[11]"},{"why":"Supplies the metric theory of Weyl sums that underlies the measure-theoretic counting of the differencing cases.","marker":"[5]"}],"fun_headline_variants":["Broken-arc Weyl sum moments gain 1/(2k+2) saving","No even-m restriction for short-interval Weyl sums","Integral mean value theorem beats trivial Weyl bound","Fixed-measure minor arcs: moment exponent improved","Weyl sum moment saving without parity condition"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is Lemma 4.2's claim that after differencing, the several Weyl sums $S_j(C_j\\alpha)$ behave like independent random variables in the single parameter $\\alpha$, so the probability that the differencing process survives to step $l$ is the product of individual probabilities; since all these sums are deterministic functions of the same $\\alpha$, the factorization is not automatic and the measure estimates in Sections 4 and 6 rest on it.","fun_headline_variants_meta":{"raw":{"variants":["Broken-arc Weyl sum moments gain 1/(2k+2) saving","No even-m restriction for short-interval Weyl sums","Integral mean value theorem beats trivial Weyl bound","Fixed-measure minor arcs: moment exponent improved","Weyl sum moment saving without parity condition"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00019,"raw_usage":{"total_tokens":1456,"prompt_tokens":1180,"completion_tokens":276,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":796,"completion_tokens_details":{"reasoning_tokens":193}},"tokens_in":796,"tokens_out":276,"duration_ms":3575,"temperature":1.0,"reasoning_tokens":193,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T13:08:36.866759+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Pick a concrete $k$, $N$, and $\\theta$, choose two of the differenced sums $S_i(C_i\\alpha)$ and $S_j(C_j\\alpha)$ from Lemma 4.2, and choose a level $h$ so that each individual set $\\{\\alpha: |S_i(C_i\\alpha)|\\le N^{h/2}\\}$ has measure about a small power of $N^{-1}$. Compute the measure of the intersection of the two sets by direct quadrature. If this joint measure differs from the product of the two individual measures by more than the allowed error, the factorization used to compute the probability in Section 4 fails, and the final exponent would need a different argument.","supporting_citations":[{"cited_title":"Proof of the main conjecture in vinogradov’s mean value theorem for degrees higher than three.Ann.Math., pages 633–682(2), 2016","cited_arxiv_id":null,"evidence_quote":"Proves the main conjecture in Vinogradov's mean value theorem for degrees higher than three, giving the bound on the Vinogradov integral used in Lemma 2.1."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the cubic case of the main conjecture, so the pointwise bound covers all degrees k at least 3."},{"cited_title":"Pan and C.B","cited_arxiv_id":null,"evidence_quote":"Provides the classical reduction of a Weyl sum to products of linear exponential sums and a Vinogradov-type integral, used in the proof of Lemma 2.1."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the classical even-s bound for the single-variable mean value theorem that the new result removes the parity restriction from."},{"cited_title":"An extended Vinogradov's mean value theorem","cited_arxiv_id":"2506.01751","evidence_quote":"Provides the broken-arc mean value theorem that motivates the setting and serves as the comparison point for the incomplete-arc results."},{"cited_title":"Metric theory of weyl sums, 2023","cited_arxiv_id":null,"evidence_quote":"Supplies the metric theory of Weyl sums that underlies the measure-theoretic counting of the differencing cases."}],"review_version":1}