{"id":"3af91894-98d9-4849-b9a1-5ae89f1c0294","arxiv_id":"1908.06292","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"For any slowly growing function f, there exists a Poissonian pair correlation sequence whose number of distinct neighboring gap lengths is at most f(n), and for every such sequence the maximal gap multiplicity is o(n).","lead":"This mathematics paper proves two results about sequences of points on a circle whose spacings look random. It shows that no single gap size can dominate, and that the number of distinct gap sizes can grow arbitrarily slowly while pair correlations stay Poissonian, answering an open question by Gerhard Larcher.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Claim 2's variance bound relies on a false pairwise-independence assertion for summands sharing an index; without a covariance estimate the PPC proof lacks its key probabilistic input.","rationale":"The paper's central achievement is Theorem 1.4, proved by a random construction that must simultaneously yield PPC almost surely and few distinct gap lengths. The PPC side rests on Claims 1-6. Claim 1 on expectations is standard. Claim 2 is the first point where the proof is not merely terse but wrong as written: it asserts pairwise independence of all summands in a quadratic form in the X_i. Independence is valid for disjoint index pairs but not for pairs sharing an index, and shared-index pairs dominate the variance calculation. The proof then ignores their covariance, so Var[Ftilde_{X,N}(s)] << 1/N is unproved. Since Claim 3 immediately converts this variance into P(Q_N) << 1/N^2 and invokes Borel-Cantelli, the almost-sure PPC of the random component is unsupported. This is a real proof gap, not a stylistic issue. It is very likely repairable: the random variables are uniform on nested dyadic grids, so shared-index covariances admit explicit bounds; for j,k in block m, the relevant product probability is about 1/(N 2^{m+a(m)}), and the O(N^3) triples contribute O(N) after normalization, yielding the same O(1/N) variance. Thus the gap should not trigger rejection, only conditional acceptance pending the missing covariance estimate. Claim 4's interpolation inequality is also not valid as printed for N in [M^2, M^2+M), but this too is repairable by choosing the upper comparison point from the adjacent square or by letting s vary with N and using monotonicity plus convergence on a dense set. Since both defects are local and repairable, the reader's CONDITIONAL verdict is appropriate; the stress-test identifies the same main concern and does not move the verdict.","tokens_in":10779,"tokens_out":20388,"duration_ms":219099,"concrete_test":"Re-derive Claim 2 without the independence claim: compute Cov(1_{|X_i-X_j| <= s/(y_N)}, 1_{|X_i-X_k| <= s/(y_N)}) for j,k in the same dyadic block I_m, using the fact that A_j and A_k are nested subgroups of A_i, and verify that the total triple-covariance sum is O(N) (or O(N log N)), giving Var[Ftilde_{X,N}(s)] = O(1/N). If the required estimate needs a(m) to grow faster than a(m) = ceil(h(m)/2), then the PPC and gap-counting constraints in Claim 7 conflict and the construction must be modified.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The most load-bearing defect is in Claim 2 (Section 3), where the proof asserts that the indicator variables 1_{[-s/(y_N), s/(y_N)]}(X_{i1}-X_{j1}) and 1_{[-s/(y_N), s/(y_N)]}(X_{i2}-X_{j2}) are independent for all distinct pairs (i1,j1) and (i2,j2), citing [6, Corollary 272L]. This is false when the pairs share an index: for example, h(X_1-X_2) and h(X_1-X_3) are both functions of X_1 and are not independent. Since Var[Ftilde_{X,N}(s)] is then written as the sum of individual variances with no covariance term, the bound Var[Ftilde_{X,N}(s)] << 1/N in Claim 2 is not established. Claim 3 uses exactly this bound to get P(Q_N) << Var[Z_N] << 1/N^2 and then applies Borel-Cantelli; without a valid variance bound, the almost-sure PPC of the random component, and hence the random+deterministic construction in Claim 6, is unsupported. The defect is internal: the written proof is invalid at this step, even though the estimate itself is plausibly repairable by a covariance/Hoeffding decomposition exploiting the nested dyadic grids, with shared-index covariance terms bounded by about 1/(N 2^{k+a(k)}) and summing to O(1/N). A second, smaller gap is the interpolation inequality in Claim 4, which fails for N between M^2 and w_{(M+1)^2}=M^2+M; this is also repairable, but it means the route from square-index PPC to full PPC is not as written.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the number of distinct gap lengths between neighboring elements of the first N terms of a sequence on the torus, under the assumption that the sequence has Poissonian pair correlations (PPC). The authors prove two main results. Theorem 1.2 improves a previous result by showing that the maximum multiplicity among neighboring gap lengths is o(n) for every PPC sequence. Theorem 1.4 answers a question of Larcher negatively: for every function f with f(n) -> infinity, there exists a sequence with PPC such that the number of distinct gap lengths g(n) is at most f(n) for all sufficiently large n. The proof of Theorem 1.4 constructs a random sequence consisting of independent random variables supported on dyadic grids, interleaved with deterministic blocks, and shows that almost surely the combined sequence has PPC while having few distinct gap lengths.","tokens_in":11164,"tokens_out":6411,"duration_ms":58655,"significance":"If both theorems are correct, the paper settles Larcher's Question 1.3 in the negative and provides a sharp necessary condition (max multiplicity o(n)) for PPC. The construction is original and the deterministic-block/random-block technique is a natural and potentially reusable tool. The proof of Theorem 1.2 is clean and self-contained. The main probabilistic construction in Theorem 1.4 is conceptually sound, and the overall structure of the estimates is plausible. However, as written, the proof of Theorem 1.4 contains two load-bearing gaps: the variance bound in Claim 2 rests on a false independence assertion, and the interpolation in Claim 4 fails for N near M^2. Both appear repairable, but the current manuscript does not establish Theorem 1.4 as written.","major_comments":[{"comment":"The proof asserts that the indicator random variables 1_{[-s/y_N,s/y_N]}(X_{i1}-X_{j1}) and 1_{[-s/y_N,s/y_N]}(X_{i2}-X_{j2}) are independent for all distinct pairs (i1,j1) and (i2,j2), citing [6, Corollary 272L]. This is false when the two pairs share an index, e.g., h(X_1-X_2) and h(X_1-X_3) are both functions of X_1 and are not independent. Consequently, the displayed identity Var[Ftilde_{X,y,N}(s)] = (4/N^2) Var(sum ...) does not reduce to the sum of individual variances; covariance terms are omitted. The claimed bound Var[Ftilde_{X,y,N}(s)] << 1/N is therefore not established. This bound is exactly what Claim 3 uses via Chebyshev's inequality and Borel-Cantelli to obtain almost sure PPC along square indices, so the central probabilistic input for Theorem 1.4 is missing as written. A covariance or Hoeffding-type decomposition is needed.","section":"Section 3, Claim 2"},{"comment":"The interpolation step from square-index PPC to full PPC contains an invalid inequality for N between M^2 and w_{(M+1)^2} = (M+1)^2 - (M+1) = M^2+M. In the displayed chain, the upper bound uses Ftilde_{X,w,(M+1)^2}(s) = (1/(M+1)^2) #{ |Xi-Xj| <= s/w_{(M+1)^2} }. For N in [M^2+1, M^2+M], we have w_{(M+1)^2} > N, so s/w_{(M+1)^2} < s/N and the count with the smaller threshold is not an upper bound for F_{X,N}(s). Thus the right-hand inequality in Claim 4 fails for such N. The passage from the almost sure limits at square indices to the full limit F_{X,N}(s) -> 2s is therefore not established as written. This is a separate but also load-bearing gap; the argument needs a different comparison, for example using two-sided bounds with both v and w at appropriately chosen indices.","section":"Section 3, Claim 4"}],"minor_comments":[{"comment":"The phrase 'gap len gths' contains a typo and should read 'gap lengths'.","section":"Abstract"},{"comment":"In the sentence 'the last O(1) comes the fact that ...', the word 'from' is missing; it should read 'comes from the fact that'.","section":"Section 3, Claim 2"},{"comment":"The name 'Cauchy–Schwartz' should be 'Cauchy–Schwarz'.","section":"Section 3, Claim 6"},{"comment":"The inclusion B_{m-1} \\subseteq {Z_1(ω), ..., Z_n(ω)} \\subseteq A_m is asserted without proof; a one-sentence justification that the deterministic blocks up to m are contained in B_m and that B_m \\subseteq A_m under the chosen a and b would improve readability.","section":"Section 3, Claim 7"},{"comment":"In the sentence 'there would exist t \\in N+ such that ...', the expression 'there would exist t' should be 'there would exist a t' or 'there exists t' for grammatical correctness.","section":"Section 2, proof of Theorem 1.2"}],"recommendation":"major_revision","confidential_remarks":"The two gaps identified in the major comments are serious but plausibly repairable: the variance bound likely can be fixed with a covariance estimate exploiting the dyadic structure, and the interpolation in Claim 4 can likely be repaired by a more careful choice of indices. The results themselves are significant and would be of interest to the community. I recommend major revision rather than rejection. I also note that the proof of Theorem 1.2 is clean and could be published independently even if the construction in Section 3 requires more work."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two genuinely new results here. Theorem 1.2 upgrades Larcher–Stockinger's liminf g(n)=∞ to the stronger max_i φ_{n,i}=o(n), and the proof is clean. Theorem 1.4 answers Larcher's question in the negative: for any f→∞ there is a PPC sequence with g(n)≤f(n). The block construction is the right idea and the high-level strategy is coherent. The paper is worth taking seriously.\n\nThe soft spots are real, though both look repairable. Claim 2 (Section 3) asserts that the random variables 1_{[-s/y_N,s/y_N]}(X_{i1}-X_{j1}) and the corresponding indicator for a different pair are independent whenever the pairs differ. That's false when the pairs share an index, e.g. (1,2) and (1,3) both involve X_1. The cited Fremlin result doesn't cover that. Without the variance bound Var[F~_{X,N}(s)] ≪ 1/N, Claim 3's Borel–Cantelli argument has no input. This is a load-bearing gap, but it's the right kind of gap: the bound itself is plausible and a covariance/Hoeffding decomposition should recover it.\n\nThe second gap is smaller. In Claim 4 the interpolation between squares uses the sequences v_n=n+⌊√n⌋ and w_n=n−⌊√n⌋, but for N in [M^2, M^2+M) the point w_{(M+1)^2} lies above N, so the upper bound inequality can fail. Same repair: take M = ⌊√N⌋ and handle the interval [M^2, M^2+M) separately. So the route from square-index PPC to full PPC is not as written.\n\nThe citation pattern looks fine; the authors cite the relevant literature and the constructions are original. There are no fitted parameters—the free functions a(m), b(m) are chosen after the fact, which is legitimate in an existence proof.\n\nBottom line: the paper deserves a serious referee. If the authors fix Claim 2 and smooth out Claim 4, I think the results go through. I'd send it to peer review and flag the two proof gaps for the referee.","headline":"Strong new results; proof gaps in Claim 2 and Claim 4 are repairable but must be fixed.","tokens_in":11656,"tokens_out":2678,"would_cite":true,"duration_ms":25963,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["11K06","11B05","11K99"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper proves two gap-statistics theorems for Poissonian pair-correlation sequences: the maximal gap multiplicity is $o(n)$, and for every $f(n)\\to\\infty$ a Poissonian sequence with at most $f(n)$ distinct gap lengths exists.","keywords":["Poissonian pair correlations","gap lengths","dyadic grids","random sequences","almost-sure convergence","equidistribution","three gap theorem","open problem on gap counts"],"falsifier":"A direct refutation of Theorem 1.2 would be a Poissonian sequence with a subsequence on which some single gap length occurs at least $cn$ times for a fixed $c>0$; no such sequence can exist if the theorem is right. For the construction, computing the true covariance between pair indicators that share an index would decide whether the claimed concentration genuinely holds.","tokens_in":10612,"feed_emoji":"📏","tokens_out":17844,"duration_ms":149637,"temperature":0.7,"pith_summary":"Sequences on the unit interval have Poissonian pair correlations when their normalized two-point spacing counts match those of a Poisson process. The paper asks how few distinct gap lengths between neighboring ordered points such a sequence can have, and how uneven the gap multiplicities can be. It establishes two things: in every Poissonian sequence the most frequent gap length occurs only $o(n)$ times, so the number of distinct gap lengths must tend to infinity; and yet, for any prescribed unbounded function $f$, there exists a Poissonian sequence with fewer than $f(n)$ distinct gaps eventually. The second statement answers negatively an open question posed at a 2019 workshop, which asked whether a universal slowly growing lower bound on the number of gaps is forced by Poissonian pair correlations.","feed_headline":"Random-like sequences can have very few gap lengths","feed_subtitle":"Even the slowest-growing bound on distinct gaps is compatible with Poissonian pair correlations.","key_machinery":"The central mechanism is a two-scale dyadic-grid construction. Random blocks $(X_i)$ are independent uniform draws from the grids $A_m=\\{j/2^{m+a(m)}\\}$, and deterministic blocks $(Y_{m,j})$ are the new points $C_m=B_m\\setminus B_{m-1}$ of the coarser dyadic grids $B_m=\\{j/2^{b(m)}\\}$, inserted between random blocks. The random blocks supply Poissonian pair correlations through expectation and variance estimates followed by a standard almost-sure convergence lemma; the deterministic blocks control the gap count through the sandwich $B_{m-1}\\subseteq\\{Z_1,\\ldots,Z_n\\}\\subseteq A_m$, which bounds the number of gap lengths by a power of two governed by the prescribed gap function $h(n)=\\lfloor\\log_2 q(n)\\rfloor$.","core_discovery":"The central discovery is that Poissonian pair correlations impose a very mild constraint on gap-length statistics. Theorem 1.2 shows that if a sequence has Poissonian pair correlations, then $\\max_{i\\le g(n)}\\phi_{n,i}=o(n)$; since at least one gap length must occur at least $n/g(n)$ times, the total number of distinct gap lengths $g(n)$ tends to infinity. Theorem 1.4 shows this is essentially the only constraint: for any function $f(n)\\to\\infty$ one can construct a sequence $(x_n)$ with Poissonian pair correlations and $g(n)\\le f(n)$ for all sufficiently large $n$. The construction interleaves deterministic dyadic grid points, which keep the gap set small, with random dyadic grid points, which force the Poissonian pair-correlation statistics. The second theorem answers the open question negatively, while the first shows the limitation is real: the gap count must still tend to infinity.","pith_inferences":["Beyond the paper, a natural next step is to make the construction explicit: the proof is probabilistic and yields almost-sure existence, so a deterministic version of the dyadic-block construction would turn the existence result into a concrete sequence.","The two-scale grid suggests a quantitative tradeoff between the prescribed gap bound and the denominator growth of the random blocks; one could ask how fast $a(m)$ and $b(m)$ must grow as a function of $f$.","The proof of $\\max \\phi_{n,i}=o(n)$ is by contradiction and gives no rate; extracting a quantitative rate would connect these gap statistics to discrepancy-type estimates for Poissonian sequences."],"forward_implications":["For every sequence with Poissonian pair correlations, the maximum multiplicity of a neighboring gap length is $o(n)$; since $g(n)\\ge n/\\max_{i\\le g(n)}\\phi_{n,i}$, the number of distinct gap lengths tends to infinity.","Combined with the Three Gap Theorem, this implies that no sequence of the form $(\\{n\\alpha\\})$ has Poissonian pair correlations, because such a sequence has at most three gap lengths and therefore a gap multiplicity at least $n/3$.","There is no universal slowly growing lower bound on $g(n)$: for every $f(n)\\to\\infty$ there is a Poissonian sequence with $g(n)\\le f(n)$ for all large $n$, answering the open question negatively.","The constructed sequences show that the typical behavior of almost all sequences, which have all $n$ adjacent gaps distinct, is not a necessary feature of Poissonian pair correlations."],"supporting_citations":[{"why":"It supplies the earlier result that the number of distinct gap lengths cannot stay bounded along any subsequence, which Theorem 1.2 strengthens.","marker":"[11]"},{"why":"It states the problem version that the paper's negative answer targets and provides the context for the open question.","marker":"[9]"},{"why":"It supplies the measure-theoretic independence corollary used in the variance estimate for the random blocks.","marker":"[6]"},{"why":"It is one of the references establishing that almost all sequences have Poissonian pair correlations, the probabilistic backdrop for the construction.","marker":"[8]"},{"why":"It is another reference for almost-all sequences having Poissonian pair correlations, used as the starting point for the random blocks.","marker":"[14]"},{"why":"It is the Three Gap Theorem, which converts Theorem 1.2 into the corollary that sequences of the form $(\\{n\\alpha\\})$ fail Poissonian pair correlations.","marker":"[13]"}],"fun_headline_variants":["Poissonian sequences tolerate very few distinct gaps","Gap counts can grow arbitrarily slowly for random-like sequences","Slow gap growth still gives Poissonian pair correlations","Few distinct gaps do not ruin Poissonian pair correlations"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof needs the random blocks' pair counts to sit very close to their averages for almost every outcome; the written variance estimate treats every pair of indicators as independent even when the pairs share an index, and without that concentration the almost-sure Poissonian step collapses.","fun_headline_variants_meta":{"raw":{"variants":["Poissonian sequences tolerate very few distinct gaps","Gap counts can grow arbitrarily slowly for random-like sequences","Slow gap growth still gives Poissonian pair correlations","Few distinct gaps do not ruin Poissonian pair correlations"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000579,"raw_usage":{"total_tokens":2731,"prompt_tokens":951,"completion_tokens":1780,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":567,"completion_tokens_details":{"reasoning_tokens":1718}},"tokens_in":567,"tokens_out":1780,"duration_ms":12569,"temperature":1.0,"reasoning_tokens":1718,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T12:52:27.765648+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A direct refutation of Theorem 1.2 would be a Poissonian sequence with a subsequence on which some single gap length occurs at least $cn$ times for a fixed $c>0$; no such sequence can exist if the theorem is right. For the construction, computing the true covariance between pair indicators that share an index would decide whether the claimed concentration genuinely holds.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"It supplies the measure-theoretic independence corollary used in the variance estimate for the random blocks."},{"cited_title":"Poissonian Pair Correlation in Higher Dimensions","cited_arxiv_id":"1812.10458","evidence_quote":"It is another reference for almost-all sequences having Poissonian pair correlations, used as the starting point for the random blocks."},{"cited_title":"Marklof and A","cited_arxiv_id":null,"evidence_quote":"It is the Three Gap Theorem, which converts Theorem 1.2 into the corollary that sequences of the form $(\\{n\\alpha\\})$ fail Poissonian pair correlations."}],"review_version":1}