{"id":"546f77bc-e5ef-46e5-8bed-b76d2abc5398","arxiv_id":"2501.10962","paper_version":2,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"For sign-valued completely multiplicative functions, the paper computes the full set of possible shifted-correlation limits for small sets of primes.","lead":"This paper gives elementary combinatorial proofs of two known results about shifted correlations of sign-valued multiplicative functions, and fully describes the set of possible correlation limits. The new part is a characterization of this \"spectrum\" of values.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Proof of Proposition 3 has an algebraic error in the α_H < 0 case: the constructed set P would need a negative κP, and the product does not equal β.","rationale":"The reader's verdict accepted the paper, identifying Lemma 6's terse density argument as the weakest assumption. That concern is real: Lemma 6 asserts that for any interval (a,b) one can find a finite set of large primes with product in (a,b), citing only divergence of reciprocal primes. This can be repaired by noting the factors 1 − 2d/(p+1) tend to 1, so for a fixed interval one can choose primes large enough that the product cannot jump over the interval, and then use single primes to fine-tune near 1. But the more serious issue is the algebraic error in Proposition 3's α_H < 0 case, which the reader did not flag. As written, that step requires a negative κH_P for a set of non-exceptional primes and fails to produce the target β in general. The result itself is likely correct — the fix is to choose κH_P = β/α_H rather than 1 − 2β/α_H — but the proof in the manuscript does not establish it. Since the spectrum characterization is a central contribution, the paper should not be accepted without correcting this argument. Theorems 1 and 2 appear unaffected, and the remaining gaps are fillable, so a conditional acceptance is appropriate.","tokens_in":12550,"tokens_out":9942,"duration_ms":98307,"concrete_test":"Take H = {0,1}, so d = 2, all primes are non-exceptional, α_H = −1/3 (attained at p = 2), and choose β = −0.2. The printed recipe gives α = β/α_H = 0.6 and demands a set P of primes > X with κH_P = 1 − 2α = −0.2. Since every prime p > X has κH_p = 1 − 4/(p+1) > 0, no such P exists. The corrected construction instead takes P with κH_P = β/α_H = 0.6, which exists by Lemma 6, and then P ∪ {2} has κ = 0.6 × (−1/3) = −0.2 = β. This arithmetic check isolates the error and confirms the repair.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"In Section 5, proof of Proposition 3, the α_H < 0 case reads: 'Choose α := β/α_H, and choose P a small set of non-exceptional primes, each larger than X, such that κH_P = 1 − 2α. Let p be such that α_H = 1 − 2ηH_p. Then from Theorem 1, the required small set of primes is P ∪ {p}.' Since β ∈ (α_H, 0) and α_H < 0, we have α = β/α_H ∈ (0,1). The intended product is κH_{P∪{p}} = κH_P · α_H = (1 − 2α)α_H = α_H − 2β, which equals β only if α_H = 3β — not generally true. Moreover, for α > 1/2 the quantity 1 − 2α is negative, but κH_P is a product of positive factors (1 − 2d/(p+1)) for primes p > X > 2d, so it cannot be negative. Thus the construction fails as written. The correct construction would require κH_P = β/α_H (or equivalently α := (1 − β/α_H)/2), which lies in (0,1) and is attainable by Lemma 6. This is load-bearing because Proposition 3, the spectrum characterization, depends on it for the interval [α_H, 0) when α_H < 0.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies shifted convolution sums of completely multiplicative functions λ_P taking values in {±1}, where P is a set of primes. For a finite shift set H, it defines κH_P as the limiting average of ΛH_P(n) = λ_P(n+h1)...λ_P(n+hd). Theorem 1 asserts that for small P (convergent sum of reciprocals), κH_P equals an Euler product ∏_{p∈P}(1-2ηH_p), with ηH_p = d/(p+1) for non-exceptional primes. Theorem 2 states that no non-constant λ_P has a correlation with absolute value 1, following from a result of Matomäki and Radziwill. Proposition 3 characterizes the spectrum Γ_H, the closure of all attainable κH_P values, as [α_H,1] ∪ [0,1], where α_H = inf_p(1-2ηH_p). The proofs use a symmetric-difference algebra for the sets N_H^P where ΛH_P = -1, and a density approximation argument.","tokens_in":12825,"tokens_out":7314,"duration_ms":75512,"significance":"The paper gives elementary, self-contained proofs of known correlation formulas for λ_P and obtains a complete description of the spectrum of possible correlations for small prime sets. The symmetric-difference viewpoint is elegant and provides a concrete way to compute ηH_p (as in Example 1). If the spectrum theorem is correct, it is a new structural result for this family of multiplicative functions. The main theorems are stated clearly, and the dependence on external deep input (Matomäki-Radziwill) is explicit and limited, which is a strength. No code or machine-checked proofs are provided, but the arguments are conceptually transparent and likely repairable.","major_comments":[{"comment":"The construction of P ∪ {p} is algebraically wrong. With α := β/α_H, choosing P such that κH_P = 1-2α gives κH_{P∪{p}} = (1-2α)α_H = α_H - 2β, which equals β only in the special case α_H = 3β. Moreover, if α > 1/2, then 1-2α is negative, impossible for primes p > X (where the factors are positive). The correct choice is a small set P with κH_P = β/α_H, which lies in (0,1) and is attainable by Lemma 6; then κH_{P∪{p}} = (β/α_H)·α_H = β. This error is load-bearing for the inclusion [α_H,1] ⊆ Γ_H and must be fixed.","section":"Section 5, proof of Proposition 3 (α_H < 0 case)"},{"comment":"The double induction is not fully specified. When the projection modulo p is constant, the proof replaces H by H1 = (H - i1)/p, which has max H1 < max H but the same cardinality as H. The later statement 'we may then apply induction on |H|' is unjustified, since |H_r| = |H| at the terminal stage. A well-founded induction on a measure such as (|H|, max H) would repair the argument, but as written the induction is incomplete and this is used to establish Theorem 1.","section":"Section 3, Proposition 2 (induction structure)"},{"comment":"The assertion 'Such a choice is possible because the sum of reciprocals of primes is divergent' is not proved, yet it is the core of the lemma's construction. It is true (the finite subset products of (1-2d/(p+1)) are dense in (0,1) because ∑ 2d/(p+1) diverges and the summands tend to 0), but the one-sentence justification is too terse for a load-bearing step in Proposition 3. A short proof or reference would make the lemma self-contained.","section":"Section 5, Lemma 6"}],"minor_comments":[{"comment":"'Combinatorical' should be 'combinatorial'.","section":"Abstract and Introduction"},{"comment":"The bound δ+(N_H^{P\\P_i}) ≤ ∑_{p∈P\\P_i} ηH_p follows from Lemma 2 and (3.5), but this is not stated; please add a sentence making the use of Lemma 2 explicit.","section":"Section 3, equation (3.9)"},{"comment":"The chain 'κH_P ≥ κH_{P\\{p}}α_H ≥ α_H' is misleading: the first inequality requires a case split on the sign of κH_{P\\{p}}, and the second uses κH_{P\\{p}} ≤ 1. The conclusion is true because every factor lies in [α_H,1] and products of such factors are ≥ α_H, but the written chain should be clarified.","section":"Section 5, proof of Proposition 3, final paragraph"},{"comment":"The empty set is denoted φ, which may be confused with the totient function; consider using ∅.","section":"Notation"}],"recommendation":"major_revision","confidential_remarks":"The main theorems are likely correct, but the spectrum proof (Proposition 3) contains a genuine algebraic error in the α_H < 0 case, and the induction in Proposition 2 is formally incomplete. Both are repairable within the scope of the paper. The paper would benefit from a careful revision of these two proofs and a fuller justification of Lemma 6."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"First thing to know: this paper re-proves two known results, Klurman's Theorem 1 and Teräväinen's Theorem 2, with a mostly elementary combinatorial method, and adds a new spectrum result (Proposition 3). The elementary machinery is the real value: the symmetric-difference algebra and the density approximation argument are clean and worth having. The spectrum statement is natural and probably true, but the proof as written has a genuine gap in the negative-α_H case.\n\nThe good parts: Theorem 1's proof is self-contained and sound. The limiting argument for small sets via the tail bound (3.9) works. Theorem 2's reduction to the Matomäki-Radziwill theorem is correct, and the lemma borrowed from Wildon is a neat way to get a two-point set. The paper is honest about what is new: both theorems are explicitly attributed to earlier work, and Proposition 3 is the only claimed new result.\n\nThe soft spots, in proportion. In Section 5, proof of Proposition 3, the α_H < 0 case: the text sets α := β/α_H (which is in (0,1)) and then chooses P with κ_P^H = 1 − 2α. That does not follow from Lemma 6, which only provides positive κ values, and 1 − 2α is negative whenever β < α_H/2. Also, for primes larger than X, the factors 1 − 2η_p^H are positive, so κ_P^H cannot be negative at all. The fix is immediate: choose P with κ_P^H = β/α_H (which does lie in (0,1)) and then multiply by the α_H factor from p. As written, the proof does not cover part of the interval, so this is load-bearing for the new result, but it is a small repair, not a deep flaw.\n\nLemma 6 is also terse: the assertion that you can hit any target in (0,1) using disjoint finite sets of large primes because the reciprocal-prime sum diverges needs more than one sentence. The construction is plausible and standard, but the write-up should expand it.\n\nThe reverse inclusion (κ_P ≥ α_H) is fine, and there is no circularity or citation weirdness.\n\nWho this is for: analytic number theorists working on Chowla-type correlations. It deserves a serious referee; it should not be desk rejected. The referee should require the Proposition 3 correction and an expanded Lemma 6. After that, it's a solid short paper.","headline":"Nice elementary proofs of two known Chowla-variant theorems, plus a new spectrum result whose proof has a fixable gap in the negative-α_H case.","tokens_in":13375,"tokens_out":7155,"would_cite":true,"duration_ms":72749,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["11N37","11P32","11N35","11T06"],"pacs":[],"model":"deepseek-v4-flash","headline":"For small prime sets, shifted correlation averages of ±1-valued completely multiplicative functions equal an Euler product of local factors, and the attainable values form a complete interval.","keywords":["Chowla's conjecture","Liouville function","completely multiplicative functions","shifted convolution sums","spectrum","small sets of primes","natural density","two-point correlations"],"falsifier":"Compute $S(x)=(1/x)\\sum_{n\\le x}\\lambda_{\\{2\\}}(n)\\lambda_{\\{2\\}}(n+4)\\lambda_{\\{2\\}}(n+6)$, where $\\lambda_{\\{2\\}}(p)=-1$ only at $p=2$ and is $1$ at all other primes. The paper's formula predicts the limit is $1-2\\cdot(1/6)=2/3$; a persistent deviation as $x$ grows would refute Theorem 1. Alternatively, exhibiting a value in $(0,1)$ that no small set $P$ realizes would refute Proposition 3.","tokens_in":12331,"feed_emoji":"🔢","tokens_out":13994,"duration_ms":128953,"temperature":0.7,"pith_summary":"This paper studies averages of shifted products $\\lambda_P(n+h_1)\\cdots\\lambda_P(n+h_d)$, where $\\lambda_P$ is the completely multiplicative function equal to $-1$ at primes in $P$ and $1$ elsewhere. It proves that when $P$ is a small set of primes, the long-run average has an exact Euler-product formula whose factors are local densities, one per prime. That makes the global correlation a product of independent local contributions even though the shifted product is not multiplicative. It also proves that only the constant function can achieve a correlation of magnitude $1$, and that the set of all possible correlation values for a fixed shift set is a full interval union. These are variants of Chowla's conjecture, which predicts such averages vanish for the Liouville function.","feed_headline":"Correlation averages become exact Euler products on small prime sets","feed_subtitle":"A combinatorial argument pins down every possible correlation value for a fixed set of shifts.","key_machinery":"The carrying object is the set $N_H^P=\\{n:\\Lambda_H^P(n)=-1\\}$, whose natural density $\\eta_H^P$ is the quantity being computed. Lemma 2 gives $N_{H_1\\triangle H_2}^{P_1\\triangle P_2}=N_{H_1}^{P_1}\\triangle N_{H_1}^{P_2}\\triangle N_{H_2}^{P_1}\\triangle N_{H_2}^{P_2}$, so primes and shifts can be added one at a time. A double induction approximates $N_H^P$ from inside and outside by finite unions of arithmetic progressions with moduli made only from primes in $P$, yielding the density recursion $\\eta_H^P=\\eta_H^{P'}(1-\\eta_p^H)+\\eta_p^H(1-\\eta_H^{P'})$, which factors into $1-2\\eta_H^P=\\prod_{p\\in P}(1-2\\eta_p^H)$. Inequality (3.9) then controls the passage from finite to small infinite $P$ by the tail sum over primes outside a finite subset. For Theorem 2, the machinery is the symmetric-difference closure of the family of shift sets with $|\\kappa_H^P|=1$, combined with a cited two-point correlation bound that rules out values of magnitude $1$.","core_discovery":"The central claim is Theorem 1: for any finite shift set $H$ and any small prime set $P$, the limit $\\kappa_H^P=\\lim_{x\\to\\infty}\\frac{1}{x}\\sum_{n\\le x}\\prod_{h\\in H}\\lambda_P(n+h)$ exists and equals $\\prod_{p\\in P}(1-2\\eta_p^H)$, where $\\eta_p^H=d/(p+1)$ whenever $p$ divides none of the differences $h_i-h_j$, and exceptional primes have explicitly computable constants. The proof works through the density of the set $N_H^P=\\{n:\\Lambda_H^P(n)=-1\\}$, which is shown to be approximable by finite unions of arithmetic progressions whose moduli use only primes from $P$. Theorem 2 then says that if some non-empty $H$ has $|\\kappa_H^P|=1$, then $P$ is empty, so non-constant functions never produce perfect correlation; the proof reduces this to the two-point case via a symmetric-difference closure property. Finally, the spectrum $\\Gamma_H$, the closure of all values $\\kappa_H^P$ over small $P$, is exactly $[\\alpha_H,1]\\cup[0,1]$.","pith_inferences":["The same one-prime-at-a-time recursion should carry over to correlations of $\\pm1$-valued functions twisted by Dirichlet characters, with $\\eta_p^H$ replaced by counts of roots modulo prime powers; the paper does not pursue this extension.","Lemma 6's use of the divergence of reciprocals suggests a stronger quantitative statement: the finite-product approximants can be chosen so that the error at stage $i$ decays like a tail of the prime harmonic series, which would give explicit convergence rates for the spectrum construction.","Theorem 2 would become fully elementary if the two-point correlation bound it invokes could be replaced by a direct combinatorial argument; Remark 2 already points out that the limsup/liminf version would require a stronger form of Lemma 4, so the dependence is genuine."],"forward_implications":["For any small prime set $P$ and finite shift set $H$, the average $\\kappa_H^P$ exists and equals $\\prod_{p\\in P}(1-2\\eta_p^H)$, so the average is fixed exactly by per-prime data.","For every non-exceptional prime, $\\eta_p^H=d/(p+1)$; therefore the convergence of the product is the same as $P$ being small, and the product vanishes exactly when the sum of reciprocals diverges.","No non-constant $\\pm1$-valued completely multiplicative function can have a finite-shift correlation of magnitude $1$; every such average satisfies $\\liminf<1$ and $\\limsup>-1$.","For a fixed $H$, the closure of all values $\\kappa_H^P$ over small $P$ is $[\\alpha_H,1]\\cup[0,1]$, so every value in that range is realized by some small set of primes.","The recursive procedure also computes exceptional factors explicitly, for example $\\eta^{\\{0,4,6\\}}_2=1/6$ and $\\eta^{\\{0,4,6\\}}_3=5/12$."],"supporting_citations":[{"why":"It is the conjecture whose variants are studied, and it defines the shifted correlation averages analyzed here.","marker":"[3]"},{"why":"It supplies the combinatorial lemma that any family of finite sets closed under symmetric difference and translation contains a two-element set, which reduces Theorem 2 to the two-point case.","marker":"[8]"},{"why":"It gives the earlier correlation result from which Theorem 1 can be deduced, and the paper compares its local constants with roots of the associated polynomial.","marker":"[10]"},{"why":"It provides the two-point short-interval bound for $\\pm1$-valued multiplicative functions that rules out correlation values of magnitude $1$ in Theorem 2.","marker":"[11]"}],"fun_headline_variants":["Correlation limits equal exact Euler products on small primes","Combinatorial proof determines all Chowla correlation values","Non-constant ±1 functions never give perfect correlation","Small primes fully control shifted correlation limits"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"Lemma 6 assumes that for any interval $(a,b)$ inside $(0,1)$ one can find finitely many new large primes whose per-prime factors multiply into $(a,b)$, and the only justification given is the divergence of the sum of reciprocals of primes; if this density statement failed, the claimed spectrum could omit attainable values.","fun_headline_variants_meta":{"raw":{"variants":["Correlation limits equal exact Euler products on small primes","Combinatorial proof determines all Chowla correlation values","Non-constant ±1 functions never give perfect correlation","Small primes fully control shifted correlation limits"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000934,"raw_usage":{"total_tokens":3928,"prompt_tokens":809,"completion_tokens":3119,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":425,"completion_tokens_details":{"reasoning_tokens":3059}},"tokens_in":425,"tokens_out":3119,"duration_ms":24285,"temperature":1.0,"reasoning_tokens":3059,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T18:49:28.175793+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute $S(x)=(1/x)\\sum_{n\\le x}\\lambda_{\\{2\\}}(n)\\lambda_{\\{2\\}}(n+4)\\lambda_{\\{2\\}}(n+6)$, where $\\lambda_{\\{2\\}}(p)=-1$ only at $p=2$ and is $1$ at all other primes. The paper's formula predicts the limit is $1-2\\cdot(1/6)=2/3$; a persistent deviation as $x$ grows would refute Theorem 1. Alternatively, exhibiting a value in $(0,1)$ that no small set $P$ realizes would refute Proposition 3.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"It is the conjecture whose variants are studied, and it defines the shifted correlation averages analyzed here."},{"cited_title":"Collection closed under sym- metric difference and translation","cited_arxiv_id":null,"evidence_quote":"It supplies the combinatorial lemma that any family of finite sets closed under symmetric difference and translation contains a two-element set, which reduces Theorem 2 to the two-point case."},{"cited_title":"Correlations of multiplicative functions and applic ations","cited_arxiv_id":null,"evidence_quote":"It gives the earlier correlation result from which Theorem 1 can be deduced, and the paper compares its local constants with roots of the associated polynomial."},{"cited_title":"Multiplicative functionsin short intervals","cited_arxiv_id":null,"evidence_quote":"It provides the two-point short-interval bound for $\\pm1$-valued multiplicative functions that rules out correlation values of magnitude $1$ in Theorem 2."}],"review_version":1}