{"id":"3c28cddd-f760-4dc5-8f4e-5f4038c15bc6","arxiv_id":"2506.17753","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"In hyperbolic space H^n (n≥3) with a cocompact lattice, the averaged error in lattice counting diverges, and under two conjectures the local average over the quotient is O(X^{n-2+ε}).","lead":"For hyperbolic spaces in dimensions 3 and up, this paper measures how the error in counting lattice points inside a ball behaves when averaged. It proves the averaged error is unbounded in all large dimensions, and gives a conditional near-optimal bound for the local average under two conjectures from spectral theory.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Proof of Theorem 1.7 aligns only e^{it_jR} to 1, not the full coefficient phase; for n=3 this makes the claimed t^{-3} lower bound an order of magnitude too strong.","rationale":"The reader's verdict CONDITIONAL captures a factor error in (4.3) and an ambiguous Dirichlet application, but the load-bearing defect is deeper: Lemma 4.1 aligns the exponential phases to 1, while the displayed lower bound silently replaces the full complex coefficient by its absolute value. For n=3 this is not a harmless factor: Re(C(t)) is O(t^{-4}) while |C(t)| is t^{-3}, so the asserted lower bound fails by one power of t in the weight, and the n=3 partial sum becomes O(1) instead of log(1/ε). For n≡3 mod 4 with n≥7 the same mechanism degrades the claimed (n-3)/(2n) exponent. The conditional local average theorem (Theorem 1.5) appears internally consistent, and the factor typo in (4.3) is repairable, but the phase-alignment gap affects the central unconditional claim. Since the result may be true with a more sophisticated Kronecker argument that the paper does not supply, the appropriate verdict is UNVERDICTED rather than outright rejection. No machine-checked proof or independent verification of the Omega result is provided, so this concern is not offset by external evidence.","tokens_in":16896,"tokens_out":51927,"duration_ms":550314,"concrete_test":"Recompute (4.4) for n=3 from (4.3) and (2.5) using the coefficient C(t)=1/[it(1+it)^2], under the only condition Lemma 4.1 gives, |e^{itR}-1|<1/N. Show that the main term is O(sum |phi_j|^2 t_j^{-4}) rather than >> sum |phi_j|^2 t_j^{-3}. Then independently check whether the paper contains any lemma that aligns t_jR to the phase -arg C(t_j) with R ≤ N^{c A^n}; if not, the lower bound in (4.4) is unsupported and Theorem 1.7 is not proved as written.","verdict_should_be":"UNVERDICTED","load_bearing_attack":"The unconditional Omega-result rests on the step from (4.3) to (4.4). After correcting the displayed factor in (4.3) to 1/(1+it) — as forced by (2.5) and by the later weight (1+t)^{-(n+3)/2} — the coefficient is C(t)=Gamma(it)/Gamma((n+1)/2+it) * 1/(1+it). Lemma 4.1 only supplies R with |e^{it_jR}-1|<1/N. For n=3, C(t)=1/[it(1+it)^2] ~ i/t^3, so Re(C(t)e^{itR}) with e^{itR}≈1 equals Re(C(t)) ~ -2/(1+t^2)^2 = O(t^{-4}), not the t^{-3} used in (4.4). The claimed lower bound sum_{t_j≤τ/ε}|phi_j|^2 t_j^{-3} >> log(1/ε) therefore does not follow; with the true real part the n=3 sum is O(1). The same phase defect affects n≡3 mod 4 with n≥7, where Re(C(t)) is one power of t smaller than |C(t)|, changing the exponent in part (b). A repair would require a Kronecker/Dirichlet statement aligning t_jR to the slowly varying phase of C(t_j), with a comparable bound on R; no such statement is stated or proved.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the error term in the hyperbolic lattice point counting problem on compact quotients Γ\\H^n for n ≥ 3. Its first main result, Theorem 1.5, is conditional: assuming a weakened quantum variance conjecture (Conjecture 1.1 with exponent q_n = n−2+ε) and a conjectural spectral exponential sum bound (Conjecture 1.2), it proves that the local average of EΓ(X;z,z) against a smooth compactly supported test function is O_f(X^{n−2+ε}). The second main result, Theorem 1.7, claims unconditional Ω-results for the normalized mean error e(T,z): Ω(log log log T) for n = 3 and Ω((log log T)^{1/2−3/(2n)−ε}) for n ≥ 4; Corollary 1.1 transfers these to the second moment of the normalized error term via Cauchy–Schwarz. The proofs use smooth approximations to the characteristic kernel, the pre-trace formula, dyadic summation, partial summation, and a Dirichlet box principle.","tokens_in":17203,"tokens_out":31474,"duration_ms":313835,"significance":"If valid, the local average result is a natural higher-dimensional analogue of the Petridis–Risager and Cherubini–Katsivelos theorems, and it clearly identifies the conjectural inputs needed for the bound: a quantum variance estimate and a spectral exponential sum estimate. The Ω-results would constitute a new contribution to the fluctuation theory of hyperbolic lattice point errors, complementing the earlier work of Phillips–Rudnick and Chatzakos. The paper is transparent about its conditional assumptions, contains no fitted free parameters, and the local-average section is a coherent dyadic-summation argument under the stated conjectures. However, as discussed below, the proof of the Ω-results has a load-bearing gap for dimensions n ≡ 3 mod 4, including the prominent n = 3 case.","major_comments":[{"comment":"There is a coefficient error in the mollified pre-trace expansion. The antiderivative of x^{it_j} over the interval [√T,T] produces a factor 1/(1+it_j), not (1+it_j); this is also forced by the later tail bound in (4.3), which uses the decay (1+|t|)^{-(n+3)/2}. After correcting this to C(t) = Γ(it)/Γ((n+1)/2+it) · (1+it)^{-1}, the step from (4.4) to the displayed lower bound is invalid for n ≡ 3 mod 4. For n = 3, C(t) = 1/[it(1+it)^2] ≈ i/t^3, so Re(C(t)) ≈ −2/(1+t^2)^2 = O(t^{-4}), which is one full power of t smaller than |C(t)|. Lemma 4.1 only guarantees e^{it_jR} ≈ 1, hence Re(C(t_j)e^{it_jR}) ≈ Re(C(t_j)); the sum Σ_{t_j≤τ/ε}|φ_j(z)|^2 t_j^{-4} is O(1) by the local Weyl law (spectral density ∼ t^2), not ≫ log(1/ε). The same obstruction occurs for every n ≡ 3 mod 4. A repair would require simultaneous alignment of t_jR to the slowly varying phase of C(t_j) (for n = 3, to −i), which is not a consequence of the stated Dirichlet lemma and may fail for spectra with Q-linear relations. Thus Theorem 1.7(a), the n ≡ 3 mod 4 portion of Theorem 1.7(b), and the corresponding cases of Corollary 1.1 are not established.","section":"§4, Eqs. (4.3)–(4.4)"},{"comment":"The text states that Lemma 4.1 gives 'R ≤ N A^n' for t_j ≤ A, but the lemma as stated gives R ≤ M N^m with m ≍ A^n, i.e. R ≤ M N^{cA^n}. The balance in (4.5), which contains the exponential expression ε^{−2kn/(2k−(n−3))}, is consistent with the exponential bound, not with a polynomial bound. As printed, the bound R ≤ N A^n would give logR ≍ log(1/ε) and would change the final exponents. This appears to be a missing superscript, but it must be corrected because the transfer from the mollified lower bound to the Ω-statement for e(T,z) depends on the relation between R and ε.","section":"§4, after Eq. (4.4)"},{"comment":"Even for dimensions n ≥ 4 with n not congruent to 3 mod 4, where Re(C(t)) has the same order of magnitude as |C(t)|, the sign of Re(C(t)) is negative for some residue classes of n. The argument currently asserts a positive lower bound of the form e* ≫ Σ |φ_j|^2 (1+|t_j|)^{−(n+3)/2}; when Re(C(t)) is negative, the correct conclusion is |e*| ≫ Σ |φ_j|^2 (1+|t_j|)^{−(n+3)/2}. The absolute value is sufficient for the Ω-statement, but the sign issue should be addressed explicitly so that the extraction of the lower bound is legitimate.","section":"§4, Eq. (4.4)"}],"minor_comments":[{"comment":"The statement writes Γ/H^n; this should be Γ\\H^n.","section":"§3.4, Lemma 3.3"},{"comment":"The displayed factor (1+it_j) should be 1/(1+it_j); the surrounding discussion and tail bound make this clear, but the equation itself is currently wrong.","section":"§4, Eq. (4.3)"},{"comment":"The lower limit of the inner x-integral is written as e^Y/2 in one place; it should be e^{Y/2} = √T. This should be corrected for consistency with the definition of e(T,z).","section":"§4, preceding Eq. (4.3)"},{"comment":"The typesetting of (4.5) is garbled; as printed it reads 'logR≪nϵ− 2kn 2k−(n−3) log(ϵ−1)' and should be written as a clear inequality of the form log R ≪ ε^{−2kn/(2k−(n−3))} log(1/ε).","section":"§4, Eq. (4.5)"},{"comment":"Conjecture 1.1 states q_n = 1+ε for every n ≥ 2, but Theorem 1.5 uses the weaker exponent q_n = n−2+ε. This is fine, but the statement could explicitly say that the full conjecture is not needed.","section":"§1.1, Conjecture 1.1"}],"recommendation":"major_revision","confidential_remarks":"The phase issue in §4 is the main concern: it affects the flagship n = 3 result and the residue class n ≡ 3 mod 4 in higher dimensions. If the author cannot supply a Kronecker/Dirichlet statement aligning the t_jR phases to the coefficient phase, the affected Ω-results should be removed or weakened. The local-average theorem appears sound and could be published separately."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: the conditional local average (Theorem 1.5) is a real, correct result, but the advertised Ω-results (Theorem 1.7 and Corollary 1.1) have a load-bearing gap. As written, the proof does not establish the claimed lower bounds.\n\nThe issue is in the step from (4.3) to (4.4). After correcting the displayed factor to 1/(1+it_j) — forced by (2.5) and by the later weight (1+|tj|)^{-(n+3)/2} — the coefficient is C(t)=Γ(it)/Γ((n+1)/2+it) · 1/(1+it). Lemma 4.1 only gives R with |e^{it_jR}-1|<1/N, i.e. it aligns the exponential to 1. Then the main sum becomes Σ |φ_j|^2 \\hatψ_ε(t_j) Re(C(t_j)). For n=3, C(t)=1/[it(1+it)^2], and a direct computation gives Re(C(t)) = -2/(1+t^2)^2 = O(t^{-4}), not the t^{-3} used in (4.4). With Weyl density t^2, the sum over t_j≤τ/ε is absolutely convergent, so the claimed log(1/ε) lower bound does not follow. The phase cancellation also affects n≡3 mod 4 in part (b), shifting the exponent. The fix would require aligning t_jR to the slowly varying phase of C(t_j) via an inhomogeneous Kronecker/Dirichlet statement with a comparable bound on R; none is stated.\n\nThat said, the conditional Theorem 1.5 is solid. The dyadic summation and partial summation in Section 3, modulo minor typos, work under the stated conjectures. The paper is clearly written, engages the literature honestly, and the local-average argument is a plausible route to the conjectured exponent. There are smaller issues: (4.3) has a factor typo, and the box-principle bound should be R≤N^{cA^n} (since m≍A^n) rather than R≤NA^n; these are minor and repairable.\n\nBottom line: the paper's headline unconditional results are not established. It still deserves a serious referee, because the conditional result is valuable and the gap, while real, may be repairable with a different diophantine argument. An editor should send it out, but flag the phase issue explicitly.","headline":"The conditional local average is fine; the advertised Ω-results in Theorem 1.7 do not follow from the proof as written because the Dirichlet step aligns t_jR to 1 while the coefficient's real part is one power of t smaller.","tokens_in":17768,"tokens_out":15550,"would_cite":false,"duration_ms":141541,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["11F72","37C35","37D40"],"pacs":[],"model":"deepseek-v4-flash","headline":"In hyperbolic spaces of dimension $n\\geq 3$, the paper proves that normalized lattice-counting errors oscillate at least $\\log\\log\\log T$ for $n=3$, and conditionally that local averages of the error are as small as $O(X^{n-2+\\epsilon})$.","keywords":["hyperbolic lattice point problem","error term","local average","quantum variance","spectral exponential sums","Omega results","cocompact lattices","pre-trace formula"],"falsifier":"Numerically test Conjecture 1.2 by computing $S(T,X)=\\sum_{|t_j|\\leq T} X^{it_j}$ for a cocompact arithmetic quotient in dimension 3 at scales up to about $10^5$; if some sequence has $|S(T,X)|\\gg X^\\epsilon T^{3-\\epsilon}$, the hypothesis behind Theorem 1.5 is untenable. For the $\\Omega$-result, proving an upper bound $e(T,z)=o(\\log\\log\\log T)$ for any single cocompact 3-fold would contradict Theorem 1.7(a).","tokens_in":16683,"feed_emoji":"📐","tokens_out":16598,"duration_ms":150631,"temperature":0.7,"pith_summary":"The paper studies the error term in the hyperbolic lattice-counting problem: for a cocompact group $\\Gamma$ acting on $n$-dimensional hyperbolic space $\\mathbb{H}^n$, one counts the $\\Gamma$-translates of a point inside a ball of radius $r\\sim \\log X$, and $E_\\Gamma(X;z,w)$ is the discrepancy between that count and its smooth main term. The first result is conditional: if a weak quantum-variance estimate with exponent $n-2+\\epsilon$ and a spectral exponential-sum bound $S(T,X)\\ll X^\\epsilon T^{n-1+\\epsilon}$ both hold, then the average of $E_\\Gamma(X;z,z)$ against any smooth compactly supported test function is $O_f(X^{n-2+\\epsilon})$. The second result is unconditional: the normalized mean error $e(T,z)$ is $\\Omega(\\log\\log\\log T)$ for $n=3$ and $\\Omega((\\log\\log T)^{(n-3)/(2n)-\\epsilon})$ for $n\\geq 4$, and these lower bounds transfer to the second moment. The appeal is that for $n=3$ the conditional local-average exponent is essentially as strong as one could hope, while the $\\Omega$-results show the $X$-variable average fluctuates far more than the radial-variable average.","feed_headline":"Dimension-3 hyperbolic counting errors climb log-log-log T","feed_subtitle":"The n=3 result is unconditional; in every dimension, two spectral conjectures would give near-optimal local averages.","key_machinery":"The load-bearing mechanism is the spectral expansion supplied by the pre-trace formula. The counting error is written as $\\sum_{t_j\\neq 0} h_X(t_j)|\\varphi_j(z)|^2$, where $h_X$ is the Selberg/Harish-Chandra transform of a smoothed indicator of a hyperbolic ball; its approximate size is $X^{(n-1)/2}|t|^{-(n+1)/2}$ times an oscillating factor $X^{it}$, up to angular constants. For the conditional local-average theorem, the two spectral inputs are a weak quantum-variance bound on the deviations of the measures $|\\varphi_j(z)|^2\\,d\\mu$ from the volume measure and a bound for the spectral exponential sum $S(T,X)=\\sum_{|t_j|\\leq T}X^{it_j}$; these control respectively the test-function fluctuation term and the constant term in the pre-trace expansion. For the unconditional $\\Omega$-results, the same transform, together with the local Weyl law and Dirichlet's box principle, is used to force the phases $X^{it_j}$ to align at a sequence of scales, producing a lower bound for the mollified mean error.","core_discovery":"On the paper's own terms, the central discovery is that in every dimension $n\\geq 3$ the large-eigenvalue part of the hyperbolic lattice-counting error satisfies two definitive statements. Unconditionally, for every cocompact lattice $\\Gamma\\subset \\mathrm{SO}^+(1,n)$ and every base point $z$, the modified mean $e(T,z)=T^{-1}\\int_{\\sqrt{T}}^{T}E_\\Gamma(x;z,z)x^{-(n-1)/2}\\,dx$ obeys $e(T,z)=\\Omega(\\log\\log\\log T)$ for $n=3$ and $e(T,z)=\\Omega((\\log\\log T)^{(n-3)/(2n)-\\epsilon})$ for $n\\geq 4$; Cauchy-Schwarz then transfers these lower bounds to the second moment of the normalized error. Conditionally, assuming the weak quantum-variance bound for the measures $|\\varphi_j(z)|^2\\,d\\mu$ with exponent $q_n=n-2+\\epsilon$ and the spectral exponential-sum bound $S(T,X)\\ll X^\\epsilon T^{n-1+\\epsilon}$, the local average $\\int f(z)E_\\Gamma(X;z,z)\\,d\\mu(z)$ is $O_f(X^{n-2+\\epsilon})$, with no improvement available from assuming a stronger quantum-variance exponent. The paper thus extends a local-average strategy previously applied to surfaces to all cocompact quotients of hyperbolic space, and it separates the $X$-variable mean behaviour from the radial-variable mean behaviour.","pith_inferences":["The local-average proof uses only the spectral expansion and the two conjectures, so the same $X^{n-2+\\epsilon}$ bound should extend from cocompact $\\Gamma$ to finite-volume non-compact quotients once the Eisenstein-series contribution is controlled; the exponential-sum conjecture is stated for essentially cuspidal groups, but the theorem is proved only in the cocompact case.","Because the $\\Omega$-results use only the local Weyl law and phase alignment, they likely hold for every cocompact quotient, arithmetic or not; testing a non-arithmetic compact hyperbolic 3-manifold would check whether the $\\log\\log\\log T$ rate is universal.","If the box-principle balance in the $\\Omega$-proof could be replaced by a sharper almost-periodicity argument, the $n=3$ lower bound might improve to $\\Omega(\\log\\log T)$, the natural analogue of the radial-variable result; this is outside the paper but suggested by its method."],"forward_implications":["For $n=3$, the normalized mean error $e(T,z)$ is $\\Omega(\\log\\log\\log T)$, so it cannot converge as $T\\to\\infty$.","For $n\\geq 4$, $e(T,z)=\\Omega((\\log\\log T)^{(n-3)/(2n)-\\epsilon})$, so the normalized error grows slowly but unboundedly in every dimension.","By Cauchy-Schwarz, the same lower bounds transfer to the second moment of the normalized error: $\\Omega((\\log\\log\\log T)^2)$ for $n=3$ and $\\Omega((\\log\\log T)^{1-3/n-\\epsilon})$ for $n\\geq 4$.","If the two conjectures hold, the local average over any compact set is $O_f(X^{n-2+\\epsilon})$; for $n=3$ this is essentially the strongest scaling available, and stronger quantum-variance assumptions would not improve it.","The results separate the $X$-variable mean from the radial-parameter mean: the latter has a finite mean value by earlier work, while the former is forced to oscillate."],"supporting_citations":[{"why":"introduces the local-average method by expanding the smoothed error spectrally and controlling it with a quantum-variance estimate","marker":"[43]"},{"why":"provides the main-term decomposition, the radial mean-value analysis, and the Dirichlet box-principle lemma used in the $\\Omega$-proof","marker":"[44]"},{"why":"studies the modified $X$-variable mean, proves prior $\\Omega$-results and the failure of the limit, which Theorem 1.7 extends to $n\\geq 3$","marker":"[11]"},{"why":"gives the analogous conditional local-average result on the three-dimensional Picard manifold under quantum-variance and exponential-sum hypotheses","marker":"[14]"},{"why":"establishes the spectral exponential-sum framework and conjecture for the Picard manifold that Conjecture 1.2 generalizes","marker":"[29]"},{"why":"proves the quantum-variance estimate on the modular surface that motivates the weak quantum-variance form of Conjecture 1.1","marker":"[38]"},{"why":"computes the variance of the hyperbolic counting function and supplies the hypergeometric formulas used for the small eigenvalues","marker":"[25]"},{"why":"supplies the large-sieve arguments and transform lemmas used for the second moment and kernel smoothing","marker":"[10]"}],"fun_headline_variants":["Hyperbolic lattice counting: Omega results for all n≥3","Error term in hyperbolic circle problem: Omega(log log log T) in n=3","Conditional near-optimal local averages for hyperbolic counting in large dimensions","Unconditional Omega and conditional near-optimal bounds for hyperbolic counting","Large-n hyperbolic counting: separating X and radial means"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The local-average theorem stands on two unproved spectral conjectures, a weak quantum-variance bound with exponent $n-2+\\epsilon$ and a spectral exponential-sum bound with exponent $n-1+\\epsilon$; the $\\Omega$-theorem stands on a Dirichlet box-principle step that must align roughly $A^n$ oscillating phases while keeping the averaging parameter within a range that is only singly exponential in the smoothing scale.","fun_headline_variants_meta":{"raw":{"variants":["Hyperbolic lattice counting: Omega results for all n≥3","Error term in hyperbolic circle problem: Omega(log log log T) in n=3","Conditional near-optimal local averages for hyperbolic counting in large dimensions","Unconditional Omega and conditional near-optimal bounds for hyperbolic counting","Large-n hyperbolic counting: separating X and radial means"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001245,"raw_usage":{"total_tokens":5101,"prompt_tokens":936,"completion_tokens":4165,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":552,"completion_tokens_details":{"reasoning_tokens":4075}},"tokens_in":552,"tokens_out":4165,"duration_ms":33645,"temperature":1.0,"reasoning_tokens":4075,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T19:03:53.519259+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Numerically test Conjecture 1.2 by computing $S(T,X)=\\sum_{|t_j|\\leq T} X^{it_j}$ for a cocompact arithmetic quotient in dimension 3 at scales up to about $10^5$; if some sequence has $|S(T,X)|\\gg X^\\epsilon T^{3-\\epsilon}$, the hypothesis behind Theorem 1.5 is untenable. For the $\\Omega$-result, proving an upper bound $e(T,z)=o(\\log\\log\\log T)$ for any single cocompact 3-fold would contradict Theorem 1.7(a).","supporting_citations":[{"cited_title":"Petridis and M","cited_arxiv_id":null,"evidence_quote":"introduces the local-average method by expanding the smoothed error spectrally and controlling it with a quantum-variance estimate"},{"cited_title":"Phillips and Z","cited_arxiv_id":null,"evidence_quote":"provides the main-term decomposition, the radial mean-value analysis, and the Dirichlet box-principle lemma used in the $\\Omega$-proof"},{"cited_title":"Chatzakos, Ω -results for the hyperbolic lattice point problem","cited_arxiv_id":null,"evidence_quote":"studies the modified $X$-variable mean, proves prior $\\Omega$-results and the failure of the limit, which Theorem 1.7 extends to $n\\geq 3$"},{"cited_title":"Cherubini, C.Katsivelos, Local average in the Hyperbolic sphere problem","cited_arxiv_id":null,"evidence_quote":"gives the analogous conditional local-average result on the three-dimensional Picard manifold under quantum-variance and exponential-sum hypotheses"},{"cited_title":"Kaneko, The prime geodesic theorem for PSL2(Z[i]) and spectral exponential sums","cited_arxiv_id":null,"evidence_quote":"establishes the spectral exponential-sum framework and conjecture for the Picard manifold that Conjecture 1.2 generalizes"},{"cited_title":"Luo and P","cited_arxiv_id":null,"evidence_quote":"proves the quantum-variance estimate on the modular surface that motivates the weak quantum-variance form of Conjecture 1.1"},{"cited_title":"Hill and L","cited_arxiv_id":null,"evidence_quote":"computes the variance of the hyperbolic counting function and supplies the hypergeometric formulas used for the small eigenvalues"},{"cited_title":"Chamizo, Some applications of large sieve in Riemann surfaces","cited_arxiv_id":null,"evidence_quote":"supplies the large-sieve arguments and transform lemmas used for the second moment and kernel smoothing"}],"review_version":1}