{"id":"67ccd21e-fe00-4787-a7e7-46fdc701d151","arxiv_id":"2502.03554","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"This paper rigorously proves that fluctuations of stationary Hastings-Levitov(0) have logarithmic spatial correlations and logarithmic maximal growth.","lead":"This paper proves that the random boundary fluctuations of a growing random cluster, the stationary Hastings-Levitov process, are correlated over long distances by a logarithm of the distance. It also proves a logarithmic upper bound on the largest upward fluctuation, advancing the rigorous theory of diffusion-limited aggregation type growth.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The lower bound in (14) is algebraically false: under the paper's own H4 event, ImF_s is approximately pi*s/2, so pi/(4*ImF_s) is approximately 1/(2s), not pi/(4s); Theorem 1's pi/4 constant is therefore unproven.","rationale":"The central claim summarized by the reader is the logarithmic-correlation structure with the specific constant pi/4. The paper's own definitions force ImF_s ~ pi*s/2 under the high-probability events H2 and H4. Lemma A.1 then gives an integrand pi/(4*ImF_s) ~ 1/(2s). Equation (14) replaces this by pi/(4s), which is a factor pi/2 larger; since the claimed lower bound is an equality statement, this algebraic error directly invalidates the proof of Theorem 1. The same mismatch appears in the upper-bound calculation (12), although there the erroneous bound is merely loose and does not break the upper direction. Because Theorem 2's covariance constant inherits the variance constant from Theorem 1, the stated quantitative form of the log-correlation result is also unsupported. The reader's identified weakest assumption, the unproved uniform event H4, is a real gap, but the constant error is more decisive: it attacks the numerical claim even under the optimistic assumption that H4 holds. I recommend REJECT for the current version: the main theorems as printed are not proven, and the displayed inequalities contain a concrete algebraic falsehood. The qualitative phenomenon of logarithmic correlations may still be true with constant 1/2, and the paper's martingale and generator ideas are potentially valuable, but a major revision is required before the results can be accepted.","tokens_in":10803,"tokens_out":24872,"duration_ms":217914,"concrete_test":"Recompute the lower bound in equation (14) using the H4 bounds: substitute ImF_s = pi*s/2 + s^(2/3) and verify that pi/(4*ImF_s) - c/(ImF_s)^3 is asymptotically 1/(2s) - O(s^(-4/3)), which is strictly less than pi/(4s) for all large s. This shows the displayed '>= pi/(4s)' step is invalid and that the proof as written yields only a 1/2 log t lower bound.","verdict_should_be":"REJECT","load_bearing_attack":"The proof of the lower bound in Theorem 1 (Section 2, equation (14)) is internally inconsistent. Under the manuscript's event H4, ImF_s(0) lies between pi*s/2 - s^(2/3) and pi*s/2 + s^(2/3). The integrand pi/(4*ImF_s(0)) from Lemma A.1 is then at least about 1/(2s) - O(s^(-4/3)) (the minimum occurs at the upper endpoint), not pi/(4s). The displayed inequality '>= pi/(4s)' in (14) is false; the correct lower bound gives E|M_t(0)|^2 >= (1/2) log t, not (pi/4) log t. The same substitution error appears in the upper-bound display (12), where pi/(4*ImF_s) is bounded by pi/(4s) although under H2 it is at most about 1/(2s). Thus the exact constant pi/4 in Theorem 1, and consequently the same constant in Theorem 2's covariance formula, is not established by the presented argument. This issue is independent of the unproved uniform event H4: even granting H4 with probability one, the inequality in (14) fails.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the stationary Hastings-Levitov(0) process in the upper half-plane and its fluctuation field M_t(x)=F_t(x)-x-iπt/2. It claims three theorems: Theorem 1 gives E|M_t(0)|^2=(π/4)log t(1+o(1)); Theorem 2 gives a logarithmic covariance formula; Theorem 3 gives a high-probability bound of order β log t for max_{x∈[0,t]} Im M_t(x), with failure probability O(t^{-1/2}). The proofs use the martingale decomposition from [BPT22], a bootstrap on Im F_s(0), a generator computation for exponential moments, and a conformal-geometry argument for the maximum.","tokens_in":11115,"tokens_out":20610,"duration_ms":188948,"significance":"If correct, the paper would provide a non-Gaussian, physically motivated example of a logarithmically correlated random field with precise variance growth and logarithmic maximal growth. The paper does not fit free parameters, and the martingale representation is anchored in the independently published construction of [BPT22]. However, the central variance constant in Theorem 1 is not established by the submitted proof because of an algebraic error in the lower bound, the uniform event H4 is asserted without proof, and Theorem 2 as printed is false in the regime b>t. The significance is therefore conditional on a substantial reworking of the main arguments.","major_comments":[{"comment":"The displayed lower bound is algebraically false. Under the event H4, Im F_s(0) ≤ πs/2 + s^{2/3}, so Lemma A.1 gives ∫ |φ_x(F_s(0))-F_s(0)|^2 dx ≥ π/(4 Im F_s(0)) - c/(Im F_s(0))^3 ≥ 1/(2s) - O(s^{-4/3}). Since 1/(2s) < π/(4s), the inequality '≥ π/(4 P[H4(η)] ∫_η^t ds/s' does not follow. At best this route yields E|M_t(0)|^2 ≥ (1/2 - o(1)) log t, which is strictly weaker than the claimed π/4 log t. The proof of Theorem 1's lower bound is therefore not valid.","section":"Section 2, Eq. (14)"},{"comment":"The uniform event H4(η) is used with the assertion P(H4(η)) ≥ 1 - η^{-1/3}, but no proof is given. The preceding bound P(H2(s)) ≥ 1 - c/s^{1+1/6} is pointwise in s; it cannot be turned into a bound for the uncountable intersection over all s > η without a maximal inequality or a dyadic/continuity argument. This unproved uniform-control estimate is load-bearing for the lower bound in Theorem 1 and is also invoked in the bootstrap argument.","section":"Section 2, Eq. (13)"},{"comment":"As printed, Theorem 2 is false. For b > t, min{log t - log b, 0} = log t - log b < 0, so the right-hand side is negative and, for fixed t, tends to -∞ as b → ∞. This contradicts Lemma 3.3, which gives |Cov(M_t(0),M_t(b))| ≤ c√(t/b) log t = o(1) in that regime. The intended expression is presumably max{log(t/b), 0}, or the theorem must be restricted to b < t with an explicit relation between b and t. The phrase 'the little-o notation is with respect to b → ∞' is also ill-posed because t is free; if t grows with b, the error may depend on t.","section":"Theorem 2 (Section 1.1)"},{"comment":"The lower-bound estimate in Lemma 3.4 is not justified for the stated range 'any t > b'. Substituting (31) into (32) gives an error term of order sqrt((log t - log(b log^3 b))(log t + b^2)) times (1/(1+(b log^3 b)^2) - 1/(1+t^2)). The displayed simplification to O(log log b) fails when t is large relative to b log^3 b; for example, with log t = b^2 the error is of order b/log^6 b, which is not O(log log b). Thus Lemma 3.4, and with it the covariance asymptotics in the intended range of Theorem 2, require a precise growth condition on t relative to b and a proof of the estimate in (31).","section":"Section 3, Eq. (32)"}],"minor_comments":[{"comment":"On the event H2(s), one only has 1 + Y_s ≥ 1 + πs/2 - s^{2/3}, so the displayed inequality e^{α/(1+Y_s)} ≤ e^{α/(1+πs/2)} is not literally correct; the s^{2/3} correction should appear in the denominator. The asymptotic conclusion is likely unaffected, but the displayed inequality should be corrected.","section":"Section 4, Eq. (35)"},{"comment":"The event H3.1(s) is used before being defined clearly for all z; please state explicitly whether it is required for all z ∈ H or only for the specific endpoints appearing in the argument.","section":"Section 3, Lemma 3.2"},{"comment":"There is a typo: 'sonstant' should be 'constant'.","section":"Section 3, after Eq. (22)"}],"recommendation":"reject","confidential_remarks":"The manuscript relies heavily on [BPT22] (same first author) for the construction, the martingale decomposition, and linear-growth results. The text often cites 'by [BPT22, Lemma 3.3]' without stating the exact quantitative content; the editor may wish to have the cited results checked for whether they indeed contain the needed versions. This is not a circularity concern because [BPT22] is independently published, but it affects verifiability. The main issue is the algebraic failure in Eq. (14), which is independent of those citations."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The main theorems as printed are not supported by the paper's own estimates. Using Lemma A.1, the variance density is pi/(4 * ImF_s). Under the paper's own H2/H4 events, ImF_s ≈ pi*s/2, so that density is ≈ 1/(2s), not pi/(4s). The inequality in (14), which claims pi/(4 ImF_s) >= pi/(4s), is therefore false; the correct lower bound would be about 1/2 log t. The upper bound display (12) has the same substitution issue, and even granting the unproved uniform event H4, the pi/4 constant does not follow. This is a load-bearing error: Theorem 1's variance constant is likely wrong (the argument actually suggests 1/2 log t), and Theorems 2 and 3 inherit the problem. The reader's weaker concern about H4 is real but secondary; the algebraic mistake is independent and fatal to the stated results.\n\nWhat the paper does well: it attacks an interesting object—the fluctuation field of stationary Hastings-Levitov(0) directly, rather than via Silvestri's scaling limit. The martingale bootstrap and the generator-based exponential moment bound are genuinely new and plausible. The covariance structure of the field is a natural question, and the overall strategy could work once the constants are corrected. The paper is also honest in building on published foundations [BPT22], and the reference list looks appropriate.\n\nSoft spots, proportionately: Theorem 2 as printed uses min instead of max for b>t, an obvious typo but still a concrete error. More importantly, the lower bound for Theorem 1 silently depends on H4, a uniform-in-s event that is asserted but not proved. But the algebra error in (12) and (14) is the real problem. It is not a matter of missing technical detail; the inequality just goes the wrong way. This cannot be fixed by a small patch.\n\nWho is this for? Specialists in stochastic planar growth and log-correlated fields. The method may be of interest, but the main result as stated is not reliable. If the authors correct the constant (likely to 1/2), the paper could be a solid contribution. As is, I would not cite the theorems. A serious referee could still help if the editor expects heavy revision, but the paper needs substantial rework before it is publishable.","headline":"The pi/4 variance constant is not just unproved; it is inconsistent with the paper's own Lemma A.1 and the known pi/2 linear growth rate.","tokens_in":11596,"tokens_out":4128,"would_cite":false,"duration_ms":37095,"reading_group":"maybe","serious_thinker":"no","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60K35","60G60","30C35"],"pacs":[],"model":"deepseek-v4-flash","headline":"The fluctuation field of stationary Hastings-Levitov(0) is log-correlated: variance (π/4) log t, covariance (π/4)(log t − log b) for b < t, and maximum O(log t) with high probability.","keywords":["stationary Hastings-Levitov","log-correlated random field","conformal slit maps","Doob martingale decomposition","fluctuation field","Diffusion Limited Aggregation","maximum of random field","Markov generator"],"falsifier":"Simulate the SHL(0) process for large $t$ (for example $t = 10^6$) and compute $\\mathbb{E}|M_t(0)|^2 / \\log t$; if it does not converge to $\\pi/4 \\approx 0.785$, the main variance claim fails. Alternatively, check the uniform corridor directly: for many samples, record whether $\\mathrm{Im}\\,F_s(0)$ ever exits $[\\pi s/2 - s^{2/3}, \\pi s/2 + s^{2/3}]$ for $s > \\eta$; a non-negligible probability of exit would disprove the asserted event $H_4(\\eta)$ on which the lower bound depends.","tokens_in":10582,"feed_emoji":"📈","tokens_out":10983,"duration_ms":90055,"temperature":0.7,"pith_summary":"Stationary Hastings-Levitov(0) is a random growth model built by composing conformal slit maps at Poisson times, and this paper studies the fluctuations around its deterministic linear growth. The central claim is that the fluctuation field $\\{M_t(x)\\}$ is log-correlated: its variance at a point is $(\\pi/4)\\log t$, and its covariance between two points separated by $b$ is $(\\pi/4)(\\log t - \\log b)$ for $b < t$. The paper also proves that the maximum imaginary fluctuation over the interval $[0,t]$ is at most of order $\\log t$ with high probability, with failure probability $O(t^{-1/2})$. If these results are correct, the model provides a non-Gaussian, physically motivated example of a log-correlated random field with logarithmic extremes. The proof works directly on the physical process rather than on a scaling limit, and it obtains the variance and covariance from an identity expressing the martingale increments as an integral over slit maps.","feed_headline":"Growth noise is log-correlated in stationary Hastings-Levitov","feed_subtitle":"Variance grows as (π/4) log t; correlations decay logarithmically with separation.","key_machinery":"The argument is carried by the martingale representation $F_t(z) = z + i\\pi t/2 + M_t(z)$, where $F_t$ is the backwards composition of slit maps $\\varphi_x(z) = x + \\sqrt{(z-x)^2 - 1}$ at Poisson-arrival times. For the variance, the key identity is $\\mathbb{E}|M_t(0)|^2 = \\mathbb{E}\\int_0^t\\int_{-\\infty}^{\\infty}|\\varphi_x(F_s(0)) - F_s(0)|^2\\,dx\\,ds$; Lemma A.1 evaluates the $x$-integral at a point $iy$ as $\\min\\{1, \\pi/(4y)\\} + O(y^{-3})$, so together with the strong law $\\mathrm{Im}\\,F_s(0) \\approx \\pi s/2$ the variance becomes $(\\pi/4)\\int ds/s = (\\pi/4)\\log t$. For the maximum bound, the Markov generator of $Y_t = \\mathrm{Im}\\,F_t(0)$ is computed as $\\mathcal{L}f(\\zeta) = \\int_{-\\infty}^{\\infty}[f(\\zeta + \\Delta(\\zeta,x)) - f(\\zeta)]\\,dx$, and applying it to $f(y) = e^{\\alpha y}$ yields an integral inequality whose solution bounds the exponential moments of $\\mathrm{Im}\\,M_t(0)$ by $\\exp(\\tfrac{\\pi}{2}\\alpha^2 e^\\alpha)\\,t^{\\alpha^2}$. Those moments give a Chernoff tail $\\mathbb{P}(\\mathrm{Im}\\,M_t(0) > \\beta\\log t) \\le \\mathrm{const}\\cdot t^{-\\beta^2/4}$, and the maximum result follows by a grid-approximation with derivative control via the half-plane Koebe distortion theorem and a planarity argument.","core_discovery":"The paper's discovery is that the fluctuation field $\\{M_t(x)\\}_{x\\in\\mathbb{R}}$ of the backwards stationary Hastings-Levitov(0) map, defined as the martingale part of the decomposition $F_t(z) = z + i\\pi t/2 + M_t(z)$, exhibits logarithmic spatial correlations: $\\mathbb{E}|M_t(0)|^2 = (\\pi/4)\\log t(1+o(1))$ and $\\mathrm{Cov}(M_t(0),M_t(b)) = (\\pi/4)(\\log t - \\log b)(1+o(1))$ for $b < t$, with covariance negligible for $b > t$. In addition, the maximum of $\\mathrm{Im}\\,M_t$ over $x\\in[0,t]$ satisfies $\\mathbb{P}(\\max_{x\\in[0,t]}\\mathrm{Im}\\,M_t(x) > \\beta\\log t) < c t^{-1/2}$ for some $\\beta > 0$ and all large $t$. These results are established for the process itself, not just for a small-particle scaling limit, through variance asymptotics obtained from an integral over slit maps and through exponential-moment bounds for the Markov generator of the imaginary part.","pith_inferences":["The proof would be completed by proving the uniform event $H_4(\\eta)$ asserted after equation (13): that $\\mathrm{Im}\\,F_s(0)$ stays within $s^{2/3}$ of $\\pi s/2$ for every $s>\\eta$ simultaneously, with probability at least $1-\\eta^{-1/3}$; a direct simulation of the trajectory could check whether this linear-growth corridor is actually maintained, and if it fails the central asymptotics may still","Because the simulated marginals of $M_t(0)$ look asymmetric, the log-correlated field here is probably not Gaussian; if the variance and covariance results hold, SHL(0) becomes a rare example of a non-Gaussian log-correlated field whose maximum still grows logarithmically, which may inform conjectures about universality of log-correlated extremes.","The paper's bound on the maximum is an upper bound only; a natural companion conjecture is matching logarithmic lower bounds (e.g., that the maximum is at least a deterministic constant times $\\log t$), which would establish that the extremes grow deterministically like the standard deviation, and this would be a testable extension of the present results."],"forward_implications":["The variance of the fluctuation field grows exactly like $(\\pi/4)\\log t$, confirming at the level of the full process (not only a scaling limit) the logarithmic growth of fluctuations expected for this aggregation model.","The two-point covariance decays as $(\\pi/4)(\\log t - \\log b)$ in separation $b$, placing the field in the log-correlated universality class alongside objects like the Gaussian free field and critical models.","The maximum imaginary part over an interval of length $t$ is $O(\\log t)$ with probability $1 - O(t^{-1/2})$, and the single-point right tail obeys $\\mathbb{P}(\\mathrm{Im}\\,M_t(0) > \\beta\\log t) \\le \\mathrm{const}\\cdot t^{-\\beta^2/4}$, so large upward fluctuations above the logarithmic scale have polynomially decaying probability.","The generator-based method yields quantitative exponential-moment bounds, giving a route to extremal statistics (e.g., tightness of the centered maximum) that does not require Gaussianity."],"supporting_citations":[{"why":"Provides the definition of stationary Hastings-Levitov(0), the martingale decomposition $F_t(z) = z + i\\pi t/2 + M_t(z)$, the bound in Lemma 3.3, the half-plane Koebe distortion theorem, and the planar path crossing results used for the maximum.","marker":"[BPT22]"},{"why":"Prior work establishing logarithmic correlations for the small-particle scaling limit of Hastings-Levitov(0); the present paper extends this to the physical process.","marker":"[Sil17]"},{"why":"Supplies the random-walk intersection estimate used in the planarity argument of Theorem 3.","marker":"[LL10]"}],"fun_headline_variants":["Log-correlated fluctuations proven for stationary Hastings-Levitov","Hastings-Levitov noise has logarithmic spatial correlations","Fluctuations in stationary Hastings-Levitov are log-correlated","Proven: logarithmic correlations for Hastings-Levitov fluctuations"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The covariance and variance lower bounds rely on an unproved uniform event: that the imaginary part of the growing map at every time $s$ stays between $\\pi s/2 - s^{2/3}$ and $\\pi s/2 + s^{2/3}$ (for all $s$ after some threshold) with high probability, a corridor the paper asserts after equation (13) but does not prove; if this corridor fails, the variance and covariance asymptotics are not established by the given argument.","fun_headline_variants_meta":{"raw":{"variants":["Log-correlated fluctuations proven for stationary Hastings-Levitov","Hastings-Levitov noise has logarithmic spatial correlations","Fluctuations in stationary Hastings-Levitov are log-correlated","Proven: logarithmic correlations for Hastings-Levitov fluctuations"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00177,"raw_usage":{"total_tokens":6939,"prompt_tokens":860,"completion_tokens":6079,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":476,"completion_tokens_details":{"reasoning_tokens":6010}},"tokens_in":476,"tokens_out":6079,"duration_ms":38913,"temperature":1.0,"reasoning_tokens":6010,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-09T04:31:47.824080+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Simulate the SHL(0) process for large $t$ (for example $t = 10^6$) and compute $\\mathbb{E}|M_t(0)|^2 / \\log t$; if it does not converge to $\\pi/4 \\approx 0.785$, the main variance claim fails. Alternatively, check the uniform corridor directly: for many samples, record whether $\\mathrm{Im}\\,F_s(0)$ ever exits $[\\pi s/2 - s^{2/3}, \\pi s/2 + s^{2/3}]$ for $s > \\eta$; a non-negligible probability of exit would disprove the asserted event $H_4(\\eta)$ on which the lower bound depends.","supporting_citations":[],"review_version":1}