{"id":"71a169ef-0370-4dd5-9241-64b6e7799be1","arxiv_id":"1909.01617","paper_version":2,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":8.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"For d≥3, the joint occupancy counts of a critical branching random walk approach a deterministic multiple of an exponential with Wasserstein rate n^{−(d−2)/(2(d+1))}; for d≥7, centered fluctuations converge to a multivariate Laplace law with rate n^{−(2d−9)/(6(2d+1))}.","lead":"For a population that reproduces and spreads on a lattice, this paper measures how fast the numbers of sites with exactly one, two, or more individuals settle into a simple exponential law. For lattices of dimension seven or higher it also shows that the fluctuations around that law converge to a Laplace distribution, with explicit error rates.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified: the proof of Theorems 1.1 and 1.2 is internally consistent; the only soft spot is the asserted algebraic cancellation in Lemma 2.9, which a symbolic recomputation can settle.","rationale":"The reader's verdict is ACCEPT with high confidence, and my independent review of the argument does not uncover a concrete mathematical error. The central claim, Theorem 1.2, rests on Lemma 2.9 for the covariance limit and the fourth-moment asymptotics, and on the random-sum approximation Theorem 2.8. I checked the moment budgets: the tail bound for E[Z_n Y_{n;1}] and the higher-moment analogue are consistent with the stated moment condition, and the choice of alpha in Lemma 2.9 works for every d >= 7. The rates in the proof of Theorem 1.2 all match the announced exponent: Lemma 2.5 gives n^{-(2d-9)/(6(2d+1))}, the replacement terms in (2.26) and (2.27) are smaller, Theorem 2.8 contributes the same critical order through its M^{1/3} E||X||^{1/3} term, and Lemma 3.3 plus the covariance rate from Lemma 2.9 gives a negligible contribution for d >= 7. The one part I could not fully verify by reading is the algebraic cancellation of the gamma_3 and gamma_4 terms in the fourth-moment recursion. The authors state the result of this computation without displaying all coefficients, and an error here would break (2.17) and hence Theorem 1.2. However, I have no positive evidence of such an error, and the structure of the recursion is consistent with the heuristic random-sum calculation E[(sum X_i)^4] = E[Z_1] E[X^4] + 3 E[Z_1(Z_1-1)] A^2, which gives exactly an increment of 3 sigma^2 A^2. The paper's explicit caveat that tilde{Sigma} may be degenerate or zero is acknowledged and does not invalidate the theorem as stated, though it tempers the strength of the Laplace conclusion. I therefore see no reason to change the verdict, and I identify the unverified algebraic step as the natural target for a concrete verification.","tokens_in":23768,"tokens_out":35434,"duration_ms":301100,"concrete_test":"Independently recompute the fourth-moment expansion in Lemma 2.9: using the displayed expressions for E[(sum M_i)^4], E[(sum M_i)^3(sum Z_i)], E[(sum M_i)^2(sum Z_i)^2], E[(sum M_i)(sum Z_i)^3], and E[(sum Z_i)^4], substitute into (2.24) and simplify symbolically. Verify that the coefficients of gamma_3 and gamma_4 are zero and that the coefficient of sigma^2 in the resulting difference is exactly 3 A_{n-1}(j,j)^2. If any additional term survives, the recursion for E[(M_n - mu_n Z_n)^4] would need revision and (2.17) would not follow.","verdict_should_be":"UNCHANGED","load_bearing_attack":"After checking the proof of Theorem 1.2, I find no gap that would invalidate the central claim. The moment condition E[X^{5+floor(18/(d-6))}] < infinity is exactly what is needed: with beta = floor(18/(d-6)) + 1, the bound E[Z_n^{4+beta}] = O(n^{3+beta}) requires E[X^{4+beta}] < infinity, which is the stated hypothesis. The choice alpha = (d-6)/6 - epsilon in Lemma 2.9 gives alpha*beta > 3 for every d >= 7, so the error terms e^{(k,4-k)}_n are o(1), and the recursion for the fourth moment yields (2.17). The covariance convergence |A_n - Sigma| <= c n^{2-d/3} is summable for d >= 7 and supplies the rates used in (2.29). Every error term in the proof of Theorem 1.2 has an exponent at least as small as the claimed -(2d-9)/(6(2d+1)); for d=7 the n^{-1/18} rate is attained exactly by the Lemma 2.5 and Theorem 2.8 first-term bounds. The only place where I would want independent verification is the algebraic simplification inside Lemma 2.9: the authors assert that after expanding the four random-sum expectations, the gamma_3 and gamma_4 coefficients cancel and only 3 sigma^2 A_{n-1}(j,j)^2 survives. They do not display the full coefficient computation. This is an omitted calculation, not an observed error. The authors also explicitly flag the possibility that the limiting covariance tilde{Sigma} is degenerate or zero; the theorem still holds in that case, but the advertised Laplace law would be a point mass. That is a limitation of the result's strength, not of its correctness.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the critical nearest-neighbor branching random walk on the d-dimensional integer lattice, conditioned on non-extinction. It proves two approximation results for the occupancy statistics M_n(j), the number of sites in generation n occupied by exactly j particles. Theorem 1.1 gives a Wasserstein rate of convergence for the joint law of (Z_n/n, M_n(1)/n, ..., M_n(r)/n) to the Lalley-Zheng limit (1, kappa_1, ..., kappa_r) Z, where Z is exponential with mean sigma^2/2; the rate is n^{-(d-2)/(2(d+1))} for d >= 3 under a third-moment assumption. Theorem 1.2, for d >= 7 and finite moments of order 5 + floor(18/(d-6)), shows that the scaled centered occupancy counts (M_n(j) - kappa_j Z_n)/sqrt(n) converge in the L1-Wasserstein metric to a multivariate symmetric Laplace distribution SL_r(tilde{Sigma}), with an explicit rate n^{-(2d-9)/(6(2d+1))}. The proofs use a size-biased tree construction, couplings that relate conditioning on Z_n > 0 to conditioning on Z_m > 0, a random-sum Laplace approximation theorem (Theorem 2.8), and a detailed analysis of the covariance and fourth-moment structure of the centered occupancy counts (Lemma 2.9).","tokens_in":24135,"tokens_out":14219,"duration_ms":118398,"significance":"If correct, Theorem 1.2 provides the first second-order fluctuation limit for these occupancy statistics and reveals a Laplace, rather than Gaussian, scaling regime in dimension d >= 7. The paper is notable for its explicit rates, its careful handling of the conditioning on non-extinction via the size-biased tree, and its general Theorem 2.8, which gives a Wasserstein-rate Laplace approximation for geometric random sums with a self-contained proof. The authors also honestly flag the possibility that the limiting covariance tilde{Sigma} is degenerate or even zero; in that case the advertised Laplace law collapses to a point mass, a limitation of strength rather than correctness. The overall structure is clear, the external results (Lyons-Pemantle-Peres, Lalley-Zheng, Vatutin-Zubkov, and the authors' earlier Peköz-Röllin Yaglom rate) are used appropriately, and the moment conditions are matched to the needs of the error estimates.","major_comments":[{"comment":"The derivation of the fourth-moment asymptotics (2.17) contains a substantial omitted calculation. After displaying the expansions of the four random-sum expectations, the text states that 'it is easiest to compute the coefficients for each of sigma^2, gamma_3, gamma_4; the last two are zero' and immediately concludes the recursion E[(M_n - mu_n Z_n)^4] - E[(M_{n-1} - mu_{n-1} Z_{n-1})^4] = 3 sigma^2 A_{n-1}(j,j)^2 + O(n^{-delta'}). This algebraic cancellation is load-bearing: the resulting bound E[(tilde{M}_{n,m}(j))^4] <= c(n-m) is used in the proof of Theorem 1.2 to control the third moments of the summands via Hölder's inequality. Since the gamma_3 and gamma_4 coefficients involve several terms each, and a failure of the stated cancellation would change the leading constant in (2.17) and could invalidate the recursion, the authors should either display the full coefficient computation or provide a verified symbolic computation in an appendix.","section":"Lemma 2.9, proof of (2.17)"}],"minor_comments":[{"comment":"In the chain bounding d_W(L(hat{Z}_m), Geo(mu^{-1})), the expressions 'Exp(sigma^2/(2m))' and 'Exp(mu^{-1})' are inconsistent with the subsequent bound |m sigma^2/2 - mu|. Since the paper uses Exp(theta) to denote an exponential distribution with mean theta, the second and third terms in that chain should read Exp(m sigma^2/2) and Exp(mu), respectively; the displayed version appears to be a typo that makes the inequality formally incorrect.","section":"Proof of Theorem 1.2, display after (2.28)"},{"comment":"In the definition of M_k(f), the denominator in the displayed derivative is written as 'partial x_{i1} ... partial x_{ir}', but it should be 'partial x_{i1} ... partial x_{ik}' since the sum runs over i_1,...,i_k; this is a notational typo that could confuse readers.","section":"Section 3, definition of M_k(f)"},{"comment":"The bound in the lemma statement, c(nm(n-m)^{-d/2} + (n-m)), is slightly stronger than what the proof immediately yields, namely c(n(1 + m sigma^2)(n-m)^{-d/2} + (n-m)). The equivalence is clear for bounded sigma^2, but stating the proof's bound would be cleaner.","section":"Lemma 2.5"}],"recommendation":"major_revision","confidential_remarks":"The paper is well within the scope of the journal and the central results are very likely correct. The only substantive request is to supply the omitted algebraic computation in Lemma 2.9, without which the proof of (2.17) is incomplete as written. The typos in the proof of Theorem 1.2 and in Section 3 are easily fixed. Once the computation is provided and the typos corrected, I would support acceptance."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is the first paper to give quantitative Wasserstein rates for the Lalley–Zheng exponential limit in d≥3, and for d≥7 it proves a genuine multivariate Laplace fluctuation limit with rate. The random-sum Laplace approximation in Theorem 2.8 is also a useful standalone tool. The proofs are built from published results—Lyons–Pemantle–Peres, Lalley–Zheng, Vatutin–Zubkov, and the authors' own Yaglom rate—so there is no circularity I can see; the conditioning couplings in Section 2 are handled carefully.\n\nWhat is good: Theorem 1.1's rate n^{-(d-2)/(2(d+1))} is new; Theorem 1.2's symmetric Laplace limit for the scaled centered occupancy vector in d≥7 is new as far as I know; Lemma 2.9 pins down covariance convergence and the fourth-moment growth. The authors also flag the caveat that the limiting covariance could be degenerate, and they state exactly where d≥7 enters.\n\nSoft spots, in proportion: the strong moment condition 5+floor(18/(d-6)) is real but appears to be exactly what the proof needs; I would not call it a flaw. The main thing I would want independently checked is the algebra inside Lemma 2.9: they assert that after expanding four random-sum expectations, the gamma_3 and gamma_4 coefficients cancel and only 3σ² A_{n-1}(j,j)² survives. They do not show the full coefficient computation. The stress-test note says this is an omitted calculation, not an observed error, and on reading the surrounding expansions I agree; a symbolic recomputation or a longer appendix would settle it. Also, the limit covariance is only shown to exist, not to be nondegenerate, so in some parameter regimes the advertised Laplace law could collapse to a point mass. The theorem is still correct in that case, but the result is weaker than the title suggests in those regimes.\n\nCitations look right: the relevant Lalley–Zheng, Peköz–Röllin, Pike–Ren, and Yaglom-rate literature is there, and the claimed gaps are genuinely gaps in that literature as far as I can tell. There are no data or empirical claims; reproducibility is proof verification.\n\nBottom line: this deserves a serious referee. The referee should spend time on the Lemma 2.9 expansion and on checking the error-exponent bookkeeping in Theorem 1.2, but I did not find a load-bearing gap. I would bring it to the reading group and would cite it if I worked on occupation statistics or second-order Yaglom limits.","headline":"Genuinely new rates and a genuinely new Laplace fluctuation limit for branching random walk occupancy statistics, with the one real soft spot being an algebraic cancellation in Lemma 2.9 that the authors summarize rather than fully display.","tokens_in":24715,"tokens_out":1783,"would_cite":true,"duration_ms":17179,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60J80","60F05"],"pacs":[],"model":"deepseek-v4-flash","headline":"In dimensions $d\\geq 7$, the centered, scaled occupancy counts of a critical nearest-neighbor branching random walk, conditioned on non-extinction, converge in $L^1$-Wasserstein distance to a multivariate symmetric Laplace distribution…","keywords":["branching random walk","occupancy statistics","critical Galton-Watson process","exponential limit","multivariate symmetric Laplace distribution","Wasserstein metric","geometric random sums","size-biased tree"],"falsifier":"Look for a counterexample within the theorem's hypotheses: in dimension $d=7$, choose an offspring law with $\\mathbb{E}[X]=1$, finite variance, and $\\mathbb{E}[X^{5+\\lfloor 18/(d-6)\\rfloor}]<\\infty$ for which the covariance sequence $A_n(j,k)$ does not converge, or for which $n^{-1}\\mathbb{E}[(M_n(j)-\\mathbb{E}[M_n(j)]Z_n)^4]$ does not converge to $3\\sigma^2\\Sigma_{jj}^2$. If such a law exists, Theorem 1.2 is false; the proof in the paper shows these limits hold under the stated moments, so checking their claimed convergence rate $n^{2-d/3}$ numerically would test the sharpness of the argument.","tokens_in":23548,"feed_emoji":"🎲","tokens_out":17753,"duration_ms":148372,"temperature":0.7,"pith_summary":"The paper studies occupancy counts of a critical nearest-neighbor branching random walk on the $d$-dimensional integer lattice, conditioned on not going extinct. Building on the known first-order result that these counts converge to deterministic multiples of a single exponential variable, the authors prove an explicit Wasserstein rate for that exponential approximation in every dimension $d\\geq 3$. Their main new claim is a second-order limit: for $d\\geq 7$, after subtracting $\\kappa_j Z_n$ from the number of sites with $j$ particles and dividing by $\\sqrt{n}$, the joint distribution converges in $L^1$-Wasserstein distance to a multivariate symmetric Laplace distribution (the law of $\\sqrt{E}Z$ with $E$ a unit exponential independent of a centered normal vector $Z$), with rate $n^{-(2d-9)/(6(2d+1))}$. This matters because it identifies a non-Gaussian scaling regime for the fluctuations and provides quantitative bounds rather than only convergence in distribution.","feed_headline":"Branching-walk occupancy fluctuations go Laplace for d≥7","feed_subtitle":"Second-order fluctuations around the classical exponential limit are non-Gaussian, with explicit Wasserstein rate.","key_machinery":"The load-bearing objects are the centered subtree contributions $\\tilde M^i_{n,m}(j)=M^i_{n,m}(j)-\\mu_{n,m}(j) Z^i_{n,m}$: for a generation-$m$ ancestor $i$, $M^i_{n,m}(j)$ counts multiplicity-$j$ sites among its generation-$n$ descendants and $Z^i_{n,m}$ is its number of generation-$n$ descendants, so subtracting $\\mu_{n,m}(j) Z^i_{n,m}$ removes the mean contribution correlated with the total population. Three mechanisms carry the argument: the size-biased tree construction that couples the conditioned process to a marked spine with independent side subtrees; Lemma 2.9, which shows the covariance sequence $A_n(j,k)$ converges at rate $n^{2-d/3}$ and that the fourth moment of each centered coordinate grows as $3n\\sigma^2\\Sigma_{jj}^2$; and Theorem 2.8, a Wasserstein-rate symmetric Laplace approximation for sums of a random number of i.i.d. zero-mean vectors. The approximation to $\\mathrm{SL}_r(\\tilde\\Sigma)$ follows because the random number of summands is approximately geometric, and a geometric sum with small success parameter is the classical route to the symmetric Laplace distribution.","core_discovery":"On the model's own terms, the central discovery is Theorem 1.2: for $d\\geq 7$, if the offspring variable $X$ has mean $1$, variance $\\sigma^2$, and finite moments of order $5+\\lfloor 18/(d-6)\\rfloor$, then for every fixed $r\\geq 1$ there are constants $\\kappa_1,\\ldots,\\kappa_r$ and a non-negative definite matrix $\\tilde\\Sigma$ such that $$d_W\\bigl(\\mathcal{L}\\bigl((M_n(1)-\\kappa_1 Z_n)/\\sqrt{n},\\ldots,(M_n(r)-\\kappa_r Z_n)/\\sqrt{n}\\mid Z_n>0\\bigr), \\mathrm{SL}_r(\\tilde\\Sigma)\\bigr)\\leq c\\, $n^{{-(2d-9)/(6(2d+1))}}$.$$ The matrix $\\tilde\\Sigma$ is $(\\sigma^2/2)$ times the limit of the unconditional covariance matrix of the centered variables $M_n(j)-\\mathbb{E}[M_n(j)]Z_n$, which the proof shows converges; the paper cannot rule out that $\\tilde\\Sigma$ is degenerate or zero. A companion theorem gives the rate $n^{-(d-2)/(2(d+1))}$ for the first-order exponential approximation for all $d\\geq 3$ under only a third moment. The proof replaces the conditioned occupancy vector by a random sum of nearly independent subtree contributions, applies a new Wasserstein-rate Laplace approximation for geometric random sums, and controls the remainder with a size-biased tree coupling.","pith_inferences":["The authors leave the minimal dimension open; testing whether the covariance limit $A_n(j,k)$ exists for $d=5,6$ would show whether the $d\\geq 7$ restriction is an artifact of the estimates or a genuine phase boundary.","Since the symmetric Laplace law is the signature of geometric random summation, the result suggests a broader principle: in the transient regime, second-order occupancy fluctuations are governed by random summation rather than by a classical Gaussian central limit; this paper provides the first instance.","A natural extension is to replace the nearest-neighbor walk by other transient random-walk kernels, or to pass to supercritical branching random walks, where the number of summands is no longer geometric and a different limit may emerge."],"forward_implications":["For $d\\geq 7$, the joint law of the centered occupancy counts has a quantitative distributional limit: the $L^1$-Wasserstein distance to $\\mathrm{SL}_r(\\tilde\\Sigma)$ is $O(n^{-(2d-9)/(6(2d+1))})$.","The same machinery gives a rate $n^{-(d-2)/(2(d+1))}$ for the first-order exponential approximation in every dimension $d\\geq 3$, assuming only $\\mathbb{E}[X^3]<\\infty$.","The limiting covariance is tied to the offspring variance: $\\tilde\\Sigma=(\\sigma^2/2)\\Sigma$, where $\\Sigma$ is the limit of the unconditional covariance $A_n(j,k)$; the theorem stays valid even if that limit is degenerate or zero.","Theorem 2.8 is a standalone approximation tool: a Wasserstein bound for replacing a sum of a random number of i.i.d. centered vectors by a symmetric Laplace variable, applicable beyond branching random walks.","Sharpening the error estimates might lower the dimension threshold, but the authors identify the upper bound in (2.23) as potentially sharp, so $d\\geq 7$ may be intrinsic to this approach."],"supporting_citations":[{"why":"supplies the first-order exponential limit for occupancy counts and the constants κ_j, and the estimate for the drift of E[M_n(j)] used in Lemma 2.4.","marker":"Lalley and Zheng (2011)"},{"why":"provides the classical exponential limit for critical Galton-Watson processes that the first-order approximation refines.","marker":"Yaglom (1947)"},{"why":"gives the Wasserstein rate for the exponential approximation of Z_n/n conditioned on survival, stated as Theorem 1.3 and used throughout.","marker":"Peköz and Röllin (2011)"},{"why":"supplies the size-biased tree construction and the distributional identity for the process conditioned on non-extinction.","marker":"Lyons et al. (1995)"},{"why":"provides the asymptotic estimate for P(Z_n>0) that controls the cost of changing the conditioning.","marker":"Vatutin and Zubkov (1985)"},{"why":"defines the multivariate symmetric Laplace distribution and collects the properties used in the geometric-sum interpretation.","marker":"Kotz et al. (2001)"},{"why":"provides the multivariate normal approximation with third-moment Wasserstein error used in Theorem 3.1.","marker":"Meckes (2009)"},{"why":"gives the smoothing bound on third derivatives used to run that normal approximation.","marker":"Raič (2018)"}],"fun_headline_variants":["Branching walk occupancy fluctuations hit Laplace for d≥7","From exponential to Laplace: refining branching walk occupancy","High-dim branching walk: occupancy oscillations follow Laplace law","Multivariate Laplace emerges in branching walk occupancy for d≥7","Occupancy statistics in branching walks: Laplace beyond exponential"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument depends on the offspring distribution having enough finite moments: in dimension $d\\geq 7$ it needs $\\mathbb{E}[X]=1$, $\\mathrm{Var}(X)=\\sigma^2$, and $\\mathbb{E}[X^{5+\\lfloor 18/(d-6)\\rfloor}]<\\infty$; if the tail is heavier, the covariance sequence $A_n(j,k)$ in Lemma 2.9 may fail to converge and the Laplace approximation is not established.","fun_headline_variants_meta":{"raw":{"variants":["Branching walk occupancy fluctuations hit Laplace for d≥7","From exponential to Laplace: refining branching walk occupancy","High-dim branching walk: occupancy oscillations follow Laplace law","Multivariate Laplace emerges in branching walk occupancy for d≥7","Occupancy statistics in branching walks: Laplace beyond exponential"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000361,"raw_usage":{"total_tokens":2005,"prompt_tokens":1052,"completion_tokens":953,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":668,"completion_tokens_details":{"reasoning_tokens":875}},"tokens_in":668,"tokens_out":953,"duration_ms":8705,"temperature":1.0,"reasoning_tokens":875,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T05:12:12.600857+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Look for a counterexample within the theorem's hypotheses: in dimension $d=7$, choose an offspring law with $\\mathbb{E}[X]=1$, finite variance, and $\\mathbb{E}[X^{5+\\lfloor 18/(d-6)\\rfloor}]<\\infty$ for which the covariance sequence $A_n(j,k)$ does not converge, or for which $n^{-1}\\mathbb{E}[(M_n(j)-\\mathbb{E}[M_n(j)]Z_n)^4]$ does not converge to $3\\sigma^2\\Sigma_{jj}^2$. If such a law exists, Theorem 1.2 is false; the proof in the paper shows these limits hold under the stated moments, so checking their claimed convergence rate $n^{2-d/3}$ numerically would test the sharpness of the argument.","supporting_citations":[],"review_version":1}