Exponential and Laplace approximation for occupation statistics of branching random walk

Adrian R\"ollin; Erol Pek\"oz; Nathan Ross

REVIEW 1 major objections 3 minor 13 references

Exponential and Laplace approximation for occupation statistics of branching random walk

T0 review · 1 major / 3 minor · reviewed 2026-08-14 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read 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…

desk verdict 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. read the letter →

arxiv 1909.01617 v2 pith:AKMHHNR7 submitted 2019-09-04 math.PR

Erol Peköz , Adrian Röllin , Nathan Ross This is my paper · ORCID

classification math.PR MSC 60J8060F05

keywords branchingrandomwalkoccupancystatisticscriticalGalton-WatsonprocessexponentiallimitmultivariatesymmetricLaplacedistributionWassersteinmetricgeometricsumssize-biasedtree

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

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.

What carries the argument

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.

What would settle it

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.

Watch

Extended reading notes

Core claim

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.

Load-bearing premise

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.

Editorial extensions

If this is right

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.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

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.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the 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.

What 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.

Soft 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.

Citations 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.

Bottom 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.

Referee Report

1 major / 3 minor

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).

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 (1)

[Lemma 2.9, proof of (2.17)] 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.

minor comments (3)

[Proof of Theorem 1.2, display after (2.28)] 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 3, definition of M_k(f)] 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.
[Lemma 2.5] 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.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the limit theorems are proved from external published results and explicit error bounds; the only soft spot is an unexpanded algebraic cancellation, not a circular step.

full rationale

The paper's central derivation chain is not circular. Theorem 1.2 is proved by decomposing the occupancy vector into a random sum of conditionally i.i.d. contributions (Lemmas 2.4-2.5), bounding the replacement error, applying the paper's own general geometric-sum Laplace approximation (Theorem 2.8), and then supplying the needed covariance limits and fourth-moment growth through Lemma 2.9. Each input is a hypothesis on the model or an independently proved statement. The constants κ_j are the limits E[M_n(j)] established by Lalley and Zheng (2011), and the covariance matrix Σ is defined as lim A_n(j,k) of model covariances; neither is fit to reproduce the Laplace law. Lemma 2.9's moment hypothesis E[X^{5+floor(18/(d-6))}]<∞ is exactly what the bound E[Z_n^{4+β}]=O(n^{3+β}) requires, so the theorem is not hiding the target assumption. The paper cites Peköz and Röllin (2011, Theorem 3.3) for the Yaglom rate; although this is a self-citation, it is a published, parameter-free rate result with assumptions that do not include occupancy statistics, so it is independent support and not a circular load-bearing self-citation. Similarly, Lalley-Zheng, Vatutin-Zubkov, and Lyons-Pemantle-Peres are external. The only soft spot is an omitted coefficient computation in Lemma 2.9: the authors state that after expanding the random sums, the gamma_3 and gamma_4 coefficients cancel and only 3 sigma^2 A_{n-1}(j,j)^2 remains, without displaying the full algebra. This is an unverified calculation, not a circular reduction; it can be checked symbolically and does not make the theorem equivalent to its inputs. The paper also explicitly flags that tilde Sigma might be degenerate or zero, which weakens the advertised conclusion but does not indicate circularity. Accordingly the score is 0.

Assumptions & free parameters 0 free parameters · 5 assumptions · 0 invented entities

No free parameters are fitted: κ_j and Σ~ are limits of model quantities. The proof rests on the criticality and moment hypotheses plus five external published theorems. No invented entities are introduced.

assumptions (5)

domain assumption Critical offspring distribution: E[X]=1, 0<Var(X)=σ^2<∞; Theorem 1.2 adds E[X^{5+⌊18/(d−6)⌋}]<∞.
Explicit hypotheses in the Introduction and Theorem statements.
standard math Kolmogorov's estimate (Lemma 2.2): P(Zn>0)=2/(nσ^2)+O(log^2 n/n^2).
Taken from Vatutin-Zubkov (1985); used to convert conditional expectations and for the Wasserstein bounds.
standard math Size-biased tree construction of Lyons-Pemantle-Peres (1995, Theorem C(i)): the tree T_n is distributed as the original Galton-Watson tree conditioned on non-extinction up to generation n.
Basis of the coupling constructions at the start of Section 2.
standard math Lalley-Zheng (2011) Proposition 21 and Corollary 20: |E[M_n(j)]−κ_j| ≤ c n^{1−d/2} and bounds on E[Y_{n;m}].
Used in Lemma 2.4 to control means and collisions.
standard math Yaglom rate (Peköz-Röllin 2011, Theorem 3.3): dW(L(Zn/n|Zn>0), Exp(σ^2/2)) ≤ c log n/n.
External published rate used as the base exponential approximation; parameter-free and independent of the current results.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Exponential and Laplace approximation for occupation statistics of branching random walk." pith.science (2026). https://pith.science/paper/AKMHHNR7

@misc{pith2026190901617,
  author       = {Pith},
  title        = {Pith review of: Exponential and Laplace approximation for occupation statistics of branching random walk},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AKMHHNR7}},
  note         = {Machine review of arXiv:1909.01617}
}

abstract

We study occupancy counts for the critical nearest-neighbor branching random walk on the $d$-dimensional lattice, conditioned on non-extinction. For $d\geq 3$, Lalley and Zheng (2011) showed that the properly scaled joint distribution of the number of sites occupied by $j$ generation-$n$ particles, $j=1,2,\ldots$, converges in distribution as $n$ goes to infinity, to a deterministic multiple of a single exponential random variable. The limiting exponential variable can be understood as the classical Yaglom limit of the total population size of generation $n$. Here we study the second order fluctuations around this limit, first, by providing a rate of convergence in the Wasserstein metric that holds for all $d\geq3$, and second, by showing that for $d\geq 7$, the weak limit of the scaled joint differences between the number of occupancy-$j$ sites and appropriate multiples of the total population size converge in the Wasserstein metric to a multivariate symmetric Laplace distribution. We also provide a rate of convergence for this latter result.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

13 extracted references · 11 canonical work pages

[1]

D. A. Dawson, 2017. Introductory lectures on stochastic population systems. Technical report. Available at https://arxiv.org/abs/1705.03781

arXiv 2017
[2]

Geiger, 2000

J. Geiger, 2000. A new proof of Y aglom's exponential limit law. In Mathematics and computer science ( V ersailles, 2000) , Trends Math., pages 245--249. Birkh\" a user, Basel

work page 2000
[3]

S. Kotz, T. J. Kozubowski, and K. Podg\'orski, 2001. The L aplace distribution and generalizations . Birkh\"auser Boston, Inc., Boston, MA. A revisit with applications to communications, economics, engineering, and finance

work page 2001
[4]

S. P. Lalley and X. Zheng, 2011. Occupation statistics of critical branching random walks in two or higher dimensions. Ann. Probab., 39 0 (1): 0 327--368

work page 2011
[5]

Lyons, R

R. Lyons, R. Pemantle, and Y. Peres, 1995. Conceptual proofs of L L criteria for mean behavior of branching processes. Ann. Probab., 23 0 (3): 0 1125--1138

work page 1995
[6]

Meckes, 2009

E. Meckes, 2009. On S tein's method for multivariate normal approximation. In High dimensional probability V : the L uminy volume , volume 5 of Inst. Math. Stat. (IMS) Collect., pages 153--178. Inst. Math. Statist., Beachwood, OH

work page 2009
[7]

o z and A. R \

E. A. Pek \"o z and A. R \"o llin, 2011. New rates for exponential approximation and the theorems of R \'e nyi and Y aglom. Ann. Probab., 39 0 (2): 0 587--608

work page 2011
[8]

Pike and H

J. Pike and H. Ren, 2014. Stein's method and the L aplace distribution. ALEA Lat. Am. J. Probab. Math. Stat., 11 0 (1): 0 571--587

work page 2014

Show all 13 references

[9]

S. T. Rachev and L. R\" u schendorf, 1998. Mass transportation problems. V ol. I . Probability and its Applications (New York). Springer-Verlag, New York. Theory

1998
[10]

Rai c , 2018

M. Rai c , 2018. A multivariate central limit theorem for L ipschitz and smooth test functions. Preprint https://arxiv.org/abs/1812.08268

2018 arXiv
[11]

V. A. Vatutin and A. M. Zubkov, 1985. Branching processes. I . In Probability theory. M athematical statistics. T heoretical cybernetics, V ol.\ 23 , Itogi Nauki i Tekhniki, pages 3--67, 154. Akad. Nauk SSSR, Vsesoyuz. Inst. Nauchn. i Tekhn. Inform., Moscow

1985
[12]

Villani, 2009

C. Villani, 2009. Optimal transport, volume 338 of Grundlehren der Mathematischen Wissenschaften [Fundamental Principles of Mathematical Sciences]. Springer-Verlag, Berlin. Old and new

2009
[13]

A. M. Yaglom, 1947. Certain limit theorems of the theory of branching random processes. Doklady Akad. Nauk SSSR (N.S.), 56: 0 795--798

1947

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.

[1] [1]

D. A. Dawson, 2017. Introductory lectures on stochastic population systems. Technical report. Available at https://arxiv.org/abs/1705.03781

arXiv 2017

[2] [2]

Geiger, 2000

J. Geiger, 2000. A new proof of Y aglom's exponential limit law. In Mathematics and computer science ( V ersailles, 2000) , Trends Math., pages 245--249. Birkh\" a user, Basel

work page 2000

[3] [3]

S. Kotz, T. J. Kozubowski, and K. Podg\'orski, 2001. The L aplace distribution and generalizations . Birkh\"auser Boston, Inc., Boston, MA. A revisit with applications to communications, economics, engineering, and finance

work page 2001

[4] [4]

S. P. Lalley and X. Zheng, 2011. Occupation statistics of critical branching random walks in two or higher dimensions. Ann. Probab., 39 0 (1): 0 327--368

work page 2011

[5] [5]

Lyons, R

R. Lyons, R. Pemantle, and Y. Peres, 1995. Conceptual proofs of L L criteria for mean behavior of branching processes. Ann. Probab., 23 0 (3): 0 1125--1138

work page 1995

[6] [6]

Meckes, 2009

E. Meckes, 2009. On S tein's method for multivariate normal approximation. In High dimensional probability V : the L uminy volume , volume 5 of Inst. Math. Stat. (IMS) Collect., pages 153--178. Inst. Math. Statist., Beachwood, OH

work page 2009

[7] [7]

o z and A. R \

E. A. Pek \"o z and A. R \"o llin, 2011. New rates for exponential approximation and the theorems of R \'e nyi and Y aglom. Ann. Probab., 39 0 (2): 0 587--608

work page 2011

[8] [8]

Pike and H

J. Pike and H. Ren, 2014. Stein's method and the L aplace distribution. ALEA Lat. Am. J. Probab. Math. Stat., 11 0 (1): 0 571--587

work page 2014