REVIEW 3 major objections 5 minor 29 references
Normal approximation for subgraph counts in age-dependent random connection models
T0 review · 3 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash
Pith's one-line read In the age-dependent random connection model, normalized clique and directed-tree counts are asymptotically normal whenever the typical degree has finite second moments, with explicit Wasserstein rates for cliques and a sharp leaf-count…
desk verdict Strong technical paper giving the Gaussian side of the phase transition for subgraph counts in the ADRCM; the clique CLT is solid, but the subtree CLT rests on a covariance decay lemma whose proof is not fully self-contained. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
For clique counts the engine is the Malliavin–Stein method adapted to Poisson functionals whose fourth moments are not uniformly bounded: first- and second-order difference operators $D_q C_{n,k}$ and $D_{u,q} C_{n,k}$ are controlled by moment bounds such as $\mathbb{E}[(D_{u,q}C_{n,k})^a]\le C(1_{\{q\in N^{\mathrm{up}}(u)\}}+v^{-a\gamma}\hat{s}(u,|y|)^{a\gamma})$ with $\hat{s}(u,r)=1\wedge(u^\gamma r)^{-1/(1-\gamma)}1_{\{r\le 2u^{-1}\}}$, reflecting the asymmetric structure in which the up-neighbourhood of a point with mark $u$ is Poisson of parameter of order $u^{-\gamma}$ while the down-neighbourhood has bounded mean. Covariance asymptotics come from a Mecke-formula decomposition into one-vertex and two-vertex contributions, with torus-to-line boundary corrections controlled to order $n^{-\zeta(\gamma)}$. For subtree counts the machinery is different: the count is split into spatially indexed slab variables $T_i^{(n)}$, which are positively associated by the Harris–FKG inequality, and the proof reduces to summability of the Cox–Grimmett coefficient $u_n(k)$; the key input is the decay $\mathrm{Cov}(T_1^{(n)},T_k^{(n)})=O(k^{-(2-2\ell\gamma)})$ derived in Section 6 from combinatorial leaf-count relations (24)–(26). Newman's inequality then bounds the characteristic-function error, and a Lindeberg check uses the moment bounds of Proposition 16.
What would settle it
Simulate or compute $\mathrm{Cov}(T_1^{(n)},T_k^{(n)})$ for a directed tree with $\ell=2$ leaves at $\gamma$ slightly below $1/4$: Lemma 7(i) requires a tail of order $k^{-(2-2\ell\gamma)}$ up to logarithmic factors, so a Monte Carlo estimate showing decay $\Omega(k^{-(2-2\ell\gamma)+\delta})$ for some $\delta>0$ at large $k$ and $n$ would falsify the lemma, break the summability of the coefficient in (11), and remove the foundation of Theorem 2, while decay at least that fast would support the exponent the proof relies on.
Extended reading notes
Core claim
The central claim is that in the regime where a typical vertex has a neighbourhood of finite $(2+\varepsilon)$-th moment, subgraph fluctuations occur at scale $\sqrt{n}$ and are Gaussian with explicit limiting variances. Theorem 1 states that for $0<\gamma<1/2$, each normalized clique count $\widehat{C}_{n,k}$ converges to $N(0,1)$, with $|n^{-1}\mathrm{Var}(C_{n,k})-\sigma_k|=O(n^{2\gamma-1})$ for an explicit $\sigma_k>0$; the multivariate version gives joint convergence to $N(0,\Sigma)$ with rate $O(n^{-(\zeta(\gamma)\wedge(\eta-1)/2)})$ where $\zeta(\gamma)=(1-\gamma)\wedge(1-2\gamma)/\gamma$. Theorem 2 states that for a directed tree $T$ with $\ell$ leaves, whenever $0<\gamma<1/(2\ell)$, the normalized subtree count converges in distribution to $N(0,1)$ and its variance is of exact linear order in $n$. The thresholds are not artifacts: above $\gamma=1/2$ for cliques and $\gamma=1/(2\ell)$ for trees, the same statistics obey stable limit theorems, so the two results together describe a Gaussian-to-stable phase transition.
Load-bearing premise
The subtree result stands on the claim, proved by the case analysis of Section 6, that the covariance between rooted subtree counts in torus slabs separated by $k$ decays like $k^{-(2-2\ell\gamma)}$ up to logarithmic factors, which the proof discards; if the true decay were slower than that by more than a logarithmic factor near the threshold $\gamma=1/(2\ell)$, the Cox–Grimmett coefficients in (11) would cease to be summable and the associated-variable argument for Theorem 2 would break.
Editorial extensions
If this is right
- For $0<\gamma<1/2$, clique counts in the ADRCM fluctuate normally at scale $\sqrt{n}$ with explicitly computed limiting variances $\sigma_k>0$, and joint convergence holds for cliques of several sizes with the covariance matrix identified in Proposition 4.
- Directed-tree counts are normal whenever $0<\gamma<1/(2\ell)$ for a tree with $\ell$ leaves, with variance of exact linear order in the domain size, and no uniform fourth-moment assumption is needed.
- Together with the companion stable-limit paper, the results locate a Gaussian-to-stable phase transition for these statistics: at $\gamma=1/2$ for cliques and at $\gamma=1/(2\ell)$ for trees.
- The normality results supply the theoretical basis for statistical inference with the ADRCM, namely asymptotic confidence intervals and hypothesis tests for subgraph-based summaries such as wedge and triangle counts.
- The paper conjectures that the clustering coefficient — the ratio of wedge counts to triangle counts — converges to a Cauchy-type limit of the form $a+bZ_1/Z_2$ with $Z_1,Z_2$ independent standard normals in the light-tailed regime.
Reading between the lines
- The covariance-decay exponent $2-2\ell\gamma$ is forced by leaf bookkeeping rather than by spatial dimension, so the threshold $\gamma=1/(2\ell)$ and the subtree CLT are plausible in dimensions $d>1$ and for unbounded-support profiles as well; the paper itself flags these extensions as open in Remark 3, which makes them a concrete testable route.
- The rate $O(n^{-(\eta-1)/2})$ in Theorem 1 is tied to the auxiliary parameter $\eta$ whose admissible range $\eta(2\gamma\vee(1-\gamma))<1$ shrinks as $\gamma$ grows; in the sub-regime $\gamma<1/4$ the paper suspects sharper rates exist, and the variance-error term $n^{2\gamma-1}$ is a plausible bottleneck worth optimizing against the constants of Proposition 5.
- The paper proves no rate in Theorem 2, and a quantitative subtree CLT would have to turn the covariance decay of Lemma 7(i) into an explicit block-size choice; the block argument in Section 6 shows any such rate would be governed by how rapidly the block length can grow against the $(2+\delta)$-moment bound of Proposition 16.
- A Monte Carlo study of the clustering coefficient ratio in the light-tailed regime could test the Cauchy-type conjecture directly; a visible risk is that the denominator's fluctuations are not bounded away from zero, so heavy tails of the ratio may appear even where each component is Gaussian.
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper studies normal approximation for subgraph counts in the one-dimensional age-dependent random connection model (ADRCM). For clique counts and 0<γ<1/2, it proves univariate and multivariate quantitative CLTs with explicit rates using Trauthwein's p-Poincaré inequalities on Poisson space, together with variance/covariance asymptotics giving n^{-1}Cov(C_{n,k},C_{n,ℓ})→σ_{k,ℓ}. For rooted subtree counts of a directed tree T with ℓ leaves, it proves a qualitative CLT in the regime γ<1/(2ℓ) by decomposing the counts over spatial slabs, establishing positive association, and using the Cox–Grimmett coefficient. The main technical content is a long series of moment and covariance estimates for first- and second-order difference operators and for rooted-subtree counts, with many case splits. Theorems 1 and 2 complement the stable-limit results in the companion paper [9] and together describe a Gaussian-to-stable phase transition for these statistics.
Significance. If the main estimates are correct, the paper is a valuable contribution: it gives the first normal approximation results for ADRCM subgraph counts in the light-tailed regime where only (2+ε) moments are available, and the clique result is quantitative. The variance asymptotics and the explicit phase-transition picture with [9] are interesting, and the proof strategy is clearly organized, with most auxiliary lemmas proved in detail. However, the subtree CLT in Theorem 2 rests on one covariance decay estimate — Lemma 7(i) — whose proof is partly imported from companion preprints and partly presented through representative cases. Because that estimate is the load-bearing step for the Cox–Grimmett bound (11), the paper is not yet fully self-contained, and the central claim of Theorem 2 must be treated as conditional until the missing details are supplied.
major comments (3)
- [Section 6, Lemma 7(i) and Eq. (11)] The covariance decay Cov(T_1^{(n)}, T_k^{(n)}) = O(k^{-(2-2ℓγ)}) is the load-bearing estimate for Theorem 2, because it yields the Cox–Grimmett bound (11) with the positive margin 1-2ℓγ. The proof of Lemma 7(i) imports the reduction to overlap size one from [9, Proposition 2.2] and the remainder R_k^{(n)} from [8, Lemma 10], and the case analysis after (27) is not complete: the text states that Cases (II) and (IV) are 'almost identical' and omits their details, and logarithmic factors are suppressed by assertion rather than by an explicit argument. Since the margin 1-2ℓγ tends to zero at the boundary γ=1/(2ℓ), a small unverified polynomial factor in the covariance decay would destroy the summability in (11) and invalidate Theorem 2. Please provide a complete, self-contained proof of Lemma 7(i), or state Theorem 2 as conditional on the companion results with precise references to the exact statements used.
- [Section 6, Eqs. (24)–(26) and (29)–(36)] The final exponent 2-2ℓγ in Lemma 7(i) is obtained by using the leaf-count relations (24)–(26) as equalities. The manuscript does not demonstrate that every possible configuration of the two copies T1 and T2, the common node, and the leaf sets falls into exactly one of (24)–(26), nor does it show that the omitted cases and the absorbed logarithmic factors cannot change the k-exponent. Because the estimate is used at a threshold where the margin is arbitrarily small, the proof should explicitly enumerate all combinatorial possibilities or provide a uniform argument that rules out any polynomial slow-down.
- [Section 6, proof of Lemma 7(ii)] In the variance lower bound, the step 'liminf_n VarpT_1^{(n)}q ≥ ∫_0^1 μ_T(u)^2 du > 0' uses Fatou's lemma, but the convergence μ_{n,T}(u) → μ_T(u) is not proved. This convergence is plausible because for each fixed u the range of the ADRCM is finite and the torus neighborhoods stabilize as n→∞, but as written it is a gap in an argument that is needed for the linear lower bound on Var(T_{n,T}) used in Theorem 2. Please add the missing stabilization argument.
minor comments (5)
- [Title and abstract] There is a typo in the title ('Approxima tion') and the phrase 'age-dependent random connection model' is repeated in the abstract; please edit.
- [Section 6, proof of Lemma 7(i), display before Eq. (29)] The sentence at Eq. (28)–(29) stating that the displayed integral is 'of order O(k^{-(2-2ℓγ)}) uniformly over n' appears before the proof of that bound; the proof later gives upper bounds, and the text should clarify that these are estimates rather than equalities.
- [Section 6, proof of Theorem 2, Eq. (39)] In the second displayed estimate after (39), the bound ∑_{q=ℓ_n+1}^n Cov(T_p^{(n)}, T_q^{(n)}) is written with u_n(p), but the argument of u_n should be the block distance ℓ_n+1-p; the displayed index is only correct after a reindexing. Please clarify this step to avoid a mistaken reading.
- [Proposition 16] Part (i) of Proposition 16 is cited to [9, Proposition 2.1] rather than proved in this paper; the proposition statement should explicitly say so, since this is the basis of the moment bounds used in Theorem 2.
- [Theorem 1(b) and Proposition 4] The statement that σ_{k,k}>0 is asserted without proof. The authors say it is 'easily established'; a short justification would improve rigor, as this positivity enters the normalization in the CLT.
Circularity Check
No significant circularity: the CLT conclusions are not used as inputs; the same-author citations provide supporting moment and remainder estimates only.
full rationale
Walking the derivation chain, the two main theorems are obtained from external machinery: Theorem 1 uses Trauthwein's second-order p-Poincare inequalities [25,26], with the variance/covariance asymptotics and Gamma_i bounds proved in Sections 4 and 5 from the ADRCM's neighborhood-size lemmas; Theorem 2 uses Oliveira's CLTs for associated variables [19], with the Cox-Grimmett coefficient (10)-(11) controlled by Lemma 7. Lemma 7(i) is proved in the manuscript by a lengthy case analysis in Section 6, and the leaf-count identities (24)-(26) are purely combinatorial bookkeeping rather than a restatement of the CLT. The same-author citations that do appear are technical rather than conclusion-bearing: Proposition 16(i) quotes the moment bound from [9, Proposition 2.1], the overlap-size reduction is imported from [9, Proposition 2.2], and the remainder term R_k^(n) is deferred to [8, Lemma 10]. None of these cited results assumes the normal approximation being proved, and the paper's own computations supply the variance bounds, covariance decay, and Lindeberg verification. The skeptical concern that the exponent 2-2*ell*gamma in Lemma 7(i) absorbs logarithmic factors and relies on combinatorial identities, with margin 1-2*ell*gamma shrinking near gamma=1/(2*ell), is a legitimate correctness-risk flag, but it is not circularity: no fitted parameter is renamed as a prediction, no definition builds in the target result, and no uniqueness claim from the authors' prior work is used to force the choice. The central claims are therefore self-contained modulo external theorems, with only minor self-citation in supporting estimates, warranting a low score.
Assumptions & free parameters
assumptions (8)
- domain assumption Unit-intensity Poisson point process on the torus T_n x [0,1]; edges are formed with probability phi(|X-Y|/(beta U^gamma V^{1-gamma})).
- domain assumption Restriction to d=1, hard profile phi(r)=1{r<=1}, and (often) beta=1.
- domain assumption Lemma 8: up- and down-neighborhood counts of a vertex are independent Poisson variables with parameters c_+ u^{-gamma} and c_-; cited from [6, Proposition 4.1].
- standard math Poisson functional CLT bounds of Trauthwein [25, Theorem 1] and [26, Theorem 3.4] hold for functionals with finite 2+epsilon moments; used as black boxes.
- standard math CLTs for positively associated random variables of Oliveira [19, Theorems 4.1 and 4.8] and Newman's inequality [17].
- domain assumption Companion paper [9] supplies stable limit theorems and the mean bound mu_{n,T}(u) <= C u^{-ell*gamma-eta} (Proposition 16(i)).
- domain assumption The remainder term R_k^{(n)} in equation (22) is O(k^{-(2-2*ell*gamma)}) as established in [8, Lemma 10].
- standard math Mecke formula for Poisson point processes and Harris-FKG inequality for positively associated Poisson variables.
Cite this review
Pith. "Pith review of Normal approximation for subgraph counts in age-dependent random connection models." pith.science (2026). https://pith.science/paper/ASRMQMWT
@misc{pith2026250509318,
author = {Pith},
title = {Pith review of: Normal approximation for subgraph counts in age-dependent random connection models},
year = {2026},
howpublished = {\url{https://pith.science/paper/ASRMQMWT}},
note = {Machine review of arXiv:2505.09318}
}
abstract
We study normal approximation of subgraph counts in a model of spatial scale-free random networks known as the age-dependent random connection model. In the light-tailed regime where only moments of order $(2 + \varepsilon)$ are finite, we study the asymptotic normality of both clique and subtree counts. For clique counts, we establish a multivariate quantitative normal approximation result through the Malliavin-Stein method. In the more delicate case of subtree counts, we obtain distributional convergence based on a central limit theorem for sequences of associated random variables.
Figures
Reference graph
Works this paper leans on
-
[9]
Limit theorems under heavy-tailed scenario in the age dependent random connection models
C. Hirsch and T. Owada. Limit theorems under heavy-tailed scenario in the age dependent random connection models. arXiv preprint arXiv:2409.05226 , 2024
work page Pith review arXiv 2024
- [1]
-
[2]
Y. Baryshnikov and J. E. Yukich. Gaussian limits for random measures in geometric probability. Ann. Appl. Probab. , 15(1A):213–253, 2005
work page 2005
-
[3]
J. T. Cox and G. Grimmett. Central limit theorems for associated random variables and the percolation model. Ann. Probab., 12:514–528, 1984
work page 1984
- [4]
- [5]
- [6]
-
[7]
L. Gugelmann, K. Panagiotou, and U. Peter. Random hyperbolic graphs: degree sequence and clustering. In Automata, Languages, and Programming, volume 7392 of Lecture Notes in Computer Science , pages 573–585. Springer, 2012
work page 2012
Show all 29 references
-
[8]
Hirsch and P
C. Hirsch and P. Juhasz. On the topology of higher-order age-dependent random connection models. Method. Comput. Appl. Probab., 2025, forthcoming
2025
-
[10]
Jacob and P
E. Jacob and P. M¨ orters. Spatial preferential attachment networks: power laws and clustering coefficients. Ann. Appl. Probab., 25(2):632–662, 2015
2015
-
[11]
Jacob and P
E. Jacob and P. M¨ orters. Robustness of scale-free spatial networks. Ann. Probab., 45(3):1680–1722, 2017
2017
-
[12]
Komj´ athy and B
J. Komj´ athy and B. Lodewijks. Explosion in weighted hyperbolic random graphs and geometric inhomogeneous random graphs. Stochastic Process. Appl. , 130(3):1309–1367, 2020
2020
-
[13]
Lachi` eze-Rey and G
R. Lachi` eze-Rey and G. Peccati. New Berry-Esseen bounds for functionals of binomial point processes. Annals of Applied Probability, 27:1992–2031, 2017
1992
-
[14]
Lachi` eze-Rey, M
R. Lachi` eze-Rey, M. Schulte, and J. E. Yukich. Normal approximation for stabilizing functionals. Ann. Appl. Probab. , 29(2):931–993, 2019
2019
-
[15]
G. Last, G. Peccati, and M. Schulte. Normal approximation on Poisson spaces: Mehler’s formula, second order Poincar´ e inequalities and stabilization. Probab. Theory Related Fields , 165(3-4):667–723, 2016
2016
-
[16]
Last and M
G. Last and M. D. Penrose. Lectures on the Poisson Process . Cambridge University Press, Cambridge, 2016
2016
-
[17]
C. M. Newman. Normal fluctuations and the FKG inequalities. Commun. Math. Phys. , 74:119–128, 1980
1980
-
[18]
Norros and H
I. Norros and H. Reittu. On a conditionally Poissonian graph process. Adv. Appl. Prob. , 38(1):59–75, 2006
2006
-
[19]
P. E. Oliveira. Asymptotics for Associated Random Variables . Springer, Berlin, Heidelberg, 2012
2012
-
[20]
M. D. Penrose. Random Geometric Graphs . Oxford University Press, Oxford, 2003
2003
-
[21]
M. D. Penrose and J. E. Yukich. Central limit theorems for some graphs in computational geometry. Ann. Appl. Probab. , 11(4):1005–1041, 2001
2001
-
[22]
Reitzner and M
M. Reitzner and M. Schulte. Central limit theorems for U-statistics of Poisson point processes. Ann. Probab., 41(6):3879– 3909, 2013
2013
-
[23]
Schulte and J
M. Schulte and J. E. Yukich. Multivariate second order Poincar´ e inequalities for Poisson functionals.Electron. J. Probab., 24, 2019
2019
-
[24]
A. N. Shiryaev. Probability, 2nd edition. Springer, New York, 1996
1996
-
[25]
Trauthwein
T. Trauthwein. Multivariate second-order p-Poincar´ e inequalities.arXiv preprint arXiv:2409.02843 , 2024
2024 arXiv
-
[26]
Trauthwein
T. Trauthwein. Quantitative CLTs on the Poisson space via Skorohod estimates and p-Poincar´ e inequalities.To appear in Ann. Appl. Prob. , 2025
2025
-
[27]
van der Hofstad
R. van der Hofstad. Random Graphs and Complex Networks, Volume 2 . Cambridge University Press, 2024
2024
-
[28]
van der Hofstad, P
R. van der Hofstad, P. van der Hoorn, and N. Maitra. Scaling of the clustering function in spatial inhomogeneous random graphs. J. Stat. Phys. , 190(6):Paper No. 110, 43, 2023
2023
-
[29]
J. E. Yukich. Surface order scaling in stochastic geometry. Ann. Appl. Prob , 25:177–210, 2015. (Christian Hirsch) Department of Mathematics, Aarhus University, Ny Munkegade, 118, 8000, Aarhus C, Denmark. Email address : hirsch@math.au.dk (Takashi Owada) Department of Statisti...
2015
Reviewed August 15, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.