{"id":"dfa223a1-96f6-48bd-9297-a6d4e47fb08d","arxiv_id":"2412.13672","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For finite-type inhomogeneous random graphs, component-density fluctuations converge to a Gaussian process solving an explicit infinite-dimensional SDE, yielding CLTs for the giant component and MST weight.","lead":"This paper proves functional central limit theorems for fluctuations of component densities in inhomogeneous random graphs with finitely many types, and derives CLTs for giant-component size, surplus, and minimum spanning tree weight. The results extend known Erdős-Rényi results to a broader class of random graphs and resolve a conjecture on infinite-dimensional fluctuation limits.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Supercritical tightness in Theorem 3.8(b) rests on Proposition 6.18, whose proof is omitted; without it, the joint FCLTs for macroscopic functionals (Theorems 3.9 and 3.10) are not established.","rationale":"The reader's weakest_assumption focused on Assumption 3.1(b,c), which is an explicit hypothesis rather than an internal proof gap; the reader's rationale did note the 'sketched supercritical tightness proof' and the 'omitted proof of part of Theorem 3.8(b)'. I agree with those specific comments and single out the omitted supercritical tightness as the most load-bearing because it is internal, essential, and currently a gap. The other candidate concerns are real but less decisive: the data-dependent centering Kn(κ) in Theorem 3.13 is a limitation explicitly acknowledged in Remark 9, but the theorem as stated remains a valid (if weaker) conditional CLT; the reliance on the unpublished MDP result [46] in Lemma 8.8 is external and could be verified separately, and it affects only the MST application, not the core FCLT. The supercritical tightness, by contrast, is needed for the main theorem that generates the macroscopic FCLTs. If Proposition 6.18 cannot be proved as stated, the central claim fails at a foundational step. Since the reader already assigned CONDITIONAL, my read does not change the verdict; it sharpens the reason. A concrete analytical check—writing out the Aldous-Kurtz verification for the supercritical interval—would settle whether the concern lands.","tokens_in":64547,"tokens_out":13994,"duration_ms":119720,"concrete_test":"Provide a complete proof of Proposition 6.18: verify the Aldous-Kurtz condition (A) for the supercritical processes {X_n, M^c_n} using the decomposition (6.57), the estimates of Theorem 6.1, Lemma 6.5, and Theorem 6.7 on I=[T1,T2], and an explicit treatment of the initial value X_n(T1) in the drift bound (6.45). Confirm that the bound holds uniformly in t in [T1,T2] when the process is started at time T1 and that the initial value does not contribute to the modulus-of-continuity estimate. If this verification cannot be completed, Theorem 3.8(b) and its corollaries (Theorems 3.9 and 3.10) are not established.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The infinite-dimensional FCLT for the supercritical regime is Theorem 3.8(b), and its proof requires tightness of X_n^{T_{M log n}} in D([T1,T2]:ℓ_{1,δ}). The only place this tightness is claimed is Proposition 6.18 in Section 6.3.2, and its proof is dismissed with 'We omit the details.' This is the most load-bearing gap in the paper: it is the sole support for convergence in the supercritical regime, and all macroscopic results (Theorems 3.9, 3.10) plus the surplus/type-count corollaries inherit this step. The omitted argument is not a routine rerun of the subcritical proof: the Aldous-Kurtz conditions must be verified for the drift A^c_n and the martingale M^c_n on [T1,T2], with the random initial value X_n(T1) at the left endpoint. The subcritical proof (Propositions 6.14 and 6.16) starts from a deterministic initial condition and uses (6.45) together with the moment bounds (Theorem 6.1, Lemma 6.5, Theorem 6.7); in the supercritical case one only has pointwise tightness of X_n(T1) (Proposition 6.17) and must show that the drift contribution over [T1,T2] does not depend on the initial value in a way that breaks the modulus-of-continuity bound. Because the paper does not spell out this verification, the central claim for t > tc is not fully proven as written.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the dynamic finite-type inhomogeneous random graph model of Bollobás–Janson–Riordan, in which edges between vertices of types i and j appear at the points of independent Poisson processes of intensity κ_n(i,j)/n. It introduces the vector of component type densities π_n(t), including the edge density as the zero-coordinate l=0, and proves a functional law of large numbers and finite-dimensional functional central limit theorems for the fluctuations X_n^{T_N}(t). These are lifted to an infinite-dimensional FCLT in the Banach space ℓ_{1,δ} for the truncated vectors X_n^{T_{M log n}}, with the limit characterized as the unique solution of an explicit infinite-dimensional linear SDE driven by a cylindrical Brownian motion; the result is stated separately in the subcritical (t<tc) and supercritical (t>tc) regimes. In the supercritical regime the paper derives joint functional CLTs for the number of components, the size and surplus of the giant component, and the type counts in the giant, and it proves a CLT for the weight of the MST on dense graphon-modulated random graphs with exponential edge weights, with a random centering that converges in probability to a deterministic constant.","tokens_in":64823,"tokens_out":23144,"duration_ms":205767,"significance":"If completed, the paper is a substantial contribution to second-order fluctuation theory for inhomogeneous random graphs. It answers the infinite-dimensional FCLT question raised in Janson's MST paper in the Erdős–Rényi case, takes a step toward Aldous's open problem on Gaussian fluctuations for coagulation systems, and provides functional CLTs for macroscopic observables of the giant in the finite-type IRG. The proof strategy is mostly self-contained: the finite-dimensional FCLT is proved via a Poisson-process semimartingale representation and the Ethier–Kurtz martingale CLT, the ℓ_{1,δ} tightness is based on Aldous–Kurtz conditions and moment bounds, and the SDE coefficients are explicit functions of the deterministic limit π(l,t). External results are used mainly as benchmarks: Bollobás–Janson–Riordan for the LLN and phase transition, [10] for branching-process moment bounds, [46] for the moderate-deviation estimate in Lemma 8.8, and [24] for the MST law of large numbers.","major_comments":[{"comment":"Proposition 6.18 is the sole tightness input for the supercritical infinite-dimensional FCLT in Theorem 3.8(b), and its proof is omitted ('We omit the details'). The supercritical case is not simply a rerun of Propositions 6.14 and 6.16: the Aldous–Kurtz conditions must be verified on [T1,T2] for the drift A_n^c and martingale M_n^c with the random initial value X_n(T1) at the left endpoint, and only pointwise tightness of X_n(T1) is available from Proposition 6.17. Since Theorems 3.9 and 3.10 both depend on Theorem 3.8(b), this omitted step is load-bearing; the proof must be supplied.","section":"§6.3.2, Proposition 6.18"},{"comment":"The proof of the claim in (6.66) is not correct as printed. The displayed bound in that proof has a factor √n multiplying the linear terms π(k2,t)|X_n(k1,t)| and ‖l‖|X_n(l,t)|; after summation over l these terms are of order √n·O_P(1), so the right side of that display does not vanish. The subsequent inequality, which is the one actually used to conclude (6.66), drops those linear terms and retains only ∑_l ‖l‖^{δ+K+1}|X_n(l,t)|^2/√n together with parameter errors. A correct proof needs an explicit Taylor expansion of √n[F_l(π_n,κ_n,µ_n)−F_l(π,κ,µ)] around (π,κ,µ) showing the cancellation of the first-order X_n terms with the Γ(t)X_n(t) part of H_n(t).","section":"§6.4, proof of claim (6.66)"},{"comment":"The final paragraph of Section 6.4 says that the proof of part (b) of Theorem 3.8 'follows along similar lines. We omit the details.' Combined with the omission of Proposition 6.18, this means that the supercritical infinite-dimensional FCLT is not fully written out. Since the supercritical FCLT is the basis for the macroscopic results in Theorems 3.9 and 3.10, the manuscript as it stands does not establish those corollaries, even if the subcritical arguments are sound.","section":"§6.4, proof of Theorem 3.8(b)"}],"minor_comments":[{"comment":"The displayed formula for P(|BP(λ)|<∞) appears to have an extra factor e^{-λ}: the k=1 term should be e^{-λ}, not λ e^{-2λ}. The subsequent exponential bound is still true, but the formula should be corrected.","section":"§8.2, Lemma 8.6(ii)"},{"comment":"The sentence 'As observed in (6.16)' refers to an unnumbered display in the proof of Lemma 6.11; the cross-reference should be fixed.","section":"§8.2, proof of Lemma 8.6(i)"},{"comment":"Because the centering K_n(κ) is a random function of U_n and only converges in probability to K(κ), the statement 'CLT for the weight of the MST' is weaker than a CLT with deterministic centering; this caveat should be stated in the abstract or introduction, not only in Remark 9.","section":"§3.4.2, Theorem 3.13 and Remark 9"},{"comment":"The proof of Proposition 3.3 is dismissed as immediate via the lower-triangular structure; since this proposition underpins the LLN, a brief verification of uniqueness would improve readability.","section":"§3.1, Proposition 3.3"}],"recommendation":"major_revision","confidential_remarks":"The main risk is the omitted supercritical proof: Proposition 6.18 and the final paragraph of Section 6.4 leave Theorem 3.8(b) unproved as written, and all downstream supercritical results inherit that gap. The flawed-looking estimate in the proof of (6.66) also needs a careful rewrite. I would recommend asking for a complete proof of the supercritical tightness and convergence before acceptance. Note also that the proof of Lemma 8.8 relies on the concurrent preprint [46]; this dependency should be stated explicitly in the paper."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's my take. This paper is a real contribution: it extends the Pittel–Janson fluctuation program to finite-type inhomogeneous random graphs, gives an infinite-dimensional FCLT for microscopic type densities in ℓ1,δ, and pulls from it joint FCLTs for macroscopic giant observables and a CLT for MST weight on graphon-modulated dense graphs. The finite-dimensional SDE characterization and the moment bounds built on multitype branching process estimates are well organized. If the main theorems are right, they answer Janson's Remark 1.3 and generalize fluctuation theory to finite-type IRGs. That's a strong paper.\n\nThe soft spot is what the stress-test note says: Proposition 6.18, the only proof of C-tightness in the supercritical regime, is asserted with \"We omit the details.\" It is not a minor omission. It is the sole support for Theorem 3.8(b), and everything in Section 3.3 — Theorems 3.9, 3.10 — inherits it. The subcritical proof starts from a deterministic initial condition; the supercritical one has a random X_n(T1) at the left endpoint, and you need to verify the Aldous–Kurtz conditions for the drift and martingale pieces on [T1,T2] with that random initial value. The paper doesn't do that, so as written the central claim is not fully proven. I don't think this sinks the paper — the strategy is plausible and the ingredients are there — but it is load-bearing and the authors need to fill it in or the referee needs to verify it.\n\nOther, smaller issues: Theorem 3.13 is a CLT with a data-dependent centering K_n(κ), which the authors acknowledge in Remark 9; it's still a result, but weaker than Janson's. Lemma 8.8 leans on an unpublished MDP preprint [46]; that's fine if it checks out, but it's another dependency. Assumption 3.1(b,c) is a real rate condition; if it fails, the Gaussian limit is not the one described.\n\nWho should read this? Probabilists working on random graph fluctuations, coagulation systems, and MST asymptotics. It deserves a serious referee. My recommendation: send to peer review, but flag Proposition 6.18 explicitly and require the proof (or a detailed sketch) before publication.","headline":"Load-bearing gap in the supercritical tightness proof (Prop 6.18) keeps this otherwise strong paper from being fully established as written.","tokens_in":65427,"tokens_out":2732,"would_cite":true,"duration_ms":25563,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60K35","05C80"],"pacs":[],"model":"deepseek-v4-flash","headline":"One Gaussian SDE governs the √n fluctuations of component densities in inhomogeneous random graphs.","keywords":["inhomogeneous random graphs","functional central limit theorems","multi-type branching processes","giant component","surplus","minimum spanning tree","graphons","Gaussian processes"],"falsifier":"Compute, in the single-type random graph with edge probability $1-\\exp(-t/n)$, the scaled variance of the number of components over a fixed supercritical interval; the paper's Theorem 3.11 predicts a specific Gaussian covariance, for instance the giant-size variance $\\rho(1-\\rho)[1-t(1-\\rho)]^{-2}$, where $\\rho$ is the survival probability of a Poisson($t$) branching process. If a simulation or exact calculation at $\\sqrt{n}$ scale shows a different variance or non-Gaussian limits, the SDE characterization fails. More directly, exhibit a sequence of kernels $\\kappa_n$ with $\\kappa_n\\to\\kappa$ pointwise but $\\sqrt{n}(\\kappa_n-\\kappa)$ not convergent; under Assumption 3.1(a) alone the paper gives no CLT, so a normal limit of $\\sqrt{n}(\\pi_n(l,t)-\\pi(l,t))$ with covariance depending on the oscillation would refute the claimed universality.","tokens_in":64301,"feed_emoji":"📈","tokens_out":8066,"duration_ms":76415,"temperature":0.7,"pith_summary":"This paper seeks to prove that the component census of a sparse inhomogeneous random graph with finitely many vertex types is Gaussian at √n scale, as a process in time. The main object is the vector of densities of connected components of each type-composition, tracked through a dynamic construction where each potential edge has an independent clock. The paper shows that, after centering by the law-of-large-numbers limit and truncating component sizes to O(log n), these fluctuations converge to the unique solution of an infinite-dimensional linear stochastic differential equation in a weighted ℓ¹ space. From that microscopic limit it derives joint functional central limit theorems for macroscopic observables of the giant component—number of components, size, surplus, and type counts—and a central limit theorem for the weight of the minimum spanning tree on dense graphon-modulated random graphs with independent exponential edge weights. A sympathetic reader should read this as a unified second-order fluctuation theory for inhomogeneous random graphs, with explicit Gaussian covariance formulas.","feed_headline":"Random graph components converge to one Gaussian process","feed_subtitle":"A single infinite-dimensional SDE describes fluctuations of component types, giant size, surplus and MST weight.","key_machinery":"The carrying object is the component type-density vector π_n(t)=(π_n(l,t): l∈T), where π_n(l,t) is the density (number per vertex, divided by ‖l‖) of connected components whose type composition is the vector l∈N^K. The limit is studied in the Banach space ℓ_{1,δ} of infinite vectors with weighted norm ‖z‖_{1,δ}=|z_0|+Σ_{l≠0}‖l‖^δ|z_l|, after truncating to sizes at most M log n; this is exactly the scale at which all non-giant components live. The argument is carried by a semi-martingale representation of π_n: each possible type-composition merge is driven by rate-one clock processes with compensators given by the quadratic forms θ_κ, and the fluctuation process is decomposed into a drift part and a martingale part. Tightness is proved from first- and second-moment bounds that rest on exponential decay of the deterministic densities π(l,t) away from the critical time t_c, obtained by viewing them as probabilities of a dual multi-type branching process; this decay is what makes the infinite sums and operators well behaved. The limit SDE's drift operator Γ(t) is lower-triangular in the ordering by ‖l‖, which gives pathwise uniqueness, and its diffusion coefficient is the square root of an explicit trace-class operator Φ(t), so the Gaussian limit is uniquely characterized and its mean and covariance can be computed in closed form.","core_discovery":"The central claim is Theorem 3.8: for a finite-type inhomogeneous random graph with type measure µ and kernel κ satisfying second-order convergence of √n(µ_n−µ) and √n(κ_n−κ), the scaled microscopic type-density fluctuations $X_n^{{T_{M log n}}$}(t)=√n(π_n(l,t)−π(l,t)), l∈T_{M log n}, converge in the space of right-continuous paths with left limits to the unique solution of the linear SDE dV(t)=[a(t)+Γ(t)V(t)]dt+G(t)dB(t), separately on subcritical intervals [0,T] with T<t_c and on supercritical intervals [T_1,T_2] with t_c<T_1<T_2. Here a(t) encodes the kernel and type-measure fluctuation parameters, Γ(t) is a lower-triangular convolution operator built from the deterministic component densities, and G(t) is the square root of a trace-class covariance operator Φ(t). The limit is conditionally Gaussian given the initial type fluctuation, with explicit mean and covariance formulas given in Proposition 3.7. The paper then uses this limit process to represent, as continuous linear functionals of X, the fluctuations of the number of components, the size and surplus of the giant, and the type counts inside the giant; these yield Theorems 3.9 and 3.10. In the dense-graph setting the same machinery gives Theorem 3.13: for the percolated graphon model, √n(W_n−K_n(κ)) converges to N(0,σ_∞), where K_n(κ) is a data-dependent centering converging to K(κ).","pith_inferences":["The same infinite-dimensional SDE is likely the correct fluctuation object for bounded-size rule and coalescent models whose laws of large numbers are governed by similar lower-triangular coagulation equations; the paper's operators give a template for those central limit theorems, though this is not established here.","The MST central limit theorem should extend to edge-weight distributions beyond exponential, as long as the density near zero is controlled, because the proof uses only the threshold representation of component counts; the paper states this belief but does not prove it.","A numerical check of σ_∞ for a two-type kernel would test the rate of convergence of the truncation σ_k→σ_∞, which the branching-process decay suggests should stabilize quickly for moderate k.","The finite-type result is likely a stepping stone for infinite type spaces by finite-type approximation, one consequence being functional central limit theorems for degree-corrected or configuration-type graph models, but the current paper only develops the finite-type tools."],"forward_implications":["For any fixed supercritical time interval, the size, surplus, number of connected components, and per-type vertex counts of the giant converge jointly, after √n scaling, to continuous linear functionals of one common Gaussian process.","In the subcritical regime, the number of components has a functional central limit theorem with limit equal to the weighted sum of the same type-density limit process.","Percolation on a dense finite-type graphon, when the underlying graphon sequence has √n-convergent fluctuations, inherits the joint central limit theorems for the giant component and the number of components.","For dense random graphs with exponential edge weights, the minimum spanning tree weight fluctuates at √n scale around a data-dependent centering that converges to K(κ), with a finite variance σ_∞ given by integrated covariances of the microscopic limit process.","For the classical single-type random graph, the general theorems specialize to a three-dimensional joint functional central limit theorem for component count, giant size, and surplus, recovering and extending fixed-time results."],"supporting_citations":[{"why":"supplies the phase-transition law of large numbers, the branching-process representation of limit densities, and the sub/supercritical regime that the CLT refines.","marker":"[13]"},{"why":"provides the MST-weight representation as an integral of component counts and the functional-CLT strategy for component densities that the MST theorem extends.","marker":"[25]"},{"why":"introduces the weighted ℓ_{1,δ} space and the component-census approach to giant-component fluctuations that the macroscopic theorems build on.","marker":"[37]"},{"why":"supplies the infinite system of differential equations for the limit component densities and the connection to inhomogeneous coagulation used for the deterministic limit.","marker":"[3]"},{"why":"gives the asymptotic equivalence between different IRG formulations used to transfer sparse results to dense graphon percolation.","marker":"[26]"},{"why":"provides the Markov-process convergence theorem used to prove the finite-dimensional functional central limit theorem.","marker":"[22]"},{"why":"supplies the infinite-dimensional martingale representation and Hilbert-Schmidt stochastic integral theory used to identify the SDE limit.","marker":"[18]"},{"why":"gives the law of large numbers for MST weight on dense graph sequences, including convergence of the centering to K(κ).","marker":"[24]"}],"fun_headline_variants":["Giant component fluctuations follow one Gaussian SDE","Inhomogeneous random graphs share a single Gaussian limit","One SDE for type densities, giant size, and MST weight","Functional CLTs unify micro and macro random graph limits","Fluctuations in random graphs: one Gaussian process for all"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole Gaussian picture collapses if the kernel fluctuations √n(κ_n−κ) do not converge pointwise to a symmetric limit matrix while the type-measure fluctuations converge; the theorem's drift, variance, and hence every CLT it derives are all built on that convergence.","fun_headline_variants_meta":{"raw":{"variants":["Giant component fluctuations follow one Gaussian SDE","Inhomogeneous random graphs share a single Gaussian limit","One SDE for type densities, giant size, and MST weight","Functional CLTs unify micro and macro random graph limits","Fluctuations in random graphs: one Gaussian process for all"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000232,"raw_usage":{"total_tokens":1543,"prompt_tokens":1051,"completion_tokens":492,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":667,"completion_tokens_details":{"reasoning_tokens":411}},"tokens_in":667,"tokens_out":492,"duration_ms":4495,"temperature":1.0,"reasoning_tokens":411,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T12:54:46.318416+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute, in the single-type random graph with edge probability $1-\\exp(-t/n)$, the scaled variance of the number of components over a fixed supercritical interval; the paper's Theorem 3.11 predicts a specific Gaussian covariance, for instance the giant-size variance $\\rho(1-\\rho)[1-t(1-\\rho)]^{-2}$, where $\\rho$ is the survival probability of a Poisson($t$) branching process. If a simulation or exact calculation at $\\sqrt{n}$ scale shows a different variance or non-Gaussian limits, the SDE characterization fails. More directly, exhibit a sequence of kernels $\\kappa_n$ with $\\kappa_n\\to\\kappa$ pointwise but $\\sqrt{n}(\\kappa_n-\\kappa)$ not convergent; under Assumption 3.1(a) alone the paper gives no CLT, so a normal limit of $\\sqrt{n}(\\pi_n(l,t)-\\pi(l,t))$ with covariance depending on the oscillation would refute the claimed universality.","supporting_citations":[{"cited_title":"Bollob ´as, Random Graphs, Second, Cambridge Studies in Advanced Mathematics, vol","cited_arxiv_id":null,"evidence_quote":"supplies the phase-transition law of large numbers, the branching-process representation of limit densities, and the sub/supercritical regime that the CLT refines."},{"cited_title":"M Frieze, On the value of a random minimum spanning tree problem , Discrete Applied Mathematics 10 (1985), no","cited_arxiv_id":null,"evidence_quote":"provides the MST-weight representation as an integral of component counts and the functional-CLT strategy for component densities that the MST theorem extends."},{"cited_title":"Molloy and B","cited_arxiv_id":null,"evidence_quote":"introduces the weighted ℓ_{1,δ} space and the component-census approach to giant-component fluctuations that the macroscopic theorems build on."},{"cited_title":"Andreis, W","cited_arxiv_id":null,"evidence_quote":"supplies the infinite system of differential equations for the limit component densities and the connection to inhomogeneous coagulation used for the deterministic limit."},{"cited_title":"Random minimum spanning tree and dense graph limits","cited_arxiv_id":"2310.11705","evidence_quote":"gives the asymptotic equivalence between different IRG formulations used to transfer sparse results to dense graphon percolation."},{"cited_title":"Enriquez, G","cited_arxiv_id":null,"evidence_quote":"provides the Markov-process convergence theorem used to prove the finite-dimensional functional central limit theorem."},{"cited_title":"Ethier and T","cited_arxiv_id":null,"evidence_quote":"gives the law of large numbers for MST weight on dense graph sequences, including convergence of the centering to K(κ)."}],"review_version":1}