{"id":"2c686532-e8a9-4369-bd01-36b93e397021","arxiv_id":"2411.14111","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Generalized preferential attachment models with random out-degrees and possibly negative affinity converge locally to a random Pólya point tree, from which percolation thresholds and Ising critical temperatures are derived.","lead":"This mathematics thesis proves that a broad family of growing random networks, preferential attachment models, locally look like a single universal random tree, and uses that tree to compute percolation and Ising model thresholds. It matters because such networks model the internet, social, and biological systems, and these thresholds govern resilience and phase behavior.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The local convergence proof is written in full only for Model (A); the extension to Models (B) and (D) rests on Remark 4.4.9, which asserts rather than demonstrates that the same density computations transfer. This is the weakest load-bearing step.","rationale":"The reader identified precisely the same load-bearing gap: the local convergence theorem is proved in detail only for Model (A), while Models (B) and (D) are delegated to Remark 4.4.9 with an assertion that the edge-connection probabilities behave similarly. I agree that this is the point where the central claim is least secure. The visible parts of Chapter 3 do give Pólya urn equivalences for all three models, and the position concentration lemma is stated for both PU and CPU, so the gap is not an absence of all tools; rather, it is the unperformed transfer of the delicate density computation. The proposed concrete test targets exactly that transfer: recomputing the vertex- and edge-marked first-moment density directly from the Model (D) and Model (B) edge probabilities and checking that the same limit density and error rates emerge. I am not changing the verdict because the reader's CONDITIONAL judgement already captures this: accept only if the missing details are supplied or if the theorem is restricted to Model (A). The Part II percolation and Ising claims are also unverifiable from the truncated text, but the Model (B)/(D) gap is the more immediate mathematical dependency, since Theorem 4.2.1 is the foundation for applying the local limit to the subsequent stochastic-process results.","tokens_in":69778,"tokens_out":7884,"duration_ms":80380,"concrete_test":"Carry out the Model (D) analogue of Proposition 4.4.5: using only the PU(NSL) edge probabilities (4.3.7), compute the conditional joint density of an edge-marked r-neighbourhood at fixed vertex positions v_omega = ceil(n a_omega), including the full no-further-edge product prod_{u in [n]} prod_{j in [m_u]} [1 - sum_{omega in V^o(t)} p^{(j)}(u, v_omega)], and then sum over edge marks. Verify that the resulting expression equals the right-hand side of (4.4.38) with the same g_{r,t} and a uniform o_P(1) error. Repeat for Model (B) with CPU(NSL) using (4.3.6). If either computation yields a nonvanishing correction or a different factorial factor, Theorem 4.2.1 for Models (B) and (D) is not established by the displayed proof.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Theorem 4.2.1 claims vertex-marked local convergence for Models (A), (B), and (D), but the proof in Section 4.4 is fully carried out only for Model (A) via its equality with CPU(SL). Model (B) is coupled to CPU(NSL) and Model (D) to PU(NSL), where the conditional edge probabilities are (4.3.6) and (4.3.7) instead of (4.3.5). Remark 4.3.6 says the differences are insignificant, and Remark 4.4.9 says the proofs follow from the same calculations, but the displayed argument does not contain those calculations. In particular, the first-moment density proof needs the no-further-edge product in (4.4.29) to have the same exponential asymptotics and the same o_P(1) error term for PU(NSL), including the m_u factors per vertex, and the edge-mark summation in Remark 4.4.7 must produce exactly the factorial and Gamma size-bias factors in (4.4.39). For Model (D) the edge-mark combinatorics differ because each vertex has m_u distinguishable out-edges rather than one edge per collapsed block. If any of these corrections fails to vanish at the right rate, or if a combinatorial factor differs, the convergence in probability for Models (B) and (D) does not follow from the supplied proof.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The thesis studies affine preferential attachment models with i.i.d. random out-degrees M, finite p-th moment for some p>1, and fitness parameter δ > -inf supp(M). Part I defines the random Pólya point tree RPPT(M,δ), proves Pólya urn representations for Models (A), (B) and (D), proves vertex-marked local convergence of these models to the RPPT by a second-moment density calculation, and derives the asymptotic degree distribution and the degree distributions of older and younger neighbours. Part II, according to the introduction and abstract, uses this local limit to compute the critical percolation threshold of the Pólya point tree, transfers it to preferential attachment models via their large-set expander property, and studies the quenched Ising model and its inverse critical temperature. The text supplied for review contains Chapters 1–4 in detail and the table of contents, but Chapters 5–8 are not present in the provided portion.","tokens_in":70014,"tokens_out":5165,"duration_ms":52965,"significance":"If the claims are correct, the paper gives a substantial generalization of the local limit results of Berger et al., covering random out-degrees and negative δ, and identifies a universal limiting object. The explicit Pólya urn representations proved by direct graph-probability matching are a notable technical strength, as is the local density limit theorem, which is strictly stronger than plain local convergence. The size-biasing effects in the limiting tree and in the degree distributions of neighbours are cleanly identified. The percolation and Ising results, if fully verified, would be valuable examples of global phase-transition parameters being determined by local structure. The manuscript is less convincing where it relies on asserted rather than displayed calculations, particularly for Models (B) and (D).","major_comments":[{"comment":"The proof of vertex-marked local convergence is carried out in full only for Model (A). For Models (B) and (D), the manuscript states in Remark 4.4.9 that the proofs follow from the same calculations, but the displayed argument does not contain those calculations. This is load-bearing: the conditional edge probabilities for CPU(NSL) and PU(NSL) in (4.3.6) and (4.3.7) differ from (4.3.5), the no-further-edge product in (4.4.29) has to be recomputed with these probabilities, and the edge-mark summation leading to (4.4.38)–(4.4.39) must reproduce the correct factorial and Gamma size-bias factors. For Model (D), the combinatorics differ because each vertex has m_u distinguishable out-edges rather than one edge per collapsed block. The manuscript should either provide the detailed first- and second-moment density proofs for Models (B) and (D), or state and prove an explicit transfer lemma showing that all error terms o_P(1) and all combinatorial factors are identical to the Model (A) case.","section":"§4.4, Remark 4.4.9 and Theorem 4.2.1"},{"comment":"The claims about percolation and the Ising model are central parts of the thesis, but the provided text contains no statements or proofs from Chapters 5–8. In particular, the claim that the critical percolation threshold equals the inverse of the spectral radius of the mean offspring operator, and the claim that this threshold transfers to preferential attachment models via the large-set expander property, cannot be checked from the submitted material. The same holds for the quenched Ising pressure and the inverse critical temperature. If these chapters are part of the manuscript, they need to be included in the review version; otherwise the thesis is incomplete with respect to its stated central claims.","section":"Chapters 5–8"},{"comment":"The power-law derivations in Section 4.5 depend on analytic tail computations for mixed Poisson distributions with Gamma mixing. The text gives the main formulas, but some steps are compressed: for example, the assertion that the sum in (4.5.39) varies regularly with the stated exponent uses Karamata's theorem without showing that the slowly varying functions satisfy the required uniformity conditions. This is a minor gap relative to the main theorem, but since these degree-distribution results are presented as consequences of Theorem 4.2.1, the proofs should be completed with the standard regularity estimates for slowly varying functions.","section":"§4.5, Lemma 4.5.1 and Theorem 4.5.2"}],"minor_comments":[{"comment":"In the change of variables between the two representations of the Pólya point tree, the line \"Define δ = 2u/m\" appears to be a typo or an unexplained redefinition; the parameter δ is already fixed and should not be redefined in this way.","section":"§2.5"},{"comment":"The paragraph before Theorem 3.5.2 says that Model (D) is equivalent to PU(SL), while the theorem statement and its proof concern PU(NSL). Please correct this inconsistency.","section":"§3.5"},{"comment":"The name \"Pólya\" is repeatedly typeset as \"P'olya\" in the chapter preambles and running text; the accent and spelling should be made consistent.","section":"Throughout Part I"},{"comment":"The sentence \"These graphs has a very rich, but still growing literature\" contains a subject-verb agreement error and should be rewritten.","section":"§1.1.1"},{"comment":"The notation Θ(L(Y)(k)) is not defined; if it is meant to denote a slowly varying function, it should be named consistently as L(Y)(k) without the unexplained Θ symbol.","section":"§4.5, Theorem 4.5.2(b)"}],"recommendation":"major_revision","confidential_remarks":"The strongest part of the thesis is the explicit Pólya urn representation and density computation for Model (A); if the same calculations for Models (B) and (D) are supplied, the local convergence theorem would be convincing. The absence of Chapters 5–8 in the review text is a major obstacle, because the percolation and Ising results are advertised as core contributions. I also note that the chapter headers state that Chapters 2–4 are based on the author's paper [85]; the overlap with that paper should be clearly disclosed in any journal submission. The thesis front matter and acknowledgements are not appropriate for a journal article and should be removed or reduced."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: the part of this thesis I can actually read is a genuine contribution for model (A) — random out-degrees and δ > -inf supp(M) with a detailed Pólya urn representation and local density convergence to the random Pólya point tree. The proof there is honest and substantive; the second-moment argument goes through with real estimates, not hand-waving. That alone makes this worth a careful read.\n\nWhat's new: the RPPT with mixed O/Y labels and size-biased out-degree distributions is a real generalization of Berger et al., and the negative-δ regime matters for applications. The degree-distribution consequences (Lemma 4.5.1, Theorem 4.5.2) follow cleanly from the local limit and match [57], which is a nice consistency check.\n\nWhere it soft: the theorem statement claims local convergence for models (A), (B), (D), but the proof in Section 4.4 is only carried out for (A). Observation 5 says so plainly, and Remark 4.4.9 says the others follow from 'the same calculations.' I believe the same calculations probably do work — the model (B) coupling to CPU(NSL) and model (D) to PU(NSL) have very similar edge-probability asymptotics, and Remark 4.3.6 is a reasonable heuristic — but 'probably' isn't a proof. The stress-test note is right that the edge-mark combinatorics for model (D) differ: each vertex has m_u distinguishable out-edges rather than one per collapsed block, and the factorial and Gamma size-bias factors in (4.4.39) need to line up exactly. That verification is absent. This is a derivational gap, not a detected error, and it's the kind of thing a referee can ask to be filled.\n\nThe second part (percolation threshold, large-set expanders, Ising critical temperature) is not in the text I was given. I can't verify those claims and won't pretend to have an opinion on them. No red flags in the visible material, and the author flags the scope honestly (no models (E)/(F), for instance).\n\nWho it's for: anyone working on local limits or processes on scale-free networks. The thesis deserves a serious referee; the model (A) proof alone justifies it. But for publication, I'd want the (B)/(D) details supplied or the theorem explicitly restricted to model (A), and the process chapters made available. My recommendation: send it to review, ask for the missing arguments in the revision.","headline":"Solid local convergence proof for model (A), but the extension to (B) and (D) is asserted, not shown, and Part II is invisible in this draft.","tokens_in":70618,"tokens_out":2628,"would_cite":true,"duration_ms":24408,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["05C80","60J80","60K35","82B20"],"pacs":[],"model":"deepseek-v4-flash","headline":"Preferential attachment networks converge locally to one random Pólya tree","keywords":["preferential attachment","local convergence","random Pólya point tree","Pólya urn","percolation threshold","spectral radius","Ising model","power-law degree distribution"],"falsifier":"Take a small marked tree $t$ and compute, from the connection rule in (1.1.6), the exact joint age density of a uniformly chosen vertex's $r$-neighborhood in Model (D); if this density does not match the RPPT density of Proposition 4.4.2 up to a $o(1)$ error as $n\\to\\infty$, then Theorem 4.2.1 fails for Models (B) and (D). The same check can be done by evaluating the self-loop and edge-mark corrections in the second-moment sum directly.","tokens_in":69499,"feed_emoji":"🌳","tokens_out":8958,"duration_ms":81789,"temperature":0.7,"pith_summary":"This thesis tries to establish that the local geometry of preferential attachment networks is universal: for a broad class of affine models with i.i.d. random out-degrees and attractiveness parameter $\\delta > -\\inf \\mathrm{supp}(M)$, the neighborhood of a uniformly chosen vertex converges in probability to one infinite random tree, the random Pólya point tree. It claims this for three standard model variants, (A), (B), and (D), extending earlier fixed-out-degree results to random out-degrees and negative $\\delta$. The same local limit is then used to compute global phase-transition quantities: the critical percolation threshold, the subcritical component size, and the quenched Ising critical temperature and thermodynamic limits. A sympathetic reader would care because this turns a hard dynamic graph problem into a branching-process calculation.","feed_headline":"Three preferential attachment variants share one local tree limit","feed_subtitle":"The same Pólya point tree fixes degree tails, percolation, and Ising phase transitions.","key_machinery":"The load-bearing object is the random Pólya point tree, a multitype branching process whose type space is a continuous age in $[0,1]$ together with a Gamma-distributed strength and an $O/Y$ label recording whether a node is older or younger than its parent. The proof has two main mechanisms: a Pólya urn representation that makes the edge-connection events in models (A), (B), and (D) conditionally independent given Beta-distributed urn weights, and an explicit density computation, closed by a second-moment method, showing that the joint age density of an $r$-neighborhood in the graph converges to the density of the $\\mathrm{RPPT}$. For percolation, the threshold is identified with the inverse of the spectral radius of the mean offspring operator, the branching-process growth rate in a continuous type space.","core_discovery":"The central claim is Theorem 4.2.1: if $M$ is an $\\mathbb{N}$-valued out-degree distribution with finite $p$-th moment for some $p>1$ and $\\delta > -\\inf \\mathrm{supp}(M)$, then the preferential attachment models (A), (B), and (D) converge vertex-marked locally in probability to the random Pólya point tree $\\mathrm{RPPT}(M,\\delta)$. The vertex mark of vertex $k$ in an $n$-vertex graph is $k/n$, so the mark converges to the age of the corresponding node in the limiting tree. The thesis further claims that the critical percolation threshold of the Pólya point tree is the inverse of the spectral radius of its mean offspring operator, and that the same threshold holds for the finite preferential attachment models because they are large-set expanders with bounded average degree. For the quenched Ising model, it claims explicit limits for pressure per particle, magnetization, and internal energy, together with an explicit inverse critical temperature.","pith_inferences":["Editorial inference: the same Pólya-urn route would likely deliver the random Pólya point tree limit for the independent and simple models (E) and (F), which the thesis leaves open.","Editorial inference: the threshold formula suggests a testable recipe for other growing network models: compute the local branching limit, take the inverse spectral radius of its mean offspring operator, and check the large-set expander condition.","Editorial inference: the exponent formula $\\min\\{\\tau_M, 3+\\delta/\\mathbb{E}[M]\\}$ predicts that heavy out-degree tails suppress the usual rich-get-richer exponent; a simulation measuring the degree tail of a uniform vertex as $M$ and $\\delta$ vary would directly test this."],"forward_implications":["The asymptotic degree of a uniformly chosen vertex follows a power law with exponent $\\min\\{\\tau_M, 3+\\delta/\\mathbb{E}[M]\\}$, so the tail is controlled by whichever of the out-degree distribution and the preferential-attachment mechanism is heavier.","Older and younger neighbors of a uniform vertex have degree tails with exponents $\\min\\{\\tau_e-1, \\tau_M-1\\}$ and $\\min\\{\\tau_e+1, \\tau_M-1\\}$, respectively, a size-biasing effect visible directly from the local limit.","The critical percolation threshold of the Pólya point tree is the inverse of the spectral radius of the mean offspring operator, and this same threshold transfers to the preferential attachment graphs through their large-set expander property.","In the subcritical percolation regime, the largest connected component is significantly larger than the maximum degree, so subcritical clusters are not bounded by the local degree scale.","The quenched Ising model on these graphs has explicit thermodynamic limits and an explicit inverse critical temperature, making the phase-transition parameters computable from the local limit."],"supporting_citations":[{"why":"Defines the Pólya point tree and proves local convergence for fixed out-degrees, the case this thesis extends.","marker":"[21]"},{"why":"Supplies the Pólya urn representation that makes edge-connection events conditionally independent, the key tool for the density computations.","marker":"[20]"},{"why":"Establishes the power-law exponent for preferential attachment models with i.i.d. out-degrees, recovered here via the local limit.","marker":"[57]"},{"why":"Provides the model definitions, the collapsing operator, and the local-convergence toolbox used throughout the thesis.","marker":"[98]"},{"why":"Proves that percolation thresholds are local for large-set expanders with bounded average degree, the property used to transfer the threshold from the tree to the graphs.","marker":"[3]"},{"why":"Gives a simple necessary and sufficient condition for locality of the percolation threshold, used in the percolation chapter.","marker":"[97]"},{"why":"Shows that Ising thermodynamic quantities are local for random graphs, the background for the quenched Ising analysis.","marker":"[58]"}],"fun_headline_variants":["Three network models converge to one Pólya point tree","Local tree limit unifies percolation and Ising on growing graphs","Pólya point tree: the universal limit for preferential attachment","Three growth models, one local tree, two phase transitions solved","Preferential attachment's hidden tree sets percolation and Ising thresholds"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The detailed proof is written only for Model (A); for Models (B) and (D) the argument assumes, without a fully written verification, that their edge-connection probabilities and self-loop corrections differ from Model (A)'s only by errors that vanish at the same rate, so the same density and second-moment computations apply.","fun_headline_variants_meta":{"raw":{"variants":["Three network models converge to one Pólya point tree","Local tree limit unifies percolation and Ising on growing graphs","Pólya point tree: the universal limit for preferential attachment","Three growth models, one local tree, two phase transitions solved","Preferential attachment's hidden tree sets percolation and Ising thresholds"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000213,"raw_usage":{"total_tokens":1417,"prompt_tokens":937,"completion_tokens":480,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":553,"completion_tokens_details":{"reasoning_tokens":391}},"tokens_in":553,"tokens_out":480,"duration_ms":5277,"temperature":1.0,"reasoning_tokens":391,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T15:31:16.328509+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a small marked tree $t$ and compute, from the connection rule in (1.1.6), the exact joint age density of a uniformly chosen vertex's $r$-neighborhood in Model (D); if this density does not match the RPPT density of Proposition 4.4.2 up to a $o(1)$ error as $n\\to\\infty$, then Theorem 4.2.1 fails for Models (B) and (D). The same check can be done by evaluating the self-loop and edge-mark corrections in the second-moment sum directly.","supporting_citations":[{"cited_title":"Random graphs and complex networks","cited_arxiv_id":null,"evidence_quote":"Provides the model definitions, the collapsing operator, and the local-convergence toolbox used throughout the thesis."}],"review_version":1}