{"id":"7888ac22-cc06-4a23-b8be-e8f8c7d6d8b5","arxiv_id":"1908.04843","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Rerooted conditioned multi-type Galton-Watson trees converge locally to multi-type sin-trees, with different limits according to which types appear infinitely often on the backwards-growing spine.","lead":"This paper proves local convergence theorems for conditioned multi-type Galton-Watson trees that are rerooted at a random vertex. The limits are multi-type sin-trees with a single infinite spine, and they differ by which types recur infinitely often along that spine.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 1's proof establishes only convergence of fringes at type-κ ancestors; it never controls fringes with non-κ root type, so Eq. (37) is not derived.","rationale":"Read in good faith: the paper's core is a clean McDiarmid concentration argument for type-κ-rooted fringe counts and a standard sin-tree construction; the applications collect several verified asymptotics. The paper has no machine-checked proofs or code, and no free parameters. The most load-bearing step is the passage from Lemma 1 to the full quenched convergence statements. The reader's identified weak assumption, (22), is real but checked in the applications for the main models (e.g. exp(o(n)) progeny probabilities in §4.2 and Eq. (138) in §4.5). The non-κ-rooted fringe issue, by contrast, is structural: the proof of Theorem 1 does not address any finite marked tree whose root type is not κ, while the limit object has positive probability of such fringes. Eq. (33) explicitly restricts the biased vertex to be non-root, so the first spine edge can end in any type of G. The equivalence in Eq. (10) requires all finite marked trees with marked type in G0, so the gap is in the central theorem, not just an application lemma. The paper may well be correct and the gap fixable with a uniform tail estimate for the sum over biased fringes, but the current argument does not supply it. Hence the verdict should move from CONDITIONAL to UNVERDICTED.","tokens_in":20599,"tokens_out":31982,"duration_ms":344022,"concrete_test":"Re-derive the quenched convergence for a pointed tree S with non-κ root type, e.g. in the reducible 2-type model of §4.3 take S to be a type-2 root with a single type-2 child as marked vertex. Attempt to express the empirical frequency of S among type-2 vertices of T_n using only the counts N_(T,v)(T_n)/#1T_n from Eq. (42); since S is not a type-1-rooted fringe, this is an infinite union over (T,v). Check whether Lemma 1 gives a uniform tail bound on this union; if not, the proof of Theorem 5 (and Theorem 1) is incomplete. As a numerical check, compute the limit frequency P(f[1](T̂(1,2))=S) and the empirical frequency for n up to 200 by exact DP; a match would suggest the claim is true but still unproven, a mismatch would refute it.","verdict_should_be":"UNVERDICTED","load_bearing_attack":"Eq. (37) asserts quenched local convergence in the space X defined in §2.2, which by Eq. (10) requires convergence of the empirical frequencies of every finite marked tree S whose marked vertex has type κ, with S's root type arbitrary. The proof of Theorem 1, however, only proves Eq. (38), i.e. convergence of N_(T,v)(T_n)/#κT_n for trees T whose root has type κ. This is exactly convergence of fκ,[h], the fringe at the h-th type-κ ancestor. It does not cover fringes at ancestors of other types, which occur with positive probability in the limit object: by Eq. (33) the marked vertex of the biased tree T̂κ is a non-root type-κ vertex, so in T̂(κ) the parent of the marked vertex is the parent of such a vertex inside T̂κ and may have any type. For a finite marked tree S with non-κ root type, the event f[H](·)=S is an infinite union of the events fκ,[h]=(T,v); pointwise convergence of each summand in Eq. (38) does not imply convergence of the union because Lemma 1 supplies no uniform tail bound. Consequently Theorems 1-3 and the applications in §4.3-4.4 with fertile non-κ types are not established.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops a concentration inequality (Lemma 1) for counts of fringe subtrees in conditioned multi-type Galton-Watson trees and uses it to claim quenched local convergence of the tree rerooted at a uniformly chosen type-κ vertex to a multi-type version of Aldous' invariant sin-tree. Theorems 2 and 3 extend the result to non-recurring types and mixtures of types, and the applications cover sesqui-type, reducible, irreducible regular, and typed-population conditioned trees. The main proof route is: Lemma 1 gives convergence of empirical fringe counts for finite trees with root type κ; identity (36) converts these into counts of certain extended fringes; and the limiting probabilities are identified with the sin-tree construction.","tokens_in":20750,"tokens_out":38377,"duration_ms":392631,"significance":"If fully established, the paper would provide a unified and quite general local-limit theory for rerooted multi-type branching trees, with a clean main lemma and broad applications. Lemma 1 is a genuinely useful concentration statement, and the sin-tree construction for multi-type processes is natural. However, the central quenched-convergence claim is not proved for extended fringes whose root type differs from κ, and this gap affects most of the applications. The paper is likely correct in its conclusions, but the missing argument is substantial.","major_comments":[{"comment":"","section":"§3.2, Theorem 1 (Eqs. (37)-(38)); §2.2 (Eq. (10))"},{"comment":"","section":"§3.3, Proposition 2"}],"minor_comments":[{"comment":"","section":"§4.2, after Eq. (70)"},{"comment":"","section":"§3.2, Eq. (33); §3.3, Eq. (40); Remark 2"},{"comment":"","section":"§4.2, Lemma 3"},{"comment":"","section":"§4.1, Proposition 4"},{"comment":"","section":"Introduction and general presentation"}],"recommendation":"major_revision","confidential_remarks":"The main issue is a genuine gap in the proof of the central quenched-convergence theorem, not a counterexample to the result. I believe the paper can be repaired with a substantial additional argument controlling extended fringes with non-κ root type, so I recommend major revision rather than rejection. The applications in Sections 4.3-4.5 are all affected by this gap, so the revision is not a matter of local rewriting."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe paper has a genuine gap that the reader's report didn't catch, and the stress-test note is right. Theorem 1 claims quenched convergence in the space X, whose metric compares fringes at every ancestor. The proof only establishes convergence for fringes at the h-th type-κ ancestor (Eq. 38). In the limit object T̂(κ), the immediate ancestors of the marked vertex lie inside the biased tree T̂κ and may have any type, so non-κ-root fringes have positive probability. The proof never controls their empirical frequencies. Eq. (38) is not equivalent to Eq. (37). The same problem propagates to Theorems 2–3 and the irreducible applications in §4.4–4.5.\n\nThat said, the core is not without value. Lemma 1's McDiarmid concentration for fringe counts of type-κ-rooted trees is clean, and the sin-tree construction is a sensible multi-type extension of Aldous. For the sesqui-type and reducible cases (Theorems 4 and 5), the gap may be benign because type-1 vertices have only type-1 parents, so the type-κ-ancestor fringes determine the full local limit. Those applications probably stand.\n\nThere is also a smaller unproved assertion in Lemma 3: the existence of a bounded truncation of (ξ,ζ) preserving E[ξ]=1 is used to identify d = m/D. The reader flagged it; it looks fixable but should be spelled out.\n\nNet: the main theorems are not proven as stated, but the machinery is likely sufficient to repair them. A referee should be sent this and asked for a revised proof of full quenched convergence in X, or a restriction of the statements to settings where the parent of a random type-κ vertex is type-κ with high probability. I'd engage with it.","headline":"Theorem 1 overstates what the proof shows: convergence is only established for fringes at type-κ ancestors, not for the full local topology, which undermines the general statement and the irreducible applications.","tokens_in":21352,"tokens_out":21005,"would_cite":true,"duration_ms":202304,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60J80","60F05","05C05"],"pacs":[],"model":"deepseek-v4-flash","headline":"A conditioned multi-type Galton–Watson tree rerooted at a random vertex converges locally to a multi-type sin-tree, with the spine's type composition governed by offspring means.","keywords":["multi-type Galton-Watson trees","fringe distributions","local convergence","sin-tree","rerooting","conditioned branching processes","sesqui-type trees"],"falsifier":"Simulate a large sesqui-type tree conditioned on total size $n$ under a law satisfying $E[\\xi]=1$ and finite covariance, and estimate the empirical probability that a uniformly chosen vertex has type 1 together with the distribution of the first few extended fringe subtrees up the spine; the theorem predicts a limit whose type-1 probability is $1/(1+E[\\zeta])$ and whose spine-fringe frequencies are those of $\\widehat T(1,\\eta)$. A systematic discrepancy in these finite-$h$ statistics at large $n$ would refute the central claim.","tokens_in":20292,"feed_emoji":"🌳","tokens_out":15675,"duration_ms":156242,"temperature":0.7,"pith_summary":"This paper proves that if you condition a multi-type Galton–Watson tree to be large in a mild way, pick a vertex uniformly at random, and inspect the finite neighborhoods around it, the picture stabilizes: the rerooted tree converges in the local sense to an infinite tree with a single infinite path running backwards from the marked vertex. The main theorems identify the limit for a type whose offspring mean is exactly 2, for types that appear only finitely often along that path, and for mixtures of types. Four applications give explicit limit laws for sesqui-type trees (two-type trees in which only one type branches), reducible critical trees, regular critical irreducible trees, and irreducible trees conditioned by typed population vectors. The paper's central message is that one object, the multi-type sin-tree, organizes the local geometry of a random vertex across these models.","feed_headline":"Rerooting a random vertex of a branching tree yields an infinite spine","feed_subtitle":"A uniformly chosen vertex sees a multi-type sin-tree; type frequencies decide which types haunt the spine.","key_machinery":"The central object is the multi-type sin-tree $\\widehat T(\\kappa)$: a rooted tree with a marked vertex $u_0$ and an infinite backwards-growing spine $u_0,u_1,\\dots$, in which each spine vertex of type $\\kappa$ carries an independent copy of the $\\kappa$-size-biased stopped tree $\\widehat T_\\kappa$, and every other type-$\\kappa$ leaf carries an independent copy of the ordinary conditioned tree $T(\\kappa)$. The argument is carried by a concentration lemma for counts of extended fringe subtrees, meaning the marked subtrees spanned by a vertex and its ancestors: the conditioned tree is decomposed into a short random sequence of independent copies of $T_\\kappa$, so a bounded-differences concentration inequality controls fluctuations of fringe-subtree counts, and Assumption (22) makes the summed tail vanish. The size-bias identity $P(\\widehat T_\\kappa=(T,u)) = P(T_\\kappa=T)$, valid exactly when $E[\\#_\\kappa T_\\kappa]=2$, is what makes the spine Markovian and converts type-$\\kappa$ fringe counts into the sin-tree distribution.","core_discovery":"The central claim, stated as Equation (37) in Theorem 1, is that the conditional law $L((T_n,v_n^\\kappa)\\mid T_n)$ converges in probability to $L(\\widehat T(\\kappa))$, where $T_n$ is the conditioned multi-type Galton–Watson tree, $v_n^\\kappa$ is a uniformly selected type-$\\kappa$ vertex, and $\\widehat T(\\kappa)$ is the $\\kappa$-biased multi-type sin-tree, provided $E[\\#_\\kappa T_\\kappa]=2$ and Assumption (22) holds. Theorem 2 extends the limit to a type $\\gamma$ that need not recur infinitely often on the spine, giving the modified sin-tree $\\widehat T(\\kappa,\\gamma)$, and Theorem 3 mixes these limits when the rerooting vertex is drawn uniformly from a set of types, with mixing weights $p(\\gamma)=E[\\#_\\gamma T_\\kappa]$. In the applications, the limit for sesqui-type and reducible trees has one infinite-recurrence type and one finitely-recurring type, while regular critical irreducible trees and typed-population conditioned trees converge to mixtures $\\widehat T(\\eta)$ in which every type recurs infinitely often along the spine; in the irreducible cases the limit does not depend on the initial root type.","pith_inferences":["I infer that the mixture formula in Theorem 3 extends to any limiting type-proportion vector $w_\\gamma$: if the empirical ratios $\\#_\\gamma T_n/\\#_\\kappa T_n$ converge to $w_\\gamma$ (rather than specifically to $E[\\#_\\gamma T_\\kappa]$), the same proof should give a limit in which the rerooted vertex has type $\\gamma$ with probability proportional to $w_\\gamma$.","I infer that the irreducibility of the limiting spine in the regular critical case is a fingerprint of the conditioning statistic: any conditioning that leaves the initial type influential should produce a limit that does depend on the starting type, so checking whether the first few spine-type frequencies match the left eigenvector of the mean matrix would test the 'forgetting' claim at a quantit","I infer that the same local-convergence scheme should apply to conditioning events with polynomially decaying probabilities, as long as the threshold sequence $s_n$ grows slowly enough for Assumption (22) to hold; the paper verifies the assumption case by case, but the mechanism itself is not tied to $n^{-3/2}$-type estimates."],"forward_implications":["For critical sesqui-type trees conditioned on a total of $n$ vertices, a uniformly selected vertex sees the limit $\\widehat T(1,\\eta)$, where the marked vertex has type 1 with probability $1/(1+E[\\zeta])$ and type 2 with probability $E[\\zeta]/(1+E[\\zeta])$; type 1 recurs infinitely often along the spine, while type 2 does not.","For reducible $d$-type trees with upper-triangular offspring, rerooting at a uniformly chosen vertex from any non-empty type subset $G_0$ yields the mixture $\\widehat T(1,\\eta)$, with $\\eta$ drawn proportionally to $E[\\#_i T^1]$.","For regular critical irreducible trees conditioned on a positive linear combination of type counts, the local limit is $\\widehat T(\\eta)$, independent of the root type chosen to start the conditioning; every type recurs infinitely often along the spine, and $\\eta$ is proportional to the asymptotic type proportions $c_\\gamma$.","For irreducible critical trees conditioned on an exact typed population vector $k(n)$ with $k(n)/\\|k(n)\\|_1\\to a$, the local limit is $\\widehat T(\\eta)$ with $\\eta$ proportional to $a$ restricted to the selected type subset.","In all four settings the convergence is quenched: the empirical distribution of local neighborhoods around a uniformly chosen vertex converges in probability to the stated sin-tree law, not merely in expectation over the random tree."],"supporting_citations":[{"why":"It constructs the monotype sin-tree whose multi-type analogue is the paper's limit object.","marker":"[2]"},{"why":"It establishes monotype local limits for rerooted conditioned Galton–Watson trees, the result being generalized.","marker":"[21]"},{"why":"It supplies the exp(o(n)) total-progeny estimates used to verify Assumption (22) in the sesqui-type and reducible settings.","marker":"[10]"},{"why":"It provides the criticality identity, lattice and support facts, and type-count concentration used by the irreducible applications.","marker":"[18]"},{"why":"It is used alongside [18] for the identity $E[\\#_\\gamma T_\\gamma]=2$ in critical irreducible branching processes.","marker":"[13]"},{"why":"It provides the gcd and lattice lemma and singularity analysis used to prove the lattice support lemma and identify constants.","marker":"[4]"},{"why":"It gives the aperiodic lattice local limit theorem that underlies the sesqui-type asymptotic in Lemma 3.","marker":"[30]"},{"why":"It gives the enriched-tree precursor whose Lemma 22 is extended to the sesqui-type covariance setting.","marker":"[19]"}],"fun_headline_variants":["Multi-type branching trees reroot to infinite spine limits","Rerooted branching trees converge to multi-type sin-trees","Infinite spine emerges in rerooted multi-type trees","Spine types decide the limit of rerooted branching trees","Rerooting branching trees: spine recurrences shape the limit"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole concentration argument rests on Assumption (22): the conditioning event defining $T_n$ must not be exponentially improbable relative to $\\exp(-\\epsilon s_n)$ for every $\\epsilon>0$, while $s_n\\to\\infty$ and $P(\\#_\\kappa T_n\\ge s_n)\\to1$; if the conditioning event decays faster than this, the tail sums of the concentration inequality need not vanish and the local limit could fail.","fun_headline_variants_meta":{"raw":{"variants":["Multi-type branching trees reroot to infinite spine limits","Rerooted branching trees converge to multi-type sin-trees","Infinite spine emerges in rerooted multi-type trees","Spine types decide the limit of rerooted branching trees","Rerooting branching trees: spine recurrences shape the limit"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000487,"raw_usage":{"total_tokens":2347,"prompt_tokens":836,"completion_tokens":1511,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":452,"completion_tokens_details":{"reasoning_tokens":1427}},"tokens_in":452,"tokens_out":1511,"duration_ms":9945,"temperature":1.0,"reasoning_tokens":1427,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T13:31:21.007759+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Simulate a large sesqui-type tree conditioned on total size $n$ under a law satisfying $E[\\xi]=1$ and finite covariance, and estimate the empirical probability that a uniformly chosen vertex has type 1 together with the distribution of the first few extended fringe subtrees up the spine; the theorem predicts a limit whose type-1 probability is $1/(1+E[\\zeta])$ and whose spine-fringe frequencies are those of $\\widehat T(1,\\eta)$. A systematic discrepancy in these finite-$h$ statistics at large $n$ would refute the central claim.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"It constructs the monotype sin-tree whose multi-type analogue is the paper's limit object."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"It establishes monotype local limits for rerooted conditioned Galton–Watson trees, the result being generalized."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"It supplies the exp(o(n)) total-progeny estimates used to verify Assumption (22) in the sesqui-type and reducible settings."},{"cited_title":"Stephenson","cited_arxiv_id":null,"evidence_quote":"It provides the criticality identity, lattice and support facts, and type-count concentration used by the irreducible applications."},{"cited_title":"Miermont","cited_arxiv_id":null,"evidence_quote":"It is used alongside [18] for the identity $E[\\#_\\gamma T_\\gamma]=2$ in critical irreducible branching processes."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"It provides the gcd and lattice lemma and singularity analysis used to prove the lattice support lemma and identify constants."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"It gives the aperiodic lattice local limit theorem that underlies the sesqui-type asymptotic in Lemma 3."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"It gives the enriched-tree precursor whose Lemma 22 is extended to the sesqui-type covariance setting."}],"review_version":1}