{"id":"d03a4e21-d453-4b67-9ff0-9e39863b4855","arxiv_id":"2501.03195","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"On a random recursive tree with n vertices, parking is supercritical at every positive density, and the first outward flux for binary car arrivals appears when the mean number of cars per vertex is about (log n)^{-2+o(1)}.","lead":"Cars park on a random recursive tree, a random family tree built one vertex at a time. The paper proves that for any positive car density, traffic spills out of the tree, and it identifies the very small density at which this outward flux starts.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The lower bound of Theorem 3 rests on unproven concentration estimates for the branching construction in Section 4; a rigorous triangular-array LLN is needed before the supercritical half is established.","rationale":"The reader's verdict is CONDITIONAL, and I agree that the paper is publishable after revision. However, the single most load-bearing gap is not the monotonicity of ψ(T_n) in n (which is true by a simple path argument and does not threaten the result), but the unproven 'successive applications of the law of large numbers' in Section 4. The lower bound for the supercritical regime constructs a multitype structure V^{(ℓ)}_{cars} whose size is claimed to grow like t^{ℓδ} with high probability. This is a triangular array of branching processes with offspring distributions depending on t, and the induction requires quantitative variance bounds at each step. The number of parents at step ℓ is only t^{ℓδ}, so the relative error of the total is controlled only if the per-parent variance is sufficiently small compared to the square of the per-parent mean; the paper does not verify this. A concrete analytical check—computing the second moment of the first-generation count and then iterating a conditional Chebyshev bound—would settle whether the claimed concentration holds. If it holds, Theorem 3 is correct; if not, the critical window could be shifted. The other issues (typos in Section 3, the c_v integration range, the 1/2 vs β* in Section 5) are cosmetic. Since the missing concentration argument is a fillable gap rather than a demonstrated contradiction, the verdict remains CONDITIONAL (UNCHANGED).","tokens_in":12245,"tokens_out":43669,"duration_ms":384417,"concrete_test":"Provide a complete proof of the concentration used in Section 4. Concretely, for the first generation, let Z_t be the number of vertices at height k* in T_t born before t/2; compute E[Z_t] and Var(Z_t) (or couple Z_t with the Poisson(t/(2k*))-offspring Galton–Watson process) and show Var(Z_t)/E[Z_t]^2 → 0. Then, for ℓ = 1,...,j, show inductively that conditional on |V^{(ℓ)}_{cars}| ≥ m t^{ℓδ}, the mean of |V^{(ℓ+1)}_{cars}| is at least c t^{δ} |V^{(ℓ)}_{cars}| and its conditional relative variance is o(1); if the variance at any step fails to be negligible, the lower bound argument in Theorem 3 collapses.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The proof of the supercritical half of Theorem 3 hinges on the assertion in Section 4 that, for a fixed integer j, the sets V^{(ℓ)}_{cars} satisfy |V^{(ℓ)}_{cars}| ≥ m t^{ℓδ} with probability tending to 1, obtained by 'successive applications of the law of large numbers'. This is a triangular array of embedded branching processes whose offspring distributions depend on t, and the induction passes from generation ℓ to ℓ+1 by conditioning on the previous count being at least m t^{ℓδ}. No variance or concentration estimates are supplied. In particular, the first step requires that the number of vertices at height k* born before time t/2 in the Yule tree is concentrated around (t/(2k*))^{k*}; this is plausible but needs a second-moment argument. More importantly, at each subsequent step the number of parents is only of order t^{ℓδ}, so the averaging that justifies the LLN requires the per-parent variance to be small relative to the square of the per-parent mean. If at any step the relative fluctuations do not vanish, the construction could fail to produce a positive flux at α_n ≫ (log n)^{-1/β* + δ}. The monotonicity of ψ(T_n) in n, flagged by the reader, is also unproved but is elementary: adding a leaf cannot decrease the number of cars reaching the root, since existing cars' paths are unchanged and new cars can only occupy empty spots or push further toward the root. Thus the concentration step, not monotonicity, is the most load-bearing gap.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the parking process on uniform random recursive trees with i.i.d. car arrivals. It proves that the critical density parameter for the parking phase transition is 0 (Theorem 1), via the Benjamini–Schramm limit and a criterion from the first author's earlier work [8]. It then identifies the critical window for a positive outgoing flux: for a general class of bounded, stochastically increasing arrival families satisfying the small-α asymptotics (2), the flux converges to 0 in probability when α_n ≪ (log n)^{-1/β*}, and to +∞ when α_n ≫ (log n)^{-1/β* + δ} for any δ > 0 (Theorem 3), where β* is determined by the arrival law. In the binary case this gives the window (log n)^{-2+o(1)} (Theorem 2). The upper bound is proved by a first-moment sum over fully parked trees; the lower bound uses a planted multi-level branching structure inside the parked cluster of the root, with continuous-time Yule estimates transferred to discrete recursive trees via an increasing coupling.","tokens_in":12434,"tokens_out":28917,"duration_ms":255519,"significance":"If the technical gaps identified below are fixed, this is a substantial contribution. It is the first parking-process result on trees with unbounded degrees, and it gives a sharp, parameter-free characterization of the critical window through the exponent β*. The upper-bound method via fully parked trees and the spine construction for α_c = 0 are elegant and likely to be reusable. The main weakness is the lower bound: the concentration estimates for the planted branching structure are asserted rather than proved, and one displayed inequality chain in Section 4 is misordered. The central claims are plausible and appear correct, but the manuscript in its current form does not yet provide rigorous proofs of all load-bearing steps.","major_comments":[{"comment":"The lower bound of Theorem 3 relies on the assertions, introduced by 'By successive applications of the law of large numbers', that P(|V^{(1)}(T_t)| ≥ (1/2)(t/(2k*))^{k*}) → 1 and, after car thinning, P(|V^{(j)}_{cars}(T_t, μ_{α_t})| ≥ m t^{jδ}) ≥ 1 − 2ε. These are concentration statements for a triangular array of embedded branching processes whose offspring distributions depend on t, and the induction passes from generation ℓ to ℓ+1 with only about t^{ℓδ} parents. The paper supplies no variance or second-moment estimates, and the phrase 'successive applications of the law of large numbers' does not by itself justify the high-probability lower bounds at polynomial scale. This is a load-bearing gap: if at any step the relative fluctuations do not vanish, the planted structure may fail to produce a positive flux. Please provide a rigorous induction (e.g., Chebyshev's inequality with explicit bounds on the conditional mean and variance given the previous generation) or replace the construction with one that has a proof.","section":"Section 4, definition of V^{(1)} and V^{(j)}_{cars}"},{"comment":"In the paragraph containing Equation (10), the displayed chain of inequalities reads P(ψ(T_t, α_t) ≥ C) ≥ P(ψ ≥ (mc/2^{j+1}) t) ≥ P(ψ ≥ (mc/2^{j+1}) t^{δ j+1} α_t^γ) ≥ 1 − 4ε. The middle step is not valid as written: by the choice of j and (9), t^{δ j} α_t^γ > 1, so for large t the threshold (mc/2^{j+1}) t^{δ(j+1)} α_t^γ exceeds (mc/2^{j+1}) t, making the second event the smaller one. The desired conclusion (P(ψ ≥ C) → 1 for every fixed C) follows directly from the previously established bound with B_t := (mc/2^{j+1}) t^{δ(j+1)} α_t^γ, since B_t → ∞; the intermediate comparison with t should be removed or corrected.","section":"Section 4, Equation (10)"},{"comment":"In the paragraph starting 'Given the family of (t^i_v ...)', the product ∏_{v∈t} ∏_{i=1}^{c_v} ∫_0^t dt^i_v is written with c_v denoting the number of cars arriving at v, but the surrounding text says the integrals are over the creation times of the i-th children of v. The total number of such creation times is n−1 (one per edge of t), not m = ∑_v c_v, and the subsequent factor t^{n-1} confirms that the intended upper index is the outdegree of v. This notational inconsistency obscures a load-bearing step: the exponent n−1 is essential for the condition t α^{β*} ≤ c in Proposition 2 (if the exponent were m, the required condition would be of order t^2 α^{β*} ≤ c). Please rewrite this calculation with a clear distinction between car counts and child counts (e.g., use d_v for the number of children).","section":"Section 3.2, computation of P(C(ρ,T_t) ⊇ t)"}],"minor_comments":[{"comment":"The monotonicity assertion 'for every fixed α, the sequence (ψ(T_n, μ_α)) is non-decreasing' is used in both directions of the discrete transfer (e.g., in the inequalities with E[ψ(T_t)1_{|T_t|≥n}] and with P(φ(T_{log n})≥K, |T_{log n}|≤n)) but is not proved. Please add a one-sentence justification (adding a leaf cannot decrease the number of cars visiting the root, since existing car paths are unchanged and new cars can only add visits).","section":"Section 5, discrete transfer"},{"comment":"The bound P(ψ_{S_k}(T∞,μ)=0) ≤ (1/(1+δ(α)))^k should be (1/(1+δ(α)))^{k+1} if the integration runs over k+1 variables τ_0,...,τ_k; the conclusion is unaffected.","section":"Section 2.3, Borel–Cantelli bound"},{"comment":"The time cutoff 't − t/2^ℓ' with ℓ=1 gives t/2, the same as in the definition of V^{(1)}; please check the intended time windows (should it be t − t/2^{ℓ+1} or similar?) to avoid ambiguity.","section":"Section 4, recursive definition of V^{(ℓ+1)}"},{"comment":"There are minor typos: 'supercritial' should be 'supercritical', and 'Bienaym´e' is written with a non-standard accent; the reference list also shows 'Bienaym ´e' in [6].","section":"Section 1 and throughout"}],"recommendation":"major_revision","confidential_remarks":"The paper relies on the first author's PhD thesis [8] for the general phase-transition characterization; this is a legitimate black box, and the RRT-specific conclusions are derived independently from that theorem. I see no circularity issue. The main risk is that the lower-bound concentration gap is nontrivial: the authors should be encouraged to fill it with explicit second-moment or Chebyshev estimates. The paper fits the scope of math.PR and is likely to be a good contribution once the missing rigour is added."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Contat and Laulin prove two things about parking on uniform random recursive trees: the critical density is 0 for any nontrivial arrival law, and for bounded arrivals the flux window sits at α_n = (log n)^{-1/β* + o(1)}. The α_c=0 result is genuinely new — the local limit is non-degenerate, so the mechanism has to be the growing degree along the spine — and the critical window calculation is not in the earlier literature. The paper is worth a serious referee.\n\nWhat is solid: the upper bound via fully parked trees and first moment is clean and, as far as I can tell, correct. The continuous-time Yule construction is well suited to the model, and the Borel-Cantelli argument over the spine vertices for α_c=0 is elegant. The self-citation to [8] for the general phase-transition criterion is fine; the RRT-specific work is independent.\n\nThe soft spot is the lower bound of Theorem 3. The construction of V_cars^{(ℓ)} is plausible, but the text says 'by successive applications of the law of large numbers' twice without giving variance or concentration estimates. This is a triangular-array LLN: the per-parent offspring means grow with t, but the number of parents at step ℓ is only t^{ℓδ}, and you need relative fluctuations to vanish at each step. I don't see an actual counterexample — the branching subtrees above distinct parents are conditionally independent, so a Chebyshev argument should go through if the right moment bounds are stated. But as written, the supercritical half is not fully proved. The monotonicity of ψ(T_n) in n, flagged as a worry, is actually elementary: adding a leaf cannot remove cars from existing vertices, and new cars can only push occupied vertices upward. That one only needs a sentence. There are also a few typos: in Section 3 the integration is over children but written with c_v (the car count), and the inequality chain in Section 5 around the discrete transfer has a garbled term. None of these change the intended arguments.\n\nWho this is for: people working on parking on random trees and on the RRT itself. The results are significant for that subfield, moderate in broader impact. I would send it to review; the referee should ask for a written proof of the concentration step in Section 4 and cleanup of the typos, not for a rewrite.","headline":"New result on parking on random recursive trees: critical density zero, critical window (log n)^{-1/β*}; the upper bound is clean, but the supercritical half needs a real LLN proof.","tokens_in":13072,"tokens_out":12464,"would_cite":true,"duration_ms":118168,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60C05","05C05","60J80"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper proves that parking on a random recursive tree is supercritical at every positive car density, with the critical window for binary arrivals at $\\alpha_n = (\\log n)^{-2+o(1)}$.","keywords":["parking process","random recursive tree","phase transition","outgoing flux","Benjamini–Schramm limit","Yule tree","critical window","random trees"],"falsifier":"Run the standard coupling where vertex $n$ attaches to a uniform earlier vertex and each vertex receives an independent car count from a fixed law of mean $\\alpha=0.1$, and plot $\\phi(T_n,\\mu)/n$ for $n$ up to $10^6$: Theorem 1 predicts a positive limit, so a steady decay toward $0$ would refute it. As a more targeted check, record $\\psi(T_n)$ along the same coupling; any single decrease in this sequence would disprove the monotonicity assertion on which the discrete transfer depends.","tokens_in":11930,"feed_emoji":"🚗","tokens_out":12009,"duration_ms":98833,"temperature":0.7,"pith_summary":"On a random recursive tree, the parking process has no nontrivial phase transition: for any fixed positive arrival density $\\alpha$, a positive fraction of cars eventually leaves the root, even though the tree sequence has a non-degenerate Benjamini–Schramm limit. The paper identifies the reason in the infinite spine: ancestors of a typical vertex have degrees growing along the spine, so infinitely many of them harbour parked cars. It then locates the window in which the flux first appears when the density is allowed to vanish with $n$. For binary arrivals the threshold is $\\alpha_n = (\\log n)^{-2+o(1)}$; for general bounded stochastically increasing families it is $(\\log n)^{-1/\\beta^*+o(1)}$. This is the first analysis of the parking process on trees that can have large-degree vertices.","feed_headline":"Random recursive trees flush cars out at any positive density","feed_subtitle":"The infinite spine drives the critical density to zero; binary arrivals turn supercritical near (log n)^{-2}.","key_machinery":"The argument runs through three constructions. (1) The Yule-tree coupling: the discrete recursive trees $(T_n)$ are realized from a continuous-time Yule tree $T_t$ cut at the times of births, so that $T_{\\vartheta_n}$ has the law of $T_n$; this lets the paper work in continuous time and then transfer results back to the discrete sequence through the law of $|T_t|$. (2) The spine tree $T_\\infty$: the Benjamini–Schramm limit is described as one infinite spine decorated with independent Yule trees along a Poisson process, and the ancestors $(S_k)$ have degrees growing like $k$; this spine is what makes the phase transition trivial. (3) Fully parked trees: for the upper bound, the flux is bounded by summing over all fully parked configurations embedded in the root's parked component, giving a first-moment estimate $\\mathbb{E}[\\psi(T_t, \\mu_\\alpha)] \\to 0$ when $\\alpha^{\\beta^*} t$ stays small. The lower bound instead constructs a generation-restricted Galton–Watson tree on vertices that receive $k^*$ cars at multiples of the height $k^*$, and shows this tree reaches a height where its occupied children force a large flux.","core_discovery":"The central claim is Theorem 1: the critical parameter for parking on the random recursive tree is $\\alpha_c = 0$, meaning that for every stochastically increasing family of arrival laws with $\\mu_\\alpha(\\{0,1\\}) < 1$ and every fixed $\\alpha > 0$, the flux per vertex $\\phi(T_n, \\mu_\\alpha)/n$ converges in probability to a positive constant $C_\\alpha$. The reason is not the root's large degree but the spine of the Benjamini–Schramm limit: the $k$-th ancestor $S_k$ of the distinguished vertex has degree of order $k$, and the probability that $S_k$ remains empty in the final configuration decays like $(1+\\delta(\\alpha))^{-k}$, so by Borel–Cantelli infinitely many ancestors are occupied and an infinite parked cluster exists. The paper's second claim, Theorem 3, quantifies the transition when $\\alpha = \\alpha_n \\to 0$: the flux vanishes in probability when $\\alpha_n \\ll (\\log n)^{-1/\\beta^*}$ and diverges to $+\\infty$ when $\\alpha_n \\gg (\\log n)^{-1/\\beta^*+\\delta}$, where $\\beta^* = \\inf\\{\\beta_k : C_k > 0\\}$ is read off from the small-$\\alpha$ asymptotics $\\mu_\\alpha(\\{k\\}) \\sim C_k \\alpha^{\\beta_k k}$; for binary arrivals this gives the critical window $(\\log n)^{-2+o(1)}$.","pith_inferences":["The spine mechanism suggests a broader principle: any finite-tree sequence whose Benjamini–Schramm limit has a one-ended spine with side subtrees accumulating along it may also have $\\alpha_c = 0$, and the critical window will be governed by the growth of degrees along the spine.","The gap between the upper and lower regimes, $\\alpha_n \\ll (\\log n)^{-1/\\beta^*}$ versus $\\alpha_n \\gg (\\log n)^{-1/\\beta^*+\\delta}$, leaves the exact constant and the sharp order of the flux inside the window open; one could test numerically whether the flux scales like a power of $\\log n$ at $\\alpha_n = c (\\log n)^{-1/\\beta^*}$.","A direct way to check the paper's transfer step is to simulate the natural coupling of $T_n$ and record $\\psi(T_n)$; monotonicity of this sequence is asserted without proof, and any decrease would isolate a gap in the proof rather than in the result.","The boundedness of arrivals is used only in the upper bound, so simulating Poisson or geometric arrivals at $\\alpha_n = c (\\log n)^{-1/\\beta^*}$ would probe how far the conjectured threshold extends beyond the bounded case."],"forward_implications":["For every fixed $\\alpha > 0$, the flux fraction $\\phi(T_n, \\mu_\\alpha)/n$ converges in probability to a positive constant, so the random recursive tree is always supercritical.","When $\\alpha_n \\ll (\\log n)^{-1/\\beta^*}$, the outgoing flux converges to $0$ in probability; when $\\alpha_n \\gg (\\log n)^{-1/\\beta^*+\\delta}$, it diverges to $+\\infty$.","For binary car arrivals the critical window is $\\alpha_n = (\\log n)^{-2+o(1)}$.","The first time the flux reaches any fixed level $C$ satisfies $\\log \\log \\theta(\\mu_\\alpha, C)/|\\log \\alpha| \\to \\beta^*$, so the waiting time grows like $\\exp(\\alpha^{-\\beta^*+o(1)})$.","The flux time-corollary makes the divergence sharp: once the density parametre is above the window, the flux is not merely positive but unbounded in probability."],"supporting_citations":[{"why":"Supplies the phase-transition characterization (Theorem 4.1) that ties the critical density to the existence of an infinite parked cluster in the Benjamini–Schramm limit; used to prove Theorem 1.","marker":"[8]"},{"why":"Provides the Yule-process construction with $T_{\\vartheta_n}$ equal in law to $T_n$ and the Benjamini–Schramm convergence of the fringe, which the discrete transfer and the limit description rely on.","marker":"[16]"},{"why":"Gives the first-moment and fully-parked-tree decomposition that the upper bound in Proposition 2 adapts.","marker":"[15]"},{"why":"Introduces the parking process and flux concept that the paper generalizes to trees.","marker":"[17]"},{"why":"Extends parking to trees and defines the flux and parked components used throughout the paper.","marker":"[18]"}],"fun_headline_variants":["Critical parking density on recursive trees is exactly zero","Recursive trees flush excess cars at any positive density","Zero parking threshold on random recursive trees","Infinite spine drives parking threshold to zero","Even tiny car arrivals overflow recursive trees"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof that bounds on the continuous-time version of the tree transfer to the discrete one-vertex-at-a-time version assumes, without proof, that the number of cars passing the root never decreases as the tree grows; if this monotonicity ever failed, the transfer of both the upper and lower bounds would collapse.","fun_headline_variants_meta":{"raw":{"variants":["Critical parking density on recursive trees is exactly zero","Recursive trees flush excess cars at any positive density","Zero parking threshold on random recursive trees","Infinite spine drives parking threshold to zero","Even tiny car arrivals overflow recursive trees"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000313,"raw_usage":{"total_tokens":1781,"prompt_tokens":948,"completion_tokens":833,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":564,"completion_tokens_details":{"reasoning_tokens":766}},"tokens_in":564,"tokens_out":833,"duration_ms":9965,"temperature":1.0,"reasoning_tokens":766,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T21:57:09.364788+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the standard coupling where vertex $n$ attaches to a uniform earlier vertex and each vertex receives an independent car count from a fixed law of mean $\\alpha=0.1$, and plot $\\phi(T_n,\\mu)/n$ for $n$ up to $10^6$: Theorem 1 predicts a positive limit, so a steady decay toward $0$ would refute it. As a more targeted check, record $\\psi(T_n)$ along the same coupling; any single decrease in this sequence would disprove the monotonicity assertion on which the discrete transfer depends.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the phase-transition characterization (Theorem 4.1) that ties the critical density to the existence of an infinite parked cluster in the Benjamini–Schramm limit; used to prove Theorem 1."},{"cited_title":"H OLMGREN AND S","cited_arxiv_id":null,"evidence_quote":"Provides the Yule-process construction with $T_{\\vartheta_n}$ equal in law to $T_n$ and the Benjamini–Schramm convergence of the fringe, which the discrete transfer and the limit description rely on."},{"cited_title":"G OLDSCHMIDT AND M","cited_arxiv_id":null,"evidence_quote":"Gives the first-moment and fully-parked-tree decomposition that the upper bound in Proposition 2 adapts."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Introduces the parking process and flux concept that the paper generalizes to trees."},{"cited_title":"L ACKNER AND A","cited_arxiv_id":null,"evidence_quote":"Extends parking to trees and defines the flux and parked components used throughout the paper."}],"review_version":1}