{"id":"ba9b9163-fa15-4ca0-be9e-0bee84c701ef","arxiv_id":"1908.08481","paper_version":4,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"A critical one-parameter family of Rayleigh random flights on Poisson line SIRSNs has a speed process that is neighborhood-recurrent, giving evidence for the conjecture that SIRSN geodesics never halt en route.","lead":"This paper builds a scale-invariant random flight process on Poisson line road networks and shows that at a special parameter value its speed keeps returning to its starting range instead of dying out or exploding. The result supports an open conjecture that fastest routes on such networks never come to a complete stop.","discovery_kind":"first_principles","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 30's ergodicity proof has a gap: the time-constant set of intersection angles of Π may be a nonconstant bounded harmonic function, which would break the Kesten-Spitzer-Whitman step in Theorem 31.","rationale":"The reader identified Theorem 30 as the weakest assumption, specifically the reduction of the bounded harmonic function to H(Π) and the application of Theorem 21. I agree that this is the soft spot, but I sharpen it: the paper's own Section 4 constructs a candidate counterexample, the time-constant angle set A. If A is measurable and nonconstant, the claimed ergodicity of Ψ is false, not merely unproven. If A is not measurable in the relevant σ-algebra, then the proof needs an explicit argument showing why such invariants do not enter the invariant σ-field, and why the conditional mean of the log-speed is zero, which the paper does not provide. Because the main theorem Theorem 31 depends on this ergodicity for the Kesten-Spitzer-Whitman recurrence, the verdict should be conditional on a rigorous resolution of this gap rather than unconditional acceptance. I do not see an independent flaw elsewhere; the abstract scattering formalism and the stationary distribution computation appear coherent, and the non-critical divergence/convergence discussion is plausible. The concern is concentrated in Theorem 30 and its use in Theorem 31.","tokens_in":34081,"tokens_out":40420,"duration_ms":452578,"concrete_test":"Check whether the angle set A(Π) is a measurable, similarity-invariant, non-constant function on the state space of the relative environment process. Concretely: for a Poisson line process with parameter γ>2, compute the law of A(Π) as a random countable dense subset of [0,π), and test whether there is a Borel set C in the space of countable subsets with 0 < P(A(Π)∈C) < 1 such that {A(Π)∈C} belongs to the σ-algebra generated by bounded-window events of Π. If such C exists, then F(A(Ψ_n)) = 1_C(A(Ψ_n)) is a nonconstant bounded harmonic function, directly contradicting Theorem 30. If no such measurable C exists, provide the missing measure-theoretic argument that the invariant σ-field of Ψ is nonetheless trivial, and show directly that E[log(V_1/V_0) | I] = 0 in the critical case, which is what Theorem 31 actually needs.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The proof of Theorem 30 is the load-bearing step: it converts the zero-mean log-speed of the critical SIRSN-RRF into neighbourhood recurrence via Theorem 29. The step where a bounded harmonic function h on the relative environment Ψ is claimed to reduce to a similarity-invariant H(Π), which is then declared non-random by 'the ergodicity of Π (Theorem 21)', is not justified as written. Section 4 explicitly notes that the set A = {∠(L1,L2) : (L1,L2) ∈ (Ψ×Ψ)∖Δ} of pairwise intersection angles is time-constant under the evolution of Ψ. Therefore, for any bounded function F on such angle sets, F(A(Ψ_n)) is a bounded harmonic function on Ψ. The angle set A(Π) is a random, similarity-invariant function of the Poisson line process: two independent realisations of Π have, almost surely, different countable dense sets of angles. If A is measurable in the relevant state-space σ-algebra, this yields a nonconstant bounded harmonic function on Ψ, so Theorem 30's ergodicity conclusion fails. The proof's appeal to Theorem 21 is not directly applicable: Theorem 21 concerns functions ξ(L,Π) of a distinguished line and the pattern, not arbitrary similarity-invariant functions of Π alone. The paper does not supply the missing argument ruling out invariants such as A. Since Theorem 31 relies on Ψ (and hence the log-speed process U) being ergodic, this gap is central.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper introduces abstract scattering representations for non-lazy Markov chains and uses them to define SIRSN-RRF, a one-parameter family of Rayleigh random flights on a Poisson line SIRSN. The main results are: (i) a characterization (Theorem 22) reducing similarity-equivariant zero-deficit scattering dynamics to a parameter α > γ - 1; (ii) an explicit stationary distribution for the relative environment process (Theorem 27), which is symmetric exactly at α = 2(γ - 1); (iii) a critical-value speed-neighbourhood-recurrence theorem (Theorem 31) based on an ergodicity claim for the relative environment (Theorem 30). The paper motivates this as evidence for the conjecture that SIRSN geodesics never come to a complete stop.","tokens_in":34400,"tokens_out":15111,"duration_ms":162861,"significance":"If valid, Theorem 31 gives a nontrivial example of a scale-invariant random process on a Poisson line SIRSN whose speed is neighbourhood-recurrent at a derived critical parameter, supporting Conjecture 1 and the doubly-infinite geodesic representation. The abstract scattering framework and the explicit calculations with Mecke-Slivnyak and Dirichlet forms are original and likely useful beyond this application. The critical value α = 2(γ - 1) is derived from a zero-mean condition rather than fitted, and the statements are falsifiable in principle. The main obstacle is the unsupported ergodicity proof of Theorem 30, on which the central theorem depends.","major_comments":[{"comment":"The proof of Theorem 30 has a load-bearing gap. From stationarity and harmonicity the paper correctly obtains P[h(Ψ1) = h(Ψ0)] = 1, but the next sentence asserts that 'using Ψ to explore the network' yields a Π-measurable H(Π) with h(Ψn) = H(Π) for all n. This does not follow from constancy along stationary sample paths; it requires an argument that the stationary chain can connect the relative environments arising from the same Π. The paper explicitly states in Section 4 that the reduced relative environment is 'very far from being irreducible', giving the time-constant set A of intersection angles as an obstruction, and no such connecting argument is supplied.","section":"§5, Theorem 30"},{"comment":"Even granting the existence of H(Π), the sentence 'It follows from the ergodicity of Π (Theorem 21) that H(Π) must be non-random' is not justified. Theorem 21 applies to functions ξ(L, Π) that are Euclidean-invariant jointly in a distinguished line L and the pattern; its proof uses translation parallel to L. H(Π) is a similarity-invariant function of the pattern alone, and Theorem 21 does not assert that all such functions are constant. In fact the angle set A(Π) noted in Section 4 is similarity-invariant and, if measurable in the state-space σ-algebra, would yield a nonconstant bounded function constant along Ψ sample paths, violating the claimed ergodicity. The manuscript must either prove the absence of such invariants or weaken Theorem 30 to a statement sufficient for Theorem 31, for example ergodicity of the log-relative-speed process U under the stationary law.","section":"§5, Theorem 30, second paragraph"},{"comment":"The proof of Theorem 31 invokes Theorem 29 for the stationary ergodic log-speed process, but the theorem is stated for an arbitrary SIRSN-RRF initial state. Recurrence for the stationary version does not automatically transfer to a fixed initial state, especially because the paper notes in Section 4 that the invariant measure for the quenched chain is infinite and that no stationary version of Z itself exists. An extension argument from stationary initial conditions to the asserted almost-sure statement for general starts is missing.","section":"§5, Theorem 31"}],"minor_comments":[{"comment":"The sentence 'exploitation of delineated structure as expressed in Theorem 2' should refer to Theorem 5, since the paper has no Theorem 2 at that point.","section":"§2, before Lemma 12"},{"comment":"Remark 15 refers to 'Definition 23' for Euclidean-invariance properties that are introduced in Definition 18; the cross-reference appears to be off by five.","section":"§3, Remark 15"},{"comment":"There are numerous typographical and formatting artifacts, such as 'rˆole', 'c` adl` ag', and inconsistent use of 'deficit' versus 'defect'; these should be cleaned up before publication.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"I agree with the skeptical assessment: the gap in Theorem 30 is real and central. The abstract framework and the Section 4 computations are strong, and the issue may be fixable by proving a weaker ergodicity statement for the log-relative-speed process or by quotienting out the invariant angle structure. I do not see grounds for rejection, but the current proof of the main theorem is incomplete."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a real contribution, and the reader's ACCEPT verdict is about right. The abstract scattering representation is genuinely new and worth having on its own; the one-parameter family of SIRSN-RRF is natural, and the identification of the critical exponent α = 2(γ−1) is a clean first-principles derivation. The main recurrence result is significant if correct, and the Kesten–Spitzer–Whitman adaptation is standard and fine.\n\nThe soft spot is exactly where the reader put it: Theorem 30. The proof is much too terse. The step where a bounded harmonic function h on the relative environment is converted into a Π-measurable H(Π) needs an explicit argument — presumably using quenched irreducibility (Lemma 24) to show h is constant on all relative environments arising from the same Π. The next step, that H(Π) must be non-random, also needs care: Theorem 21 is stated for functions of a distinguished line and the pattern, not directly for functions of the pattern alone. That can be patched by embedding H into that framework with a dummy line, and then translation-ergodicity of the line process does the job, but the paper should say so.\n\nOn the stress-test note: I think the proposed counterexample does not land. The set of pairwise intersection angles A(Π) is time-constant and similarity-invariant, but it is also a translation-invariant measurable function of Π, and the Poisson line process is ergodic under translations. So for any bounded measurable function F of that angle set, F(A(Π)) is almost surely constant. Different realisations having different angle sets does not create a nonconstant bounded harmonic function; ergodicity forbids it. The concern would only bite if there were a nonconstant invariant that is not a function of Π alone, and the paper gives no such example. So I do not think the central argument is broken — but it is underproved, and a referee should insist on a full proof of the reduction to H(Π) and of the application of ergodicity to pattern-only functions.\n\nMinor quibbles: the discussion of the non-critical cases in the conclusion is a bit sketchy, and the paper would benefit from a cleaner statement of the ergodicity theorem for pattern-only invariants. None of this undermines the main result.\n\nWho is this for: people working on random walks in random environments, stochastic geometry, and the SIRSN literature. It deserves a serious referee, not a desk rejection. I would send it out, with a request to expand Theorem 30.","headline":"A serious, novel paper whose main theorem likely holds; the ergodicity proof in Theorem 30 is too compressed but the stress-test counterexample does not actually work.","tokens_in":34895,"tokens_out":5233,"would_cite":true,"duration_ms":57190,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60D05","60G50","37A50"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper constructs a natural family of scale-invariant random flights on Poisson-line SIRSN road networks and proves that, at a critical parameter, their speed is neighbourhood-recurrent, never slowing to zero or escaping to…","keywords":["scale-invariant random spatial networks","Rayleigh random flight","Poisson line process","abstract scattering representation","speed-neighbourhood recurrence","geodesics on SIRSN","Palm conditioning","random walk in random environment"],"falsifier":"Run the critical SIRSN-RRF with α=2(γ-1) on a large finite approximation of a γ-line SIRSN and record the log-speed after each switch; if the sample mean of log-relative speeds is nonzero over long runs for positive-probability environments, or if the particle's speed eventually becomes trapped below a fixed threshold, the recurrence claim is false. A direct theoretical falsifier would be a bounded harmonic function on the relative environment that is constant on sample paths but depends on more than the Poisson line pattern, contradicting the ergodicity step.","tokens_in":33836,"feed_emoji":"🛣️","tokens_out":9807,"duration_ms":90627,"temperature":0.7,"pith_summary":"Scale-invariant random spatial networks (SIRSN) built from speed-marked random lines model the route-finding behaviour of online maps, but it is open whether their fastest paths can ever come to a complete stop. This paper defines a one-parameter family of Rayleigh random flights on such networks—randomly-broken local geodesics that switch lines at intersections with probabilities depending on relative speeds—and proves that at the critical parameter the speed process is neighbourhood-recurrent: it returns infinitely often to any neighbourhood of its starting speed and neither decays to zero nor diverges. Because these flights are the natural randomized analogue of an actual geodesic, their failure to stall is presented as evidence for the conjecture that true geodesics never halt. If the conjecture is true, geodesics can be represented as doubly-infinite sequences of connected line segments, which would justify computing them from finite-line approximations.","feed_headline":"Random flights on Poisson road networks never stall or blow up","feed_subtitle":"At the critical parameter, speed returns near its start infinitely often, backing the no-stall geodesic conjecture.","key_machinery":"The load-bearing construction is the abstract scattering representation of a non-lazy Markov chain, in which transition probabilities are factorized as p_{a,b}=ω_{a,b}s_b, with s_b the probability of scattering at state b and ω_{a,b} the probability of arriving there. For the RRF, states are ordered pairs of distinct lines (L_-,L_0) at an intersection, scattering classes are the lines themselves, and zero-deficit similarity-equivariant dynamics force s(L_1,L_2)=min{1,(v(L_2)/v(L_1))^α} with κ(L)=v(L)^α. The analysis then uses Slivnyak-Mecke Palm conditioning to identify the stationary relative environment, Birkhoff's ergodic theorem to exclude non-critical exponents, a variation of an averaging argument for walks in inhomogeneous environments to prove ergodicity of the relative environment, and an adaptation of the Kesten-Spitzer-Whitman range theorem to deduce neighbourhood recurrence of the log-speed from its zero mean.","core_discovery":"The central discovery is Theorem 31: for every Poisson-line SIRSN or SIRSN candidate with parameter γ≥2, there is a discrete-time SIRSN-RRF whose speed process V_n=v(L_0) at successive line switches almost surely returns infinitely often to every neighbourhood of the initial speed V_0. This occurs at the critical exponent α=2(γ-1), where the stationary law of the logarithm of the relative speed of consecutive lines is symmetric Laplace with zero mean. The proof passes through the relative environment seen from the moving particle: its stationary distribution has independent components (an asymmetric-Laplace log-speed, a sine-weighted angle, and an independent copy of the Poisson line process), and ergodicity of that environment, combined with a real-valued adaptation of the Kesten-Spitzer-Whitman range theorem, converts zero mean into neighbourhood recurrence. Non-critical exponents are excluded: α>2(γ-1) would make the speed almost surely diverge exponentially, while α<2(γ-1) would make it almost surely converge to zero. The result is offered as evidence for Conjecture 1, that a true Π-geodesic never comes to a complete stop en route.","pith_inferences":["The leap from randomly-broken local geodesics to true geodesics is not proved; a direct analogue of Theorem 31 for the actual fastest path would be needed to settle Conjecture 1, so this paper's result is evidence rather than proof.","The critical condition α=2(γ-1) can be read as a zero-log-drift balance: the rate at which faster lines are encountered exactly compensates for slower lines. One could test whether true Π-geodesics show the same balance by measuring, on finite approximations, whether logarithmic speed fluctuations along long geodesic segments have zero mean.","The axiomatic scattering representation may find use beyond SIRSN, for instance in Markov chain Monte Carlo algorithms built from piecewise-deterministic dynamics; the zero-deficit construction suggests a template for designing reversible scattering chains from a prescribed function on scattering classes."],"forward_implications":["The critical SIRSN-RRF supplies a concrete stochastic model of a broken local geodesic whose speed is neighbourhood-recurrent; if the analogy to true geodesics holds, Conjecture 1 follows and Π-geodesics are doubly-infinite sequences of line segments.","Ergodicity of the relative environment forces a sharp trichotomy among SIRSN-RRF: critical speeds are recurrent, supercritical speeds almost surely diverge to infinity, and subcritical speeds almost surely converge to zero with the particle becoming trapped in ever-shrinking cells.","In the converging case the continuous-time flight reaches zero speed in finite time, while in the diverging case it runs for all time; the critical case is the only one consistent with non-halting geodesics.","The abstract scattering representation gives a general algebraic characterization of reversible scattering Markov chains through scattering and transmission probabilities, so the SIRSN-RRF construction can be exported to other reversible dynamics, including piecewise-deterministic Markov processes."],"supporting_citations":[{"why":"defines SIRSN mechanisms and poses the geodesic no-stop conjecture that motivates the result.","marker":"(Aldous, 2014)"},{"why":"establishes that the speed-marked Poisson line process yields a SIRSN and provides the path-comparison estimates used to control explosions.","marker":"(Kendall, 2017)"},{"why":"supplies the rigorous construction of the Poisson-line SIRSN in general dimension, underpinning the planar geodesic framework.","marker":"(Kahn, 2016)"},{"why":"is the source for the Slivnyak-Mecke theorem and Palm conditioning used to compute the stationary relative environment.","marker":"(Chiu et al., 2013)"},{"why":"provides the averaging argument adapted here to prove ergodicity of the relative environment process.","marker":"(Kozlov, 1985)"},{"why":"states the Kesten-Spitzer-Whitman range theorem whose real-valued adaptation yields neighbourhood recurrence from zero-mean ergodic increments.","marker":"(Spitzer, 1976)"}],"fun_headline_variants":["Random flights on Poisson road networks revisit every speed","Speed of random flights on SIRSN returns infinitely often","Poisson-line flights: speed recurrence backs no-stall geodesics","Critical exponent: random flight speed never diverges or stalls","Random walks on Poisson lines: speed recurs, supporting geodesic conjecture"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument stands or falls on the assertion that the random environment seen by the moving particle is ergodic; the proof assumes that any bounded harmonic function constant along particle paths is already determined by the Poisson line process, despite the reduced environment being far from irreducible. If that ergodicity fails, the zero-mean log-speed does not force the speed to return to its starting neighbourhood.","fun_headline_variants_meta":{"raw":{"variants":["Random flights on Poisson road networks revisit every speed","Speed of random flights on SIRSN returns infinitely often","Poisson-line flights: speed recurrence backs no-stall geodesics","Critical exponent: random flight speed never diverges or stalls","Random walks on Poisson lines: speed recurs, supporting geodesic conjecture"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000172,"raw_usage":{"total_tokens":1308,"prompt_tokens":1011,"completion_tokens":297,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":627,"completion_tokens_details":{"reasoning_tokens":214}},"tokens_in":627,"tokens_out":297,"duration_ms":3782,"temperature":1.0,"reasoning_tokens":214,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T11:38:36.081349+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the critical SIRSN-RRF with α=2(γ-1) on a large finite approximation of a γ-line SIRSN and record the log-speed after each switch; if the sample mean of log-relative speeds is nonzero over long runs for positive-probability environments, or if the particle's speed eventually becomes trapped below a fixed threshold, the recurrence claim is false. A direct theoretical falsifier would be a bounded harmonic function on the relative environment that is constant on sample paths but depends on more than the Poisson line pattern, contradicting the ergodicity step.","supporting_citations":[{"cited_title":"Scale- Invariant Random Spatial Networks","cited_arxiv_id":null,"evidence_quote":"defines SIRSN mechanisms and poses the geodesic no-stop conjecture that motivates the result."},{"cited_title":"Improper Poisson line process as SIRSN in any dimension","cited_arxiv_id":null,"evidence_quote":"supplies the rigorous construction of the Poisson-line SIRSN in general dimension, underpinning the planar geodesic framework."},{"cited_title":"Principles of Random Walk, volume 34 of Graduate Texts in Mathematics","cited_arxiv_id":null,"evidence_quote":"states the Kesten-Spitzer-Whitman range theorem whose real-valued adaptation yields neighbourhood recurrence from zero-mean ergodic increments."}],"review_version":1}