{"id":"9ed89717-a57e-45fa-8fd4-db61523d53be","arxiv_id":"2506.18827","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":9.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"A random walk that reflects off the boundary at infinity yields new algorithmic constructions of the free uniform spanning forest and a conjectural embedding framework for supercritical Liouville quantum gravity.","lead":"This paper constructs a random walk that reaches infinity in finite time and then bounces back, repeatedly, on any infinite graph where the ordinary walk escapes. It shows this process generates the free uniform spanning forest via the Aldous-Broder algorithm, and it provides a way to embed infinite-ended maps for supercritical Liouville quantum gravity.","discovery_kind":"first_principles","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Aldous–Broder FUSF theorem is only proven for locally finite G; the abstract overstates scope.","rationale":"The reader's weakest-assumption analysis is exactly the local finiteness requirement in the Aldous–Broder construction, and my reading agrees: this is the single most load-bearing limitation for the paper's headline application. The core Theorem 1.5 on the reflected walk appears well supported: the construction via consistent finite-state Markov chains is detailed, the choice of w* in Lemma 3.5 is justified by a Borel–Cantelli argument, and the uniqueness proof's time-change step, while terse, can be made rigorous by noting that the total time spent outside G_n before any fixed time tends to zero a.s. once Lemma 3.5 is applied to each return segment. No internal inconsistency was found in the proofs of Theorem 1.5. The local finiteness issue is not an error in Theorem 1.14 as stated, since Section 1.4 explicitly imposes it, but it is a genuine scope limitation relative to the abstract's unqualified claim. The paper's conjectural Section 6 is clearly labeled as conjectures, and the Tutte embedding's finite-boundary assumption is another limitation but mainly affects the general definition, not the main FUSF theorem. Therefore the reader's CONDITIONAL verdict is appropriate, and my independent read does not change it.","tokens_in":45309,"tokens_out":39239,"duration_ms":398482,"concrete_test":"Take a non-locally-finite transient graph, e.g., a root r joined by edges to infinitely many disjoint copies of Z^3 (or a tree with one vertex of infinite degree whose simple random walk is transient). Construct or simulate the reflected walk X of Theorem 1.5 and examine the first hitting time τ_x of a fixed vertex x of infinite degree. Determine whether, with positive probability, X visits infinitely many distinct neighbors of x at times arbitrarily close to τ_x without a last distinct neighbor before τ_x. If such a path occurs, Lemma 3.13 fails for this graph, so Definition 1.13's parent map is ill-defined and Theorem 1.14 cannot hold as stated for non-locally-finite G.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The headline FUSF application (Theorem 1.14) is stated in Section 1.4 under the explicit assumption that G is locally finite. This assumption is load-bearing: Definition 1.13 defines the parent a(x) as the vertex visited immediately before the first hitting time τ_x, and Lemma 3.13 is what guarantees this vertex exists. The proof of Lemma 3.13 invokes Lemma 3.12 with V equal to the set of neighbors of x, and Lemma 3.12 requires V to be finite. If G is not locally finite, V can be infinite, so the set X^{-1}(V)∩[0,t) need not be a finite union of intervals and may accumulate at τ_x with no final interval during which X is at a single neighboring vertex. In that case a(x) is not well-defined and the Aldous–Broder forest T^AB of Definition 1.13 does not exist. The abstract, however, states the Aldous–Broder result for an arbitrary infinite transient graph without the local-finiteness qualification. Thus the paper's advertised scope for the FUSF construction is broader than what is actually proved; Theorem 1.14 is correct as a locally finite statement, but the central application does not extend to all graphs covered by the abstract.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces a continuous-time process on an infinite transient graph that is intended to reach infinity in finite time and reflect off it. The main theorem (Theorem 1.5) asserts existence and uniqueness of such a process satisfying structural properties and an energy-minimizing harmonic-measure hitting property. The authors then use this process to construct the free uniform spanning forest via the Aldous–Broder algorithm (Theorem 1.14, under local finiteness), to give a Wilson-type construction (Theorem 1.16), to define a Green's function and a discrete Gaussian free field (Section 1.5), to define a Tutte embedding for multi-ended planar maps (Section 1.6), and to pose scaling-limit conjectures for supercritical Liouville quantum gravity (Section 6). The paper is written in a clear style, and the finite-dimensional portions of the argument are coherent, but the central projective-limit construction of the reflected walk has a fundamental flaw that invalidates Theorem 1.5 and everything built on it.","tokens_in":45542,"tokens_out":40069,"duration_ms":375889,"significance":"If the construction worked, the paper would provide a novel and potentially powerful representation of the free uniform spanning forest, a natural definition of a Green's function and Gaussian free field with free boundary conditions at infinity, and a concrete embedding for multi-ended random planar maps, thereby opening a route to conjectures about supercritical LQG. The conjectural framework in Section 6 is stimulating. However, the load-bearing existence theorem (Theorem 1.5) is not established: as constructed, the discrete-time process collapses to ordinary random walk, so the claimed reflection mechanism is absent. Because all later results—the FUSF theorems, the Green's function/GFF definitions, and the Tutte embedding—call on Theorem 1.5, the paper's central claims are currently unsupported.","major_comments":[{"comment":"The claim that there are infinitely many ξ with Y_ξ = z follows from recurrence of each Y^n, but as shown in the previous comment, Y is the ordinary random walk, which on a transient graph does not return to z infinitely often. Hence the proof of Property (v) is invalid. Moreover, Property (vi) is proven only for the finite approximations Y^n (Lemma 3.2); the passage to Y uses the asserted agreement of hitting sequences, which fails because Y does not have the harmonic-measure transitions. Thus the existence half of Theorem 1.5 is not established.","section":"Section 3.4, proof of Theorem 1.5 existence"}],"minor_comments":[{"comment":"The abstract states that the Aldous–Broder algorithm gives the FUSF on 'an infinite graph' with no local-finiteness qualification, but Theorem 1.14 and Definition 1.13 require local finiteness (Section 1.4 explicitly assumes it). The abstract should either assume local finiteness or state that the FUSF application is proved only in the locally finite case.","section":"Abstract and Section 1.4"},{"comment":"The definition of X_t = infinity for t outside the intervals [τ_η, τ_{hatη}) is vacuous as written: since the equivalence classes are in bijection with N and the sum of all T_ξ is infinite, the intervals cover [0,∞). The authors should clarify whether they intend the process to visit infinity at some times; as written, X_t is always in VG.","section":"Section 3.3, Eq. (3.26)"},{"comment":"The Tutte embedding is defined using harmonic measure on ∂G, but Definition 1.4 requires a finite set A. The text does not state that ∂G is finite; for a general infinite planar map the boundary may be infinite. Please clarify the standing assumption or extend the definition.","section":"Section 1.6 and 5.2"}],"recommendation":"reject","confidential_remarks":"The projective-limit construction in Section 3 collapses to ordinary random walk, so Theorem 1.5 is not proved and the subsequent FUSF and GFF results do not have a valid foundation. This is not a matter of a small repair; the construction of the reflected walk is the core contribution. I recommend rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nQuick take on 2506.18827. The headline: the paper constructs a new process, random walk reflected off infinity, on transient graphs and uses it to give the first direct Aldous-Broder algorithm for the free uniform spanning forest. I think the main existence and uniqueness result is right, and the FUSF theorems look substantively correct for locally finite graphs. The catch is that the abstract promises more than the theorems deliver.\n\nWhat is genuinely new and good: the process is distinct from random interlacements, and the energy-minimizing harmonic measure hitting property is natural. The consistency-with-finite-subgraphs argument in Section 3 is clever and appears to work. Theorem 1.14 (Aldous-Broder equals the FUSF) fills a real gap, since the wired forest had both Aldous-Broder and Wilson constructions via interlacements, while the free forest previously came only through limits over finite graphs. The Wilson variant on the connected event, plus the Green's function and GFF consequences, are nice additions. The supercritical LQG conjectures are clearly labeled and plausible, and they address a question of Ang and Gwynne at least at the level of conjecture.\n\nSoft spots. The most important is the local-finiteness overstatement. Section 1.4 assumes G is locally finite, and Lemma 3.13, which makes the parents in Definition 1.13 well-defined, uses finiteness of the neighbor set in an essential way. Without local finiteness, the Aldous-Broder construction can fail: visits to a neighbor set can accumulate at the hitting time, so \"the vertex immediately before tau_x\" need not exist. The abstract, however, states the Aldous-Broder result for arbitrary transient graphs. This is a real, though easily fixable, mismatch: qualify the abstract and the overview and the paper is fine. Other small things: Lemma 3.5's proof of the existence of w* is terse; it works, but I would ask the authors to spell out the definition of w* and the Borel-Cantelli step more fully. The multi-ended Tutte embedding in Sections 1.6 and 5.2 implicitly assumes the boundary is finite; the enumeration y_1,...,y_m and the convergence Lemma 5.4 need that. In the random planar map application the boundary is finite, so this does not damage the conjecture, but the general definition should be made more careful.\n\nWho this is for: people working on uniform spanning forests, random interlacements, or random planar maps. I would send it to a serious referee. The central contribution is solid and the edges are fixable. I expect acceptance after revision.\n\nBest,\n\n[your name]","headline":"A genuinely new random-walk construction of the free spanning forest, with a clean proof for locally finite graphs and an abstract that overstates that scope.","tokens_in":46064,"tokens_out":3924,"would_cite":true,"duration_ms":39752,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60J45","31C20","05C81","60J27"],"pacs":[],"model":"deepseek-v4-flash","headline":"An infinite-graph random walk that reaches infinity and reflects back is constructed, and its Aldous–Broder forest is shown to be the free uniform spanning forest.","keywords":["random walk reflected off infinity","energy-minimizing harmonic functions","free uniform spanning forest","Aldous-Broder algorithm","Wilson's algorithm","Tutte embedding","supercritical Liouville quantum gravity","discrete Gaussian free field"],"falsifier":"Take a transient graph with a nontrivial boundary at infinity, for instance two copies of $Z^{3}$ joined by a single edge, and run the reflected walk with any rate w≥w* from a fixed vertex; record the first vertex of a fixed finite set A hit by the walk over many trials. If the empirical hitting frequencies do not converge to the energy-minimizing harmonic measure hm^z_A(y), then Property (vi) of Theorem 1.5 fails and the construction is not the claimed one.","tokens_in":45115,"feed_emoji":"🌲","tokens_out":9113,"duration_ms":82714,"temperature":0.7,"pith_summary":"The paper constructs a continuous-time random walk on any countably infinite, transient graph that almost surely reaches infinity in finite time and bounces back infinitely often. The walk is canonical: its hitting distribution on every finite set is the unique energy-minimizing harmonic measure, and under a sufficiently large vertex rate function it is unique in law. Because the walk is recurrent, the Aldous–Broder algorithm can be run on a single infinite sample path, and the paper proves the resulting forest is the free uniform spanning forest; Wilson's algorithm gives the same forest only when that forest is a single tree. The same machinery defines a Green's function and a discrete Gaussian free field with free boundary at infinity, and a Tutte embedding for infinite planar maps with infinitely many ends, which the paper conjectures converges to supercritical Liouville quantum gravity.","feed_headline":"Random walk that bounces off infinity builds spanning forests","feed_subtitle":"A new infinite-graph process samples the free uniform spanning forest and points toward supercritical quantum gravity","key_machinery":"The carrying object is the energy-minimizing discrete harmonic function: for a finite set A and boundary data φ, the unique function agreeing with φ on A that minimizes the Dirichlet energy over all of V_G. Its values define the harmonic measure hm^x_A(y)=h_y(x), which Property (vi) makes the hitting law of the reflected walk. These functions define transition probabilities of finite approximating Markov chains on one-neighbourhoods B1G_n; the consistency of these chains on nested sets lets the authors couple all approximations and take a limit to get the continuous-time process. Local finiteness is then used to make the Aldous–Broder parent map well-defined.","core_discovery":"The paper's central claim is that on any countably infinite connected graph with finite vertex conductances and transient simple random walk, there is a rate function w* such that for every w≥w* and every starting vertex there is a unique-in-law continuous-time process X taking values in V_G∪{∞}, which is constant on intervals at vertices, is right-continuous, has exponential holding times with the graph's transition probabilities, satisfies the Markov property, returns to its starting point at arbitrarily large times, and has first hitting distribution of every finite set A equal to the energy-minimizing harmonic measure on A. The process reaches ∞ in finite time and reflects infinitely often, with the 'point at infinity' carrying information about which end was hit. From this process, the Aldous–Broder construction has the law of the c-free spanning forest, and a version of Wilson's algorithm agrees with the c-FSF exactly on the event that the c-FSF is a single tree; the Green's function and a discrete GFF with free boundary at infinity are defined, and a Tutte embedding is introduced for infinite maps with infinitely many ends.","pith_inferences":["A natural next step, not pursued by the paper, is to find a loop erasure that is itself allowed to reflect off infinity; such an object would make Wilson's algorithm produce the c-FSF without conditioning on connectedness.","The same energy-minimizing harmonic measure could define reflected processes in continuum settings, where the boundary at infinity is the Martin boundary and the discrete construction suggests a canonical 'reflected Brownian motion' on non-compact spaces that returns from infinity in finite time.","Numerically, one could test the predicted c=16 FUSF phase transition by simulating the reflected walk on the explicitly constructed supercritical maps and checking whether the Aldous–Broder forest is connected, since the walk visits every vertex almost surely.","The free-boundary GFF from Definition 1.20 may give a natural way to define embeddings of multi-ended maps via level-line geometry, which the paper does not explore."],"forward_implications":["Because the reflected walk is recurrent, the Aldous–Broder algorithm can be run forever on a single sample path, giving a path-by-path way to simulate and study the c-free spanning forest.","Wilson's algorithm with the reflected walk produces a spanning forest that agrees with the c-FSF if and only if the c-FSF is connected; on the non-connected event the unconditioned Wilson forest is not the FSF because it can create finite components.","The Green's function reflected off infinity satisfies the finite-graph identities: harmonicity, symmetry up to factors of π, and a Kirchhoff-type formula P[{x,y}∈FSF] = c(x,y) G_y(x,x)/π(x).","The Tutte embedding extends to infinite, multi-ended planar maps, with convex faces and finite-submap approximation, and the paper conjectures that under this embedding supercritical-LQG random planar maps converge to LQG decorated by CLE4.","On these maps the paper predicts sharp phase transitions: the free uniform spanning forest is connected for c>16 and has infinitely many components for c<16, and critical percolation occurs for c<95/4 but not for c>95/4."],"supporting_citations":[{"why":"Supplies the Aldous–Broder algorithm that generates the c-spanning tree of a finite graph from a random walk; Theorem 1.14 takes its infinite-graph limit.","marker":"[Bro89,Ald90]"},{"why":"Supplies Wilson's algorithm via loop-erased random walk; Theorem 1.16 is its reflected-walk analogue.","marker":"[Wil96]"},{"why":"Provides the definitions and properties of free and wired uniform spanning forests, including constructions of the wired forest.","marker":"[BLPS01]"},{"why":"Provides standard facts on the FSF, loop erasures, Green's functions, and effective resistance used in Sections 1.4 and 5.3.","marker":"[LP16]"},{"why":"Introduces the Tutte embedding of a finite planar graph with convex faces, which Lemma 5.5 extends to infinite maps.","marker":"[Tut63]"},{"why":"Defines the supercritical LQG/CLE4 coupling and poses Problem 4.4 on embeddings of multi-ended maps, which Conjecture 6.7 addresses.","marker":"[AG23]"}],"fun_headline_variants":["Walk bounces off infinity to grow spanning forests","Random walk reflecting off infinity yields free spanning forests","Infinity-bouncing walk builds spanning forests, probes quantum gravity","Reflect at infinity: random walk constructs free uniform spanning forests"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The spanning-forest results require G to be locally finite, because the Aldous–Broder parent map is defined as the vertex visited immediately before a first hitting time, and only local finiteness guarantees that such a vertex exists.","fun_headline_variants_meta":{"raw":{"variants":["Walk bounces off infinity to grow spanning forests","Random walk reflecting off infinity yields free spanning forests","Infinity-bouncing walk builds spanning forests, probes quantum gravity","Reflect at infinity: random walk constructs free uniform spanning forests"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001774,"raw_usage":{"total_tokens":7021,"prompt_tokens":995,"completion_tokens":6026,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":611,"completion_tokens_details":{"reasoning_tokens":5961}},"tokens_in":611,"tokens_out":6026,"duration_ms":41829,"temperature":1.0,"reasoning_tokens":5961,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T18:42:57.709770+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a transient graph with a nontrivial boundary at infinity, for instance two copies of $Z^{3}$ joined by a single edge, and run the reflected walk with any rate w≥w* from a fixed vertex; record the first vertex of a fixed finite set A hit by the walk over many trials. If the empirical hitting frequencies do not converge to the energy-minimizing harmonic measure hm^z_A(y), then Property (vi) of Theorem 1.5 fails and the construction is not the claimed one.","supporting_citations":[],"review_version":2}