{"id":"cbd3f7db-cee1-4e7e-be56-da58791fb671","arxiv_id":"2502.09853","paper_version":1,"verdict":"UNVERDICTED","confidence":"HIGH","novelty_score":2.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"Expanded lecture notes proving that unvisited sites of a 2D simple random walk at sub-cover times converge to the thick points of the discrete Gaussian free field.","lead":"These lecture notes explain a recent theorem: the set of points that a two-dimensional random walk has not yet visited by a fixed fraction of the cover time converges to a random fractal measure. The notes connect this to the Gaussian free field, showing the unvisited points match the field's high-value 'thick points'.","discovery_kind":"review","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified: Theorem 1.7 is a cited theorem from Abe-Biskup, and the notes' proof sketches are consistent with the published arguments; all deferred steps are explicit and referenced.","rationale":"The reader correctly identified the admissibility hypothesis as the key technical premise, and I agree that Theorem 1.7 should not be read beyond that class of domains. However, because the paper explicitly restricts to admissible domains and the theorem is an established result from the cited literature, I do not regard this as a load-bearing objection to the notes. I also checked the internal logic of the main proof: the normalization identities (1.34)-(1.35) and (4.32)-(4.33) are consistent; the measure mu in Theorem 4.8 indeed has the claimed Laplace transform; and Lemma 4.9 controls the small-positive local time contributions needed to pass from the extended process to the avoided-point measure. The only notable caveat is that the proof of Theorem 1.5 is presented only for lambda < 1/sqrt(2), while Theorem 1.7 for theta > 1/2 invokes the complementary regime; the notes flag this and cite [7,12]. This affects self-containedness, not correctness. I also note the stray copy-pasted block in Section 5.1 containing text from another paper; it is a typographical artifact and does not affect the mathematics. Overall, the reader's UNVERDICTED verdict remains appropriate: the submission is a lecture-note exposition of known results, not a research preprint, and no mathematical defect was found.","tokens_in":35927,"tokens_out":9716,"duration_ms":97370,"concrete_test":"Verify the deferred truncation argument for Lemma 4.5 by consulting Abe-Biskup [2, Lemma 7.1] and checking that it proves the reverse inequality (4.28) uniformly in s in [0,1]; this is the single step the notes explicitly defer, and its validity closes the only gap in the lecture-notes derivation of Theorem 1.7.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim, Theorem 1.7, is not a new research assertion; it is a theorem of Abe and Biskup [2], and the notes present it as a lecture-note exposition. The proof strategy is coherent: use the Second Ray-Knight coupling (Theorem 2.7) to convert avoided points of the random-walk local time into thick points of a DGFF, identify the subsequential limit via the extended process (Lemma 4.5), and then remove the absolutely continuous part of the limit measure using Lemma 4.9. Each step is either proved in the notes or explicitly deferred to the cited literature. In particular, the restriction to admissible domains (Definitions 1.3 and 1.4) is exactly the setting in which the Green-function asymptotics of Theorem 2.1 and the subsequent moment estimates (Lemmas 3.5, 3.6) are valid; the paper does not claim the theorem for non-admissible domains. The one potentially load-bearing gap is that the proof of Theorem 1.5 is only carried out for lambda < 1/sqrt(2), while Theorem 1.7 for theta > 1/2 uses lambda = sqrt(theta) > 1/sqrt(2); however, the notes explicitly state this restriction and refer to [7,12] for the full result, and the full theorem is established in the cited papers. Thus there is no internal inconsistency and no reason to doubt the mathematical claim.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"These expanded lecture notes provide an expository account of the recent theorem of Abe and Biskup (Theorem 1.7) that, for a simple random walk on an admissible lattice approximation of a bounded planar domain, the set of points still unvisited at a θ-fraction of the cover time, parametrized by local time at the boundary vertex and normalized by N^2 e^{-t_N/(g log N)}, converges in law to the √θ-thick point measure of the discrete Gaussian free field. The notes develop the necessary background: Green function asymptotics with the conformal radius, the second Ray-Knight theorem via the Kac moment formula, the Gibbs-Markov property of the DGFF, the existence and characterization of thick-point limits, and finally the thinning argument that converts avoided points into DGFF thick points. Theorems are stated in the full generality of the original papers, while several technical steps are explicitly marked as deferred to the literature.","tokens_in":1032,"tokens_out":1133,"duration_ms":74912,"significance":"If the result holds—and it is an already published theorem in [2]—these notes make a technically demanding circle of ideas accessible to a graduate-level audience. The main strength is transparency: the author repeatedly and explicitly marks where a proof is sketched, where a step is omitted, and where a cited result is used (see, e.g., Section 4.3 'modulo a technical step', Section 3.2 'only carry out the proof under the assumption λ < 1/√2', and Lemma 3.8 'Proof (idea)'). This is exactly the right tone for lecture notes and makes the document a valuable companion to [7] and [12]. The paper also records open questions and conjectures (Sections 5.2–5.3) that may stimulate further research. For a research journal, the novelty is limited because the central theorems are not new, but as an expository contribution the notes are of high quality and fill a pedagogical gap.","major_comments":[],"minor_comments":[{"comment":"There is a typo in 'Alfr ´ed R ´enyi Insititute'; it should read 'Institute'.","section":"Title page"},{"comment":"The definition of r_D(x) via the integral over the boundary is followed by a remark that it coincides with the conformal radius for simply connected D; a one-line justification or a precise reference would help the reader appreciate why this quantity is natural.","section":"§2.1, Eq. (2.3)"},{"comment":"The proof is explicitly 'modulo a technical step' and refers to [2, Lemma 7.1]; since this lemma is the key bridge between the local-time measure and the extended process, the introduction to Lecture 4 should state at the outset that a complete proof of this step is deferred to [2].","section":"§4.3, Lemma 4.5"},{"comment":"The event in the probability is typeset awkwardly ('Pϱ d τcov/deg(DN) ≤ ...'); adding parentheses to make the event unambiguous would improve readability.","section":"§5.3, Eq. (5.20)"},{"comment":"The surname of the coauthor of [27] is misspelled as 'Fitzimmons' both in the bibliography and in the body text; it should be 'Fitzsimmons'.","section":"References"}],"recommendation":"minor_revision","confidential_remarks":"The paper contains no new mathematical results; its value is purely expository. If the journal's scope is strictly limited to original research, the editor may consider whether lecture-notes-style submissions fit. That said, the exposition is careful, the caveats are honestly stated, and the author is a leading expert in the area, so the editorial decision should hinge on fit rather than correctness."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The one thing to know: this is not a research paper. It is an expanded minicourse presenting the author's own theorems (Biskup–Louidor on DGFF thick points, Abe–Biskup on avoided points of 2D random walk) with proof sketches, exercises, and open problems. There are no new theorems, and the paper says so outright. If you pick it up expecting new results, you'll be disappointed. But as lecture notes, it's very good.\n\nThe notes' real value is pedagogical. The route from the Second Ray–Knight coupling to the avoided-point limit is laid out coherently: set up the extended process, identify subsequential limits, factor the measure, then remove the absolutely continuous part to isolate the atom at zero. The author is honest about which steps are sketched: the proof of Theorem 1.5 is carried out only for lambda < 1/sqrt(2), and the key technical step in Lemma 4.5 is deferred to [2]. That is acceptable in this format. The figures are helpful, and the appendix gives a nice summary of related results and conjectures (frequent points, cover time).\n\nSoft spots, in proportion: the novelty is zero by design, so the paper's worth depends entirely on the quality of the exposition, which is high. The deferred technical steps are flagged explicitly and point to the right references, so a reader can fill gaps. The admissibility conditions on the domain are restrictive but stated clearly; the proofs do not overclaim. The citation pattern is fine — the author cites his own prior work because the results are his, and also gives proper credit to Eisenbaum–Kaspi–Marcus–Rosen–Shi, Dynkin, etc.\n\nWho is this for? Graduate students or probabilists entering the area who want a fast but reliable introduction to the thick-point/avoided-point connection. It would make a good reading group source. For a journal, the right question is whether the venue publishes expository lecture notes. If yes, this deserves a serious referee — the exposition is subtle and easy to get wrong, so an expert check is worthwhile. If the venue only takes original research, desk reject is defensible, but I would not call the paper deficient on those grounds.\n\nI would bring it to a reading group, but I wouldn't cite it in my own work beyond a background reference; I'd cite the original papers instead.","headline":"Well-crafted lecture notes on a major result in 2D random walk local time and DGFF thick points, but no new mathematics; review as exposition, not research.","tokens_in":36742,"tokens_out":1725,"would_cite":false,"duration_ms":19449,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60J55","60G60","60F05","60J65"],"pacs":[],"model":"deepseek-v4-flash","headline":"For a planar simple random walk, the set of points still unvisited when the boundary clock reaches a θ-fraction of the cover time converges, after rescaling, to the same random fractal measure that describes the √θ-thick points of the…","keywords":["random walk local time","avoided points","cover time","discrete Gaussian free field","thick points","Second Ray-Knight theorem","two-dimensional simple random walk","admissible domain"],"falsifier":"Simulate a simple random walk on a large square lattice approximation, run it until the boundary local time reaches t_N with θ=1/2, and record the avoided points; if the normalized empirical measure does not converge (along a sequence of squares) to a diffuse measure with total mass matching E[$Z_D^{{1/√2}}$(D)]—or if the set fails to have Hausdorff dimension near 2(1-1/2)=1—the central equivalence with DGFF thick points would be refuted.","tokens_in":35713,"feed_emoji":"🎲","tokens_out":3015,"duration_ms":33697,"temperature":0.7,"pith_summary":"This paper establishes that, for a two-dimensional simple random walk on a lattice approximation of a planar domain, the spatial distribution of the points not yet visited at a fixed fraction θ of the cover time has a non-trivial scaling limit. Time is measured by the local time accumulated at a distinguished boundary vertex rather than by the walk's own clock. In that parametrization, the normalized counting measure of avoided points converges in law to the measure $Z_D^{{√θ}}$ that describes the √θ-thick points of the discrete Gaussian free field. This connects an extremal quantity of the random walk trajectory to a canonical logarithmically correlated Gaussian field. The course lectures develop the proof through the Second Ray-Knight theorem, Kac moment formulas, and first- and second-moment estimates of the avoided-point process.","feed_headline":"2D random walk avoided sites match DGFF thick points","feed_subtitle":"At a θ-fraction of cover time, unvisited points converge to the √θ-thick point measure of the discrete Gaussian free field.","key_machinery":"The load-bearing identity is the Second Ray-Knight theorem, which couples the boundary-time-parametrized local time L_t of the walk with two independent discrete Gaussian free fields h and h̃ on the same graph through L_t + ½h² = ½(h̃ + √(2t))². This turns zero-local-time points into points where h̃ is near -√(2t), i.e., into thick points of a DGFF. Supporting machinery includes the Kac moment formula for Laplace transforms of local time, the Gibbs-Markov property of the DGFF (harmonic decomposition of the field), Green-function asymptotics with the conformal radius r_D(x), and an extended process that tracks both small local time and small field values, whose joint convergence is established by first- and second-moment estimates.","core_discovery":"The central claim is Theorem 1.7: for admissible lattice approximations D_N of an admissible planar domain D, if t_N is chosen so that θ = lim t_N/(2g(log N)^2) lies in (0,1), then the measure $N^{{-2}}$$e^{{t_N/(g log N)}}$ times the sum of point masses at vertices with zero local time at the boundary-parametrized time t_N converges in law to $Z_D^{{√θ}}$. The punchline is that avoided points at a θ-fraction of the cover time are asymptotically distributed exactly as the √θ-thick points of the DGFF, where a point is thick if the field value exceeds roughly 2√(gλ) log N. Because the normalization exponent is $N^{{2(1-θ)+o(1)}}$, the measure vanishes when θ>1, consistent with the cover-time scale. The result also yields a limit law for the cardinality of the avoided set, normalized by the same factor, converging to the total mass of $Z_D^{{√θ}}$.","pith_inferences":["A direct corollary not spelled out in the notes is that the fractal dimension of the avoided set should be 2(1-θ) almost surely, matching the known carrier dimension of Z_D^λ; this is a testable prediction for simulations in the boundary-time parametrization.","Under the natural (walk) time parametrization, the limit becomes a differently tilted measure built from the DGFF conditioned to have zero spatial average; this shows that the boundary-clock parametrization is not just a convenience but changes the answer.","The argument is restricted to the wired-boundary construction via a single boundary vertex; extending the same statement to free boundary conditions or the lattice torus, where the Second Ray-Knight coupling is unavailable, remains an open problem suggested by the notes."],"forward_implications":["At a θ-fraction of the boundary-clock cover time, the avoided points of a 2D random walk form the same random fractal, in law, as the √θ-thick points of the DGFF; their normalized number converges to the total mass of Z_D^{√θ}.","For θ>1, the avoided set is asymptotically empty, pinpointing the leading order of the cover time in this parametrization.","The same machinery yields distributions for other exceptional level sets of the local time, including λ-thick and λ-thin points, again described by the measures Z_D^λ.","The result exhibits universality of the DGFF for extremal problems of logarithmically correlated processes in two dimensions."],"supporting_citations":[{"why":"Original joint paper of Abe and Biskup that proved Theorem 1.7, the central claim of these notes.","marker":"[2]"},{"why":"Biskup-Louidor theorem on intermediate level sets of the DGFF that supplies the limit measure Z_D^λ and its characterization.","marker":"[12]"},{"why":"Eisenbaum-Kaspi-Marcus-Rosen-Shi Second Ray-Knight theorem, the key distributional identity connecting local time to DGFF.","marker":"[25]"},{"why":"Zhai's coupling that upgrades the Ray-Knight identity to an almost-sure coupling used in the proof of Lemma 4.6.","marker":"[44]"},{"why":"The author's PIMS lecture notes providing full technical details for the thick-point convergence, first and second moment computations.","marker":"[7]"},{"why":"Lawler-Limić monograph supplying the potential kernel framework and asymptotic estimates behind the Green function asymptotics.","marker":"[34]"},{"why":"Stöhr's early result on the asymptotic expansion of the lattice potential kernel that fixes the constants g and c0.","marker":"[42]"},{"why":"Kahane's criterion used to prove uniqueness of the Gaussian multiplicative chaos measure Z_D^λ in the factorization step.","marker":"[32]"}],"fun_headline_variants":["Random walk gaps match DGFF thick points","Avoided sites mimic DGFF thick points","2D walk holes equal DGFF thick points","Local time zeros become thick points","Random walk skipped points follow DGFF"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument collapses if the Green-function asymptotics of Theorem 2.1 fail, which requires the domain and its lattice approximations to be admissible; rough boundaries or non-admissible discretizations can break the conformal-radius term r_D(x) and the logarithmic covariance structure, so the limiting measure would not have the stated form.","fun_headline_variants_meta":{"raw":{"variants":["Random walk gaps match DGFF thick points","Avoided sites mimic DGFF thick points","2D walk holes equal DGFF thick points","Local time zeros become thick points","Random walk skipped points follow DGFF"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.0001,"raw_usage":{"total_tokens":928,"prompt_tokens":763,"completion_tokens":165,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":379,"completion_tokens_details":{"reasoning_tokens":101}},"tokens_in":379,"tokens_out":165,"duration_ms":2170,"temperature":1.0,"reasoning_tokens":101,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T20:17:30.846969+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Simulate a simple random walk on a large square lattice approximation, run it until the boundary local time reaches t_N with θ=1/2, and record the avoided points; if the normalized empirical measure does not converge (along a sequence of squares) to a diffuse measure with total mass matching E[$Z_D^{{1/√2}}$(D)]—or if the set fails to have Hausdorff dimension near 2(1-1/2)=1—the central equivalence with DGFF thick points would be refuted.","supporting_citations":[{"cited_title":"Abe and M","cited_arxiv_id":null,"evidence_quote":"Original joint paper of Abe and Biskup that proved Theorem 1.7, the central claim of these notes."},{"cited_title":"Biskup and O","cited_arxiv_id":null,"evidence_quote":"Biskup-Louidor theorem on intermediate level sets of the DGFF that supplies the limit measure Z_D^λ and its characterization."},{"cited_title":"Eisenbaum, H","cited_arxiv_id":null,"evidence_quote":"Eisenbaum-Kaspi-Marcus-Rosen-Shi Second Ray-Knight theorem, the key distributional identity connecting local time to DGFF."},{"cited_title":"Zhai (2018)","cited_arxiv_id":null,"evidence_quote":"Zhai's coupling that upgrades the Ray-Knight identity to an almost-sure coupling used in the proof of Lemma 4.6."},{"cited_title":"Lawler and V","cited_arxiv_id":null,"evidence_quote":"Lawler-Limić monograph supplying the potential kernel framework and asymptotic estimates behind the Green function asymptotics."},{"cited_title":"St ¨ohr (1950)","cited_arxiv_id":null,"evidence_quote":"Stöhr's early result on the asymptotic expansion of the lattice potential kernel that fixes the constants g and c0."},{"cited_title":"Kahane (1985)","cited_arxiv_id":null,"evidence_quote":"Kahane's criterion used to prove uniqueness of the Gaussian multiplicative chaos measure Z_D^λ in the factorization step."}],"review_version":1}