{"id":"82937e5d-8fed-46e8-9d9f-9e9f01d45a1e","arxiv_id":"2509.01797","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Restricted odd Wick powers of the 2D GFF appear as coefficients of the half-integer logarithmic asymptotics of conformal-radius neighborhoods of first passage sets and sign clusters; even powers appear only after compensation.","lead":"Odd Wick powers of the 2D Gaussian free field can be restricted to first passage sets and excursion clusters, and these restricted fields give the correction terms in an asymptotic expansion of small neighborhoods of those fractal sets. The expansion, in half-integer powers of 1/log(1/epsilon), is the GFF analogue of Le Gall's Wiener sausage expansion, with a clean separation between odd and even powers.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 8.10, the excursion-cluster analogue of the main FPS expansion, rests on the explicitly unproved boundary-size condition (8.2); Theorem 5.1 itself is not affected by this gap.","rationale":"The reader's weakest-assumption identification is exactly the concern that survives my independent stress test. The FPS expansion Theorem 5.1 is the technically central result and appears to be supported by a coherent proof: the TVS conditioning, the six-term error decomposition, the tail bounds, and the choice of ε-dependent δ and b all fit together, and I found no obvious unproved step there. The excursion-cluster extension Theorem 8.10, however, explicitly depends on condition (8.2), whose proof is deferred and described as 'very standard' but lengthy. Since the abstract promises expansions for both first passage sets and excursion clusters, Theorem 8.10 is part of the central claim rather than a peripheral remark. The gap does not make the result false or the paper unsound; it makes the excursion-cluster half of the central claim conditional on a missing but very plausible estimate. Therefore the appropriate verdict remains CONDITIONAL, and my review does not require changing the reader's verdict.","tokens_in":83292,"tokens_out":5031,"duration_ms":62978,"concrete_test":"Prove (8.2) as a standalone lemma: for a CLE4 loop Γ (the outer boundary of an excursion cluster), establish E[Leb{z ∈ Int(Γ) : d(z,Γ) < ε}] ≤ C_a ε^{1/2-a} for every a > 0 (or directly o(|log ε|^{-β})). The natural route is to use the a.s. upper Minkowski dimension 3/2 of SLE4, cover Γ by O(ε^{-3/2+o(1)}) balls of radius ε, and bound the inner one-sided neighborhood by the full two-sided ε-neighborhood. If this covering argument cannot be made rigorous because the inner-side geometry of the SLE4 outer boundary is insufficiently controlled, then Theorem 8.10 should be restated as conditional on (8.2).","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim includes asymptotic expansions for both first passage sets and excursion clusters. The FPS part, Theorem 5.1, has a detailed proof and no apparent internal gap. The excursion-cluster part, Theorem 8.10, transfers the deterministic-domain proof to the random domain Int(Γ_j). Most domain-dependent estimates are monotone and carry over, but the boundary control provided by Lemma 5.15 for deterministic domains becomes precisely condition (8.2): E[Leb{z ∈ Int(Γ_j) : d(z, Γ_j) < ε}] = o(|log ε|^{-β}) for all β. The paper explicitly states that no proof of (8.2) is given, only that it should follow from the Hausdorff dimension 3/2 of SLE4 curves. This is load-bearing because the six-term error decomposition used in the proof of Theorem 5.1 cannot be controlled for random Γ_j without this condition; in particular, terms involving the inner ε-neighborhood of ∂Int(Γ_j) = Γ_j require faster-than-any-power-of-|log ε| decay. The condition is highly plausible from Minkowski-dimension considerations, so this is not an internal inconsistency, but it is an acknowledged missing lemma in a substantial part of the paper's headline claim.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper investigates the relation between Wick powers of the 2D Gaussian free field (GFF) and its first passage sets (FPS) and excursion clusters. For an FPS A, the author defines ψ_{n,A}=E[:Φ^n:|A] and shows that odd ψ_{2k+1,A} are generalized functions supported on A, while even ψ_{2k,A} require a non-integrable exterior compensation. The main result, Theorem 5.1, states an asymptotic expansion for the conformal-radius neighborhood N_ε(A) in half-integer powers of 1/|log ε|, with coefficients the restricted odd Wick powers; even powers do not appear. The proof uses conditioning on two-valued sets (TVS) and gives explicit error bounds. Section 7 gives a Gaussian multiplicative chaos description of ψ_{n,A}. Section 8 extends the decomposition to excursion clusters, and Theorem 8.10 states the analogue of Theorem 5.1 for individual clusters. Section 9 contains conjectures relating the expansion to Le Gall's Wiener sausage expansion and to SLE-loop measures.","tokens_in":83675,"tokens_out":6576,"duration_ms":74088,"significance":"If the main theorem is correct, it provides a new and surprisingly clean asymptotic link between fractal local sets of the GFF and Wick renormalization, with coefficients explicitly identified and with the odd/even distinction explained. The proof of Theorem 5.1 is a substantial technical achievement: it avoids explicit two-point correlation functions and instead uses TVS conditioning, with careful Sobolev-space estimates. The paper also makes a thought-provoking connection to umbral calculus and to Le Gall's Wiener sausage expansion. However, the excursion-cluster analogue, which is part of the advertised scope, is not proved to the same standard: it depends on an explicitly unproved boundary-decay condition (8.2). This prevents the paper from being fully acceptable in its current form.","major_comments":[{"comment":"Theorem 8.10 is stated as an unconditional analogue of Theorem 5.1 for excursion clusters, but its proof relies on the unproved bound (8.2): E[Leb{z∈Int(Γ_j): d(z,Γ_j)<ε}] = o(|log ε|^{-β}) for all β. The paper explicitly says that no proof is given and that this 'should follow' from the Hausdorff dimension 3/2 of SLE4 curves. This is load-bearing: in the proof of Theorem 5.1, Lemma 5.15 is used to control the boundary contribution in the six-term error decomposition. For the random domain Int(Γ_j), condition (8.2) is exactly the replacement for that lemma. A heuristic dimension count is not a proof, especially because (8.2) involves an expectation over random CLE4-type boundaries and must hold uniformly in the small-ε regime. The theorem should either be proved with a full argument for (8.2), or explicitly stated as conditional on (8.2).","section":"§8.2"},{"comment":"The transfer of the FPS results from deterministic domains to the random domains Int(Γ_j) is not fully justified. The paper says that 'it is easy to check that all the domain-dependent bounds are monotonic in the domains' and that Int(Γ_j)⊂D. This is not a complete proof: several estimates in Sections 3 and 5 (e.g., Corollaries 3.3–3.4 and Lemma 5.15) depend on distances to the boundary, on the conformal radius, and on H^{-η}(C) norms that are not obviously monotone under random domain reduction. Since Γ_j is random and correlated with the field, one needs a uniform-in-domain argument or a separate proof for the random case. This gap is closely related to (8.2), and together they undermine the current formulation of Theorem 8.10.","section":"§8.2"},{"comment":"Section 6 is explicitly heuristic, but it contains statements labelled as 'Claim' rather than 'Conjecture'. Claim 6.1 (the coupling with positive probability) and Claim 6.3 (the asymptotic expansion with ε replaced by a deterministic function) are used to formulate the consistency principle, but no proofs are given and the text says 'we leave the details for the reader'. Since these claims are not used in the proof of Theorem 5.1, this is not a fatal issue, but it is misleading to label them as claims. They should either be proved or explicitly labelled as heuristic assumptions.","section":"§6.1"}],"minor_comments":[{"comment":"The constant in the bound (5.17) appears to be off by a factor of π: from the displayed estimate P_0(T_{-b,b}>t) ≤ (4/π) e^{-π^2t/(8b^2)}, taking square roots gives a prefactor 2/√π, not 2√π. The error is harmless because only the exponential rate matters, but the constants should be corrected or explained.","section":"§5.3"},{"comment":"In part (1) the limit is written as 'lim_{q→0}', but from the proof and from definition (4.9) the intended limit is q→∞. Please fix.","section":"§4.3"},{"comment":"The paper contains several 'we omit the details' or 'tedious but standard' statements (e.g., in Propositions 4.18, 8.4, 8.12). Most are believable, but in a paper of this length it would help the reader if the omitted computations were at least summarized in an appendix or in footnotes with the main algebraic steps.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The FPS part, and particularly Theorem 5.1, appears technically sound and is a strong contribution. The main risk is the excursion-cluster analogue: Theorem 8.10 is conditional on an unproved estimate (8.2) that the author acknowledges but does not establish. If the author can supply a proof of (8.2), or alternatively restate Theorem 8.10 as a conditional theorem with the condition prominently displayed, I would view the paper as close to acceptable. The Section 6 'Claims' should also be demoted to heuristics or proved."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague, quick read. The core of this paper, Theorem 5.1, is genuinely new and, as far as I can tell, correct: the indicator of a conformal-radius epsilon-neighborhood of an FPS expands in half-integer powers of 1/|log epsilon|, with coefficients given by the restricted odd Wick powers. The proof is long but careful, going through TVS conditioning and keeping explicit error bounds. The FPS decomposition theorems in Section 4 and the GMC description in Section 7 also look solid. This is the GFF analogue of Le Gall's Wiener sausage expansion, with the expected change of exponents. No circularity: it builds on ALS20a/ALS22 as external inputs, and Le Gall's expansion is used only for comparison.\n\nWhat is new: the half-integer powers, the odd-only support, and the compensation picture for even Wick powers. The section on umbral calculus is a nice algebraic wrapper, though not load-bearing.\n\nSoft spots, in proportion. The excursion-cluster analogue, Theorem 8.10, depends on condition (8.2): the area of the epsilon-neighborhood of an SLE4 outer boundary decays faster than any power of |log epsilon|. The paper explicitly states this is not proved and says it should follow from Hausdorff dimension 3/2. That condition is load-bearing for the cluster half of the abstract claim. It is likely true, but it is a missing lemma, and a potentially lengthy one. The FPS half is unaffected. Section 6 is explicitly heuristic and not used in the main proofs. The many conjectures are clearly labeled.\n\nWho this is for: people working in SLE/CLE, GFF renormalization, and 2D CFT. I would send it to a serious referee, but with the instruction to either supply a proof of (8.2) or state Theorem 8.10 as conditional. My own verdict: Thm 5.1 is a clear yes; the excursion-cluster theorem needs work. I would cite this paper for Thm 5.1.","headline":"The FPS expansion (Thm 5.1) is a well-proved new result, but the excursion-cluster analogue (Thm 8.10) rests on an unproved boundary-size condition the paper acknowledges.","tokens_in":84062,"tokens_out":1761,"would_cite":true,"duration_ms":23714,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60J67","60G60","60J65"],"pacs":[],"model":"deepseek-v4-flash","headline":"For the 2D Gaussian free field, odd Wick powers can be restricted to first passage sets and excursion clusters, and the small conformal-radius neighborhoods of those sets expand in half-integer powers of 1/|log eps|, with coefficients given","keywords":["Gaussian free field","first passage set","excursion cluster","Wick power","conformal radius","asymptotic expansion","two-valued set","SLE4"],"falsifier":"Measure the area of the conformal-radius neighborhood $N_\\epsilon(A)$ for a first passage set in a discretized GFF, at several $\\epsilon$, and compare the leading coefficient with the Minkowski content of $A$ and the $|\\log \\epsilon|^{-3/2}$ coefficient with the predicted expression involving $\\psi_{3,A}$. Alternatively, compute $\\mathbb{E}[\\mathbf{1}_{N_\\epsilon(A)}]$ directly from the Brownian hitting-time density: the expansion predicts exactly $\\mathbb{E}[\\mathbf{1}_{N_\\epsilon(A)}] = \\sum (-1)^k v^{2k+1}/(\\sqrt{2\\pi} 2^k k!(k+1/2)) (2\\pi|\\log \\epsilon|)^{-(k+1/2)}$; any $|\\log \\epsilon|^{-1}$ term or contribution from an even Wick power would disprove Theorem 5.1.","tokens_in":1895,"feed_emoji":"🌀","tokens_out":2012,"duration_ms":93202,"temperature":0.7,"texified_at":"2026-08-05T20:21:04.885708+00:00","pith_summary":"This paper claims that the small-scale geometry of the 2D Gaussian free field's random level components is controlled by its odd Wick powers. For a first passage set or excursion cluster, the indicator of its conformal-radius epsilon-neighborhood admits an $L^2$ asymptotic expansion in half-integer powers of $1/|\\log \\epsilon|$, and the coefficients are the restrictions of the odd Wick powers to the fractal set. The even Wick powers are provably absent from these expansions. The paper also proves that odd Wick powers can be restricted to these non-thin random sets, while even powers diverge on them and need an external compensating term. If correct, this gives a direct geometric route to the odd Wick powers and explains why the two parity classes of renormalized field powers behave so differently.","texify_model":"deepseek-v4-flash","texify_usage":{"total_tokens":5817,"prompt_tokens":986,"completion_tokens":4831,"prompt_tokens_details":{"cached_tokens":0},"prompt_cache_hit_tokens":0,"prompt_cache_miss_tokens":986,"completion_tokens_details":{"reasoning_tokens":3839}},"feed_headline":"Half-integer log powers govern GFF cluster neighborhoods","feed_subtitle":"Neighborhoods of first-passage sets expand in |log eps|^{-(k+1/2)}; even Wick powers never appear.","key_machinery":"The argument is carried by the two-variable Hermite polynomials $Q_n(x,u) = u^{n/2}H_n(xu^{-1/2})$ and the change-of-variance identity $Q_n(x,u_1+u_2) = \\sum (-1)^k n!/[2^k k!(n-2k)!] Q_{n-2k}(x,u_1) u_2^k$. This identity re-expresses Wick powers when the domain is cut by a two-valued set, and it forces the combinatorial coefficients $(-1)^k/[2^k k!(k+1/2)]$. Conditioning on a two-valued set reduces the conformal-radius neighborhood indicator to a first-hitting time of a one-dimensional Brownian motion, whose density produces the half-integer powers after expansion. To go from two-valued sets to first passage sets, the proof controls the interchange of $\\epsilon$ to 0 with the level of the two-valued set usin","core_discovery":"For a first passage set $A$ of the 2D GFF with constant boundary condition, write $N_\\epsilon(A) = \\{z \\in D\\setminus A : \\operatorname{CR}(z,D\\setminus A) < \\epsilon \\operatorname{CR}(z,D)\\}$. Theorem 5.1 asserts that, in $L^2(dP, H^{-\\eta})$, $\\mathbf{1}_{N_\\epsilon(A)} = (2\\pi)^{-1/2} \\sum_{k=0}^{N} (-1)^k [2^k k!(k+1/2)]^{-1} \\psi_{2k+1,A} (2\\pi|\\log \\epsilon|)^{-(k+1/2)} + R_{N,\\epsilon}$, with $\\mathbb{E}[\\|R_{N,\\epsilon}\\|^2_{H^{-\\eta}}]^{1/2} = o(|\\log \\epsilon|^{-(N+1/2)})$. Here $\\psi_{2k+1,A}$ are the conditional expectations of the odd Wick powers $\\colon \\Phi^{2k+1} \\colon$ given $A$, hence generalized functions supported on $A$. Corollary 1.3 extracts each $\\psi_{2n+1,A}$ as a limit of linear combinations of indicators at multiple scales, and Section 8 transfers the same expansion to each excursion cluster, yielding a full cl","pith_inferences":["If the loop-soup conjecture in Section 9.4 is correct, the exponents |log eps|^{-(n-c/2)} interpolate between the GFF result (c=1) and the Wiener sausage (c to 0), suggesting a one-parameter family of renormalized fractional powers of the loop-soup occupation field for every central charge c in (0,1).","The geometric multi-scale extraction could serve as an alternative definition of odd Wick powers on non-thin sets, bypassing Hermite-polynomial regularization; testing this in a metric-graph approximation of the GFF would be a natural numerical check.","The absence of even powers suggests a general selection rule: local-set functionals whose leading area is |log eps|^{-1/2} couple only to observables that are supported on and signed on the set, a principle that may extend to other GFF local sets and loop-soup clusters."],"forward_implications":["The odd Wick powers of the 2D GFF are determined by the geometry of first passage sets and excursion clusters alone, and can be recovered by multi-scale counting of small conformal-radius neighborhoods.","For even powers, no version supported on these non-thin sets exists; any convergent renormalized even power must include a smooth compensating function outside the set that diverges non-integrably near the set.","The full collection of excursion clusters gives an orthogonal-in-law decomposition of every Wick power, generation by generation, so renormalized powers are assembled from independent cluster contributions.","The exponents differ from the Wiener-sausage case: half-integer instead of integer, so two-dimensional log-scaling carries a parity-dependent structure not visible in Brownian self-intersection local time expansions.","The conformal-radius expansion and the Euclidean-distance version conjectured in Section 9.1 are connected through a scale-invariant loop measure expected to describe microscopic holes of the first passage set."],"supporting_citations":[{"why":"Constructs two-valued sets and gives the hitting-time law of the conformal-radius ratio used to condition the expansion.","marker":"[ASW19]"},{"why":"Introduces first passage sets, identifies the Minkowski-content measure, and gives the Brownian hitting law for the FPS conformal-radius ratio.","marker":"[ALS20a]"},{"why":"Supplies the joint law involving the extremal distance to the two-valued set's surrounding loop, used for the exponential error bounds.","marker":"[ALS22]"},{"why":"Relates first passage sets to clusters of Brownian loops and excursions, motivating the analogy with the Wiener sausage and the loop-soup picture.","marker":"[ALS20b]"},{"why":"Provides the reference Wiener-sausage expansion with integer powers that the new half-integer expansion is compared against.","marker":"[LG90, LG92]"},{"why":"Supplies the dimension-3/2 fact invoked for the shell-area estimate (8.2) used for excursion clusters.","marker":"[Bef08]"},{"why":"Provides the Gaussian multiplicative chaos change-of-domain identity used to express conditional Wick powers as germs at gamma to 0.","marker":"[APS20]"},{"why":"Define the height gap and level-line structure underlying the GFF level-set decomposition used throughout.","marker":"[SS09, SS13]"}],"fun_headline_variants":["GFF cluster neighborhoods expand in half-log powers","Odd Wick powers surface in GFF cluster expansions","Even Wick powers absent from GFF cluster neighborhoods","Half-integer log terms govern 2D GFF cluster geometry","Excursion clusters yield odd Wick power asymptotics"],"cache_read_input_tokens":85888,"weakest_assumption_plain":"For the excursion-cluster version, the proof assumes without a complete proof that the area of a thin shell around an SLE4 cluster boundary decays faster than any negative power of $|\\log \\epsilon|$; the paper derives this estimate from the known fractal dimension of the boundary but does not write the full proof.","fun_headline_variants_meta":{"raw":{"variants":["GFF cluster neighborhoods expand in half-log powers","Odd Wick powers surface in GFF cluster expansions","Even Wick powers absent from GFF cluster neighborhoods","Half-integer log terms govern 2D GFF cluster geometry","Excursion clusters yield odd Wick power asymptotics"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000178,"raw_usage":{"total_tokens":1272,"prompt_tokens":1024,"completion_tokens":248,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":768,"completion_tokens_details":{"reasoning_tokens":173}},"tokens_in":768,"tokens_out":248,"duration_ms":3722,"temperature":1.0,"reasoning_tokens":173,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-05T12:10:33.351923+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure the area of the conformal-radius neighborhood $N_\\epsilon(A)$ for a first passage set in a discretized GFF, at several $\\epsilon$, and compare the leading coefficient with the Minkowski content of $A$ and the $|\\log \\epsilon|^{-3/2}$ coefficient with the predicted expression involving $\\psi_{3,A}$. Alternatively, compute $\\mathbb{E}[\\mathbf{1}_{N_\\epsilon(A)}]$ directly from the Brownian hitting-time density: the expansion predicts exactly $\\mathbb{E}[\\mathbf{1}_{N_\\epsilon(A)}] = \\sum (-1)^k v^{2k+1}/(\\sqrt{2\\pi} 2^k k!(k+1/2)) (2\\pi|\\log \\epsilon|)^{-(k+1/2)}$; any $|\\log \\epsilon|^{-1}$ term or contribution from an even Wick power would disprove Theorem 5.1.","supporting_citations":[],"review_version":1}