{"id":"3b8390e1-5d48-47a1-a29a-cda388e023d0","arxiv_id":"2607.27786","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Explicit theta-function formulas give the probability that all GFF level lines cross an annulus, with a new sqrt(r) correction as the inner hole shrinks.","lead":"This paper computes, by explicit formula, the probability that all level lines of a Gaussian free field cross a ring-shaped domain rather than stopping on its outer edge. The result is the first exact crossing-probability statement for GFF level lines in a doubly connected domain, and it exposes a polynomial sqrt(r) correction to the known exponential decay.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"BPZ verification rests on unverified ChatGPT-generated algebra in Appendices A-B; a coefficient error there would destroy the martingale property and with it the terminal value (5.20) behind Theorem 1.2.","rationale":"The reader judged the paper CONDITIONAL and flagged the ChatGPT-compiled appendices as the main reason; I agree that the unverified algebra in Appendices A-B is a load-bearing correctness risk. I differ from the reader's stated weakest_assumption: the named assumption (Lemma 4.2 terminal bounds) is also sketchy, but the algebraic verification of Prop. 1.1 is more central, since it underpins the martingale property for both partition functions and feeds directly into the terminal computation via (3.37)/(4.6). The paper is otherwise carefully written, uses independent published inputs ([PW19], Lawler/Zhan), and contains no evident internal contradiction. No reason to change the reader's CONDITIONAL verdict.","tokens_in":65914,"tokens_out":11479,"duration_ms":113497,"concrete_test":"Independently re-derive identity (4.6) for n=2 and n=3, using a computer algebra system (e.g., SymPy) that expands G_ε in (4.4) after substituting the product forms (1.5) and checks double-periodicity, pole cancellation, and the degeneration limit. Also re-compute the Itô drift calculation (3.37) symbolically, verifying that the q_j/p_j^2 terms vanish and that the remaining modulus terms are exactly a_j^2 F_n(R_t)-F_1(s_j). If either check yields a nonzero residual, Theorem 1.2's martingale argument collapses; if both pass, the main remaining risk is Lemma 4.2's terminal estimates.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The route to Theorem 1.2 is: (i) Z_n-ann and Z^{(ℓ)}_{n-cro} solve the annulus BPZ system (Prop. 1.1); (ii) this makes the ratio M_t in (5.15) a bounded martingale; (iii) Lemma 4.2 gives the terminal value, reducing to the [PW19] rainbow probability. Step (i) is the algebraic core. Its proof reduces to the identity (4.6): G_± = G_+ = -(3/4)(n-1)E(r), established in Appendix B by double-periodicity (Lemma B.1) and pole-cancellation (Lemma B.2). Appendix B opens: 'The following calculations are compiled from ChatGPT 5.5 Pro (OpenAI)'s computation drafts.' Appendix A, which produces the crucial drift formula (3.37), carries the same disclaimer. The long elliptic-function identities involve cancellations of singular terms in G_ε and the evaluation of the constant via degeneration; these are exactly the places a wrong sign or coefficient would be invisible to a casual check. A single such error would make M_t a local super/sub-martingale rather than a martingale, so the optional-stopping identity E[M_T]=M_0, and hence (1.13), would fail. No independent human or symbolic verification is provided. This is not a claim that the formula is false; it is a claim that the central theorem's most computation-heavy foundation is currently unverified.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies level lines of the Gaussian free field in an annulus with alternating boundary data and computes the probability that all n level lines cross the annulus. The two central objects are explicit partition functions Z_n-ann and Z^{(ℓ)}_{n-cro}, built from Jacobi theta functions via Dubédat's regularized Dirichlet energy. The paper claims these solve the annulus BPZ equations (Proposition 1.1), that the ratio Z^{(ℓ)}_{n-cro}/Z_n-ann is a bounded martingale, and that its terminal value reduces to the known rainbow probability in the slit polygon, yielding the exact identity P[cross, wind=(m;ℓ)] = Z^{(ℓ)}_{n-cro}/Z_n-ann (Theorem 1.2) and the asymptotics with a √r prefactor (Proposition 1.3). The proof machinery includes a multi-time martingale framework for annulus SLE, a Brownian-loop cascade construction, and an analytic-lift argument for the GFF coupling.","tokens_in":66073,"tokens_out":9673,"duration_ms":98290,"significance":"If the algebraic identities are correct, this is a substantial advance: it gives the first explicit annulus crossing probabilities for multiple GFF level lines and exhibits a nontrivial polynomial correction relative to the disc limit. The paper is honest about the key structural difficulty—the annulus BPZ system is underdetermined—and the main proof architecture is original and credible: the regularized-Dirichlet construction supplies the correct solution, the Liouville argument reduces BPZ verification to periodicity and pole cancellation, and the crossing probability is obtained without fitting parameters. The precise claims are falsifiable and of clear interest to the SLE/GFF community. However, the two computational appendices are the load-bearing algebraic core, and as written they are AI-draft computations without independent human or machine verification; this is a serious gap for a proof journal.","major_comments":[{"comment":"The stated prefactors do not follow from the lemmas cited for their proof. From (2.46), Z_n-ann ~ sqrt(2)^n Z_n(α)Z_n(β) exp(nr/4 + ...); from (2.48), Z^{(ℓ)}_{n-cro} ~ sqrt(2)^{n(n-1)} Z_n-rad(α)Z_n-rad(β) exp(n(2-n)r/8 + ...). Their ratio has prefactor sqrt(2)^{n(n-2)}, not sqrt(2)^{n^2}. Likewise (2.50) gives sqrt(2)^{n(n-2)}/sqrt(2π) rather than sqrt(2)^{n^2}/sqrt(2π) in (1.17). For n=2 the discrepancy is concrete: the exact ratio at (1.13), combined with (2.46) and (2.48), gives prefactor 1, while (1.16) gives 4. This is not cosmetic: Proposition 1.3 is a stated main result and is used in the comparison with the disc analogue. Please correct the prefactors or justify the printed form.","section":"Eq. (1.16)–(1.17), Prop. 1.3, vs Lemmas 2.10–2.12"},{"comment":"The identity G_± = G_+ = -(3/4)(n-1)E(r) in (4.6) is the algebraic core of Proposition 1.1, and the drift computation (3.37) of Proposition 3.8 is the core of the multi-time martingale argument. Appendix B, which proves (4.6) via double-periodicity and pole cancellation, and Appendix A, which produces (3.37), both state that the calculations are 'compiled from ChatGPT 5.5 Pro (OpenAI)’s computation drafts.' No independent human-readable derivation, symbolic algebra file, or machine-checkable certificate is provided. A single wrong sign or coefficient in these cancellations would make the ratio in (5.15) a super/submartingale rather than a martingale, and would invalidate the terminal-value identity (5.20) and Theorem 1.2. This is therefore a load-bearing gap in verification, not merely a presentation issue. I am not claiming the formula is false, but the manuscript should provide a compl","section":"Appendices A and B; Eq. (4.6); Prop. 1.1"},{"comment":"The terminal-time estimates are the mechanism that forces the correction factors R_{1,t}, R_{2,t}, L_{j,t} in (4.15) and (5.16) to converge to 1, so this lemma is central to the proof of Lemma 4.1 and Theorem 1.2. The argument is only sketched: after (4.22)–(4.25) are proved for a fixed z, the uniform bounds (4.26) are asserted to follow from harmonic-measure estimates, but the details of how the constant C=C(γ) is obtained and how the passage from a fixed interior point z to the boundary points α_j and β_j is made are omitted. Given that these estimates control exactly the terms that determine the terminal value (5.20), I would like the proof expanded with all Beurling-estimate constants and boundary limits made explicit.","section":"Lemma 4.2, Eqs. (4.18)–(4.20)"}],"minor_comments":[{"comment":"Several small typographical errors occur, e.g. 'Propositon 3.8' in the Section 3.3 heading; please proofread.","section":"Throughout"},{"comment":"The notation '2∤(k−i)' is used without definition; state explicitly that it means the product is over pairs with odd index difference. Also, the factorization in (2.60) would be easier to follow if the products were grouped with matching factors.","section":"Eq. (2.59)–(2.61)"},{"comment":"If the ChatGPT-generated computations are retained, the manuscript should state in the introduction or in each appendix what independent verification was performed, since the current wording leaves the status of the central algebraic identities unclear.","section":"Appendix A/B disclosures"}],"recommendation":"major_revision","confidential_remarks":"I believe the paper's overall strategy and probabilistic structure are sound and the result is likely correct, but the present state is not acceptable for a proof journal. The most urgent issue is the unverified AI-drafted algebra in Appendices A and B: because Proposition 1.1 and hence the entire martingale apparatus depend on it, the editor should require either full human-verifiable computations or machine-checkable certificates. The prefactor mismatch in Proposition 1.3 is concrete and must be fixed. I would not reject: nothing here suggests an irreparable error, but the load-bearing verification gap is serious enough to require another round before acceptance."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague —\n\nI read the full text. The paper delivers something genuinely new: explicit crossing and winding probabilities for all n GFF level lines in an annulus, as a ratio of two partition functions built from Jacobi theta functions. The annulus BPZ system is underdetermined, so the construction is essential; the regularized Dirichlet energy picks the right solutions, and the sqrt(r) correction in (1.17) versus the disc asymptotics is a real new qualitative effect. This is not a repackaging of [PW19] or [KWW24]. The proof strategy is also serious: construct a bounded martingale from the partition-function ratio, control the terminal value when the first curve crosses, then hand the remaining problem to the PW19 rainbow probability. The analytic-lift argument in Lemma 5.3 is the right tool and cleans up the martingale verification. The paper is honest: (1.21) is labelled 'we believe', Remark 3.15 flags other solution families, and no parameter is fitted to the target event. I see no circularity.\n\nNow the soft spots, in order of size.\n\nFirst, the explicit caveat in the text: Appendices A and B are 'compiled from ChatGPT 5.5 Pro (OpenAI)'s computation drafts.' These appendices are load-bearing. Appendix B supplies the crucial identity (4.6), the constant value of G_epsilon that makes Z_n-ann and Z_n-cro solutions of the annulus BPZ equations. Appendix A supplies the drift formula (3.37) behind Proposition 3.8. A sign or coefficient error anywhere in those elliptic-function cancellations would turn M_t into a super/sub-martingale and break the optional-stopping identity that yields (1.13). I am not saying the formula is wrong; the structure and the degeneration checks are plausible. But an unverified computer-generated algebra block is not enough for a paper whose central theorem hangs on it. A referee should ask for a human-written or symbolic-verified version of those calculations.\n\nSecond, Lemma 4.2's terminal estimates (4.18)-(4.20) are sketched rather than proved. The Beurling-estimate/harmonic-measure strategy is standard and I don't see an obvious error, but the constants depend on the realized curve and the argument is compressed. This is a moderate gap, not a red flag on its own.\n\nThe external inputs are solid: the PW19 rainbow formula and Zhan/Lawler single-annulus SLE are published, with one shared co-author but the annulus answer is not encoded in them. The underdetermined BPZ space means uniqueness is not established, but the paper doesn't claim uniqueness; it claims the constructed functions give the level-line probabilities.\n\nWho benefits: people working on GFF/SLE crossing probabilities, conformal-field-theory martingale observables, and multiply connected SLE. I would want to see this in the literature after revision. Send it to a serious referee, and make the verification request explicit.","headline":"First explicit annulus GFF level-line crossing/winding probabilities, genuinely new and seriously argued; deserves review, but the BPZ verification leans on ChatGPT-compiled appendices that need independent checking before the central theorem is fully trusted.","tokens_in":66772,"tokens_out":3848,"would_cite":false,"duration_ms":40882,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60J67"],"pacs":[],"model":"deepseek-v4-flash","headline":"In an annulus, the probability that all GFF level lines cross with a given winding is exactly a ratio of two theta-function partition functions.","keywords":["Gaussian free field","level lines","annulus","crossing probability","BPZ equations","partition functions","Jacobi theta functions","SLE"],"falsifier":"Simulate the GFF level lines in an annulus for n=2 (two alternating arcs) at several radii and boundary angles, and compare the empirical crossing-winding frequency to Z^(0)_2-cro / Z_2-ann. A discrepancy larger than Monte Carlo error would refute the exact formula; equivalently, measuring the large-radius prefactor of the crossing probability and finding a power different from √r would refute Proposition 1.3.","tokens_in":65629,"feed_emoji":"🌀","tokens_out":4719,"duration_ms":52361,"temperature":0.7,"pith_summary":"The paper establishes an exact formula for the probability that all n level lines of a Gaussian free field in an annulus with alternating boundary values cross from the outer to the inner boundary, including a prescribed winding number. The claim is that this probability equals Z^(ℓ)_n-cro / Z_n-ann, where both partition functions are built from Jacobi theta functions and arise as exponentials of regularized Dirichlet energies. The same functions are shown to satisfy the annulus BPZ equations, a system of PDEs that ensures the associated stochastic processes are martingales. Because the annulus has one more parameter than the equations, the BPZ solution space is infinite-dimensional, so the explicit energy construction is what pins down the physically relevant answer. If correct, the formula also yields a large-radius asymptotic with a nontrivial √r polynomial correction that is absent when the inner boundary is replaced by a point.","feed_headline":"Theta ratio gives exact level-line crossing odds","feed_subtitle":"All level lines cross with probability given by a computable theta-function ratio, winding included.","key_machinery":"The load-bearing objects are the two partition functions Z_n-ann and Z^(ℓ)_n-cro, defined via Jacobi theta functions. The theta functions supply the Green's function and boundary Poisson kernel of the annulus, so exponentiating their regularized Dirichlet energies yields functions with the correct conformal covariance and modulus dependence. These functions satisfy the annulus BPZ equations — the PDE system expressing the conformal Ward identity in the presence of a moving modulus — which makes the ratio of the two a local martingale along the first level line. Proving the BPZ identities uses the periodicity of the logarithmic derivatives of the theta functions on the covering strip to reduc","core_discovery":"The central discovery is a closed-form identity: for an even number n=2N of marked points alternating between boundary values π and 0, the probability of the crossing-winding event is the ratio Z^(ℓ)_n-cro(r; α, β_{2m+1},...,β_{2m+n}) / Z_n-ann(r; α, β). The functions in the ratio are explicit products and exponentials of rescaled Jacobi theta functions; they are the exponentials of regularized Dirichlet energies of harmonic functions with alternating boundary jumps, and they satisfy annulus BPZ equations. The proof identifies the ratio as the terminal value of a bounded martingale, and shows that when the first curve makes the prescribed crossing, this martingale degenerates to the known ra","pith_inferences":["Editorial: the torus-periodicity argument for verifying the BPZ residual term suggests the same strategy could handle partition functions in higher-genus or multi-hole domains, where explicit solutions are even harder to guess.","Editorial: the √r factor likely reflects the fluctuation of the average of the field on the inner boundary; if so, similar polynomial corrections should appear when a small hole is inserted into any bounded domain with fixed boundary values.","Editorial: the inequality Z^(ℓ)_n-cro ≤ Z_n-ann, proved here to bound the martingale, looks like a general principle — 'crossing costs no more than total energy' — and if it holds for other connectivity patterns it would give a systematic way to identify the correct BPZ solution.","Editorial: a discrete-GFF simulation for n=2 with several radii would give a direct numerical check of both the exact formula and the √r asymptotic."],"forward_implications":["The law of multiple GFF level lines in an annulus is encoded by the explicitly computable partition function Z_n-ann, so the same construction can be used to write other crossing and connectivity probabilities.","For fixed winding, the crossing probability is exactly a theta-function ratio, and summing over windings gives a closed-form total crossing probability.","The large-radius asymptotics display a √r polynomial correction that is absent in the disc limit; this changes the subleading order of the probability when the inner hole boundary carries bounded boundary data.","The proof supplies a route to verify annulus BPZ equations even when the solution space is infinite-dimensional, by combining elliptic-function identities with explicit energy computations.","The multi-time martingale construction extends the chordal and radial framework to the annulus, giving a unified way to define multi-curve SLE in doubly connected domains."],"fun_headline_variants":["Annulus level-line crossing odds in theta-function ratio","Exact crossing probability from regularized Dirichlet energy ratio","Closed-form odds: GFF level lines crossing annulus","Annular GFF crossing probability as theta-function ratio"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The terminal-time bounds of Lemma 4.2 — that as the first curve reaches the inner boundary the marked points separate at controlled rates — are the step on which the martingale terminal value rests; if any of these estimates fail, the crossing-winding formula does not follow.","fun_headline_variants_meta":{"raw":{"variants":["Annulus level-line crossing odds in theta-function ratio","Exact crossing probability from regularized Dirichlet energy ratio","Closed-form odds: GFF level lines crossing annulus","Annular GFF crossing probability as theta-function ratio"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000592,"raw_usage":{"total_tokens":2568,"prompt_tokens":659,"completion_tokens":1909,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":403,"completion_tokens_details":{"reasoning_tokens":1844}},"tokens_in":403,"tokens_out":1909,"duration_ms":13874,"temperature":1.0,"reasoning_tokens":1844,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T01:24:39.913736+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Simulate the GFF level lines in an annulus for n=2 (two alternating arcs) at several radii and boundary angles, and compare the empirical crossing-winding frequency to Z^(0)_2-cro / Z_2-ann. A discrepancy larger than Monte Carlo error would refute the exact formula; equivalently, measuring the large-radius prefactor of the crossing probability and finding a power different from √r would refute Proposition 1.3.","supporting_citations":[],"review_version":1}