{"id":"cb25ba8c-5651-4ea4-99cd-76c71b514ee3","arxiv_id":"2509.04333","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"The top level-line of the low-temperature Discrete Gaussian surface above a floor converges, after L^(1/3) rescaling, to a Ferrari-Spohn diffusion, and multiple top level-lines converge to independent copies.","lead":"For a crystal-like integer-height surface above a hard floor, this paper proves that the boundary of its top plateau wiggles on a scale of about L^(1/3) and, after rescaling, follows a known universal random curve law (Ferrari-Spohn). It settles a 2016 conjecture and provides the first such limit theorem for a whole family of related surface models.","discovery_kind":"first_principles","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Lower bound depends on unproved extension of [27] depinning to disagreement polymers; Corollary 6.4 must be proved before the exact Ferrari–Spohn law is established.","rationale":"The reader's weakest_assumption identifies exactly the same load-bearing gap: the extension of the depinning theorem from Ising polymers to the disagreement-polymer model with an area tilt and non-flat boundaries. My reading of Sections 5–6 confirms this. The upper bound is elaborate but self-contained: it constructs a polymer representation, proves cone-point/irreducible-component structure, and reduces the upper stochastic domination to a Ferrari–Spohn comparison. The lower bound, however, begins with Corollary 6.4, which is stated as an immediate consequence of a theorem about Ising polymers. The text says 'The same is true for the class of polymer models considered in this paper' but does not show the modifications. Since the disagreement polymer is not a simple contour and carries integer labels and enclosed regions, this is a genuine nontrivial transfer. The final paragraph of Section 6 even acknowledges that extending depinning to wiggly boundaries is unclear, which shows the authors are aware of the fragility of this point. If Corollary 6.4 is false or requires additional assumptions, the lower bound collapses: one would only have Y_n ⎷ FS_{σ_n}, not Y_n ≍ FS_{σ_n}, so the exact law in Theorem 1.1 and the product law in Theorem 1.5 would not be established. I do not see evidence that the concern is merely a disagreement with prior consensus; it is an unverified step internal to the proof. I therefore agree with the reader's CONDITIONAL verdict and would not change it: the paper is substantial, the reductions are coherent, and the upper bound plus the O-Z machinery provide real support, but the depinning extension must be written out before the central claim is fully verified.","tokens_in":74197,"tokens_out":4048,"duration_ms":42536,"concrete_test":"Independently re-derive Corollary 6.4 for the disagreement-polymer weights qn_{Q;Q,1} from Eq. (6.1) without invoking Ising-specific contour structure: write out [27, Theorem 2] for the general O-Z polymer model satisfying Lemma 3.11 and Proposition 3.19, and identify every step that uses the three-basic-irreducible-components assumption. If the comparison bZn_{H+,H+}(A,B) ≍ bZn_{H+,Z2}(A,B) cannot be closed with a β-uniform constant for labeled disagreement polymers, the lower bound is unproved. As a complementary numerical check, compute this partition-function ratio on finite rectangles with N_n^{1/3} ranging over, say, 2^6 to 2^{14} and β large; if the ratio is not 1+o(1) uniformly in the area tilt, the asserted depinning extension fails.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The exact limiting law in Theorem 1.1 requires both the upper bound of Section 5 and the lower bound of Section 6. The lower bound hinges on Corollary 6.4, which asserts that for the disagreement-polymer model the flat-boundary depinning comparison bZn_{H+,H+}(A,B) = (1+o(1)) bZn_{H+,Z2}(A,B) holds. This is presented as an immediate consequence of [27, Theorem 2], but the move from Ising polymers to disagreement polymers is asserted, not proved. The paper's own text concedes the issue: in Section 6, after 'Forgetting the area term for now', it says the [27] proof 'is more robust' and 'allows for more complicated geometries', but no derivation is given. The final paragraph of Section 6 similarly notes that extending depinning to a wiggly boundary 'is unclear'. This matters because [27] uses Ising-specific structure, e.g. simple-contour geometry and the fact that the increment measure has 1−εβ mass on three basic irreducible components; disagreement polymers are connected bond sets with integer labels and additional enclosed regions, and the current paper does not supply the analogue. Corollary 6.4 feeds directly into the Brownian-excursion estimate and into the stochastic-domination-from-below step Pn_{Q,Q}(·|G⊓) ⎷ FSσn. If the depinning ratio is not uniformly O(1), the lower-bound reduction at Eq. (6.1) fails and the exact FS limit is not obtained. This is a missing proof, not an internal contradiction, but it is the most load-bearing gap.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the low-temperature (2+1)-dimensional Discrete Gaussian / integer-valued Gaussian free field in an L x L box above a hard floor, for side lengths outside an explicit exceptional set B. With H(L) defined from single-site probabilities and N_n as the reciprocal single-site probability of height H-n, it claims that the top m level-lines, rescaled horizontally by N_n^{2/3} and vertically by N_n^{1/3}, converge weakly to a product of independent stationary Ferrari--Spohn diffusions. The proof is a long reduction: a cluster expansion for disagreement polymers (Section 2), refined large-deviation estimates (Theorem 2.5), an Ornstein--Zernike analysis including existence and convexity of the surface tension (Section 3), Wulff-shape growth (Section 4), and upper and lower stochastic-domination bounds (Sections 5 and 6) that are matched against the known Ferrari--Spohn convergence for area-tilted random walks. The paper further extends the statement to all |\\nabla\\phi|^p models with p>1.","tokens_in":74624,"tokens_out":9816,"duration_ms":100172,"significance":"If the claims hold, this is a major result: it confirms the conjecture of Lubetzky--Martinelli--Sly on the L^{1/3+o(1)} level-line fluctuations and, more strongly, identifies the exact Ferrari--Spohn scaling limit for the top level-line of the ZGFF above a floor. It also gives the first exact level-line limit for a (2+1)-dimensional |\\nabla\\phi|^p model away from the SOS case, and proves asymptotic independence of finitely many level-lines. The definitions of N_n and sigma_n are model-derived rather than fitted: N_n is a reciprocal single-site probability and sigma_n is the variance of an effective Ornstein--Zernike random-walk increment. The paper is very ambitious and contains a substantial amount of novel technical machinery. The upper-bound chain and the Ornstein--Zernike/cluster-expansion parts are developed in detail. The main weakness is that the lower bound depends on an extension of the depinning theorem of [27] to disagreement polymers in a wiggly domain, and that extension is asserted rather than proved.","major_comments":[{"comment":"The lower bound in Theorem 6.1 reduces the problem to the comparison bZn_{Q,Q}(A,B | G⊓) = (1+o(1)) bZn_{H+,H+}(A,B), stated as Corollary 6.4. This is load-bearing: it is used to transfer the no-area depinning/repulsion estimates and then, via the area-tilt argument around Eq. (5.8), to obtain the stochastic-domination-from-below by FS_{sigma_n}. The proof of Corollary 6.4, however, rests on an asserted extension of [27, Theorem 2] from Ising polymers in a half-plane with a flat boundary to disagreement polymers in the domain Q, whose top and side boundaries are only wiggly approximations of a rectangle. The text says the proof of [27] 'is more robust and allows for more complicated geometries', but no derivation is given. The disagreement polymers here are connected sets of dual bonds with integer labels and enclosed regions D_i, and the increment measure of the associated effective ran","section":"Section 6, Corollary 6.4; also Section 5, Proposition 5.9"},{"comment":"The proof of Corollary 6.4 also uses a reduction from G⊓ to the unconditioned measure via [26, Theorem 5.3] and Proposition 5.9. The sentence around Eq. (6.3) says that if the domain restriction only forced cone-points to be nonnegative, then convergence to a Brownian excursion would make the event G⊓ have probability o(1). But the object whose cone-points are controlled is the disagreement polymer, and the equivalence between the cone-point process of the disagreement polymer and the effective 2D random walk has only been sketched through the Ornstein--Zernike results of Section 3. If the [27]-type depinning extension is not available, this step is also unsupported. The lower-bound proof thus has two linked missing pieces: the flat-boundary depinning transfer and the wiggly-domain comparison. Both need to be supplied before Theorem 1.1's lower bound can be regarded as proved.","section":"Section 6, proof of Corollary 6.4 and Theorem 6.1"}],"minor_comments":[{"comment":"The sentence 'yet they are o(L^{-1/3}) for infinitely many other values of L' appears to be a typo: the scale is N^{1/3}=L^{1/3-o(1)}, so it should read o(L^{1/3}), not o(L^{-1/3}).","section":"Remark 1.2"},{"comment":"Theorem 1.1 says 'for a fixed sigma > 0', but the matching statements in Theorems 5.1 and 6.1 use the model-dependent sigma_n from Definition 3.20. The introduction should state explicitly that the limiting diffusion is FS_{sigma_n} and that sigma_n is the constant defined in Section 3, not an arbitrary fixed parameter.","section":"Theorem 1.1 and Definition 3.20"},{"comment":"Equation (1.2) defines H(L) using bπ∞(φ_o = h), while the statement of Theorem 4.4 and the text around it use bπ∞(φ_o ≥ h) and N_n = 1/bπ∞(φ_o ≥ H-n). These are asymptotically close given Eq. (2.3), but the notation should be made consistent.","section":"Theorem 4.4 and Eq. (1.2)"},{"comment":"The caption writes 'N_n^{(p)} ≍ 1/bπ^{(p)}∞(φ_o = L)'; the argument of bπ should be a height h, not the box side L. This looks like a typographical error.","section":"Figure 3 caption"},{"comment":"Lemmas 4.7 and 6.2 are each postponed to 'Appendix B'. If the appendices are not included in the submitted version, the proofs of those two geometric reduction lemmas are missing; they should be part of the manuscript.","section":"Appendix references"}],"recommendation":"major_revision","confidential_remarks":"The main gap is the depinning-transfer issue identified above. It is not a circularity or a disagreement with consensus; it is a technical assertion that the authors themselves flag as delicate. If the authors can prove Corollary 6.4 (or otherwise replace the lower-bound reduction), the paper is likely to be a strong addition to the probability/statistical-mechanics literature. I do not see a reason to doubt the upper-bound machinery, and the target law is genuinely external to the model. The present version, however, is incomplete as a proof of Theorem 1.1."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"This paper resolves the LMS16 conjecture: it proves the L^{1/3+o(1)} fluctuation scale and, more strongly, the Ferrari-Spohn limit law for the top level-line of the (2+1)D ZGFF above a floor, and extends the result to the full |∇φ|^p family for p>1. That is a genuine breakthrough, the first FS limit law for any of these level-line models.\n\nThe paper does a lot of real work and ships most of it. The disagreement-polymer machinery is a genuine extension of Ising polymer theory, not a routine transfer: Prop 2.3 (cluster expansion), Thm 2.5 (refined large deviations), and Prop 3.12/3.14 (surface tension via the new key bound) are all proved in the text. The stochastic-domination scheme, upper bound by revealing exterior-to-interior and lower bound by interior-to-exterior, is clever and, with one exception, handles pinning convincingly. The scales are defined by the model, not fit to the target law: N_n is the reciprocal single-site probability, sigma_n is the effective random walk variance, and the FS law is an external limit from [25]. No circularity.\n\nThe soft spot is exactly where the reader and the stress-test note point: the lower bound in Section 6 depends on Corollary 6.4, an extension of the depinning comparison [27, Thm 2] from Ising polymers to disagreement polymers with a wiggly boundary. This extension is asserted, not proved. The paper's own text flags it twice: after \"Forgetting the area term for now\" it says the [27] proof \"is more robust,\" and the last paragraph of Section 6 says extending to a wiggly boundary \"is unclear.\" Those concessions are honest, but they do not make the proof. The move is not cosmetic: disagreement polymers are connected bond sets with integer labels and enclosed regions, not simple contours, and [27] uses Ising-specific structure (e.g., 1-epsilon_beta mass on three basic irreducible components). Corollary 6.4 feeds directly into the Brownian-excursion estimate and the domination-from-below step that produces the FS lower bound. If the depinning ratio is not uniformly O(1), Eq. (6.1) fails and the exact law is not established. This is a missing proof, not an internal contradiction, but it is load-bearing.\n\nThe upper bound also has delicate steps, but those are supported by detailed arguments and are less fragile.\n\nWho this is for: people working on random interfaces, level-line fluctuations, and the SOS/ZGFF family. It deserves a serious referee and likely a conditional accept after the Section 6 gap is closed. I would not desk-reject it; I would send it to an expert with a specific request to check Corollary 6.4 and the depinning extension. The result is important enough that the referee time is warranted even if the final version needs substantial revision.","headline":"A major paper that plausibly resolves the LMS16 conjecture and proves the Ferrari-Spohn limit law for ZGFF level-lines, but the lower bound leans on an explicitly flagged, unproved extension of a depinning theorem.","tokens_in":75091,"tokens_out":2580,"would_cite":true,"duration_ms":25632,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60K35","82B20","82B41","60J65"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper proves that the top level-lines of the (2+1)d Discrete Gaussian model above a hard floor have N^{1/3} fluctuations, with rescale limits given by independent Ferrari-Spohn diffusions.","keywords":["integer-valued Gaussian free field","entropic repulsion","level-line fluctuations","Ferrari-Spohn diffusion","disagreement polymer","cluster expansion","random surface","absolutely continuous gradient models"],"falsifier":"Simulate the ZGFF at large beta for non-exceptional side-lengths L, extract the top level-line, and compare the empirical distribution of N^{-1/3} rho(t N^{2/3}) at the center of the side interval with the stationary FS_sigma law. If the fluctuation exponent deviates from 1/3, or the centered marginal fails to match the squared-Airy stationary density, the central claim is wrong; a second check is whether the level-line separation N_{n+1}/N_n decays at the predicted rate exp(-Theta(sqrt(beta log L / log log L))).","tokens_in":74106,"feed_emoji":"📐","tokens_out":5284,"duration_ms":53505,"temperature":0.7,"pith_summary":"The paper aims to settle the conjecture from LMS16 that the boundary of the top plateau in the integer-valued Gaussian free field (ZGFF) on an L x L box above a hard floor fluctuates on scale L^{1/3+o(1)}. It proves a sharper statement: for most side-lengths L, after choosing the natural mesoscopic scale N = L^{1-o(1)}, the vertical distance of the top level-line from an interval on the side boundary, rescaled by N^{1/3} vertically and N^{2/3} horizontally, converges to the stationary Ferrari-Spohn diffusion. The same holds jointly for any finite number of top level-lines, and the limit is a product of independent Ferrari-Spohn diffusions. This is the first confirmation of a Ferrari-Spohn limit among the (2+1)d |grad phi|^p random surface models, and the result extends to every fixed p>1.","feed_headline":"Top level-lines get a proven L^{1/3} Ferrari-Spohn law","feed_subtitle":"The boundary of a random crystal plateau shows cube-root fluctuations with an Airy-function limit law.","key_machinery":"The argument is carried by a polymer representation of level-lines: a level-line is surrounded by a labeled disagreement polymer, a maximal connected component of dual bonds where neighboring heights differ, whose law is written via cluster expansion as an area-tilted polymer. Ornstein-Zernike theory, cone-points, and irreducible components turn this polymer into a two-dimensional random walk on cone-points with an area tilt, and the Ferrari-Spohn diffusion appears as the rescaling limit of that area-tilted random walk.","core_discovery":"Theorem 1.1: fix beta large and take the ZGFF on an L x L box above a floor with zero boundary conditions, for side-lengths L outside an explicit exceptional set of zero logarithmic density. Let H(L) be the height where the single-site probability under the no-floor measure first drops below 5 beta / L, and let N = 1 / bpi_infty(phi_o = H), which is L^{1-o(1)}. The distance rho(x) from an interval of length N^{2/3} centered on the bottom side up to the top macroscopic level-line, rescaled as Y_0(t) = N^{-1/3} rho(t N^{2/3}), converges weakly to the stationary Ferrari-Spohn diffusion FS_sigma on [-1,1]. The same statement holds jointly for the top m level-lines, with independent FS diffusions","pith_inferences":["The scale separation proved here suggests that the p=1 (SOS) case fails to have a product of FS limits not by a technical gap but because its level-lines all sit at the same scale; the conjectured correlated line ensemble is the natural contrast case.","Near the corners, the same framework points to fluctuations of order L^{1/2}, as the paper notes; a direct simulation of level-line endpoints near corners could test whether the exponent is indeed 1/2 rather than 1/3.","A concrete numerical check of the central claim: simulate the ZGFF at large beta for non-exceptional L, record the top level-line distance at the center of the side interval, and compare the empirical law of N^{-1/3} rho(0) with the FS_sigma stationary density proportional to the squared Airy first eigenfunction.","If the asserted extension of flat-boundary depinning to wiggly boundaries fails, the lower bound in Section 6 would need a new mechanism; the stochastic-domination asymmetry between upper and lower bounds makes that extension the point to scrutinize."],"forward_implications":["Confirms the LMS16 conjecture that the top level-line of the ZGFF fluctuates on scale L^{1/3+o(1)}.","Recovers the exact limiting law: the rescaled top level-line is the stationary Ferrari-Spohn diffusion, with Airy-function marginals.","Establishes that any finite number of top level-lines are asymptotically independent Ferrari-Spohn diffusions, due to a strong separation of scales between lines.","Extends the same limit law to the full |grad phi|^p family for every fixed p>1, marking the p=1 (solid-on-solid) case as the exceptional one.","Shows that the fluctuation scale is exactly L^{1/3} for infinitely many side-lengths and o(L^{1/3}) for infinitely many others, controlled by the bounds on N."],"supporting_citations":[{"why":"Established the plateau concentration and large-deviation estimates for the ZGFF, and conjectured the L^{1/3+o(1)} level-line fluctuations that this paper confirms.","marker":"[35]"},{"why":"Supplies the flat-boundary depinning result whose extension to wiggly boundaries is the load-bearing step in the lower bound.","marker":"[27]"},{"why":"Provides the recent proof that area-tilted 2d random walks converge to the Ferrari-Spohn diffusion, used as the final convergence input.","marker":"[25]"},{"why":"Gives the Ornstein-Zernike and surface-tension framework for polymer models that Section 3 adapts to disagreement polymers.","marker":"[17]"},{"why":"Established the growth-gadget and dropping-lemma strategy for SOS level-lines that the upper-bound proof follows and refines.","marker":"[14]"},{"why":"Identified the entropic repulsion effect that creates the random surface above the floor, the starting physical mechanism for the whole problem.","marker":"[5]"},{"why":"Defined the stationary Ferrari-Spohn diffusion and its Airy-function representation used as the claimed limit object.","marker":"[18]"},{"why":"Established the L^{1/3} fluctuations and Ferrari-Spohn scaling limit for area-tilted random walks, the discrete ancestor of the diffusion.","marker":"[28]"},{"why":"Showed a Ferrari-Spohn limit for SOS level-lines with deterministic boundary conditions, providing the template adapted here.","marker":"[7]"}],"fun_headline_variants":["Cube-root fluctuations proven for discrete Gaussian level lines","Ferrari-Spohn law confirmed for top level-line of discrete Gaussian","Discrete Gaussian top level-line has cube-root law: proof","Level-line fluctuation law: L^{1/3} diffusion confirmed","Joint law for multiple level-lines of discrete Gaussian"],"cache_read_input_tokens":2688,"weakest_assumption_plain":"The proof assumes that the depinning theorem for flat-boundary polymers extends to the disagreement-polymer model with a wiggly or random boundary; this extension is stated rather than proved, and the lower-bound half of the main theorem relies on it.","fun_headline_variants_meta":{"raw":{"variants":["Cube-root fluctuations proven for discrete Gaussian level lines","Ferrari-Spohn law confirmed for top level-line of discrete Gaussian","Discrete Gaussian top level-line has cube-root law: proof","Level-line fluctuation law: L^{1/3} diffusion confirmed","Joint law for multiple level-lines of discrete Gaussian"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000802,"raw_usage":{"total_tokens":3493,"prompt_tokens":1010,"completion_tokens":2483,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":754,"completion_tokens_details":{"reasoning_tokens":2408}},"tokens_in":754,"tokens_out":2483,"duration_ms":16595,"temperature":1.0,"reasoning_tokens":2408,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-05T10:14:00.893075+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Simulate the ZGFF at large beta for non-exceptional side-lengths L, extract the top level-line, and compare the empirical distribution of N^{-1/3} rho(t N^{2/3}) at the center of the side interval with the stationary FS_sigma law. If the fluctuation exponent deviates from 1/3, or the centered marginal fails to match the squared-Airy stationary density, the central claim is wrong; a second check is whether the level-line separation N_{n+1}/N_n decays at the predicted rate exp(-Theta(sqrt(beta log L / log log L))).","supporting_citations":[{"cited_title":"Lubetzky, F","cited_arxiv_id":null,"evidence_quote":"Established the plateau concentration and large-deviation estimates for the ZGFF, and conjectured the L^{1/3+o(1)} level-line fluctuations that this paper confirms."},{"cited_title":"Ioffe, S","cited_arxiv_id":null,"evidence_quote":"Supplies the flat-boundary depinning result whose extension to wiggly boundaries is the load-bearing step in the lower bound."},{"cited_title":"Ioffe, S","cited_arxiv_id":null,"evidence_quote":"Provides the recent proof that area-tilted 2d random walks converge to the Ferrari-Spohn diffusion, used as the final convergence input."},{"cited_title":"Dobrushin, R","cited_arxiv_id":null,"evidence_quote":"Gives the Ornstein-Zernike and surface-tension framework for polymer models that Section 3 adapts to disagreement polymers."},{"cited_title":"Caputo, E","cited_arxiv_id":null,"evidence_quote":"Established the growth-gadget and dropping-lemma strategy for SOS level-lines that the upper-bound proof follows and refines."},{"cited_title":"Bricmont, A","cited_arxiv_id":null,"evidence_quote":"Identified the entropic repulsion effect that creates the random surface above the floor, the starting physical mechanism for the whole problem."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Defined the stationary Ferrari-Spohn diffusion and its Airy-function representation used as the claimed limit object."},{"cited_title":"Ioffe, S","cited_arxiv_id":null,"evidence_quote":"Established the L^{1/3} fluctuations and Ferrari-Spohn scaling limit for area-tilted random walks, the discrete ancestor of the diffusion."},{"cited_title":"Caddeo, Y","cited_arxiv_id":null,"evidence_quote":"Showed a Ferrari-Spohn limit for SOS level-lines with deterministic boundary conditions, providing the template adapted here."}],"review_version":1}