REVIEW 2 major objections 4 minor 1 cited by
Quantitative delocalization for solid-on-solid models at high temperature and arbitrary tilt
T0 review · 2 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash
Pith's one-line read This paper proves that every 2D p-SOS random interface, including the solid-on-solid model, is logarithmically delocalized at high temperature uniformly in boundary data and fiber shifts.
desk verdict A serious, likely-correct paper that closes the SOS gap and proves quantitative delocalization for p-SOS at arbitrary tilt; the one place to push hard is the unproved modification of the imported multipole expansion. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The carrying mechanism is a multiscale multipole expansion: the periodic density λ(φ−ζ) is written as a convex combination over ensembles of disjoint, well-separated charge densities, each with controlled activity coefficients. A contour-shift construction, built from spin waves, renormalizes those activities down to exponentially small values, after which the partition-function ratio is Taylor-expanded. The paper's new technical step is expanding not only in the shift σ·ρ but also in the observable-error terms f·a_ρ, so that no multiplicative constant is lost; a reflection symmetry φ ↦ −φ_rfl handles arbitrary boundary data, and expanding in εf with Cauchy–Schwarz turns the exponential-moment bound into the variance bound. The central named objects are the charge densities, their square coverings, the envelopes D_Λ(ρ), and the spin waves a_ρ; the work they do is to reduce the partition function to a positive convex combination of renormalized partition functions and then to extract the Gaussian Dirichlet-Laplacian covariance.
What would settle it
A direct check would be an exact evaluation of the partition sum for a small box, say 4×4 or 6×6, at p=1 and β=$10^{-3}$ with strongly tilted boundary data ξ_i = u·i and zero fiber shift, testing inequality (1.1) for several test functions f; any violation with β_eff = β^c for a universal c would disprove the theorem. A weaker, still concrete check is to verify that the analytic extension of exp(−β|n|^p) constructed in Lemma 5.1 satisfies Assumption 2 with c_β = $Cβ^{{1/3}}$ in the strip of width $β^{{1/3}}$/2, since that parameter choice is what forces β0>0.
Extended reading notes
Core claim
On the paper's own terms, the central discovery is Theorem 1.1: for d=2 there is β0>0 such that for every 0<p≤2, every β<β0, every finite Λ⊂$Z^{2}$, every boundary datum ξ, every fiber shift ζ, and every test function f, the covariance in the p-SOS measure satisfies $µ^{{p-SOS;ξ,ζ}}$_{Λ;β}(f·φ; f·φ) ≥ (1/β_eff) f·$Δ^{{-1}}$_Λ f, with β_eff ≤ β^c for a universal c>0. The bound is uniform in ξ and ζ, so neither tilting the boundary nor offsetting the integer lattice along fibers can localize the interface at high temperature. In particular, on boxes Λ_N={-N,...,N}^2, the single-site variance obeys Var(φ0) ≥ c log N. The theorem is deduced from more general statements for continuum measures with periodic trigonometric-polynomial weights; those statements hold for a broad class of height functions whose single-site weight extends analytically to a strip with controlled derivatives. The proof also gives an exponential-moment, Laplace-transform bound under reflection-symmetry assumptions, from which the variance bound follows by replacing f with εf and expanding.
Load-bearing premise
The load-bearing premise is the imported multipole decomposition (Theorem 3.2): the periodic density can be split into disjoint, well-separated charge blocs whose coefficients satisfy the three stated bounds; if that split fails anywhere, the renormalization that makes all activities small cannot be controlled.
Editorial extensions
If this is right
- For every p ∈ (0,2], every small β, and every boundary tilt and fiber shift, the variance of the field at the centre of an N×N box grows at least as c log N, so the interface is genuinely rough in finite volume.
- The variance bound is uniform in the boundary data and fiber shifts, so tilting the boundary cannot localize the high-temperature p-SOS interface.
- Under reflection symmetry, the exponential-moment bound of Theorem 2.4 implies the variance bound by expansion in εf, and it provides the finite-volume Laplace-transform control that is the natural input for further Gibbsian analysis.
- The result applies to the ordinary SOS model (p=1), the integer-valued Gaussian free field (p=2), and the height function dual to the XY model, so logarithmic delocalization holds for all of these interfaces at high temperature.
- Because β_eff ≤ β^c for a universal c, the lower bound is polynomial in β, so the variance diverges at least as (1/β^c) log N as β decreases.
Reading between the lines
- The no-constant-loss Taylor expansion is a technique that should transfer to other height functions whose single-site weights satisfy the same analytic strip bounds; plausible targets are long-range or random-conductance interface models, which the paper does not treat.
- Combined with low-temperature localization results, the uniform finite-volume bound may be enough to locate the roughening transition for tilted p-SOS interfaces, but the paper itself does not attempt such a phase-transition statement.
- Under the reflection symmetry of Theorem 2.4, the exponential-moment bound is the type of input used to prove convergence of fluctuations to the continuum Gaussian free field; the paper stops short of a scaling-limit claim.
- For 0<p<1, the paper's regularization by Gaussian tails suggests that moment bounds of any order hold for the regularized model and pass to the limit, which would give stronger control than the variance alone.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper studies two-dimensional integer-valued p-SOS interface models, 0<p≤2, at high temperature and with arbitrary tilt, boundary data, and fiber shifts. The main result (Theorem 1.1) gives a quantitative covariance lower bound of the form μ(f·φ; f·φ) ≥ (1/β_eff) f·Δ_Λ^{-1}f with β_eff ≤ β^c for a universal c>0, uniformly in boundary data and fiber shifts; in particular the variance at the origin grows at least like c ln N on N×N boxes. The proof extends the Fröhlich-Spencer multiscale argument using a multipole expansion, complex translations, and activity renormalization, with two new ingredients: a Taylor expansion that includes the terms f·a_ρ (avoiding the loss of a constant factor in [FS81]) and a reflection symmetry argument for boundary data and domains satisfying antisymmetry. The paper also states general theorems (Theorems 2.3 and 2.4) for a class of height functions with analytic extensions, and applies them to p-SOS models, the integer-valued GFF, and the height function dual to the XY model.
Significance. If the proof is correct, this is a substantial advance: it supplies the first quantitative, uniform-in-boundary-data logarithmic delocalization for SOS-type models at nonzero tilt, fills a gap in the classical Fröhlich-Spencer treatment, and covers the full range 0<p≤2 in a unified way. The paper is careful and largely explicit, with a detailed appendix for the analytic extension and a candid discussion of where the authors improve on [FS81]. A particular strength is that the main theorems are stated in a general abstract framework, making the method potentially reusable. The main residual risk is the imported multipole expansion (Theorem 3.2), which is the combinatorial backbone of the activity renormalization and is not proved in the manuscript.
major comments (2)
- [§3.2 (Theorem 3.2), used in §3.4 (3.11) and §4] The multipole expansion is the combinatorial backbone of the proof, and the only justification for Theorem 3.2 is the sentence in its proof: “Properties (i), (ii) and (iii) are essentially [Wir19, Theorem 21]. The only difference is that in the first step of the proof we do not use the weights e^{q^2}, but e^q (which might lead to an additional factor of 2).” This is not a proof: replacing e^{q^2} by e^q changes the summation estimates that produce the coefficient bound (iii), and the “additional factor of 2” is not traced through the separation and chargedness arguments. Property (iii) is then used directly in (3.11) to obtain |z_ρ|≤1/8 and the exponential decay e^{-c4 γ_β(∥ρ∥_1+A_Λ(ρ))} on which the positivity of the renormalized measure and the error bounds in Claims 1–2 depend. If the e^q version fails to deliver (iii) with the stated constants, the central argument collapses. The authors should either give a self-contained proof of Theorem 3.2 with e^q, or provide a detailed verification that the proof in [Wir19, KP17] survives the substitution and yields exactly the bound stated in (iii).
- [§5.3] The reduction for 0<p<1 via the Gaussian regularization I_β^ε(x)=e^{-εx²}I_β(x) is incomplete. Theorem 2.3 is applied to the regularized model, giving a lower bound with constants c1, c2, γβ, and hence β_eff(ε), that a priori depend on ε. The paragraph then says “Letting ε→0 and using dominated convergence then gives Theorem 1.1 also for 0<p<1,” but no argument shows that the constants in (2.5) (cβ, c′_β, εβ) for I_β^ε can be chosen uniformly for small ε, nor that liminf_{ε→0} β_eff(ε) is positive. Without such uniformity the limit does not yield the asserted bound for the original model. The authors should specify a dependence ε=ε(β) and verify the assumptions of Section 2.2 uniformly over that range.
minor comments (4)
- [§1.3] In the statement of Theorem 1.1, “for any for any” should be “for any.”
- [§2.3, Theorem 2.4] The opening phrase “In the setting of Theorem 2.4 assume additionally” should refer to Theorem 2.3, not Theorem 2.4.
- [§3.4, Lemma 3.5] The statement begins “suppose we are given some γβ ∈ (0, ϵβ.” and is missing a closing parenthesis; it should read γβ ∈ (0, εβ).
- [§5.1, Lemma 5.1] In the bullet for the XY height function, the notation I_β(n) conflicts with the standard modified Bessel function I_n(1/β); using a different symbol such as W_β(n) would avoid ambiguity.
Circularity Check
No circularity: the variance lower bound is derived from an external multipole expansion and explicit analytic estimates; the only self-citation is contextual and not load-bearing.
full rationale
The paper's central claim, Theorem 1.1, is a quantitative delocalization bound for p-SOS models. The proof chain is self-contained in the relevant sense: the model is defined independently (Section 1.2), the analytic extension I_beta is constructed in Lemma 5.1 and Appendix C, the trigonometric-polynomial approximation is justified in Sections 5.2-5.3, and the variance lower bound is proved in Theorems 2.3-2.4 via Sections 3-4. The constants in the bound are universal and are chosen from the hypotheses (2.5); no parameter is fitted to the quantity being predicted, and (1.1) is not assumed. The only self-citation, [LO24], appears in the introduction as contextual background ('Lammers and the first author [LO24] have shown qualitative delocalization') and is explicitly said not to apply in finite volume; it is not used as a black box in the proof. The main external imports, [FS81], [Wir19], and [KP17], are standard independent references authored by others. The most delicate such import, Theorem 3.2, is the multipole expansion from [Wir19] and [KP17]; the paper modifies weights e^{q^2} to e^q, and this modification is not proved in the paper. That is a genuine correctness or completeness risk, but it is not circularity: the missing theorem is external to the present authors' work, is not the target result, and the paper's own argument does not define the multipole expansion in terms of delocalization. No equation in the paper reduces by construction to an earlier fitted or assumed quantity, and no known empirical pattern is merely renamed as a theorem. Therefore the circularity score is 0.
Assumptions & free parameters
assumptions (4)
- domain assumption The model weights Iβ admit an analytic extension to a complex strip satisfying (2.2)-(2.4) with cβ=c'_β=Cβ^{1/3} and ϵβ=β^{1/3}/2, uniformly for 0<p≤2 (Lemma 5.1, Appendix C).
- domain assumption The multipole expansion of trigonometric polynomials (Theorem 3.2), imported from [Wir19, KP17], produces a convex combination of charge ensembles with separation and coefficient bounds (i)-(iii).
- standard math Standard complex analysis: contour shifts of integrals (Lemma B.1) are valid under the stated analyticity and decay conditions.
- standard math The integer-valued measure is the limit of continuum measures with trigonometric polynomial densities (Lemma 5.2, tempered distribution convergence of combs).
Cite this review
Pith. "Pith review of Quantitative delocalization for solid-on-solid models at high temperature and arbitrary tilt." pith.science (2026). https://pith.science/paper/Y7DXOZMW
@misc{pith2026250516804,
author = {Pith},
title = {Pith review of: Quantitative delocalization for solid-on-solid models at high temperature and arbitrary tilt},
year = {2026},
howpublished = {\url{https://pith.science/paper/Y7DXOZMW}},
note = {Machine review of arXiv:2505.16804}
}
abstract
We study a family of integer-valued random interface models on the two-dimensional square lattice that include the solid-on-solid model and more generally $p$-SOS models for $0<p\le2$, and prove that at sufficiently high temperature the interface is delocalized logarithmically uniformly in the boundary data. Fr\"ohlich and Spencer had studied the analogous problem with free boundary data, and our proof is based on their multi-scale argument, with various technical improvements.
Forward citations
Cited by 1 Pith paper
-
The limiting law of the Discrete Gaussian level lines
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.
Reference graph
Works this paper leans on
-
[1]
Depinning in integer-restricted Gaussian Fields and BKT phases of two-component spin models , 2021
Michael Aizenman, Matan Harel, Ron Peled, and Jacob Shapiro. Depinning in integer-restricted Gaussian Fields and BKT phases of two-component spin models , 2021. arXiv:2110.09498
arXiv 2021
-
[2]
The Discrete Gaussian model, I
Roland Bauerschmidt, Jiwoon Park, and Pierre-Fran c ois Rodriguez. The Discrete Gaussian model, I. Renormalisation group flow at high temperature , 2022. arXiv:2202.02286
arXiv 2022
-
[3]
The Discrete Gaussian model, II
Roland Bauerschmidt, Jiwoon Park, and Pierre-Fran c ois Rodriguez. The Discrete Gaussian model, II. Infinite-volume scaling limit at high temperature , 2022. arXiv:2202.02287
arXiv 2022
-
[4]
Decay of correlations in surface models
R Brandenberger and CE Wayne. Decay of correlations in surface models. Journal of Statistical Physics , 27(3):425--440, 1982
work page 1982
-
[5]
Quantitative delocalisation for the Gaussian and $q$-SOS long-range chains
Loren Coquille, Paul Dario, and Arnaud Le Ny. Quantitative delocalisation for the G aussian and q - SOS long-range chains, 2024. arXiv:2412.15782
work page Pith review arXiv 2024
-
[6]
Delocalization of uniform graph homomorphisms from Z^2 to Z
Nishant Chandgotia, Ron Peled, Scott Sheffield, and Martin Tassy. Delocalization of uniform graph homomorphisms from Z^2 to Z . Comm. Math. Phys. , 387(2):621--647, 2021
work page 2021
-
[7]
Karrila, Ioan Manolescu, and Mendes Oulamara
Hugo Duminil-Copin, Alex M. Karrila, Ioan Manolescu, and Mendes Oulamara. Delocalization of the height function of the six-vertex model. J. Eur. Math. Soc. (JEMS) , 26(11):4131--4190, 2024
work page 2024
-
[8]
R. L. Dobru s in. The G ibbs state that describes the coexistence of phases for a three-dimensional I sing model. Teor. Verojatnost. i Primenen. , 17:619--639, 1972
work page 1972
Show all 28 references
-
[9]
Nonexistence of one-and twodimensional G ibbs fields with noncompact group of continuous symmetries
Roland L Dobrushin and Senya B Shlosman. Nonexistence of one-and twodimensional G ibbs fields with noncompact group of continuous symmetries. Multicomponent random systems , 6:199--210, 1980
1980
-
[10]
Gerald B. Folland. Real analysis. Modern techniques and their applications . Pure and Applied Mathematics (New York). John Wiley & Sons, Inc., New York, second edition, 1999
1999
-
[11]
u rg Fr \
J \"u rg Fr \"o hlich and Charles Pfister. On the absence of spontaneous symmetry breaking and of crystalline ordering in two-dimensional systems. Communications in Mathematical Physics , 81:277--298, 1981
1981
-
[12]
u rg Fr\
J\" u rg Fr\" o hlich and Thomas Spencer. The K osterlitz- T houless transition in two-dimensional abelian spin systems and the C oulomb gas. Comm. Math. Phys. , 81(4):527--602, 1981
1981
-
[13]
Invisibility of the integers for the discrete Gaussian chain via a Caffarelli-Silvestre extension of the discrete fractional Laplacian , 2023
Christophe Garban. Invisibility of the integers for the discrete Gaussian chain via a Caffarelli-Silvestre extension of the discrete fractional Laplacian , 2023. arXiv:2312.04536
2023 arXiv
-
[14]
Delocalisation and C ontinuity in 2 D : L oop O (2) , S ix- V ertex, and R andom- C luster M odels
Alexander Glazman and Piet Lammers. Delocalisation and C ontinuity in 2 D : L oop O (2) , S ix- V ertex, and R andom- C luster M odels. Comm. Math. Phys. , 406(5):Paper No. 108, 2025
2025
-
[15]
Uniform L ipschitz functions on the triangular lattice have logarithmic variations
Alexander Glazman and Ioan Manolescu. Uniform L ipschitz functions on the triangular lattice have logarithmic variations. Comm. Math. Phys. , 381(3):1153--1221, 2021
2021
-
[16]
Statistical reconstruction of the GFF and KT transition
Christophe Garban and Avelio Sep\' u lveda. Statistical reconstruction of the GFF and KT transition. J. Eur. Math. Soc. (JEMS) , 26(2):639--694, 2024
2024
-
[17]
2D models of statistical physics with continuous symmetry: the case of singular interactions
Dima Ioffe, Senya Shlosman, and Yvan Velenik. 2D models of statistical physics with continuous symmetry: the case of singular interactions. Communications in mathematical physics , 226:433--454, 2002
2002
-
[18]
The Fr \"o hlich-Spencer Proof of the Berezinskii-Kosterlitz-Thouless Transition , 2017
Vital Kharash and Ron Peled. The Fr \"o hlich-Spencer Proof of the Berezinskii-Kosterlitz-Thouless Transition , 2017. arXiv:1711.04720
2017 arXiv
-
[19]
A dichotomy theory for height functions, 2022
Piet Lammers. A dichotomy theory for height functions, 2022. arXiv:2211.14365
2022 arXiv
-
[20]
Height function delocalisation on cubic planar graphs
Piet Lammers. Height function delocalisation on cubic planar graphs. Probab. Theory Related Fields , 182(1-2):531--550, 2022
2022
-
[21]
Tilted solid-on-solid is liquid: scaling limit of SOS with a potential on a slope, 2024
Benoît Laslier and Eyal Lubetzky. Tilted solid-on-solid is liquid: scaling limit of SOS with a potential on a slope, 2024. arXiv:2409.08745
2024 arXiv
-
[22]
Delocalisation and absolute-value- FKG in the solid-on-solid model
Piet Lammers and S\' e bastien Ott. Delocalisation and absolute-value- FKG in the solid-on-solid model. Probab. Theory Related Fields , 188(1-2):63--87, 2024
2024
-
[23]
Delocalization of two-dimensional random surfaces with hard-core constraints
Piotr Mi o \'s and Ron Peled. Delocalization of two-dimensional random surfaces with hard-core constraints. Communications in Mathematical Physics , 340(1):1--46, 2015
2015
-
[24]
van Beijeren
H. van Beijeren. Interface sharpness in the I sing system. Comm. Math. Phys. , 40(1):1--6, 1975
1975
-
[25]
An elementary proof of phase transition in the planar XY model
Diederik van Engelenburg and Marcin Lis. An elementary proof of phase transition in the planar XY model. Comm. Math. Phys. , 399(1):85--104, 2023
2023
-
[26]
On the duality between height functions and continuous spin models, 2023
Diederik van Engelenburg and Marcin Lis. On the duality between height functions and continuous spin models, 2023. arXiv:2303.08596
2023
-
[27]
Weeks, George H
John D. Weeks, George H. Gilmer, and Harry J. Leamy. Structural transition in the ising-model interface. Phys. Rev. Lett. , 31:549--551, 1973
1973
-
[28]
Maximum of the integer-valued Gaussian free field , 2019
Mateo Wirth. Maximum of the integer-valued Gaussian free field , 2019. arXiv:1907.08868
2019 arXiv
Reviewed August 7, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.