Pith. sign in

REVIEW 2 major objections 6 minor 10 references

Disordered layer confirmed between ordered phases in 2D Potts model

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · glm-5.2

2026-07-04 22:17 UTC pith:34YIT4AB

load-bearing objection First rigorous geometric description of interfacial wetting in a lattice spin model; the main vulnerability is structural dependence on a companion paper for OZ theory and mixing estimates. the 2 major comments →

arxiv 2604.21668 v2 pith:34YIT4AB submitted 2026-04-23 cs.CV

Encoder-Free Human Motion Understanding via Structured Motion Descriptions

Yao Zhang , Zhuchenyang Liu , Thomas Ploetz , Yu Xiao This is my paper
classification cs.CV MSC 60K3582B2082B43 PACS 05.50.+q64.60.Cn68.08.-p
keywords Potts modelinterfacial wettingBrownian watermelonOrnstein-Zernike theoryAshkin-Teller modelentropic repulsionphase transitionstatistical mechanics
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper proves that in the two-dimensional Potts model with q>4 states, at the critical temperature where the phase transition is discontinuous, two different ordered phases do not meet directly. Instead, a mesoscopic layer of the disordered phase spontaneously emerges between them. Under diffusive (square-root) scaling, the two boundaries of this layer converge to a Brownian watermelon — a pair of Brownian bridges conditioned never to intersect. This is the first rigorous geometric description of interfacial wetting in a lattice spin model. The proof proceeds by mapping the Potts model through a chain of couplings (via the six-vertex model) to the Ashkin-Teller random-cluster (ATRC) model, where the two interfaces become crossing clusters in dual percolation configurations. The key structural insight is that these clusters exhibit entropic repulsion — they are not energetically attracted to each other, so for purely entropic reasons they stay far apart. This separation allows each interface to be treated independently using Ornstein-Zernike renewal theory (developed in a companion paper), and their joint law is then coupled to a pair of random walk bridges conditioned on mutual non-intersection, which is known to converge to the Brownian watermelon. The same convergence is established for the FK-percolation representation and for level lines in the six-vertex model.

Core claim

The central object is the Brownian watermelon — two Brownian bridges conditioned not to cross — which emerges as the universal scaling limit of the pair of interfaces bounding the wetting layer. The mechanism carrying the argument is entropic repulsion between two simultaneously constrained interfaces in the ATRC model: because the model is positively associated (FKG), proximity of the two interfaces brings no energetic benefit, so entropy drives them apart. Once separated, each interface behaves like an independent renewal process with exponential mixing, and the problem reduces to coupling two conditioned random walks. This is the only regime among all Potts Dobrushin interfaces that does,

What carries the argument

Chain of couplings (Potts → six-vertex → ATRC), Ornstein-Zernike renewal theory for single interfaces, entropic repulsion via FKG monotonicity, coupling to non-intersecting random walk bridges, invariance principle for walks in cones

Load-bearing premise

The proof imports the Ornstein-Zernike theory for a single order-disorder interface in the ATRC model — including the renewal picture, infinite-volume decomposition, and strong mixing estimates — from a companion paper. If those imported results have gaps, the coupling of the two interfaces to non-intersecting random walks would not hold.

What would settle it

If the mixing or renewal estimates for the ATRC model fail (e.g., if correlations decay slower than exponential, or if the renewal structure does not hold), the coupling to random walks breaks and the convergence to the Brownian watermelon would be unproven.

Watch this falsifier — get emailed when new claim-graph text bears on it.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

2 major / 6 minor

Summary. The paper studies the q-state Potts model on the square lattice at the critical temperature T_c(q) for q > 4, under order-order Dobrushin boundary conditions. The main result (Theorem 1.1) establishes that a mesoscopic layer of disordered phase emerges between the two ordered phases, and that under diffusive scaling, the boundaries of this layer converge to a Brownian watermelon (two Brownian bridges conditioned not to intersect). An analogous result is proved for the FK percolation model (Theorem 1.2). The proof proceeds in three steps: (1) coupling the FK model to a modified ATRC model via the six-vertex model (Section 3), (2) showing proximity of the respective interfaces (Section 4), and (3) coupling the two ATRC crossing clusters to a pair of non-intersecting random walk bridges via a total variation bound (Section 5, Theorem 5.5), combined with an invariance principle for non-intersecting walks (Theorem B.1, imported from [DA24]). The key novel contributions are the entropic repulsion argument (Lemma 5.17) and the total variation bound (Theorem 5.5), which adapt the method of [IOVW20] to the ATRC setting.

Significance. This is the first rigorous geometric description of interfacial wetting in a lattice spin model at the critical point, in the full regime q > 4. The result completes the study of Dobrushin interfaces in the Potts model: the subcritical regime was treated in [CIV08], the order-disorder case in [DGO25], and this paper handles the most involved case of order-order interfaces at T_c(q), which is the only case exhibiting the wetting phenomenon and a non-Gaussian limiting process. The proof strategy of mapping the Potts model to the ATRC model and using its positive association properties to establish effective repulsion between interfaces is a notable methodological contribution. The chain of total variation bounds in the proof of Theorem 5.5 (equations 57–61) is intricate but each step is justified by a specific lemma, and the verification that the measure μ_Λ satisfies Hypotheses (I)–(V) of [DGO25, Section 8] is carried out explicitly (Section 5.2). The diffusivity constant σ is externally characterized via [DGO25, Remark 5.5], not fitted.

major comments (2)
  1. The paper's central argument depends heavily on the companion paper [DGO25] for the Ornstein–Zernike theory of single interfaces (Theorem 5.1), mixing estimates (Lemma 5.4), and exponential relaxation of quasi-FK measures (Lemma 4.5). This is a structural dependency, not an internal inconsistency: the paper verifies in Section 5.2 that μ_Λ satisfies the required hypotheses (positive association via Lemma 3.8, ratio weak mixing, finite energy). However, the correctness of the main result is conditional on the results in [DGO25] being correct. This is acceptable for a companion-paper structure, but the authors should state this dependency explicitly in the introduction (it is currently mentioned only in passing in Section 1.4 and Section 5.2).
  2. Lemma 4.1 (proximity of upper FK and AT interfaces) is stated with a proof that is entirely omitted ('The proof [DGO25, Lemma 9.1] extends readily to this case and we omit the details'). While the claim that the extension is straightforward may be correct, this is a load-bearing result for Step 1 of the proof of Theorem 1.2. At minimum, the authors should indicate which modifications are needed compared to [DGO25, Lemma 9.1] (e.g., the change from order-disorder to order-order boundary conditions) and why they do not affect the argument.
minor comments (6)
  1. The constants α, β in Theorem 5.5 are acknowledged as non-optimal (the authors note they 'lose powers of log everywhere'). This is fine for the purpose of the result, but the specific values of α and β are never stated explicitly in the theorem. It would help the reader to state them, even as 'for some α, β > 0 depending only on q.'
  2. In the proof of Lemma 5.17, the reduction to Lemma 5.13 involves several intermediate steps (equations around lines 47–48). The role of the parameter ε_0 (the 'arbitrarily fixed constant, taken large enough') could be stated more precisely — specifically, what lower bound on ε_0 is needed for the argument to work.
  3. The notation for the modified distance d_mod(·,·) in Lemma 5.17 uses a parameter ε_0 that is distinct from the ε_0 in the definition of good pairs G_ε. Using different symbols would avoid confusion.
  4. In Section 5.2, the verification that μ_Λ satisfies Hypotheses (I)–(V) is stated but the verification of each hypothesis is brief. A short remark indicating which hypothesis corresponds to which property (positive association, mixing, etc.) would improve readability.
  5. The paper uses the notation Γ_1, Γ_2 for crossing clusters in Section 5, but these are defined differently in different subsections (sometimes as clusters, sometimes as paths). A consistent notation would help.
  6. Reference [OV25] (arXiv:2511.09274) has a future-looking arXiv number. Please verify this reference is correct and publicly available.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading and the positive assessment. Both comments are well-taken and will be addressed in the revision.

read point-by-point responses

  1. Referee: The paper's central argument depends heavily on the companion paper [DGO25] for the Ornstein–Zernike theory of single interfaces (Theorem 5.1), mixing estimates (Lemma 5.4), and exponential relaxation of quasi-FK measures (Lemma 4.5). The authors should state this dependency explicitly in the introduction.

    Authors: We agree. The dependency on [DGO25] is structural: the OZ theory for single ATRC interfaces (Theorem 5.1), the ratio weak mixing property of the ATRC model (used in Lemma 5.4 and Lemma 4.5), and the verification that μ_Λ satisfies Hypotheses (I)–(V) of [DGO25, Section 8] (carried out in Section 5.2) are all essential inputs. Currently, this is mentioned in Section 1.4 and Section 5.2 but not flagged prominently in the introduction. We will add an explicit paragraph in Section 1 (after the statement of Theorem 1.1) clarifying that the correctness of the main results is contingent on the results of [DGO25], and listing the specific imported statements: the infinite-volume OZ decomposition (Theorem 5.1), the mixing estimates (Lemma 5.4), and the exponential relaxation of quasi-FK measures (Lemma 4.5). We will also note that Section 5.2 verifies the hypotheses required by [DGO25, Theorem 8.1]. revision: yes

  2. Referee: Lemma 4.1 (proximity of upper FK and AT interfaces) is stated with a proof that is entirely omitted. The authors should indicate which modifications are needed compared to [DGO25, Lemma 9.1] and why they do not affect the argument.

    Authors: We agree that the omission is not appropriate for a load-bearing result. The proof of Lemma 4.1 follows the same strategy as [DGO25, Lemma 9.1], but with order-order (wired-wired) Dobrushin boundary conditions in place of order-disorder (wired-free) conditions. The key modifications are: (1) the six-vertex spin boundary conditions are 1 on the lower boundary and 2 on the upper boundary (instead of 1 below and free above), which changes the boundary conditions on the six-vertex spins above the upper AT interface from 'free' to '2'; (2) conditional on a realisation of the upper AT interface Γ^AT, the distribution of the six-vertex spins above Γ^AT has '+' boundary conditions (in the notation of Definition 3.2), and paths of ω^1 above induce circuits in ω^2, which has uniform exponential decay by [DGO25, Theorem 3.6] — exactly as in the order-disorder case. The reason the argument goes through unchanged is that the exponential decay of dual connections above the interface depends only on the local boundary condition being 'ordered' (which holds in both cases: free in the order-disorder case, and colour 2 in the order-order case), and the Peierls contour between σ^1 and σ^2 in the six-vertex spin is contained in Γ^AT by the same deterministic inclusion as in [DGO25]. We will add a proof sketch of approximately one paragraph to the manuscript, specifying these modifications and referencing the corresponding steps in [DGO25, Lemma 9.1]. revision: yes

Circularity Check

0 steps flagged

No significant circularity. Heavy but legitimate self-citation to companion paper [DGO25]; the imported results concern single-interface OZ theory, while the paper's central contribution is the novel two-interface entropic repulsion and coupling.

full rationale

The paper proves that order-order Potts interfaces at T_c(q) converge to a Brownian watermelon. The derivation chain is: (1) couple FK to modified ATRC (Section 3, Lemmas 3.3, 3.6), (2) show FK interfaces are close to ATRC interfaces (Section 4, Lemmas 4.1, 4.2), (3) couple ATRC crossing clusters to random walk bridges conditioned on non-intersection (Section 5, Theorem 5.5, Corollary 5.6), and (4) invoke convergence of non-intersecting walks to Brownian watermelon (Theorem B.1, from [DA24] by a different author). The companion paper [DGO25] is cited extensively for: the infinite-volume OZ decomposition (Theorem 5.1 = [DGO25, Theorem 7.1]), mixing (Lemma 5.4 = [DGO25, Lemma 8.3]), exponential relaxation (Lemma 4.5 from [DGO25, Proposition 5.8]), and several structural lemmas (5.15, 5.16, 5.19, 5.20, 5.22). However, [DGO25] develops OZ theory for a SINGLE order-disorder interface, while this paper's central novelty is handling TWO interacting interfaces via the entropic repulsion argument (Lemma 5.17) and the total variation bound (Theorem 5.5). The paper explicitly verifies that its measure satisfies Hypotheses (I)-(V) of [DGO25, Section 8] (positive association via Lemma 3.8, ratio weak mixing, finite energy), so the application of [DGO25] is condition-checked, not assumed. The diffusivity constant σ has an explicit characterization ([DGO25, Remark 5.5]) and is not fitted to the target result. No step reduces to its inputs by construction: the OZ decomposition describes single-interface geometry, while the output describes coupled two-interface geometry. The self-citation is load-bearing but provides genuinely independent building blocks, not a restatement of the target result. Score 2 reflects the heavy structural dependence on [DGO25] without any actual circular reduction.

Axiom & Free-Parameter Ledger

0 free parameters · 3 axioms · 0 invented entities

The paper introduces no free parameters or invented entities. It works entirely within the established framework of the Potts, FK percolation, six-vertex, and ATRC models. The axioms are standard results from the statistical mechanics literature, primarily imported from the authors' own companion paper, but these are parameter-free derivations, not ad hoc assumptions.

axioms (3)
  • domain assumption The ATRC model has a unique infinite-volume Gibbs measure and satisfies exponential ratio mixing.
    Imported from [ADG24] and [DGO25]. This is the foundation for the OZ theory and mixing properties used throughout the paper.
  • domain assumption The Ornstein-Zernike theory for a single interface in the ATRC model holds as developed in the companion paper.
    Imported from [DGO25, Theorem 7.1 and 8.1]. This provides the renewal picture and cone-point structure for single crossings, which is extended to pairs of crossings here.
  • standard math The invariance principle for a pair of random walks conditioned on non-intersection.
    Imported from [DA24] and [DW20]. This is the final step in the convergence to the Brownian watermelon.

pith-pipeline@v1.1.0-glm · 52268 in / 1795 out tokens · 295571 ms · 2026-07-04T22:17:55.956722+00:00 · methodology

0 comments
read the original abstract

The world knowledge and reasoning capabilities of text-based large language models (LLMs) are advancing rapidly, yet current approaches to human motion understanding, including motion question answering and captioning, have not fully exploited these capabilities. Existing LLM-based methods typically learn motion-language alignment through dedicated encoders that project motion features into the LLM's embedding space, remaining constrained by cross-modal representation and alignment. Inspired by biomechanical analysis, where joint angles and body-part kinematics have long served as a precise descriptive language for human movement, we propose \textbf{Structured Motion Description (SMD)}, a rule-based, deterministic approach that converts joint position sequences into structured natural language descriptions of joint angles, body part movements, and global trajectory. By representing motion as text, SMD enables LLMs to apply their pretrained knowledge of body parts, spatial directions, and movement semantics directly to motion reasoning, without requiring learned encoders or alignment modules. We show that this approach goes beyond state-of-the-art results on both motion question answering (66.7\% on BABEL-QA, 90.1\% on HuMMan-QA) and motion captioning (R@1 of 0.584, CIDEr of 53.16 on HumanML3D), surpassing all prior methods. SMD additionally offers practical benefits: the same text input works across different LLMs with only lightweight LoRA adaptation (validated on 8 LLMs from 6 model families), and its human-readable representation enables interpretable attention analysis over motion descriptions. Code, data, and pretrained LoRA adapters are available at https://yaozhang182.github.io/motion-smd/.

Figures

Figures reproduced from arXiv: 2604.21668 by Thomas Ploetz, Yao Zhang, Yu Xiao, Zhuchenyang Liu.

Figure 1
Figure 1. Figure 1: Comparison of (a) the previous encoder-based paradigm, which requires a complex learned [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of our approach. (a) Stage 1 (top): a deterministic, rule-based pipeline processes [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Prompt structure for (a) motion QA and (b) motion captioning. [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Zero-shot captioning examples with motion visualizations and ground-truth captions. (a) [PITH_FULL_IMAGE:figures/full_fig_p010_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Attention heatmaps for two captioning examples with their corresponding motion visual [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Illustration of the body-local and joint-local coordinate frames used for joint angle compu [PITH_FULL_IMAGE:figures/full_fig_p015_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Complete All-26 SMD for motion 014160 (“a person waves with his right hand,” 4.1s). [PITH_FULL_IMAGE:figures/full_fig_p017_7.png] view at source ↗

Review history (2 revisions) →

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

10 extracted references · 10 canonical work pages · 1 internal anchor

  1. [1]

    infinitevolumeOZtheory

    � � �� � ���� � � � ��� � ���� � ��� � �� �� � � �� ��� � � �� � � ��� � ���� � ��� � ���� �� Notice that��� ��� ��� ��. Moreover, by Lemma 4.5, we have ��� ���� � � ���� �������� ������ �� ����� ����(15) Hence, we can assume that�����occurs. Now, consider the event�that all vertices on the top�� � �� � are connected to each other outside�� �� � and that ...

    work page 2024

  2. [2]

    [AOV24] Y. Aoun, S. Ott, and Y. Velenik,Ornstein–Zernike behavior for Ising models with infinite-range interactions, Annales de l’Institut Henri Poincare (B) Probabilites et statistiques��(2024), no. 1, 167–207. [AT43] J. Ashkin and E. Teller,Statistics of two-dimensional lattices with four components, Physical Review��(1943), no. 5-6, 178–184. [BD12] V. ...

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  3. [3]

    [DS86] Bernard Derrida and M Schick,Interfacial wetting in the q-state Potts model, Journal of physics A: mathematical and general��(1986), no. 8,

    work page 1986

  4. [4]

    [DST17] Hugo Duminil-Copin, Vladas Sidoravicius, and Vincent Tassion,Continuity of the phase transition for planar random-cluster and Potts models with1�q�4, Comm. Math. Phys.���(2017), no. 1, 47–107. [DW15] Denis Denisov and Vitali Wachtel,Random walks in cones, Annals of probability�� (2015), no. 3, 992–1044. [DW20] Jetlir Duraj and Vitali Wachtel,Invar...

    work page 2017

  5. [5]

    [Fan72b] Chungpeng Fan,Remarks on the eight-vertex model and the ashkin-teller model of lattice statistics, Phys. Rev. Lett.��(1972), 158–160. [Fel91] William Feller,An introduction to probability theory and its applications, volume 2, vol. 2, John Wiley & Sons,

    work page 1972

  6. [6]

    [FK72] C. M. Fortuin and P. W. Kasteleyn,On the random-cluster model. I. Introduction and relation to other models, Physica��(1972), 536–564. [GL23] Alexander Glazman and Piet Lammers,Delocalisation and continuity in 2D: loop O(2), six-vertex, and random-cluster models, arXiv preprint arXiv:2306.01527 (2023). [GM23] Alexander Glazman and Ioan Manolescu,St...

    work page arXiv 1972

  7. [7]

    3-4, 431–476

    [HK02] Ostap Hryniv and Roman Koteck` y,Surface tension and the Ornstein–Zernike be- haviour for the 2D Blume–Capel model, Journal of statistical physics���(2002), no. 3-4, 431–476. [Hol74] Richard Holley,Remarks on theFKGinequalities, Comm. Math. Phys.��(1974), 227–231. MR 0341552 (49 #6300) [IOVW20] Dmitry Ioffe, Sébastien Ott, Yvan Velenik, and Vitali ...

    work page 2002

  8. [8]

    Kadanoff and Franz J

    MR 3313616 [KW71] Leo P. Kadanoff and Franz J. Wegner,Some critical properties of the eight-vertex model, Phys. Rev. B�(1971), 3989–3993, 10.1103/PhysRevB.4.3989. [Lis22] Marcin Lis,Spins, percolation and height functions, Electronic Journal of Probability ��(2022), 1–21. [LMMS� 91] Lahoussine Laanait, Alain Messager, Salvador Miracle-Solé, Jean Ruiz, and...

    work page doi:10.1103/physrevb.4.3989 1971

  9. [9]

    1, 309–340

    [OV19] Sébastien Ott and Yvan Velenik,Asymptotics of even–even correlations in the Ising model, Probability Theory and Related Fields���(2019), no. 1, 309–340. [OV25] ,Finite-time trajectorial estimates for inhomogeneous random walks, arXiv preprint arXiv:2511.09274 (2025). [Pau35] Linus Pauling,The structure and entropy of ice and of other crystals with ...

    work page arXiv 2019

  10. [10]

    [Weg72] F

    [Vel06] Yvan Velenik,Localization and delocalization of random interfaces. [Weg72] F. Wegner,Duality relation between the Ashkin-Teller and the eight-vertex model, Journal of physics. C, Solid state physics�(1972), no. 11, L131–L132. [Wu71] F.W. Wu,Ising model with four-spin interactions, Phys. Rev. B�(1971), 2312–2314. F akultät für Mathematik, Universit...

    work page 1972