Pith.

Dynamical Systems · arXiv:2611.16294 v1 · submitted 12 Nov 2026 · 29 references

Nondifferentiability on Irrational Invariant Graphs of Smooth Area-Preserving Twist Maps

Lin Wang

The paper says

“For a fixed irrational rotation number and every 1 ≤ r < ∞, we construct an exact area-preserving Cr twist map preserving a Lipschitz graph. The set of nondifferentiability points has Hausdorff dimension d = 2/(r + 1) and zero d-dimensional Hausdorff measure.”

more of the abstract

At the same rotation number, we construct a C∞ twist map whose invariant graph is nondifferentiable on exactly one two-sided orbit. The dynamics on these graphs has wandering intervals.

from the abstract

Pith says

Machine review · deepseek-v4-flash · 13 Nov 2026 · what was read was not recorded · not checked by a person

It can be wrong. Challenge this review or reply to it.

A serious construction paper that delivers what it advertises — prescribed nondifferentiability dimension for every finite regularity plus a C^∞ example with one bad orbit — and deserves a real referee, with the arithmetic estimates getting the closest scrutiny. This line is wrong

the referee pass did not run

How often objections from reviews like this one are right: not yet measured

Are you an author? What you can do.

Not checked: figures, code, what the 29 references say.

What would settle it

Independently recompute the maximal Birkhoff sums M_n for the rotation number (1.3) using a computer algebra system for n up to, say, 10 (where q_{2n} is manageable). Verify that M_n ≍ n^5 and that the maximizing times k_n lie strictly between q_{2n} and q_{2n+2} with separation at least M_n - 4. If the numerical evidence contradicts the claimed asymptotics or the location of k_n, then Lemma 2.6 is false and the construction collapses. If the numerical evidence supports Lemma 2.6, then the concern is mitigated but not fully resolved, as the proof may still have gaps that do not affect the result.

from the skeptic's reading

How the review went

3 readings by deepseek-v4-flash · 13 Nov 2026

  1. Reader holds up confidence moderate
  2. Skeptic agrees 3 min 33 s · 43,873 tokens read, 1,697 written
  3. Editor wrote the verdict above 1 min 21 s · 45,280 tokens read, 12,465 written

Not run on this paper: the referee, the simulated authors' defence.

Pith did not check this review's objections against the paper.

The review in full

The review in brief, machine-written

The headline: this is a serious construction paper that delivers what it advertises, and it deserves a real referee. Theorem 1 gives, for every finite regularity 1 ≤ r < ∞, a Cr standard twist map preserving a Lipschitz graph whose nondifferentiability set has Hausdorff dimension d = 2/(r+1) and zero d-dimensional measure; Theorem 2 gives a C^∞ map whose graph fails differentiability on exactly one two-sided orbit. Together these answer Mather's Problem 3.1.1 in the nonminimal setting at every finite regularity and in the smooth class, and they go genuinely beyond Arnaud, Avila–Fayad, and Guo–Liu–Wang, which stopped at C1, C2, or C3−ε. The paper is honest too: the upper bound in Appendix C holds only under the local affine cancellation condition, and the C^∞ minimal-dynamics and real-analytic cases remain open.

What I liked: the architecture is clean. The Herman–Mather formula reduces everything to a circle lift g; the second-order recurrence (3.3) with two positive solutions Aj and Aj Tj makes the symmetric sum g + g−1 − 2Id cancel near the singular parameter set E while the individual slopes refuse to converge; Lemma 2.6 supplies large Birkhoff sums at the cutoff times. I checked the summability bookkeeping in Section 5 and the exponents work: hn = n9/2 dominates Mn+1 − Mn = O(n4), yet is swallowed by Mnτ with τ = 19/20.

The stress-test worry about grid points crossing the discontinuity in Lemma 2.6 does not land. Since h = q2n is odd, the grid points sit at least 1/(2h) from 1/2, and the displacement is at most about 1/(8h); the note's own arithmetic shows the slack. Signs are preserved and each block sums to +1.

Soft spots, in proportion. The load-bearing estimates — Lemma 2.6 and the interpolation bounds (3.15)–(3.18) — are hand-verified only: no code, no machine-checked proofs, and an arithmetic slip would cascade into length summability and the Cr−1 extension. That is where referee time should go. Minor: the comparison with prior work leans on several of the author's own arXiv preprints ([13], [19], [21], [23]); nothing wrong per se, but an independent check of the quoted results is warranted. And the sharpness of d = 2/(r+1) is conditional in a real way — stated plainly, so a scope limit, not a hidden flaw.

For anyone working on non-KAM invariant curves, Denjoy-type dynamics on graphs, or Mather's problems. Send to peer review; ask the referee to verify Lemma 2.6 in full and spot-check Section 5.

The strongest objection

The central construction depends on Lemma 2.6, which claims that for the rotation number (1.3) the maximal Birkhoff sums M_n satisfy M_n ≍ n^5 and that the maximizing times k_n lie strictly between consecutive return times with separation at least M_n - 4. These estimates enter the summability of interval lengths (3.14), the recurrence-error bounds (3.16)-(3.18), and the C^{r-1} extension across the minimal set (Lemma 6.5). The proof of Lemma 2.6 is a lengthy arithmetic argument that occupies about a page (lines 2.5-2.7). The lower bound uses a block construction with L = floor(a/8) blocks, each of length h = q_{2n}, and claims that the sum over each block is +1. This relies on the assertion that the displacement (l + i/h)ε is less than 1/(2h) for all l ≤ L and i < h, so that no grid point crosses 1/2. However, the displacement is (l + i/h)ε, and ε < 1/H = 1/(ah + q_{2n-1}). For l up to a/8, the maximum displacement is (a/8 + 1) * (1/(ah)) ≈ 1/(8h) + 1/(ah). This is indeed less than 1/(2h) for large a, but the argument also requires that the grid point (h-1)/(2h) does not cross 1/2. The displacement of this point is (l + (h-1)/(2h))ε. For l = a/8, this is about (a/8 + 1/2) * (1/(ah)) ≈ 1/(8h) + 1/(2ah). This is still less than 1/(2h) for large a. However, the proof also needs to ensure that the signs of the step function σ are exactly +1 for all points in the block. The step function σ is +1 on [0, 1/2) and -1 on [1/2, 1). The grid points are ip_m/h mod 1, where p_m/h is a convergent. The displacement shifts them to the right. The claim is that for l ≤ a/8, all points remain in [0, 1/2). But the grid points include points near 1/2 from below. The displacement could push a point that was just below 1/2 to just above 1/2, changing its sign. The proof asserts that the displacement is less than 1/(2h) and that the grid points are at least 1/(2h) from 1/2. However, the grid points are at distances that are multiples of 1/h from 0.

The reader's verdict identifies Lemma 2.6 as the weakest assumption, and I agree. The proof of Lemma 2.6 is the most fragile part of the paper because it is a lengthy arithmetic argument that is not fully rigorous. The rest of the construction depends on the precise asymptotics M_n ≍ n^5 and the location of k_n. If Lemma 2.6 were false, the constructed densities might not produce finite total length or a C^r perturbation. The paper acknowledges that the proof is lengthy but does not provide a more detailed justification. The concrete test would help to verify the claim numerically for small n, which would increase confidence in the result. However, the paper does not provide any numerical verification, and the proof is not machine-checked. Therefore, the concern is valid but does not necessarily invalidate the paper. The verdict should remain UNCHANGED because the paper provides a proof, albeit one that could be more rigorous. The reader's verdict of ACCEPT with MODERATE confidence is appropriate.

Pith's reading of the paper (a machine's summary, not the authors')

This paper addresses a well-known problem in dynamical systems: how smooth can an area-preserving twist map be if its invariant graph (a curve that the map preserves) is nonsmooth? The author constructs, for any prescribed finite regularity $C^r$ ($1\le r<\infty$), an exact area-preserving twist map that preserves a Lipschitz graph whose set of points where the graph fails to be differentiable has Hausdorff dimension exactly $2/(r+1)$ and zero $d$-dimensional Hausdorff measure. This is the maximum possible dimension within a natural class of perturbations. At the same rotation number, a $C^\infty$ map is built whose invariant graph is nondifferentiable on exactly one two-sided orbit, giving the first $C^\infty$ example of this kind. The dynamics on these graphs has wandering intervals, so the restricted dynamics is not minimal. These results answer Mather's Problem 3.1.1 for finite regularity and provide a smooth example for Problem 3.1.3, leaving open the cases of $C^\infty$ with minimal dynamics and real-analytic maps.

What carries the argument. The central object is a second-order linear recurrence (3.3) with coefficients $a_j^A = q(j\alpha)$ (the recurrence coefficient of the multiplier $\lambda=e^{\sigma}$). Two linearly independent positive solutions, $A_j$ and $A_jT_j$, are constructed from the orbit sums of the step function $\sigma$. The densities $\rho_j(t)$ on the $j$th wandering interval are taken as a combination $A_j(1+T_j\Theta(t)/4)$ near the singular parameter set $E$, where $\Theta$ oscillates between $-1$ and $1$. This makes the ratio $\rho_{j+1}/\rho_j$ have two distinct limits at each $t\in E$, so the circle map $g$ is nondifferentiable at those points. However, the symmetric sum $\rho_{j+1}+\rho_{j-1}-2\rho_j$ simplifies to $a_j^A\rho_j$ by linearity of the recurrence, so $\phi$ is affine near each singular point. Cutoff radii and interpolation at selected times (the maximizing times $k_n$) ensure summability of the interval lengths and control of the $C^{r-1}$ norm of the error terms, yielding $\phi\in C^r$.

Core claim. For every $1\le r<\infty$, there exists a periodic function $\phi\in C^r$ of mean zero such that the standard twist map $F_\phi(x,y)=(x+y,\,y+\phi(x+y))$ preserves the graph of a Lipschitz function $u:\mathbb{T}\to\mathbb{R}$ with induced rotation number $\alpha$ (a fixed irrational with super-exponential partial quotients). The nondifferentiability set $\mathcal{N}(u)$ has Hausdorff dimension $d=2/(r+1)$ and zero $d$-dimensional Hausdorff measure. For $r=\infty$, there is a $C^\infty$ map whose $\mathcal{N}(u)$ is exactly one two-sided orbit. The construction uses the Herman–Mather formula to reduce the problem to building an increasing circle lift $g$ with prescribed wandering intervals such that $\phi=g+g^{-1}-2\mathrm{Id}$ has the required regularity, even though $g$ itself is nondifferentiable on a large set. The key is a second-order linear recurrence that forces cancellation between the forward and backward slopes, so the symmetric sum is smooth where the individual slopes are not.

The premise it rests on. The construction depends on delicate arithmetic estimates for the maximal Birkhoff sums $M_n$ of the step function $\sigma$ on the orbit of zero, in particular Lemma 2.6 which asserts that $M_n\asymp n^5$ and that the maximizing times $k_n$ lie strictly between consecutive return times with separation at least $M_n-4$. These estimates enter the summability of interval lengths and the $C^{r-1}$ bounds on the error terms. If Lemma 2.6 were false, the constructed densities might not produce finite total length or a $C^r$ perturbation.

If the paper is right (Pith's inferences, not the authors')

The dimension $d=2/(r+1)$ is sharp in the class of standard twist maps whose perturbation is locally affine near the orbit of every nondifferentiability point in a wandering interval (Proposition C.1).

For $r=1$, the nondifferentiability set can have Hausdorff dimension 1 and zero Lebesgue measure, showing that the generic differentiability result of Arnaud does not imply a dimension bound below 1.

The $C^\infty$ example with a single nondifferentiable orbit provides a smooth twist map whose invariant graph is not $C^1$ and carries non-minimal (Denjoy-type) dynamics, answering Mather's Problems 3.1.1 and 3.1.3 simultaneously.

The construction separates the regularity of the map from that of its invariant graph in a sharp way: the map can be $C^r$ while the graph is Lipschitz but nondifferentiable on a set of prescribed dimension up to $2/(r+1)$.

The method of using a second-order recurrence to enforce cancellation in the symmetric sum $g+g^{-1}-2\mathrm{Id}$ may be adaptable to other settings where one wants a smooth function expressed as a sum of two nonsmooth functions.

The local affine condition (C.1) used in the sharpness proof is satisfied by the constructed examples but may not hold for all possible $C^r$ twist maps with nondifferentiable invariant graphs. If one could construct a $C^r$ map whose nondifferentiability set has dimension larger than $2/(r+1)$ without satisfying (C.1), the bound would be violated, so the condition is essential for the upper bound

The construction relies on a fixed rotation number $\alpha$ with very fast growth of partial quotients ($a_{2n+1}\sim 2n^4$). It is plausible that similar results hold for a larger set of frequencies (the paper defines an admissible set $\mathcal{S}$ of Hausdorff dimension 1), but the dimension of $\mathcal{N}(u)$ may depend on the Diophantine properties of $\alpha$ in a more refined way.

The $C^\infty$ example has nondifferentiability on a single orbit, which is the minimal possible nondifferentiability set for a non-$C^1$ invariant graph (since the graph must be nondifferentiable on at least one orbit by the dynamics). The remaining open case of a $C^\infty$ map with minimal dynamics (no wandering intervals) would require a different mechanism, because the interval insertion meth

The real-analytic case is fundamentally blocked by the fact that a non-constant real-analytic function cannot be affine on an interval. Thus any construction for the real-analytic setting would need a different cancellation mechanism, perhaps using a non-local identity.

Check this review

Pith stored no record of this review's passes and took no fingerprint of it. Reviews made by code from 13 Nov 2026 on carry both.

How often reviews like this one are right

Reviews like this one: written by deepseek-v4-flash, where what was read was not recorded.

How often objections from reviews like this one are right: not yet measured

Every value above, in the review's public record (JSON)

Reviews by people

This review was written by a machine. No person has reviewed this paper on Pith yet. A scientist listed on Pith can review it for a fee, and their review is published here whatever it concludes.

Request a human review

How Pith checks papers · The full audit of this paper

The paper

Abstract

For a fixed irrational rotation number and every 1 ≤ r < ∞, we construct an exact area-preserving Cr twist map preserving a Lipschitz graph. The set of nondifferentiability points has Hausdorff dimension d = 2/(r + 1) and zero d-dimensional Hausdorff measure. At the same rotation number, we construct a C∞ twist map whose invariant graph is nondifferentiable on exactly one two-sided orbit. The dynamics on these graphs has wandering intervals.

The paper's own sentences, arranged, kept beside the review and separate from it. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. No sentence in this record was written by a model.

abstract compiled · 2026-11-13

The authors' main claim, as the classifier picked it out 0.41

“The set of nondifferentiability points has Hausdorff dimension $d=2/(r+1)$ and zero $d$-dimensional Hausdorff measure.”

How the main claim is stated: asserted 1.00

What the abstract itself states, as probabilities: claims first 0.03 · states limitation 0.28 · human participants 0.01 · controlled comparison 0.01 · randomized assignment 0.01 · states quantitative result 0.97 · compares with prior results 0.07 · mentions code or data availability 0.01

The abstract, sentence by sentence

This table labels the abstract only (arXiv listing abstract v1). The number is the probability the classifier gave the label.

SentenceRoleThe authors' own workStates the main conclusionStrength
For a fixed irrational rotation number and every $1\le r<\infty$, we construct an exact area-preserving $C^r$ twist map preserving a Lipschitz graph. contribution 0.69 0.94 0.30 asserted 1.00
The set of nondifferentiability points has Hausdorff dimension $d=2/(r+1)$ and zero $d$-dimensional Hausdorff measure. result 1.00 0.88 0.34 asserted 1.00
At the same rotation number, we construct a $C^\infty$ twist map whose invariant graph is nondifferentiable on exactly one two-sided orbit. contribution 0.50 0.93 0.27 asserted 1.00
The dynamics on these graphs has wandering intervals. result 1.00 0.83 0.12 asserted 1.00

What kind of paper the classifier reports this is: theoretical result 1.00 · evidence from proof or derivation 1.00

What was examined, and what was not

Sources read

  • arXiv listing abstract v1

Not examined

  • full text
  • figures and tables
  • references and citing sentences
  • relations to other papers

Levels run

  • P abstract

Not run

  • S sections
  • T body sentences
  • C citations
  • R cross-paper relations

How the labels were made

Compiled by classifier-1.13.0 via TypeSafe API, question bank abstract-p1-20260923, 2026-11-13: 1 request, 3,604 input tokens.

Every quoted sentence on this record is the authors'. Every label is from Pith's closed vocabulary, with the probability shown beside it. This record as JSON

Figures (3)
Figure 1 of the paper
Figure 1. The three cutoffs in (3.15). If D(t) ≤ Rn+1, all three equal one and the difference Qj − q(jα) vanishes. We estimate all terms in (3.15) on a component of [0, 1] \ E0. For a smooth function f there, set [f]m,D(t) = Pm i=0 D(t) i |f (i) (t)|. The product rule gives [fg]m,D ≤ Cm[f]m,D[g]m,D. For f = Z and f = 1/Z one has v k |f (k) (v)| ≤ Ckf(v): this follows from the formulas outside [1, 2] and from positivity on the compact interval 1 ≤ v ≤ 2. For f(D/R) the same chain rule as in (3.9) gives [f(D/R)]m,D ≤ Cmf(D/R), with Cm independent of R. For the reciprocal of Hi differentiat
Figure 2 of the paper
Figure 2. The interpolation blocks on the two orbits, for positive indices
Figure 3 of the paper
Figure 3. For the same two parameter values t, t + h, the three chord lengths are ∆hXi = Xi(t + h) − Xi(t), i = j − 1, j, j + 1. Lemma 6.2 (The forward and backward derivatives). For each j, the following derivatives exist at almost every t ∈ (0, 1): (6.11) Dg(Xj (t)) = X′ j+1(t) X′ j (t) , D(g −1 )(Xj (t)) = X′ j−1 (t) X′ j (t)

Figures are the authors', reproduced from the paper.

References (29)
  1. [13] J. Guo, Y. Liu and L. Wang, On non-KAM invariant circles for area-preserving twist maps, preprint (2025). arXiv:2510.21098 on PitharXiv
  2. [19] Y. Huang, J. Li and L. Wang, On the sharpness of theC 1-norm threshold for perturbations in the normally hyperbolic invariant manifold theorem—a toy model perspective, preprint, arXiv:2608.04862 (2026) on PithPith's reviewarXiv
  3. [21] D. Kong, J. E. Massetti and L. Wang, Quantitative persistence and regularity of invariant graphs for dissipative twist maps, preprint, arXiv:2610.35068 (2026) on PithPith's reviewarXiv
  4. [23] Q. Li and L. Wang, On the dynamics of invariant graphs for dissipative twist maps, preprint, arXiv:2506.04938v3 (2026) on PithPith's reviewarXiv
  5. [1] M.-C. Arnaud, Three results on the regularity of the curves that are invariant by an exact symplectic twist map, Publ. Math. Inst. Hautes ´Etudes Sci.109(2009), 1–17 on Pith
  6. [2] M.-C. Arnaud, A non-differentiable essential irrational invariant curve for aC 1 symplectic twist map,J. Mod. Dyn.5(2011), 583–591 on Pith
  7. [3] M.-C. Arnaud, The link between the shape of the irrational Aubry–Mather sets and their Lyapunov exponents, Ann. of Math.174(2011), 1571–1601 on Pith
  8. [4] M.-C. Arnaud, Boundaries of instability zones for symplectic twist maps,J. Inst. Math. Jussieu13(2014), 19–41 on Pith
  9. [5] M.-C. Arnaud and P. Berger, The non-hyperbolicity of irrational invariant curves for twist maps and all that follows,Rev. Mat. Iberoam.32(2016), no. 4, 1295–1310 on Pith
  10. [6] M.-C. Arnaud and P. Le Calvez, A notion of Denjoy sub-system,C. R. Math. Acad. Sci. Paris355(2017), 914–919 on Pith
  11. [7] V. I. Arnol’d, Proof of a theorem of A. N. Kolmogorov on the invariance of quasi-periodic motions under small perturbations of the Hamiltonian,Russian Math. Surveys18(1963), no. 5, 9–36 on Pith
  12. [8] S. Aubry and P. Y. Le Daeron, The discrete Frenkel–Kontorova model and its extensions. I. Exact results for the ground-states,Physica D8(1983), 381–422 on Pith
  13. [9] A. Avila and B. Fayad, Non-differentiable irrational curves forC 1 twist map,Ergodic Theory Dynam. Systems 42(2022), 491–499 on Pith
  14. [10] V. Bangert, Mather sets for twist maps and geodesics on tori, inDynamics Reported, Vol. 1, U. Kirchgraber and H.-O. Walther (eds.), Wiley, Chichester, and Teubner, Stuttgart, 1988, 1–56 on Pith
  15. [11] M. Boshernitzan and D. Ralston, Continued fractions and heavy sequences,Proc. Amer. Math. Soc.137(2009), 3177–3185 on Pith
  16. [12] B. Fayad and R. Krikorian, Some questions around quasi-periodic dynamics, inProceedings of the International Congress of Mathematicians—Rio de Janeiro 2018, Vol. III, World Scientific, 2018, 1909–1932 on Pith
  17. [14] G. R. Hall, Bifurcation of an attracting invariant circle: a Denjoy attractor,Ergodic Theory Dynam. Systems3 (1983), 87–118 on Pith
  18. [15] D. Hensley, Continued fraction Cantor sets, Hausdorff dimension, and functional analysis,J. Number Theory40 (1992), 336–358 on Pith
  19. [16] M. R. Herman, Sur la conjugaison diff´ erentiable des diff´ eomorphismes du cercle ` a des rotations,Publ. Math. Inst. Hautes ´Etudes Sci.49(1979), 5–233 on Pith
  20. [17] M. R. Herman,Sur les courbes invariantes par les diff´ eomorphismes de l’anneau, Vol. I, Ast´ erisque103–104, Soci´ et´ e Math´ ematique de France, 1983, 1–221. 44 LIN W ANG on Pith
show 9 more
  1. [18] M. R. Herman,Sur les courbes invariantes par les diff´ eomorphismes de l’anneau, Vol. II, Ast´ erisque144, Soci´ et´ e Math´ ematique de France, 1986 on Pith
  2. [20] A. N. Kolmogorov, On conservation of conditionally periodic motions for a small change in Hamilton’s function, Dokl. Akad. Nauk SSSR98(1954), 527–530 on Pith
  3. [22] P. Le Calvez, Propri´ et´ es des attracteurs de Birkhoff,Ergodic Theory Dynam. Systems8(1988), 241–310 on Pith
  4. [24] S. Marmi and J.-C. Yoccoz, Some open problems related to small divisors, inDynamical Systems and Small Divisors(Cetraro, 13–20 June 1998), Lecture Notes in Math., vol. 1784, Springer, Berlin, 2002, 175–191 on Pith
  5. [25] J. N. Mather, Existence of quasi-periodic orbits for twist homeomorphisms of the annulus,Topology21(1982), 457–467 on Pith
  6. [26] J. N. Mather, Non-existence of invariant circles,Ergodic Theory Dynam. Systems4(1984), 301–309 on Pith
  7. [27] J. Moser, On invariant curves of area-preserving mappings of an annulus,Nachr. Akad. Wiss. G¨ ottingen Math.- Phys. Kl. II(1962), 1–20 on Pith
  8. [28] Ya. G. Sinai, Some problems in the theory of dynamical systems and mathematical physics,Geom. Funct. Anal., GAF A 2000 Special Volume, Part I (2000), 425–433. Lecture atVisions in Mathematics, Tel Aviv, 25 August–3 September 1999 on Pith
  9. [29] A. Sorrentino and L. Wang, On the destruction of invariant Lagrangian graphs for conformal symplectic twist maps,Calc. Var. Partial Differential Equations65(2026), article 153. School of Mathematics and Statistics, Beijing Institute of Technology, Beijing 100081, China Email a... on Pith

Challenges and discussion

Disagree with the review, or know something it missed? Leave a comment, or challenge this review formally: a challenge becomes a permanent entry in this paper's record, and its decision is published here.

No comments yet.

Sign in to comment