Pith. sign in

REVIEW 3 major objections 6 minor 9 references

The coordinate functions of the Heighway dragon curve

T0 review · 3 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read For every admissible dragon angle, the coordinate graphs' box dimension is exactly 1 - log(cos alpha)/log 2.

desk verdict The rational-angle theorem is real, but the irrational-angle extension has a multiplicative covering error that leaves the headline claim unproved. read the letter →

arxiv 2506.05541 v2 pith:U6D2QQMF submitted 2025-06-05 math.DS

Danilo Antonio Caprio This is my paper
classification math.DS MSC 28A8037C45
keywords Heighwaydragonpaper-foldingcurvebox-countingdimensioncoordinatefunctionsgeneralizedbinomial-coefficienttriangleself-similarfractalL-system
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper's claim is that for the generalized Heighway dragon curve $D_\theta$, with turning angle $\theta \in (\pi/3,5\pi/3)$ and half-angle $\alpha=(\pi-\theta)/2$, the graphs of the two coordinate functions $X_\theta$ and $Y_\theta$ have the same box-counting dimension, exactly $1 - \log(\cos\alpha)/\log 2$. This formula interpolates between dimension $1$ when the curve degenerates to a straight line at $\theta=\pi$ and dimension $2$ as $\theta$ approaches $\pi/3$, and it gives $3/2$ for the classical Heighway dragon at $\theta=\pi/2$. If correct, the result pins down the graph dimension of both coordinate functions for every admissible turning angle, rational or irrational, in terms of a single trigonometric quantity. A reader would care because exact dimensions of coordinate graphs for a whole continuous family of fractals are rare, and the argument reduces the problem to counting how many line-segment directions appear at each generation.

What carries the argument

The mechanism is the pair of stage functions $x_{\theta,k}, y_{\theta,k}$ parametrizing the $2^k$ segments of the $k$-th dragon generation, together with the substitution rule that generates the sequence of segment directions $b_{k,i}$. At generation $k$ the direction word is built by replacing each direction $a$ alternately by $a+\alpha, a-\alpha$; Lemma 3.4 shows that for rational $\alpha=(p/q)2\pi$, at generation $k=nq-1$ the directions are $k\alpha-2i\alpha$, and direction $i$ occurs $\binom{k}{i}$ times. This binomial count is what allows the exact covering estimates: the width of each piece is $2^{-k}$ and its vertical height is controlled by $(\cos\alpha)^{-k}$, so the number of mesh cubes needed grows like $(2/\cos\alpha)^k$, whose logarithm divided by $k\log 2$ gives $1 - \log(\cos\alpha)/\log 2$.

What would settle it

Pick an irrational $\alpha$, for example $\alpha=\pi/(2\sqrt2)$, and compute the number of $2^{-k}$-mesh squares hitting $X_\theta$ for $k$ from about 12 to 24; if the log-slope $\log N_{2^{-k}}/\log(2^k)$ does not approach $1-\log(\cos\alpha)/\log 2$, the theorem fails. For a rational $\alpha=p/q$, the stage counts at $k=nq-1$ are pinned by Lemma 3.5, so a direct enumeration of those $2^{-k}$-mesh squares for small $n$ would settle whether the lower-bound estimate holds.

Watch

Extended reading notes

Core claim

The central discovery is Theorem 2.2: for every $\theta \in (\pi/3,5\pi/3)$, writing $\alpha=(\pi-\theta)/2$, one has $\dim_B X_\theta = \dim_B Y_\theta = 1 - \log(\cos\alpha)/\log 2$. The proof splits by rationality. For rational $\alpha=(p/q)2\pi$, the angle words produced by the substitution rule are periodic modulo $2\pi$, and at generations $k=nq-1$ the segments' directions are $k\alpha-2i\alpha$ with multiplicities given by the binomial-coefficient triangle; this exact structure yields a covering of $X_\theta$ and $Y_\theta$ by mesh cubes of side $2^{-k}$ in Lemma 3.2 and a matching lower bound in Lemma 3.5. For irrational $\alpha$, the paper approximates $\theta$ by rational angles, proves uniform convergence of the corresponding coordinate functions in Lemma 3.6, and transfers the covering estimates to the limit. The foundational invariant, Lemma 3.1, is that every stage function agrees with the limit at dyadic rationals, so exact endpoint values are stable under refinement.

Load-bearing premise

The load-bearing premise is that approximating an irrational-angle dragon by rational-angle dragons with a fixed vertical error of $2^{-(m+1)}$ changes the count of needed $2^{-k}$ grid squares by only a constant at every finer scale, even where those squares are much smaller than the error; without that transfer, the lower-bound proof does not reach irrational angles.

Editorial extensions

If this is right

  • For $\theta=\pi/2$, the formula returns $3/2$, matching the known value for the classical Heighway dragon and for the coordinate graphs of the related dragon curve.
  • For all $\theta$, $\dim_B X_\theta=\dim_B Y_\theta$; the two coordinate graphs are always dimensionally identical even though they are different sets.
  • The dimension varies continuously and monotonically from $2$ (approached as $\theta\to\pi/3$) down to $1$ (attained at the straight-line case $\theta=\pi$).
  • For rational $\alpha=p/q$, the proof gives explicit checkable counting bounds: at scales $2^k$ with $k=nq-1$, at least $2^{k-1}(\lambda/(2\cos^k\alpha)-1)$ mesh boxes are required, where $\lambda$ is the smallest nonzero absolute cosine of a multiple of $\alpha$.
  • The result makes the coordinate functions' graph dimension depend only on the turning angle, so the whole family of generalized dragon curves has known box dimension without any numerical approximation.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A direct numerical test of the formula for one irrational angle, such as $\alpha/\pi=1/(2\sqrt2)$, would check the limit; the rational-angle estimates provide exact box counts at special dyadic scales, giving a fixed point for the comparison.
  • The binomial-coefficient structure suggests a transferable method: for any self-similar curve whose generation directions are produced by an alternating two-letter substitution and whose angle set is finite modulo $2\pi$, the same binomial counts may yield an exact coordinate-graph dimension.
  • The paper leaves Hausdorff dimension open; a next step would be to see whether the same covering estimates can be sharpened to arbitrary covers, starting with rational $\alpha$, where the binomial counts give the cleanest control.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. This paper studies the coordinate functions x_θ and y_θ of the generalized Heighway dragon curve D_θ for θ ∈ (π/3, 5π/3), with α = (π − θ)/2. The main theorem (Theorem 2.2) asserts that the box-counting dimensions of the graphs X_θ and Y_θ are both 1 − log(cos α)/log 2. The proof proceeds by constructing piecewise-linear approximations x_{θ,n}, y_{θ,n}, proving their uniform convergence (Lemma 2.1), establishing exact endpoint values at dyadic rationals (Lemma 3.1), giving a general upper cover estimate (Lemma 3.2), deriving a Pascal-triangle description of segment-angle multiplicities (Lemma 3.4), using it to obtain a lower bound at rational angles (Lemma 3.5), and finally attempting to pass from rational to irrational α via uniform convergence (Lemma 3.6 and Section 4). The classical case θ = π/2 gives the known value 3/2.

Significance. If the main theorem were fully proved, the paper would provide an exact closed-form graph dimension for both coordinate functions over the whole admissible angle range, interpolating between dimension 1 and dimension 2 and reproducing the known Heighway-dragon value 3/2. The rational-angle part is a genuine structural argument: Lemma 3.4 reduces angle multiplicities to Pascal-triangle rows, Lemma 3.5 uses exact dyadic endpoint values, and the classical case serves as an independent benchmark. There are no fitted constants and no obvious circularity. The weakness is concentrated in the irrational-angle extension, which is not established by the argument as written; hence the full theorem stated in the abstract remains unproved.

major comments (3)
  1. [§4, irrational lower-bound transfer] The claim that X_θ cannot be covered by 2^{k−1}(⌊λ_1/(2 cos α_m)^k⌋ − 1) boxes of side 2^{−k} does not follow from ‖x_{θ_m} − x_θ‖_∞ < 2^{−m−1}. A vertical perturbation of size ε = 2^{−m−1} at scale δ = 2^{−k} costs a multiplicative factor, not one box per column. If X_{θ_m} requires N_{m,k} δ-boxes, then any cover of X_θ by M δ-boxes gives, after vertical inflation by ε, a cover of X_{θ_m} by at most M(1 + 2⌈ε/δ⌉) boxes; hence M ≥ N_{m,k}/(1 + 2⌈ε/δ⌉) ≍ N_{m,k} 2^{m−k}. Since N_{m,k} ≍ 2^k/(2 cos α_m)^k, the transferred lower bound is only of order 2^m/(cos α_m)^k, whose box dimension is −log(cos α_m)/log 2, one full dimension below the claimed value. The +1/−1 accounting in the displayed formulas is therefore not a harmless roundoff.
  2. [§4, irrational upper-bound transfer] The same multiplicative loss invalidates the upper bound. Starting from the rational cover of X_{θ_m} of size N_{m,k} = 2^k(⌊2^{k+1}/(|2 cos α_m|^{k−1}(|2 cos α_m| − 1))⌋ + 1), vertical inflation to X_θ multiplies the required number of boxes by a factor 1 + 2⌈ε/δ⌉ ≍ 2^{k−m}, not by an additive constant. For fixed m and k → ∞, the resulting estimate has dimension tending to 2, not to 1 − log(cos α_m)/log 2. Thus the claimed upper bound for irrational θ is also unsupported.
  3. [§4, final inference] The final step 'dim X_θ ≥ 1 − log(cos α_m)/log 2 for all m, hence dim X_θ ≥ 1 − log(cos α)/log 2' is logically invalid. For each fixed m the estimates are only asserted for k ≥ m, and the factors 2^{m−k} or 2^{k−m} do not disappear after taking k → ∞. No estimate uniform in m is established. Since box-counting dimension is not continuous under uniform convergence, Lemma 3.6 alone cannot bridge rational and irrational angles; a genuinely different argument is required for the irrational case.
minor comments (6)
  1. [Remark 1] The definition of box-counting dimension has the limit written as δ → ∞; it should be δ → 0.
  2. [Lemma 3.5] The symbol Q_{k,i} is used without definition, and the letter m is overloaded (first as an index in '0 < m < q' and later as 'm = 2j'); please clarify the notation.
  3. [Definition 3.3] The relation ∼ is informal; since it underlies the Pascal-triangle lemma, a precise multiset definition would improve readability.
  4. [§4] In the irrational lower-bound argument, the displayed constant λ_1 should be λ_m; as written it is inconsistent with the definition of λ_m in the preceding sentence.
  5. [§5] There is a typo: 'Haussdorf' should be 'Hausdorff'.
  6. [Figure 7] The figure uses θ = 25π/18, which is outside the range θ ∈ (π/3, π] retained by the WLOG reduction; the caption should state the reflected equivalent angle.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the dimension formula is derived from the explicit recursive construction, and the classical θ=π/2 value serves only as an independent consistency check.

full rationale

The derivation is self-contained: Xθ and Yθ are defined from the explicit recursive segment-angle construction (Proposition 1, Lemmas 3.1 and 3.4), and the box-counting dimension formula is obtained by direct δ-mesh cube counts for rational α in Lemmas 3.2 and 3.5, then transferred to irrational α by uniform approximation in Lemma 3.6. No parameter is fitted to the claimed dimension, no equation is assumed in order to prove itself, and the proof contains no self-citations. The classical θ=π/2 case reproduces the known value 3/2 and functions as an independent benchmark, not as an input. There is a serious correctness gap in the irrational-angle transfer in §4: a uniform error 2^{-m-1} costs about 2^{k-m} boxes per column at scale 2^{-k}, not one box, so the ±1 adjustment is not an equivalence; box-counting dimension is not continuous under uniform convergence. That is a mathematical flaw in the transfer argument, not circularity, because the transferred estimates are not assumed to equal the target and the target formula is not smuggled back into the proof. The 'same way' and 'not hard to prove' remarks in Lemmas 3.5 and 3.6 are omitted proofs, but they do not create circular dependencies.

Assumptions & free parameters 0 free parameters · 6 assumptions · 0 invented entities

No free parameters are fitted: alpha and theta are geometric family parameters, and the auxiliary constants lambda1 and lambda2 (Lemma 3.5) are defined from the angle set and cancel out of the final formula. The axioms are mostly standard fractal-geometry facts plus the substitution-word encoding of the construction; the only ad hoc assumption is the irrational-angle transfer step, which is also the paper's weakest point. No new particles, forces, dimensions, or entities are introduced.

assumptions (6)
  • domain assumption The substitution word psi with rule (a_i + alpha, a_i - alpha) for odd positions and (a_i - alpha, a_i + alpha) for even positions faithfully encodes the geometric alternating-rotation dragon construction (Section 2).
    The paper defines the geometric recursion by alternating right/left rotations and then proves Lemma 3.4 (Pascal multiplicities) from the word substitution. The equivalence of the geometric rule and the word rule is asserted rather than derived, and the lemma chain depends on it.
  • standard math Box-counting dimension may be computed on the dyadic subsequence of meshes 2^(-k) with 2^(-k)-mesh cubes (Falconer [5]).
    Used in Remark 1 and throughout Lemmas 3.2 and 3.5; the dyadic-subsequence reduction is standard.
  • standard math Cauchy completeness of C[0,1] gives the limits x_theta and y_theta (Lemma 2.1).
    The uniform Cauchy estimate 4/(2 cos alpha)^n requires 2 cos alpha > 1, which follows from alpha in (0, pi/3).
  • domain assumption The case theta in (pi, 5*pi/3) reduces to 2*pi - theta by reflection across the horizontal axis (Section 2).
    Stated as WLOG with no proof or reference; plausible but unexamined, and the coordinate graphs' dimensions are invariant under the reflection only if the claimed symmetry is exact.
  • standard math Endpoint stability x_theta(j/2^k) = x_k(j/2^k) (Lemma 3.1).
    Proven in the paper from the substitution rule; it is the key mechanism that makes the rational lower bound exact at dyadic points.
  • ad hoc to paper Transfer of covering-number bounds from x_theta_m to x_theta under uniform closeness, with O(1) boxes lost per column for all k (Section 4, irrational case).
    This is the load-bearing unproven step: the fixed error 2^(-m-1) costs 2^(k-m-1) boxes per column at mesh 2^(-k), and box dimension is not continuous under uniform convergence. The statement is asserted via 'by the same idea of the proof of Lemma 3.5' with no quantitative justification.

how reviews work

0 comments
Cite this review

Pith. "Pith review of The coordinate functions of the Heighway dragon curve." pith.science (2026). https://pith.science/paper/U6D2QQMF

@misc{pith2026250605541,
  author       = {Pith},
  title        = {Pith review of: The coordinate functions of the Heighway dragon curve},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/U6D2QQMF}},
  note         = {Machine review of arXiv:2506.05541}
}
abstract

In this work, we study properties of the coordinate functions $x_\theta$ and $y_\theta$ of the dragon curve associated with the angle $\frac{\pi}{3} < \theta < \frac{5\pi}{3}$, and we prove that the box-counting dimension of its graph is equal to $1 - \frac{\log \cos\alpha}{\log 2}$, where $\alpha = \frac{\pi - \theta}{2}$.

Figures

Figures reproduced from arXiv: 2506.05541 by the authors.

Figure 1
Figure 1. Examples of dragon curve. In this work, we analyze the parametrization (xθ(t), yθ(t)) of the dragon curve Dθ for t ∈ [0, 1]. Our main focus concerns the fractal properties of the coordinate functions, particularly their box-counting dimensions (Minkowski-Bouligand dimensions). We establish that for the graphs Xθ := {(t, xθ(t)) : t ∈ [0, 1]} and Yθ := {(t, yθ(t)) : t ∈ [0, 1]}, their box-counting dimensions coincide … view at source ↗
Figure 2
Figure 2. illustrates the first six iterations of this construction for θ = π 2 , with each subsequent step shown in red [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Graphs of x π 2 ,1, x π 2 ,2 and x π 2 ,3, respectively [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Graphs of y π 2 ,1, y π 2 ,2 and y π 2 ,3, respectively. In order to define the coordinate functions xn and yn of Dn for each n ≥ 0, we first observe that by construction, each Dn consists of 2n line segments, each of length Ln = (2 cos α) −n . Let B := {nα : n ≥ 0} be…
Figure 5
Figure 5. Figure 5: Graphs of x π 2 and y π 2 [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Graphs of x 4π 9 and y 4π 9 [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Graphs of x 25π 18 and y 25π 18 . Consider π 3 < θ < 5π 3 , α = π−θ 2 and (xθ(t), yθ(t)) the parametrization of the dragon curve Dθ associated to the angle θ. The main result of this paper is the following [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 8
Figure 8. Figure 8: Graphs of x 5π 6 and y 5π 6 . Theorem 2.2. We have that the box-counting dimensions of Xθ and Yθ are dimB Xθ = dimB Yθ = 1 − log(cos α) log 2 . Remark 1. For proving that dimB Xθ = dimB Yθ = 1 − log(cos α) log 2 , we will use the following definition of the box-countin…
Figure 9
Figure 9. Figure 9: 1 2 k -mesh cubes covering the image of j−1 2 k , j 2 k [PITH_FULL_IMAGE:figures/full_fig_p010_9.png]
Figure 10
Figure 10. Figure 10: Boxes with side 1 2 k covering the image of [PITH_FULL_IMAGE:figures/full_fig_p014_10.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

9 extracted references · 9 canonical work pages

  1. [6]

    Falconer, (2021)

    K. Falconer, (2021). Intermediate Dimensions: A Survey. In: Pollicott, M., Vaienti, S. (eds) Thermodynamic Formalism. Lecture Notes in Mathemat- ics, vol 2290. Springer, Cham. https://doi.org/10.1007/978-3-030-74863- 0W 14 20 THE COORDINATE FUNCTIONS OF THE HEIGHW AY DRAGON CUR VE

    work page doi:10.1007/978-3-030-74863- 2021

  2. [7]

    Falconer and T

    K. Falconer and T. Kempton, Planar self-affine sets with equal Hausdorff, box and affinity dimensions, Ergodic Theory and Dynamical Systems, 38 (4), 1369-1388 (2018)

    work page 2018

  3. [1]

    Akiyama, Y

    S. Akiyama, Y. Kamiya, F. Wen, Non-self-intersective Dragon curves, Indagationes Mathematicae (2025),ISSN 0019-3577, https://doi.org/10.1016/j.indag.2025.03.006

    work page doi:10.1016/j.indag.2025.03.006 2025

  4. [2]

    Albers, Papierfalten, Doctoral thesis, Bremen Uni- versity, (2006)

    R. Albers, Papierfalten, Doctoral thesis, Bremen Uni- versity, (2006). Available at http://www. math.uni- bremen.de/didaktik/ma/ralbers/Publikationen/Dissertation/index.html

    work page 2006

  5. [3]

    Allaart and K

    P. Allaart and K. Kawamura, The coordinate functions of L´ evy Dragon curve, Real Analysis Exchange, 31(1), 295–308 (2005/2006)

    work page 2005

  6. [4]

    Allouche, M

    J.-P. Allouche, M. Mend` es France, G. Skordev, Non-intersectivity of Paper- folding Dragon Curves and of Curves Generated by Automatic Sequences, Vol. 18A, Integers (2018)

    work page 2018

  7. [5]

    Falconer, Fractal Geometry - Mathematical Foundations and Applica- tions, 3rd Ed., John Wiley, 2014

    K. Falconer, Fractal Geometry - Mathematical Foundations and Applica- tions, 3rd Ed., John Wiley, 2014

    work page 2014

  8. [8]

    Nguyen, The Heighway dragon revisited, Discrete and Computational Geometry (2003) doi: 10.1007/S00454-003-0778-7

    N. Nguyen, The Heighway dragon revisited, Discrete and Computational Geometry (2003) doi: 10.1007/S00454-003-0778-7

    work page doi:10.1007/s00454-003-0778-7 2003

Show all 9 references
  1. [9]

    Tabachnikov, Dragon curves revisited, Math

    S. Tabachnikov, Dragon curves revisited, Math. Intelligencer 36 (2014) 13–17

    2014

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.