Convergence to a line
Diverging from lineStraightness S = ‖end − start‖ / L. Closed form for the Dragon Curve: S(n) = (1/√2)ⁿ. Convergence criterion: S ≥ 0.999. Analysis frame is capped at N_max = 100 iterations.
0.000.250.500.751.00
space-fillingstraight
Predicted S(n)n = 12.00 · (1/√2)^n
1.563e-2Total turningclosed form 90°·(2^n − 1)
368550.0°CapS(100) = 8.88e-16
n ≤ 100S(n) over n ∈ [0, 100]log scale
Iterations to drop below threshold
S < 0.5n ≥ 2
S < 0.1n ≥ 7
S < 0.01n ≥ 14
S < 10^-3n ≥ 20
S < 10^-6n ≥ 40
S < 10^-15n ≥ 100
Derived from (1/√2)ⁿ < τ ⇔ n > −2·log₂(τ).
Latest: “Recognizable dragon” at n = 10. Next: “A dragon” at n = 14.
✓
First creasen = 1
One fold. Still mostly straight.✓
L-bendn = 2
S drops below 0.5. (τ = 0.5)✓
Curling upn = 4
Sixteen segments. Shape emerging.✓
Decisively coiledn = 7
S drops below 0.1. (τ = 0.1)✓
Recognizable dragonn = 10
1,024 segments. Unmistakable silhouette.·
A dragonn = 14
S drops below 0.01. (τ = 0.01)·
Space-filler in trainingn = 20
S drops below 10⁻³. (τ = 10^-3)·
Beyond sightn = 40
S drops below 10⁻⁶. (τ = 10^-6)·
Numerical zeron = 100
Past double-precision epsilon. (τ = 10^-15)
Target iterations for straight-line state
n = 0 only. Unreachable for 1 ≤ n ≤ 100.
Each fold injects 2ⁿ − 1 fixed 90° turns. The summability test Σ 2ⁿ·θₙ < ∞ fails for constant θₙ = 90°, so straightness decays geometrically and cannot recover within the 100-iteration cap.
Variable sensitivity
- n — only lever; S decays as 2^(−n/2). Monotonic, one-way.
- ℓ (segment length) — uniform scale; cancels in ‖Δ‖/L. No effect on n.
- θ₀ (initial heading) — rigid rotation; geometry-invariant. No effect on n.
- t, easing, speed — re-time the fold but never alter integer-iteration geometry. No effect on n.
- stroke, palette, mode — render only. No effect on n.
Underlying math
RecurrenceDn+1 = Dn ∪ R−π/2, p(Dn)
Segment count|Dn| = 2n
Bounding ratio / step≈ √2 per iteration
Hausdorff dimdimH(boundary) ≈ 1.5236
Tiling factFour dragons tile the plane around their shared start point.