Abstract Art: Strange Attractor Painter

Millions of orbit points of a Clifford, de Jong or Svensson map are counted per pixel and painted by log-density. Live chaos, dimension and confinement checks.

Switch to 3D

Presets

Playback

Map

Parameters

Layers

Model check

The science

What is modelled

A picture grown from one equation. A point is moved again and again by a two-dimensional map. After a short burn-in the point never settles, it wanders over a thin, folded set called a strange attractor. Every pixel counts how often the orbit landed in it; the colour comes from the logarithm of that count, and the hue from how fast the orbit was moving there.

Equations

  • Clifford: x' = sin(a y) + c cos(a x), y' = sin(b x) + d cos(b y). de Jong: x' = sin(a y) − cos(b x), y' = sin(c x) − cos(d y). Svensson: x' = d sin(a x) − sin(b y), y' = c cos(a x) + cos(b y).
  • Confinement: sines and cosines are bounded by 1, so Clifford orbits stay inside |x| ≤ 1+|c|, |y| ≤ 1+|d|; de Jong inside 2 × 2. The page uses this box to scale the picture and checks that no orbit point ever leaves it.
  • Largest Lyapunov exponent: λ1 = lim (1/n) ∑ ln |Jk vk| with the tangent vector renormalised each step. λ1 > 0 means nearby orbits separate exponentially, i.e. chaos. The mean of ln |det J| equals λ1+λ2.
  • Kaplan-Yorke dimension: DKY = 1 + λ1/|λ2| (capped at 2 when the map does not contract area on average).
  • Box counting: N(ε) ∝ ε−D; the slope of ln N against ln(1/ε) estimates the fractal dimension D.
  • Tone mapping: brightness b = [ln(1 + g n) / ln(1 + g nmax)]γ for hit count n, so a million-hit core and a ten-hit filament are both visible.
  • Symmetry copies rotate every plotted point by 2πk/m about the picture centre; they do not change the map.

Checks run in the page

  • Two independent Lyapunov estimates (Jacobian tangent method and the two-orbit Benettin method that uses only the map itself) must agree within 0.02.
  • All orbit points stay inside the analytic box and the histogram contains exactly the number of hits that were counted (exact integers).
  • The Kaplan-Yorke dimension from the exponents is compared with a box-counting dimension measured on a separate 400,000-point sample (tolerance 0.3, because finite-sample box counting under-estimates thin fractal sets).

Honest simplifications

This is a discrete map, not a physical system, and the colours are artistic. The box-counting slope is measured over a limited range of scales (about one decade), so it is an estimate, not a proof of fractality. The Kaplan-Yorke formula is a conjecture that holds for many but not all attractors. Presets were chosen because they are chaotic at the stated parameters; many random parameter sets give only a point or a closed curve, and the page says so in the third check.

What to try

  • Press Surprise me several times; the page only accepts parameters whose measured λ1 is positive.
  • Take the Nebula preset and move a by 0.01: the picture changes character while the Lyapunov exponent barely does.
  • Choose Quiet: fixed point and see the orbit collapse to a dot with a negative exponent.
  • Set Parameter drift above zero to watch the attractor breathe; light fades with a 7 % decay per frame.
Box counting: ln N vs ln(1/eps)
Lyapunov exponent convergence
Distribution of hits per pixel
Return map x(n) vs x(n+1)