← All exhibits

Fourier drawing

How can rotating circles draw a shape?

Rotating arrows are joined tip to tail. The end of the last arrow follows the reconstructed outline.
  • Original outline
  • Reconstruction
  • Rotating circles
6

RMS sample error —

How to read this

Start with one circle, then add more. Watch the gold curve bring out the plane’s nose and notch as it approaches the dashed outline.

What to look for

Move the slider to one circle. Its arrow sweeps around at a steady rate, so the gold point traces a loop.

As you add circles, each new arrow starts where the last one ends. The tip of the chain follows the combined motion. Watch the nose and the notch as the gold curve approaches the dashed outline.

The display keeps the strongest rotations first. A circle’s size shows how much it contributes. Its speed tells you how many times it turns per lap. Negative frequencies turn clockwise.

The stationary term sets the center of the drawing and stays in place even when the slider is at zero.

The rule

The exhibit samples 129 points at equal distances along the closed outline. It reads each point as a complex number with its horizontal coordinate as the real part and its vertical coordinate as the imaginary part. The discrete Fourier transform finds a coefficient for each rotation by comparing that rotation to the samples.

z(t)= ∑k∈S ck e2πikt

Here t is the fraction of a lap, i is the imaginary unit, and S is the set of frequencies currently included. Each coefficient ck gives its arrow a length and a starting angle. The factor after it rotates that arrow k times per lap. Adding the arrows gives the reconstructed position z.

Reading the error

The RMS sample error is the root mean square of the distances between the gold curve and the original points at the 129 sample times. It uses the drawing’s coordinate units. At the maximum, 128 rotating terms plus the stationary term recover those sample points within floating-point rounding. Between sample points the reconstruction is a smooth periodic curve, so it can round a sharp corner or overshoot it slightly.

Where the idea appears

Frequency analysis uses the same change of representation. For example, an engineer can examine the strongest frequencies in a vibration recording to see which repeating motions dominate. Here, the signal happens to be a position moving around a plane.

Kuramoto synchronization uses rotating arrows too. There, the arrows describe rhythms that influence one another.

Further reading