Toothed-Wheel Light Gate (2D)

Follow pulses on a space-time diagram: a pulse that returns while a tooth is in the way is lost.

Switch to 3D

Presets

Playback

Experiment: measure c

Wheel

Light path

Wheel speed

Observer and slow motion

Layers

Model check

The science

What you are looking at

The 2D view is a space-time diagram. Left: the wheel face with its N teeth and two magnified strips of the rim, one for the moment a pulse leaves and one for the moment it returns (the pattern has slid on by delta pitches). Right: distance from the wheel runs across, time runs up; every pulse is a pair of straight lines whose slope is c/n, and the bar at the left edge shows whether the wheel is open (green) or blocked (brown) at each time. The pictures of the bench, the wheel and the far mirror are Blender renders (Cycles); everything else is drawn in Canvas. A lamp shines through a gap of a rotating toothed wheel, runs a long way to a distant mirror and comes back to the same spot on the wheel. Light needs a round-trip time tau; during that time the wheel turns. If the wheel has moved by a whole tooth pitch the returning light finds a gap again and the eyepiece is bright; if it has moved by half a pitch a tooth is in the way and the eyepiece goes dark (an eclipse). Reading the wheel speed at the first eclipse gives the speed of light without any clock faster than a counter on the axle. This is the arrangement Hippolyte Fizeau used in 1849 over a path of about 8.6 km; the model here is a teaching version of it, not a reproduction of his exact apparatus.

Equations used

  • Round-trip time tau = 2 L n / c0 with the exact SI value c0 = 299,792,458 m/s and the refractive index n of the path (air about 1.000293, a water-filled tube 1.333).
  • Pattern coordinate: one tooth pitch is the unit, so the wheel advances delta = N f tau = 2 N f L n / c0 pitches during the round trip (N teeth, f revolutions per second).
  • Gate: the wheel is open for the fraction g of every pitch. The time-averaged transmitted fraction of the lamp light is the overlap of the gate with itself shifted by delta: A(delta) = max(0, g - d) + max(0, g - (1 - d)), d = delta mod 1. For g = 0.5 this is a triangle wave, maximum g at integer delta, zero at half-integers.
  • Eclipses (g = 0.5): delta = k + 1/2 gives f_k = (2k + 1) c0 / (4 N L n), so the speed of light follows from c = 4 N L f_1 / n and, for a measured eclipse k, c = 4 N L f / (2k + 1) (in air this gives c0/n). Bright orders sit at f = k c0 / (2 N L n).
  • Smear of the returning image (width w in pitches) is a box average of A: T = [F(delta + w/2) - F(delta - w/2)] / w with F the exact antiderivative of A (F(1) = g^2). Wheel-speed jitter (relative sigma) is a Gaussian average over delta with spread delta * sigma. The displayed brightness is T / g: 100 % means the wheel stopped with a gap on the beam.

What every control does

  • Teeth N and gap fraction g set the gate. More teeth make the first eclipse come at a lower wheel speed (f_1 is proportional to 1/N); a gap fraction below 0.5 gives a wide dark plateau instead of a sharp minimum, above 0.5 the eclipse never gets fully dark (the dip only reaches (2g - 1)/g of the maximum brightness).
  • Path L and index n set tau. A longer path or a slower medium (n above 1) lowers every eclipse speed by the factor 1/(L n).
  • Wheel speed is the sum of a coarse and a fine slider; the fine one exists because eclipses are narrow (about 0.1 rev/s wide at 12 rev/s). The Speed ramp button sweeps from 0 to 3.5 f_1 and back.
  • Return-image smear rounds the eclipse minima (poor focusing); jitter blurs them in proportion to delta; detector noise is the eye/photometer noise added to the scan; the tachometer calibration error is a single random scale error on the speed reading (one draw per scan), which dominates the scatter of c.
  • Slow motion changes how many seconds one round trip lasts in the animation; it does not change any number, because every distance, speed and angle in the page keeps the same delta.
  • Experiment: a scan sweeps the wheel speed from 0 to 3.5 times the first eclipse speed, finds each eclipse centre from the two half-brightness crossings (this works for plateaus and for partial eclipses) and turns it into c. Repeating 60 scans shows the scatter.

Numerical checks (shown live on the page)

  • Brute-force scan of the gate model versus the closed form f_1 = c0 / (4 N L n).
  • Photon Monte Carlo with explicit wheel angles at departure and return (including blur and jitter) versus the closed-form brightness.
  • The mean of T over one period of delta must equal g^2, and an exact interval-arithmetic overlap must equal the triangle formula.
  • 2D view only: a time-domain worldline count (explicit event times at the wheel, no use of the delta formula) versus the closed-form overlap, and the slope of the drawn rays versus c0.

Simplifications, honestly

  • The light is treated as a beam of rays at a point on the rim; real beams have a finite width, diffraction at the gaps and lenses with aberrations. Their effect is lumped into the single smear parameter.
  • The wheel is an ideal two-state gate; the finite thickness of the teeth, vignetting and the time the beam needs to cross a gap are ignored.
  • The medium is lossless and non-dispersive; group and phase velocity are not distinguished (in air the difference is below 0.1 percent). Eye adaptation, flicker fusion and the personal judgement of a minimum are not modelled except as noise.
  • The scene is not to scale: the distant mirror is drawn 22 m away; the path length used in the formulae is the L slider (8.633 km by default). The Fizeau-style default matches the often quoted wheel with 720 teeth and a path of 8.633 km; the exact wheel speed of the historical first eclipse (about 12.6 rev/s, giving c of about 3.13e8 m/s, roughly 4 percent high) is quoted from memory and should be spot-checked. The model's own first eclipse for these numbers is 12.05 rev/s because it uses the true c.
Eyepiece brightness versus wheel speed
Out and return gates over the tooth pattern
Eclipse speeds versus odd order 2k+1
Distribution of c from repeated scans
Made with 🟠 Blender · Three.js