Thought Toys · Strategy & computation · Exhibit 46

The wheel that spins backward

A wheel turns one way, but on film it can crawl, freeze, or turn the other way. It isn't the wheel playing tricks — it's the camera looking too rarely. Sample any signal fewer than twice per cycle and it comes back as a slower, wrong one that you can't tell apart from the real thing.

The wheel — truly turning, and how the strobe sees it truestrobed

Same samples, two waves that fit them true fsamplesreconstructed

your turn — pull the sample rate below 2f and watch the wheel reverse

What you're seeing

The disc really rotates at f turns per second — the faint spoke. The strobe only glimpses it fs times a second, freezing the amber spoke at each glimpse and holding it until the next. String those glimpses together and the amber wheel seems to rotate at its own rate — the apparent frequency. When you photograph often enough (fs above 2f, the Nyquist rate — twice the highest frequency present) the amber wheel tracks the real one. Drop below it and the glimpses land a little short of a full turn each time, so the wheel appears to drift backward — the wagon-wheel effect from old Westerns and any wheel under mains lighting.

The lower plot is the same story as a wave: the amber dots are your samples, and the cyan curve is the slowest wave that threads through them. Above Nyquist it lies exactly on the true wave; below, it's a genuinely different, slower wave — and no measurement of the dots alone could ever tell the two apart. That indistinguishability is the whole reason sampling has a hard floor: CD audio samples at 44,100/s to clear the ~20,000 Hz ceiling of hearing, and cameras add shutter tricks so spokes and helicopter blades don't lie. Once you've undersampled, the true frequency isn't hiding in the data — it is simply gone.

The rule, exactly. A phasor at frequency f sampled at fs advances 2πf/fs radians per sample; what you see is that angle folded into (−π, π], so the apparent signed frequency is fapp = ffs · round(f/fs) and its magnitude is f folded into [0, fs/2]. Sampling is faithful exactly when fs > 2f (Nyquist–Shannon). Verified in node (improve/verify/46-aliasing.js): the phase-wrap and this closed form agree to 1e-15 off the half-integer tie set; above Nyquist the apparent frequency is exactly f; the aliased magnitude always lands in [0, fs/2]; and cosine samples of f, of f+fs, and of the fold are bit-for-bit identical. Negative controls: a best-fit reconstruction recovers the true frequency above Nyquist but returns the alias below it (aliasing is real, not a display trick); f < fs < 2f gives a strictly negative apparent frequency (backward); fs = f freezes it; and a very fast sampler never aliases — oversampling is always safe.

Also in Strategy & computation: The secretary problem →

All 23 in Strategy & computation
  1. 10The evolution of trust
  2. 23Sorting algorithms
  3. 24PageRank & the random surfer
  4. 25Huffman coding
  5. 26Dijkstra's shortest path
  6. 27Nash equilibria
  7. 33The learning-rate cliff
  8. 36A* pathfinding
  9. 37Braess's paradox
  10. 44Diffie–Hellman key exchange
  11. 45Preferential attachment
  12. 46Aliasing & the Nyquist limit — you are here
  13. 47The secretary problem
  14. 49Freeze too fast, stay stuck
  15. 51Cross one line, and its territory closes
  16. 56Catch one error, miss the next
  17. 57Why more processors stop helping
  18. 58Why a busy line explodes
  19. 59The set that's only sure when it says no
  20. 60The fit that memorizes instead of learns
  21. 61When the wire breaks, pick one
  22. 67Better at both, and still better off trading
  23. 69Everyone was consistent. The vote wasn't.

← the cabinet · Thought Toys — a cabinet of explorable explanations. Exhibit 46.