Frontier math · see it to get it
The Kakeya needle: how much room does it take to turn a needle around?
The Duowenwen team · Updated
A needle is just a line. The area is everything it sweeps over
What each of the six steps shows
- A needle is a line; area is what it sweeps:The needle itself has no width. Slide it up by 1 and the thin strip it sweeps fills a unit square, area 1.000. Sliding never turns it, so the compass lights up only 0°.
- Half a turn around the middle sweeps a whole disc:Each end sweeps out a sector, and together they make a disc of diameter 1, area π/4 ≈ 0.785. Every direction from 0° to 180° is covered, and the ends swap places.
- Turn it a different way: a triangle is enough:Pál's route: pivot 60° about a vertex, slide along a side, then pivot about another vertex, three times in all. Sliding adds no area, so the total is 1/√3 ≈ 0.577.
- Cut and shift: every direction stays, the area shrinks:Cut the triangle into 2, 4, 8, 16, 32 pieces and slide neighbors into overlap, Perron-tree style. The area, computed live, falls to 0.385, 0.289, 0.231, 0.193, 0.165, and not one of those 60° directions goes missing.
- Three trees fill 180°, and the area can still shrink:Each of three trees covers 60° without overlapping the others, so the total is three times one tree's area. At 4096 pieces it is 0.248. The curve plots only computed values, and it falls more slowly the further it goes.
- In three and four dimensions, how thin can it get?:A bundle of thin tubes of length 1, each pointing a different way, with the same needle spiraling along those directions. How little volume is enough in three or four dimensions? Here we only ask the question; we don't claim an answer.
Take a needle of length 1 and turn it all the way around in the plane (a half turn, ends swapped). How much area must it sweep? Japanese mathematician Sōichi Kakeya posed the question in 1917.
Rotate it inside a disc and the area is π/4 ≈ 0.785. Use an equilateral triangle of height 1 and it drops to 1/√3 ≈ 0.577. Then in 1928 Besicovitch showed something startling: the area can be made as small as you like. The animation above shows this by cutting the triangle up and sliding the pieces into overlap (a Perron tree), and every number in it is computed from the real geometry.
Area can shrink to almost nothing, but another way of measuring size will not budge: a set in the plane that holds a segment in every direction must have dimension 2. The same result in three dimensions (dimension 3) was only proved in 2025, by Hong Wang and Joshua Zahl.
Sources
- Hong Wang, Joshua Zahl, Volume estimates for unions of convex sets, and the Kakeya set conjecture in three dimensions, arXiv:2502.17655 (2025)
- A. S. Besicovitch, On Kakeya's problem and a similar one, Mathematische Zeitschrift 27 (1928)
- Four dimensions: a manuscript credited to OpenAI in 2026 claims a proof, but it hasn't been peer reviewed. We note the claim here and don't treat it as established.
How Claude traced the NIST curve seeds in half an hour
Tesla's new Roadster jumps on gas jets: why would a car need thrusters?
AI-designed quantum circuits: how Claude rebuilt eight circuits without keeping its scratch work