The Kakeya Needle

Turn a needle around

Put a needle of length 1 on a table. You want to spin it half a turn so it points the other way, using as little floor as possible.

The obvious move is to spin it around its middle. That sweeps a disk with area about 0.785.

You can do better by sliding while you turn. Trace a three-cornered curve called a deltoid and the needle turns inside it with exactly half that area, about 0.393. Switch the panel to see it.

For a while people suspected the deltoid was the best possible. It isn't. In 1928 Abram Besicovitch showed the needle can turn inside a region of any area you name, however small, if you let it take long detours.

Shrink the area toward zero

Now drop the rule that the needle has to move. Ask only for a set that contains a needle pointing in every direction. That is a Kakeya set.

The panel shows the idea behind Besicovitch's construction, in a version called a Perron tree. A triangle already holds a needle for every direction inside its angle. Cut it into thin slivers, then slide the slivers sideways so they pile onto each other.

Sliding doesn't rotate anything, so every direction is still there. But the overlaps eat area. More slivers, less area left. Carry this on forever and you get a Kakeya set with zero area.

So area is the wrong ruler. Something can contain a needle in every direction and still take up no room at all.

This construction is slow: the area falls roughly like 1/log of the number of slivers, which is why the panel stops near a third of the original.

A better ruler: dimension

If area can be zero, we need a finer measure of size. That measure is dimension, and box counting is the easiest way to see it.

Lay a grid of boxes with side δ over the shape and count the boxes it touches. Then halve δ. A line needs about 2 times as many boxes (2 = 2¹, so dimension 1). A filled square needs 4 times as many (4 = 2², dimension 2). A Sierpinski triangle needs 3 times as many, so its dimension is log 3 / log 2 ≈ 1.585. It has zero area but it's bigger than any line.

The Kakeya conjecture: a Kakeya set in n-dimensional space always has dimension n. It can have zero volume, but it can never be thin in this sense.

The 4D claim uses Hausdorff dimension, a stricter cousin of box counting. Proving full Hausdorff dimension is the harder of the two.

From needles to tubes

To actually prove things, mathematicians thicken each needle into a tube of width δ. One tube per direction, and the question becomes: how much can these tubes pile up on each other?

Try the arrangements. A bush runs every tube through one point. They overlap at the center, then fan out and stop overlapping. A hairbrush threads tubes through a single stem; Thomas Wolff used this picture in 1995 to prove dimension at least 2.5 in 3D. Slabs pack tubes into thin flat layers, the kind of structure Wang and Zahl had to rule out.

The conjecture in this language: no arrangement wins by much. As the tubes get thinner, the overlap factor can grow, but slower than any power of 1/δ.

Volumes are Monte Carlo estimates from 15,000 random points, recomputed each time you change the setup.

Seeing four dimensions

Nobody can picture 4D directly, so we cheat in two ways.

A shadow: rotate the 4D set and project it into 3D, the way a spinning cube casts a changing 2D shadow. Color stands in for the hidden fourth coordinate.

A slice: cut the 4D set with a 3D cross-section, like an MRI cutting a body. Each needle crosses the slice at a single point, so a slice of a 4D Kakeya set looks like dust. Move the slice and watch the dust shift.

Why 4D is hard: Larry Guth has pointed out that the central theorem in the 3D proof stops working in four dimensions. You can't just copy the argument up a dimension.