Higher Dimensions Are Weird

Volume collapses, corners grow spikes, and intuition stops working entirely

Intuition Has a Domain of Validity

You have spent your whole life in three dimensions, and your sense of how shapes behave was trained there. That training is worth something — the intuition is not wrong, it is just local. Ask it about dimension 40 and it will answer confidently, and it will be wrong about almost everything: how big a ball is, where a cube keeps its volume, how far apart randomly scattered points are. Every one of those failures has a direct consequence for sphere packing, which is why the problem gets harder rather than easier as the dimension climbs.

The three-dimensional case took 387 years to settle. Dimensions 8 and 24 fell in 2016 for reasons that have nothing to do with being close to three. Everything else remains open. The demos below are the reason.

The Ball That Runs Out of Room

The volume of the unit ball is πn/2 ÷ Γ(n/2 + 1). In one dimension it is 2, in two it is π ≈ 3.14, in three it is 4π/3 ≈ 4.19. It keeps growing — until dimension 5, where it reaches 8π²/15 ≈ 5.264 and stops. From there it falls, and it never recovers. By dimension 20 the unit ball has volume 0.0258; by dimension 100 it is about 10⁻⁴⁰. The unit ball in a hundred dimensions has more room in every direction than the ball in three, and less volume than a rounding error.

Compare it to the cube it sits inside, [−1, 1]n, and the collapse gets worse. That cube has volume 2n, which grows forever. In three dimensions the ball fills 52% of it. In ten, 0.25%. In fifty, one part in 1028. The volume has not gone anywhere — it has moved into the corners, of which there are 2n, each one a long spike reaching out to distance √n while the ball only ever reaches 1.

Volume of the unit ball, dimension 1 to 100
110⁻¹⁰10⁻²⁰10⁻³⁰10⁻⁴⁰10⁻⁵⁰10⁻⁶⁰10⁻⁷⁰120406080100dimension npeak: n = 5, V = 5.264unit ball volumeball ÷ surrounding cube
Unit ball volume
5.264
Share of the cube
16.4%
Cube ÷ ball
6.08×
Share of the n = 5 peak
1

The volume of the unit ball is πn/2 ÷ Γ(n/2 + 1). The numerator grows geometrically, the denominator factorially, and the factorial wins from dimension 6 onward. The cyan curve divides that volume by 2n, the volume of the cube the ball is inscribed in, and falls faster still: whatever a high-dimensional cube contains, almost none of it is anywhere near the middle.

Try it: Start on the linear axis and drag the slider — both curves flatline into the floor by dimension 15 and the picture stops saying anything. Switch to the log axis and the real behaviour appears: two straight-ish lines heading down forever, the cyan one twice as steep. The peak at n = 5 is the only feature either curve ever has.

The Sphere That Escapes Its Box

Here is the construction that ends the argument. Take a cube of side 4 and place a sphere of radius 1 in each corner. There are 2n of them, each tangent to its neighbours and to the walls — a perfectly ordinary arrangement that you can picture without effort in two dimensions and in three. Now drop one more sphere at the centre and inflate it until it touches all the others.

A corner sphere is centred at (±1, ±1, …, ±1), which is at distance √n from the middle. Subtract its radius and the central sphere has radius √n − 1. In two dimensions that is 0.414, a speck between four circles. In three it is 0.732. And it keeps growing, because √n does, while the cube stays exactly four units wide forever. At n = 9 the central sphere has radius exactly 2 — the half-width of the cube — and past that it reaches straight through the walls. The sphere wedged in the gap between the corner spheres is bigger than the box that holds all of them.

Dimension 2 — inner radius √2 − 1 = 0.414Inside the box
cube wall at 2centrecorner centre at √n = 1.414√n − 1 = 0.414
Inner radius
0.414
Cube half-width
2.000
Corner spheres
2² = 4
Inner ball ÷ cube
0.0337
29 — the crossover20
radius ÷ half-width = 0.207

A corner centre sits at distance √n from the middle of the cube, so the sphere squeezed between the corner spheres has radius √n − 1 — unbounded, while the box holding it never changes size. The square above is a slice through two of the n axes: it draws the cube and the radii truthfully, but it cannot show the corner spheres at their real distance, which is exactly the information two dimensions cannot hold. The radial diagram is the honest one. And note the last figure: even in dimension 20, where the sphere reaches far outside the cube, its volume is a vanishing fraction of the cube it escapes from.

Try it: Climb from 2 up to 20 and watch the moment of contact at n = 9, then keep going. Check the last statistic while you do: even in dimension 20, where the central sphere is sticking far outside the cube, its volume is a tiny fraction of the cube’s. Reaching everywhere and filling nothing is what high-dimensional shapes do.

What We Actually Know, Dimension by Dimension

Given all that, it is worth being blunt about the state of knowledge. The densest sphere packing is known — proven, not conjectured — in exactly five dimensions: 1, 2, 3, 8 and 24. For a handful of others there is a champion arrangement that has stood unbeaten for decades and is widely believed to be optimal, with no proof. Beyond that the record books thin out to constructions with nothing to recommend them except that nobody has done better.

Kissing numbers tell the same story in miniature. Dimension 4 was settled in 2003. Dimension 5 is still known only to lie somewhere between 40 and 44 — a gap of four, in a question you can state to a child.

Dimension8
Solved
proven optimalbest known packing tabulatedno record on file
Geometry of the unit ball
Volume V84.059
Share of its cube1.59%

Past the peak at dimension 5 — the ball has shed a factor of 1.3 since then.

Densest packing
Density25.3670%
ArrangementE₈
Closed formπ⁴/384

Proven optimal — Viazovska 2016. No packing in this dimension can do better.

Kissing number
Exact value240
FromE₈ root system

The 240 roots of E₈, each touching the centre

What is actually settled

Dimension 8 is one of the five where the packing problem is finished: 1, 2, 3, 8 and 24. Everywhere else the answer is a construction plus a hope.

Five dimensions solved out of infinitely many, and four of the five are 3 or below or else were cracked in a single year, 2016.

Try it: Sweep the dial across all 48 dimensions and watch the colour strip. Green means solved and there are five green cells in the whole row. Compare dimension 8 with dimension 9: one is finished mathematics, the next along has a kissing number pinned only to somewhere between 306 and 363.

Everything Is the Same Distance Away

The last failure of intuition is the one that hurts most. Scatter points at random in a high-dimensional cube and ask which are close together. The answer is that the question dissolves: the distance between two random points is an average over n independent coordinates, and averages concentrate. The typical distance grows like √(n/6) while the spread around it stays near a fixed 0.24, so the histogram of pairwise distances slides to the right and narrows into a blade.

This is why packing in high dimensions is so hard to reason about. There is no local structure to exploit — no cluster of close points to tighten, no obvious empty region to fill, because everything is equidistant from everything else. It is also why the central question is still open in a deeper sense than “we have not found the answer yet.” Nobody knows whether the densest packings in high dimensions are beautifully ordered lattices like E₈, or amorphous disordered arrangements with no symmetry at all. In a world where every distance is the same, order stops being obviously better.

2,415 pairwise distances among 70 random pointsn = 2
1/√6 ≈ 0.4080.00.20.40.60.81.0distance ÷ √n (0 = same point, 1 = opposite corners of the cube)distances in dimension 2
Typical distance
0.531
Closest / farthest
0.00 / 1.23
Range ÷ typical
230.6%
Farthest vs closest
+33742.3%

Each coordinate contributes an independent bite to the squared distance, so the total is an average of n independent numbers — and averages concentrate. The mean distance grows like √(n/6) while the standard deviation around it stays near 0.24 no matter how high you climb, which is why the histogram slides right and thins to a blade. In the plane, the full range of distances is more than twice the typical distance itself, and the farthest pair of points is hundreds of times farther apart than the closest. By dimension 200 the range has fallen under a third of the typical distance and the farthest pair is barely a third farther than the closest: these points have no near neighbours and no far ones, only neighbours.

Try it: Press Raise the dimension and watch the broad hump of dimension 2 — kept as a dashed outline for comparison — collapse into a spike at 1/√6. Track Range ÷ typical: it sits above 200% in the plane and falls below 30% by dimension 200. The farthest pair of points, hundreds of times farther away than the closest when n = 2, ends up barely a third farther.

Key Takeaways

  • Volume peaks and dies — The unit ball is largest in dimension 5 at 8π²/15 ≈ 5.264, then collapses: about 10⁻⁴⁰ by dimension 100, because πn/2 grows geometrically and Γ(n/2 + 1) grows factorially
  • All the volume is in the corners — A ball fills 52% of its cube in three dimensions and one part in 1028 by dimension 50; the 2n corners reach out to distance √n while the ball never gets past 1
  • The sphere escapes the box — Wedge a sphere between the 2n corner spheres of a cube of side 4 and its radius is √n − 1, which passes the cube’s half-width of 2 at dimension 9 and grows without bound afterwards
  • Five dimensions are finished — Optimal packings are proven only in dimensions 1, 2, 3, 8 and 24; everywhere else there is a best known construction and an unclosed gap between what has been built and what has been ruled out
  • Distances concentrate — Random points in dimension n are all roughly √(n/6) apart with spread near 0.24, so nothing has near neighbours; with no local structure to exploit, it is still unknown whether the best high-dimensional packings are ordered or disordered