BORSUK'S PUZZLE IN 4D: THE ANSWER IS FIVE OR SIX
Borsuk’s question is easy to state. Pick a bounded shape and measure its diameter: the widest gap between any two of its points. Then cut the shape into pieces, each strictly narrower than that diameter. How many pieces will be enough in every case?
In 1933, Borsuk asked whether, in n dimensions, n + 1 pieces always suffice. Fewer cannot work: in an equilateral triangle of side one, every pair of corners sits exactly one diameter apart, so each corner needs a piece of its own. A regular simplex — the n-dimensional version of the triangle — likewise demands n + 1 pieces for its corners. The smallest number that works for every shape goes by the name of Borsuk number, b_n.
Solved up to three dimensions
Borsuk proved that b₂ = 3: in the plane, three pieces always do. In space, b₃ = 4 was announced by Perkal in 1947 and proved by Eggleston in 1955. In 1993, though, Kahn and Kalai proved the guess wrong in high enough dimensions.
That leaves an awkward gap. Dimension four is now the smallest case still unanswered.
A bound that keeps falling
For four dimensions, the answer is at least five (the corners of the four-dimensional simplex). The question is how many pieces are enough. Lassak’s 1982 bound, b_n ≤ 2^(n−1) + 1, gave nine. Tolmachev and Voronov cut it to eight in May 2026. In October 2026, Arman, Bondarenko, Prymak and Radchenko covered every four-dimensional shape of diameter one with seven balls, each of diameter less than 0.9999: seven.
Logan Chalmers, at the University of Otago in New Zealand, now proves that six pieces always suffice. So the four-dimensional Borsuk number is five or six.
Pieces do not have to be balls
The key remark fits in one line: a piece narrower than one does not have to fit inside a ball narrower than one. The previous proof used only balls. This one lets the last piece take any shape.
The proof first scales the shape so that its diameter is exactly one, then encloses it in the tightest possible ball; shapes of the same diameter can need a bigger or a smaller ball, and the proof sorts them by the size of that ball.
- Large enclosing ball: five balls, each of radius 9999/20000 — so of diameter 0.9999 — are placed near the points where the shape touches its enclosing ball, pulled towards the centre. Whatever the five balls miss becomes one single sixth piece. The proof shows that this leftover can never contain two points exactly one apart, so it too is narrower than the whole. Where the previous construction needed two extra balls, one leftover piece is enough.
- Small enclosing ball: the whole ball is cut into six identical cones from its centre, built from two equilateral triples of directions lying in two perpendicular planes.
- Everything in between: here the computer takes over. The proof picks three points where the shape touches its enclosing ball; the possible shapes of that triangle are described by four numbers, and that range is split into small boxes. On 1,512 of them, six simple formulas with fractional coefficients decide which piece each point goes to; the other boxes are shown to describe configurations that cannot occur. In each piece, the width is bounded by the width of a polytope — a flat-sided solid — whose corners have exact fractional coordinates. Every piece comes out narrower than one by a guaranteed margin: its squared diameter is at most 1 − 86 × 10⁻¹⁰.
Checked by machine
A proof built on 1,512 boxes of exact fractions is hard to check by hand. The theorem has been formalised in Lean: the full proof was checked by that program, with its mathematics library; the data and the formalisation are public. Python programs built the partitions and computed the bounds. In an “AI declaration”, the author states that OpenAI’s Astra helped explore proofs, prepare the Lean formalisation and prepare the article, and takes full responsibility for it.
Five or six?
The paper does not settle the question, and it says what the answer would mean. If the true number is six, four is the lowest dimension in which Borsuk’s guess fails.
It ends on a sharper question. Could a shape of diameter one, lying on a sphere of radius at most 1/√2 in four dimensions, need six pieces? If such a shape exists, a known construction would push the failure up through every higher dimension. Dimensions one, two and three would then be exactly the only ones where n + 1 pieces always suffice.
