Why ten smaller pieces may not cover a set in nine dimensions

← All digests

Family 156 · Combinatorics
Importance 4/5 · Provisional editorial rating
Sources consulted 2026-10-11

1. The question

How many smaller pieces are needed to cover a geometric object? Here “smaller” means smaller diameter: the greatest distance between any two points. An interval can be split into two shorter intervals. In the plane, three smaller-diameter pieces suffice for every bounded set. Borsuk asked whether the corresponding bound in dimension $d$ is always $d+1$.

That assertion is already known to fail in high dimensions. Family 156 claims a dramatic reduction of the dimension: a compact set in $\mathbb R^9$ that cannot be covered by ten subsets of strictly smaller diameter. The example is described using four-dimensional lines and elementary matrix algebra, even though the obstruction to a cover uses topology.

The source’s witness is an entire compact space of projection matrices. It does not present a small finite cloud of points, and it does not claim that nine is the smallest possible dimension of failure.

2. Background and essential definitions

For a bounded set $Y$ in Euclidean space, its diameter is

$$\operatorname{diam}(Y)=\sup\{\|x-y\|:x,y\in Y\}.$$

Let $b(Y)$ be the least number of subsets with diameter strictly below $\operatorname{diam}(Y)$ whose union contains $Y$. Covering sets may overlap. This gives the same minimum as partitioning: assign each point in an overlap to just one piece, which cannot increase any diameter.

The word “strictly” matters. A piece can be almost as wide as the original set, but it cannot retain its full diameter. The problem imposes no requirement that the pieces be convex, connected, measurable or attractive shapes.

Lines represented by matrices

Take a unit column vector $u\in\mathbb R^4$. The matrix

$$P_u=uu^{\mathsf T}$$

is the orthogonal projector onto its line: it sends a vector $v$ to $u(u^{\mathsf T}v)$. Replacing $u$ by $-u$ gives the same matrix, as it should, because the underlying line is unchanged. The space of lines through the origin is real projective three-space, written $\mathbb{RP}^3$.

A symmetric $4\times4$ matrix has four independent diagonal entries and six entries above the diagonal, so these matrices form a ten-dimensional real vector space. Every $P_u$ has trace one. The trace-one condition removes one dimension, putting all these projectors in a nine-dimensional affine space.

Distances are measured by the Frobenius norm, the square root of the sum of squared matrix entries. For unit vectors $u,v$,

$$\|P_u-P_v\|_F^2=2-2(u\cdot v)^2.$$

Consequently the projector set has diameter $\sqrt2$. Two of its points attain this distance exactly when the corresponding lines are perpendicular. A smaller-diameter piece must therefore keep its lines uniformly away from perpendicularity; merely excluding an attained perpendicular pair would not by itself control a supremum for an arbitrary nonclosed piece.

3. Development history and previous partial results

Borsuk posed the problem in 1933. Its appeal came from the simple dimensional prediction and its compatibility with low-dimensional geometry. The general conjecture survived until Kahn and Kalai’s 1993 counterexample. Their argument used a theorem on intersections of finite sets to produce point configurations needing too many smaller-diameter parts.

After that, a central question became how far down in dimension counterexamples could be brought. Different constructions exploited spherical codes and exceptional lattice configurations. Another important approach used strongly regular graphs, whose very uniform adjacency patterns produce Euclidean configurations with only two pairwise distances.

Bondarenko obtained a two-distance example in dimension 65. Jenrich and Brouwer then gave one in dimension 64. The new manuscript also credits a 2026 author manuscript by Grinsztajn giving a finite example in dimension 63. That is the recent comparison stated by M1; the earlier published 64-dimensional paper supplies a particularly clear established milestone.

The projection-matrix representation used here is not itself new. Conway, Hardin and Sloane used it for packing subspaces, and related rank-one configurations appear in accounts of the Kahn–Kalai construction. The claimed new contribution is a sufficiently strong obstruction to covering the specific projector space associated with $\mathbb R^4$.

Topology provides another antecedent. Results on how many vertices are needed to triangulate projective spaces suggest why ten labels might be insufficient. But a complex recording overlaps of a cover is not automatically a triangulation. The manuscript’s main task is to derive the needed combinatorial structure from the geometric cover, rather than simply importing a triangulation bound.

4. What the new manuscripts claim

M1’s Theorem 1.1 concerns the compact set

$$X=\{uu^{\mathsf T}:u\in\mathbb R^4,\ \|u\|=1\}.$$

It asserts that no ten subsets of diameter strictly smaller than $\sqrt2$ cover $X$. In the notation above,

$$b(X)>10.$$

Since the surrounding trace-one affine space is Euclidean nine-space, this contradicts Borsuk’s proposed bound of ten pieces in dimension nine. The metric is precisely the Frobenius metric; changing its normalization by a common positive factor would not affect the covering question.

The manuscript also states a construction of compact counterexamples in every dimension $d\ge9$. This extension involves its further construction, not just viewing the same set inside a larger space: the required number of pieces rises with the dimension.

The headline theorem does not give the exact value of $b(X)$. It also leaves open which smaller dimensions support counterexamples. Its quantified assertion is that every ten-piece cover fails, including covers by highly irregular subsets.

5. Why the advance matters

Importance: 4/5. A counterexample in dimension nine would sharply change the low-dimensional landscape of a classical problem. It brings the obstruction much closer to dimensions where positive geometric intuition has remained influential.

The proposed mechanism is also valuable beyond the number nine. It converts a metric covering question into an incompatibility among finite collections of labels, with topology supplying the constraints. Linear algebra makes the witness easy to describe, while the topological structure prevents the desired decomposition.

The rating reflects a major advance within discrete geometry. Borsuk’s general assertion was already disproved, so the central new achievement is the very low dimension and the method supporting it. The result should not be described as the first refutation of the original conjecture.

6. The main idea

A cover becomes a map

Suppose ten smaller-diameter pieces cover $X$. Because their diameters are strictly smaller, the manuscript first enlarges them to suitable open sets without losing the separation from perpendicular pairs. It then constructs continuous, indeed smooth, weights

$$f_i(x)\ge0,\qquad \sum_{i=1}^{10}f_i(x)=1,$$

for lines $x\in\mathbb{RP}^3$. Each weight is supported within the corresponding open set. Such weights are called a partition of unity: instead of assigning a point abruptly to one piece, they distribute its membership continuously.

If two lines are perpendicular, they cannot have the same positive coordinate. Thus geometry has become a rule about which subsets of the ten labels can occur together.

Record every possible set of simultaneously positive coordinates as a face of a finite simplicial complex. A simplicial complex is a collection of finite sets closed under taking subsets. One can think of its one-, two-, three- and four-element faces as vertices, edges, triangles and tetrahedra, without assuming they form an ordinary solid in space.

Matrix topology constrains the faces

The source extends the weights from rank-one projectors to positive semidefinite matrices, then compares the positive and negative parts of arbitrary symmetric matrices. After normalization, this produces an odd map between spheres. “Odd” means that changing the input sign changes the output sign.

The relevant topological tool is degree modulo two, which records whether a suitably chosen output has an odd or even number of preimages. It is stable under continuous deformations and can detect an obstruction invisible to individual coordinates.

At the ten-label threshold, the manuscript uses this parity information to force maximal positive supports to have four labels. It also forces the resulting tetrahedral faces to fit together as a cycle: their triangular boundaries cancel in pairs when counted modulo two. Deriving these properties is essential; they cannot be assumed from the definition of a cover.

A finite incompatibility finishes the argument

A related three-dimensional analysis produces rigid triangle systems on six labels. Each tetrahedron among the ten labels leaves a six-label complement, where one of these systems must occur. The systems associated with neighboring tetrahedra must agree where their requirements overlap.

The final combinatorial argument shows that all these compatibility conditions cannot hold simultaneously. The hypothetical cover would therefore create a finite structure that cannot exist. Returning through the map and the projector representation excludes the ten-piece cover.

This explanation separates the roles of the ingredients: the distance formula identifies perpendicular lines, continuity enables topology, and the final finite argument converts topological restrictions into a contradiction.

7. Scope, limitations, and open questions

The witness is a compact infinite set, and the source does not supply an optimized finite replacement or an exact Borsuk number for this projector space. It is important to preserve that form of the claim rather than silently turn it into an explicit finite configuration.

The result would establish failure in dimension nine, not minimality of nine. Lower dimensions require their own positive theorems or counterexamples. Nor does the argument classify all sets whose covering number exceeds the dimensional prediction.

The role of arbitrary covering subsets is substantive: restricting to convex or measurable pieces would be a different, weaker obstruction. M1 starts with arbitrary smaller-diameter covers and explains the passage to open sets before applying its smooth machinery.

Further natural questions concern the exact covering number of $X$, finite configurations capturing comparable obstructions, and whether other spaces of subspaces yield counterexamples in still smaller dimensions. These are directions suggested by the construction, not consequences already supplied by its theorem.

8. Sources and further reading

M1’s Theorem 1.1 states the nine-dimensional result. Lemma 2.1 gives the projector distance formula; Sections 2–4 turn a cover into a constrained complex, and Sections 5–7 give the finite obstruction and dimensional extension. Its introduction provides the more detailed historical sequence.

The primary papers linked in Section 3 explain the original Kahn–Kalai breakthrough and the later 64-dimensional example. Optional background on real projective space and partitions of unity can help readers pursue the topological part beyond this digest.

Source manuscripts

• M1. OpenAI, A nine-dimensional counterexample to Borsuk’s covering assertion (September 23, 2026).

Sources consulted on 11 October 2026 at repository revision fd4aeeb. Download the manuscript citations.