Rank-one projectors and Borsuk’s covering problem

Source and verification

This article explains family 156 of the October 6, 2026 OpenAI collection. Research conclusions are attributed to its manuscripts. The review covers statements, introductions, and selected proof mechanisms; it is not an independent verification of the complete proofs.

🏷️ Diameter and rank-one projectors

For a bounded set $X$ in Euclidean space, its diameter is ${\sup }_{x,y\in X}\Vert x-y\Vert$. Borsuk’s covering assertion asks whether every bounded subset of ${\mathbb{R}}^d$ can be covered by $d+1$ subsets of strictly smaller diameter. The covering pieces need not be open, measurable, or disjoint.

The manuscript studies a particularly concrete compact set. For a unit vector $u\in {\mathbb{R}}^4$, let $P_u=u{u}^{\mathsf{T}}$ be the orthogonal projector onto its line, and put

$$X=\{P_u:\Vert u\Vert =1\}.$$

The matrices $P_u$ are symmetric and have trace one. Symmetric $4\times 4$ matrices form a ten-dimensional Euclidean space, so the trace-one affine hyperplane is nine-dimensional. Its Euclidean metric is the Frobenius norm, $\Vert M{\Vert }_F^2=\mathrm{tr}(M^{\mathsf{T}}M)$.

The identity

$$\begin{array}{rl}\Vert P_u-P_v{\Vert }_F^2& =\mathrm{tr}(P_u)+\mathrm{tr}(P_v)\\ & -2\mathrm{tr}(P_u{P}_v)\\ & =2-2(u\cdot v{)}^2\end{array}$$

shows that $X$ has diameter $\sqrt{2}$, attained exactly by orthogonal lines. Also $P_u=P_{-u}$, so $X$ is a concrete model of real projective three-space, the space of unoriented lines through the origin in ${\mathbb{R}}^4$.

Manuscript claim

The compact set $X$ cannot be covered by ten subsets of diameter strictly less than $\sqrt{2}$. Thus Borsuk’s assertion fails in dimension nine. (OpenAI, 2026)

The paper also constructs counterexamples in all higher dimensions. This does not identify the least dimension of failure. Nor does simply embedding $X$ into a higher-dimensional space prove the higher-dimensional assertion: the allowed number of pieces increases with the dimension.

🏷️ From an arbitrary cover to continuous functions

Suppose a ten-piece cover existed. Since there are finitely many pieces and each has diameter strictly below $\sqrt{2}$, there is a common positive margin below that threshold. Taking sufficiently small open neighborhoods inside the compact space $X$ preserves this margin. A smooth partition of unity subordinate to the resulting open cover gives nonnegative functions $f_1,\dots ,f_{10}$ satisfying

$$\sum _{i=1}^{10}{f}_i(x)=1.$$

If $x$ and $y$ represent orthogonal lines, they cannot both belong to the positive support of the same $f_i$. Thus a metric covering problem produces continuous functions with a strong exclusion rule.

This reduction explains why no regularity assumption is needed on the original pieces. Smoothness is introduced after using the strict diameter gap. It also identifies the central topological object: the pattern of subsets of labels that can be simultaneously positive at a point.

That pattern is a support complex, not automatically a triangulation of projective space. A lower bound on the number of vertices needed to triangulate projective space cannot therefore be applied directly. The manuscript develops additional structure connecting the supports to projective geometry.

🏷️ Spectral decomposition and an odd map

For a symmetric matrix $M$, its positive and negative parts are the positive semidefinite matrices $M_{+},M_{-}$ with orthogonal ranges and $M=M_{+}-M_{-}$. The partition functions can be extended to positive semidefinite matrices through integration over line directions. Orthogonality of the two ranges then keeps the positive labels associated with $M_{+}$ separate from those associated with $M_{-}$.

Taking the difference of their extended label vectors produces an odd map: replacing $M$ by $-M$ changes its sign. Its target ${\ell }_1$ norm agrees with the appropriate trace norm of $M$. After normalization this becomes a map between spheres, where parity of degree constrains the possible support patterns.

At the critical number $k(k+1)/2$ of labels for lines in ${\mathbb{R}}^k$, the argument forces maximal supports of size $k$ and organizes their facets into a cycle. The parity calculation uses level sets lying in an affine chart and compares preimages in a sphere bundle. Its purpose is to turn the analytic support data into a rigid combinatorial structure; it is not merely an appeal to the dimension of the ambient matrix space.

🏷️ The four-dimensional compatibility obstruction

For $k=4$, the forced maximal supports behave like tetrahedral faces. The labels complementary to such a face inherit a six-label configuration governed by the corresponding three-dimensional argument. These complementary configurations satisfy stringent triangle relations.

The final obstruction propagates those relations along shared edges and cycles. Local choices that individually satisfy the support rules cannot be assembled into the global ten-label configuration required by the cover. This is the manuscript’s new topological-combinatorial ingredient.

The elementary projector calculation explains the dimension and the diameter exactly. The non-covering conclusion additionally depends on the full support and compatibility argument. Computing a few pairwise distances, or checking a finite sample of projectors, would not verify that conclusion for the compact set.

🏷️ Key steps and proof outlines

The strict diameter margin permits smooth support data

Given finitely many covering pieces of diameter strictly below $\sqrt{2}$, first replace them by sufficiently small open neighborhoods in the compact projector space. The minimum of the finitely many positive diameter margins remains positive. A subordinate smooth partition of unity $f=(f_1,\dots ,f_m)$ then satisfies $f_i\ge 0$ and ${\sum }_i{f}_i=1$, and orthogonal lines have disjoint positive-label sets. This is the precise admissibility property used in the proof; it does not require the original pieces to be measurable.

For a vector space $V$ of dimension $k$, define the degree-two homogeneous extension $H(v)=\Vert v{\Vert }^{2}f([v])$ for $v\ne 0$, and $H(0)=0$. Continuity at zero follows from $\Vert H(v){\Vert }_1=\Vert v{\Vert }^2$. For a positive semidefinite matrix $Q$, put

$$E_V(Q)=k{\int }_{S(V)}H(Q^{1/2}u)d{\mu }_V(u),$$

where ${\mu }_V$ is uniform sphere probability. The matrix square root is continuous even at singular matrices, so this extension remains continuous on the closed positive cone. (OpenAI, 2026)

Trace normalization and support are exact

Since $\int u{u}^{\mathsf{T}}d{\mu }_V=I/k$,

$$\sum _i{E}_V(Q{)}_i=k\int \Vert Q^{1/2}u{\Vert }^{2}d{\mu }_V(u)=\mathrm{tr}Q.$$

Every image vector lies in $\mathrm{ran}Q$. Conversely, every line in that range is attained by some nonzero $Q^{1/2}u$. If a label is positive at that line, continuity makes the integrand positive on an open sphere set of positive measure. Thus the positive coordinates of $E_V(Q)$ are exactly the union of labels used on lines in the range. In particular, $E_V(P_x)=f(x)$ for a rank-one projector.

The normalization also survives restriction to a smaller subspace $W$. The direction of the projection of a uniform vector onto $W$, weighted by the squared projection length, is rotationally invariant and has total mass $\dim W/k$. Consequently the ambient factor $k$ becomes exactly $\dim W$, and the same extension computed intrinsically in $W$ gives the same result. This subspace consistency is needed when complementary spectral ranges are analyzed at different dimensions.

Orthogonal spectral ranges prohibit cancellation

For symmetric $A$, let $A_{+},A_{-}$ be its positive and negative parts and set $F(A)=E_V(A_{+})-E_V(A_{-})$. Their ranges are orthogonal, so admissibility makes their positive-label supports disjoint. There is therefore no coordinate cancellation, and

$$\Vert F(A){\Vert }_1=\mathrm{tr}|A|,\sum _{i}F(A{)}_i=\mathrm{tr}A.$$

Replacing $A$ by $-A$ swaps the two parts, making $F$ odd. On the trace-norm unit sphere, it maps into the ${\ell }_1$ unit sphere. Radial normalization identifies these topological spheres with ordinary Euclidean spheres; no claim that the norm spheres are globally smooth is needed.

A useful nondegeneracy test follows from the support identity. If $A$ has a kernel line $x$, any label positive at $x$ is excluded from both spectral ranges and hence is a zero coordinate of $F(A)$. Thus every preimage of a vector with all coordinates nonzero is nonsingular. Near such a preimage the positive and negative spectral subspaces vary smoothly, so local differential and degree calculations are justified. At the critical label count $m=k(k+1)/2$, the source applies odd-map degree parity on spheres of the same dimension.

The combinatorial obstruction is an additional theorem

The degree and support arguments force maximal supports of size $k$ and their facet-cycle relations. For $k=4$ and ten labels, complementary label sets inherit the six-label structure from the three-dimensional case. The final shared-edge and cycle compatibility argument rules out their simultaneous realization. This last step uses more information than the mere dimension of projective space, and the support complex need not be its triangulation.

Once that incompatibility is established, the original ten-piece smaller-diameter cover is impossible by the smooth-support reduction. The projector identity fixes the ambient dimension nine and the exact forbidden distance $\sqrt{2}$; the extension, degree calculation, and finite-label compatibility are what turn that metric observation into a covering obstruction. Higher-dimensional examples require the separate construction accounting for their larger allowed number of pieces, not just an embedding of this same set.

See Also

• Graph embeddings gives another example in which an embedding translates geometric restrictions into combinatorial structure.
• Euclidean Ramsey configurations asks a different covering question: which fixed distance patterns every finite coloring must contain.

🏷️ Manuscripts and verification scope

• A nine-dimensional counterexample to Borsuk’s covering assertion — pinned manuscript and companion source.

The family scope page describes selected formalization coverage. The selected statement is the nine-dimensional projector counterexample; it does not assert a least dimension of failure. No Lean build, axiom audit, or manuscript-to-formal-statement equivalence check was performed for these notes.

OpenAI collection provides prerequisites and the reading sequence; coverage register records completed and pending families.

🐻  OpenAI 2026. A nine-dimensional counterexample to Borsuk’s covering assertion.