Equipartition of a measure by \((Z_{p})^{k}\)-invariant fans (Q848668)

On the plane a point with rays sprouting from it which simultaneously equipartition a measure is called a \textit{fan}. In this article a measure, on a Euclidean space, is assumed absolutely continuous and the fan is constructed using a finite group to act by rotating a closed convex cone, assuming that there is at least one ray common to the cones. The author actually assumes that there is a unique common ray, though the reviewer's opinion is that his proof should be valid even if there are other common rays. To prove the existence of an equipartition the author finds a suitable rotation of the common ray and a translation, possibly zero, of the apex to a suitable position. The author relies on co-incidence theorems in the following sense. The Borsuk-Ulam theorem, conjectured by Ulam and proved by \textit{K. Borsuk} [Fundam. Math. 20, 177--190 (1933; Zbl 0006.42403)] asserts that, for the unit sphere \(S^{n}\), corresponding to every continuous mapping \(f : S^{n}\to {\mathbb{R}}^{n}\) there is a point \(x\) such that \(f (x) = f(-x)\), i.e., some pair of antipodal points has the same image. \textit{B. Knaster}'s conjecture [``Problem 4'', Colloq. Math. 30, 30--31 (1947)] was that for any continuous map \(f : S^{n} \to {\mathbb{R}}^{m}\) and certain discrete configurations in a general position within a domain \(D \subset S^{n}\), there exists a rotation \(\rho\) in \(S^{n}\) such that \(f(\rho(D))\) is a single point; the conjecture is not true for arbitrary \(m\) and \(n\). \textit{S. Kakutani} [Ann. Math. (2) 43, 739--741 (1942; Zbl 0061.38309)] had developed a similar `coincidence theorem' but using rectangular cones, originally for the plane, then generalised by others. \textit{A. Volovikov} [Russ. Acad. Sci., Sb., Math. 76, No.~2, 361--387 (1993); translation from Mat. Sb. 183, No.~7, 115--144 (1992; Zbl 0791.57023)] proved that Knaster's conjecture holds in Euclidean space using a rotation representating a discrete cyclic group \(({\mathbb{Z}}_{p})^{k}\) for \(p\) an odd prime, \(k\) the number of factors. The author furthers Volovikov's model to obtain existence of an equipartition of a given measure on \({\mathbb{R}}^{d}\) where \(d = p^{k}\). Given an absolutely continuous probability measure \(\mu\) and a solid closed cone \(C \subset {\mathbb{R}}^{d}\), the author proves the existence of a rigid motion \(\rho\) in \({\mathbb{R}}^{d}\) such that the family of cones \(\pm g(C)\), \(g \in G\) divides \(\mathbb{R}^{d}\) into sectors of equal measure \(\mu(\rho(g(C))) = \mu(\rho(-g(C))) = {1 \over {2d}}\). He follows Volovikov to construct a related fibre bundle and calculating Euler cohomology-classes to be non-zero so that there are obstructions to the existence of a vanishing section. This is done by counting zeros or self-intersections mod \(p\) to get the (Brouwer) degree for continuous mappings between compact manifolds of the same dimension. The manifolds being finite-dimensional there are no further obstructions other to a vanishing section. However, to prove equipartion the author needs to find a vanishing section which he does by extending Volovikov's results further. The author takes his `configuration space' \(E\) to be a group of orientation-preserving rigid motions \(\rho\) of \({\mathbb{R}}^{d}\) in which he hopes to find candidates for a solution. His \(\rho\) is \(\rho(\gamma,\tau), \gamma \in SO(d)\) and \(\tau\) a finite translation in \({\mathbb{R}}^{d}\). It can be shown from Volovikov's theorem that no solution can be in \(G \times {\mathbb{R}}^{d}\). He constructs \(\phi\) mapping \(E\) into to a \(2d\)-dimensional `target space' \(V \subset {\mathbb{R}}^{2d}\) which is the direct sum of two copies of the group algebra of \(G\). He gives \(V\) co-ordinates \((a_g, b_g)\) where \(a_{g}= \mu(\rho(g(C))), b_{g} = \mu(\rho(-g(C)))\), \(g \in G\), are the measures of the cones. The group \(G\) acts on \(V\) by left multiplication and on \(V\) and on \(E\) by multiplication on the right by the rotation \(g^{-1}\), so \(\phi\) is equivariant. The mapping \(\phi\) is shown to be continuous in the Hausdorff metric. Indeed the intersection, say \(K\), of the cone with a bounded ball \(B \subset {\mathbb{R}}^{d}\) is convex and compact. So \(\mu(\partial K) = 0\). One may then allow the radius of a ball \(B\) to tend to infinity, first considering small \(\epsilon\)-neighbourhoods of \(B\) separating the interior and exterior of the ball, then letting \(\epsilon \to 0+\). As the radius of \(B\) increases the measures \(\mu(\rho(g(C)) \cap B)\) will tend uniformly to \(\mu(\rho(g(C)))\). The author defines new coordinates \(s_{g} = a_{g} + b_{g}, t_{g} = a_{g} - b_{g}\). He denotes the \(d\)-dimensional hull of the \(t_{g}\) by \(U\) and the (\(d-1\))-dimensional hull of the \(s_{g}\) by \(W\). He is in fact using the regular representation of \(G\) on the left \(G\) and a reduced regular representation on the right. His \(W\) is isomorphic to the group ring modulo \(\sum_{g\in G} g\). He then quotients out the line, denoted \(L\), extending the ray \(s_g = 0\) and \(t_{g}\) constant for \(g \in G\) and denotes the restriction of \(\phi\) to \(U \oplus V\) by \(f\). As an illustration for the plane, though this case does not fulfill the author's conditions for \(d\), with \(G\) the cyclic group having generators \(e\) and \(\sigma\), \(U\) will be isomorphic to the group algebra and \(W\) will be a line in \({\mathbb{R}}^{4}\) perpendicular to \(U\). The author considers the vector bundle \(E \times (U \oplus W)\to E\), the fibres being the copies of space \(U \oplus W\). By counting iself-intersections mod \(p\), the sub-bundle over to \(U\) has nonvanishing Euler class in the equivariant cohomology of the relative cohomology pair \((E, SO(d)\times \partial B)\) where \(B\) is a large enough ball centred at the origin. He refers to Volovik for a proof that the Euler class of the sub-bundle over \(W\) is non-zero. The fact that \(W\) is \(d-1\)-dimensional and has non-zero Euler class is the crux of the proof of the equipartition. There is one dimension left free so now one has a decomposition of \({\mathbb{R}}^{2d}\) as \( U \oplus W \oplus (1,1,\dots, 1)\), the last summand being the diagonal line in \({\mathbb{R}}^{2d}\) consisting of vectors with all coordinates equal. This line corresponds to the configurations that give an equipartition. Back to using \(\phi\) as a section, and using multicativity of Euler classes, the bundle itself has zero Euler class because the diagonal line, being trivial, has zero Euler characteristic class. There will thus be a section vanishing on the line diagonal line so that an equipartition exists for some \(\rho \in E\). The author has informed the reviewer that there is an error in his proof of Theorem 2 in section 4, on inscribing a regular crosspolytope, and that he has since corrected it on the archive network.