Partitions of points into simplices with \(k\)-dimensional intersection. I: The conic Tverberg's theorem (Q5944012)

Tverberg's theorem, first proved in 1966, states that for any \(m,d>0\), any set of more than \((m-1)(d+1)+1\) points in \(\mathbb{R}^d\) may always be partitioned into \(m\) subsets such that the intersection of the convex hulls of the subsets is nonempty. That is, there is a point \(x\) such that \(x=\Sigma_j a_{ij}x_{ij}\) for each subset \(X_i\) and appropriate sets of non-negative numbers \(\{a_{ij}: j=1,2,\dots, n_i\}\) with \(\Sigma_j a_{ij}=1\). This has been generalized in various ways. In particular, Reay conjectured in 1968 that if the threshold were increased by \(k\) and the points were required to be in general position, the partition could be chosen to make the set of intersection points at least \(k\)-dimensional (the set is \((m,k)\)-divisible). While special cases of this have been proved (notable by the author, in this paper and the sequel), the general conjecture remains open. In this paper, the author proves a version of Tverberg's theorem in which some of the points are replaced by flats \(L_i\) (with a \(k\)-flat counting as \(k+1\) points) and the convex hulls by cones \([L_i,X_i)\). Informally, the theorem says that a person intending to perform the partition guaranteed by Tverberg's theorem may permit a ``hostile'' opponent to preassign any or all of the points to sets \(X_i\), provided that the coefficients \(a_{ij}\) corresponding to those points are thereafter permitted to be negative, and will still be able to obtain an intersection. This ``conic Tverberg's theorem'' is then used to prove several results on the \((m,k)\)-divisibility of sets of points, including the cases \(m=3\) and \(m=4\) of Reay's conjecture, and a Reay-type result with a threshold somewhat larger than that conjectured.