Permutations from ranking independent random variables

Abstract

Let $X_1,\ldots,X_n$ be almost surely distinct independent real-valued random variables. Let $σ$ be the random permutation of $\{1,\ldots,n\}$ such that $X_{σ(1)}<X_{σ(2)}<\cdots< X_{σ(n)}$. We show that the set of laws of $σ$, as the laws of $X_1,\ldots,X_n$ vary, has semialgebraic dimension \[ \sum_{k=2}^n {n \choose k}(k-1)! \] as a subset of the $(n!-1)$-simplex. This establishes a conjecture of Babson, Duchin, Iseli, Poggi-Corradini, Thurston, and Tucker-Foltz who proved that the dimension is upper bounded by the above expression. We prove their conjecture by constructing a family of finitely supported laws for $X_1,\ldots,X_n$ that provides the correct dimension. We also give an alternative proof of the upper bound using a theorem of Radford. Furthermore, we show that the Mallows law on permutations cannot arise as a law of $σ$ for $n\geq 4$, answering a question of Ed Crane.