The Eulerian distribution on involutions is indeed unimodal (Q2500618): Difference between revisions

Let \( I_{n,k} \) and \( J_{n,k} \) be the number of involutions resp. fixed point free involutions of \( \{1,\dots,n\} \) having \(k\) descents. The polynomials \(I_{n}(t)\) and \(J_{n}(t)\) are defined by \(I_n(t) = \sum I_{n,k} t^k\) and \( J_n(t) = \sum J_{n,k}t^k \). Up to now what was known about these polynomials is: 1) they were symmetric, proved by Strehl (1981); 2) they were conjectured to be log concave by Brenti (1994); 3) some partial results on unimodularity by Dukes (2006). By considering the generating functions as obtained by Désarménian and Foata (1985) and separately by Gessel and Reutenauer (1993), the authors derive recurrence relations for \( I_{n,k} \) and \( J_{2n,k} \). Using these recurrence relations they were able to prove the unimodularity of these sequences by induction, applying a very usefull Lemma (Lemma 3.1). Rewriting \( I_n(t) \) and \( J_{2n}(t) \) in terms of \( t^k((1+t)^{(n-2k-1)} \) resp. \( t^k(1+t)^{(2n-2k)} \) the authors derive recurrence relations for the corresponding coefficients and conjecture the nonnegativity of these coefficients for almost all values of \(n\) and \(k\). To prove this conjecture it would be enough to prove the nonnegativity in some boundary cases. While working through the paper I verified all formulas up to Section 4, not finding any errors.