# Symmetric distribution of maj over des in pattern avoidance classes

It appears from computation to be the case (and would prove at least one clause of a conjecture advanced by Bruce Sagan and collaborators in a recent preprint) that in some pattern avoidance classes of permutations, the distribution of the major index is symmetric among permutations with a given descent number. For instance, $$\sum_{S_6(1234)} q^{maj(\sigma)}t^{des(\sigma)} = \dots + (10 q^6 + 35 q^7 + 66 q^8 + 80 q^9 + 66 q^{10} + 35 q^{11} + 10 q^{12}) t^3 + \dots.$$

As you can see, the coefficient of $t^3$ is a symmetric polynomial.

This is not the case for all avoidance classes: for instance, $$\sum_{S_6(2134)} q^{maj(\sigma)}t^{des(\sigma)} = \dots + (4 q^6 + 21 q^7 + 42 q^8 + 61 q^9 + 56 q^{10} + 35 q^{11} + 10 q^{12}) t^3 + \dots.$$

Before I start hammering at this, I was wondering if this was known to be the case for any particular avoidance classes, and if so, which. Since I have not even found any papers that seem to deal with the subject, pointers to one you know of would also be gratefully received.

-
My vague impression is that there are papers on permutation statistics for pattern avoidance classes, but I don't know specific references. Sadly, your best bet may be to show up at Permutation Patterns 2012 (in Glasgow in June) and ask around. – Alexander Woo Feb 28 '12 at 22:18

Let $w=a_1 a_2\cdots a_n\in S_n$. Set $w' =n+1-a_n,n+1-a_{n-1},\dots,n+1-a_1$. Then (1) $\mathrm{des}(w)=\mathrm{des}(w')$, (2) $\mathrm{maj}(w)+\mathrm{maj}(w')=dn$, where $d=\mathrm{des}(w)$, and (3) $\mathrm{is}(w)=\mathrm{is}(w')$, where $\mathrm{is}(w)$ denotes the length of the longest increasing subsequence of $w$. This implies the suggested symmetry property for $S_n(12\cdots p)$. Since the map $w\mapsto w'$ also preserves the length of the longest decreasing subsequence (in fact, it preserves the RSK insertion tableau), the symmetry property also holds for $S_n(12\cdots p, q\cdots 21)$.
Thanks Richard. Your argument also proves the property for any other class preserved by reverse-complement, such as $S_n(2143)$, since that's the purpose of (3) and the properties (1) and (2) hold generally. Also seems to hold for a few other singleton classes, but given the discussion so far I think I can attack it with confidence and at least know I'm not reinventing the wheel. – William J. Keith Mar 1 '12 at 11:09