4
$\begingroup$

I am computing the volumes of certain polytopes and it turns out that knowing a "closed formula" for the following number would help a lot.

Determine the number of permutations $\sigma\in \mathfrak{S}_{n}$ such that $\sigma$ has exactly $k$ descents and such that in the segment $\sigma_1\sigma_2\cdots\sigma_s$ there are at most $t$ descents.

A clarification: a descent in a permutation $\sigma\in \mathfrak{S}_n$ is an index $i\in [n-1]$ such that $\sigma_i > \sigma_{i+1}$.

At first glance one can think of "choosing" the first $s$ numbers for $\sigma$ (inline notation) and counting in how many ways one can reorder them so that there are $0\leq j\leq t$ descents in that part, and then order the remaining $n-s$ numbers producing $k-j$ descents. Namely, do something like $$ \sum_{j=0}^t \binom{n}{s} A_{s,j} A_{n-s,t-j},$$ where $A_{n,k} = \#\{\sigma\in \mathfrak{S}_n : \operatorname{des}(\sigma) = k\}$ is the classical Eulerian number.

This, however, does not work because we lose track of how the numbers $\sigma_s$ and $\sigma_{s+1}$ compare (that could introduce an extra descent).

By pretending a "closed formula" I intend a (possibly multiple) sum of products of binomial, Stirling and Eulerian numbers.

$\endgroup$
3
  • 2
    $\begingroup$ There is a way to count permutations which have given descent set in terms of inner products with ribbon Schur functions (see Stanley EC 2, Chapter 7), but this is probably not the kind of formula you are after. $\endgroup$ Commented Apr 5, 2022 at 13:52
  • 2
    $\begingroup$ (You can also write the number of permutations with descents in a given position as an inclusion-exclusion type alternating sum of multinomials, which amounts to the same thing as my previous comment.) $\endgroup$ Commented Apr 5, 2022 at 22:19
  • $\begingroup$ Thanks Sam. I will verify if a formula of that type does indeed simplify the work. $\endgroup$ Commented Apr 6, 2022 at 6:19

1 Answer 1

3
$\begingroup$

This question requires a refinement of Eulerian numbers. Namely, let $A_{n,k,\ell}$ be the number of permutations of order $n$ with $k$ descents and ending in element $\ell$. Under subtracting each element of a permutation from $n+1$, we get a property $A_{n,k,\ell}=A_{n,n-1-k,n+1-\ell}$. Similarly, denoting by $A'_{n,k,\ell}$ the number of permutations of order $n$ with $k$ descents and starting with element $\ell$, we have $A'_{n,k,\ell} = A_{n,n-1-k,\ell}$ (by reading permutations backwards).

Then an answer to the question can be given by the following formula: $$\sum_{j=0}^t \sum_{\ell=1}^n \sum_{f=\ell+1}^n \sum_{u+v+w=s-1} \binom{\ell-1}u\binom{f-\ell-1}v\binom{n-f}w A_{s,j,u+1} A'_{n-s,k-j,f-1-u-v}$$ $$+\sum_{j=0}^t \sum_{\ell=1}^n \sum_{f=1}^{\ell-1} \sum_{u+v+w=s-1} \binom{f-1}u\binom{\ell-f-1}v\binom{n-\ell}w A_{s,j,u+v+1} A'_{n-s,k-j-1,f-u},$$ where $\ell$ stands for $\sigma_s$, $f$ stands for $\sigma_{s+1}$, and $u,v,w$ are the number of elements of $\sigma_1,\dots,\sigma_{s-1}$ in each of the three subintervals of $[1,n]$ obtained by removal of $\ell$ and $f$. As explained above, we have $$A'_{n-s,k-j,f-1-u-v} = A_{n-s,n-s-1-(k-j),f-1-u-v} = A_{n-s,k-j,n+1-f-w}$$ and $$A'_{n-s,k-j-1,f-u} = A_{n-s,n-s-1-(k-j-1),f-u} = A_{n-s,k-j-1,n-s+1-f+u}.$$ So, it now amounts to having a formula for $A_{n,k,\ell}$.


First, we have a recurrence formula for $n\geq\ell>1$: $$A_{n,k,\ell} = A_{n,k,\ell-1} + A_{n-1,k,\ell-1} - A_{n-1,k-1,\ell-1},$$ which translated into a linear first-order ODE: $$(1-z)\frac{\partial}{\partial x} F(x,y,z) = z(1-y)F(x,y,z) + yz\frac{y-1}{y - e^{(y-1)x}} - z^2 \frac{y-1}{y - e^{(y-1)xz}}$$ for the generating function: $$F(x,y,z) := \sum_{n,k,\ell\geq 0} A_{n,k,\ell}\frac{x^n}{n!} y^k z^\ell.$$ From its solution, we obtain the following formula: $$A_{n,k,\ell} = \begin{cases} \left\langle n-1\atop k\right\rangle & \text{if }\ell=n\\ (-1)^n \sum\limits_{j=1}^n (-1)^j j^{n-\ell} (j-1)^{\ell-1} \bigg( (-1)^k\binom{n}{k+j} - \binom{n}j\bigg) & \text{if }\ell < n. \end{cases} $$

$\endgroup$

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .