1
$\begingroup$

Let $\mathrm{CO}(m)$ be the set of all compositions of the positive integer $m$. By a composition of $m$, I mean a finite sequence $(m_1,\ldots,m_k)$ of positive integers with sum $\sum m_i=m$. There are $2^{m-1}$ compositions of $m$. The $4$ compositions of $3$ are

$$ \mathrm{CO}(3) =\{ (1,1,1), (1,2), (2,1), (3)\} $$

Let also $q$ be a prime power. Put $[t]_q = \frac{q^t-1}{q-1}$. Write $[m]_q! = \prod_{1 \le t \le m} [t]_q$ for the $q$-factorial of $m$ and

$$ \binom{m}{m_1,\ldots,m_k}_q = \frac{[m]_q!}{[m_1]_q! \cdots [m_k]_q!} $$

for the $q$-multinomial coefficient of the composition $(m_1,\ldots, m_k)$ of $m$. Let $S(m,q)$ be the sequence whose $d$th element, $d=0,1,\ldots$, is $$ S(m,q)(d)= \sum_{(m_1,\ldots,m_k) \in \mathrm{CO}(m)} (-1)^{m-k} \binom{m}{m_1,\ldots,m_k}_q q^{\sum \binom{m_i}{2}}\binom{q^{\binom{m}{2}- \sum \binom{m_i}{2}}}{d} $$ For $m=3$ we get the sequence with $d$th element \begin{equation*} S(3,q)(d) =\binom{3}{1,1,1}_q q^0\binom{q^3}{d} -2 \binom{3}{1,2}_q q^1\binom{q^2}{d} + \binom{3}{3}_q q^3 \binom{q^0}{d} \end{equation*} which for $q=2$ evaluates to \begin{equation*} S(3,2)=1, 64, 420, 1064, 1442, 1176, 588, 168, 21, 0, 0, \ldots \end{equation*} Note that $S(3,2)$ is a log-concave sequence. My question is: Is the sequence $S(m,q)$ always log-concave?

$S(m,q)(d)$, for $q$ a power of the prime $p$, is the number of $d$-subsets of $\mathrm{SL}_m(\mathbf{F}_q)$ generating a $p$-subgroup. These subsets form a contractible simplicial complex with reduced Euler characteristic $\sum_{d \ge 0} (-1)^d S(m,q)(d)=0$.

$\endgroup$
5
  • $\begingroup$ Huh, these expressions looks very close to something a coworker asked me about (without the qs). Looks like power-sums expanded into monomial basis, for symmetric functions, or similar. $\endgroup$ Commented Jan 30 at 20:47
  • $\begingroup$ I replaced "ordered partition of $m$" by "composition of $m$" and explained the significance of the number $S(m,q)(d)$. $\endgroup$ Commented Jan 31 at 9:56
  • $\begingroup$ how far did you check this numerically? The same question about real-rootedness of the generating polynomial $\sum_d S(m,q)(d)t^d$. $\endgroup$ Commented Jan 31 at 10:24
  • $\begingroup$ The sequences S(m,2) and S(m,4) are log-concave for $m \le 4$. I know nothing about the generating polynomial $\sum_d S(m,q)(d)t^d$. $\endgroup$ Commented Jan 31 at 10:52
  • 1
    $\begingroup$ @FedorPetrov, my approach to numerical tests is to calculate $S(m, q)(d)^2 - S(m, q)(d-1)S(m, q)(d+1)$ as a polynomial in $q$, bound its largest real root, and test all integers from $2$ up to that bound. Using that approach I've found no counterexamples for $2 \le m \le 8$, $1 \le d < 100$. As a bonus observation, for $m > 4$ the largest upper bound I've found was about $4.4$, so not many values of $q$ need to be tested. $\endgroup$ Commented Jan 31 at 11:41

0

You must log in to answer this question.

Browse other questions tagged .