8
$\begingroup$

Let us consider the set $[n]=\{1,\ldots,n\}$ and all of its partitions into exactly $m$ blocks, but let us allow each block to be internally ordered. For example, taking $n=6$ and $m=2$, we will distinguish between $\{(1,2),(3,5,4,6)\}$ and $\{(1,2),(5,3,4,6)\}$, but we will consider $\{(1,2),(3,5,4,6)\}$ and $\{(3,5,4,6),(1,2)\}$ as the same object.

For each partition $\pi$ of $[n]$ into $m$ parts as above, we will define a notion of weight as follows: for each part of $P$ of $\pi$, consider the number of elements in $P$ that are smaller (as positive integers) than the first element of $P$, this will be denoted by $w(P)$. Then, the weight of $\pi$ is defined by $$w(\pi) := \sum_{P\in \pi} w(P)$$

For instance, the weight of $\pi_1=\{(1,2),(3,5,4,6)\}$ is $w(\pi_1) = 0 + 0 = 0$, the weight of $\pi_2=\{(1,2),(5,3,4,6)\}$ is $w(\pi_2) = 0 + 2 = 2$, etc.

Denote by $W(\ell,n,m)$ the number of partitions $P$ of $[n]$ into $m$ parts and having weight $\ell$.

Problem: Show that for each $n,m$, the polynomial: $$P_{n,m}(x) := \sum_{\ell=0}^{n-m} W(\ell,n,m)\, x^{\ell}$$ has all of its complex roots lying on the unit circle $|z|=1$.

For example, if $n=6$ and $m=2$, one may calculate: $$P_{6,2}(x) = 274 + 404x + 444x^2+404x^3+274x^4.$$

For $n=7$ and $m=3$,

$$P_{7,3}(x)=1624 + 2954x + 3444x^2 + 2954x^3 + 1624x^4$$

I was originally interested in proving that $P_{n,m}(x)$ was log-concave, and I discovered that this unit-circle-rootedness pattern holds for all small values of $n$ and $m$.

$\endgroup$
8
  • 1
    $\begingroup$ The Lee-Yang Circle Theorem gives a general class of polynomials whose zeros all lie on the unit circle, but I don't know if it's related to the question. For a clear statement not involving physics, see arxiv.org/pdf/1201.3169.pdf. $\endgroup$ Commented Jun 27, 2023 at 19:19
  • 3
    $\begingroup$ One can show that $\sum_{n\geq 0}\sum_{m\geq 1}P_{n,m}(x) t^m\frac{q^n}{n!}=\left( \frac{1-qx}{1-q}\right)^{t/(1-x)}$. $\endgroup$ Commented Jul 5, 2023 at 16:24
  • 1
    $\begingroup$ @RichardStanley That sounds interesting! I tried to relate this with the unit-circle-rootedness problem, but it doesn't seem to be easy. Relatedly, I was able to prove the following recurrence $P_{n,m}(x) = (n-1)(x+1)\,P_{n-1,m}(x)+P_{n-1,m-1}(x)-(n-1)(n-2)x\,P_{n-2,m}(x)$. Maybe this is useful somehow (it reminds me a lot to recurrences that people use to prove that polynomials are real-rooted, but here I want unit-circle-rootedness and I have this annoying minus sign in the recurrence). $\endgroup$ Commented Jul 6, 2023 at 9:54
  • 3
    $\begingroup$ It seems that with some small exceptions, either $P_{n,m}(x)$ is irreducible or $P_{n,m}(x)=(x+1)Q_{n,m}(x)$, where $Q_{n,m}(x)$ is irreducible. Moreover, the Galois groups of $P_{m,n}(x)$ (when irreducible) and $Q_{n,m}(x)$ are always hyperoctahedral groups. More computation would be desirable. $\endgroup$ Commented Jul 6, 2023 at 19:39
  • 1
    $\begingroup$ An empirical observation: the roots of $P_{n,m}(x)$ are approximately $-\exp(2\pi i t_j)$ where $t_j=\frac{j}{n-(m-1)/2}$ for $j=-\frac{n-m-1}{2}, \ldots, \frac{n-m-1}{2}$ (incrementing by $1$ each time). It's exact when $m=1$. This implies a lot of interlacing. Roughly, it says $P_{n, m}(x)$ is approximately a "truncated" q-integer, where some initial and final roots have been left off. Very curious. $\endgroup$ Commented Nov 23, 2024 at 8:58

0

You must log in to answer this question.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.