3
$\begingroup$

For any fixed $c>0$ consider the polynomials \begin{align*} & p_n(X_1,X_2,\ldots) := \frac{n!}{c} \sum\limits_{b=1}^n \frac{c^b}{b!(n+1-b)!} \sum\limits_{\substack{l_1,\ldots,l_b \geq 1 \\ l_1+\cdots+l_b=n}} X_{l_1} \cdots X_{l_b} \ , \tag{1} \end{align*} which we can also write as a sum over partitions as \begin{align*} & p_n(X_\bullet) = \frac{n!}{c} \sum\limits_{\pi \vdash n} \frac{\operatorname{sym}(\pi) c^{\ell(\pi)}}{\ell(\pi)!(n+1-\ell(\pi))!} X_{\pi_1} \cdots X_{\pi_{\ell(\pi)}} \ , \tag{2} \end{align*} where $\ell(\pi)$ is the length of the partition and $\operatorname{sym}(\pi)$ denotes the number of distinct ways in which the numbers $(\pi_1,\ldots,\pi_{\ell(\pi)})$ can be arranged. I was looking for polynomials $q_n$ with the 'inversion'-property \begin{align*} & X_n = q_n(p_1(X_\bullet),p_2(X_\bullet),\ldots) \tag{3} \end{align*} and with some help from the computer and OEIS have found the candidate \begin{align*} q_n(Y_1,Y_2,\ldots) = & -\frac{1}{c} \sum\limits_{a=1}^n \frac{(-c)^a (n+a-2)!}{a!(n-1)!} \sum\limits_{\substack{k_1,\ldots,k_a \geq 1 \\ k_1+\cdots+k_a=n}} Y_{k_1} \cdots Y_{k_a} \tag{4}\\ = & -\frac{1}{c} \sum\limits_{\tau \vdash n} \frac{(-c)^{\ell(\tau)} \operatorname{sym}(\tau) (n+\ell(\tau)-2)!}{\ell(\tau)! (n-1)!} Y_{\tau_1} \cdots Y_{\tau_{\ell(\tau)}} \ . \tag{5} \end{align*} With the computer I have checked the property (3) up to $n=25$, where I have used formulas (2) and (5) for the polynomials. Unfortunately, proving this seems to require methods for partitions that I have little experience with.

Any help is much appreciated!


Edit: No idea if this helps, but as $q_n(p_1(X_\bullet),p_2(X_\bullet),\ldots)$ has the form $\sum\limits_{\eta \vdash n} c^{\ell(\eta)-1} C_\eta X_{\eta_1} \cdots X_{\eta_{\ell(\eta)}}$ for some coefficients $C_\eta$ and one can relatively easily see $C_{(n)} = 1$, it suffices to show that the map \begin{align*} & c \mapsto q_n(p_1(X_\bullet),p_2(X_\bullet),\ldots) \end{align*} is constant.

Also here are the first five polynomials: \begin{align*} & p_1(X) = X_1\\ & p_2(X) = X_2 + c X_1^2\\ & p_3(X) = X_3 + 3c X_2X_1 + c^2 X_1^3\\ & p_4(X) = X_4 + 4c X_3X_1 + 2c X_2^2 + 6c^2 X_2X_1^2 + c^3 X_1^4\\ & p_5(X) = X_5 + 5c X_4X_1 + 5c X_3X_2 + 10c^2 X_3X_1^2 + 10c^2 X_2^2X_1 + 10c^3 X_2 X_1^3 + c^4 X_1^5 \end{align*} \begin{align*} & q_1(Y) = Y_1\\ & q_2(Y) = Y_2 - c Y_1^2\\ & q_3(Y) = Y_3 - 3c Y_2Y_1 + 2c^2 Y_1^3\\ & q_4(Y) = Y_4 - 4c Y_3Y_1 - 2c Y_2^2 + 10c^2 Y_2Y_1^2 - 5c^3 Y_1^4\\ & q_5(Y) = Y_5 - 5c Y_4Y_1 - 5c Y_3Y_2 + 15c^2 Y_3Y_1^2 + 15c^2 Y_2^2Y_1 - 35c^3 Y_2Y_1^3 + 14c^4 Y_1^5 \end{align*}

$\endgroup$
4
  • $\begingroup$ It might help to give explicitly the first four or so $p_n$ with the actual real numbers. I've looked at several such inverse sets of partition polynomials, so I might be able to identify your set from the first few examples. $\endgroup$ Commented Sep 13 at 14:23
  • 1
    $\begingroup$ @TomCopeland Sure, I've added the first five terms to the question! $\endgroup$ Commented Sep 13 at 14:50
  • $\begingroup$ Suppose you write a+...+b in LaTeX or MathJax. In MathJax you will see $a+...+b$ with the dots asymmetrically positions to be closer to $a$ than to $b,$ and without proper spacing to the right and left of the "plus" signs. By contrast, with a+\cdots+b and with a+\ldots+b you see $a+\cdots+b$ and $a+\ldots+b,$ with an amount of space surrounding the "plus" signs as in $a+b$ rather than as in $a{+}b.$ In LaTeX (as opposed to MathJax), you'll see something that looks about like $a+\text{...}+b,$ with the dots uncomfortably close together. Some readers won't notice any of this, but$\,\ldots$ $\endgroup$ Commented Sep 13 at 18:56
  • $\begingroup$ $\ldots\,$but instead will say "This font doesn't look as good as that font," without awareness of the specific differences. $\endgroup$ Commented Sep 13 at 18:56

1 Answer 1

3
$\begingroup$

These are the inverse pair OEIs A134264 and OEIS A350499.

$\endgroup$
7
  • $\begingroup$ Btw, to find such sequences with integer coefficients in the OEIS, it often works to reduce the partition polynomials to a regular polynomial by letting each $x_k$ = $y$; e.g., $P_4$ becomes $y + 6y^2+6y^3+y^4$. Then searching on 1,3,1; 1,6,6,1; 1,10,20,10,1 will lead you to the Narayana polynomials A001263 which will lead you to A134264. Alternatively, you can try to order the partitions as in Abramowitz & Stegun or reverse to that and search on the unreduced numbers. (For me, I've used these partition polynomials dozens of times over the years, so it was easy to recognize them.) $\endgroup$ Commented Sep 13 at 21:46
  • 2
    $\begingroup$ Thank you! This, and especially your question from 2021 (mathoverflow.net/questions/412573/…) has been very helpful! $\endgroup$ Commented Sep 13 at 22:04
  • $\begingroup$ You're welcome. I may ask you some questions on random matrix theory in the future. $\endgroup$ Commented Sep 13 at 22:13
  • $\begingroup$ I tried looking in FindStat and drew a blank, so I'm surprised that it was something so well known. $\endgroup$ Commented Sep 14 at 0:03
  • $\begingroup$ @PeterTaylor, define so well known. $\endgroup$ Commented Sep 14 at 0:55

You must log in to answer this question.

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