8
$\begingroup$

For $n=1,2,3,\ldots$ let $a_n$ denote the determinant $\det[(i^2+j^2)^n]_{0\le i,j\le n-1}$. Then $$a_1=0,\ a_2=-1,\ a_3=-17280,\ a_4= 1168415539200.$$

QUESTION: Is it true that $(2n)!\mid a_n$ for all $n=3,4,\ldots$?

I even conjecture that $$a_n'=\frac{(-1)^{n(n-1)/2}a_n}{2\prod_{k=1}^n(k!(2k-1)!)}$$ is a positive integer for every integer $n>2$. Note that \begin{gather*}a_3'=1,\ a_4'=559,\ a_5'=10767500,\ a_6'=9372614611500. \end{gather*}

The question is similar to my previous question http://mathoverflow.net/questions/302130. But it seems that darij grinberg's method there does not work for the present question.

Any comments are welcome!

$\endgroup$
2
  • 2
    $\begingroup$ My method easily generalizes to showing that $\det\left(\left(x_i+y_j\right)^n\right)_{0\leq i \leq n-1,\ 0 \leq j\leq n-1}$ is divisible by $n^2 V\left(x_0, x_1, \ldots, x_{n-1}\right) V\left(y_0, y_1, \ldots, y_{n-1}\right)$ whenever $x_0, x_1, \ldots, x_{n-1}, y_0, y_1, \ldots, y_{n-1}$ are $2n$ integers and $n \geq 3$. (Here, $V\left(a_0, a_1, \ldots, a_{n-1}\right)$ means the Vandermonde determinant $\prod\limits_{0 \leq i < j \leq n-1} \left(a_j - a_i\right)$.) Also, ... $\endgroup$ Jun 6, 2018 at 7:17
  • 1
    $\begingroup$ ... we have $V\left(0^2, 1^2, \ldots, \left(n-1\right)^2\right) = \prod\limits_{k=1}^{n-1} \left(k!^2 \dbinom{2k-1}{k}\right) = \prod\limits_{k=1}^{n-1} \left(k \left(2k-1\right)!\right)$. Since you have two of these $V$ factors around, you can kill almost all of your denominator; only $2 n \left(2n-1\right)!$ remains. I'm not sure where to get the $\left(2n-1\right)!$ from, though, particularly if $2n-1$ is prime. $\endgroup$ Jun 6, 2018 at 7:23

2 Answers 2

6
$\begingroup$

The conjecture in my posting has been proved finally! Its solution and further generalization now appear in the following preprint:

Darij Grinberg, Zhi-Wei Sun and Lilu Zhao, Proof of three conjectures on determinants related to quadratic residues, arXiv:2007.06453, 2020.

$\endgroup$
1
  • $\begingroup$ Published yet? or posted to arXiv? $\endgroup$ Dec 25, 2019 at 14:50
2
$\begingroup$

This is not an answer but a claim regarding some generalized determinant: $$\det\left[(x_i+x_j)^n\right]_{i,j=0}^{n-1}$$ is divisible by the square of the Vandermonde $$\prod_{i<j}^{0,n-1}(x_i-x_j)^2.$$ Letting $x_i=i^2$ recovers your matrix.

$\endgroup$
5
  • $\begingroup$ It is symmetric polynomial and equals to 0 when two variables are equal. This implies your claim. $\endgroup$ Jun 6, 2018 at 5:57
  • $\begingroup$ Further, using the suggested substitution gives the square of the Vandermonde as the square of the product of the odd factorials up to 2n-1!. This comes close to resolving the posted conjecture. Gerhard "Lots Of Multiplication Going On" Paseman, 2018.06.05. $\endgroup$ Jun 6, 2018 at 6:17
  • 1
    $\begingroup$ when we divide by the square of the Vandermonde we get a symmetric polynomial in $x_i$'s of degree $n$ which has degree 2 in each specific variable. Maybe it is explicitly computable? $\endgroup$ Jun 6, 2018 at 9:45
  • $\begingroup$ @FedorPetrov: Let $x_0, x_1, \ldots, x_{n-1}, y_0, y_1, \ldots, y_{n-1}$ be $2n$ elements of a commutative ring $R$. Let $\overrightarrow{x} = \left(x_0, x_1, \ldots, x_{n-1}\right)$ and $\overrightarrow{y} = \left(y_0, y_1, \ldots, y_{n-1}\right)$. For any $\overrightarrow{z} = \left(z_0, z_1, \ldots, z_{n-1}\right) \in R^n$, we let $V \left( \overrightarrow{z} \right)$ be the Vandermonde determinant $\prod\limits_{0 \leq i < j \leq n-1}\left(z_j - z_i\right)$, and we let $e_i\left(\overrightarrow{z}\right)$ be the $i$-th elementary symmetric polynomial in ... $\endgroup$ Jun 6, 2018 at 13:23
  • 1
    $\begingroup$ ... the $z_0, z_1, \ldots, z_{n-1}$ whenever $i$ is a nonnegative integer. (Note that $e_0\left(\overrightarrow{z}\right) = 1$.) Then, $\det \left( \left(x_i + y_j\right)^n \right)_{0 \leq i \leq n-1,\ 0 \leq j \leq n-1}$ $= V\left(\overrightarrow{x}\right) V\left(\overrightarrow{y}\right) \sum\limits_{k=0}^n \left(\prod\limits_{i \in \left\{0,1,\ldots,n\right\};\ i \neq k} \dbinom{n}{i}\right) e_{n-k}\left(\overrightarrow{x}\right) e_k\left(\overrightarrow{y}\right)$. This follows from the argument using Cauchy-Binet that I gave in the comments to mathoverflow.net/questions/302130 . $\endgroup$ Jun 6, 2018 at 13:25

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.