13
$\begingroup$

(EDIT: I have removed the denominators I had in a previous version as they were superfluous)

The $N\times N$ determinant $$D(a,\vec{b})=\det\left((2N+a+b_j-i-j)!\right)$$ has the nice form $$D(a,\vec{b})=\prod_{j=1}^N(N+a+b_j-j)!\prod_{i=j+1}^N(b_j-b_i-j+i).$$

Since from the definition it is clear that $D(a,\vec{b})$ is antisymmetric in the variables $x_j=(N+b_j-j)$, it should be proportional to the Vandermonde of the $x$.

I would like to know if the generalization where $a$ is allowed to vary with $i$ has a nice expression as well, $$D(\vec{a},\vec{b})=\det\left((2N+a_i+b_j-i-j)!\right)=?$$

This is antisymmetric in both $x_j=(N+b_j-j)$ and $y_i=(N+a_i-i)$, so it should be proportional to both Vandermondes of the $x$ and the $y$...

I know Krattenthaler has this great paper about determinants, but I was not able to find help there.

$\endgroup$
5
  • $\begingroup$ Does it factor? I mean, a CAS can factor some expressions.. $\endgroup$ Commented Apr 12, 2018 at 16:41
  • $\begingroup$ Math experiment done with Maple brings nothing of the kind for $N=3, N=4.$ The executed code on demand. $\endgroup$
    – user64494
    Commented Apr 12, 2018 at 16:43
  • 1
    $\begingroup$ Your $D(\vec{a},\vec{b})=\det\left((2N+a_i+b_j-i-j)!\right)=? $ was changed without indicating it in your edit. My pevious comment was about the previous version of$D(\vec{a},\vec{b})=\det\left((2N+a_i+b_j-i-j)!/(\left( N-i \right) !\, \left( N+a_{{i}}-i \right) !)\right)$. $\endgroup$
    – user64494
    Commented Apr 12, 2018 at 19:58
  • $\begingroup$ @user64494 As I said in the edit, all I did was remove some superfluous denominators. Nothing substantial has changed. $\endgroup$
    – Marcel
    Commented Apr 12, 2018 at 23:24
  • 2
    $\begingroup$ With $x_i = N+a_i-i$, $y_j = N+b_j-j$, you are asking for $\det((x_i+y_j)!)$. If you multiply the $i$th row by $(x_i!)^{-1}$ and the $j$th column by $(y_j!)^{-1}$, then up to a factor $\prod_i x_i! \prod_j y_j!$, you just need $\det(\binom{x_i+y_j}{x_i})$. The Lindström-Gessel-Viennot lemma gives a combinatorial interpretation of the latter determinant in terms of lattice paths between the set of points $\{(-x_i,0) \mid i=1,\dotsc,N\}$ and the set of points $\{(0,y_j) \mid j=1,\dotsc,N\}$. $\endgroup$ Commented Apr 13, 2018 at 3:24

1 Answer 1

5
$\begingroup$

I have found a solution myself, at least in the case when $a$ and $b$ are partitions.

The determinant can be written as $$ D=\det((x_i+y_j)!)=\det\left( \int z^{x_i+y_j}e^{-z}dz\right)$$ We resort to the Andreief identity, $$ \int dz \det(f_{i}(z_{j}))\det(g_{i}(z_{j}))=N! \det\left(\int f_{i}(z)g_{j}(z)dz\right).$$ This is usually used from left to right, but I used in reverse to write $$D=\frac{1}{N!}\int dz \det(z_{j}^{x_j})\det(z_{j}^{y_i})\prod_{i=1}^Ne^{-z_i}$$ Now, if $x_i=a_i-i+N$ and $y_i=b_i-i+N$ and if $a\vdash n$ and $b\vdash m$ are partitions, then $\det(z_{j}^{x_j})\det(z_{j}^{y_i})=(V(z))^2s_a(z)s_b(z)$, where $V$ is the Vandermonde and $s$ are the Schur functions. Then $$D=\frac{1}{N!}\int dz (V(z))^2s_a(z)s_b(z)\prod_{i=1}^Ne^{-z_i}$$

The Littlewood-Richardson coefficients are defined by $s_as_b=\sum_{\rho\vdash n+m} c^\rho_{ab}s_\rho$. Using them we have $$D=\frac{1}{N!}\sum_{\rho\vdash n+m} c^\rho_{ab}\int dz (V(z))^2s_\rho(z)\prod_{i=1}^Ne^{-z_i}.$$ This is an integral of the Selberg type, and the result is known: $$\int dz (V(z))^2s_\rho(z)\prod_{i=1}^Ne^{-z_i}=N!\frac{d_\rho}{(n+m)!}\prod_{j=1}^N (\rho_i+N-i)!^2 \quad (\ell(\rho)\le N),$$ where $d_\rho$ is the dimension of the irreducible representation of the permutation group labeled by $\rho$.

Therefore, $$D=\frac{1}{(n+m)!}\sum_{\substack{\rho\vdash n+m\\\ell(\rho)\le N}} d_\rho c^\rho_{ab}\prod_{j=1}^N (\rho_i+N-i)!^2.$$

Curiously, there are no Vandermondes in this solution.

$\endgroup$
0

You must log in to answer this question.

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