Let $z_{1},\dots,z_{k}$ be distinct complex numbers with $\left|z_{j}\right|=1,\;j=1,\dots,k$. For any natural $N\geqslant k$ consider the rectangular Vandermonde matrix $$ V_{N}=\begin{pmatrix}1 & 1 & \dots & 1\\ z_{1} & z_{2} & & z_{k}\\ \vdots & \vdots & & \vdots\\ z_{1}^{N-1} & z_{2}^{N-1} & & z_{k}^{N-1} \end{pmatrix}. $$

Let $V_{N}^{*}$ denote the conjugate transpose of $V_{N}$. Since $V_{N}$ has full rank, the square $k\times k$ matrix $V_{N}^{*}V_N$ is nonsingular. We are interested in the quantity $$ D\left(N\right)=\det\left(V_{N}^{*}V_N\right). $$ For $N=k$ we have by the well-known explicit formula $$ D\left(k\right)=\prod_{i<j}\left|z_{i}-z_{j}\right|^{2}. $$

Question: Does there exist an explicit ``simple'' expression for $D\left(N\right)$ with arbitrary $N>k$?

Example of a simple expression in the special case $k=2$:

Without loss of generality $z_{1}=1$ and $z_{2}=\exp\left(\imath x\right)$ for $x\in\left[-\pi,\pi\right]\setminus\{0\}.$ An explicit computation gives the following: $$ D\left(N\right)=N^{2}-\frac{\sin^{2}\frac{N}{2}x}{\sin^{2}\frac{x}{2}}. $$


This is a partial answer only - but there is a clear place to start.

The Cauchy-Binet theorem gives the answer as a sum of products of all maximal minors of the two matrices, where the minors are taken with matching sets of rows/columns. All of these minors are themselves ordinary, N by N Vandermonde determinants. So the answer is the symmetrization of the squared Vandermonde determinant, over all possible choices of N of the k variables.

This is probably not simple enough for you, but at least it's a formula. I haven't assumed anything about the z variables, nor attempted to do any simplification.

  • $\begingroup$ Benjamin, thanks for the reference to Cauchy-Binet. However it does not seem to be helpful straight away. First notice that $N>k$ and thus the maximal minors would be $k\times k$. Second, these minors are definitely NOT the usual Vandermonde because they would involve non-consecutive powers of the $z$ variables. For instance, if $k=3$ and $N>5$ then there would be the minor $$\begin{pmatrix} z_1 & z_2 & z_3\\ z_1^3 & z_2^3 & z_3^3\\z_1^4 & z_2^4 & z_3^4 \end{pmatrix}.$$ $\endgroup$
    – dima
    Jan 22 '14 at 19:45
  • 2
    $\begingroup$ @dima: Determinants of that form are Schur polynomials times the Vandermonde determinant. $\endgroup$ Jan 22 '14 at 22:18
  • 1
    $\begingroup$ @MichaelJoyce: So we get $D(k)$ times a sum of norm-squareds of Schur polynomials. Specifically, it is the sum of the norm-square of all Schur polynomials corresponding to partitions whose largest part is $<N-k$. Is there a simpler expression for this? $\endgroup$
    – Will Sawin
    Jan 22 '14 at 23:14
  • $\begingroup$ We can rewrite $|s_{\lambda}(z_1, z_2, \ldots, z_k)|^2$ as $s_{\lambda}(z_1, \ldots, z_l) s_{\lambda}(\bar{z}_1, \bar{z}_2, \ldots, \bar{z}_k)$. This recalls Cauchy's identity: $\sum_{\lambda} s_{\lambda}(x_1, x_2, \ldots, x_k) s_{\lambda}(y_1, y_2, \ldots, y_k) = \prod_{i,j} (1-x_i y_j)^{-1}$. But I don't know what happens when you bound the size of the top row. $\endgroup$ Feb 21 '14 at 20:58

I guess that the determinant only factors if there is a stronger relation between the $z_j$. As example one can give a closed expression for $D_k(N)$ if $z_j = \exp(\iota (j-1) x)$. For $N$ even and $k=3$ one finds $$ D_k(N) = N^k -N \left[ 2 \frac{\sin^2\frac{Nx}{2}}{\sin^2\frac{x}{2}} + \frac{\sin^2 (Nx)}{\sin^2 x} \right] + 2\frac{\sin^2\frac{Nx}{2}}{\sin^2\frac{x}{2}} \frac{\sin(Nx)}{\sin x}. $$ I think the cases $k>3$ can also be derived. Note that for equidistant $z_j$ the Vandermonde becomes the discrete Fourier transform, at least if $N=k$.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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