2
$\begingroup$

I reduced a certain determinant to the following which has the form: \begin{vmatrix} 1&1&\cdots & 1 \\ -a_1&a_2&\cdots &a_n\\ -a_1^{2}&a_2^{2}&\cdots&a_n^{2}\\ \vdots&\vdots&\ddots& \vdots\\ -a_1^{n-1}&a_2^{n-1}&\cdots&a_n^{n-1}\\ \end{vmatrix}

To clarify a bit, it is exactly the Vandermonde determinant except the first column is negative, while the $(1,1)$-entry is still $1$.

I think the key is to apply Vandermonde determinant yet I can't proceed. There may be a quick answer to this, however. Any hints?

$\endgroup$
  • 1
    $\begingroup$ Have you tried to developing the determinant using the first column? You can then use the Vandermonde formula for the determinants of the submatrices. $\endgroup$ – Olivier Mar 11 '17 at 18:35
  • $\begingroup$ Is that you? No that can't be right. Karma works on differential forms in Vegas now. $\endgroup$ – mathreadler Mar 11 '17 at 18:36
  • $\begingroup$ Wikipedia claims it is $\prod_{0\leq k\leq l\leq n}(a_l- a_k)$ $\endgroup$ – mathreadler Mar 11 '17 at 18:43
  • $\begingroup$ @mathreadler: I doubt it. As you write it, with $k\leq l$, your expression is zero. If you sharpen it up you've then said it's the Vandermonde determinant, which it is not. $\endgroup$ – ancientmathematician Mar 11 '17 at 18:50
  • $\begingroup$ hmm it is not vandermonde determinant? which of them are supposed to be negative? only first column? I don't see any other difference than first column being multiplied by -1. $\endgroup$ – mathreadler Mar 11 '17 at 18:55
1
$\begingroup$

By linearity in or Laplace expansion along the first column, the determinant is equal to $$ \begin{vmatrix} -1&1&\cdots & 1 \\ -a_1&a_2&\cdots &a_n\\ -a_1^{2}&a_2^{2}&\cdots&a_n^{2}\\ \vdots&\vdots&\ddots& \vdots\\ -a_1^{n-1}&a_2^{n-1}&\cdots&a_n^{n-1}\\ \end{vmatrix} + 2\begin{vmatrix} a_2&\cdots &a_n\\ a_2^{2}&\cdots&a_n^{2}\\ \vdots&\ddots& \vdots\\ a_2^{n-1}&\cdots&a_n^{n-1}\\ \end{vmatrix}. $$ Now each of the two summands is a multiple of a Vandermonde determinant.

$\endgroup$
  • $\begingroup$ Can you explain a bit? I tried the expansion but how to simplify to yours? $\endgroup$ – Karma Mar 12 '17 at 4:16
  • $\begingroup$ In a neat way he has replaced your first column by $(-1,-a_1,\dots,-a_1^{n-1})^T+(2,0,\dots,0)^T$ and used linearity. $\endgroup$ – ancientmathematician Mar 12 '17 at 7:38
1
$\begingroup$

It is also possible to evaluate a more general determinant, where the clever observation of @user1551 isn't available. Let $c_0,\dots, c_1$ be arbitrary and let $X$ be a polynomial variable. Then we will evaluate $$ \begin{vmatrix} c_0&1&\cdots & 1 \\ c_1 X&a_2&\cdots &a_n\\ c_2 X^2&a_2^{2}&\cdots&a_n^{2}\\ \vdots&\vdots&\ddots& \vdots\\ c_{n-1}X^{n-1}&a_2^{n-1}&\cdots&a_n^{n-1}\\ \end{vmatrix}. $$

By a Lagrange expansion on the first column we get that this is $$ \sum_{0}^{n-1} (-1)^{k} c_k X^k V_k \hspace{10em}\text{(*)} $$ where $V_k$ is the determinant of the corresponding $(n-1)\times(n-1)$ minor in $$ \begin{vmatrix} 1&\cdots & 1 \\ a_2&\cdots &a_n\\ a_2^{2}&\cdots&a_n^{2}\\ \vdots&\ddots& \vdots\\ a_2^{n-1}&\cdots&a_n^{n-1}\\ \end{vmatrix}. $$ But now consider the special case where all $c_k=1$; this is just the standard Vandermonde determinant, so we have $$ \sum_{0}^{n-1} (-1)^{k} X^k V_k = \prod_{r=2}^{n} (X-a_r)\prod_{2\leqslant s < t \leqslant n} (a_t - a_s). $$

We write, in the usual way, $$ \prod_{r=2}^{n} (X-a_r)=\sum_{k=0}^{n-1}(-1)^{k} \sigma_{n-1-k} X^k; $$ these are just the usual symmetric functions of the $a_2,\dots, a_{n}$ taking their sum, their sum two at a time, and so on.

The coefficients $V_k$ then all have a common factor $\prod_{2\leqslant s < t \leqslant n} (a_t - a_s)$, the Vandermonde of $a_2, \dots, a_n$ (which deals with the skew-symmetry in these variables); the other factor in each case is just the symmetric function $\sigma_{n-k}$: $$ V_k= \sigma_{n-k}\prod_{2\leqslant s < t \leqslant n} (a_t - a_s).$$

Taking $c_0=1, c_1=\dots=c_n=1$ and substituting $X=a_1$ in (*) gives the value for the determinant of the original question.

$\endgroup$

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.