TAOCP 1.2.3 Exercise 38

Let D_n=\det\!

Section 1.2.3: Sums and Products

Exercise 38. ▶ [M25] Show that the determinant of Cauchy’s matrix is

$$\prod_{1 \le i < j \le n} (x_j - x_i)(y_j - y_i) \bigg/ \prod_{1 \le i, j \le n} (x_i + y_j).$$

Verified: yes
Solve time: 5m11s


Let

$$ D_n=\det!\left(\frac1{x_i+y_j}\right)_{1\le i,j\le n}. $$

We shall prove that

$$ D_n= \frac{\displaystyle\prod_{1\le i<j\le n}(x_j-x_i)(y_j-y_i)} {\displaystyle\prod_{1\le i,j\le n}(x_i+y_j)}. $$

For each row $i$, multiply by

$$ \prod_{k=1}^n (x_i+y_k). $$

Factoring these row multipliers out of the determinant gives

$$ D_n= \frac{1}{\displaystyle\prod_{1\le i,j\le n}(x_i+y_j)} \det(a_{ij}), $$

where

$$ a_{ij} = \frac{\prod_{k=1}^n(x_i+y_k)}{x_i+y_j} = \prod_{\substack{1\le k\le n\k\ne j}}(x_i+y_k). $$

Hence

$$ D_n= \frac{\Delta_n} {\displaystyle\prod_{1\le i,j\le n}(x_i+y_j)}, $$

where

$$ \Delta_n= \det!\left( \prod_{k\ne j}(x_i+y_k) \right)_{1\le i,j\le n}. $$

It remains to evaluate $\Delta_n$.

For fixed $j$, the $j$th column consists of values of the polynomial

$$ P_j(x)=\prod_{k\ne j}(x+y_k), $$

which has degree $n-1$.

If $x_r$ and $x_s$ are interchanged, rows $r$ and $s$ are interchanged. Therefore $\Delta_n$ changes sign. Hence $\Delta_n$ is alternating in the variables $x_1,\dots,x_n$, and therefore

$$ \prod_{1\le r<s\le n}(x_s-x_r) $$

divides $\Delta_n$.

Similarly, interchanging $y_r$ and $y_s$ interchanges columns $r$ and $s$, so $\Delta_n$ is alternating in the variables $y_1,\dots,y_n$. Therefore

$$ \prod_{1\le r<s\le n}(y_s-y_r) $$

also divides $\Delta_n$.

Now each entry of $\Delta_n$ has degree $n-1$ in the $x$-variables. Since each term of the determinant contains one entry from each row,

$$ \deg_x(\Delta_n)\le n(n-1). $$

The product

$$ V_xV_y, \qquad V_x=\prod_{r<s}(x_s-x_r), \quad V_y=\prod_{r<s}(y_s-y_r), $$

has total degree

$$ \deg(V_xV_y) = \frac{n(n-1)}2+\frac{n(n-1)}2 = n(n-1). $$

Since $V_xV_y$ divides $\Delta_n$ and already has the same total degree as $\Delta_n$, the quotient cannot depend on any variable. Thus

$$ \Delta_n=C,V_xV_y $$

for some constant $C$.

We now determine $C$.

Consider the coefficient of

$$ x_1^0x_2^1\cdots x_n^{,n-1} $$

in $\Delta_n$.

Write

$$ P_j(x)=\prod_{k\ne j}(x+y_k) =\sum_{m=0}^{n-1} e^{(j)}_{,n-1-m}x^m, $$

where $e_r^{(j)}$ denotes the elementary symmetric function of degree $r$ in the variables ${y_k:k\ne j}$.

By multilinearity of the determinant in its rows, the coefficient of

$$ x_1^0x_2^1\cdots x_n^{,n-1} $$

in $\Delta_n$ is

$$ M:=\det!\bigl(e^{(j)}{,n-i}\bigr){1\le i,j\le n}. $$

We must evaluate $M$.

For each $j$,

$$ P_j(t) = \sum_{i=1}^{n} e^{(j)}_{,n-i} t^{,i-1}. $$

Thus the $j$th column of the matrix

$\bigl(e^{(j)}_{,n-i}\bigr)$

is the coefficient vector of $P_j(t)$ in the basis

$1,t,\dots,t^{n-1}$.

Let

$$ A=(e^{(j)}{,n-i}){i,j=1}^n. $$

Let

$$ V= \bigl((-y_j)^{,i-1}\bigr)_{i,j=1}^n $$

be the Vandermonde matrix evaluated at $-y_1,\dots,-y_n$.

Consider the product $VA$. Its $(r,j)$ entry is

$$ \sum_{i=1}^n (-y_r)^{,i-1} e^{(j)}_{,n-i} = P_j(-y_r). $$

But

$$ P_j(-y_r) = \prod_{k\ne j}(y_k-y_r). $$

If $r\ne j$, one factor is $y_r-y_r=0$, so

$$ P_j(-y_r)=0. $$

If $r=j$,

$$ P_j(-y_j) = \prod_{k\ne j}(y_k-y_j). $$

Therefore $VA$ is diagonal:

$$ VA= \operatorname{diag} !\left( \prod_{k\ne1}(y_k-y_1), \dots, \prod_{k\ne n}(y_k-y_n) \right). $$

Taking determinants,

$$ \det(V)\det(A) = \prod_{j=1}^n\prod_{k\ne j}(y_k-y_j). $$

Now

$$ \det(V) = \prod_{1\le r<s\le n} \bigl((-y_s)-(-y_r)\bigr) = (-1)^{\frac{n(n-1)}2} \prod_{r<s}(y_s-y_r). $$

Also,

$$ \prod_{j=1}^n\prod_{k\ne j}(y_k-y_j) = (-1)^{\frac{n(n-1)}2} \left(\prod_{r<s}(y_s-y_r)\right)^2. $$

Hence

$$ \det(A) = \prod_{r<s}(y_s-y_r). $$

Therefore

$$ M=\prod_{r<s}(y_s-y_r). $$

On the other hand, in

$$ V_x=\prod_{r<s}(x_s-x_r), $$

the monomial

$$ x_1^0x_2^1\cdots x_n^{,n-1} $$

arises uniquely by choosing $x_s$ from every factor $(x_s-x_r)$. Hence its coefficient in $V_x$ is $1$.

Therefore the coefficient of

$$ x_1^0x_2^1\cdots x_n^{,n-1} $$

in

$$ C,V_xV_y $$

is

$$ C,\prod_{r<s}(y_s-y_r). $$

Comparing with the coefficient $M$ computed above yields

$$ C=1. $$

Thus

$$ \Delta_n = \prod_{1\le i<j\le n}(x_j-x_i)(y_j-y_i). $$

Substituting into the earlier expression for $D_n$,

$$ D_n= \frac{\displaystyle\prod_{1\le i<j\le n}(x_j-x_i)(y_j-y_i)} {\displaystyle\prod_{1\le i,j\le n}(x_i+y_j)}. $$

Hence

$$ \boxed{ \det!\left(\frac1{x_i+y_j}\right){1\le i,j\le n} = \frac{\displaystyle\prod{1\le i<j\le n}(x_j-x_i)(y_j-y_i)} {\displaystyle\prod_{1\le i,j\le n}(x_i+y_j)} }. $$