Consider the following matrix:

$$\left[\begin{array}{cccc} 1+x_1y_1 & 1+x_1y_2 & \ldots & 1+x_1y_n \\ 1+x_2y_1 & 1+x_2y_2 & \ldots & 1+x_2y_n \\ 1+x_3y_1 & 1+x_3y_2 & \ldots & 1+x_3 y_n \\ \vdots & & & \\ 1+x_ny_1 & 1+x_ny_2 & \ldots & 1+x_n y_n \\ \end{array} \right]$$ with $x_k, y_k$ being real numbers. I suppose that there must be some neat way of computing the determinant of this matrix. Those ones are annoying to me.

  • $\begingroup$ This appears to be singular, after doing several matrices out by hand and doing some symbolic calcs in Octave $\endgroup$ – costrom May 16 '16 at 20:30
  • 1
    $\begingroup$ @costrom this heavily depends on your choice of the $x_k$ and $y_k$ respectively $\endgroup$ – user190080 May 16 '16 at 20:34

The matrix can be expressed as the sum of two rank $1$ matrices, each of which is the outer product of vectors $$M=\mathbf{11}^T+\mathbf{xy}^T$$ where $\mathbf{1}=[1,1,1,\cdots,1]^T\in\mathbb{R}^{n\times 1}$, $\mathbf{x}=[x_1,x_2,\cdots,x_n]^T\in\mathbb{R}^{n\times 1}$ and $\mathbf{y}=[y_1,y_2,\cdots,y_n]^T\in\mathbb{R}^{n\times 1}$.

Using the subadditive property of the rank, where $\text{rank}(A+B)\leq\text{rank}(A)+\text{rank}(B)$, we have $$\text{rank}(M)\leq\text{rank}(\mathbf{11}^T)+\text{rank}(\mathbf{xy}^T)\Rightarrow \text{rank}(M)\leq 2$$ Thus, only when $n=2$ the matrix can be full rank, and so potentially have a non-zero determinant, which is easily evaluated as $(1+x_1y_1)(1+x_2y_2)-(1+x_2y_1)(1+x_1y_2)$. However for $n\geq3$, the matrix will be rank deficient, in which case the determinant will be $0$.

  • $\begingroup$ This is impressively simple. Well done! (+1) $\endgroup$ – Semiclassical May 16 '16 at 21:08

Here are two approaches.

1) By inspecting the pattern of this matrix, one sees that it is of the form $A+x y^T$ where $A$ is the matrix of ones and $x,y$ are appropriate column vectors. The matrix determinant lemma then gives $\det(A+x y^T)=\det(A)+y^T \text{adj}(A)x$ where $\text{adj}(A)$ is the adjugate matrix of $A$. But the determinant of $A$ vanishes; moreover, if $n\geq 3$ then then so too does the the adjugate, so the determinant is equal to zero for $n\geq 3$.

2) If we imagine trying to compute this determinant by cofactors, we see that should be some degree polynomial of degree $2n$ in the variables $x_k,y_k$. Furthermore, said polynomial must be antisymmetric with respect to swaps of $x$- or $y$- variables. This suggests an answer of the form $\displaystyle A\prod_{i<j}(x_i-x_j)(y_i-y_j)$ where $A$ is some overall constant. To normalize this, we take $x_k=y_k=\omega^k$ where $\omega=e^{2\pi i/k}$. Then the matrix is circulant and for $n\geq 3$ has a zero eigenvalue, but the above polynomial doesn't vanish. Consequently both $A$ and the determinant itself must vanish if $n\geq 3$.


If the matrix is $M$, then


where $J$ is the vector of all ones. $X$ is an $n*n$ matrix that has $x_1,x_2,...,x_n$ on the diagonal and zeros elsewhere. $Y$ is an $n*n$ matrix that has $y_1,y_2,...,y_n$, with the same order, on every row.

$rank(J)=1, rank(Y)=1,rank(X)= n$

$rank(XY)\leqslant min(rank(X),rank(Y))=min(1,n)=1$



$rank(J+XY)\leqslant rank(J)+rank(XY)=2$

If $2<n$, then we can conclude

$rank(J+XY)\leqslant 2 < n$

So, the rank of $M=J+XY$ is less that $n$ and consequently, its determinant is zero. But the correctness depends on the assumption that $2<n$


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.