I would like to compute the kernel of the following matrix.

Let $W \in \mathbb{R}^n$ have only $1$'s on its main diagonal and $w_{i,j} \in (-1,0)$ (open interval) off the main diagonal, so that the columns sum up to $-1$. I would like to show that

$$\dim \ker W = 1$$

If this is not true, I ask for a counterexample.


It is clear that $\dim\ker W \geq 1$, since adding for all $i=1,...,n-1$ the $i$-th row to the last row of $W$ kills the last row.

Consider the Minor $W_n$, which is $W$ without the last column and the last row. It is left to show that $W_n$ has $\ker = 0$. Consider the decomposition (set $\tilde W_n = W_n - I_{n-1})$ $W_n = I_{n-1} + \tilde W_n$. Now $W_n x=0$ implies $\tilde W_n x = -x$, e.g., $\tilde W_n$ has eigenvalue $-1$.

So the goal is to show that $\tilde W_n$ has eigenvalue $-1$. I wanted to estimate the spectral raduius and show that is (in absolute value) strictly smaller $1$. This would finish the proof, but I dont know how to do that

I need help here or a different approach :)

  • 1
    $\begingroup$ I think you mean that the sum of the off-diagonal entries in each column is $-1$, so that the entries in each columns sum up to zero. $\endgroup$ – daw Apr 29 at 8:24

The minor $W_n$ has diagonal entries $1$, and the sum of the off-diagonal entries in each columns is strictly larger than $-1$ (because deleting the last row of the matrix increases this sum). Then Gershgorin's theorem tells you that the eigenvalues of the remaining matrix are all non-zero.

If the off-diagonal entries would allowed to be taken from $[-1,0]$ then $$ \pmatrix{ 1 & -1 &0&0\\ -1 & 1 &0&0\\ 0&0&1&-1\\0&0&-1&1} $$ would be a counter-example.

  • $\begingroup$ its not a counter-example... since i asked off-diagonal entires to be not 0 and not -1. However i solved my own question. The answer is to used diagonal dominance of the Minor matrix $W_n$ (kill the last column and last row) to see that $rang$ of $\tilde W$ is at least n-1. $\endgroup$ – mathemagier May 10 at 16:30
  • $\begingroup$ I did not claim this to be a counterexample... $\endgroup$ – daw May 11 at 7:32

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.