Rank of a specific-type matrix

57 Views Asked by At

Consider an N by N matrix with all main-diagonal elements negative, all off-diagonal elements non-negative, and the sum of all N columns equal to a zero vector. Clearly, the rank of the matrix is less than N; how does one prove that it is exactly N-1?

In view of the comments below, I have to add the following condition: visualize a graph with N nodes and our N by N matrix indicating which of these are adjacent (when the matrix has a non-zero value) and which are not (zero value). Consider only matrices resulting in a connected graph.

Subsidiary (an easier question): how would the proof go if all off-diagonal elements were required to be positive?

As pointed out by the Answer below, rather than messing up with graphs, the extra condition can be simply stated as being irreducible.

1

There are 1 best solutions below

0
On BEST ANSWER

As pointed out in a comment, your claim cannot be universally true: if it is true for some matrix $A$, it will be false for $A\oplus A$.

It is true, however, if your matrix is irreducible (i.e., if it is not similar, via any permutation matrix, to a block triangular matrix). Call your matrix $A$. Let $B=kI-A$ where $k>0$ is sufficiently large, so that $B$ is entrywise non-negative. It is also irreducible because $A$ is irreducible. Therefore $\rho(B)$ is a simple eigenvalue of $B$, by Perron-Frobenius theorem. However, since $\mathbf1^TB=\mathbf1^T(kI-A)=k\mathbf1^T$, $\mathbf1$ is a positive left eigenvector of $B$ corresponding to the eigenvalue $k$. Hence $k$ is necessarily equal to $\rho(B)$. In turn, $k-\rho(B)=0$ is a simple eigenvalue of $A=kI-B$, meaning that the nullity of $A$ is $1$.