1
$\begingroup$

Prove that there exists some $k $ with $1\leq k\leq n$ such that $C_{kk}\neq 0$ where $C_{kk}$ is the cofactor of $a_{kk}$ where $A= (a_{ij})$ is a symmetric matrix of rank $n-1.$

First approach: Suppose not. Let $\text{rank}(A) = n-1.$ Then all the diagonal entries of $B =\text{adj}A$ are $0$. Since $AB= 0$ (as $A\times \text{adj}A= \det A \times I_n$ and $\det A= 0$ as $A$ is not full rank), $\text{rank} AB=0$ so by Sylvester's rank inequality $(\text{rank}A + \text{rank}B -n \leq \text{rank}AB)$ it follows that $\text{rank} B \leq 1.$ If $\text{rank} B=0$ then $B=0$ contradicting the fact that since $\text{rank}A = n-1$ there is some $(n-1)\times (n-1)$ submatrix with non-zero determinant. Therefore $\text{rank} B=1.$ Since $A$ is symmetric, $B$ is symmetric so $B$ must be of the form

$B= \left(\begin{matrix}0&&x_1 &&\cdots &&x_{n-1}\\x_1&&0&&\cdots &&x_{n-2}\\\vdots&& &&\ddots &&\\x_{n-1}&&x_{n-2}&& \cdots&&0 \end{matrix}\right)$

Since $\text{rank}B =1$ we can assume WLOG that rows $2$ to $n$ are scalar multiples of row $1$ but this will imply that $x_i=0$ for $i=1, \ldots n-1$ implying that $B=0$ which we already showed leads to a contradiction.

My first question: is this approach right?

My second question is: can the second approach below be salvaged somehow?

Second approach: Since $A$ is real and symmetric it is orthogonally diagonalisable and since its rank is $n-1$ it has $0$ as an eigenvalue with algebraic (and hence geometric) multiplicity $1$. If $A = PDP^{T}$ where $D =\text{diag}(d_{11}, \ldots, d_{nn})$ then there is some $k$ such that $d_{kk}=0.$ Can we say something about the rank of $A'$ where $A'$ is obtained from $A$ by deleting the $k^{th}$ row and column? (Again we know it is orthogonally diagonalisable but can we say for sure that $A'$ does not have $0$ as an eigenvalue? This might be a little farfetched, but can the interlacing inequalities of eigenvalues of principal submatrices of Hermitian matrices be applied here?)

$\endgroup$
1
$\begingroup$

Your first proof looks fine to me. I am not sure about the second one, but you may follow a third approach: the sum of all principal $(n-1)\times(n-1)$ minors of $A$ is $\sum_i\prod_{j\ne i}\lambda_j$.

$\endgroup$

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.