eigenvalues of sum of matrices: $A$+block matrix are strictly less than eigenvalues of $A$+identity

268 Views Asked by At

Let $A$ whose sum for rows is 0, can I prove that the $ \lambda_i \left ( \begin{bmatrix} I & 0\\ 0 & 0 \end{bmatrix} +A\right ) $ are strictly less than $\lambda_i \left ( \begin{bmatrix} I & 0\\ 0 & I \end{bmatrix} +A\right )$?

Any help would be greatly appreciated .

I know that $\lambda_i \left ( \begin{bmatrix} I & 0\\ 0 & I \end{bmatrix} +A\right )=1+\lambda_i(A)$ but I can't see any relation with $\lambda_i \left ( \begin{bmatrix} I & 0\\ 0 & 0 \end{bmatrix} +A\right )$.

1

There are 1 best solutions below

1
On

I think we can use the following inequality due to Weyl. Let $A$ and $B$ be Hermitian matrices with eigenvalues $\lambda_i$ and $\mu_i$, respectively (indexed in descending order so that $\lambda_1$ and $\mu_1$ are the top eigenvalues). Let $\nu_i$ denote the eigenvalues of $A+B$, indexed in the same way. Then: $$\nu_{i+j+1} \leq \lambda_{i+1} + \mu_{j+1}$$ for all $i$ and $j$ for which both sides of the inequality makes sense.

In your case, let $A$ be any Hermitian $2n \times 2n$ matrix and let $B = (I_n \oplus 0_n)$. Setting $j = 0$ in Weil's inequality, we get: $$\nu_{i+1} \leq \lambda_{i+1} + \mu_1 = \lambda_{i+1} + 1$$ The right-hand side, as you observed, is the eigenvalue of $A + I_{2n}$.

However, you wanted the inequality to be strict. I unfortunately don't know the full analysis of the equality case of Weyl's estimates; all I know off the top of my head is that $\nu_1 = \lambda_1 + \mu_1$ if and only if $A$ and $B$ share a principal eigenvector. This is clearly ruled out by your condition that the row sums vanish, but I don't know what to say about the other eigenvalues.