Eigenvalues of arrowhead-like Hermitian matrix

146 Views Asked by At

Let $\lambda \in \mathbb C, \ell \in \mathbb R, x \in \mathbb C^{n},$ and $$\mathcal H=\begin{pmatrix}\lambda I_n & x\\ x^{\ast} &\ell\end{pmatrix}.$$ Prove that $\lambda$ is an eigenvalue of $\mathcal H$ with multiplicity at least $n − 1$. What are the other two eigenvalues?

Attempt: I tried to use Cauchy Interlacing Theorem but I'm getting nowhere. Any hint and help would be much appreciated!

2

There are 2 best solutions below

0
On

The Cauchy interlacing theorem is exactly the right way to go. In this case, the theorem guarantees that the eigenvalues $\lambda_1 \leq \cdots \leq \lambda_{n+1}$ of $\mathcal H$ satisfy $$ \lambda_1 \leq \lambda \leq \lambda_2 \leq \lambda \leq \cdots \leq \lambda \leq \lambda_n \leq \lambda \leq \lambda_{n+1}, $$ which implies that we have $\lambda_2 = \cdots = \lambda_n = \lambda$.

Now, the only step left is to compute the other two eigenvalues. One way to do this is as follows: it is equivalent to consider the eigenvalues of the matrix $$ A = \mathcal H - \lambda I_{n+1} = \pmatrix{0&x\\ x^* & \ell - \lambda}. $$ Note that $\mu$ is an eigenvalue of $A$ if and only if $\mu + \lambda$ is an eigenvalue of $\mathcal H$. Notably, $A$ is a rank-2 matrix. With that in mind, there are two nice strategies that you can use here to compute the non-zero eigenvalues:

Option 1: Calculate $\operatorname{tr}(A)$ and $\operatorname{tr}(A^2)$, then use the relationship between the trace of a matrix and the matrix's eigenvalues to solve for the two non-zero eigenvalues of $A$. We find that $$ \operatorname{tr}(A) = \ell - \lambda, \quad \operatorname{tr}(A^2) = (\ell - \lambda)^2 + 2\|x\|^2. $$

Option 2: Compute a "rank-factorization" $A = BC$, i.e. a factorization in which $B,C$ have sizes $(n+1) \times 2, 2 \times (n + 1)$. Then, note that the non-zero eigenvalues of $A$ are equal to the non-zero eigenvalues of the $2 \times 2$ matrix $CB$. One such factorization is given by $$ B = \pmatrix{x & 0\\0 & 1}, \quad C = \pmatrix{0_{1 \times n} & 1\\ x^* & \ell - \lambda}. $$

0
On

$$\det \left( s {\bf I}_{n+1} - {\bf H} \right) = \det \begin{bmatrix} (s - \lambda) \, {\bf I}_n & - {\bf x}\\ - {\bf x}^{\ast} & s - \ell\end{bmatrix} = \cdots = (s - \lambda)^{n-1} \left( (s - \lambda) (s - \ell) - \| {\bf x} \|_2^2 \right)$$