Given some upper Hessenberg matrix $H \in R^{n \text{x} n}$, i know how to find an orthogonal matrix which is a product of Givens rotations such that $P^THP$ is also upper Hessenberg, but I'm not sure how to find $P$ so that

$P^THP=\begin{bmatrix}\lambda & w^T \\ 0 & H_1\end{bmatrix}$

where $H_1 \in R^{(n-1) \text {x} (n-1)}$ and $\lambda$ is the eigenvalue of $H$ and its corresponding eigenvector is $x$.

  • $\begingroup$ the first column of $P$ must be the eigenvector of $H$ corresponding to the eigenvalue $\lambda$. the rest of the columns are chosen to make $P$ orthogonal. $\endgroup$ – abel Apr 20 '15 at 0:18

Since you know $\lambda$ you can apply standard Hessenberg QR algorithm with shift and deflate H.

Standard steps would result in

$P^THP = \begin{bmatrix}H_1 & w\\ 0 &\lambda\end{bmatrix}$

However I think if we apply the $QR$ step from the left we can get the desired form for the solution

(Another solution)

Say $(\lambda,x)$ is the eigenvalue,eigenvector pair for $H$.
I guess another way to solve this is to apply Givens rotations on the $(n-1,n),(n-2,n-1),\cdots,(1,2)\space$ entries of $x$ like so

$G_{(1,2)}\cdots G_{(n-2,n-1)}G_{(n-1,n)}Hx = \lambda G_{(1,2)}\cdots G_{(n-2,n-1)}G_{(n-1,n)}x$

$\implies GHx = \lambda Gx\quad\quad$ where $G = G_{(1,2)}\cdots G_{(n-2,n-1)}G_{(n-1,n)}$

$\implies GHG^TGx = \lambda Gx$

$GHG^T$ is a unitary transform and preserves form via the Implicit-Q theorem and $\lambda Gx = \lambda e_1$


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.