MathOverflow is a question and answer site for professional mathematicians. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Suppose that $B$ is a Hermitian matrix with one known eigenpair $(\lambda,v)$. (assume its the smallest or largest pair, if you like). Form the rank one update $B+\rho bb^{T}$.

Now I'm interested in a (possibly approximate) formula for the updated eigenpair $(\widetilde{\lambda},\widetilde{v})$, that depends only on the entries of $B$ and $\lambda,v$. There are such formulas derived in a classic 1978 paper by Bunch, Nielsen and Sorensen (building on an earlier classic work by Golub), but they involve the full eigenbasis of $B$.

Any pointers?

P.S. (Assume $\lambda,\widetilde{\lambda}$ are simple, to keep things simple).

share|cite|improve this question
By $A$ do you mean $B$? Edit: You will probably also want to know quantitatively how `simple' $\lambda$ is, for example something like the distance to its nearest eigenvalue neighbor. – tergi Aug 6 '12 at 23:24
I changed $A$ to $B$, thanks! I think I can obtain this information - what next? I eagerly await further enlightenment! :) – Felix Goldberg Aug 6 '12 at 23:50

Here is a first-order approximation. I'll write $\tilde{v} = v + \rho w + O(\rho^2)$ and $\tilde{\lambda} = \lambda + \rho \mu + O(\rho^2)$. We may assume $\|v\| = \|\tilde{v}\| = 1$, so $v^T w = 0$. If $\lambda$ is a simple eigenvalue of $B$, $B - \lambda I$ is invertible on the orthogonal complement $v^\perp$ of $v$. Taking $(B + \rho b b^T) \tilde{v} = \tilde{\lambda} \tilde{v}$ to first order in $\rho$, we have $$ B w + b b^T v = \mu v + \lambda w $$ Let $c = b - (b^T v) v$ which is orthogonal to $v$, so that this equation splits into $$ B w + (b^T v) c = \lambda w \ \text{and} (b^T v)^2 v = \mu v$$ Thus $\mu = (b^T v)^2$ and $w = - (b^T v) (B - \lambda I)|_{v^\perp}^{-1} c$.

Higher-order approximations along the same lines should be possible.

share|cite|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.