1
$\begingroup$

Take two matrices $A,B \in \mathbb{R}^{n\times n}$. I am considering the generalized eigenvalue problem ($\gamma \in \mathbb{R}, \phi \in \mathbb{R}^n$)

$$A\phi = \gamma B\phi.$$

Is there a systematic way of modifying the matrix $B$ such that $\phi$ acquires desired properties?

I have a specific example in mind: let $n=4m$ and require $\phi$ to be in the m-fold product of the unit sphere in $\mathbb{R}^4$, i.e. when $\phi$ is chopped into blocks of 4 entries, these vectors each have Euclidean norm 1.

For a fixed $A\in GL(n)$, can one construct a diagonal matrix $B$ such that there exists an eigenpair $(\gamma,\phi)$ with the desired property on $\phi$?

$\endgroup$
3
  • $\begingroup$ Your exact goal is unclear to me. I presume you want to find the pairs $(\phi,\gamma)$ that solve your equation, but do you want to do it for all $A$ and $B$? Or for some nice fixed $B$ and all $A$? What modifications of $B$ do you allow? I assume "replace $B$ with $I$" is not the kind of systematic modification you have in mind. $\endgroup$ Aug 26, 2014 at 10:51
  • 1
    $\begingroup$ @JoonasIlmavirta Wiki: generalized eigenproblem should make the setting clearer and answer your first two questions. I second the third one though: what kind of modifications? What kind of properties? This looks too general; can you be more specific or make an example? $\endgroup$ Aug 26, 2014 at 11:03
  • $\begingroup$ I added a specific example with a rather complicated condition on $\phi$. $\endgroup$
    – madison54
    Aug 26, 2014 at 11:34

1 Answer 1

2
$\begingroup$

(Probably this is not what you are asking for, but it might be a cue to modify your question in the right direction.)

If you fix $A$, $\gamma$, $\phi$, then what you get is a linear system in the entries of $B$.

If $B$ has to be diagonal, then you have $n$ equations and $n$ unknowns, and it is easy to see that it is solvable whenever $\gamma\neq 0$ and $\phi$ has no zero entries.

If $B$ has a more general structure, then you may end up with an underdetermined system, or with a nonlinear one, but there are algorithm for those as well.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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