2
$\begingroup$

Let us consider two arbitrary Hermitian square matrices $\mathbf{A,B}$ with the same dimension. Given $\mathbf{v}$ the eigenvector associated to the maximum eigenvalue of $\mathbf{A}$:

Are there any conditions appart from $\mathbf{A} = \mathbf{B}$ or $\mathbf{B} = \mathbf{v}\mathbf{v}^H$ so that it can be ensured that $\mathbf{B}$ has the same eigenvector associated to the maximum eigenvalue?

Note that, in Dan Shemesh, Common eigenvectors of two matrices, Linear Algebra and its Applications, Volume 62, November 1984, Pages 11-18, ISSN 0024-3795, http://dx.doi.org/10.1016/0024-3795(84)90085-5. an arbitrary eigenvector was considered. I cross check the derivation, and I do not see how to promote the maximum eigenvector constraint.

Thank you in advance

$\endgroup$
7
  • 1
    $\begingroup$ The question is unclear. Are they real matrices? (Otherwise what do you mean by "largest" --- largest in absolute value?) What kind of condition would you want? $\endgroup$
    – Nik Weaver
    Jun 29, 2015 at 12:14
  • $\begingroup$ Thank you Nik, I add the condition of A and B being Hermitian. I look for a more 'exhaustive' condition rather than the ones I metion in the question. $\endgroup$
    – mikitov
    Jun 29, 2015 at 12:19
  • 1
    $\begingroup$ One such condition is that (for normal matrices) $\| AB\| = \| A \| \cdot \|B\|$ (where obviously we take the $2-2$ norm). $\endgroup$ Jun 29, 2015 at 13:34
  • $\begingroup$ @DavidHandelman: maybe require $A$ and $B$ to be positive? I'm unsure whether the "maximum" eigenvalue is allowed to be negative ... $\endgroup$
    – Nik Weaver
    Jun 29, 2015 at 15:19
  • $\begingroup$ $A$, $B$ are only required to be Hermitian. With this, the eigenvalues are always real, then there are no worries about the ordering. $\endgroup$
    – mikitov
    Jun 29, 2015 at 15:56

1 Answer 1

3
$\begingroup$

This is true iff the largest eigenvalue of $A+B$ is the sum of that of $A$ and that of $B$.

$\endgroup$
2
  • $\begingroup$ Could you elaborate your response a little? $\endgroup$
    – mikitov
    Jun 30, 2015 at 9:04
  • $\begingroup$ For an appropriate interpretation of the question, at least. The OP said $v$ is "the" eigenvector associated to the maximum eigenvalue of $A$, but if the eigenspace is not 1-dimensional you may need to replace $v$ with another eigenvector. $\endgroup$ Jun 30, 2015 at 11:01

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.