Following this question, let $A$ and $B$ be square $n\times n$ matrices. It is a well known fact (e.g., 12.3 at [1]) that $AB$ and $BA$ have the same eigenvalues (spectra):

The proof is trivial for $\lambda=0$ in case it is an eigenvalue. If $v$ is an eigenvector of $AB$ which belongs to $\lambda\neq 0$ then $A(Bv)=\lambda v\neq\vec{0}$, and so $Bv\neq \vec{0}$. It is then easy to see that $Bv$ is an eigenvector of $BA$ which belongs to the very same $\lambda$.

While the proof is straightforward, why should one expect this to be true? I'd be glad to have a geometric explanation to this fact.

[1] Golan, Jonathan S., The linear algebra. A beginning graduate student ought to know, Dordrecht: Springer (ISBN 978-94-007-2635-2/pbk; 978-94-007-2636-9/ebook). xii, 497 p. (2012). ZBL1237.15002.

  • 2
    $\begingroup$ In the case that $A$ is invertible, the operation $A^{-1}(AB)A = BA$ can be interpreted geometrically as a change of basis. This doesn't quite account for the case in which both matrices are singular, though. $\endgroup$ – Omnomnomnom Oct 9 '17 at 13:18

It is easy to see that $AB$ and $BA$ have the same non-zero eigenvalues. Say $ABv = \lambda v$, with $v\ne 0$. Then $B(AB)v = \lambda Bv$. Assume that $\lambda \ne 0$. Then from the first equation we get $Bv \ne 0$. So now we got $Bv$ a $\lambda$ eigenvector for $BA$. This trick unfortunately does not account for multiplicities.

  • $\begingroup$ Yes, that's exactly how Golan proves that $AB$ and $BA$ have the same spectra. But that does not answer my questions. $\endgroup$ – Shlomi A Oct 9 '17 at 20:33

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.