Suppose that you've got an $n \times n$ irreducible matrix $A$ with strictly positive real entries and eigenvalues $\lambda_i$, $i=1,...,m$, arranged so that $|\lambda_1| > \cdots > |\lambda_m|$. Suppose that $D$ is a diagonal matrix with strictly positive real entries, and define $$ \tilde{A} := D^{-1} A' D. $$ What can be said about the eigenvalues of the matrix $A \tilde{A}$? Note that $\tilde{A}$ is self adjoint with respect to the inner product $\langle \cdot, \cdot \rangle_D$ given by $$ \langle f, g \rangle_D = f' \overline{g} D $$ and that one could build a linear isometry between this inner-product space and the one with the usual Euclidean inner product, under which $A\tilde{A}$ would be Hermitian (real symmetric in fact).

My sense is that the eigenvalues of $A\tilde{A}$ would be the squares of the absolute values of the eigenvalues of $A$. Am I right? And does it share the eigenvalues of $\tilde{A} A$?



The eigenvalues of $AB$ and $BA$ are always the same.

As for the eigenvalues of $A \tilde{A}$ being the squares of the absolute values of the eigenvalues of $A$, have you tried some examples? Try e.g. $A = \pmatrix{1 & 2\cr 1 & 1\cr}$ and $D = \pmatrix{2 & 0\cr 0 & 1\cr}$.

  • $\begingroup$ Thanks Robert. I tried to remove the specifics of the problem I was working on, hoping something more general than I needed was true, but now I realize that I said something silly. This was helpful for knocking me back into reality though. $\endgroup$ – user139388 Apr 12 '14 at 1:37

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.