I have two polynomials, $p$ and $q$, and a square matrix $A$. I also have that $q(A)$ is invertible. I want to prove the following statement.

Suppose that $\mu$ is an eigenvalue of $p(A)q(A)^{-1}$. Then $A$ has an eigenvalue $\lambda$ such that $\mu =\frac{p(\lambda)}{q(\lambda)}$.

I have gotten this far:

Since $q(A)$ and $p(A)$ commute, $p(A)q(A)^{-1} = q(A)^{-1}p(A)$. If $\mu$ is an eigenvalue of $p(A)q(A)^{-1}$, then I can write

$$p(A)q(A)^{-1}v = \mu v$$ for some $v \neq 0$. Hence, $q(A)^{-1}p(A)v = \mu v$, which would then imply $r(A)v = 0$ where $r(x) = p(x) -\mu q(x)$ is another polynomial. So $(0,v)$ is an eigenpair for $r(A)$. Not sure how to carry this forward. I thought about minimal polynomials but I don't think they are of any use here.

Maybe I could also use the fact that I can find two matrices in Jordan form, $P$ and $Q$, and an invertible matrix $T$, such that $p(A) = TPT^{-1}$ and $q(A) = TQT^{-1}$ thanks to $p(A)$ and $q(A)$ commuting. This gives right to the fact that

$$QPw = PQw = \mu w$$ where $w = T^{-1}v$. The fact that $(\mu,w)$ is an eigenpair for both $QP$ and $PQ$ seems noteworthy but I don't see how it is useful here.

I would very much appreciate seeing different kinds of answers; some based on results for finite dimensional spaces and some based on operator theory or whatever abstraction level is appropriate.

  • $\begingroup$ Do you mean $q (A^{-1}) $ or $[q (A)]^{-1} $? $\endgroup$ – user May 23 '18 at 18:43
  • $\begingroup$ @user The second one. $\endgroup$ – Calculon May 23 '18 at 18:44
  • $\begingroup$ How do you know that $p(A)$ and $q(A)$ commute? $\endgroup$ – Sean Roberson May 23 '18 at 19:19
  • $\begingroup$ @SeanRoberson $A$ commutes with itself and all its powers. $\endgroup$ – Calculon May 23 '18 at 19:20

Assume you are working over the complex numbers. The spectrum $\sigma(A)$ consists of the eigenvalues of $A$. The spectral mapping theorem states that $$ \sigma(p(A))= p(\sigma(A)). $$ For the given problem, $\mu$ is an eigenvalue of $p(A)q(A)^{-1}$ iff $\mu I-p(A)q(A)^{-1}$ is singular, which is true iff $\mu q(A)-p(A)$ is singular because $q(A)$ is assumed to be invertible. This in turn is equivalent to knowing that $\mu q(\lambda)-p(\lambda)$ maps one of the eigenvalues $\lambda$ of $A$ to $0$, by the Spectral Mapping Theorem. So $\mu q(\lambda)-p(\lambda)=0$ for some eigenvalue $\lambda$ of $A$. Because $q(A)$ is invertible, then $q(\lambda)\ne 0$ for every eigenvalue of $A$. Hence, $$ \mu = \frac{p(\lambda)}{q(\lambda)} $$ for some $\lambda\in\sigma(A)$.


Let $B=q(A)$. Then $B^{-1}=h(B)$ for some polynomial $h$. (See a proof here.)

Therefore, $q(A)^{-1}=h(q(A))$ and $p(A)q(A)^{-1}=f(A)$ for $f(x)=p(x)h(q(x))$, also a polynomial.

The result is now well known.

You need to assume that the scalar field is algebraically closed or at least that $A$ is triangularizable.

  • $\begingroup$ Thank you. Your first remark follows from Cayley-Hamilton, right? $\endgroup$ – Calculon May 24 '18 at 9:49
  • 1
    $\begingroup$ @Calculon, yes, but not necessarily. See math.stackexchange.com/a/2785699/589. I've edited my answer. $\endgroup$ – lhf May 24 '18 at 11:45

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.