I am looking for a proof that two matrix share the same eigenvalues, even if they have different size. Similar matrices share the same eigenvalues and the similarity relationship is defined for $A,B,C \in K^{n,n}$ where $\operatorname{rank} (C)=n$ as:

$AC = CB \Leftrightarrow A = C^-1BC$

Now, if the matrices have different sizes, $A \in K^{(n,n)}$, $B \in K^{(m,m)} $ and $C \in K^{(m,n)}$ where $n>=m$ and $\operatorname{rank} (A)=m$. It can also be stated that


And that all eigenvalues for $B$ are eigenvalues for $A$.

I just can't come up with a way to map them to a common diagonal matrix the same way you would do if $C$ is invertible. I have tried using the pseudo inverse of $C$ and applying the SVD decomposition but I am not able to prove any relationship between eigenvalues.


As it stands, what you want to prove is false (e.g. when $C=0$). The followings are true, though:

  1. If $AC=CB$ and $C$ is a "tall" matrix with full column rank, then all eigenvalues of $B$ (counting multiplicities) are eigenvalues of $A$. You may use the rank of $C$ and the equality $(A-\lambda I)^kC=C(B-\lambda I)^k$ to prove that every generalised eigenpair $(\lambda, v)$ of $B$ gives rise to a generalised eigenpair $(\lambda, Cv)$ of $A$.
  2. If $AC=CB$ and $C$ is nonzero, then $A$ and $B$ shares at least one eigenvalue over the algebraic closure $\overline{K}$ of $K$. More specifically, from $AC=CB$, we get $A\pmatrix{C&0}=\pmatrix{C&0}\pmatrix{B\\ &D}$ for every $(n-m)\times(n-m)$ matrix $D$ over $\overline{K}$. Now the conclusion follows from the analogous result for square matrices and the infinite cardinality of $\overline{K}$ (which implies the infinitude of choices of the spectrum of $D$).
  • $\begingroup$ Thanks! Both the idea of extending it to square matrices and the reference really helped. $\endgroup$ – ebabio Nov 9 '16 at 21:24

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.