Let $V$ be a vector space over a field $F$ (not necessarily algebraically closed), and $T:V\rightarrow W$ be a $F$-linear map. If $c(T)$ and $m(T)$ denote the characteristic and minimal polynomial respectively, then I want to show that $c(T)$ and $m(T)$ have the same set of irreducible factors.

I know that the claim is true if $F$ is algebraically closed, because then the two polynomials have the same roots. The typical reasoning one gives when $F$ is not algebraically closed is: Take an algebraic closure of $F$, say $K$, and look at $m(T)$ and $c(T)$ in $K$ to conclude the proof. But as far as I know, the proof that any root of $c(T)$ is a root of $m(T)$ requires the fact that $V$ is a $K$-vector space. But we only know that $V$ is an $F$-vector space. I can't understand how to proceed after this.

  • 2
    $\begingroup$ You must assume $V=W$ ! $\endgroup$ – Georges Elencwajg Sep 6 '13 at 8:27

It is easy to see that any root in the (non-algebraically closed) field $F$ of the characteristic polynomial is also a root of the minimal polynomial; algebraic closure plays no role here. Such a root$~\lambda$ is a (true) eigenvalue with an actual eigenvector$~v$ in the original vector space: $T\cdot v=\lambda v$. But then $P[T]\cdot v=P[\lambda]v$ for any polynomial $P[T]$ of$~T$, an in particular $0=0\cdot v= m[T]\cdot v=m[\lambda]v$, and since $v\neq0$ it must be that $m[\lambda]=0$ so $\lambda$ is a root of the minimal polynomial $~m$.

However this does not show that other irreducible factors $p$ of the characteristic polynomial divide the minimal polynomial. For that the easiest argument is to extend the field to $K$ so that $p$ admits a root$~\lambda$ in$~K$ (there is no need for a further extension); choosing a matrix for $T$ it defines an endomorphism of $K^n$ with characteristic polynomial $c$ and (somewhat less obviously) minimal polynomial$~m$. Then by the above $\lambda\in K$ is a root of$~m$, which therefore over$~F$ must have a factor$~p$ (as this is the minimal polynomial of$~\lambda$ in the sense of field theory). However, one can give a proof avoiding field extensions, though it is a bit more work. I gave one in (the last part of) this answer.


In basis-free jargon, what you do is consider the $K$-linear map induced by $T$ on the tensor product $V \otimes K$. If you don't know about tensor products yet, this amounts to using a basis for $V$ to represent $T$ as an $n \times n$-matrix $M$ (where $n$ is the dimension of $V$) and then consider the $K$-linear map on $K^n$ represented by $M$.


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.