I've been thinking about my previous question a bit more, and I'm afraid I still don't quite understand. See: Can the natural embedding $K\to K[X]/(f)$ be extended to form an isomorphism $L/K\to K[X]/(f)$?

The theorem I'm trying to prove is as follows: given a field $K$ and an irreducible polynomial $f\in K[X]$, there always exists a (finite) extension $L$ of $K$ such that $L$ contains at least one root of $f$.

Most proofs that I've seen on the internet and in algebra textbooks run like this: $K$ can be embedded in $L'=K[X]/(f)$, which is a field (because $(f)$ is a maximal ideal). So $L'$ is isomorphic to a field extension $L$ of $K$. Then define $\alpha=X\pmod{f}$, and substitute $\alpha$ in $f$. This gives you $0$, so $\alpha$ is a root of $f$.

I get the basic idea, sure enough. But when I try to write a fully rigorous proof, I get stuck.

First of all, literally substituting $\alpha$ in $f$ is impossible, because $\alpha$ is a coset, and not an element of $L$. So, we have to replace the coefficients $a_i$ of $f$ by the corresponding cosets $\overline{a_i}$. This works fine.

But how on Earth do we know for certain that this new polynomial $\sum_i\overline{a_i}X^i$ (which is in $L'[X]$) corresponds, as far as roots are concerned, to the old $f$ (which is in $K[X]$)?

It seems that all proofs I've seen so far implicitly assume that there is an isomorphism $L\to L'$ such that $a_i\mapsto\overline{a_i}$ for all $a_i\in K$. But this is highly non-trivial. Or am I missing something?


The field generated by the companion matrix of $f$ over $K$ gives you the desired extension.

  • $\begingroup$ Thanks for offering an alternative line of attack. But the main point is that I want to know why the traditional proofs (with $X\pmod{f}$) work when you look through the `handwaving'. I just don't see it. $\endgroup$ – Kim Fierens Sep 3 '14 at 18:04
  • $\begingroup$ The extension I describe is essentially isomorphic to the $K[X]/(f)$ construct. The isomorphism is given by the map $M^n \leftrightarrow X^n mod (f)$ (with $0\leq i <\deg(f)$) essentially because $f$ is the mininimal polynomial of $M$, or otherwise put $f(M)=0$. "M" is a "root" of $f$, not really a number, but an algebraic entity nevertheless. The embedding of $K$ is done by associating $k \in K$ with the scalar matrix with $k$'s on the diagonal. $\endgroup$ – Marc Bogaerts Sep 3 '14 at 18:28

The elements of the quotient ring are of the form $p(x)+(f)$ where $p(x) \in K[x]$. And we identify $a$ with $a+(f)$ for $a \in K$ (this isnt a problem is it ?) Now if $q(y) \in K[y] $ and $\overline{q}(y)$ the image in $L[x]$ then $$\overline{q}(p(x)+(f))=q(p(x))+(f)$$

Now this for $q=f$ and $p(x)=x$, this gives


So this shows that $x+(f)$ is a root of $\overline{f}$. Under the identification of $K$ with its image $f$ and $\overline{f}$ correspond.

  • $\begingroup$ My problem is that the identification, let's say $\phi:K\to K[X]/(f)$, of $a$ with $a+(f)$ isn't an isomorphism between the extension $L$ (of $K$) and $K[X]/(f)$. It's only an embedding. So how do we know for sure whether there is an isomorphism $\Psi:K[X]/(f)\to K$ that is compatible with $\phi$? (By compatible I mean: $\Psi^{-1}|_K=\phi$, so that $f(\Psi(\overline{x}))=0\in K$.) $\endgroup$ – Kim Fierens Sep 3 '14 at 19:12
  • $\begingroup$ $\phi$ is an injective homomorphism, so $K[x]/(f)$ is an extension of $\phi(K)$ and $K$ is isomorphic to $\phi(K)$. $\endgroup$ – Rene Schipperus Sep 3 '14 at 19:31
  • $\begingroup$ Yes, up until that point it's perfectly clear to me. But where it gets hazy is this: suppose that $\alpha=\overline{x}$ is our desired root (you and the algebra books say it is). Then there should be an isomorphism $\Psi:K[X]/(f)\to K$ such that $f(\Psi(\alpha))=0$. Right? But at the same time we want that $\Psi(\overline{a_i})X^n+\cdots+\Psi(\overline{a_0})=a_iX^n+\cdots+a_0$. In other words: $\Psi^{-1}|_K =\phi$. It is completely non-obvious to me why such a $\Psi$ should exist. $\endgroup$ – Kim Fierens Sep 3 '14 at 19:46
  • $\begingroup$ There is no such isomorphism $\Phi$. The property that $\alpha$ is a zero is as explained in by answer. $\endgroup$ – Rene Schipperus Sep 3 '14 at 19:51
  • $\begingroup$ I'm very sorry. You've obviously done your best to explain it to me, and I still don't understand. This must be due to some kind of incompetence or blind-spot on my part. I will upvote your answer. $\endgroup$ – Kim Fierens Sep 3 '14 at 19:56

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.