Consider the following proposition with its relative proof:

Let $k$ be an algebraically closed field of characteristic $0$.

a) If $L$ is a subfield of $k$, then every elements of $\operatorname {Aut} (L)$ extends to an element of $\operatorname{Aut} (k)$

b) $\operatorname{Fix}_k\left(\operatorname{Gal}(k/L)\right)=L$

Proof: $a)$ Let $S$ be a transcendence basis for $k/L$, then every element $\sigma\in\operatorname{Aut}(L)$ extends naturally to an element $\widetilde\sigma\in\operatorname{Aut}(L(S))$; since $k$ is an algebraic closure of $L(S)$, we use the isomorphism extension theorem to conclude that $\widetilde\sigma$ extends to an element of $\operatorname{Aut}(k)$.

$b)$ Obviously it is enough to prove that for every $x\in k\setminus L$ there is an element $\sigma\in\operatorname{Gal}(k/L)$ such that $\sigma(x)\neq x$. We distinguish two cases:

  1. $x$ is transcendental over $L$. Consider the field $L(x)$, then the assignment $x\mapsto-x$ induces a unique element $\sigma\in\operatorname{Gal}(L(x)/L)$ that moves $x$. By the point $a)$ this $\sigma$ extends to an element of $\operatorname{Gal}(k/L)$.
  2. $x$ is algebraic over $L$. Since in characteristic $0$ every irreducible polynomial is separable, if $f=\textrm{min}\left(x,L\right)$ then there exists in $k$ (remember that $f$ splits over $k$) a root $y$ of $f$ such that $x\neq y$. Let $M$ be the splitting field of $f$ over $L$ and look at the inclusions
    $$L\subseteq L(x)\subseteq M\subseteq k$$ $M$ is normal over $L$ and the canonical $L$-isomorphism $\sigma:L(x)\longrightarrow L(y)$ can be viewed as an immersion $\sigma: L(x)\longrightarrow k$. By the characterization of normal extensions there is an element $\tau\in\operatorname{Gal}(M/L)$ such that $\tau_{|L(x)}=\sigma$ (in particular $\tau(x)=y$), therefore by the point $a)$ $\tau$ extends to an element of $\operatorname{Gal}(k/L)$ which moves $x$.

Now my question: Is the point $b)$ of the proposition true also when the characteristic of $k$ is $p\neq 0$? The above proof uses the separability of all irreducible polynomials!

Edit: Moreover I'd like to know if there are shorter or more elegant proofs of the proposition.

Thanks in advance.

  • 1
    $\begingroup$ I don't think so. The following is not a counterexample, because the bigger field is not algebraically closed, but I bet it may be extended. Take $\alpha\in\mathbb{F_p}\setminus\mathbb{F}_p^p$, and consider $\beta\in\overline{\mathbb{F}_P}$ such that $\beta^p=\alpha$. Now take $L=\mathbb{F}_p$ and $k=L(\beta)$, the only automorphism of $k$ over $L$ is the identity, so the fixed field is $k$. What you need is some hypothesis to guarantee the separability of $k/L$: $\operatorname{char}(k)=0$ is only used to ensure it. $\endgroup$ Mar 19, 2014 at 23:50
  • $\begingroup$ @Giulio Every element of $\Bbb F_p$ is a $p$th power, perhaps you mean $\Bbb F_p(T)$? $\endgroup$
    – blue
    Mar 20, 2014 at 19:19
  • $\begingroup$ Uh, yes, sorry. Replace $\mathbb{F}_p$ with some field with characteristic $p$ and without $p$-roots. $\endgroup$ Mar 20, 2014 at 19:41

1 Answer 1


Let $k=\Bbb F_p(T)$, denote by $K$ its algebraic closure, and $G={\rm Aut}_k(K)$ the automorphisms of $K$ which fix $k$ pointwise. Let $l=\Bbb F_p(T^{1/p})$ be the splitting field of $X^p-T\in\Bbb F_p(T)[X]$. Since it is a splitting field it must be the case that $\sigma l=l$ for all $\sigma \in G$. An action $\sigma$ must send $T^{1/p}$ to a root of the polynomial $X^p-T=(X-T^{1/p})^p$, but this polynomial's only root is $T^{1/p}$, so $\sigma$ fixes $T^{1/p}$ and hence fixes $\Bbb F_p(T^{1/p})$ pointwise. Therefore, $\Bbb F_p(T^{1/p})$ is contained in the fixed field $K^G$, which means that $K^G=k$ is impossible. In fact, $\Bbb F_p(T^{1/p^\infty})\subseteq K^G$.

So we see why separability is important: inseparability of an element means a Galois action can't move it anywhere else (it has no conjugates), thus exhibiting a nontrivial element fixed by $G$.


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.