Let $F$ be a field. By a Galois algebra over $F$ I mean a finite etale extension, that is, a product $K = K_1 \times \cdots \times K_r$ of finite (separable) field extensions, of total degree $[K : F] = n$, equipped with a subgroup $G \subseteq \operatorname{Aut}_F K$ of $n$ linearly independent automorphisms. For example, $\mathbb{C} \times \mathbb{C}$ equipped with the set of four automorphisms $(z,w) \mapsto \{(z,w), (\bar{w},z), (\bar{z},\bar{w}), (w,\bar{z})\}$ is a cyclic Galois algebra over $\mathbb{R}$ of degree $4$.

I'd like to find a reference (or counterexample) to the following simple assertions:

  1. $K$ is a Galois algebra iff it is a product $K = K_1^r$ of copies of a field that is Galois over $F$, and the $G$-action permutes the $n$ ring maps from $K$ to $K_1$ simply transitively;
  2. If $H \leq G$ is a subgroup, then $K/K^H$ is also Galois under the natural $H$-action (more precisely, for each field factor of $K^H$, the portion of $K$ above it is Galois under the $H$-action);
  3. Also, if $H$ is normal, then $K^H/F$ is Galois under the natural $G/H$-action;
  4. If $E$ is any extension field of $F$, then $L \otimes E$ is Galois over $E$, with the $G$-action extended linearly;
  5. If $K$ and $K'$ are Galois algebras over $F$ with Galois groups $G$ and $G'$ respectively, then $K \otimes K'$ is Galois under the natural action of $G \times G'$.

A construction due to Bhargava embeds any finite etale extension of degree $n$ (in characteristic $0$) into an $S_n$-Galois algebra.

It seems that the theory of such algebras is similar in depth to, but not a trivial consequence of, Galois theory of fields (e.g. for part 2, how do we compute $[K^H : F]$?) We no longer have that every subalgebra of $K$ is the fixed field of some subgroup; but on the other hand, the elegant properties 4 and 5 are missing from the classical presentation (because the tensor product of fields need not be a field).

It could be that this is all a special case of Grothendieck's theory of finite etale covers of schemes, but I wouldn't like to dredge up Grothendieckian formalism while discussing mainly about elementary properties of number fields, with a bit of class field theory.

  • 4
    $\begingroup$ The notion you're missing is that of a $G$-torsor. For a finite group $G$ and ring $K$, an injective finite etale map $K \to E$ (concrete for $K$ a finite product of fields) is a $G$-torsor if $E$ is equipped with a $G$-action over $K$ so $G\times {\rm{Spec}}(E) \to {\rm{Spec}}(E)\times_{{\rm{Spec}}(K)} {\rm{Spec}}(E)$ defined by $(g,x)\mapsto (x, g.x)$ is an isomorphism, or equivalently $E \otimes_K E \to \prod_{g \in G} E$ defined by $a\otimes b\mapsto (ag(b))_g$ is an isomorphism. This gives geometric insight leading to quick easy proofs of 1--5 when $K$ is a finite product of fields. $\endgroup$
    – nfdc23
    Mar 3 '17 at 3:57
  • 3
    $\begingroup$ Your comment about tensor products of fields not generally being fields is precisely why the geometric language is more convenient than the algebraic one: it is easier to visualize keeping track of connected components than primitive idempotents. And yes, this is all a special case of the theory of finite etale covers of schemes. A real virtue of the geometric language is that the analogous constructions and facts for possibly disconnected finite-degree covering spaces of a connected topological space $X$ (with other nice properties if you wish) provide a very vivid way to "see" what is true. $\endgroup$
    – nfdc23
    Mar 3 '17 at 6:00
  • 2
    $\begingroup$ Is Bhargava's construction something other than assigning to any degree-$n$ finite etale cover $X \rightarrow Y$ the $S_n$-torsor $\underline{\rm{Isom}}_Y({\rm{Aut}}_{X/Y}, (S_n)_Y)$ (i.e., the Isom-scheme over $Y$ between two finite etale $Y$-groups of order $n!$, namely the Aut-scheme of $X$ over $Y$ and the constant $Y$-group $S_n$)? $\endgroup$
    – nfdc23
    Mar 3 '17 at 9:10
  • 1
    $\begingroup$ Thank you. I was considering changing the definition to one very similar to nfdc23's first comment, but I missed the connection with torsors. $\endgroup$ Mar 4 '17 at 17:43
  • 1
    $\begingroup$ @nfdc23 Bhargava's construction is defined for arbitrary finite covers and not just etale ones. In the etale case, it is equivalent to what you state, but it provides a good generalization of this to arbitrary covers. $\endgroup$
    – Will Sawin
    Jun 7 '17 at 0:59

Let $K$ be a separable algebraic closure of $k$. Then the functor sending an etale $k$-algebra $A$ to $Hom(A,K)$ is an equivalence to the category of finite sets with a continuous action of the Galois group of $k$ (baby form of Grothendieck's theory). There is an elementary exposition of this in Milne's notes on Fields and Galois theory. I think you can answer all your questions by looking at the sets with Galois action.


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.