MathOverflow is a question and answer site for professional mathematicians. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

My question is as follows:

Let $k$ be a field of characteristic zero and let $\overline{k}$ be an algebraic closure. Let $V$ be an algebraic variety over $k$ and let $\overline{V}=V \times_k \overline{k}$. Suppose that $\overline{V}$ admits the structure of an algebraic group. Then is $V$ itself a principal homogeneous space for some algebraic group?

Note that the answer to my question is yes when $V$ is projective; in this case it is well-known that such a $V$ is a principal homogeneous space for its Albanese variety. So I am really interested in the case where $V$ is affine. Here I am not aware of an analogue of the Albanese variety for linear algebraic groups.

Even if the answer is no in general, I would still be interested in some positive results for special cases, e.g. for reductive groups or semisimple groups.

share|cite|improve this question
Since the structure of an algebraic group can be defined by writing down finitely many polynomials, it will always show up over some finite field extension, so a twist will show up as a map $Gal \to Aut(G)$ with finite image. I think this will descend to a homogeneous space if and only if $Lie(G)$ inside $H^0(G,T_G)$ is an invariant subspace under the action of this group. We can check this in some basic cases like the torus, but I'm not sure in general. – Will Sawin May 24 '13 at 16:38
I should probably also mention that I have come across proofs of the fact that the answer is also yes when $\overline{V}$ is an algebraic torus, see e.g. Skorobogatov - Torsors and rational points - Lemma 2.4.4. The proof here however makes key use of the fact that $\overline{V}$ is an algebraic torus, namely it uses the fact that one can recover an algebraic torus from its character group. – Daniel Loughran May 24 '13 at 17:03
I don't think that restricting to simple groups will help. Some simple groups are isomorphic as a variety of a product of a vector space with another space. So if you had a strange form of a vector space, you could take a product to get a candidate strange form of the simple group. – Ben Wieland May 29 '13 at 21:15

Here is a counterexample with $V$ finite.

Suppose that $k'$ is a separable extension of degree $5$ of $k$, and set $V = \mathop{\rm Spec} k \sqcup \mathop{\rm Spec} k'$. Suppose that $V$ is a homogenous space; then it is a group scheme, since $V(k) \neq \emptyset$.

If $\overline k$ is the separable closure of $k$, then $V_{\overline k}$ is the disjoint union of six copies of $\mathop{\rm Spec} \overline k$, and these copies form a group of order $6$. The Galois group of $k$ acts on $V_{\overline k}$ by permuting the non-identity component transitively. But a group of order $6$ contains elements of order $2$ and $3$, so this is impossible.

I am convinced that one can get geometrically connected examples using forms of $\mathbb G_{\rm a}^n$.

share|cite|improve this answer
This works whenever the Galois group, as a subgroup of $S_n$, does not normalize any simply transitive subgroups. – Will Sawin May 25 '13 at 0:01
Thanks Angelo. If anybody could say anything at all about geometrically connected examples (except for tori) I would be most interested. – Daniel Loughran May 25 '13 at 9:26

I think the answer should be "no" if $V$ has finite order automorphisms as a variety which are not conjugate to group automorphisms. Principal homogenous spaces for $G$ are described by $H^1(Gal(\bar{k}/k), G(\bar{k}))$. I believe groups over $k$ which be come isomorphic to $G$ over $\bar{k}$ are described by $H^1(Gal(\bar{k}/k), Out(G))$ and principal homogenous spaces for groups of this sort are described by $H^1(Gal(\bar{k}/k), Aut_{grp}(G))$. On the other hand, spaces isomorphic to $G$ as a variety are $H^1(Gal(\bar{k}/k), Aut_{var}(G))$.

So a good place to look should be algebraic groups $G$ whose automorpism group as varieties is much larger than their automorphism group as groups. The automorphism group of $\mathbb{G}_a^n$ is huge, but most of it is elements have infinite order and Galois cocycles have finite image. So I don't have an example to propose yet.

share|cite|improve this answer
(For $A^1$ it is easy and for $A^2$ it follows (not easy, though!) from the amalgamated product structure of the automorphism group, that finite order automorphisms are linearizable. There is an example by Asanuma of a non-linerlizable automorphism in higher dimension, but in positive characteristic.) – Mariano Suárez-Alvarez May 24 '13 at 17:22
Abelian varieties have many finite-order automorphisms of the underlying variety that are not group automorphisms. Why does your argument fail for that case? (To be sure, I am quite convinced that what you say must be close to true. I just don't fully understand it yet.) – René May 24 '13 at 17:28
All automorphisms of an abelian variety are of the form $x \mapsto Ax + b$ where $A$ is an automorphism of the abelian variety and $b$ is a point of the abelian variety. If $A - \mathrm{Id}$ is invertible, at least, that's conjugate (in the group of variety automorphisms) to $x \mapsto Ax$. I haven't actually worked through the Galois cohomology I sketch above, but I suspect that things conjugate to group automorphisms won't do it. It isn't clear to me what happens if $A - \mathrm{Id}$ is not invertible. – David Speyer May 24 '13 at 19:46
I think any $Ax+b$ is clearly fine. In the nonabelian case, an automorphism composed with a left/right multiplication will be fine. You can mod out by the translation subgroup to get the action by automorphism, which gives a twist of the group. But then the action of that twist on your variety is Galois-equivariant, by construction, so your variety is a homogeneous space. – Will Sawin May 24 '13 at 21:16
I think forms of $G$ use not $Out(G)$, but $Aut(G)$; and homogeneous spaces over forms of $G$ use $G\rtimes Aut(G)$ (aka $(G\times G)/\Delta(Z(G))$, I think). As a check, $Out(SL_2)$ is trivial, but there are forms of $SL_2$. – Ben Wieland May 25 '13 at 21:34

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.