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

I'm trying to understand the following basic property of the restriction of scalars:

Given an absolutely simple algebraic groups $G$ defined over a number field $k$, are there at most finitely (up-to $k$-isomorphism) many absolutely simple algebraic groups $H$ defined over $k$ with $$Res_{k/\mathbb Q}G\cong_{\mathbb Q} Res_{k/\mathbb Q}H?$$ (i.e. their restrictions are isomorphic over $\mathbb Q$)

I understand that it follows that $G$ and $H$ are $k$-forms of each other and good understanding of the possible forms over the completions of $k$ should lead to an answer. This is because there is a theorem by Borel and Serre that establish the finiteness of forms that agree at all but finitely many completions.

Therefore my vague intuition is that the answer is yes, but I can't find a precise proof. Thinking about it also directed me to an even more basic question:

Given $G$ and $H$, two forms of each other (i.e., $G$ and $H$ both defined over $k$, and are isomorphic over the algebraic closure $\bar k$.) Is $G$ and $H$ isomorphic over $k_v$ for all but finitely many valuations $v$ of $k$?

I think that a positive answer to the latter should imply a positive answer to the former, but I think the latter might have a negative answer...

I'll be a happy if someone can shed some light about it and\or guide me to the relevant results in the literature. Thanks!

share|cite|improve this question
For the latter question, take G=SL_3 and H=SU_3. They're isomorphic over the algebraic closure, but are only isomorphic at half of the completions, hence answer is no. – Peter McNamara Dec 2 '10 at 21:26
Indeed. But these are outer forms? (or maybe not? I'm quite confused). Is it also not true for inner forms? – Menny Dec 2 '10 at 21:44
[They are indeed outer forms] – Kevin Buzzard Dec 2 '10 at 21:45
Arithmetic & Galois actions on root data are irrelevant. First I give easy proof of something very general: for any field $K$, finite etale $K$-algebras $K'$ and $K''$, and ss gps $G'$ and $G''$ over $K'$ and $K''$ resp. with all fibers connd and abs. simple, every $K$-gp isom. R$_{K'/K}(G')\simeq$ R$_{K''/K}(G'')$ comes from a unique pair $(f,\alpha)$ consisting of a $K$-algebra isom $\alpha:K''\simeq K'$ and a gp isom $f:G'\simeq G''$ over Spec($\alpha$). Pf: By uniqueness and Galois descent, enough to check after base change to $K_s$. Then $K'$ and $K''$ are powers of $K$, so it's obvious! – BCnrd Dec 3 '10 at 6:38
OK, that being said, we now dispose of the original finiteness question using nothing subtle or arithmetic. Let $K'/K$ be a finite separable extension of fields and let $G'$ and $H'$ be two connected semisimple $K'$-groups that are abs. simple. Suppose the $K$-groups R$_{K'/K}(G')$ and R$_{K'/K}(H')$ are $K$-isomorphic. Previous comment shows $H'$ is an Aut($K'/K$)-twist of $G'$ (which is clearly "best possible"), and Aut($K'/K$) is a finite group. Hence, only finitely many possibilities for $H'$, and we've nailed down exactly which ones actually occur (e.g., if Aut($K'/K) = 1$ then $H'=G'$!) – BCnrd Dec 3 '10 at 6:43

For the question in the comments, Kevin is right: every connected reductive group over k is an inner form of a unique quasi-split group, up to isomorphism. This follows from an argument using Galois cohomology and the splitting

$$ 1\to\operatorname{Int}(G)\to\operatorname{Aut}(G)\to\operatorname{Aut}(\Psi_0(G))\to 1,$$ where $\Psi_0(G)$ is the based root datum of G. More details can be found on the first article of Corvallis, and surely in some other place (maybe Borel-Tits?).

share|cite|improve this answer

If I may be allowed, I will use a "high powered" theorem to deduce this is the case of number fields. It is indeed true that given an absolutely simple $k$-algebraic group $G$, the number of absolutely simple $k$-algebraic groups $H$ such that the restriction of scalars $R_{k/{\mathbb Q}}G$ and $R_{k/{\mathbb Q}}H$ are ${\mathbb Q}$-isomorphic, is finite. The hypothesis implies that as abstract groups $G(k)$ and $H(k)$ are isomorphic, since they are both ${\mathbb Q}$-rational points of the restriction of scalars group. Now by the Margulis superrigidity theorem (it was initially proved for ($S$)-arithmetic groups, but there is a version in his book for $k$-rational points which may be thought of as irreducible lattices in the adelic groups) such an abstract isomorphism $\theta $ arises from an isomorphism $\sigma:k \rightarrow k$ of the number field $k$, and a $k$-isomorphism $\phi: ^{\sigma }G \rightarrow H$ of $k$ algebraic groups. This means that $\theta (g)= (\phi (\sigma(g))$ for all $g\in G(k)$. $^{\sigma }G$ is the same group as $G$, twisted by the automorphism $\sigma$ on scalars.

In particular, $H$ is isomorphic to $^{\sigma }G$ for some $\sigma$. Since the number of the $\sigma$'s is finite, it follows that the number of $H$ is finite.

share|cite|improve this answer
Don't the comments to the original question provide a short elementary proof of the finiteness over any field using nothing beyond Galois descent (in particular, no input from number theory)? – user28172 Nov 18 '12 at 16:11
Indeed, but I find this answer very interesting also. Thanks, Askumadula! – Menny Nov 19 '12 at 20:57
@Menny: OK, but doesn't Margulis' theorem (say in its adelic form) require an isotropicity hypothesis (or more)? It would be good to clarify this issue. – user28172 Nov 20 '12 at 4:14
No, Margulis' theorem does not require isotropy hypothesis; if $G$ is isotropic over $K$, then this result is a theorem of Borel and Tits (abstract homomorphisms paper). The Margulis theorem, on the other hand, does not use cocompact or non-cocompact (the proof in his book). – Venkataramana Nov 20 '12 at 9:00
And thanks Menny for your remarks. – Venkataramana Nov 20 '12 at 9:01

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.