7
$\begingroup$

Suppose that $G$ is a linear group of positive dimension, defined over some field $k$. Is that true, that $G$ admits a (closed) one-dimensional subgroup?

I'm pretty much sure this is true in characteristic 0, or at least for $k=\mathbb{C}$. It seems that the main obstacle in positive characteristic is that there may not exist elements of infinite order.

EDIT1: As @Daniel Loughran pointed out in the answer below, one need at least to assume that $k$ is algebraically closed (or it may not even be necessary, see the comments of @Marguax abaut $k$ being separably closed, or real closed in characteristic 0).

EDIT2: Since my first statement of the problem seems to be genereting a lot of confusion (after getting a comment from @Jim Humphreys I'm even a little ashamed), here is the final version of the question:

Suppose that $G$ is a linear algebraic group, of positive dimension, defined over an algebraically closed field $k$ (but of arbitrary characteristic). Is it true, that $G$ posses a one dimensional (closed) subgroup? (so in fact, either $\mathbb{G}_a$ or $\mathbb{G}_m$)

$\endgroup$
2
  • 1
    $\begingroup$ As indicated in the answers, your language is too loose to be clear: the meaning of "linear group defined over some field" needs to be made more precise, at which point the question is easy to answer. $\endgroup$ Sep 26, 2013 at 0:21
  • $\begingroup$ It is not necessary to assume $k = \overline{k}$ if allowing any characteristic; separably closed is enough. In char. 0 "real closed" is enough; e.g., the conclusion holds for $k = \mathbf{R}$. By the way, why do you highlight points of infinite order? That doesn't seem germane, since every nontrivial torus over a field $k$ not algebraic over a finite field (even such a $k$-torus with no nontrivial proper $k$-subtori) contains a $k$-point generating a Zariski-dense subgroup, so one cannot produce a 1-dimensional smooth closed $k$-subgroup by using any old point of infinite order. $\endgroup$
    – Marguax
    Sep 26, 2013 at 2:29

3 Answers 3

11
$\begingroup$

Every positive dimensional linear algebraic group $G$ over an algebraically closed field has a one dimensional subgroup.

Case 1 $G$ is reductive. In that case, $G$ contains a torus $T$. Since we are over an algebraically closed field, $T \cong \mathbb{G}_m^r$ for some $r>0$. In particular, $\mathbb{G}_m \subseteq T \subseteq G$.

Case 2 $G$ has a nontrivial unipotent radical $U$. In turn, $U$ has a nontrivial center $Z$. We have $Z \cong \mathbb{G}_a^s$ for some $s>0$ (not quite right, see Peter McNamara's correction below). In particular, $\mathbb{G}_a \subseteq Z \subseteq U \subseteq G$.

$\endgroup$
5
  • 5
    $\begingroup$ In Case 2, if the field is of positive characteristic, then Z is not necessarily a power of the additive group. (counterexample: length two Witt vectors). However it will still have a filtration by additive groups, so you still have a G_a as a subgroup $\endgroup$ Sep 26, 2013 at 0:56
  • 1
    $\begingroup$ The same conclusion holds over any separably closed field. Indeed, if $G\ne 1$ is a smooth connected affine group over a field $k$, $G$ contains a torus $T \ne 1$ or $G$ is unipotent. The first case is clear for $k=k_s$ ($T$ splits). For any $k$ and unipotent $G\ne 1$, WLOG $G$ is commutative, and also $p$-torsion if char($k)=p>0$. Thus, $G_K=\underline{V}$ for a finite $K/k$ and finite-dimensional $K$-vector space $V \ne 0$. For $j:G\hookrightarrow {\rm{R}}_{K/k}(G_K)$ and $W\subsetneq V$ complementary to Lie($G$) inside Lie(R$_{K/k}(G_K))=V$, $G\rightarrow \underline{V/W}$ is etale. QED $\endgroup$
    – Marguax
    Sep 26, 2013 at 2:21
  • $\begingroup$ I still have a problem to understand why a reductive group must contain a nontrivial torus. I suppose this is trivial to specialists, but for me this subject is totally new. $\endgroup$ Sep 26, 2013 at 11:05
  • 2
    $\begingroup$ Here's a proof that every positive-dimensional L.A.G. $G$ has a positive dimensional abelian subgroup, over an algebraically closed field. You can suppose $G$ closed connected of $GL_n$, and the field uncountable. If the function $\chi$ "characteristic polynomial" on $G$ is constant, every element in $G$ is unipotent, so $G$ is unipotent. [...->] $\endgroup$
    – YCor
    Sep 26, 2013 at 14:04
  • 2
    $\begingroup$ [--->] Otherwise $\chi$ has uncountable image, so contains polynomials with at least one root not root of unity. Hence $G$ has an element of infinite order and hence the Zariski closure of the subgroup it generates is a positive-dimensional (and not unipotent!) abelian subgroup of $G$. $\endgroup$
    – YCor
    Sep 26, 2013 at 14:04
11
$\begingroup$

Here is a counter-example for $k$ not algebraically closed (for simplicitly I assume that $k$ is perfect). Recall that there is an anti-equivalence of categories between the category of algebraic tori over $k$ and the category of free $\mathbb{Z}$-modules with a continuous action of $Gal(\overline{k}/k)$ (given by the character group $\widehat{T}$).

Choose an algebraic torus $T$ of dimension larger than one, whose character group $\widehat{T}$ is simple as a Galois module (these exist if e.g. $k=\mathbb{Q}$). Suppose that $T$ contains a one dimensional irreducible algebraic subgroup $C$. Then it is well known that such a group is isomorphic to either $\mathbb{G}_a,\mathbb{G}_m$ or an elliptic curve over $\overline{k}$. As $T$ is affine and every element is semi-simple, we see that $C$ is itself an algebraic torus. The inclusion $C \subset T$ therefore induces a non-trivial homorphism $\widehat{T} \to \widehat{C}$. The kernel of this homorphism is a Galois sub-module of $\widehat{T}$. However by assumption $\widehat{T}$ was simple, thus we deduce that $\widehat{C} = 0$ or $\widehat{T}$, which is a contradiction.

Conclusion: Such a torus $T$ cannot contain a one dimensional algebraic subgroup.

Edit: As Marguax points out I do in fact need $\widehat{T} \otimes_{\mathbb{Z}} \mathbb{Q}$ to be simple as a Galois module. I believe that such examples may also be given be given by the norm one torus $$R^1_{K/k} \mathbb{G}_m : N_{K/k}(x_1,\ldots,x_n) = 1.$$ Here $k \subset K$ is a cyclic Galois field extension of prime degree $n > 2$ and $N_{K/k}$ denotes the norm form for $K/k$. This may be defined by choosing an isomorphism $K \cong k^n$ and considering the usual field norm as a homogeneous polynomial of degree $n$ on $k^n$.

$\endgroup$
3
  • 1
    $\begingroup$ Can you provide an more explicit example of such a torus? $\endgroup$ Sep 25, 2013 at 21:32
  • 1
    $\begingroup$ You meant that $\widehat{T}_{\mathbf{Q}}$ is simple as a Galois module. Your argument works for $k$ (perfect or not) not separably closed or real closed, as such $k$ has a finite Galois extension $k'$ not a compositum of quadratic fields (so some $k$-simple isogeny factor of R$_{k'/k}(\mathbf{G}_m)$ has dimension $> 1$). Deeper is that a nontrivial smooth connected unipotent group over a field $k$ has a 1-dimensional smooth connected (central) $k$-subgroup. A connected semisimple example is the norm-1 unit group of a central division algebra of dimension $\ell^2$ for a prime $\ell > 2$. $\endgroup$
    – Marguax
    Sep 25, 2013 at 21:35
  • 4
    $\begingroup$ @Tomasz: Let $k'/k$ be a finite Galois extension of prime degree $p$. The torus $T = {\rm{R}}_{k'/k}(\mathbf{G}_m)/\mathbf{G}_m$ of dimension $p-1$ has rationalized character group given by the permutation action of cyclic group of order $p$ on the hyperplane of rational $p$-tuples whose coordinates sum to 0, and that's irreducible over $\mathbf{Q}$ (!), so $T$ is $k$-simple. So use $p > 2$. $\endgroup$
    – Marguax
    Sep 25, 2013 at 21:38
7
$\begingroup$

Over an algebraically closed field, the answer is yes. I assume that by "linear group" you mean a smooth affine algebraic group scheme. Such a group has a filtration whose quotients (modulo finite group schemes) are successively a unipotent group, a torus, and a semisimple group, so it is only a question of checking each of these cases.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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