1
$\begingroup$

I'm confused about the precise definition of an inner automorphism of an algebraic group. Here is what Milne says in his book on algebraic groups:

Let $k$ be a field, let $\overline{k}$ be an algebraic closure, and let $G$ be an algebraic group over $k$. Let $Z$ (or $Z(G)$) denote the center of $G$. Then an automorphism of $G$ is said to be inner if it becomes of the form inn$(g)$ over $\overline{k}$, where $g \in G(\overline{k})$. The action of $G$ on itself by inner automorphisms, $(x,y) \mapsto xyx^{-1}$, is invariant under $Z(G) \times e$ acting by translation, and so it factors through $G/Z(G)$. The inner automorphisms of $G$ are exactly the automorphisms of $G$ defined by elements of $G/Z(k)$.

I don't understand why we don't simply say that the inner automorphisms of $G$ can be identified with $G/Z$. I think Milne is identifying $G/Z(k)$ with a subset of the underlying topological space $G/Z$.

$\endgroup$
  • 2
    $\begingroup$ I don't have the print copy of Milne to hand, and I can't find this in the last edition of the online notes that he posted. The point is that $G/Z$ has points valued over all field extensions over $k$ (and more), and, if we want to regard these as inducing automorphisms of $G$ (regarded as a $k$-group), then they had better preserve $G(k)$. An arbitrary-field-valued point of $G/Z$ doesn't behave this way, but a $k$-valued point $(G/Z)(k)$, which may lift only to a point of $G(\overline k)$, does. Is that related to your question? $\endgroup$ – LSpice Sep 7 '18 at 1:28
  • $\begingroup$ Inner automorphisms are exactly the $k$-points of $G/Z$, the definition in terms of points over $\bar k$ is just an explanation how the points of the quotient look like. $\endgroup$ – Victor Petrov Sep 7 '18 at 17:13
  • 1
    $\begingroup$ When they make that identification, they are regarding $\mathrm{Inn}(G)$ as the identity component of the algebraic group $\mathrm{Aut}(G)$ of algebraic automorphisms—that is, in particular, as an algebraic group—whereas Milne is regarding the inner automorphism group as an abstract group (as @anon points out below). $\endgroup$ – LSpice Sep 9 '18 at 23:56
  • 1
    $\begingroup$ In particular, Milne's inner automorphism group can change after base change, whereas Borel's can't; and, more precisely, we have $\mathrm{Inn}_{\text{Milne}}(G) = (\mathrm{Inn}_{\text{Borel}}(G))(k)$, or we could perhaps even write more explicitly $\mathrm{Inn}_{\text{Milne}}(G_{/k}) = (\mathrm{Inn}_{\text{Borel}}(G_{/k}))(k)$. $\endgroup$ – LSpice Sep 9 '18 at 23:57
  • 1
    $\begingroup$ @LSpice Shouldn't $\mathrm{Aut}(\mathbb{G}_m^n)=\mathrm{GL}_n(\mathbb{Z})$? $\endgroup$ – Not a grad student Nov 9 '18 at 20:14
4
$\begingroup$

The inner automorphisms of $G$ form an abstract group, whereas $G/Z$ is an algebraic group (i.e., group scheme of finite type over the field $k$), so you can't say that one is equal to the other --- they are different types of objects. By $(G/Z)(k)$ Milne means the group of $k$-rational points of $G/Z$, which is an abstract group. Each element of $(G/Z)(k)$ defines an automorphism of $G$, and these are exactly the inner automorphisms

Here is the passage in Milne's book 3.51: Let $G$ be an algebraic group over $k$. An automorphism of $G$ is inner if it becomes of the form $ inn(g)$ over $k^{\mathrm{a}}$. Later (17.63), we shall see that the group $\mathcal{I}(k)$ of inner automorphisms of $G$ is equal to $\bar{G}(k)$, where $\bar{G}$ is the quotient of $G$ by its centre. For example, for $t\in k^{\times}$, the automorphism $\left( \begin{smallmatrix} a & b\\ c & d \end{smallmatrix} \right) \mapsto\left( \begin{smallmatrix} a & tb\\ t^{-1}c & d \end{smallmatrix} \right) $ of $SL_{2}$ is inner because it becomes the inner automorphism defined by $diag(\sqrt{t},\sqrt{t}\,^{-1})$ over $k^{\mathrm{a}}$. It is also the inner automorphism defined by the element $diag(t,1)$ of $PGL_{2}(k)$.

| cite | improve this answer | |
$\endgroup$
  • 1
    $\begingroup$ Thanks. In 3.51 I guess Milne regards the set of automorphisms of G as an abstract group. Later he says that if G is semisimple then Aut(G) has an algebraic group structure. $\endgroup$ – Not a grad student Sep 10 '18 at 2:06

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.