1
$\begingroup$

Let $\mathbb{k}$ be a totally real number field and let $$\text{SL}^{\pm}(n+1,\mathbb{R}) = \{ A \in \text{GL}(n+1,\mathbb{R}) | \det A = \pm 1 \}.$$ Let $$\text{SL}^{\pm}(n+1,\mathbb{k}) = \{ A \in \text{GL}(n+1,\mathbb{k}) | \det A = \pm 1 \}.$$
Consider the adjoint representation as a map from $\text{SL}^{\pm}(n+1,\mathbb{R})$ to the automorphism group of the Lie algebra $\mathfrak{sl}(n+1,\mathbb{R})$. The kernel of this map is $\{ I, -I \}$. Let $\gamma \in \text{SL}^{\pm}(n+1,\mathbb{k})$. Then since $\text{Ad}$ is a $\mathbb{Q}$-morphism, clearly the image $\text{Ad}(\gamma)$ is an $n^{2}+2n$ by $n^{2}+2n$ matrix whose entries are in $\mathbb{k}$.

Now conversely suppose that $Ad(\gamma)$ is an $n^{2}+2n$ by $n^{2}+2n$ matrix whose entries are in $\mathbb{k}$. Then does $\gamma$ have to be in $\text{SL}^{\pm}(n+1,\mathbb{k})$? If not, can anyone give an counter example?

$\endgroup$
6
  • 2
    $\begingroup$ Do you mean : if $\bar k$ is an algebraic closure of $k$, $\gamma\in {\rm SL}^{\pm}(n+1,{\bar k})$, ${\rm Ad}(\gamma )$ has coefficients in $k$, THEN $\gamma$ has coefficients in $k$ ? $\endgroup$ Nov 25, 2011 at 8:59
  • 1
    $\begingroup$ The kernel of the adjoint representation contains all $n+1$-roots of unity. Take any one of those, which is not in $\mathbb R$ and there's your counterexample. $\endgroup$
    – user1688
    Nov 25, 2011 at 15:00
  • $\begingroup$ The kernel of $Ad$:$SL^{\pm}(n+1,\mathbb{R}) \arrow Ad(SL^{\pm}(n+1,\mathbb{R}))$ is $\{Id, -Id\}$. $\endgroup$
    – kchoi
    Nov 25, 2011 at 23:49
  • 1
    $\begingroup$ I cleaned up some of your TeX. It is not clear to me what you mean by "conversely." $\endgroup$ Nov 26, 2011 at 0:05
  • 2
    $\begingroup$ You need to define the domain of $Ad$ for your question to make sense. As written, it is tautologically true. $\endgroup$
    – S. Carnahan
    Nov 26, 2011 at 3:32

1 Answer 1

1
$\begingroup$

The answer is Yes. The only reason for the appearance of formally real fields and $\mathbb R$ in the question is to rule out roots of unity; in fact, we have the following (cleaner and more general) statement:

Theorem 1. Let $L$ be a field of characteristic $0$, and $K$ be a subfield of $L$. Let $N\in\mathbb N$. Let $U\in\mathrm{GL}_N\left(L\right)$ be such that every $V\in\mathfrak{sl}_N\left(K\right)$ satisfies $UVU^{-1}\in \mathrm{M}_N\left(K\right)$. Then, there exists a $\lambda\in L^{\times}$ such that $\lambda U \in\mathrm{GL}_N\left(K\right)$.

[You might need a moment to see why this answers your question. Here is a dictionary: My $N$ is your $n+1$. My $U$ is your $\gamma$. My $K$ is your $\mathbb k$. My $L$ is your $\mathbb R$. Your condition that $\mathrm{Ad}\left(\gamma\right)$ is my condition that every $V\in\mathfrak{sl}_N\left(K\right)$ satisfies $UVU^{-1}\in \mathrm{M}_N\left(K\right)$. I'm leaving to you to check that, in your case, the $\lambda$ must be $\pm 1$; this is where $\mathbb R$ and the determinant of $\gamma$ being $\pm 1$ come into play, and it is completely trivial.]

Proof of Theorem 1. First, let us show that

(1) every $V\in M_N\left(K\right)$ satisfies $UVU^{-1}\in \mathrm{M}_N\left(K\right)$.

Proof of (1). Since $K$ has characteristic $0$ (this is the only place where we are using this assumption!), we have $\mathrm{M}_N\left(K\right)=\mathfrak{sl}_N\left(K\right)+K\cdot I_N$, where $I_N$ is the $N\times N$ identity matrix. Since (1) is linear in $V$, we are therefore done with the proof of (1) once we have shown that (1) holds for all $V\in\mathfrak{sl}_N\left(K\right)$ and for all $V\in K\cdot I_N$. But (1) holds for all $V\in\mathfrak{sl}_N\left(K\right)$ by assumption, and for all $V\in K\cdot I_N$ by inspection. Thus, (1) is proven.

Now, consider the map $r : M_N\left(K\right) \to M_N\left(K\right)$ which maps every $V$ to $UVU^{-1}$. This $r$ is well-defined due to (1), and a $K$-algebra isomorphism as can be easily seen; hence, $r$ is a $K$-algebra automorphism of $M_N\left(K\right)$. But it is known (see, e. g., the line after Corollary 2.12 in Chapter IV of Milne's Class Field Theory, or see Corollary 1 in http://ysharifi.wordpress.com/2011/01/26/automorphisms-of-central-simple-algebras-2/ , or any other source on central simple algebras and Brauer groups) that all $K$-algebra automorphisms of $M_N\left(K\right)$ are inner. Hence, $r$ is inner, so there exists some $P\in\mathrm{GL}_N\left(K\right)$ such that every $V\in\mathrm{M}_N\left(K\right)$ satisfies $r\left(V\right) = PVP^{-1}$. So every $V\in\mathrm{M}_N\left(K\right)$ satisfies

$PVP^{-1} = r\left(V\right) = UVU^{-1}$.

This can be easily rewritten as $U^{-1}PV\left(U^{-1}P\right)^{-1}=V$. In other words, $U^{-1}PV=VU^{-1}P$. Since this holds for all $V\in\mathrm{M}_N\left(K\right)$, it must also hold for all $V\in\mathrm{M}_N\left(L\right)$ (because it is a linear equation in $V$, so it is enough to check it on an $L$-basis of $\mathrm{M}_N\left(L\right)$, but such a basis can be chosen to lie in $\mathrm{M}_N\left(K\right)$). In other words, the matrix $U^{-1}P$ lies in the center of $\mathrm{M}_N\left(L\right)$. But this center is known to be $L\cdot I_N$ (this is, again, a known fact from the theory of central simple algebras, but it also appears in most linear algebra books). Thus, $U^{-1}P\in L\cdot I_N$. In other words, there exists some $\lambda\in L$ such that $U^{-1}P = \lambda I_N$. This $\lambda$ is nonzero (else, $P$ would be zero, contradicting $P\in\mathrm{GL}_N\left(K\right)$), so this becomes $P=\lambda U$. Hence, $\lambda U=P\in\mathrm{GL}_N\left(K\right)$, proving Theorem 1.

$\endgroup$
2
  • $\begingroup$ Thanks for your answer. But I don't see why $\lambda$ has to be $\pm 1$. I suppose you are saying that $U \in SL^{\pm}(n+1, \mathbb{R})$ and $P \in GL(n+1,\mathbb{k})$ imply $U \in SL^{\pm}(n+1, \mathbb{k})$? $\endgroup$
    – kchoi
    Dec 26, 2011 at 19:05
  • $\begingroup$ Okay, it seems that I got confused by the notations. But then your claim is wrong, as you can see by the example $n=1$, $\gamma = \left(\begin{array}{cc} \sqrt2 / 2 & 0 \\ 0 & \sqrt2 \end{array}\right)$ and $\mathbb k=\mathbb Q$. I am pretty sure that Theorem 1 is as much as one can get in this context. $\endgroup$ Dec 26, 2011 at 19:26

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.