3
$\begingroup$

What is the classification of (Zariski) closed subgroups in $\operatorname{GL}_n(\mathbb C)$ (viewed as a linear algebraic group) with Lie algebra $\mathfrak{so}_n(\mathbb C)$?

Is it true that every such group is a product of $\operatorname{SO}_n(\mathbb C)$ with a cyclic subgroup generated by $\mu I$, where $\mu$ is some root of unity and $I$ is the identity matrix?

Correction in response to Mikhail Borovoi's comment: Is it true that for $n \neq 8$ every such group is a product of either $\operatorname{SO}_n(\mathbb C)$ or $\operatorname{O}_n(\mathbb C)$ with a cyclic subgroup generated by $\mu I$, where $\mu$ is some root of unity and $I$ is the identity matrix?

$\endgroup$
1
  • 5
    $\begingroup$ No, the group $G={\rm O}_{2m}(\Bbb C)$ is not such a product. Indeed, an element $g\in G$ with $\det(g)=-1$ acts on $\mathrm{SO}_{2m}(\mathbb C)$ by an outer automorphism. It follows that $g$ is not of the form $\mu I$. $\endgroup$ Feb 1, 2022 at 17:07

2 Answers 2

3
$\begingroup$

Such a subgroup contains $\mathrm{SO}_n(\mathbf{C})$ and is contained in the normalizer $N_n$ of $\mathfrak{so}_n(\mathbf{C})$.

What is $N_n$? It contains scalars acts on the 1-dimensional space of bilinear forms preserved by $\mathfrak{so}_n(\mathbf{C})$. It follows that $N_n=D\mathrm{O}_n(\mathbf{C})$ where $D\simeq\mathbf{C}^*$ is the group of scalars. If $n$ is odd, it follows that $N_n=D\mathrm{SO}_n(\mathbf{C})$ thus the description is exactly the one described: every closed subgroup with Lie algebra $\mathfrak{so}_n(\mathbf{C})$ is generated by $\mu\mathrm{Id}_n$ for some root of unity $\mu$.

If $n$ is even, it still holds that the possible groups are parameterized by finite subgroups of $D\mathrm{O}_n(\mathbf{C})/\mathrm{SO}_n(\mathbf{C})$. This group is abelian and is direct product of $D\mathrm{SO}_n(\mathbf{C})/\mathrm{SO}_n(\mathbf{C})\simeq\mathbf{C}^*$ by a group $W_2$ of order 2 (generated by any matrix of determinant $-1$ in $\mathrm{O}_n(\mathbf{C})$. So the resulting subgroups $H$ are parameterized by finite subgroups of $\mathbf{C}^*\times W_2$ and only those corresponding to a subgroup of $\mathbf{C}^\times \{1\}$ have the given form (generated by $\mathrm{SO}_n(\mathbf{C})$ and scalar matrices). In particular, $H/\mathrm{SO}_n(\mathbf{C})$ can be non-cyclic.

If $H$ contains $\mathrm{O}_n(\mathbf{C})$, it therefore happens that $H$ is generated by $\mathrm{O}_n(\mathbf{C})$ and $\mu I_n$ for some root of unity.

However when $n$ is even, it can happen that $H$ doesn't contain $\mathrm{O}_n(\mathbf{C})$, and also is not generated by $\mathrm{SO}_n(\mathbf{C})$ and scalar matrices. The simplest example is the group generated by $\mathrm{SO}_2(\mathbf{C})$ and the matrix $\begin{pmatrix}0 & i\\ i & 0\end{pmatrix}$ (note that this one is contained in $\mathrm{SL}_2(\mathbf{C})$).

(Note: $n=8$ is not a special case. Indeed, in general the "unusual" behavior of $\mathfrak{so}_n$ for $n=8$ disappears when $\mathrm{so}_n$ it is endowed with its standard $n$-dimensional representation.)

$\endgroup$
1
  • 2
    $\begingroup$ The question was modified from its original form (in response to @MikhailBorovoi's comment) to allow a subgroup generated by $\operatorname O_n(\mathbb C)$ and orthogonal matrices. \\ Is there any chance that "It contains scalars acts on the 1-dimensional space of bilinear forms preserved by $\mathfrak{so}_n(\mathbb C)$" is a typo? I'm not quite sure how to parse it. $\endgroup$
    – LSpice
    Feb 1, 2022 at 20:54
1
$\begingroup$

There's nothing special about $\mathbb C$ here except that it is a characteristic-$0$ field, so I will work over an arbitrary such field.

The identity component of a group $G$ with Lie algebra $\mathfrak{so}_n$ is $\operatorname{SO}_n$, so you're asking about the finite extensions of $\operatorname{SO}_n$ in $\operatorname{GL}_n$.

If $n = 1$, these are the finite subgroups of $\operatorname{GL}_1$, which are as you want.

If $n = 2$, these are the finite extensions of the (algebraic) torus $\operatorname{SO}_2$. Note that $\operatorname{SO}_2$ is a maximal torus in $\operatorname{GL}_2$, hence self-centralising. There is thus an injection from $\pi_0(G)$ to the Weyl group $\operatorname N_{\operatorname{GL}_2}(\operatorname{SO}_2)/{\operatorname{SO}_2}$, which is cyclic of order $2$. It is easy to check that the representatives of the non-identity coset of the Weyl group lie in $\operatorname O_2$, so your conjecture works in this case.

There is an injection from $\pi_0(G)/{\operatorname C_G(\operatorname{SO}_n)}$ to $\operatorname{Out}(\operatorname{SO}_n)$, which, since $n \ne 8$, is cyclic of order $1$ (for $n > 2$ even) or $2$ (for $n > 2$ odd).

We have that $\operatorname C_G(\operatorname{SO}_n)$ is contained in $\operatorname C_{\operatorname{GL}_n}(\operatorname{SO}_n)$. For $n > 2$, the latter equals $\operatorname Z(\operatorname{GL}_n)$ (as can be tested by computing with diagonal matrices, for example). Thus, for $n > 2$ odd, we are done.

For $n > 2$ even (with $n \ne 8$), the inner-automorphism map is an isomorphism $\operatorname O_n/{\operatorname{SO}_n} \to \operatorname{Out}(\operatorname{SO}_n)$. We thus have that $G$ is contained in $\operatorname Z(\operatorname{GL}_n)\cdot\operatorname O_n$, as desired.

$\endgroup$
1
  • 1
    $\begingroup$ @‍YCor's answer suggests that there is no need to treat $n = 8$ separately, but perhaps this Dynkin-diagram-focussed perspective, in the spirit of @‍MikhailBorovoi's comment, is nonetheless of some minor value. $\endgroup$
    – LSpice
    Feb 1, 2022 at 20:56

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.