7
$\begingroup$

This weird problem popped up in my research:

Let $\ell$ be a prime. Is there a description of the smallest normal subgroup of $GL_2(\mathbb{Z}_\ell)$ containing $GL_2(\mathbb{Z})$?

Is there a description of the corresponding quotient group?

$\endgroup$

1 Answer 1

12
$\begingroup$

It's the subgroup of these elements in $\mathrm{GL}_2(\mathbf{Z}_\ell)$ with determinant in $\pm 1$. Thus the quotient is naturally isomorphic to the quotient of $\mathbf{Z}_\ell^\times$ by $\{\pm 1\}$.

Indeed, denote by $e_{ij}(x)$ (for $i\neq j$) the matrix with entry $(i,j)$ equal to $x$, diagonal entries equal to 1, and other entries 0, and $d_i(x)$, for $x$ invertible, the diagonal matrix with $(i,i)$ entry equal to $x$, and other diagonal entries equal to $1$; write $s_{ij}(x)=d_i(x)d_j(x^{-1})$. Then, for $\{i,j\}=\{1,2\}$, and $x\in \mathbf{Z}_\ell^\times$, we have $s_{ij}(x)e_{ij}(1)s_{ij}(x)^{-1}=e_{ij}(x^2)$. Since any element in $\mathbf{Z}_\ell$ is a sum of squares (of 2 squares for $\ell$ odd, 4 squares for $\ell=2$, if I remember correctly), it follows that the normal closure of $\mathrm{SL}_2(\mathbf{Z})$ contains all matrices $e_{ij}(y)$ for any $y\in\mathbf{Z}_\ell$. Since such matrices generate $\mathrm{SL}_2(\mathbf{Z}_\ell)$ (because it's a Euclidean ring), we deduce that the normal closure of $\mathrm{SL}_2(\mathbf{Z})$ is all of $\mathrm{SL}_2(\mathbf{Z}_\ell)$. (The same simple argument work in $\mathrm{SL}_d$ for $d\ge 2$.) Now since $\det(\mathrm{GL}_2(\mathbf{Z})=\{\pm 1\}$, we get the result.

(Thanks to grghxy for a correction, I was initially essentially thinking of $\mathrm{SL}_2$ and $\mathrm{GL}_2$ makes an issue with the determinant.)

$\endgroup$
9
  • 2
    $\begingroup$ One has to be careful about the "torus" quotient. More specifically, this argument shows that one gets at least ${\rm{SL}}_2(\mathbf{Z}_{\ell})$, so the answer is actually the image of ${\rm{GL}}_2(\mathbf{Z})$ in the abelian ${\rm{GL}}_2(\mathbf{Z}_{\ell})/{\rm{SL}}_2(\mathbf{Z}_{\ell}) = \mathbf{Z}_{\ell}^{\times}$ (via det), and this image is $\mathbf{Z}^{\times}$. So the normal closure is generated by ${\rm{SL}}_2(\mathbf{Z}_{\ell})$ and diag($u, 1$) for $u \in \mathbf{Z}^{\times}$, with quotient $\mathbf{Z}_{\ell}^{\times}/\mathbf{Z}^{\times}$. $\endgroup$
    – grghxy
    Jul 22, 2015 at 16:34
  • $\begingroup$ Also, a softer way to handle the squares (which works with more general integer rings in place of $\mathbf{Z}$) is to note that since the subset of squares in $\mathbf{Z}_{\ell}$ is open, the $\mathbf{Z}$-submodule they generate inside $\mathbf{Z}_{\ell}$ is open and hence closed, so it is a $\mathbf{Z}_{\ell}$-submodule. But it contains 1, so that $\mathbf{Z}$-submodule coincides with $\mathbf{Z}_{\ell}$. $\endgroup$
    – grghxy
    Jul 22, 2015 at 16:39
  • $\begingroup$ Since ${\rm{diag}}(t,1/t)=u^+(t)u^{-}(-1/t)u^+(t-1)u^{-}(1)u^+(-1)$ for units $t$ and standard parameterizations $u^{\pm}$ of standard root groups of ${\rm{SL}}_2$, for any ring $R$ and simply connected Chevalley $R$-group $G$ the subgroup of $G(R)$ generated by $U_a(R)$'s for roots $a$ contains $\Omega(R)$ for an open cell $\Omega$. If $R$ is local then $\ker(G(R)\rightarrow G(k))\subset \Omega(R)$ (as $\Omega$ is an open subscheme of $G$), and translates of $\Omega$ by Weyl-representatives cover $G$ (enough on geometric fibers!), so $U_a(R)$'s generate $G(R)$ (no Euclidean stuff). $\endgroup$
    – grghxy
    Jul 22, 2015 at 16:54
  • $\begingroup$ The Euclidean argument works; yes I'm aware there are plenty of variants and generalizations, and since the question was about $\mathrm{SL}_2$ I chose the most elementary argument I had in mind. $\endgroup$
    – YCor
    Jul 22, 2015 at 17:09
  • $\begingroup$ (But it's true that both $R$ local or $R$ Euclidean imply $\mathrm{SL}_2(R)$ generated by elementary matrices, and the argument when $R$ is local is the simpler one.) $\endgroup$
    – YCor
    Jul 22, 2015 at 18:47

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.