19
$\begingroup$

Consider $SL_2$ embedded into $SL_3$ as upper left block matrices. The quotient $SL_3/SL_2$ is an affine variety, as is any quotient of reductive groups. How does one describe $SL_3/SL_2$? What are the equations for it in some affine space?

(One can also pose the same question more generally for $SL_{n}/SL_{n - 1}$.)

$\endgroup$
  • $\begingroup$ I think that it can just be cut out of $\mathbb A^6$ by thinking of it as $\begin{pmatrix} a & 0 & c \\ 0 & 1 & f \\ g & h & i \end{pmatrix}$, with $a$ constrained so that the determinant is 1. $\endgroup$ – LSpice Dec 28 '18 at 2:20
  • 2
    $\begingroup$ @LSpice I don't think it can; consider the matrix $\begin{pmatrix} 1 & 0 & 1 \\ 0 & 0 & -1 \\ 0 & 1 & 0 \end{pmatrix}$. Additionally, even on that set, the "constraint" isn't exactly "on $a$"; if the bottom-right 2 by 2 has determinant $0$, then $a$ doesn't affect the determinant. $\endgroup$ – user44191 Dec 28 '18 at 2:37
  • 1
    $\begingroup$ @user44191, thanks; I knew that it was a good decision to leave it as a comment rather than an answer. (Maybe a better decision still would have been to wait to comment until after some computation.) $\endgroup$ – LSpice Dec 28 '18 at 4:13
26
$\begingroup$

In general, $SL_n/SL_{n - 1}$ is isomorphic to an affine subvariety $X$ of $\mathbb{A}^{2n}$ with coordinate equations given by $\sum_i x_i y_i = 1$.

The map $SL_n/SL_{n-1} \rightarrow X$ is given by: the coordinates $y_i$ are given by the "last" vector (i.e. the last column), which is unaffected by $SL_{n-1}$, and the coordinates $x_i$ are given by the $(n-1)$-minors which ignore the last column (and the $i$th row), with some sign switching. Note that both of these are invariant, and so this is a well-defined map. The equation then comes from the equation for determinants from expansion by minors.

This map can also be reversed: given coordinates $x_i, y_i$ such that $\sum_i x_i y_i = 1$, we can find a matrix $A \in SL_n$ such that the last column of $A$ is $y_i$, and such that the $(n-1)$-minors ignoring the last column are $x_i$. Clearly, we can "fill in" the last column as-is, so we can focus on the minors, using the matrix $A_{n-1}$ with $n - 1$ columns. At least one of the $x_i$ must be nonzero. Fill in the rows of $A_{n-1}$ other than $i$ with a diagonal matrix with first entry equal to $x_i$, and the rest equal to $1$. We then only need to fill the $i$th row. But clearly the equations for each minor tell us one coordinate in that row - try it for yourself to see why. That "fills in" the coordinates, and each of the minors works, so we now have such a matrix $A$.

I claim that this map is well-defined - that it gives us the same class of $SL_n/SL_{n-1}$ no matter which nonzero $x_i$ we choose. Assume we have some $B$ with the same $(n-1)$-minors ignoring the last column (and the same last column) as the $A$ constructed above. Then its $i$th $(n-1)$-submatrix must have determinant $x_i$, which is nonzero, and so must be invertible. Therefore, there is a unique element $M \in SL_{n-1}$ such that $B_{n-1}M = A_{n-1}$ for all rows but the $i$th row. But by checking the other minors, we can then see that $B_{n-1}M = A_{n-1}$ - again, try and see for yourself. Therefore, this map is well-defined. We therefore have that if $M'$ is the extension of $M$ by adding a single $1$-block to $M$, then $BM' = A$.


Incorporating an idea from YCor's comment:

In general, a homogenous space $G/H$ is isomorphic to the $G$-orbit of a point with stabilizer $H$. As such, we only need to find an affine space $X$ with a $SL_n$-action and a point $x \in X$ such that $SL_{n-1}$ is the stabilizer of $x$. We can consider $SL_{n-1}$ as the intersection of two subgroups of $SL_n$: the subgroup that leaves the last column fixed (which consists of all matrices where the last column is $\begin{pmatrix} 0 \\ 0 \\ \vdots \\ 1 \end{pmatrix}$), and the subgroup that leaves the $(n-1)$-minors fixed (which consists of all matrices where the last row is $\begin{pmatrix} 0 & 0 & \dots & 1\end{pmatrix}$). Let $V$ be the natural representation of $SL_n$, and $V^*$ be its dual. The first of these subgroups can be considered the stabilizer of the vector $\vec{v} := \begin{pmatrix} 0 \\ 0 \\ \vdots \\ 1 \end{pmatrix} \in V$, while the second of the subgroups can be considered the stabilizer of the vector $\vec{v}^* := \begin{pmatrix} 0 & 0 & \dots & 1\end{pmatrix} \in V^*$. As such, $SL_{n-1}$ is the intersection of two stabilizers, and therefore is the stabilizer of $(\vec{v}, \vec{v^*}) \in V \oplus V^*$. Correspondingly, its orbit is isomorphic to $SL_n/SL_{n-1}$.

$\endgroup$
  • 8
    $\begingroup$ In a more group-theoretic point of view: $SL(V)$ acts linearly on $V\oplus V^*$. If $\dim(V)\ge 3$, this action has exactly 5 orbits: $\{(0,0)\}$, $(V-\{0\})\oplus\{0\}$, $\{0\}\oplus (V^*-\{0\})$, the set of pairs $(v,\ell)$ such that $v\neq 0$, $\ell(v)=0$, and finally the the set of pairs $(v,\ell)$ such that $\ell(v)\neq 0$. For this latter orbit, the point stabilizer of $(v,\ell)$ preserves the decomposition $Kv\oplus \mathrm{Ker}(\ell)$, fixes $v$ and acts on the hyperplane as an element of determinant 1, so the orbit of $(v,\ell)$ is the required homogeneous space. $\endgroup$ – YCor Dec 28 '18 at 8:08

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.