1

Here is the description of the reflection group of type $D_n$ in Humphreys' book Reflection Groups and Coxeter Groups:

($D_n$, $n \ge 4)$ Let $V = \Bbb{R}^n$, and define $\Phi$ to be the set of vectors of squared length $2$ in the standard lattice. So $\Phi$ consists of the $2n(n-1)$ roots $\pm e_i \pm e_j$ ($1 \le i < j \le n$). For $\Delta$ take $\alpha_1 = e_1 - e_2$, $\alpha_2 = e_2 - e_3$,..., $\alpha_{n-1} = e_{n-1} - e_n$, $\alpha_n = e_{n-1} + e_n$. Then $\widetilde{\alpha} = e_1 + e_2$ [is the longest root?]. $W$ is the semidirect product of $S_n$ (permuting the $e_i$) and $(\Bbb{Z}/2 \Bbb{Z})^{n-1}$ (acting by an even number of sign changes), the latter normal in $W$.

My first question is, in this case are all the roots long roots, in which case the size of the orbit of $\widetilde{\alpha}$ under $W$ is $2n(n-1)$? Because $W$ acts transitively on the long roots? My second question is, what exactly does "acting by an even number of sign changes" mean?

user193319
  • 8,318

1 Answers1

1

For every vector $α$ in $V$, the reflection associated to $α$ is the map $$ s_α \colon V \to V \,, \quad x \mapsto x - 2 \frac{(x, α)}{(α, α)} α $$ There are two special cases of this general formula that one should be aware off here:

  • If the vector $α$ is of the form $α = e_i - e_j$ for two standard basis vectors $e_i$ and $e_j$ with $i ≠ j$, then the reflection $s_α$ swaps the two standard basis vectors $e_i$ and $e_j$, and leaves all other standard basis vectors untouched. In a formula, this reads as $$ s_{e_i - e_j}(e_k) = \begin{cases} e_j & \text{if $k = i$,} \\ e_i & \text{if $k = j$,} \\ e_k & \text{otherwise.} \\ \end{cases} $$

  • If the vector $α$ is of the form $e_i + e_j$ for two standard basis vectors $e_i$ and $e_j$ with $i ≠ j$, then the reflection $s_α$ swaps the sign of both $e_i$ and $e_j$, and leaves all other standard bases vectors untouched. As a formula, $$ s_{e_i + e_j}(e_k) = \begin{cases} -e_i & \text{if $k = i$,} \\ -e_j & \text{if $k = j$,} \\ \phantom{-}e_k & \text{otherwise.} \\ \end{cases} $$

(One should also be aware of the case $α = e_i$, but we won’t need this here.)


The group of type $D_n$ is the Weyl group of the root system of type $D_n$. It is therefore the subgroup of $\operatorname{GL}(ℝ^n)$ generated by all reflections $$ s_α \quad\text{with}\quad α = \pm e_i \pm e_j \,, i ≠ j \,. $$ We always have $s_{-α} = s_α$, so it suffices to consider the reflections $$ s_α \quad\text{with}\quad α = e_i \pm e_j \,, i ≠ j \,. $$ We see that these generators of $D_n$ fall into the two categories of reflections discussed above:

  • The reflections of the first kind (coming from $α = e_i - e_j$) swap the standard basis vectors $e_1, \dotsc, e_n$ around. They generated a subgroup of $W$ isomorphic to the symmetric group $S_n$

  • The reflections of the second kind (coming from $α = e_i + e_j$) swap the signs of the standard basis vectors. However, each such reflection always swaps two signs at once! So if we consider the subgroup of $W$ generated by the reflections of the second kind, then we can only ever swap an even number of signs. (This subgroup of $W$ is isomorphic to the subgroup of $(ℤ / 2ℤ)^n$ whose elements have an even number of non-zero entries. This subgroup is in turn isomorphic to $(ℤ / 2ℤ)^{n - 1}$.)


Regarding the first part of the question: all roots have the same length, and the root system $D_n$ is irreducible (because $n ≥ 3$). Therefore, the Weyl group acts transitively on $D_n$: the orbit of any root is the entire root system.

We can also see this from the explicit description of the group $D_n$ above: we can go from any root $± e_i ± e_j$ with $i ≠ j$ to any other root $± e_k ± e_l$ with $k ≠ l$ by changes indices and by swapping an even number of signs of all standard basis vectors $e_1, \dotsc, e_n$. (We need here that $n ≥ 3$. This allows us, for example, to go from $e_1 - e_2$ to $e_1 + e_2$ by applying the reflection $s_{e_2 + e_3}$ that flips the signs of $e_2$ and of $e_3$. We need here the additional third standard basis vector $e_3$ to flip only a single sign in the given expression $e_1 - e_2$, but flip overall an even number of signs in all of $e_1, \dotsc, e_n$.)