There is a canonical faithful orthogonal representation of the symmetric group $S_{n+1}$, for $n\geq 1$: $$ S_{n+1}\to O(n) $$ given as follows.

(1). I regard $O(n)$ as the isometry group of the unit sphere $S^{n-1}$ in $\mathbb{R}^{n}$.

(2). Let $\Delta^n$ be a regular $n$-simplex embedded in $\mathbb{R}^{n}$ such that

(i). all its edges are of the same length;
(ii). all its $(n+1)$-vertices are in $S^n$;
(iii). its center is the 0.

(3). I observe that any permutation on the $(n+1)$-vertices of $\Delta^n$ can be uniquely extended to an isometry of $S^n$. Hence I get an embedding of $S_{n+1}$ into $O(n)$.

Regarding $O(n)$ as a manifold, we have a canonical action of $S_{n+1}$ on $O(n)$. Hence we have a covering map $$ O(n)\to O(n)/S_{n+1}. $$ Let the vector bundle (the action of $S_{n+1}$ on $\mathbb{R}^{n+1}$ is given by permutation of coordinates of $\mathbb{R}^{n+1}$) $$ \eta: \mathbb{R}^{n+1}\to O(n)\times _{S_{n+1}}\mathbb{R}^{n+1}\to O(n)/S_{n+1}. $$

Question: I want to know the Stiefel-Whitney class of $w(\eta)$. How to compute it?

  • 1
    $\begingroup$ I can see a canonical homomorphism from $S_n$ to $O(n)$, namely by imbedding as a subgroup of the hyper-octahedral group; how do you define one from $S_{n+1}$ to $O(n)$? $\endgroup$ – John Jiang Sep 21 '15 at 4:19
  • $\begingroup$ Hi Prof., could I embed $S_{n+1}$ in such a way? $\endgroup$ – Shi Q. Sep 21 '15 at 6:26
  • $\begingroup$ @JohnJiang The map is in the original post. Another way to see this is the canonical homomorphism you mention, which is usually called a \it{unreduced regular representation}, fixes a hyperplane $x_1+x_2+⋯x_n=0$, so by restricting, you get a map from Sn to $O(n−1)$, which is called {\it reduced regular representation}. $\endgroup$ – user43326 Sep 21 '15 at 8:43
  • $\begingroup$ @JohnJiang: I think the embedding exists, and this is my way to see it. If $x$ and $y$ are two vertices of the $n$-simplex $\Delta$, I can define the `axis' $\pi(x,y)$ to be the hyperplane joining the middle point between $x$ and $y$, with the remaining vertices of $\Delta$. Then you can map the transposition $(i,j)$ from $S_n$ into the reflection with respect to the hyperplane $\pi(x_i,x_j)$ ($x_i$ vert. of $\Delta$). Since the transpositions generates $S_n$ (mathoverflow.net/questions/24101/…) this is (almost) enough to define the embedding. $\endgroup$ – Giovanni Moreno Sep 21 '15 at 14:25
  • 1
    $\begingroup$ What do you mean by "compute"? As I said elsewhere, there is a spectral sequence for $H^*(O(n)/S_{n+1})$. The SW classes in question are the images of the SW classes in $H^*BS_{n+1}$ of the standard representation of $S_{n+1}$, so they lie on the bottom line of the spectral sequence. You need to understand something about the behaviour of the spectral sequence before you can formulate any further questions. $\endgroup$ – Neil Strickland Sep 22 '15 at 7:04

I think the bundle you consider is trivial, because the representation of $S_{n+1}$ on $\mathbb R^{n+1}$ by permutation of coordinates extends to a representation of $O(n)$. I am not completely sure about the first part of the argument, however.

I think the map $S_{n+1}\to O(n)$ you describe actually comes from the $n$-dimensional standard representation of $S_{n+1}$. Indeed, permutation of coordinates gives a homomorphism $S_{n+1}\to O(n+1)$, but this has actually values in the stabilizer of the vector $v:=(1,\dots,1)$ and that stabilizer is isomorphic to $O(n)$, via the action on $v^\perp$. Now I think that if you orthogonally project the unit vectors to this hyperplane, you get (up to a multiplication by a constant factor depending on $n$) the $n+1$ points in the sphere that you describe.

If this is correct, then the representation of $S_{n+1}$ on $\mathbb R^{n+1}$ indeed extends to $O(n)$ as the direct sum of the defining representation of $O(n)$ on $v^\perp$ and the trivial representation on the line spanned by $v$. But then it follows from general principles that the bundle $O(n)\times_{S_{n+1}}\mathbb R^{n+1}$ can be trivialized. To do this, consider the map $O(n)\times\mathbb R^{n+1}\to (O(n)/S_{n+1})\times\mathbb R^{n+1}$ defined by $(g,v)\mapsto (gS_{n+1},g\cdot v)$. This evidently factorizes to an isomorphism on the induced bundle.


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.