I am having a question about finding the metric on $SO(n+1)/SO(n)$ such that $\pi:SO(n+1)\rightarrow SO(n+1)/SO(n)$ is a Riemannian submersion, which $SO(n)$ are equipped with the standard bi-invariant metric. I try to use Schur's Lemma to show that it is isometric to the canonical metric on $S^n$ multiplied by a constant. I am confusing about how to find this constant directly.
1 Answers
First, you are right that there is a unique (up to scaling) metric on $S^{n-1}$ that makes $\pi$ into a Riemannian submersion, as I explain in my answer here. Note that the isotropy action in this case is transitive on the unit sphere, so is definitely irreducible.
Now that we know there is a constant we can scale by, let's figure it out. I'm not exactly sure what you mean by the "standard bi-invariant metric" on $SO(n)$, but the bi-invariant metric I like to use is defined on $T_I SO(n)$ by $\langle X,Y\rangle =-Tr(XY)$.
The function $\pi:SO(n)\rightarrow S^{n-1}$ I'm going to use is $\pi(A) = A_n$ where $A_n$ denotes the last column of $A$. This means that the preimage of the point $p=(0,...,0,1)\in S^n$ corresponds to matrices of the block form $diag(B,1)$ with $B\in SO(n-1)$.
Consider the tangent vector $\alpha'(0)\in T_p S^{n-1}$ with $\alpha(t) = (0,....,\sin(t),\cos(t))$. Note that $\|\alpha'(0)\| = 1$ in the usual metric on $S^{n-1}$.
Now, the identity matrix $I\in SO(n)$ is an element of $\pi^{-1}(p)$, so let's find a tangent vector in$ (\ker \pi_\ast)^\bot\subseteq T_I SO(n) = \mathfrak{so}(n)$ which projects to $\alpha'(0)$. (The notation $\pi_\ast$ refers to the differential $\pi_\ast: T_I SO(n)\rightarrow T_p S^{n-1}$.) Then we can compute the length of this tangent vector to find out the scaling we need to have a Riemannian submersion.
To that end, first note that because $\pi$ is constant on the orbit $I \,\cdot SO(n-1)$, it follows that $\ker \pi_\ast$ contains $\mathfrak{so}(n-1)$, embedded in $\mathfrak{so(n)}$ as matrices with the block form $diag(B,0)$ with $B\in \mathfrak{so}(n-1)$. Since $\pi$ is a subermsion, the kernel of $\pi_\ast$ cannot be any larger, so $\ker \pi_\ast = \mathfrak{so}(n-1)$. A reasonable straightforward calculation now shows that $(\ker \pi_\ast)^\bot = \{M = (M)_{ij}\in \mathfrak{so}(n): M_{ij} = 0$ if both $i,j < n\}.$ In other words, $\ker \pi_\ast^\bot$ consists of matrices of the form $$M = \begin{bmatrix} 0 & \cdots & 0 & m_{1,n}\\ 0 & \cdots & 0 & m_{2,n}\\ \vdots & & \ddots & \vdots\\ -m_{1,n} & -m_{2,n} & \cdots & 0\end{bmatrix}.$$
Now, consider $\gamma:\mathbb{R}\rightarrow SO(n)$ with $\gamma(t) = diag\left(1,...,1, \begin{bmatrix} \cos t & \sin t\\ -\sin t & \cos t\end{bmatrix}\right)$. Then $\gamma(0) = I$ and $\gamma'(0)$ is a matrix whose only non-zero entries are $\gamma'(0)_{n-1,n} = -\gamma'(0)_{n,n-1} = 1$. It follows that $\gamma'(0)\in (\ker\pi_\ast)^\bot.$
Finally, note that $\pi \circ \gamma = \alpha$, so $\pi_\ast(\gamma'(0)) = \alpha'(0)$.
Now, an easy calculation shows that $\langle \gamma'(0),\gamma'(0)\rangle = 2$. Since $\langle \alpha'(0), \alpha'(0)\rangle = 1$, we see that the submersion metric on $S^{n-1}$ is the usual metric scaled by a factor of $2$.
- 50,257
-
I do not understand what you have done without considering details. For an arbitrary submersion what is the common trick to construct a metric on the image? – C.F.G Nov 04 '20 at 21:04
-
Thank you so much! – EGRCC Nov 05 '20 at 05:24
-
@HarryWang: If the above answer, resolve your question you can click on the green mark below the downvote triangle and accept it as a best answer. – C.F.G Nov 05 '20 at 14:24
-
@C.F.G: I don't know if I'd say there is a common trick. Given a submersion $M\rightarrow N$, and any Riemannian metric on $N$, you can construct a metric on $M$ which makes the projection a Riemannian submersion. However, given a metric on $M$, it need not be possible to construct a metric on $N$ for which the projection is a Riemannian submersion. The trick here is that the Riemannian submersion is nice - a bi-invariant metric on $SO(n)$ is preserves by the natural action of $SO(n-1)$ given by multiplying on the right. So the quotient inherits a metric. – Jason DeVito - on hiatus Nov 05 '20 at 14:25
-
@JasonDeVito: Thank you. So there is no general method. Yeah? and that form $N$ to $M$ is constructed by pullback? – C.F.G Nov 05 '20 at 14:27
-
1@C.F.G: Pulling back doesn't work because submersions are rarely immersions. In other words, pulling back a metric gives a positive-semidefinite symmetric bilinear form on each tangent space, but there is no reason it must be positive definition. (For immersions, pullbacks automatically give positive definite things). The way the construction works is by first picking an auxiliary Riemannian metric on $M$ which allows you to decompose each $T_m M$ as $\ker \pi_\ast \oplus H_m$ for some complimentary subspace $H_m$. $\pi_\ast|{H_m}:H_m\rightarrow T{\pi(m)}N$ is an isomorphism.... – Jason DeVito - on hiatus Nov 05 '20 at 14:59
-
1Now, modify your auxilary metric on $M$ on $H_m$ by declaring $\pi_\ast$ to be an isometry on $H_m$. But keep the metric the same on $\ker \pi_\ast$, and keep $\ker\pi_\ast \bot H_m$. The resulting metric turns $\pi$ into a Riemannian submersion. – Jason DeVito - on hiatus Nov 05 '20 at 15:01