Use connection to give local trivialization of fibration

49 Views Asked by At

I'm trying to figure out the following question:

Consider a smooth submersion $$f: M^{2n} \mapsto \mathbb{R}^{n}$$

Then for $x_0 \in M$ with $f(x_0)=c_0$, we can find an open neighborhood $U_0$ of $x_0$ and an open neighborhood $B_0$ of $c_0$ s.t. there exists a diffeomorphism $\phi: B_0 \times f^{-1}(c_0) \mapsto U_0$ with

$f(\phi (c,x))=c$ and $\phi(c_0,x)=x$

i.e. we have over $U_0$, $f$ is a trivial fibration.

My idea is to use connection to ifentify nearby fibers, precisely:

  1. Choose a connection $\nabla$ on M, denote its exponential map as exp.
  2. For each $v \in \mathbb{R}^n$ as tangent vector at $c_0 \in \mathbb{R}^n$, choose a splitting $T_{x_0}M=\text{ker}(df)\bigoplus V$ and define a lifting map $L: T_{c_0}\mathbb{R}^n \mapsto V$ by reversing the isomorphism $df\mid_{V}$
  3. Define $\phi$ as $\phi (b,f)=\text{exp}_f(L(b-b_0))$

I have a vague idea that this should work, but I still have some confusions in the following parts:

  1. The local diffeomorphism comes from the fact that the exponential map itself is a local diffemorphism. But to ensure this, do we have to choose the connection in a certain way? i.e. how do I make sure that the exponential map takes a vector in $\text{ker}df$ to some other point in the same fiber?

  2. The choice of the splitting should be done by use of a metric. Is this necessary? i.e. can we choose a splitting that solely relies on the connection itself?