## Parallelism determines the connection November 2, 2009

Posted by Akhil Mathew in differential geometry, MaBloWriMo.
Tags: , , ,

I’m going to try participating in Charles Siegel’s MaBloWriMo project of writing a short post a day for a month.  In particular, I’m categorizing yesterday’s post that way too.  I’m making no promises about meeting that every day, but much of the material I talk about lends itself to bite-sized pieces anyway.

There is a nice way to tie together (dare I say connect?) the material yesterday on parallelism with the axiomatic scheme for a Koszul connection. In particular, it shows that connections can be recovered from parallelism.

So, let’s pick a nonzero tangent vector ${Y \in T_p(M)}$, where ${M}$ is a smooth manifold endowed with a connection ${\nabla}$, and a vector field ${X}$. Then ${\nabla_Y X \in T_p(M)}$ makes sense from the axiomatic definition. We want to make this look more like a normal derivative.

Now choose a curve ${c: (-1,1) \rightarrow M}$ with ${c(0)=p,c'(0) = Y}$. Then I claim that

$\displaystyle \nabla_Y X = \lim_{s \rightarrow 0} \frac{ \tau_{p, c(s)}^{-1} X(c(s)) - X(p) }{s}.$

This can be proved by a direct calculation in local coordinates, but here is a slightly more devious argument. By shrinking, assume ${c}$ is an embedding and not just an immersion. Now let ${v_1, \dots, v_n}$ be a basis for the tangent space ${T_p(M)}$, and consider the parallel vector fields ${X_1, \dots, X_n}$ on the curve ${c}$ with ${X_i(0) = v_i}$; extend them to vector fields on some neighborhood of ${c((-1,1))}$ with the same notation. Then, if necessary by shrinking the neighborhood of definition, we can assume the ${X_i}$ are linearly independent, and ${\nabla_Y X_i = 0}$ by assumption of parallelism.

We can write ${X = \sum f_i X_i}$ for some smooth ${f_i}$. Then

$\displaystyle \nabla_Y X = \sum (Yf_i)X_i (p)$

and by parallelism

$\displaystyle \lim_{s \rightarrow 0} \frac{ \tau_{p, c(s)}^{-1} (fX_i)(c(s)) - X_i(p) }{s} = X_i \lim_{s \rightarrow 0} \frac{ f(c(s)) - f(p)}{s} = (Yf_i) X_i.$

Note that parallel translation is linear, so ${\tau_{p, c(s)}^{-1} (fX_i)(c(s)) = f(c(s)) X_i(p)}$. Summing, we’re done.

Now thus parallel translation induces isomorphisms on the tangent spaces. In particular, we can consider the tensor bundles ${T^{r.s}(M)}$ and find isomorphisms between the fibers at any two points ${p,q}$ depending upon a curve between them. Then we can use a limit definition to define a connection on tensor fields too. But I will come back to this tomorrow.