1

$\newcommand{\al}{\alpha}$ Let $M$ be a smooth Riemannian manifold. Fix $p \in M, v \in T_pM$. Define $$ \mathcal{A}_v:=\{ w \in T_pM\,|\,\exists\alpha:(-\epsilon,\epsilon) \to M, \, \, \alpha(0)=p, \dot \alpha(0)=v, (\nabla_{\frac{\partial}{\partial_t}}\dot \alpha)(0)=w\}, $$ where $\nabla_{\frac{\partial}{\partial_t}}\dot \alpha$ is the covariant derivative along the path $\alpha$ w.r.t the Levi-Civita connection.

Conjecture: For every $v \in T_pM$, we have $\mathcal{A}_v=T_pM$.

Question: Is there a coordinate-free proof?

A possible solution:

Set $\gamma(t) = \exp_p(t(v + \frac{t}{2}w))$. Then $(\nabla_{\frac{\partial}{\partial_t}}\dot \alpha)(0)=w$. Here is an attempt to find a coordinate-free proof of that.


A coordinate-based approach:

If $\dot \alpha(t)=a^i(t)\partial_i( \alpha(t))$, then $$ \nabla_{\frac{\partial}{\partial_t}}\dot \alpha(0)=\dot a^i(0)\partial_i+a^i(0)a^j(0)\Gamma_{ij}^k\partial_k. $$ Now the $a^i(0)$ are uniquely determined by requiring $v=\dot \alpha(0)=a^i(0)\partial_i$.

Since we can freely choose $\dot a^i(0)$, we can always arrange for $\nabla_{\frac{\partial}{\partial_t}}\dot \alpha(0)$ to equal any given vector $w$, regardless of the value of $a^i(0)a^j(0)\Gamma_{ij}^k\partial_k$.

Asaf Shachar
  • 25,111
  • 1
    I think something like $\gamma(t) = \exp_p(t(v + \frac{t}{2}w))$ works, no? – Didier May 07 '22 at 12:07
  • Well, this might look promising on a first glance, but unfortunately it doesn't work universally on all manifolds: Take for instance $M=\text{SO}_n$, $p=Id$. Then the Riemannian exponential map is just the standard exponential map, and taking $\alpha(t)=\exp(vt+\tfrac12 wt^2),$ one gets $\dot \alpha(0)=v, \ddot \alpha(0)=v^2+w.$ (See here: https://math.stackexchange.com/a/3056589/104576) So instead of $\ddot \alpha(0)=w$, we got $\ddot \alpha(0)=v^2+w$, which does not even have a reasonable interpretation for more general manifolds: – Asaf Shachar May 07 '22 at 14:35
  • After all $v^2$ denotes the multiplication of $v$ with itself; this notion exists for matrices, but not for "general tangent vectors" on arbitrary manifolds having less structure. – Asaf Shachar May 07 '22 at 14:36
  • 1
    I might be wrong, but I think there is some error somewhere in your example since $v^2$ is not even skew symmetric (so it is not in the tangent space of $SO(n)$). This second order derivative seems not to be a Levi-Civita derivative (one has to project it onto the tangent space) – Didier May 07 '22 at 14:40
  • Oh sorry, you are right! I confused the extrinsic acceleration, as measured in $\mathbb{R}^{n^2}$ with the intrinsic one in $SO(n)$. You are completely right. So now I wonder whether your guess is correct. (Indeed after projecting the extrinsic acceleration $v^2+w$ onto the tangent space of skew-symmetric matrices, one is left with $w$ as you wanted. So I wonder whether that is true in general. – Asaf Shachar May 07 '22 at 14:44

1 Answers1

2

In geodesic normal coordinates constructed at a point, the Christoffel symbols vanish there. This makes it easy to see that the answer to your question is always Yes

MathFont
  • 4,837