Let $(M,g)$ be a Riemannian manifold, $\gamma:I\rightarrow M$ a geodesic. Let $c:(-\varepsilon,\varepsilon)$ is defined to be another geodesic with $$c(0)= \gamma(0),\ \ \ c^\prime(0)=X.$$ Let $V(s)$,$W(s)$ be parallel vector fields along $c$.

The derivative that I want to compute is $$\frac{\partial}{\partial s}\exp_{c(s)} \big(t\cdot(V(s)+sW(s))\big)$$ at $s=0$.

I looked at examples, but there is no such thing as the $c(t)$ there so I can not figure it out.

Thank you very much, I hope it is not a too simple calculation question for this forum.


I encountered this problem several weeks ago when I read the paper

"A Riemannian interpolation inequality a la Borell, Brascamp and Lieb"

There is more or less a way to get rid of the dependence to the base point. The drawback is that we have to calculate the derivative of the distance function and consider $t$ not so big here (So it might not be very useful).

Let $y = y(t) =\exp_x tV$, where $x = c(0)$. If $x$ is not in the cut locus of $y$ (So $t$ is not big), then the function

$$ d^2_{y} (z) = d^2(y, z)$$

is differntiable at a neigborhood of $x$. Moreover, we can use Gauss lemma to show

$$ (*) \ \ \ \exp_{c(s)} \bigg( - \frac 12 \nabla d^2_{y} (c(s))\bigg) = y$$

For simplicity call $Z(s) = t(V(s) + sW(s))$. Then write

$$\exp_{c(s)}(Z(s)) = \exp_{c(s)} \bigg(- \frac 12 \nabla d^2_{y} (c(s)) + \frac 12 \nabla d^2_{y} (c(s)) + Z(s) \bigg)$$

Then using (*) and chain rule we have

$$\frac\partial{\partial s}\exp_{c(s)}(Z(s))\big|_{s=0} = (d\exp_x)_{tV} \bigg(\frac 12 \nabla_X \nabla d^2_{y} (x) + (\nabla_X Z)(0)\bigg)$$

In your case $V, W$ are parallel along $c$, so

$$\frac\partial{\partial s}\exp_{c(s)}(Z(s))\big|_{s=0} = (d\exp_x)_{tV} \bigg(\frac 12 \nabla_X \nabla d^2_{y} (x) + tW \bigg)$$

  • $\begingroup$ I'm probably missing something trivial here, but I am reading the same paper and I am stuck at this point. Could you please clarify your next-to-last equation (which is the first one in page 230 of the paper)? How do you actually fix the fact that the exponential map depends on $c(s)$? (and, but contrast, why can't you do that directly with $\exp_{c(s)}(Z(s))$?) $\endgroup$ – Giovanni Mascellani Oct 2 '14 at 8:44

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.