For a Riemannian manifold $(M,g)$, if $F: M\to M$ is a diffeomorphism and preserves distances, I would like to show that $F$ is an isometry. By "distance" I mean $d(x,y)=\inf \int_0^1 |\gamma'(t)| dt$ where the infemum is taken over all admissible curves $\gamma$ (with $\gamma(0)=x$ and $\gamma(1)=y$). By "isometry", I mean that each differential $dF_p : T_p M\to T_{F(p)} M$ is a linear isometry. I know that this is true (at least with mild assumptions on $M$, like possibly connectedness), but I am not sure if there is a "simple" proof of it. For instance, problem 7.7 in Lee's "Introduction to Riemannian Geometry" asks you to show that a homeomorphism that is a metric space isometry is an isometry (so clearly $F$ falls into this category), but the proof is very long. Does anyone know if there is a simple way of showing this?


Given a minimizing geodesic segment $\gamma : [0,\epsilon] \to M,$ consider the restrictions $$\gamma|_{[0,t]} : [0,t] \to M$$ for small $t$. Each of these is also a minimizing geodesic, and thus we have the inequality

$$\ell_\gamma(t)=L(\gamma|_{[0,t]}) = d(\gamma(0), \gamma(t)) = d(F(\gamma(0)), F(\gamma(t))) \le L((F\circ\gamma)|_{[0,t]})=\ell_{F\circ\gamma}(t).$$

Since $\ell_\gamma(0) = \ell_{F\circ\gamma}(0) = 0,$ we conclude from calculus that $\ell_{\gamma}'(0) \le \ell_{F \circ \gamma}'(0).$ Since $\ell_\gamma(t) = \int_0^t |\gamma'|,$ we have $\ell_{\gamma}'(t) = |\gamma'(t)|;$ so the inequality of derivatives is $$|\gamma'(0)| \le |(F\circ \gamma)'(0)| = |DF(\gamma'(0))|.$$

By the local existence of geodesics, this implies that $|v|_g \le |DF(v)|_g = |v|_{F^*g}$ for all vectors $v\in TM.$

Since $F^{-1}$ is also a distance-preserving diffeomorphism, repeating the same argument with $w = DF(v)$ shows that we also have the opposite inequality $$|v|_g=|DF^{-1}(w)|_g \ge |w|_g = |v|_{F^*g};$$ so in fact we have equality: $$|v|_g = |v|_{F^* g}.$$ By polarization, this equivalence of norms implies the equivalence of inner products; so $g=F^*g$ as desired.

  • $\begingroup$ From just knowing that distances between points are preserved, how do you conclude that lengths of curves are preserved? $\endgroup$ Mar 4 '20 at 18:17
  • 1
    $\begingroup$ @AdamMartens: you're right, total brainfart there; my answer was wrong. Hopefully it's correct now. $\endgroup$ Mar 4 '20 at 21:31

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.