MathOverflow is a question and answer site for professional mathematicians. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Hi guys,

I am able to prove that any symmetric manifold is complete (Consider a local geodesic and use the symmetry to flip it, effectively doubling the length of the geodesic, ad infinitum). I want to use a similar procedure to prove that a manifold whose isometries act transitively is complete, i.e there is always an isometry which maps the start point of a local geodesic to its end point, preserving the geodesic. I am, however, unable to ensure that it is not `rotated' in the process, i.e I want the pushforward of the initial tangent, by the isometry, to be the final tangent, ensuring the resultant doubled geodesic is smooth.

My Lie group theory is a bit scratchy but I assume there is a method which allows me to construct the correct pushforward using only transitivity.

Any ideas would be great,



share|cite|improve this question
The result you want is not true in indefinite signature, so you need to assume that you are working with positive-definite riemannian manifolds. (Perhaps this is implicit in your tag, but a growing number of people use riemannian geometry to include also the indefinite case.) – José Figueroa-O'Farrill Jul 8 '10 at 15:56
up vote 11 down vote accepted

By the Hopf-Rinow theorem, you only have to prove that the manifold is a complete metric space. By homogeneity, the injectivity radius is bounded from below by a uniform positive constant. Using this and the compacity of balls whose radius is smaller than the injectivity radius of their center it is easy to check the convergence of Cauchy sequences.

Another way to do this is to interpret the bound on injectivity radius, $r$ say, in term of geodesic extension: a geodesic $\gamma$ defined on $[a,b]$ can be extended to a geodesic defined on $(a-r,b+r)$. From this the conclusion follows.

share|cite|improve this answer
Needless to say, but just in case and since the OP has not made this clear: this will not work in indefinite signature, where Hopf-Rinow fails. There are incomplete homogeneous lorentzian manifolds, for example. – José Figueroa-O'Farrill Jul 8 '10 at 15:54

It is easy to see that any metrically homogeneous, locally compact, metric space, $X$, is complete. If $p$ is some point of $X$ then, by local compactness, for some $\epsilon > 0$, the closed $\epsilon$-ball about $p$ is compact and hence complete. Then, by metric homogeneity, the closed $\epsilon$-ball about every point is complete. Then, if $x_n$ is a Cauchy sequence in X, eventually the $x_n$ are all within $\epsilon/2$ of eachother, and so by the triangle inequality they eventually lie in a closed $\epsilon$ ball about one of them. qed (Of course, the same argument shows more generally that if a metrically homogeneous metric space has one point with a complete neighborhood then it is complete.)

share|cite|improve this answer
thank you to everybody. – kangdon Jul 11 '10 at 1:38

Your original approach for extending the geodesic is in fact possible, after a slight modification. (Credint: I heard this Idea from Eran Assaf).

(You do not pushforward the original geodesic, but instead you pushforward another geodesic which starts at $p$).

Let $p \in M$. It's known there exists a neighbourhood (of $p$) $V$ in $M$, and $\epsilon > 0$ such that for every $q \in V \, , w \in T_qM , |w|<\epsilon$ the (unique) geodesic which starts at $q$ with velocity $w$ is defined on $(-2,2)$.

Let $c$ be geodesic: $c(0)=p,|\dot c(0)|<\epsilon$. By the above statement, we can assume $c$ is defined on $(-2,2)$,So we can define $q = c(1)$.

By assumption there is an isometry $f:M \to M$ such that $f(p)=q$.

Now let $\gamma$ be the geodesic satisfying: $\gamma(0)=p,\dot \gamma(0)=(df_p)^{-1}(\dot c(1))$. Since $f$ is an isometry: $|\dot \gamma(0)|=|\dot c(1)|\overset{ c \, \text{geodesic} }{=}|\dot c(0)|<\epsilon$. Hence $\gamma$ is also defined on $(-2,2)$.

Now consider the curve $\beta = f \circ \gamma$:

$\beta(0) = f(\gamma(0))=f(p)=q=c(1)$

$\dot \beta(0) = df_p(\dot \gamma(0))=\dot c(1)$

Since $f$ is an isometry and $\gamma$ is a geodesic, $\beta$ is also a geodesic. Hence, $\beta$ is the unique geodesic starting at $q = c(1)$ with velocity $\dot c(1)$. But so is $\tilde c(t)=c(t+1)$, hence they are equal, whenever both are defined.

So, we can now concatenate, to obtain a geodsic $\alpha:(-2,3) \to M$,

$\alpha(t) = \begin{cases} c(t), & t \in (-2,2) \\ \beta(t-1), & t \in (-1,3) \end{cases}$

$\alpha$ is well-defined, and a geodesic. Now we may repeat this procedure ad infinitum.

share|cite|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.