Let $(M,\langle\cdot,\cdot\rangle)$ be a riemannian manifold. Let $S \subset M$ be a compact submanifold with corners and $p \in M$ a point belonging to $\partial A$. Can we ensure that exists $\varepsilon > 0$ and $\delta > 0$ (depending on $p$) such that for every $q \in B(p,\varepsilon)\cap S$ and $0 \leq t \leq \delta$ it is satisfied $$ \exp_p(t\exp_{p}^{-1}(q)) \in S \cap B(p,\varepsilon)? $$

I know that if $p \in \operatorname{int}(S)$ the result is obvious. If $S$ is locally convex the result is obvious too. Although, I was drawing some polyhedras in $\mathbb{R}^3$ and I thought that it is true in a this context too. Anyone can help me?

PS. I've just realized that the condition that I'm asking is: are compact submanifold with corners locally star-shaped?

  • $\begingroup$ Presumably $A=S$? $\endgroup$
    – Didier
    Commented Nov 25, 2022 at 12:55
  • $\begingroup$ Yes, sorry :). I edit it $\endgroup$ Commented Nov 25, 2022 at 12:57
  • 1
    $\begingroup$ there are some other things you may want to fix. I assume $x= p$? Are you assuming that $q\in S$? And is $\exp$ the exponentional map or $M$, or of $S$? If I look at the northern half of the $n-1$ dimensional sphere in $\mathbb{R}^n$ (which does not even have corners) I wonder why such a result should be true, anyway? $\endgroup$
    – Thomas
    Commented Nov 25, 2022 at 13:10
  • $\begingroup$ What about $S=[-1,1]^2$ in $M=\Bbb R^2$, $p = (0,1)$, $q = (0,1+\varepsilon/2)$? In that case, $\exp_p(tv) = p + tv$, to that $\exp_p(t\exp^{-1}(q)) = p+t(0,\varepsilon/2)$, which is never in $S$ for $t>0$. $\endgroup$
    – Didier
    Commented Nov 25, 2022 at 13:10
  • $\begingroup$ The example it not valid because I said $q \in B(p,\varepsilon) \cap S$. Your example, in fact, has the property that I want to prove. It is a very interesting question and I can't find anything about it. $\endgroup$ Commented Nov 25, 2022 at 13:14

1 Answer 1


Consider $M=\Bbb R^2$ endowed with the euclidean metric, and $$ S = \left\{(x,y) \in M \mid x\in [-1,1] \quad \text{and} \quad 0\leqslant y \leqslant 1 + x^2\right\}. $$

$S$ is a compact submanifold with corners (the corners are $(\pm 1,0)$ and $(\pm 1,2)$). Let $p= (0,1)\in S$, and $q = (x,1+x^2)\in S$ for some $x\in [-1,1]$. By strict convexity of the function $f(x) = 1+x^2$, we have $$ S \cap [p,q] = \{p,q\}. $$ From this, it should be clear that the statement you wish to be true is false.


You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .