Let $(M,g)$ be a Riemannian manifold and let $N$ be a compact hypersurface isometrically embedded into $M$ and let $\eta$ denote a choice of unit normal vector field on $N$. It is then true that $N$ admits an $\varepsilon$-tubular neighborhood. That is, there exists some $\varepsilon>0$ such that the normal exponential $\exp(t\eta)$ is a diffeomorphism onto its image for $t\in (-\varepsilon, \varepsilon)$. A generalization of the Gauss lemma implies that for each $t$, we have a "parallel" hypersurface $N_t$ which intersects the normal geodesics orthogonally and are distance $t$ from $N$.

I am interested if there has been a result relating the injectivity radius of $N_t$ to the injectivity radius of $N$. More specifically, it seems like it should be the case that the cut locus of $N_t$ varies continuously as one travels along the normal geodesic flow. That is, small change in $t$ should imply small change in the cut time of a point $p(t)$ moving along the normal geodesic. I have not seen a proof of this and to my knowledge the arguments for continuity of the injectivity radius of a manifold do not seem to generalize.

Has this been shown anywhere? The best result that I have found so far is the Riccati equation for tubular neighborhoods in Grays "Tubes" book: $$ S'(t) = S^2(t)+R(t) $$ where $S(t)$ is the shape operator of $N_t$ and $R(t)$ is defined by $R[X,Y] = \nabla_{[X,Y]} - [\nabla_X, \nabla_Y]$ on $N_t$. In this way, it seems like one should be able to relate the second fundamental form of $N_t$ to $N$ via integration. However, since the cut locus is not uniquely determined by the shape operator, this does not seem to be sufficient.

Has this result been shown before? Perhaps a weaker result: if we take $t$ to be on a compact interval $[-\varepsilon, \varepsilon]$, can we guarantee that the injectivity radius of $N_t$ does not shrink to zero as $t\rightarrow \varepsilon$?

Any help would be greatly appreciated!


Yes, the two injectivity radii are related.

Let $r>0$ be the injectivity radius from $N$. Then there is a neighborhood $U \subset M$ of $N$ and a Riemannian isometry $\phi : U \to (-r,r) \times N$. Each hypersurface $N_t$ as you describe, for $|t|<r$ will correspond to the slice $\{t\} \times N$ with the identification provided by $\phi$.

Identifying $U$ with $(-r,r)\times N$, the Riemannian metric has the form $dx^2 + h_x$ where, for all $x\in (-r,r)$, $h_x$ denotes a smooth one-parameter family of Riemannian metrics on $N$. This fact is just a generalization of the concept of normal coordinates (this is for sure proved in Gray's book, I think).

Furthermore, for all $p\in N$, the rays $\tau \mapsto (\tau,p)$, for $\tau \in (-r,r)$ are unit speed geodesics that realize the distance to $N$, that is $d_N((\tau,p)) = |\tau|$.

Therefore, if $|t|<r$, we have that the injectivity radius from $N_t$ is equal to $r-|t|$.


After Ryan's comment, I realize what the OP was really asking concerns the injectivity radius of the embedded submanifold $N_t$ with the metric induced by $(M,g)$, that is the injectivity radius of $(N,h_t)$, where $h_t$ is a one-parameter family of metrics on a compact manifold $N$ (induced by the ambient metric $g$, in this case, but this is not really important).

Since the function $h \mapsto i_h(N)$, defined on the space of smooth Riemannian structure equipped with the $C^2$ topology, is continous, and since the aforementioned family $h_t$ is smooth in $t$, for small $t$, the result easily follows. The continuity of $h \mapsto i_h(N)$ is proved here:

Ehrlich, Paul E., Continuity properties of the injectivity radius function, Compos. Math. 29, 151-178 (1974). ZBL0289.53034.

and see also the following reference for perhaps simpler statements and a more accessible proof:

Sakai, Takashi, On continuity of injectivity radius function, Math. J. Okayama Univ. 25, 91-97 (1983). ZBL0525.53053.

Since the proof above is not simple, let me show how one can prove your weaker claim. For a given manifold $(N,h)$ there exist a beautiful universal lower bound for its injectivity radius given explicitly in terms of a lower bound $\delta$ for the Ricci curvature, an upper bound $\Delta$ for the sectional curvature, the diameter $D$ and the volume $V$. This is originally due to Cheeger, but the lower bound can be recovered by more "elementary" arguments such as (i) the Klingenberg Lemma and (ii) the Heintze-Karcher inequality (cf. Note and references therein in Berger's "Panoramic view of Riemannian geometry). In particular one obtains

$$ i_h(N) \geq \inf\left\{ \frac{\pi}{\sqrt{\Delta}}, c_n \left(\frac{\sqrt{|\delta|}}{\sinh(\sqrt{|\delta|} D)}\right)^{n-1} V \right\} $$

where $c_n$ is a universal dimensional consant.

Applyting this estimate to your specific case $(N,h_t)$ shows that the injectivity radious cannot become zero for small $t$.

Clearly $t \mapsto i_{h_t}(N)$ can become zero when your family $N_t$ approaches the normal cut locus, as $N_t$ can collapse. Consider for example the unit circle in $\mathbb{R}^2$, for which $N_t$ collapses to a point as $t \to 1$.

  • $\begingroup$ Interesting. I don't quite follow the last sentence. It seems you are talking about the "normal cut locus" to $N$, that is, the distance $r$ for which the geodesics intersecting $N$ orthogonally are unique and length minimizing. I am interested in the injectivity radius of the submanifold $N_t$, i.e. the relationship between the injectivity radius of $N$ endowed with the metric $h_x$ as compared to the metric on $N$. $\endgroup$ – Ryan Vaughn Apr 2 '20 at 17:21
  • $\begingroup$ Thank you for the great edit! This was exactly what I was looking for. $\endgroup$ – Ryan Vaughn Apr 3 '20 at 13:46

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.