5
$\begingroup$

Let $M$ be an $n$-dimensional complete Riemannian manifold and $r$ is the distance function to a fixed point.

The Hessian comparison theorem says that if the sectional curvature of $M$ is bounded (precisely $k\le \operatorname{sec}\le K$), then the Hessian of $r$ is bounded by the Hessians of the distance function for the space form with constant sectional curvature $k$ and $K$ (precisely $\operatorname{Hess}_Kr\le\operatorname{Hess}r\le \operatorname{Hess}_kr$).

My question: What is the difference between the Hessian of $r$ and the Hessian of the distance function for the space form with constant sectional curvature 0? It's better to have a Taylor expansion.

Following the notation in Petersen's Riemannian geometry, the Hessian of the distance function for the space form with constant sectional curvature $k$ is $\operatorname{Hess}_kr=\frac{\operatorname{sn}_k'(r)}{\operatorname{sn}_k(r)}g_r$ where $g_r=g-(\operatorname{d}r)^2$ and $\frac{\operatorname{sn}_k'(r)}{\operatorname{sn}_k(r)}=\frac{1}{r}-\frac{k}{3}r+O(r^2)$. Note that $\operatorname{Hess}_0r=\frac{1}{r}g_r$. Based on the Hessian comparison theorem, I guess that $\operatorname{Hess}r-\operatorname{Hess}_0r=-\frac{\operatorname{sec}}{3}rg_r+O(r^2)$.


I originally guessed that the difference $\operatorname{Hess}r-\operatorname{Hess}_0r$ is a multiplication of the Ricci tensor, big thanks to MySheperd who excluded this possibility and suggested that the difference is $O(r)$ in the answer below. Then I started to ask the above question.

Thank you!

$\endgroup$
13
  • $\begingroup$ If $r$ is a distance function both for $g$ and $g_{0}$ (something that in general need not be true), note how $\mathrm{Hess}_{0}\,r\,=\frac{g_{0}-(dr)^{2}}{r}$ in your flat metric $g_{0}$. I think you cannot replace $g_{0}$ with $g$ there. What I said in my answer below is still valid, I think: in general, the difference $\mathrm{Hess}\,{r}-\mathrm{Hess}_{0}\,r$ will depend on the specfic choice of a flat metric $g_{0}$. $\endgroup$
    – MyShepherd
    Commented Apr 29, 2022 at 21:02
  • $\begingroup$ Please refrain from making substantial changes to your question which may render an answer invalid. I rolled back your question to the original version. $\endgroup$
    – Stefan Kohl
    Commented Apr 30, 2022 at 8:07
  • $\begingroup$ Sorry, I didn't mean to render the answer invalid. I just wanted to make the question more meaningful. $\endgroup$
    – Borromean
    Commented Apr 30, 2022 at 13:15
  • $\begingroup$ @StefanKohl The general idea of my question has never changed. The first few versions are unclear and even the answerer did not see these versions. I just improved my question to make it concise and clear and even added my new thoughts. I made such a great effort, but you rolled back my question to the origin, it's really unfair. $\endgroup$
    – Borromean
    Commented May 2, 2022 at 6:09
  • 1
    $\begingroup$ Let us continue this discussion in chat. $\endgroup$
    – MyShepherd
    Commented May 3, 2022 at 7:59

1 Answer 1

5
$\begingroup$

$dr$ is defined regardless of the metric. If $g$ and $g_{0}$ are two metrics, and $\nabla^{g}$ and $\nabla^{g_{0}}$ are their corresponding Levi-Civita connections, then the connection difference yields a tensorial operation $D:T^*M\rightarrow T^*{}M\otimes T^{*}M$ such that $\mathrm{Hess}\,r-\mathrm{Hess}_{0}\,r=\nabla^{g}dr-\nabla^{g_{0}}dr=D(dr)$. If one now assumes that $g_{0}$ is locally-flat, i.e., that its curvature tensor has 0-constant sectional cruvature, then one can pick parallel coordinates for this metric which makes $(\nabla^{g_{0}}dr)_{ij}=\partial_{i}\partial_{j}r$. Therefore, in these coordiantes $D(dr)_{ij}=-\Gamma_{ij}^{k}\partial_{k}r$, where $\Gamma_{ij}^{k}$ are the chirstoffel symbols of the metric $g$. This includes only first derivatives of the metric, hence in general it won't be e.g. a multipication of the Ricci-Tensor.

In fact, if one picks $r$ to be a distance function, and picks polar coordinates with respect to this function, then an example of a locally-flat metric is $g_{0}=dr\otimes dr+\sum_{i=1}^{d-1}dx_{i}\otimes dx_{i}$. These polar coordinates are then parallel coordinates for $g_{0}$, and $D(dr)_{ij}=-\delta_{k}^{r}\Gamma_{ij}^{k}$. Since in polar coordinates $\Gamma_{ij}^{k}=o(r)$, this would imply that the difference vanishes as r goes to zero. This example also shows that the expression would depend on the choice of the particular locally-flat metric $g_{0}$.


Edit: Regarding @Borromean follow up question, the difference does not have to be $o(r)$ either. Let us take a unit disc $\mathcal{D}\subset \mathbb{R^{2}}$ and pick for it the usual polar coordinates $(r,\theta)$. Write two different metrics: $g=dr\otimes dr+r^{2}d\theta\otimes d\theta$ and $g_{0}=d\theta\otimes d\theta+\theta^{2}dr\otimes dr$. Note how both of these metrics are flat: they are simply eucliden metrics written in polar coordinates, with the roles of $r$ and $\theta$ exchanged. As such, both have constant sectional curvature 0. Note how $r$ is a distance function in one metric but not in the other.

However, $\mathrm{Hess}\,r=\nabla^{g} dr=-\Gamma_{\theta\theta}^{r}d\theta\,\otimes d\theta=-\frac{1}{r} d\theta\,\otimes d\theta$ while $\mathrm{Hess}_{0}r=\nabla^{g_{0}}dr=-\frac{1}{\theta}(dr\otimes d\theta+d\theta\otimes dr)$. So in this case, the difference will explode as $r\rightarrow 0$. Moreover, since the curvature of both metrics vanishes, the coefficents in the expansion of this difference in either $r$ or $\theta$ will include no expressions involving the curvature, as all curvature ingredients vanish identically.

$\endgroup$
10
  • $\begingroup$ In polar coordinates $g^{ij}=\delta^{ij}+o(r^2)$ so in this example we still have $\Delta r-\Delta_{0}r=-g^{ij}\Gamma_{ij}^{r}=o(r)$. In general, if the Hessian vanishes identitacly so must the Laplacian. $\endgroup$
    – MyShepherd
    Commented Apr 26, 2022 at 8:53
  • $\begingroup$ I forgot to note how $\Delta_{0}r=0$ in the polar coordinates, when we are using the locally flat metric $g_{0}$. This is becuase $\partial_{i}\partial_{j}r=0$ and the christoffel symbols of this metric in these coordinates vanish identically. $\endgroup$
    – MyShepherd
    Commented Apr 26, 2022 at 9:02
  • $\begingroup$ Maybe I did not understand correctly your question, but I think there is no escape from this, since in order to discuss the expression $\mathrm{Hess}\,r-\mathrm{Hess}_{0}\,r$, both of these objects must be sections of $T^{*}M\otimes T^{*}M$. Even if you take another space of constant sectional curvature 0, say $\mathbb{R}^{d}$, and map a neighborhood in it locally to a neighborhood in $M$ where $r$ is defined, this is where this difference is understood. Every such a map will induce, by declaration, a different flat metric on that neighborhood by pulling back the euclidean metric. $\endgroup$
    – MyShepherd
    Commented Apr 26, 2022 at 14:46
  • $\begingroup$ Did you read the Hessian comparison theorem? The metrics are on different manifolds and the distance functions are also on different manifolds. When we compare the Hessians, we take the distance functions to be both $r$. Did you read the page 175 of the 2nd edition of Petersen's Riemannian geometry? It says that the difference is $O(r)$. $\endgroup$
    – Borromean
    Commented May 3, 2022 at 10:53
  • $\begingroup$ The Hessians are defined on different tangent bundles, see Theorem 1.1 of Schoen-Yau's lectures on differential geometry. $\endgroup$
    – Borromean
    Commented May 3, 2022 at 11:12

You must log in to answer this question.

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