3
$\begingroup$

Let $(M,g)$ be a simply connected 2-dimensional Riemannian manifold without boundary, and let $K$ be the Gausssian curvature defined on $M$. If $M$ is compact, then by Gauss-Bonnet Theorem, we have $$\int_M K dA = 4\pi,$$where $dA$ is the area element of $M$ under the metric $g$.

If $M$ is not compact, then the above equality is no longer true. For example, let $M=\mathbb{R}^2$ and we define the conformal metric $g=e^{2u}\delta$, where $u=\ln(sech x)$ and $\delta$ is the Euclidean metric, then one can verify that $K \equiv 1$ on $M$, but the total area of $M$ is $\infty$, not $4\pi$.

Motivated by this example, my question is that, if we assume $(M,g)$ is a noncompact simply connected surface without boundary, and the total area of $M$ is finite, then is it true that
$$\int_M K dA =4\pi?$$ Or is it true at least for the conformal case?

$\endgroup$

3 Answers 3

8
$\begingroup$

Consider the metric on $\mathbb R^2$ that is rotationally symmetric metric outside a compact set, namely, it is $dr^2+m(r)^2 d\phi^2$ for $r>R>0$. Here $m$ is a positive function on $[R,\infty)$.

The area form at points with $r>R$ is $dA=m(r)drd\phi$, so the surface has finite area if and only if $m$ is integrable on $[R,\infty)$. The total curvature of the rotationally symmetric end is $$\int_R^\infty -\frac{m^{\prime\prime}}{\!\!\! m} dA=-2\pi\int_R^\infty m^{\prime\prime} dr=2\pi\left(m'(R)-\lim_{r\to\infty} m'(r)\right).$$ It is easy to find examples where the limit on the right hand side does not exist but $m$ is integrable. If the limit exists and $m$ is integrable, the limit is zero, and the total curvature of the end is $2\pi m^\prime(R)$.

The metric is arbitrary if $r<R$, and the total curvature of this region can be computed with the usual Gauss-Bonnet for surfaces with boundary. The geodesic curvature of the boundary is easy to compute from the $r\ge R$ side (I don't remember the answer).

EDIT: As Willie Wong points out by Gauss-Bonnet every smooth metric on the region $\{r<R\}$ will have the same total curvature. So just extend $m$ to a smooth function on $[0,R]$ so that $m(r)=r$ near $0$ and consider the metric $dr^2+m(r)^2d\phi^2$ for all $r>0$. Its metric completion is smooth at the origin. (More generally, the metric is smooth at the origin if and only if $m^\prime(0)=1$ and $m$ extends to an odd smooth function on $\mathbb R$). Now the above computation gives total curvature as $2\pi(m^\prime(0)-m^\prime(\infty))=2\pi(1-m^\prime(\infty))$ and if $m^\prime(\infty)$ exists and the area is finite, the total curvature is $2\pi$.

$\endgroup$
6
  • $\begingroup$ To be lazy, you can set $m'(R) = 0$ and you have that the circle $r = R$ is geodesic. Then the total curvature is exactly $2\pi$. $\endgroup$ Jun 24, 2020 at 0:51
  • $\begingroup$ Right, but it would be nice to compute the general case. Here is a sketch (to be checked !). The vector field $X=\frac{1}{m(R)}\frac{d}{d\phi}$ is a unit vector field tangent to the curve $r=R$. The geodesic curvature is the length of $\nabla_X X=-\frac{m^\prime(R)}{m(R)}$. So it looks like the geodesic curvature of the boundary is $-2\pi \frac{|m^\prime(R)|}{m(R)}$. $\endgroup$ Jun 24, 2020 at 1:00
  • $\begingroup$ Thank you very much for this inspirational post. I'm just curious that, does there exist a positive function $m(r)$ such that $m(r)$ is integrable, while $\lim_{r \rightarrow \infty} m'(r) \ne 0$? Also, I think the metric cannot be arbitary if $r< R$, since we require the metric is smooth, right? $\endgroup$
    – student
    Jun 24, 2020 at 1:11
  • $\begingroup$ Taking my laziness further: by Gauss Bonnet it doesn't matter what the portion $< R$ is, so might as well be a spherical cap and assume $R < \pi \rho$ where $\rho$ is the radius of the spherical cap. The curvature integral of the cap is $2 \pi (1- \cos(R/\rho))$. Then since $m(r) = \rho \sin(r/ \rho)$ we have $m'(R)$ is simply $\cos(R)$. So in fact even in the general case you will have that that any surface that you constructed will have total curvature integral exactly $2\pi$. $\endgroup$ Jun 24, 2020 at 1:22
  • $\begingroup$ @student: if the limit $\lim m'(r)$ exists, it must be zero (otherwise $m(r)$ is unbounded). The limit may fail to exist (imagine of $m(r)$ containing a train of skinnier and skinnier bumps). // Sufficient regularity of the metric (for Gauss Bonnet to apply) is implicitly assumed. You just need the metric to be piecewise $C^2$ (IIRC) for Gauss-Bonnet to work. $\endgroup$ Jun 24, 2020 at 1:30
6
$\begingroup$

If $M$ has finite total curvature, then $$\int_M KdA\leq 2\pi$$ by the Cohn-Vossen inequality. The $2\pi$ arises since the Euler characteristic of a noncompact simply-connected surface without boundary is $1$.

$\endgroup$
2
  • 1
    $\begingroup$ I don't think this is right. A simple example is to consider $\mathbb{R}^2, \frac{4}{1+x^2+y^2}\delta$. This is simply-connected without boundary, and it is not compact (we don't compatify it at $\infty$), then the total curvature is $4\pi$. I checked th eCohn-VOssen inequality, and it requires that the manifold is complete. $\endgroup$
    – student
    Jun 24, 2020 at 0:29
  • 3
    $\begingroup$ For Cohn-Vossen you need to assume that the surface is complete. $\endgroup$ Jun 24, 2020 at 0:30
4
$\begingroup$

Since you asked about the conformal case: consider the metric $g = e^{2\phi} \delta$. The area element is $e^{2\phi} ~dx$. The Gauss curvature is $$ K = - e^{-2\phi} \Delta \phi $$ and so the curvature integral is equal to $$ - \int_{\mathbb{R}^2} \Delta \phi ~dx $$

Suppose now that $\phi = \phi(|x|)$ is radial. Then the total integral can be evaluated using Gauss-Green as $$ - \lim_{R\to\infty} 2\pi R\phi'(R). $$ Consider the case that $\phi$ is a smooth function such that for all $|x|$ sufficiently large we have $\phi(|x|) = - \kappa \ln(|x|)$. Notice that when $\kappa > 1$ we have that $M$ has finite total area. Observe also that by a direct computation that the total curvature integral can be evaluated to equal exactly $2\pi \kappa$.

And hence we have that in the conformal case the valid range of values of the total curvature integral contains at least the full range $(2\pi, \infty]$. (The $\infty$ endpoint is attained for, e.g., $\phi = - |x|^2$.)

$\endgroup$
1
  • $\begingroup$ Thank you very much! $\endgroup$
    – student
    Jun 24, 2020 at 2:57

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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