The term geodesic essentially means the shortest path between a pair of points. For example, a straight line in a plane is a geodesic, and any great circle on a sphere is also a geodesic. The geodesic curvature of a regular curve on a surface measures how the curve deviates from being a geodesic. To begin, we should first understand how to find, or characterize, a geodesic. It turns out it is related to the covariant derivative.
Suppose \(\gamma_0(t):[0,1]\to\Sigma^n\subset\mathbb{R}^{n+1}\) is a regular curve on a regular hypersurface \(\Sigma^n\), so that it is the shortest path on \(\Sigma^n\) between the points \(\gamma_0(0)\) and \(\gamma_0(1)\). Then, we have \[(4.1)\qquad \nabla_{\gamma_0'(t)}\!\left(\frac{\gamma_0'(t)}{|\gamma_0'(t)|}\right)=0\quad\forall\,t\in[0,1].\] Recall that \(\nabla\) denotes the covariant derivative, which is the projection of the directional derivative onto the tangent space.
The technique used in the proof is commonly called Calculus of Variations — which frequently appears in physics. We consider a family of curves \(\{\gamma_u(t):[0,1]\to\Sigma^n\}_{u\in(-\varepsilon,\varepsilon)}\) so that \(\gamma_u(0)=p\in\Sigma\) and \(\gamma_u(1)=q\in\Sigma\) for any \(u\in(-\varepsilon,\varepsilon)\). When \(u=0\) we get the curve \(\gamma_0(t)\) given in the proposition. The parameter \(u\) gives the label of a curve in the family; varying \(u\) is like a deformation of the center curve \(\gamma_0(t)\).
For any fixed \(u\), the arc-length of \(\gamma_u\) is \(L(\gamma_u)=\int_0^1\left|\frac{d}{dt}\gamma_u(t)\right|dt\). We only assume \(\gamma_0\) is arc-length parametrized. For \(\gamma_0\) to be the shortest path we need \(\frac{d}{du}L(\gamma_u)\big|_{u=0}=0\). Differentiating under the integral and using the identity \(\frac{d}{du}\left|\frac{d}{dt}\gamma_u\right|=\frac{1}{|\frac{d}{dt}\gamma_u|}\left\langle\frac{d}{dt}\frac{d}{du}\gamma_u,\frac{d}{dt}\gamma_u\right\rangle\) (where \(\frac{d}{dt}\) and \(\frac{d}{du}\) commute as \(u,t\) are independent), we get \[\frac{d}{du}L(\gamma_u)=\int_0^1\left\langle\frac{d}{dt}\frac{d}{du}\gamma_u(t),\frac{\frac{d}{dt}\gamma_u(t)}{\left|\frac{d}{dt}\gamma_u(t)\right|}\right\rangle dt.\]
We use integration by parts to move the \(\frac{d}{dt}\) derivative onto the other factor: \[\frac{d}{du}L(\gamma_u)=\left\langle\frac{d}{du}\gamma_u,\frac{\frac{d}{dt}\gamma_u}{\left|\frac{d}{dt}\gamma_u\right|}\right\rangle\bigg|_{t=1}-\left\langle\frac{d}{du}\gamma_u,\frac{\frac{d}{dt}\gamma_u}{\left|\frac{d}{dt}\gamma_u\right|}\right\rangle\bigg|_{t=0}-\int_0^1\left\langle\frac{d}{du}\gamma_u,\frac{d}{dt}\frac{\frac{d}{dt}\gamma_u}{\left|\frac{d}{dt}\gamma_u\right|}\right\rangle dt.\] Since \(\gamma_u(0)=p\) and \(\gamma_u(1)=q\) for all \(u\), the boundary terms \(\frac{d}{du}\gamma_u(0)=0\) and \(\frac{d}{du}\gamma_u(1)=0\), giving \[\frac{d}{du}L(\gamma_u)=-\int_0^1\left\langle\frac{d}{du}\gamma_u(t),D_{\gamma_u'(t)}\frac{\frac{d}{dt}\gamma_u}{\left|\frac{d}{dt}\gamma_u\right|}\right\rangle dt,\] where \(D\) denotes the directional derivative.
If \(\frac{d}{du}L(\gamma_u)\big|_{u=0}=0\) for any variation, then \(D_{\gamma_0'(t)}\frac{\frac{d}{dt}\gamma_0}{|\frac{d}{dt}\gamma_0|}\big|_{u=0}\) must be orthogonal to every tangent vector of \(\Sigma\). Its projection onto the tangent space is zero, which is exactly \(\nabla_{\gamma_0'(t)}\frac{\frac{d}{dt}\gamma_0(t)}{|\frac{d}{dt}\gamma_0(t)|}=0\), completing the proof. \(\square\)
From high school calculus, we know that \(f'(a)=0\) does not always guarantee that \(f\) has a minimum at \(a\), as \((a,f(a))\) could be a maximum or a saddle point. At the same token, a curve \(\gamma_0\) satisfying (4.1) does not guarantee that it is the shortest path between \(\gamma_0(0)\) and \(\gamma_0(1)\). In high school calculus we call \((a,f(a))\) a critical point if \(f'(a)=0\); now we call a curve satisfying (4.1) to be:
A regular curve \(\gamma(t):[a,b]\to\Sigma^n\subset\mathbb{R}^{n+1}\) on a regular hypersurface \(\Sigma\) is said to be a geodesic on \(\Sigma\) if \[\nabla_{\gamma'(t)}\!\left(\frac{\gamma'(t)}{|\gamma'(t)|}\right)=0\quad\forall\,t\in[a,b].\] If further that \(\gamma\) is the shortest path on \(\Sigma\) between \(\gamma(a)\) and \(\gamma(b)\), then we call \(\gamma\) to be a minimizing geodesic.
Consider any great circle \(\gamma(s)\), parametrized by arc-length, on a round sphere. It is easy to see that \(\gamma''(s)\) is an inward-pointing normal vector of the sphere. Recall that \(\nabla_{\gamma'(s)}\gamma'(s)\) is the projection of \(D_{\gamma'(s)}\gamma'(s)\) onto the tangent plane. Note that \(D_{\gamma'(s)}\gamma'(s)\) is just the normal vector \(\gamma''(s)\), and so \(D_{\gamma'(s)}\gamma'(s)\) has no tangent component. This shows \[\nabla_{\gamma'(s)}\gamma'(s)=0.\] Hence \(\gamma(s)\) is a geodesic of the round sphere.
Consider a rotational symmetric surface generated by the profile curve \(\gamma(t)=(x(t),0,z(t))\), where \(t\in(a,b)\), on the \(xz\)-plane such that \(x(t)>0\) for any \(t\in(a,b)\). The surface of revolution generated by \(\gamma\) is obtained by revolving \(\gamma\) about the \(z\)-axis, and it can be parametrized by \[F(t,\theta)=(x(t)\cos\theta,\,x(t)\sin\theta,\,z(t)),\qquad(t,\theta)\in(a,b)\times[0,2\pi].\] According to the calculations in p69, the first fundamental forms are given by: \[g_{tt}=x'(t)^2+z'(t)^2,\quad g_{t\theta}=0,\quad g_{\theta t}=0,\quad g_{\theta\theta}=x(t)^2.\] The components of the inverse matrix \([g]^{-1}\) are: \[g^{tt}=\frac{1}{x'(t)^2+z'(t)^2},\quad g^{t\theta}=0,\quad g^{\theta t}=0,\quad g^{\theta\theta}=\frac{1}{x(t)^2}.\] The Christoffel symbols are: \[\Gamma^t_{tt}=\frac{x'(t)x''(t)+z'(t)z''(t)}{x'(t)^2+z'(t)^2},\quad\Gamma^\theta_{tt}=0,\quad\Gamma^t_{t\theta}=0,\quad\Gamma^\theta_{\theta\theta}=0,\] \[\Gamma^\theta_{t\theta}=\frac{x'(t)}{x(t)},\quad\Gamma^t_{\theta\theta}=-\frac{x(t)x'(t)}{x'(t)^2+z'(t)^2}.\] For a curve of the form \(r(s)=F(t_0,\theta(s))\) with \(t_0\) fixed and \(\theta'(s)>0\), using properties (a)–(d) from p77 we compute \[\nabla_{r'(s)}\!\left(\frac{r'(s)}{|r'(s)|}\right)=-\frac{\theta'(s)x'(t_0)}{x'(t_0)^2+z'(t_0)^2}\,\frac{\partial F}{\partial t}.\] This shows \(r(s)\) is a geodesic if and only if \(x'(t_0)=0\). Geometrically, the \(\theta\)-curve generated by rotating \((x(t_0),0,z(t_0))\) about the \(z\)-axis is a geodesic precisely when \(x(t_0)\) achieves a local maximum or minimum.
Verify the Christoffel symbols of the above example.
Show that a regular curve of the form \(r(s)=F(t(s),\theta_0)\) where \(\theta_0\) is fixed must be a geodesic of the above surface of revolution.
Show that an arc-length parametrized geodesic \(r(s)=F(t(s),\theta(s))\) of a surface of revolution satisfies the following ODE system: \[\theta''(s)+\frac{2x'(t(s))}{x(t(s))}\theta'(s)t'(s)=0,\] \[t''(s)-\frac{x(t(s))x'(t(s))}{x'(t(s))^2+z'(t(s))^2}\theta'(s)^2+\frac{x'(t(s))x''(t(s))+z'(t(s))z''(t(s))}{x'(t(s))^2+z'(t(s))^2}t'(s)^2=0.\]