Indexed text attachment
connections-curvature-higher-geometry.tex
Source for Connections, Curvature, and Higher Geometry: From covariant differentiation to differential cohomology
View paper · Download original file
Lines 521–620 of 4453. Source text is displayed literally.
521\chapter{Geodesics, Jacobi fields, and surfaces}522\label{ch:geodesics}523 524The covariant derivative turns geodesic motion into a differential equation. A variation of one geodesic through other geodesics satisfies a new, linear equation containing curvature. On surfaces the same curvature has a global integral constrained by the topology. The local motion and the global identity thus give two ways of detecting the curvature defined in Chapter~\ref{ch:intrinsic}. We use Tu's conventions for the geodesic equation and his presentation of the surface theorem as a comparison \cite[\S\S\,13--17]{tu}.525 526\section{Energy variation and the exponential map}527 528Let \(\Gamma(s,t)\) be a smooth variation of a path \(\gamma(t)=\Gamma(0,t)\) on \([a,b]\). Put \(T(s,t)=\partial_t\Gamma(s,t)\) and \(V(s,t)=\partial_s\Gamma(s,t)\). The energy of the path \(t\mapsto\Gamma(s,t)\) is529\[530 E(s)=\frac12\int_a^b531 g_{\Gamma(s,t)}\bigl(T(s,t),T(s,t)\bigr)\,dt ,532\]533with \(s\) held fixed during integration. Both fields are sections of \(\Gamma^*TM\). The torsion-free identity \(\nabla_sT=\nabla_tV\) holds because the mixed partials of \(\Gamma\) agree and the Christoffel contributions are symmetric. Therefore534\begin{equation}535 E'(s)=\bigl[g(V,T)\bigr]_a^b-536 \int_a^b g(V,\nabla_tT)\,dt .537 \label{eq:geo-first-var}538\end{equation}539Here metric compatibility first replaces the derivative of the squared speed by \(2g(\nabla_sT,T)\), and then gives the integration-by-parts formula. For variations fixing both endpoints, the boundary term vanishes. Every smooth field supported in a small segment of the interior serves as a variation vector there, using coordinate displacement multiplied by a cutoff. Thus \(E'(0)=0\) for every fixed-endpoint variation implies the geodesic equation along \(\gamma\) by the fundamental lemma of the calculus of variations. The reverse implication follows from \eqref{eq:geo-first-var}.540 541Let \(\gamma_v\) start at \(p\) with velocity \(v\). Put \(\exp_p(v)=\gamma_v(1)\) when the solution exists on \([0,1]\). The solution depends smoothly on the initial conditions. The rescaling relation \(\gamma_{tv}(1)=\gamma_v(t)\) on its domain gives542\((d\exp_p)_0=\operatorname{id}_{T_pM}\), under the canonical identification of the tangent space of \(T_pM\) at zero. Thus \(\exp_p\) maps a sufficiently small open ball diffeomorphically onto a normal neighborhood. The coordinates from a linear orthonormal identification \(T_pM\simeq\mathbb R^n\) then have radial lines as geodesics.543 544\begin{proposition}[Radial orthogonality]545At a point \(v\) in a normal ball and for \(w\in T_pM\), one has546\[547 g_{\exp_p(v)}\big((d\exp_p)_v v,(d\exp_p)_v w\big)548 =g_p(v,w).549\]550\end{proposition}551 552\begin{proof}553Use \(\Gamma(s,t)=\exp_p(t(v+sw))\) for small \(s\) and \(0\leq t\leq1\). Along a radial geodesic \(\nabla_tT=0\), so554\[555 \partial_t g(V,T)=g(\nabla_sT,T)556 =\tfrac12\partial_s |v+sw|_{g_p}^2.557\]558As \(V(s,0)=0\), integration to \(t=1\) and evaluation at \(s=0\) give the formula. The vector \(T(s,1)\) at that endpoint is \((d\exp_p)_{v+sw}(v+sw)\).559\end{proof}560 561Consequently radial directions have unit length under the radial distance coordinate and are perpendicular to variations tangent to its spheres. For a path in the normal neighborhood the speed is at least the absolute derivative of the radius \(|\exp_p^{-1}(\gamma(t))|\) wherever the radius is positive. Integrating, with limiting endpoints at zero allowed, proves minimality of the radial segment among paths in the neighborhood. If the endpoint has radius \(r\) smaller than the radius \(R\) of a compactly contained normal ball, any path leaving that ball has initial length at least \(R\) by the same estimate up to its first exit. Thus the radial segment minimizes among all paths to the sufficiently near endpoint. These arguments are local and do not assume geodesic completeness.562 563\section{The linearized geodesic equation}564 565Suppose now that every \(t\mapsto\Gamma(s,t)\) is geodesic. The curvature commutator on \(\Gamma^*TM\) gives566\[567 \nabla_t^2 V=\nabla_t\nabla_s T568 =\nabla_s\nabla_t T+R(T,V)T569 =-R(V,T)T .570\]571At \(s=0\), the field \(J=V|_{s=0}\) thus satisfies572\begin{equation}573 \nabla_t^2 J+R(J,\dot\gamma)\dot\gamma=0 .574 \label{eq:geo-jacobi}575\end{equation}576Conversely, the linear equation has unique solutions prescribed by \(J(0),\nabla_t J(0)\). Choose a smooth curve of initial positions with derivative \(J(0)\), and vary the initial velocities along it with covariant derivative \(\nabla_t J(0)\). To construct the varied velocities, use parallel transport along the curve of positions and add the desired first-order term in that trivialization. Smooth dependence of the geodesic flow gives a variation, at least locally on the geodesic interval, whose variation field has those two initial values. Uniqueness of the linear equation identifies it with \(J\). This exhibits Jacobi fields as the linear response of geodesic motion \cite{lee-riemannian}.577 578For a unit-speed geodesic in constant sectional curvature \(K\), the operator \(J\mapsto R(J,\dot\gamma)\dot\gamma\) on the orthogonal complement of \(\dot\gamma\) is \(K\operatorname{id}\). Indeed it is self-adjoint by the curvature symmetries, and its quadratic form on perpendicular vectors is \(K|J|^2\). In a parallel orthonormal frame perpendicular to \(\dot\gamma\), a field with \(J(0)=0\) has coefficients proportional to579\[580 s_K(t)=581 \begin{cases}582 \sin(\sqrt Kt)/\sqrt K,&K>0,\\583 t,&K=0,\\584 \sinh(\sqrt{-K}t)/\sqrt{-K},&K<0 .585 \end{cases}586\]587The tangential component with zero initial value is linear in \(t\). On the round \(n\)-sphere of radius \(a\) with \(n\geq2\), the first positive time for a nonzero such field to vanish is \(\pi a\), so the antipode is the first conjugate point. Varying the initial velocity in a geodesic with initial point held fixed identifies these vanishing fields with the kernel of the derivative of the exponential map at that time. For \(K\leq0\) the displayed solutions give no positive conjugate time in a constant-curvature model.588 589\section{Second variation and local stability}590 591Let \(\gamma\) be a geodesic and suppose the variation fixes both endpoints. Differentiate \eqref{eq:geo-first-var} at \(s=0\). The term involving \(\nabla_sV\) vanishes there because \(\nabla_tT=0\), so592\[593 E''(0)=-\int_a^b g(V,\nabla_s\nabla_tT)\,dt .594\]595The curvature commutator and \(\nabla_sT=\nabla_tV\) give596\(\nabla_s\nabla_tT=\nabla_t^2V+R(V,T)T\) at \(s=0\).597Integration by parts, using \(V(a)=V(b)=0\), yields the quadratic form598\[599 I(V,V)=\int_a^b600 \bigl(|\nabla_t V|^2-g(R(V,T)T,V)\bigr)\,dt.601\]602Its bilinear polarization is symmetric by the metric curvature identities. A Jacobi field \(J\) vanishing at the endpoints pairs to zero with every such \(V\), by \eqref{eq:geo-jacobi} and integration by parts. The converse follows by testing against fields supported on interior segments. Thus conjugate endpoint data are detected by the kernel of the second variation.603 604For a unit-speed segment of length \(L\) on the round sphere, choose a perpendicular parallel unit field \(e\), and set \(V(t)=\sin(\pi t/L)e(t)\) for \(0\leq t\leq L\). The two terms in \(I\) integrate explicitly to605\[606 I(V,V)=\frac L2\bigl((\pi/L)^2-a^{-2}\bigr).607\]608The variation \((s,t)\mapsto\exp_{\gamma(t)}(sV(t))\) is defined for sufficiently small \(s\), by compactness of the path interval and the local existence theorem. When \(L>\pi a\) the quadratic form is negative on \(V\), giving a decrease in energy through that variation. At \(L=\pi a\), the displayed field is the conjugate Jacobi field already computed.609 610\begin{example}[The upper half-plane]611On \(y>0\), the metric \(g=y^{-2}(dx^2+dy^2)\) has oriented orthonormal frame \(e_1=y\partial_x\), \(e_2=y\partial_y\). Its bracket is \([e_1,e_2]=-e_1\), so the Koszul formula yields \(\nabla_{e_1}e_1=e_2\) and \(\nabla_{e_2}e_1=0\). Hence \(\omega=dx/y\) and \(d\omega=dx\wedge dy/y^2=dV_g\), giving \(K=-1\). Curves \(t\mapsto x_0+iy_0e^t\) have unit velocity \(e_2\) and are geodesics. For \(a,b,c,d\in\mathbb R\) with \(ad-bc>0\), the map \(z\mapsto (az+b)/(cz+d)\) multiplies the imaginary part and the norm of the Euclidean differential by the same factor \((ad-bc)/|cz+d|^2\). It therefore preserves \(g\). The images of the imaginary axis under \(z\mapsto x_0+r(z-1)/(z+1)\), \(r>0\), are the semicircles centered at \(x_0\). These are geodesics as well, by the naturality result of Chapter~\ref{ch:levi}.612\end{example}613 614\section{From boundary turning to Gauss--Bonnet}615 616For a unit oriented boundary tangent \(T=\cos\theta\,e_1+\sin\theta\,e_2\) on a smooth arc, define \(k_g=g(\nabla_TT,JT)\), where \(J\) is positive rotation by \(\pi/2\). In a local positive frame the derivative of the expression for \(T\) gives617\(k_g=d\theta/ds+\omega(T)\). On a positively oriented embedded triangular disk \(D\) in an oriented chart, the boundary tangent, with signed exterior turns inserted at its three corners, turns by \(2\pi\). One may compute this using the planar turning theorem in the chart, since a positive frame change extending over a disk contributes zero winding. The turning identity and Stokes' theorem, with \(d\omega=-K\,dV_g\), give618\begin{equation}619 \int_D K\,dV_g+\int_{\partial D} k_g\,ds620 +\sum_{\mathrm{corners}}(\pi-\alpha)=2\pi ,