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 261–360 of 4453. Source text is displayed literally.
261\]262in an oriented chart. Here \(\det(J^{\mathsf T}gJ)=(\det J)^2\det g\). Consequently the formula transforms as an \(n\)-form under orientation-preserving changes of chart. Forgetting orientation gives the well-defined density \(|dV_g|\). For a map \(f\) that is an orientation-preserving local diffeomorphism, the same computation gives \(dV_{f^*h}=f^*dV_h\).263 264This construction depends on the metric algebraically at each point. If \(g'=e^{2u}g\) for a smooth real function \(u\), the length of a tangent vector is multiplied by \(e^u\) and \(dV_{g'}=e^{nu}dV_g\). Sectional curvature will depend on derivatives of \(u\) as well. The different orders of dependence, first for connections and second for curvature, reappear in the local description by differential forms.265 266The length metric on each connected component is defined by267\(d_g(p,q)=\inf_\gamma L_g(\gamma)\) over piecewise smooth paths from \(p\) to \(q\). In a relatively compact coordinate ball about \(p\), the positive metric matrix admits uniform bounds \(cI\leq(g_{ij})\leq CI\) for some \(0<c\leq C\). Paths staying in the chart obey the same two-sided bounds for lengths up to factors \(\sqrt c,\sqrt C\). Paths leaving a smaller ball incur a positive lower length bound before their first exit. Coordinate line segments supply the local upper estimate in the smaller ball. These facts prove nondegeneracy and show that \(d_g\) induces the manifold topology. Finding length minimizers is the next question. Its local equation is a connection equation, as treated in Chapter~\ref{ch:geodesics}.268\chapter{Levi-Civita connection and torsion}269\label{ch:levi}270 271A connection furnishes a derivative of tangent-vector fields. It compares vectors above adjacent points by specifying an infinitesimal rule with a product law for multiplying sections by functions. The construction recovers the directional derivative on Euclidean space and gives an intrinsic rule when the tangent bundle is presented only by local coordinates. We use the curvature convention \(R(X,Y)=[\nabla_X,\nabla_Y]-\nabla_{[X,Y]}\) throughout this chapter and Chapter~\ref{ch:intrinsic}.272 273\section{Affine connections and the derivative of a section}274 275\begin{definition}276An affine connection on \(M\) is an \(\mathbb R\)-bilinear operation \((X,Y)\mapsto\nabla_XY\) on smooth vector fields such that277\[278 \nabla_{fX}Y=f\nabla_XY,\qquad279 \nabla_X(fY)=X(f)Y+f\nabla_XY .280\]281Its torsion is the operation282\[283 T(X,Y)=\nabla_XY-\nabla_YX-[X,Y].284\]285\end{definition}286 287These identities make all expressions local. For example, if \(Y\) vanishes near a given point \(p\), choose a cutoff that is one near \(p\) with support in the vanishing neighborhood. Applying the product rule to the zero field obtained by multiplying by this cutoff gives \(\nabla_XY|_p=0\). Smooth local formulas for \(\nabla\) therefore restrict to smaller coordinate domains. In a local frame \(e_1,\ldots,e_n\), put288\(\nabla_{e_i}e_j=\Gamma^k{}_{ij}e_k\). Expanding the connection rules gives289\[290 \nabla_XY=\sum_k\left(X(Y^k)+291 \sum_{i,j}X^iY^j\Gamma^k{}_{ij}\right)e_k .292\]293The coefficients are arbitrary smooth functions for a local connection. A second connection differs from the first by a section of \(T^*M\otimes T^*M\otimes TM\), since the terms differentiating a multiplier cancel. For torsion the bracket satisfies \([X,fY]=X(f)Y+f[X,Y]\). The same cancellation proves \(C^\infty(M)\)-bilinearity and skew symmetry. In a coordinate frame \(T^k{}_{ij}=\Gamma^k{}_{ij}-\Gamma^k{}_{ji}\). In a general frame there is also the contribution from \([e_i,e_j]\).294 295Torsion can be changed without changing the equation for a curve.296For any affine connection, put297\(\nabla^\circ_XY=\nabla_XY-\tfrac12T(X,Y)\). The subtracted term298is a tensor, so the Leibniz rule for the connection is preserved.299The antisymmetric part of that tensor has exactly the size required300to cancel \(T\). Thus \(\nabla^\circ\) is torsion-free. Since301\(T(v,v)=0\), the two connections have the same affine geodesics.302This construction uses the antisymmetry of torsion and requires no303metric compatibility.304 305\section{The metric and symmetry determine the connection}306 307\begin{theorem}[Levi-Civita]308For a Riemannian metric \(g\), exactly one affine connection satisfies309\[310 Xg(Y,Z)=g(\nabla_XY,Z)+g(Y,\nabla_XZ),311 \qquad T(X,Y)=0.312\]313\end{theorem}314 315\begin{proof}316Write the compatibility identity in the cyclic orders \((X,Y,Z)\), \((Y,Z,X)\), and \((Z,X,Y)\). Add the first two and subtract the third. Torsion-freeness replaces each reversed derivative by the given derivative and a bracket. The result is the Koszul formula317\begin{align}318 2g(\nabla_XY,Z)319 &= Xg(Y,Z)+Yg(Z,X)-Zg(X,Y)\notag\\320 &\quad -g(X,[Y,Z])+g(Y,[Z,X])+g(Z,[X,Y]).321 \label{eq:levi-koszul}322\end{align}323Nondegeneracy gives uniqueness. To see existence, denote the right side by \(K(X,Y,Z)\). The product rule and the bracket identity show that \(K(X,Y,fZ)=fK(X,Y,Z)\): the coefficients of \(X(f)\) and \(Y(f)\) from differentiating the metric evaluations cancel the corresponding bracket terms. The same expansion gives \(K(fX,Y,Z)=fK(X,Y,Z)\) and \(K(X,fY,Z)=fK(X,Y,Z)+2X(f)g(Y,Z)\). Thus \eqref{eq:levi-koszul} defines a smooth affine connection using \(g\) to recover a vector from a covector. Subtracting \(K(Y,X,Z)\) from \(K(X,Y,Z)\) leaves \(2g([X,Y],Z)\), which proves zero torsion. Adding \(K(X,Y,Z)\) to \(K(X,Z,Y)\) leaves \(2Xg(Y,Z)\), which proves metric compatibility.324\end{proof}325 326With coordinate fields all brackets vanish. Inverting the metric matrix gives the explicit formula327\begin{equation}328 \Gamma^k{}_{ij}=\frac12\sum_\ell g^{k\ell}329 \bigl(\partial_i g_{\ell j}+\partial_j g_{\ell i}330 -\partial_\ell g_{ij}\bigr).331 \label{eq:levi-christoffel}332\end{equation}333In particular, the connection depends on the first derivatives of the metric. Formula \eqref{eq:levi-koszul} applies also to orthonormal frames with nonzero brackets, where the derivative terms of the metric evaluations are then zero. These are two complementary ways to compute the same connection \cite[\S\S\,6,13]{tu}.334 335\begin{proposition}336Let \(f:(M,g)\to(N,h)\) be an isometric local diffeomorphism. If \(X,Y\) on \(M\) correspond locally to \(df(X),df(Y)\) on \(N\), then \(df(\nabla^g_XY)=\nabla^h_{df(X)}df(Y)\) there.337\end{proposition}338 339\begin{proof}340The derivative of a local diffeomorphism preserves the vector-field bracket because the conjugated derivations have conjugated commutators. Transport the connection on \(N\) to \(M\) in any neighborhood where \(f\) is a diffeomorphism. The transported connection has zero torsion. It is compatible with \(f^*h=g\) by differentiation of the pulled-back inner products. Uniqueness gives the asserted identity. The local computations agree on overlaps.341\end{proof}342 343There is a closely related rule for an isometric immersion into Euclidean space. Decompose the ambient derivative \(D_XY\) into its tangential and normal parts. The tangential projection is compatible with the induced metric because normal fields are perpendicular to tangent fields. Its torsion is the tangential projection of \(D_XY-D_YX-[X,Y]=0\). Thus the tangential projection is the Levi-Civita connection. The normal projection defines the second fundamental form used in Chapter~\ref{ch:intrinsic}.344 345\section{Covariant differentiation along a path}346 347Let \(\gamma:I\to M\) and let \(V(t)\in T_{\gamma(t)}M\) be a smooth section of the pullback tangent bundle. In coordinates around a segment of \(\gamma\) define348\begin{equation}349 (\nabla_t V)^k350 =\frac{d V^k}{dt}+351 \sum_{i,j}\Gamma^k{}_{ij}(\gamma)\dot\gamma^i V^j .352 \label{eq:levi-path}353\end{equation}354This can also be derived locally by expressing \(V=\sum_jV^j(t)e_j(\gamma(t))\) in any frame and applying the product rule with355\(\nabla_t(e_j\circ\gamma)=(\nabla e_j)_{\gamma(t)}(\dot\gamma(t))\). If a frame is changed by a matrix \(B(\gamma(t))\), differentiating \(B\) by the ordinary chain rule gives the same change as applying \(\nabla_{\dot\gamma}\) to the new frame fields. Hence the result is independent of the frame. Formula \eqref{eq:levi-path} also applies when the path crosses itself or is momentarily stationary. It requires only the pulled-back section and no extension as a field on \(M\).356 357A parallel section has \(\nabla_t V=0\). This is a linear differential equation in \(V\), so an initial vector determines a unique solution over the path interval. The solution operators compose on concatenation, and reversing the path gives the inverse. Here \(\gamma_{12}*\gamma_{01}\) denotes first \(\gamma_{01}\) and then \(\gamma_{12}\).358\[359 \begin{tikzcd}[column sep=3.3em]360 T_{p_0}M\arrow[r,"P_{\gamma_{01}}"]\arrow[rr,bend right=25,"{P_{\gamma_{12}*\gamma_{01}}}"']&