[Tu Differential Geometry] 7. Vector Bundles
This article is one of Manifold, Differential Geometry, Fibre Bundle.
The set \mathfrak{X}(M) of all C^\infty vector fields on a manifold M has the structure of a real vector space, which is the same as a module over the field \mathbb{R} of real numbers. Let \mathcal{F}=C^\infty(M) again be the ring of C^\infty functions on M. Since we can multiply a vector field by a C^\infty function. the vector space \mathfrak{X}(M) is also a module over \mathcal{F}. Thus the set \mathfrak{X}(M) has two module structure, over \mathbb{R} and over \mathfrak{F}. In spaeking of a linear map: \mathfrak{X}(M)\rightarrow \mathfrak{X}(M) one should be careful to specify whether it is \mathbb{R}-linear or \mathcal{F}-linear. An \mathcal{F}-linear map is of course \mathbb{R}-linear, but the converse is not true.
The \mathcal{F}-linearity of the torsion T(X,Y) and the curvature R(X,Y)Z from the preceding section has an important consequence, namely that these two constructions make sense pointwise. For example, if X_p, Y_p, and Z_p are tangent vectors to a manifold M at p, then one can define R(X_p,Y_p)Z_p to be (R(X,Y)Z)_p\in T_pM for any vector fields X, Y, and Z on M that extend X_p, Y_p, and Z_p, respectively. While it is possible to explain this fact strictly whithin the framework of vector fields, it is most natural to study it in the context of vector bundles. For this reason, we make a digression on vector bundles in this section.
We will try to understand \mathcal{F}-linear maps from the point of view of vector bundles. The main result (Theorem 7.26) asserts the existence of a one-to-one correspondence between \mathcal{F}-linear maps \alpha:\Gamma(E)\rightarrow \Gamma(F) of sections of vector bundles and bundle maps \varphi:E\rightarrow F.
7.1 Definition of a Vector Bundle
Given an open subset U of a manifold M, one can think of U\times \mathbb{R}^r as a family of vector spaces \mathbb{R}^r parametrized by the points in U. A vector bundle, intuitively speaking, is a family of vector spaces that locally "looks" liker U\times \mathbb{R}^r.
Definition 7.1. A C^\infty surjection \pi:E\rightarrow M is a C^\infty vector bundle of rank r if
(i) For every p\in M, the set E_p:=\pi^{-1}(p) is areal vector space of dimension r;
(ii) every point p\in M has an open neighborhood U such that there is a fiber-preserving diffeomorphism \phi_U:\pi^{-1}(U)\rightarrow U\times \mathbb{R}^r that restricts to a linear isomorphism E_p\rightarrow \{p\}\times \mathbb{R}^r on each fiber.
The space E is called the total space, the space M the base space, and the space E_p the fiber above p of the vector bundle. We often say that E is a vector bundle over M. A vector bundle of rank 1 is also called a line bundle.
Condition (i) says that \pi:E\rightarrow M is a family of vector spaces, while condition (ii) formalizes the fact that this family is locally looks like \mathbb{R}^n. We call the open set U in (ii) a trivializing open subset for the vector bundle, and \phi_U a trivialization of \pi^{-1}(U). A trivializing open cover for the vector bundle is an open cover \{U_\alpha\} of M consisting of trivializing open sets U_\alpha together with trivializations \phi_\alpha:\pi^{-1}(U_\alpha)\rightarrow U_\alpha\times \mathbb{R}^r.
Definition 7.5. Let \pi_E:E\rightarrow M and \pi_F:F\rightarrow N be C^\infty vector bundles. A C^\infty bundle map from E to F is a pair of C^\infty maps (\varphi:E\rightarrow F, \varphi_-:M\rightarrow N) such that
(i) The diagram
commutes,
(ii) \varphi restricts to a linear map \varphi_p:E_p\rightarrow F_{\varphi_-(p)} of fibers for each p\in M.
Abusing language, we often call the map \varphi:E\rightarrow F alone the bundle map.
An important special case of a bundle map occurs when E and F are vector bundles over the same manifold M and the base map \varphi_- is the identity map 1_M, In this case we call the bundle map (\varphi:E\rightarrow F,1_M) a bundle map over M.
If there is a bundel map \psi:F\rightarrow E over M such that \psi\circ \varphi=1_E and \varphi\circ \psi=1_F, then \varphi is called a bundle isomorphism over M, and the vector bundles E and F are said to be isomorphic over M.
Definition 7.6. A vector bundle \pi:E\rightarrow M is said to be trivial if it is isomorphic to a product bundle M\times \mathbb{R}^r\rightarrow M over M.
7.2 The Vector Space of Sections
A section of a bector bundle \pi:E\rightarrow M over an open set U is a function s:U\rightarrow E such that \pi\circ s=1_U, the identity map on U. For each p\in U, the section s picks out one element of the fiber E_p. The set of all C^\infty sections of E over U is denoted by \Gamma(U,E). If U is the manifold M, we also write \Gamma(E) instead of \Gamma(M,E).
The set \Gamma(U,E) of C^\infty sections of E over U is clearly a vector space over \mathbb{R}. It is in fact a module over the ring C^\infty(U) of C^\infty functions, for if f is a C^\infty function over U and s is a C^\infty section of E over U, then the definition (fs)(p):= f(p)s(p)\in E_p,\quad p\in U, makes fs into a C^\infty section of E over U.
Definition 7.12. A bundle map \varphi:E\rightarrow F over a manifold M (meaning that the base map is the identity 1_M) induces a map on the space of sections: \varphi_\#: \Gamma(E)\rightarrow \Gamma(F),\\ \varphi_\#(s)=\varphi\circ s. This induced map \varphi_\# is \mathcal{F}-linear because if f\in \mathcal{F}, then \left( \varphi_\#(fs)\right) (p)=\left( \varphi\circ (fs)\right) (p)=\varphi \left( f(p)s(p)\right)=f(p)\varphi \left(s(p)\right)=\left( f\varphi_\#(s)\right) (p).
Our goal in the rest of this chapter is to prove that conversely, every \mathcal{F}-linear map \alpha:\Gamma(E)\rightarrow \Gamma(F) comes from a bundle map \varphi:E\rightarrow F, i.e., \alpha=\varphi_\#.
7.3 Extending a Local Section to a Global Section
Proposition 7.13 (Existence of Global Section). Let E\rightarrow M be a C^\infty vector bundle, s a C^\infty section of E over some open set U in M, and p a point in U. Then there exists a C^\infty global section \bar{s}\in \Gamma(M,E) that agrees with s over some neighborhood of p.
Proof. Choose a C^\infty bump function f on M such that f\equiv 1 on a neighborhood W of p contained in U and \mbox{supp} f\subset U (Figure 7.3). Define \bar{s}:M\rightarrow E by \bar{s}(q)=\begin{cases}f(q)s(q)& \mbox{for }q\in U,\\ 0 & \mbox{for }q\notin U.\end{cases}
On U the section \bar{s} is clearly C^\infty for it is the product of a C^\infty function f and a C^\infty section s.
If p\notin U, then p\notin \mbox{supp} f. Since \mbox{supp} f is a closed set, there is a neighborhood V of p contained in its complement M\backslash \mbox{supp} f. On V the section \bar{s} is identically zero. Hence, \bar{s} is C^\infty at p. This proves that \bar{s} is C^\infty on M.
On W, since f\equiv 1, the section \bar{s} agrees with s.
7.4 Local Operators
In this section. E and F are C^\infty vector bundles over a manifold M, and \mathcal{F} is the ring C^\infty(M) of C^\infty functions on M.
Definition 7.14. Let E and F be vector bundles over a manifold M. An \mathbb{R}-linear map \alpha:\Gamma(E)\rightarrow \Gamma(F) is a local operator if whenever a section s\in \Gamma(E) vanishes on an open set U in M, then \alpha(s)\in \Gamma(F) also vanishes on U. It is a point operator if whenever a section s\in \Gamma(F) vanishes at a point p in M, then \alpha(s)\in \Gamma(F) also vanishes at p.
Proposition 7.17 (\mathcal{F}-linear is Local Operator). Let E and F be C^\infty vector bundles over a manifold M, and \mathcal{F}=C^\infty(M). If a map \alpha:\Gamma(E)\rightarrow \Gamma(F) is \mathcal{F}-linear, then it is a local operator.
Proof. Suppose the section s\in \Gamma(E) vanishes on the open set U. Let p\in U and let f be a C^\infty bump function such that f(p)=1 and \mbox{supp} f\subset U (Figure 7.3). Then fs\in \Gamma(E) and fs\equiv 0 on M (FIgure 7.4). So \alpha(fs)\equiv 0. By \mathcal{F}-linearity, 0=\alpha(fs)=f\alpha(s). Evaluating at p gives \alpha(s)(p)=0. Since p is an arbitrary point of U, \alpha(s)\equiv 0 on U.
7.5 Restriction of a Local Operator to an Open Subset
Theorem 7.20 (Existence of Local Section Map). Let E and F be vector bundles over a manifold M. If \alpha:\Gamma(E)\rightarrow \Gamma(F) is a local operator, then for each open subset U of M there is a unique linear map, called the restriction of \alpha of \alpha to U, \alpha_U:\Gamma(U,E)\rightarrow \Gamma(U,F) such that for any global section t\in \Gamma(E), \begin{align}\alpha_U(t|_U)=\alpha(t)|_U.\end{align}
Proof. Let s\in \Gamma(U,E) and p\in U. By Proposition 7.13, there exists a global section \bar{s} of E that agrees with s in some neighborhood W of p in U. We define \alpha_U(s)(p) using (1): \alpha_U(s)(p)=\alpha(\bar{s})(p). Suppose \tilde{s}\in \Gamma(E) is another global section that agrees with s in the neighborhood W of p. Then \bar{s}=\tilde{s} in W. Since \alpha is a local operator, \alpha(\bar{s})=\alpha(\tilde{s}) on W. Hence, \alpha(\bar{s})(p)=\alpha(\tilde{s})(p). This shows that \alpha_U(s)(p) is independent of the choice of \bar{s}, so \alpha_U is well defined and unique. Fix p\in U. If s\in \Gamma(U,E) and \bar{s}\in \Gamma(M,E) agree on a neighborhood W of p, then \alpha_U(s)=\alpha(\bar{s}) on W. Hence, \alpha_U(s) is C^\infty as a section of F.
If t\in \Gamma(M,E) is a global section, then it is a global extension of its restriction t|_U. Hence, \alpha_U(t|_U)(p)=\alpha(t)(p) for all p\in U. This proves that \alpha_U(t|_U)=\alpha(t)|_U.
Proposition 7.21 (\mathcal{F}-linear become Local). Let E and F be C^\infty vector bundles over a manifold M, let U be an open subset of M, let \mathcal{F}(U)=C^\infty(U), the ring of C^\infty functions on U. If \alpha:\Gamma(E)\rightarrow \Gamma(F) is \mathcal{F}-linear, then the restriction \alpha_U:\Gamma(U,E)\rightarrow \Gamma(U,F) is \mathcal{F}(U)-linear.
Proof. Let s\in \Gamma(U,E) and f\in \mathcal{F}(U). Fix p\in U and let \bar{s} and \bar{f} be global extensions of s and f that agree with s and f, respectivelty, on a neighborhood of p (Proposition 7.13). Then \alpha_U(fs)(p)=\alpha(\bar{f}\bar{s})(p)=\bar{f}(p)\alpha(\bar{s})(p)=f(p)\alpha_U(s)(p). (Use definition of \alpha_U, \mathcal{F}-linearity of \alpha)
Since p is an arbitrary point of U, \alpha_U(fs)=f\alpha_U(s), proving that \alpha_U is \mathcal{F}(U)-linear.
7.6 Frames
A frame for a vector bundle E of rank r over an open set U is a collection of sections e_1,\cdots,e_r of E over U such that at each point p\in U, the element e_1(p),\cdots,e_r(p) form a basis for the fiber E_p.
Proposition 7.22 (Frame on Trivial Bundle). A C^\infty vector bundle \pi:E\rightarrow M is trivial if and only if it has a C^\infty frame.
Proof. Suppose E is trivial, with C^\infty trivialization \phi:E\rightarrow M\times \mathbb{R}^r. Let v_1,\cdots, v_r be the standard basis for \mathbb{R}^r. Then the elements (p,v_i), i=1,\cdots,r, form a basis for \{p\}\times \mathbb{R}^r for each p\in M, and so the sections of E e_i(p)=\phi^{-1}(p,v_i),\quad i=1,\cdots,r, provide a basis for E_p at each point p\in M.
Conversely, suppose e_1,\cdots,e_r is a frame for E\rightarrow M. Then every point e\in E is a linear combination e=\sum a^ie_i. The map \phi(e)=(\pi(e),a^1,\cdots,a^r):E\rightarrow M\times \mathbb{R}^r is a bundle map with inverse \psi:M\times \mathbb{R}^r\rightarrow E,\\ \psi(p,a^1,\cdots,a^r)=\sum a^i(p)e_i(p).
It follows from this proposition that over any trivializing open set U of a vector bundle Em there is always a frame.
7.7 \mathcal{F}-Linearity and Bundle Maps
Throughout this subsection, E and F are C^\infty vector bundles over a manifold M, and \mathcal{F}=C^\infty(M) is the ring of C^\infty real-valued functions on M. We will show that an \mathcal{F}-linear map \alpha:\Gamma(E)\rightarrow \Gamma(F) can be defined pointwise and therefore corresponds uniquely to a bundle map E\rightarrow F.
Lemma 7.23 (\mathcal{F}-linear Section Map is Point Operator). An \mathcal{F}-linear map \alpha:\Gamma(E)\rightarrow \Gamma(F) is a point operator.
Proof. We need to show that if s\in \Gamma(E) vanishes at a point p in M, then \alpha(s)\in \Gamma(F) also vanishes at p. Let U be an open neighborhood of p over which E is trivial. Thus, over U it is possible to find a frame e_1,\cdots,e_r for E. We write s|_U=\sum a^ie_i,\quad a^i\in C^\infty(U)=\mathcal{F}(U). Because s vanishes at p, all a^i(p)=0. Since \alpha is \mathcal{F}-linear, it is a local operator (Proposition 7.17) and by Theorem 7.20 its restriction \alpha_U:\Gamma(U,E)\rightarrow \Gamma(U,F) is defined. Then \alpha(s)(p)=\alpha_U(s|_U)(p)=\alpha_U\left( \sum a^ie_i \right) (p)=\left( \sum a^i\alpha_U(e_i)\right) (p) = \sum a^i(p)\alpha_U(e_i)(p)=0.
Lemma 7.24 (Fibre-Preserving Map and Section Map). Let E and F be C^\infty vector bundles over a manifold M. A fiber-preserving map \varphi:E\rightarrow F that is linear on each fiber is C^\infty if and only if \varphi_\# takes C^\infty sections of E to C^\infty sections of F.
Proof. (\Rightarrow) If \varphi is C^\infty, then \varphi_\#(s)=\varphi\circ s clearly takes a C^\infty section s of E to a C^\infty section of F.
(\Leftarrow) Fix p\in M and let (U,x^1,\cdots,x^n) be a chart about p over which E and F are both trivial. Let e_1,\cdots,e_r\in \Gamma(E) be a frame for E over U. Likewise, let f_1,\cdots,f_m\in \Gamma(F) be a frame for F over U. A point of E|_U can be written as a unique linear combination \sum a^je_j. Suppose \varphi\circ e_j=\sum_i b^i_jf_i. In this expression the b^i_j's are C^\infty functions on U, because by hypothesis \varphi\circ e_j=\varphi_\#(e_j) is a C^\infty section of F. Then \varphi\circ \left( \sum_j a^j e_j\right) = \sum_{i,j}a^jb^i_jf_i. One can take local coordinates on E|_U to be (x^1,\cdots,x^n,a^1,\cdots,a^r). In terms of these local coordinates, \varphi(x^1,\cdots,x^n,a^1,\cdots,a^r)=\left( x^1,\cdots,x^n,\sum_j a^jb^1_j,\cdots,\sum_j a^jb^m_j\right), which is a C^\infty map.
Proposition 7.25 (Existence and Uniqueness of Linear Bundle Map). If \alpha: \Gamma(E)\rightarrow \Gamma(F) is \mathcal{F}-linear, then for each p\in M, there is a unique linear map \varphi_p:E_p\rightarrow F_p such that for all s\in \Gamma(E), \varphi_p\left( s(p)\right)=\alpha(s)(p).
Proof. Given e\in E_p, to define \varphi_p(e), choose any section s\in \Gamma(E) such that s(p)=e and define \varphi_p(e)=\alpha(s)(p)\in F_p. This definition is independent of the choice of the section s, because if s' is another section of E with s'(p)=e, then (s-s')(p)=0 and so by Lemma 7.23, we have \alpha(s-s')(p)=0, i.e., \alpha(s)(p)=\alpha(s')(p).
Let us show that \varphi_p:E_p\rightarrow F_p is linear. Suppose e_1,e_2\in E_p and a_1,a_2\in \mathbb{R}. Let s_1,s_2 be global sections of E such that s_i(p)=e_i. Then (a_1s_1+a_2s_2)(p)=a_1e_1+a_2e_2, so \varphi_p(a_1e_1+a_2e_2)=\alpha(a_1s_2+a_2s_2)(p)=a_1\alpha(s_1)(p)+a_2\alpha(s_2)(p)=a_1\varphi_p(e_1)+a_2\varphi_p(e_2).
Theorem 7.26 (Bundle Map and \mathcal{F}-linear Section Map). There is a one-to-one correspondence \{\mbox{bundle maps }\varphi:E\rightarrow F\}\leftrightarrow \{ \mathcal{F}\mbox{-linear maps }\alpha:\Gamma(E)\rightarrow \Gamma(F)\},\\ \varphi\mapsto \varphi_\#.
Proof. We first show surjectivity. Suppose \alpha: \Gamma(E)\rightarrow \Gamma(F) is \mathcal{F}-linear. By the preceding proposition, for each p\in M there is a linear map \varphi_p:E_p\rightarrow F_p such that for any s\in \Gamma(E), \varphi_p(s(p))=\alpha(s)(p). Define \varphi:E\rightarrow F by \varphi(e)=\varphi_p(e) if e\in E_p.
For any s\in \Gamma(E) and for every p\in M, \left( \varphi_\#(s)\right)(p)=\varphi \left( s(p)\right)=\alpha(s)(p), which shows that \alpha=\varphi_\#. Since \varphi_\# takes C^\infty sections of E to C^\infty sections of F, by Lemma 7.24 the map \varphi:E\rightarrow F is C^\infty. Thus, \varphi is a bundle map.
Next we prove the injectivity of the correspondence. Suppose \varphi,\psi:E\rightarrow F are two bundel maps such that \varphi_\#=\psi_\#:\Gamma(E)\rightarrow \Gamma(F). For any e\in E_p, choose a section s\in \Gamma(E) such that s(p)=e. Then \varphi(e)=\varphi\left( s(p)\right)=\left( \varphi_\#(s)\right) (p)=\left( \psi_\#(s)\right) (p)=(\psi\circ s)(p)=\psi(e). Hence, \varphi=\psi.
Corollary 7.27 (Defining of 1-Form). An \mathcal{F}-linear map \omega:\mathfrak{X}(M)\rightarrow C^\infty(M) is a C^\infty 1-form on M.
Proof. By proposition 7.25, one can define for each p\in M, a linear map \omega_p:T_pM\rightarrow \mathbb{R} such that for all X\in \mathfrak{X}(M), \omega_p(X_p)=\omega(X)(p). This shows that \omega is a 1-form on M.
For every C^\infty vector field X on M, \omega(X) is a C^\infty function on M. This shows that as a 1-form, \omega is C^\infty.
7.8 Multilinear Maps over Smooth Functions
By Proposition 7.25, if \alpha:\Gamma(E)\rightarrow \Gamma(F) is an \mathcal{F}-linear map of sections of vector bundles over M, then at each p\in M, it is possible to define a linear map \varphi_p:E_p\rightarrow F_p such that for any s\in \Gamma(E), \varphi_p(s(p))=\alpha(s)(p). This can be generalized to \mathcal{F}-multilinear maps.
Proposition 7.28 (Multi-Section Map). Let E,E',F be vector bundles over a manifold M. If \alpha:\Gamma(E)\times \Gamma(E')\rightarrow \Gamma(F) is \mathcal{F}-bilinear, then for each p\in M there is a unique \mathbb{R}-bilinear map \varphi_p:E_p\times E'_p\rightarrow F_p such that for all s\in \Gamma(E) and s'\in \Gamma(E'), \varphi_p(s(p),s'(p))=(\alpha(s,s'))(p).
Since the proof is similar to that of Proposition 7.25.
Of course, Proposition 7.28 generalizes to \mathcal{F}-linear maps with k argyments. Just as in Corollary 7.27, we conclude that if an alternating map \omega:\mathfrak{X}(M)\times \cdots\times \mathfrak{X}(M)(k\mbox{ times})\rightarrow C^\infty(M) is \mathcal{F}-linear in each argument, then \omega induces a k-form \tilde{\omega} on M such that for X_1,\cdots,X_k\in \mathfrak{X}(M), \tilde{\omega}_p(X_{1,p},\cdots,X_{k,p})=(\omega(X_1,\cdots,X_k))(p). It is customary to write the k-form \tilde{\omega} as \omega.