Vector bundles answer a basic question that appears whenever geometry carries data from point to point: how can a space have a [vector space](/page/Vector%20Space) attached at every point, while still allowing those vector spaces to twist as the point moves? A tangent plane to a surface, a line of possible phases over each point, or a family of solution spaces over parameters all look locally like a product, but the local products may glue together in a way that has global content. The first failure to keep in mind is that local coordinates do not give global coordinates for free. A circle has a tangent line at every point, and each small arc makes the tangent lines look like the arc times $\mathbb{R}$. On a more complicated base, the local identifications can rotate, reflect, or mix coordinates as they pass from one chart to another. The theory of vector bundles records exactly this local-product structure and the rules by which the local products are compared. [example: The Mobius Line Bundle] Start with $[0,1]\times \mathbb{R}$ and impose the [equivalence relation](/page/Equivalence%20Relation) generated by \begin{align*} (0,v)\sim (1,-v). \end{align*} The base is obtained by identifying $0$ with $1$, so it is a circle, and the fiber over the identified endpoint is not two copies of $\mathbb{R}$: the point represented by $(0,v)$ is the same fiber vector as the point represented by $(1,-v)$. On an open arc that does not contain the identified endpoint, the bundle is just the product arc $\times \mathbb{R}$. On an arc crossing the endpoint, use the coordinate $t$ near $0$ on one side and $t-1$ near $1$ on the other side; the fiber coordinate changes by \begin{align*} v\mapsto -v. \end{align*} Thus the transition function on the overlap is the constant element $-1\in \operatorname{GL}_1(\mathbb{R})$. If a continuous nonzero section existed, cutting the circle open at the endpoint would represent it by a [continuous function](/page/Continuous%20Function) \begin{align*} f:[0,1]\to \mathbb{R}\setminus\{0\} \end{align*} satisfying the endpoint compatibility condition \begin{align*} f(1)=-f(0). \end{align*} Since $[0,1]$ is connected and $f$ never equals $0$, the image of $f$ lies entirely in $(0,\infty)$ or entirely in $(-\infty,0)$. Hence $f(0)$ and $f(1)$ have the same sign, while the equation $f(1)=-f(0)$ forces them to have opposite signs. This contradiction shows that the bundle is locally a product but globally twisted. [/example] The Mobius example is not a defect in the definition. It is the point of the definition: a vector bundle should forget unnecessary coordinates inside each fiber while remembering how those coordinates transform from one local patch to another. The base space controls where the fibers live, the total space contains all vectors in all fibers, and the projection records the point under each vector. ## Definition The central definition separates the geometric object into three pieces. The total space is where the moving vectors live, the base is the space over which they move, and the projection assigns every vector to its base point. [definition: Real Vector Bundle] A real vector bundle of rank $r\in \mathbb{N}$ over a [topological space](/page/Topological%20Space) $B$ consists of a topological space $E$, a continuous map $\pi:E\to B$, and for each point $b\in B$ a real vector space structure on the fiber \begin{align*} E_b&=\pi^{-1}(\{b\}) \end{align*} such that every $b\in B$ has an open neighbourhood $U\subset B$ and a homeomorphism \begin{align*} \Phi_U:\pi^{-1}(U)&\to U\times \mathbb{R}^r \end{align*} with the property that the first projection satisfies $\operatorname{pr}_1\circ \Phi_U=\pi$ on $\pi^{-1}(U)$ and each restricted map $E_b\to \{b\}\times \mathbb{R}^r$ is a linear isomorphism. [/definition] The definition is deliberately local. It does not say that all fibers have been identified with one fixed copy of $\mathbb{R}^r$ once and for all; it says only that this can be done near each base point. The rest of the theory studies how these temporary identifications compare. ## Local Coordinates and Gluing ### Local Trivializations The bundle definition guarantees product coordinates only one [open set](/page/Open%20Set) at a time. To compare fibers, write sections locally, or build new bundles, one must refer to a chosen product coordinate system over a particular patch, not merely to the existence of such a system. The following definition names this chosen coordinate system. [definition: Local Trivialization] Let $\pi:E\to B$ be a real vector bundle of rank $r$. A local trivialization over an open set $U\subset B$ is a homeomorphism \begin{align*} \Phi_U:\pi^{-1}(U)&\to U\times \mathbb{R}^r \end{align*} such that $\operatorname{pr}_1\circ \Phi_U=\pi$ and, for every $b\in U$, the map $E_b\to \mathbb{R}^r$ obtained by projecting to the second factor is linear. [/definition] ### Transition Functions Local trivializations are not unique, so the same vector in a fiber may have two different coordinate columns on an overlap. The obstruction to treating all local products as one global product is hidden in how these coordinate columns change from one patch to another. The following definition packages that change of coordinates as a matrix-valued function. [definition: Transition Function] Let $\Phi_U$ and $\Phi_V$ be local trivializations of a rank $r$ real vector bundle $\pi:E\to B$. On an overlap $U\cap V$, the transition function $g_{VU}:U\cap V\to \operatorname{GL}_r(\mathbb{R})$ is defined by the identity \begin{align*} \Phi_V\circ \Phi_U^{-1}(b,x)&=(b,g_{VU}(b)x) \end{align*} for $b\in U\cap V$ and $x\in \mathbb{R}^r$. [/definition] The transition functions are not extra decoration. If local products are to glue into one total space, then changing coordinates from $U$ to $V$ and then from $V$ to $W$ must agree with changing directly from $U$ to $W$. Thus the essential question is which algebraic rules on overlaps are exactly strong enough to reconstruct a vector bundle from local product pieces. [quotetheorem:6080] The theorem says that vector bundles can be encoded by linear gluing data. This viewpoint is often the most efficient one: construct a bundle by writing compatible matrices on overlaps, then recover the total space by gluing the local products. ### Tangent Lines as Local Products The easiest geometric test case is a bundle whose fibers we already understand: tangent lines to a curve. The point of the example is not that anything mysterious happens on the circle, but that the transition functions arise from comparing ordinary coordinate descriptions of the same moving line. [example: Tangent Bundle of the Circle] Let $S^1\subset \mathbb{R}^2$ and write $p(t)=(\cos t,\sin t)$. The vector \begin{align*} e(t)=(-\sin t,\cos t) \end{align*} is tangent to $S^1$ at $p(t)$ because it is the derivative of the parametrization: \begin{align*} p'(t)=\left(\frac{d}{dt}\cos t,\frac{d}{dt}\sin t\right)=(-\sin t,\cos t)=e(t). \end{align*} Thus $T_{p(t)}S^1=\operatorname{span}_{\mathbb{R}}\{e(t)\}$, and every tangent vector at $p(t)$ has a unique form $a e(t)$ with $a\in\mathbb{R}$. On an arc with angle coordinate $t$, the local trivialization is \begin{align*} \Phi_t(p(t),a e(t))=(p(t),a). \end{align*} Now take another coordinate $u$ on an overlapping arc, with $t=\theta(u)$ and $\theta'(u)\neq 0$. The same point is \begin{align*} p(\theta(u))=(\cos \theta(u),\sin \theta(u)). \end{align*} The coordinate tangent vector in the $u$-coordinate is \begin{align*} \frac{d}{du}p(\theta(u))=(-\sin \theta(u)\theta'(u),\cos \theta(u)\theta'(u))=\theta'(u)e(\theta(u)). \end{align*} If the same tangent vector is written as $a e(\theta(u))$ in the $t$-coordinate and as $b\,\frac{d}{du}p(\theta(u))$ in the $u$-coordinate, then \begin{align*} a e(\theta(u))=b\theta'(u)e(\theta(u)). \end{align*} Since $e(\theta(u))\neq 0$, this gives \begin{align*} a=b\theta'(u). \end{align*} Equivalently, \begin{align*} b=(\theta'(u))^{-1}a. \end{align*} Therefore the transition function on the overlap is multiplication by the nonzero continuous function $(\theta'(u))^{-1}$, so it takes values in $\operatorname{GL}_1(\mathbb{R})=\mathbb{R}\setminus\{0\}$. The tangent bundle of $S^1$ is globally a product because the formula \begin{align*} (p(t),a)\mapsto (p(t),a e(t)) \end{align*} is well-defined when $t$ is replaced by $t+2\pi$: both $p(t+2\pi)=p(t)$ and $e(t+2\pi)=e(t)$. This example shows that transition functions are exactly the fiberwise linear changes of scalar coordinate used to describe the same tangent vector on overlapping arcs. [/example] ## Sections and Fiberwise Maps ### Sections A vector bundle is useful because it supports fields of vectors. A section chooses one vector in each fiber, just as a vector field on a manifold chooses one tangent vector at each point. Sections convert the bundle from a static object into a space of geometric data. [definition: Section of a Vector Bundle] Let $\pi:E\to B$ be a vector bundle. A section of $E$ over an open set $U\subset B$ is a continuous map $s:U\to E$ satisfying \begin{align*} \pi\circ s&=\operatorname{id}_U. \end{align*} A global section is a section over $B$. [/definition] In a local trivialization, a section becomes an ordinary function from the base patch to $\mathbb{R}^r$. The difficulty is that two such functions on overlapping patches may describe the same geometric section only after applying the transition function between the two coordinate systems. The local-to-global problem for sections is therefore to decide when these locally written functions transform compatibly on every overlap. [quotetheorem:9947] This result is the basic dictionary between global bundle language and local coordinates. It also explains why a section can fail to exist globally even when local sections are abundant: the local functions must agree after the prescribed fiberwise changes of coordinates. [example: No Nonvanishing Section of the Mobius Bundle] For the Mobius line bundle over $S^1$, cut the circle at the identified endpoint and represent the remaining circle by $t\in[0,1]$. A section over this cut interval is a continuous choice of one scalar in each fiber, so a nonvanishing section would give a continuous function \begin{align*} f:[0,1]\to \mathbb{R}\setminus\{0\}. \end{align*} The gluing rule for the Mobius bundle identifies the endpoint fiber by $(0,v)\sim (1,-v)$, hence the same section must satisfy \begin{align*} f(1)=-f(0). \end{align*} Now set $A=f^{-1}((0,\infty))$ and $B=f^{-1}((-\infty,0))$. Since $f$ is continuous, both $A$ and $B$ are open in $[0,1]$; since $f$ never equals $0$, they are disjoint and satisfy $[0,1]=A\cup B$. The interval $[0,1]$ is connected, so exactly one of $A$ or $B$ is all of $[0,1]$. Therefore $f(0)$ and $f(1)$ have the same sign. But the endpoint equation gives two cases: if $f(0)>0$, then \begin{align*} f(1)=-f(0)<0, \end{align*} and if $f(0)<0$, then \begin{align*} f(1)=-f(0)>0. \end{align*} In both cases $f(0)$ and $f(1)$ have opposite signs, contradicting the connectedness argument. Thus the Mobius line bundle has no global nonvanishing section; the obstruction comes from the sign-changing transition function around the loop. [/example] ### Fiberwise Linear Maps After sections expose individual vectors in each fiber, the next comparison problem is to map every fiber of one bundle into the corresponding fiber of another bundle. This motivates bundle homomorphisms, where continuity of total spaces is paired with linearity on each fiber. [definition: Bundle Homomorphism] Let $\pi_E:E\to B$ and $\pi_F:F\to B$ be real vector bundles over the same base. A bundle homomorphism from $E$ to $F$ is a continuous map $A:E\to F$ such that $\pi_F\circ A=\pi_E$ and each restricted map \begin{align*} A_b:E_b&\to F_b \end{align*} is linear. [/definition] If every fiber map is a linear isomorphism and the total map has a continuous inverse of the same kind, the bundles are isomorphic. For finite-rank vector bundles, this inverse condition is often packaged into the phrase "fiberwise linear isomorphism": local trivializations turn the map into a continuous matrix-valued function with invertible matrices, and the local inverse matrices vary continuously. Stating the inverse condition explicitly is a useful safeguard, since isomorphism is the equality notion that preserves the projection, the topology, and the linear structure of every fiber. ## Constructions from Old Bundles ### Direct Sums Many geometric constructions are performed fiber by fiber. Direct sums, duals, tensor products, and pullbacks start with familiar vector-space operations, then ask whether the resulting family again has compatible local trivializations. [definition: Direct Sum of Vector Bundles] Let $\pi_E:E\to B$ and $\pi_F:F\to B$ be real vector bundles. Their [direct sum](/page/Direct%20Sum) is the vector bundle $E\oplus F\to B$ whose total space is \begin{align*} E\oplus F&=\{(e,f)\in E\times F\mid \pi_E(e)=\pi_F(f)\}, \end{align*} whose projection is \begin{align*} (e,f)&\mapsto \pi_E(e), \end{align*} and whose fiber over $b\in B$ is \begin{align*} (E\oplus F)_b&=E_b\oplus F_b. \end{align*} [/definition] The direct sum turns two independent moving vector spaces into a single larger moving vector space. The induced local trivializations come from taking the direct sum of the local fiber coordinates of $E$ and $F$ over common open sets. ### Dual Bundles Direct sums combine two moving vector spaces, but many geometric operations go in the opposite direction: they test vectors by pairing them with linear functionals. A vector field is measured by a one-form, a differential sends tangent vectors to [real numbers](/page/Real%20Numbers), and local coordinates are often recovered by applying covectors to vectors. If those functionals were chosen independently in each fiber, they would not form a usable geometric object; their variation has to respect the same transition data that made the original bundle locally product-shaped. The dual bundle is the construction that keeps track of these moving dual spaces coherently. [definition: Dual Vector Bundle] Let $\pi:E\to B$ be a real vector bundle. The dual vector bundle is the vector bundle $E^*\to B$ whose total space is the disjoint union \begin{align*} E^*&=\bigsqcup_{b\in B}(E_b)^*, \end{align*} whose projection sends each functional in $(E_b)^*$ to $b$, and whose local trivializations are obtained by dualizing the fiber coordinates of local trivializations of $E$. [/definition] The dual bundle is the natural home for covector fields and bundle-valued linear functionals. If $E$ has transition matrix $g_{VU}(b)$, the dual bundle has transition matrix $(g_{VU}(b)^{-1})^\top$, because linear functionals transform contragrediently. ### Pullbacks Direct sums and duals operate over a fixed base, but geometry often changes the base by a continuous map. The next definition gives the construction that transports the fibers of a bundle on $B$ to a new bundle on a space $X$ mapping into $B$. [definition: Pullback Vector Bundle] Let $f:X\to B$ be a continuous map and let $\pi:E\to B$ be a real vector bundle. The pullback bundle $f^*E\to X$ is the subspace \begin{align*} f^*E&=\{(x,e)\in X\times E\mid f(x)=\pi(e)\} \end{align*} with projection $(x,e)\mapsto x$ and fiberwise vector space operations inherited from the fibers of $E$. [/definition] The pullback fiber over $x$ is canonically the fiber $E_{f(x)}$, but a fiber-by-fiber construction alone does not automatically give a vector bundle topology with local product charts. The point to check is that local trivializations of $E$ over open sets in $B$ induce local trivializations over their inverse images in $X$, with the same rank and compatible linear fiber structure. This verification is what makes pullback a usable operation rather than just a set-theoretic construction. Before using pullbacks to transport bundles along maps, one needs the assurance that the construction really produces another vector bundle over the new base. [quotetheorem:9948] This theorem is often used without comment, but it is a structural reason that bundles appear throughout geometry. Any map into a parameter space imports the vector data living over that parameter space. ## Smooth Vector Bundles ### Smooth Local Triviality When the base is a smooth manifold, the same idea is refined by replacing continuous local trivializations with smooth ones. The total space becomes a smooth manifold, and the transition functions are smooth maps into a [general linear group](/page/General%20Linear%20Group). [definition: Smooth Vector Bundle] Let $M$ be a smooth manifold. A smooth real vector bundle of rank $r$ over $M$ is a real vector bundle $\pi:E\to M$ such that $E$ is a smooth manifold, $\pi$ is a smooth map, and every point of $M$ has a local trivialization whose total map and inverse are smooth and whose fiber maps are linear. [/definition] In a chart $(U,\varphi)$ on $M$ and a local trivialization of $E$ over $U$, a smooth section is represented by a smooth map from $\varphi(U)\subset \mathbb{R}^n$ to $\mathbb{R}^r$. This local description lets differential calculus act on sections once additional structure, such as a connection, is supplied. ### Tangent and Cotangent Bundles On a smooth manifold, each point has its own tangent space, but these spaces are initially separate vector spaces rather than one global object. To treat vector fields as sections and apply smooth calculus to them, the tangent spaces must be assembled into a smooth vector bundle whose local coordinates come from manifold charts. The following definition names that assembled bundle. [definition: Tangent Bundle] Let $M$ be a smooth manifold. The tangent bundle is the smooth vector bundle $TM\to M$ whose total space is \begin{align*} TM&=\bigsqcup_{p\in M}T_pM, \end{align*} whose projection sends each tangent vector in $T_pM$ to $p$, and whose smooth vector bundle structure is induced by coordinate-chart trivializations of tangent vectors. [/definition] Once tangent vectors are organized as a smooth bundle, the next natural bundle is obtained by taking dual spaces in every fiber. The cotangent bundle packages differential one-forms, which evaluate vector fields and support the language of integration and differential equations on manifolds. [definition: Cotangent Bundle] Let $M$ be a smooth manifold. The cotangent bundle is the smooth vector bundle $T^*M\to M$ whose total space is \begin{align*} T^*M&=\bigsqcup_{p\in M}T_p^*M, \end{align*} whose projection sends each covector in $T_p^*M$ to $p$, and whose smooth vector bundle structure is the dual vector bundle structure of $TM\to M$. [/definition] ## Global Product Form and Obstructions ### Global Product Bundles Every vector bundle is locally a product, but the local coordinate choices may twist so that no single product description works over the whole base. The global question is whether all fibers can be identified coherently with one fixed copy of $\mathbb{R}^r$ while respecting the projection to the base. The following definition records that stronger condition. [definition: Globally Product Vector Bundle] A rank $r$ vector bundle $\pi:E\to B$ is globally product if there exists a bundle isomorphism \begin{align*} E&\cong B\times \mathbb{R}^r \end{align*} over $B$, where $B\times \mathbb{R}^r\to B$ is the projection onto the first factor. [/definition] ### Global Frames To prove that a bundle is globally product, it is not enough to know that each fiber is individually isomorphic to $\mathbb{R}^r$; those isomorphisms must vary continuously over the base. A practical way to test this is to ask whether one can choose $r$ global sections that form a basis in every fiber at the same time. Such a global frame is exactly the data needed to build a product identification. In the criterion below, $\Gamma(E)$ denotes the space of smooth sections of the vector bundle $E$. [quotetheorem:6136] This criterion explains why the Mobius bundle is the first meaningful example. A nonzero section of a line bundle is already a frame, so the sign reversal around the circle blocks global product form. Characteristic classes, orientability, and connections refine this obstruction theory, but the starting point remains the same local-to-global question: can the local vector coordinates be chosen coherently on the entire base? ## Beyond and Connected Topics Vector bundles sit near several other parts of geometry. The tangent and cotangent bundles connect this page to [Smooth Manifold](/page/Smooth%20Manifold) and the course-level treatment in [Cambridge III Differential Geometry](/page/Cambridge%20III%20Differential%20Geometry), while transition functions and cocycles place them in the broader local-to-global language of gluing. In each case, the same pattern appears: local data are simple, but global compatibility carries information. For analysis on manifolds, smooth sections of vector bundles supply the spaces on which differential operators act. The Riemannian setting developed in [Cambridge III Riemannian Geometry](/page/Cambridge%20III%20Riemannian%20Geometry) uses tangent and cotangent bundles as basic objects: a connection differentiates sections, a metric measures them fiber by fiber, and curvature records the failure of local parallel choices to be globally compatible. ## References - Androma, [Fibre Bundles I: Bundles, Sections, and Transition Data](/page/Fibre%20Bundles%20I%3A%20Bundles%2C%20Sections%2C%20and%20Transition%20Data). - Androma, [Cambridge III Differential Geometry](/page/Cambridge%20III%20Differential%20Geometry). - Androma, [Cambridge III Riemannian Geometry](/page/Cambridge%20III%20Riemannian%20Geometry). - Androma, [Smooth Manifold](/page/Smooth%20Manifold). - Androma, [Several Complex Variables IV: Complex Geometry and Curvature](/page/Several%20Complex%20Variables%20IV%3A%20Complex%20Geometry%20and%20Curvature). - Androma, [Several Complex Variables V: CR Geometry and Boundary Behavior](/page/Several%20Complex%20Variables%20V%3A%20CR%20Geometry%20and%20Boundary%20Behavior). - John M. Lee, *Introduction to Smooth Manifolds* (2013). - Dale Husemoller, *Fibre Bundles* (1994). - Raoul Bott and Loring W. Tu, *Differential Forms in Algebraic Topology* (1982).