Abstract. A fiber bundle formalizes a deceptively simple idea: attach a copy of some space (a fiber) to every point of a manifold (the base), in a way that looks locally like a product but may be globally twisted. We build up the definitions — fiber bundles, vector bundles, principal bundles, and the frame bundle — and show how they unify familiar physics objects: a vector field is a section of the tangent bundle, a "moving frame" is a local section of the frame bundle, and the failure of a frame to stay constant under transport is encoded by a connection, whose curvature is the field strength of a gauge theory.

§1 Motivation: Why "Just a Vector Space" Isn't Enough

In elementary mechanics and electromagnetism, a vector field $\mathbf{V}(x)$ is usually introduced as an assignment of an arrow to each point of $\mathbb{R}^3$. This works because $\mathbb{R}^3$ comes with a single, global, constant frame $(\hat{\mathbf{e}}_1, \hat{\mathbf{e}}_2, \hat{\mathbf{e}}_3)$: the same three orthonormal directions at every point. Components of $\mathbf{V}$ at different points can be compared directly, because the basis vectors used to express them are literally the same vectors.

On a curved manifold $M$ — the surface of a sphere, spacetime in general relativity, or the configuration space of a mechanical system — there is no such global constant frame. The tangent space $T_pM$ at one point $p$ and the tangent space $T_qM$ at another point $q$ are different vector spaces; there is no canonical identification between them. A "vector field" must therefore be reformulated as a smooth choice of element of $T_pM$ for each $p$, and comparing vectors at different points becomes a nontrivial geometric operation.

Key intuition. A fiber bundle is the bookkeeping device that keeps track of "a copy of some space, attached at each point of $M$, possibly twisted." A vector field is then not a function $M \to \mathbb{R}^3$, but a section of a bundle whose fiber over $p$ is $T_pM$. The question "is this frame constant?" becomes the question "does this bundle admit a flat connection?"

§2 Fiber Bundles: The General Definition

Let $M$ be a smooth manifold (the base space) and $F$ another manifold (the typical fiber). A fiber bundle is a smooth manifold $E$ (the total space) together with a smooth surjective map

\begin{equation} \pi : E \longrightarrow M, \label{eq:projection} \end{equation}

called the projection, such that $E$ is locally a product of $M$ and $F$. Precisely, every point $p \in M$ has an open neighbourhood $U \subset M$ and a diffeomorphism

\begin{equation} \Phi_U : \pi^{-1}(U) \;\xrightarrow{\;\sim\;}\; U \times F, \label{eq:localtriv} \end{equation}

called a local trivialization, satisfying $\mathrm{pr}_1 \circ \Phi_U = \pi$, where $\mathrm{pr}_1$ is projection onto the first factor. The set $E_p \equiv \pi^{-1}(p)$ is the fiber over $p$; the local trivialization restricts to a diffeomorphism $E_p \cong F$ for each $p \in U$.

If $\{U_\alpha\}$ is an open cover of $M$ with local trivializations $\Phi_\alpha$, then on overlaps $U_\alpha \cap U_\beta$ the two trivializations differ by a map into the diffeomorphism group of $F$:

\begin{equation} \Phi_\alpha \circ \Phi_\beta^{-1}(p, f) = \big(p,\, g_{\alpha\beta}(p)\cdot f\big), \qquad g_{\alpha\beta} : U_\alpha \cap U_\beta \to \mathrm{Diff}(F). \label{eq:transition} \end{equation}

The functions $g_{\alpha\beta}$ are the transition functions of the bundle. They satisfy the cocycle condition $g_{\alpha\beta}g_{\beta\gamma} = g_{\alpha\gamma}$ on triple overlaps. A bundle is trivial ($E \cong M \times F$ globally) precisely when the $g_{\alpha\beta}$ can be removed by a global change of trivialization; otherwise the bundle is twisted, and the transition functions encode exactly how.

A section of $\pi: E \to M$ is a smooth map $s: M \to E$ with $\pi \circ s = \mathrm{id}_M$ — a continuous choice of one element of $E_p$ for every $p$. Sections are the rigorous incarnation of "fields": the bundle is the space of all possible field configurations at every point, and a section is one particular field.

Local vs. global. Locally, every fiber bundle looks trivial — that's the content of $\Phi_U$. All the interesting topology and physics (monopoles, instantons, the Aharonov–Bohm effect, the impossibility of combing a hairy ball) lives in the transition functions $g_{\alpha\beta}$, i.e. in how the local pieces are glued.

§3 Vector Bundles and Vector Fields as Sections

A vector bundle of rank $n$ is a fiber bundle $\pi: E \to M$ whose typical fiber is $\mathbb{R}^n$ (or $\mathbb{C}^n$) and whose transition functions take values in $GL(n,\mathbb{R})$ (or $GL(n,\mathbb{C})$) acting linearly:

\begin{equation} g_{\alpha\beta} : U_\alpha \cap U_\beta \longrightarrow GL(n,\mathbb{R}). \label{eq:vbtransition} \end{equation}

The linearity means each fiber $E_p$ inherits the structure of a genuine $n$-dimensional vector space — addition and scalar multiplication of elements of $E_p$ are well-defined, because the transition functions are linear isomorphisms between the local copies of $\mathbb{R}^n$.

With this language, a vector field on $M$ is precisely a smooth section of a vector bundle $E \to M$. In a local trivialization over $U_\alpha$, the section is represented by a smooth function $V_\alpha : U_\alpha \to \mathbb{R}^n$; on the overlap $U_\alpha \cap U_\beta$, the same physical section is represented in the two trivializations by component functions related as

\begin{equation} V_\alpha(p) = g_{\alpha\beta}(p)\, V_\beta(p). \label{eq:components} \end{equation}

This is the modern statement of the elementary rule "vector components transform contravariantly under a change of basis" — except now $g_{\alpha\beta}(p)$ can vary smoothly from point to point, because the "basis" itself is allowed to vary.

§4 The Tangent Bundle and Transition Functions

The most important vector bundle in physics is the tangent bundle $TM$, defined as the disjoint union of all tangent spaces,

\begin{equation} TM = \bigsqcup_{p \in M} T_pM, \qquad \pi(v) = p \ \text{ for } v \in T_pM. \label{eq:tangentbundle} \end{equation}

A coordinate chart $x^\mu : U \to \mathbb{R}^n$ on $M$ provides a local trivialization of $TM$ over $U$, by associating to $v \in T_pM$ its components $v^\mu$ in the coordinate basis $\{\partial/\partial x^\mu\}_p$. If $y^\nu(x)$ is a second chart on $U' $ overlapping $U$, the two component descriptions of the same vector $v$ are related by the Jacobian matrix:

\begin{equation} v^{\nu}_{(y)} = \frac{\partial y^\nu}{\partial x^\mu}\Big|_p \, v^{\mu}_{(x)}. \label{eq:jacobian} \end{equation}

Thus for the tangent bundle, the transition functions $g_{\alpha\beta}$ of Eq. \eqref{eq:vbtransition} are precisely the Jacobian matrices $\partial y^\nu/\partial x^\mu$ of the coordinate change. A "vector field" $V = V^\mu \partial_\mu$ is a section of $TM$: at every point it is a tangent vector, but its component functions $V^\mu(x)$ depend on the choice of chart according to Eq. \eqref{eq:jacobian}.

Cotangent and tensor bundles. The same construction with the dual spaces $T_p^*M$ gives the cotangent bundle $T^*M$, whose sections are one-forms, with transition functions given by the inverse-transpose Jacobian. Tensor products $TM^{\otimes r} \otimes T^*M^{\otimes s}$ give bundles whose sections are tensor fields of type $(r,s)$ — all governed by the same gluing logic.

§5 The Frame Bundle: Framing Every Point

A frame at $p \in M$ is an ordered basis $(e_1, \dots, e_n)$ of $T_pM$. The set of all frames at all points of $M$ assembles into the frame bundle $FM$:

\begin{equation} FM = \bigsqcup_{p \in M} \big\{\, (e_1,\dots,e_n) \ \big|\ (e_i) \text{ is a basis of } T_pM \,\big\}, \qquad \pi(e_1,\dots,e_n) = p. \label{eq:framebundle} \end{equation}

The fiber $\pi^{-1}(p)$ is the set of all bases of $T_pM$. Crucially, this fiber is not a vector space: there is no natural "zero frame" or way to add two frames. Instead, $GL(n,\mathbb{R})$ acts freely and transitively on the fiber by changing basis,

\begin{equation} (e_1,\dots,e_n)\cdot A = \Big(\textstyle\sum_i e_i A^i{}_1,\ \dots,\ \sum_i e_i A^i{}_n\Big), \qquad A \in GL(n,\mathbb{R}), \label{eq:framerel} \end{equation}

so that, having fixed any one frame at $p$, every other frame at $p$ is obtained uniquely by acting with some $A \in GL(n,\mathbb{R})$. This makes $FM$ the prototypical example of a principal bundle with structure group $GL(n,\mathbb{R})$, defined precisely in §6.

A moving frame (or vielbein, or repère mobile) on an open set $U \subset M$ is a smooth choice of basis $(e_1(p),\dots,e_n(p))$ for every $p \in U$ — that is, a local section $s: U \to FM$ of the frame bundle. Equivalently, it is $n$ vector fields $e_1,\dots,e_n$ on $U$ that are linearly independent at every point.

The Euclidean case revisited. In $\mathbb{R}^n$ with the standard flat metric, the constant Cartesian frame $(\hat{e}_1,\dots,\hat{e}_n)$ is a global section of $FM$ that is moreover covariantly constant — parallel transporting $\hat{e}_i$ along any curve returns the same vector $\hat{e}_i$. This single global, parallel section is what allows components of vectors at different points to be compared directly. On a general manifold, $FM$ may admit local sections (moving frames) but typically no global section, and even a local section is generally not parallel — this failure is measured by the connection, introduced next.

§6 Principal Bundles and the Structure Group

A principal $G$-bundle is a fiber bundle $\pi: P \to M$ with typical fiber a Lie group $G$, equipped with a smooth right action

\begin{equation} P \times G \to P, \qquad (u,g) \mapsto u\cdot g, \label{eq:principalaction} \end{equation}

that is free (only the identity has a fixed point) and acts transitively on each fiber, with $\pi(u\cdot g) = \pi(u)$. The transition functions $g_{\alpha\beta}: U_\alpha \cap U_\beta \to G$ act on $P$ by left multiplication, and $G$ is called the structure group.

The frame bundle $FM$ of §5 is the principal $GL(n,\mathbb{R})$-bundle naturally associated with $TM$. More generally, given any vector bundle $E \to M$ of rank $n$ with structure group $G \subseteq GL(n,\mathbb{R})$, one can construct a principal $G$-bundle $P \to M$ (the frame bundle of $E$) whose fiber over $p$ consists of the linear isomorphisms $\mathbb{R}^n \to E_p$ compatible with whatever extra structure (orthonormality, orientation, ...) reduces the structure group to $G$. Conversely, given a principal $G$-bundle $P$ and a representation $\rho: G \to GL(V)$, one builds the associated vector bundle

\begin{equation} E = P \times_\rho V \;=\; \big(P \times V\big)\big/{\sim}, \qquad (u,v)\sim(u\cdot g,\, \rho(g)^{-1}v). \label{eq:associated} \end{equation}

This dictionary — principal bundle with structure group $G$ $\leftrightarrow$ vector bundles via representations of $G$ — is the precise mathematical content behind the statement "matter fields transform in representations of the gauge group."

Reduction of structure group. Choosing a Riemannian metric on $M$ reduces the structure group of $FM$ from $GL(n,\mathbb{R})$ to $O(n)$ — one restricts to orthonormal frames, giving the orthonormal frame bundle $OM$. Choosing also an orientation reduces further to $SO(n)$. In general relativity, choosing a spin structure further lifts $SO(3,1)$ to $\mathrm{Spin}(3,1) \cong SL(2,\mathbb{C})$, the bundle whose associated spinor representation describes fermions.

§7 Connections: Comparing Frames at Nearby Points

The local trivializations of a fiber bundle let us compare points within a single fiber via the structure group, but they give no canonical way to compare an element of $E_p$ with an element of $E_q$ for $p \neq q$ — the bundle, by itself, has no notion of "the same vector, translated to a nearby point." A connection supplies exactly this: an infinitesimal rule for identifying nearby fibers.

Concretely, on a vector bundle $E \to M$, a connection is a differential operator

\begin{equation} \nabla : \Gamma(E) \longrightarrow \Gamma(T^*M \otimes E), \label{eq:connectionop} \end{equation}

satisfying the Leibniz rule $\nabla(fs) = df\otimes s + f\nabla s$ for any smooth function $f$ and section $s$. In a local frame $(e_1,\dots,e_n)$ — that is, a local section of the frame bundle $FM$ — the covariant derivative of each frame vector is itself expanded in the same frame:

\begin{equation} \nabla_\mu e_i = \omega^j{}_{i\mu}\, e_j, \label{eq:connectioncoeff} \end{equation}

defining the connection coefficients (or connection one-form) $\omega^j{}_{i\mu}$, valued in the Lie algebra $\mathfrak{g}$ of the structure group. For the tangent bundle with the coordinate frame, these are the familiar Christoffel symbols $\Gamma^\nu_{\mu\lambda}$. The covariant derivative of a section $V = V^i e_i$ then follows from the Leibniz rule:

\begin{equation} \nabla_\mu V = \big(\partial_\mu V^i + \omega^i{}_{j\mu} V^j\big)\, e_i. \label{eq:covderiv} \end{equation}

Under a change of frame $e_i \to e_i' = e_j (A^{-1})^j{}_i$ given by $A: U \to G$, the connection coefficients transform inhomogeneously,

\begin{equation} \omega' = A\,\omega\, A^{-1} + A\, dA^{-1}, \label{eq:gaugetransform} \end{equation}

which is exactly the gauge transformation law of a Yang–Mills connection $A_\mu \to A A_\mu A^{-1} - (\partial_\mu A) A^{-1}$. The connection one-form is, mathematically, the same object as the gauge potential.

§8 Parallel Transport and "Constant" Frames

Given a connection, a section $V$ is said to be parallel along a curve $\gamma(t)$ if $\nabla_{\dot\gamma} V = 0$ along $\gamma$. Solving this first-order ODE for $V(t)$ given an initial condition $V(0) \in E_{\gamma(0)}$ defines the parallel transport map

\begin{equation} \mathcal{P}_\gamma : E_{\gamma(0)} \longrightarrow E_{\gamma(1)}, \label{eq:paralleltransport} \end{equation}

a linear isomorphism between the fibers at the endpoints of $\gamma$. This is precisely the rigorous replacement for "moving a vector to a nearby point without changing it." A frame $(e_1,\dots,e_n)$ is a parallel frame along $\gamma$ if every $e_i$ is parallel-transported into itself, i.e. $\omega_{\mu}\dot\gamma^\mu = 0$ along $\gamma$ in that frame — equivalently, the connection coefficients vanish identically in that frame along the curve.

In flat $\mathbb{R}^n$ with the Levi-Civita connection of the Euclidean metric, the Christoffel symbols vanish identically in Cartesian coordinates: $\Gamma^\nu_{\mu\lambda}=0$ everywhere. The Cartesian frame is therefore parallel along every curve and at every point simultaneously — it is a single global parallel section of $FM$. This is the rigorous meaning of "a moving frame that is constant in Euclidean space": constancy is the statement that the frame is covariantly constant for the (flat) connection, which happens to coincide with ordinary constancy because $\omega \equiv 0$ in that frame.

Path dependence on a curved manifold. On a curved manifold there is generally no frame for which $\omega \equiv 0$ everywhere. Parallel-transporting a vector around a closed loop $\gamma$ returns a different vector, related to the original by an element $\mathcal{P}_\gamma \in G$ — the holonomy of the loop. The set of all holonomies forms the holonomy group, a subgroup of the structure group measuring how far the bundle is from admitting a global parallel frame.

§9 Curvature: The Field Strength of a Connection

The infinitesimal version of holonomy around an infinitesimal loop is the curvature of the connection, defined as the failure of covariant derivatives to commute:

\begin{equation} F_{\mu\nu} \equiv [\nabla_\mu, \nabla_\nu] = \partial_\mu \omega_\nu - \partial_\nu \omega_\mu + [\omega_\mu, \omega_\nu], \label{eq:curvature} \end{equation}

a $\mathfrak{g}$-valued two-form on $M$. For the Levi-Civita connection on $TM$, $F_{\mu\nu}$ is (up to index conventions) the Riemann curvature tensor $R^\rho{}_{\sigma\mu\nu}$. For a $U(1)$ principal bundle with its associated line bundle, $F_{\mu\nu}$ is the electromagnetic field-strength tensor; for a general $G$, it is the Yang–Mills field strength. In every case, $F_{\mu\nu}=0$ everywhere is precisely the condition for the connection to be flat — for a frame to exist that is parallel along every curve, i.e. for a genuinely "constant" frame to be globally definable, at least locally (global flatness additionally requires trivial holonomy around non-contractible loops).

Bianchi identity. Curvature is not arbitrary: it satisfies $\nabla_{[\mu}F_{\nu\rho]} = 0$, the Bianchi identity, which in electromagnetism is equivalent to two of Maxwell's four equations ($\nabla\cdot\mathbf{B}=0$ and Faraday's law). The Bianchi identity is a purely geometric fact about curvature two-forms, independent of any equation of motion.

§10 Physical Applications

The dictionary built above translates directly into the major frameworks of theoretical physics:

Bundle-theoretic object Physical incarnation
Tangent bundle $TM$, section Velocity / vector field on configuration or spacetime
Frame bundle $FM$, section Local rest frames / vielbeins for observers
Levi-Civita connection on $TM$ Gravitational field (general relativity)
Connection on principal $U(1)$-bundle Electromagnetic vector potential $A_\mu$
Connection on principal $SU(N)$-bundle Yang–Mills gauge field (strong/weak forces)
Curvature $F_{\mu\nu}$ Riemann tensor / field-strength tensor
Holonomy of a closed loop Berry phase, Aharonov–Bohm phase, geodesic deviation

Returning to the Bloch-electron picture familiar from band theory: the Bloch wavevector $k$ ranging over the Brillouin zone parametrizes a base manifold (a torus), and the cell-periodic functions $u_{nk}$ at each $k$ span a Hilbert-space fiber. The Berry connection $\mathcal{A}_n(k) = i\langle u_{nk}|\nabla_k|u_{nk}\rangle$ is exactly a connection one-form on this bundle in the sense of Eq. \eqref{eq:connectionop}, and its curvature — the Berry curvature — integrates over closed surfaces in the Brillouin zone to give the topological invariants (Chern numbers) underlying the quantum Hall effect. Fiber bundles thus provide the single geometric language spanning gravity, gauge theory, and the topology of band structures.

Outlook. Everything above generalises by replacing $GL(n,\mathbb{R})$ with an arbitrary Lie group $G$ and the vector-space fiber with an arbitrary manifold on which $G$ acts. Characteristic classes (Chern, Pontryagin, Euler classes) — built from polynomials in the curvature $F$ — classify bundles up to topological equivalence and underlie anomaly cancellation, the index theorems relating geometry to the spectrum of Dirac operators, and the classification of topological phases of matter.

References

  1. Nakahara, M. (2003). Geometry, Topology and Physics, 2nd ed. IOP Publishing.
  2. Frankel, T. (2011). The Geometry of Physics: An Introduction, 3rd ed. Cambridge University Press.
  3. Kobayashi, S. & Nomizu, K. (1963). Foundations of Differential Geometry, Vol. 1. Wiley.
  4. Eguchi, T., Gilkey, P. B. & Hanson, A. J. (1980). Gravitation, Gauge Theories and Differential Geometry. Physics Reports, 66(6), 213–393.
  5. Xiao, D., Chang, M.-C. & Niu, Q. (2010). Berry phase effects on electronic properties. Reviews of Modern Physics, 82, 1959.
  6. Bleecker, D. (1981). Gauge Theory and Variational Principles. Addison-Wesley.