Fact-checked by Grok 2 weeks ago

Parallel transport

In differential geometry, parallel transport is a method for moving tangent vectors along smooth curves on a manifold such that the vector remains "parallel" with respect to a given affine connection, meaning its covariant derivative along the curve vanishes.^[1] This process defines a unique vector field along the curve that starts with a given initial vector and satisfies the condition \frac{DV}{dt} = 0, where V is the vector field and \frac{D}{dt} denotes the covariant derivative along the curve.^[1] Formally introduced by Tullio Levi-Civita in 1917 as part of the development of the Levi-Civita connection on Riemannian manifolds, parallel transport builds on earlier ideas from Bernhard Riemann's 1854 work on metric geometry and Gregorio Ricci-Curbastro's absolute differential calculus.^[2] It provides an isometry between tangent spaces at the curve's endpoints when the connection is metric-compatible, enabling comparisons of geometric objects across different points on the manifold.^[1] A key consequence is its role in defining holonomy, the transformation resulting from transporting vectors around closed loops, which measures the intrinsic curvature of the manifold and is central to understanding phenomena like geodesic deviation.^[3] In broader contexts, such as principal bundles, parallel transport encodes the action of a connection by mapping fibers over path endpoints via horizontal lifts, facilitating applications in gauge theory and geometric statistics.^[3]

Fundamentals of Parallel Transport

Definition on Manifolds

A smooth manifold M is a topological space locally diffeomorphic to Euclidean space \mathbb{R}^n, equipped with an atlas of charts that ensure transition maps are smooth. At each point p \in M, the tangent space T_p M consists of all tangent vectors at p, which can be identified with the derivations of the space of smooth functions C^\infty(M) at p or, equivalently, the velocities of smooth curves passing through p.^[4] The tangent bundle TM = \bigcup_{p \in M} T_p M assembles these vector spaces into a smooth vector bundle over M.^[5] An affine connection \nabla on M is a rule for differentiating vector fields, formally defined as an \mathbb{R}-bilinear map \nabla: \mathfrak{X}(M) \times \mathfrak{X}(M) \to \mathfrak{X}(M), where \mathfrak{X}(M) denotes the space of smooth vector fields on M, satisfying the Leibniz rules \nabla_{fX} Y = f \nabla_X Y and \nabla_X (f Y) = X(f) Y + f \nabla_X Y for all smooth vector fields X, Y \in \mathfrak{X}(M) and smooth functions f \in C^\infty(M).^[4] This connection extends to define the covariant derivative of sections of tensor bundles and provides a means to compare vectors from different tangent spaces.^[5] Given a smooth curve \gamma: I \to M with I \subset \mathbb{R} an interval, a vector field V along \gamma—meaning V(t) \in T_{\gamma(t)} M for each t \in I—is said to be parallel along \gamma if its covariant derivative along the curve vanishes, i.e., \nabla_{\gamma'(t)} V = 0 for all t \in I.^[4] This condition encodes the notion that V does not "twist" relative to the connection as it is transported along the curve.^[5] For a smooth curve \gamma: [a, b] \to M and points s, t \in [a, b] with \gamma(s) = p and \gamma(t) = q, the parallel transport map P_\gamma^{t,s}: T_p M \to T_q M is the linear isomorphism that sends a tangent vector v \in T_p M to V(t) \in T_q M, where V is the unique parallel vector field along \gamma satisfying V(s) = v.^[4] Under standard smoothness assumptions on \gamma and the connection (e.g., C^1 curve and smooth \nabla), the existence and uniqueness of such a parallel vector field V along \gamma with prescribed initial value V(s) = v follow from solving the associated first-order ordinary differential equation, ensuring the parallel transport map is well-defined and invertible.^[5] In local coordinates (x^i) on M, with Christoffel symbols \Gamma^i_{jk} defined by \nabla_{\partial_j} \partial_k = \Gamma^i_{jk} \partial_i, the parallel transport equation for a vector field V = V^i \partial_i along \gamma(t) = (x^j(t)) becomes the system of ODEs

\frac{d V^i}{dt} + \Gamma^i_{jk}(\gamma(t)) \frac{d x^j}{dt} V^k = 0.

This linear equation determines the components V^i(t) uniquely given initial conditions at t = s.^[6]

Basic Examples

In Euclidean space \mathbb{R}^n equipped with the standard flat metric, parallel transport along any smooth curve is simply the identity map on tangent vectors, preserving both direction and magnitude without any rotation or scaling, as the connection is trivial and vector fields that are constant in Cartesian coordinates satisfy the parallel transport equation \nabla_{\dot{\gamma}} V = 0.^[7] On the unit sphere S^2 with its round metric, consider parallel transport of a tangent vector starting at the north pole, pointing horizontally eastward. Transporting this vector along a meridian (great circle) southward to the equator keeps it tangent to the surface and aligned eastward relative to the local parallel; at the equator, it points due east. Continuing along the equator westward by 90 degrees maintains the vector pointing east (now southward relative to the initial direction), and returning northward along another meridian to the north pole results in the vector pointing westward, rotated 90 degrees counterclockwise from its starting orientation due to the sphere's curvature.^[8] The cylinder S^1 \times \mathbb{R}, endowed with the flat metric ds^2 = d\theta^2 + dz^2 (where \theta is the angular coordinate), provides a contrast as it is a flat manifold isometric to the plane via unwrapping. Parallel transport along any curve, such as a helical path or a circle around the axis, yields no net rotation or change beyond what occurs in the unwrapped Euclidean plane; for instance, transporting a vector around a closed circumferential loop rotates it by exactly $2\pi in a manner identical to a straight-line translation in the plane, with no additional holonomy from curvature.^[7] Transporting a vector around a closed loop on the sphere, such as the triangular path described above enclosing one-eighth of the surface area, introduces an angle defect: the final vector is misaligned by 90 degrees relative to the initial one, reflecting the enclosed Gaussian curvature without invoking full holonomy group theory, as the mismatch arises solely from the path's geometry on the curved surface.^[8] To compute parallel transport explicitly on the sphere of radius r, use spherical coordinates (\theta, \phi) with metric ds^2 = r^2 d\theta^2 + r^2 \sin^2 \theta \, d\phi^2. The non-vanishing Christoffel symbols for the Levi-Civita connection are \Gamma^\theta_{\phi\phi} = -\sin\theta \cos\theta and \Gamma^\phi_{\theta\phi} = \Gamma^\phi_{\phi\theta} = \cot\theta. Along a latitude curve at fixed \theta = \theta_0 parametrized by \phi, with tangent V = \partial_\phi, the parallel transport equations for a vector W = W^\theta \partial_\theta + W^\phi \partial_\phi become the system

\frac{d W^\theta}{d\phi} = \sin\theta_0 \cos\theta_0 \, W^\phi, \quad \frac{d W^\phi}{d\phi} = -\frac{\cos\theta_0}{\sin\theta_0} W^\theta.

This is a coupled linear ODE solvable as harmonic motion, yielding after one full loop \Delta\phi = 2\pi a rotation of the vector by angle $2\pi \cos\theta_0 in the tangent plane.^[9]

Connections and Their Properties

Affine Connections

An affine connection on a smooth manifold M is a map \nabla: \mathfrak{X}(M) \times \mathfrak{X}(M) \to \mathfrak{X}(M) that assigns to each pair of smooth vector fields X, Y \in \mathfrak{X}(M) a vector field \nabla_X Y, interpreted as the covariant derivative of Y in the direction of X. This map is required to be bilinear over the ring of smooth functions C^\infty(M), meaning \nabla_{fX + gY} Z = f \nabla_X Z + g \nabla_Y Z and \nabla_X (fY) = (Xf) Y + f \nabla_X Y for all f, g \in C^\infty(M) and Z \in \mathfrak{X}(M), thereby satisfying the Leibniz rule for differentiation of tensor fields. Such a connection equips the tangent bundle TM with a structure for differentiating sections, enabling the extension of directional derivatives from Euclidean space to curved manifolds.^[10] In local coordinates (x^i) on an open set U \subset M, the affine connection is expressed through Christoffel symbols \Gamma^k_{ij}, which are smooth functions on U. Specifically, for the coordinate basis vector fields \partial_i = \frac{\partial}{\partial x^i}, the covariant derivative takes the form

\nabla_{\partial_i} \partial_j = \sum_k \Gamma^k_{ij} \partial_k,

and for a general vector field Y = Y^j \partial_j, it is \nabla_{\partial_i} Y = (\partial_i Y^j + Y^l \Gamma^j_{li}) \partial_j. These symbols provide the local coordinates for the connection and transform under coordinate changes according to the rule \tilde{\Gamma}^k_{ij} = \frac{\partial \tilde{x}^k}{\partial x^p} \frac{\partial x^l}{\partial \tilde{x}^i} \frac{\partial x^m}{\partial \tilde{x}^j} \Gamma^p_{lm} + \frac{\partial^2 x^p}{\partial \tilde{x}^i \partial \tilde{x}^j} \frac{\partial \tilde{x}^k}{\partial x^p},^[11] ensuring the connection is well-defined globally on M.^[10] The torsion tensor of an affine connection \nabla is the tensor field T \in \mathfrak{X}^2(M, TM) defined by T(X, Y) = \nabla_X Y - \nabla_Y X - [X, Y] for all X, Y \in \mathfrak{X}(M), where [X, Y] is the Lie bracket measuring the commutator of the vector fields. In components, T(\partial_i, \partial_j) = (\Gamma^k_{ij} - \Gamma^k_{ji}) \partial_k, so the torsion vanishes if and only if the Christoffel symbols are symmetric in the lower indices, \Gamma^k_{ij} = \Gamma^k_{ji}. Torsion quantifies the failure of the connection to preserve the Lie bracket and plays a key role in non-symmetric connections, where it introduces an antisymmetric component that affects the transport of vectors along non-commuting directions, distinguishing them from torsion-free cases common in classical geometry.^[10]^[12] Affine connections are compatible with the smooth structure of the manifold, meaning the Christoffel symbols \Gamma^k_{ij} depend smoothly on the point p \in M, ensuring that \nabla_X Y is a smooth vector field whenever X and Y are. Parallel transport arises directly from the connection through the covariant derivative along a smooth curve \gamma: I \to M. For a vector field V along \gamma, the parallel transport condition is \frac{DV}{dt} = \nabla_{\gamma'(t)} V = 0, which is a first-order ordinary differential equation whose unique solution (by Picard-Lindelöf) defines an isomorphism between tangent spaces T_{\gamma(a)}M and T_{\gamma(b)}M, preserving the linear structure of fibers in TM.^[10] A canonical example is the flat affine connection on the Euclidean space \mathbb{R}^n, where the standard coordinate basis yields zero Christoffel symbols, \Gamma^k_{ij} = 0 for all i, j, k. In this case, the torsion tensor vanishes, and parallel transport reduces to ordinary vector translation, reflecting the absence of curvature or torsion in flat space.^[10]

Metric Connections

In Riemannian geometry, a Riemannian metric g on a smooth manifold M is a smooth, positive-definite inner product g_p: T_pM \times T_pM \to \mathbb{R} defined at each point p \in M, varying smoothly with p. This metric equips the manifold with a geometric structure that allows the definition of lengths, angles, and volumes: the length of a curve \gamma: [a,b] \to M is given by \int_a^b \sqrt{g_{\gamma(t)}(\dot{\gamma}(t), \dot{\gamma}(t))} \, dt, and the angle between two vectors u, v \in T_pM is \cos^{-1} \left( \frac{g_p(u,v)}{\sqrt{g_p(u,u) g_p(v,v)}} \right). An affine connection \nabla on a Riemannian manifold (M, [g](/page/G)) is said to be metric-compatible if it preserves the metric tensor under parallel transport, meaning \nabla [g](/page/G) = 0. In local coordinates, this condition is expressed as X([g](/page/G)(Y,Z)) = [g](/page/G)(\nabla_X Y, Z) + [g](/page/G)(Y, \nabla_X Z) for all vector fields X, Y, Z, ensuring that inner products are invariant along curves: if V, W are parallel transported along \gamma, then [g](/page/G)(V,W) remains constant. This compatibility implies that parallel transport along \gamma preserves the lengths of vectors, so |P_\gamma V| = |V| where |\cdot| denotes the norm induced by [g](/page/G), and similarly for angles between vectors. The Levi-Civita connection is the unique torsion-free, metric-compatible affine connection on (M, g), often called the fundamental theorem of Riemannian geometry. Torsion-freeness requires \nabla_X Y - \nabla_Y X = [X,Y], and together with metric compatibility, this uniqueness guarantees a canonical way to define covariant differentiation compatible with the geometry. The connection coefficients, or Christoffel symbols \Gamma^k_{ij}, are determined by the Koszul formula:

\Gamma^k_{ij} = \frac{1}{2} g^{kl} \left( \partial_i g_{jl} + \partial_j g_{il} - \partial_l g_{ij} \right),

where g^{kl} is the inverse metric tensor, and \partial_i denotes partial differentiation with respect to the i-th coordinate. This explicit construction allows computation of parallel transport in coordinates, preserving the metric structure. The concept of the Levi-Civita connection was introduced by Tullio Levi-Civita in 1917, formalizing parallel displacement in general manifolds and specifying the Riemannian curvature geometrically through this unique connection.

Relation to Geodesics

In the framework of an affine connection on a manifold, geodesics are defined as auto-parallel curves, meaning that the tangent vector to the curve is parallel transported along itself.^[13] This property ensures that the curve follows the "straightest" possible path consistent with the connection's notion of parallelism. Specifically, for a curve \gamma: I \to M with tangent vector \gamma', the condition for \gamma to be a geodesic is that the covariant derivative satisfies \nabla_{\gamma'} \gamma' = 0, indicating that \gamma' remains unchanged under parallel transport along \gamma.^[14] In local coordinates, this geodesic equation takes the form

\frac{d^2 \gamma^k}{dt^2} + \Gamma^k_{ij}(\gamma) \frac{d \gamma^i}{dt} \frac{d \gamma^j}{dt} = 0,

where \Gamma^k_{ij} are the Christoffel symbols of the connection, and t is an affine parameter along the curve.^[15] The affine parameterization is crucial, as it preserves the form of the equation under reparametrizations of the form t \mapsto at + b.^[14] Within Riemannian geometry, where the connection is the metric-compatible Levi-Civita connection, geodesics additionally represent locally shortest paths between points, as the metric compatibility ensures that parallel transport preserves lengths and angles, minimizing the arc length functional along the curve.^[6]^[16] The exponential map \exp_p: T_p M \to M, defined by \exp_p(v) = \gamma(1) where \gamma is the geodesic starting at p with initial velocity v, formalizes this relation by mapping initial tangent vectors to endpoint positions via geodesic flow, providing a local chart around p through radial geodesics.^[17] To analyze variations of geodesics, Jacobi fields arise as vector fields J along a geodesic \gamma satisfying the Jacobi equation \nabla_{\dot{\gamma}} \nabla_{\dot{\gamma}} J + R(\dot{\gamma}, J) \dot{\gamma} = 0, where R is the Riemann curvature tensor; these fields describe the infinitesimal transport of nearby tangent vectors and are precisely the variation fields of geodesic variations of \gamma.^[18] In an orthonormal frame parallel transported along \gamma, the equation decouples into ordinary differential equations for the components of J, highlighting how curvature influences the evolution of these transported vectors.^[18]

Parallel Transport in Vector Bundles

Definition and Construction

A vector bundle E over a smooth manifold M consists of a total space E, a projection \pi: E \to M, and fibers E_x \cong V for each x \in M, where V is a fixed vector space; smooth sections of E are referred to as vector fields over M.^[19] An Ehresmann connection on E is defined by assigning to each point u \in E a horizontal subspace H_u \subset T_u E such that T_u E = H_u \oplus V_u, where V_u = \ker(d\pi_u) is the vertical subspace, and this assignment varies smoothly with u.^[20] This horizontal distribution provides a way to lift tangent vectors from the base manifold M to the total space E, complementing the vertical directions along the fibers. Parallel transport along a smooth curve \gamma: [0,1] \to M with \gamma(0) = x is constructed by lifting \gamma to a horizontal curve \tilde{\gamma} in E starting at a given point w \in E_x, ensuring \tilde{\gamma}'(t) \in H_{\tilde{\gamma}(t)} for all t.^[21] This defines a linear isomorphism P_\gamma: E_x \to E_{\gamma(1)}, mapping w to \tilde{\gamma}(1), which preserves the vector space structure of the fibers.^[19] Equivalently, for a section s along \gamma, parallel transport solves the equation \nabla_{\dot{\gamma}(t)} s(t) = 0, where \nabla is the covariant derivative induced by the connection, yielding a unique solution linear in the initial condition s(0).^[20] In a local trivialization \phi: \pi^{-1}(U) \to U \times V over an open set U \subset M, parallel transport appears as a path-dependent linear map on the fiber coordinates V, evolving according to an ordinary differential equation determined by the connection form.^[21] Specifically, if \gamma lies in U, the transport P_\gamma satisfies \frac{d}{dt} (\phi \circ \tilde{\gamma})(t) = -A(\dot{\gamma}(t)) \cdot (\phi \circ \tilde{\gamma})(t), where A is the connection matrix in these coordinates, with solution P_\gamma = T\exp\left(-\int_0^1 A(\dot{\gamma}(t)) dt\right).^[19] The curvature form \Omega \in \Omega^2(M, \End(E)), defined by \Omega(X,Y)s = [\nabla_X, \nabla_Y]s - \nabla_{[X,Y]}s for sections s and vector fields X,Y, measures the failure of path-independence in parallel transport.^[20] For composable paths \gamma and \delta meeting at a point, the commutator [P_\gamma, P_\delta] approximates the identity plus an infinitesimal action of \Omega along the enclosed loop, with vanishing \Omega implying local flatness but not necessarily global path-independence due to topology.^[21] As an example, consider the trivial bundle E = M \times V over \mathbb{R}^n, equipped with the flat connection where horizontal subspaces are trivial; here, parallel transport reduces to constant translation in the fibers, recovering the standard parallel transport in the tangent bundle TM when V = \mathbb{R}^n.^[19]

Recovering the Connection

Parallel transport on a vector bundle E \to M is defined by a family of fiberwise linear isomorphisms P_\gamma^{t,s}: E_{\gamma(s)} \to E_{\gamma(t)} for each smooth curve \gamma: [a,b] \to M and parameters s,t \in [a,b], satisfying the cocycle condition P_\gamma^{u,t} \circ P_\gamma^{t,s} = P_\gamma^{u,s} for a \leq s \leq t \leq u \leq b. This family of maps admits an infinitesimal generator given by an \mathfrak{gl}(E)-valued 1-form \omega, called the connection form, such that the parallel transport satisfies the differential equation \frac{d}{dt} P_\gamma^{t,s} = -\omega(\gamma'(t)) \cdot P_\gamma^{t,s} with initial condition P_\gamma^{s,s} = \mathrm{id}. From this, the covariant derivative associated to the connection can be recovered: for a smooth section s \in \Gamma(E) and a curve \gamma with \gamma(0) = p and \gamma'(0) = X_p \in T_p M, \nabla_X s = \lim_{t \to 0} \frac{1}{t} \left( s_p - P_{\gamma_t}^0 (s_{\gamma(t)}) \right), where P_{\gamma_t}^0: E_{\gamma(t)} \to E_p denotes transport backward along \gamma from time t to 0. In a local trivialization of the bundle over an open set U \subset M, the connection form \omega is represented by a smooth \mathfrak{gl}(n,\mathbb{R})-valued 1-form A (the gauge potential), and the parallel transport maps are solutions to the ordinary differential equation \frac{d}{dt} P(t) = -A(\dot{\gamma}(t)) P(t). Conversely, given such a local A, the connection form \omega on the frame bundle pulls back accordingly to define the global connection. Any smooth connection on E induces a unique parallel transport satisfying the above properties, and under suitable smoothness assumptions, every such family of isomorphisms arises uniquely from a connection. Specifically, a map assigning parallel transports along curves defines a connection if and only if it is smooth (C^\infty) in the endpoints and curve reparametrization parameters. This establishes the equivalence between connections and parallel transport on vector bundles.

Holonomy and Global Aspects

Local vs. Global Transport

In a contractible neighborhood of a point on a manifold, parallel transport exhibits path-independent behavior, meaning that transporting a vector along any two homotopic paths between the same endpoints yields the same result. This local flatness stems from the Poincaré lemma for connections, which guarantees that the connection form can be locally gauged to zero, allowing for a trivialization where parallel sections exist and transport is unambiguous within such neighborhoods.^[22] Globally, however, parallel transport becomes path-dependent on manifolds that are not simply connected, as different paths from a point p to a point q can produce distinct end vectors unless the connection's curvature vanishes identically. This inconsistency reflects the topological obstructions to extending local trivializations across the entire manifold, leading to non-trivial holonomy effects that accumulate along non-contractible loops.^[22] The Ambrose–Singer theorem links the infinitesimal curvature to the global holonomy by asserting that the Lie algebra of the holonomy group at a base point is spanned by the values of the curvature endomorphisms (and their covariant derivatives) at that point, thereby linking the infinitesimal curvature to the overall group generated by path-dependent transports. An illustrative example of curvature's influence on local transport is the case of an infinitesimal closed loop formed by paths \gamma and \delta tangent to vectors X and Y, respectively. The parallel transport P_\gamma along \gamma followed by P_\delta along \delta, and vice versa, differs by an amount captured by the curvature tensor:

P_\delta^{-1} P_\gamma V - V \approx R(X, Y) V,

where V is the initial vector; this approximation highlights how curvature quantifies the failure of closed-loop transport to return V unchanged, even on tiny scales.^[23] For flat connections, where curvature vanishes, the Poincaré lemma ensures local triviality, implying path-independent transport in contractible regions. The moduli spaces of flat connections carry topological invariants that can be studied via de Rham cohomology, which helps encode properties of the possible holonomy representations.^[24]

Holonomy Groups

The holonomy map associated to a closed loop \gamma based at a point p in a manifold M with connection \nabla on the tangent bundle TM is the automorphism Hol_\gamma: T_pM \to T_pM induced by parallel transport along \gamma, often denoted as P_\gamma^{T,0}. This map encapsulates the net effect of parallel transport around the loop, measuring the failure of the connection to be flat.^[25] The holonomy group at p, denoted Hol_p(\nabla), is the subgroup of \mathrm{[GL](/page/GL)}(T_pM) generated by all such Hol_\gamma for loops \gamma based at p; it forms a Lie subgroup whose Lie algebra is spanned by curvature endomorphisms.^[26] In the Riemannian setting with the Levi-Civita connection, the restricted holonomy group (the connected component of the identity) lies in \mathrm{SO}(n) for an orientable n-manifold, preserving the metric and orientation. Special cases include \mathrm{SU}(m) for Kähler manifolds with Ricci-flat metrics, corresponding to Calabi-Yau structures where the holonomy preserves a holomorphic volume form.^[27]^[28] Marcel Berger classified the possible irreducible holonomy groups for simply connected Riemannian manifolds of dimension greater than 3, excluding reducible cases and symmetric spaces; the list includes \mathrm{SO}(n), \mathrm{U}(m), \mathrm{SU}(m), \mathrm{Sp}(m), \mathrm{Sp}(m)\mathrm{Sp}(1), and the exceptional groups \mathrm{G_2} (dimension 7) and \mathrm{Spin}(7) (dimension 8). These groups determine rich geometric structures, such as nearly Kähler or quaternionic Kähler metrics for the symplectic cases, and exceptional calibrations for \mathrm{G_2} and \mathrm{Spin}(7).^[29]^[30] The holonomy group relates to curvature via the Ambrose-Singer theorem, where the Lie algebra is generated by integrals of the curvature tensor R along loops; infinitesimally, for small loops enclosing a surface D, Hol_\gamma \approx \exp\left(\int_D R\right), with the full group reflecting integrated curvature effects. De Rham's decomposition theorem states that if the restricted holonomy representation is reducible, the manifold decomposes locally as a Riemannian product of factors with irreducible holonomy, linking flat (trivial) holonomy to Euclidean topology.^[31]^[32]^[33] A concrete computation arises on the unit 2-sphere with its round metric, where parallel transport of a tangent vector around a closed curve yields a rotation in \mathrm{SO}(2) by an angle equal to the Gaussian curvature (1) times the enclosed area, demonstrating how holonomy encodes global topology via local curvature.^[34]

Approximations and Visualizations

Schild's Ladder

Schild's ladder provides a geometric method for approximating the parallel transport of a tangent vector along a curve on a Riemannian manifold by constructing successive geodesic parallelograms.^[35] To construct one step, begin with points A and B on a base geodesic \gamma, and a vector at A represented by the endpoint C of a perpendicular geodesic segment from A. From B, draw another perpendicular geodesic of equal length to point D. The vector from B to D then approximates the parallel transport of the original vector from A to C.^[36] This process iterates along the curve, forming a "ladder" of such segments to transport vectors over finite distances.^[35] The method originated in the context of general relativity pedagogy, introduced by physicist Alfred Schild in a 1972 chapter co-authored with Jürgen Ehlers and Felix Pirani, where it illustrates the geometry of free fall and light propagation on curved spacetimes.^[36] Although no earlier published reference exists, it has since become a standard tool for discretizing parallel transport without requiring explicit knowledge of the connection.^[35] As a first-order approximation, Schild's ladder incurs an error that is second-order in the step size, with the discrepancy between the approximated and exact transport proportional to the Riemann curvature tensor integrated over the area enclosed by the geodesic parallelogram.^[35] Specifically, for a small displacement along the base geodesic with tangent \gamma' and normal direction n, the Taylor expansion yields an initial velocity approximation u = v + \frac{1}{2} R(w, v) v + O(4), where v and w are tangent vectors defining the parallelogram sides, and higher-order terms involve covariant derivatives of the curvature.^[35] Iterating with step size scaled appropriately achieves quadratic convergence overall.^[35] Visually, the ladder's "rungs"—the perpendicular segments—appear to twist relative to the base due to manifold curvature, qualitatively demonstrating holonomy as the cumulative mismatch upon closing a loop.^[35] In the infinitesimal limit, this twisting is captured by the displacement \delta V \approx -R(\gamma', n) V \, ds \wedge dn, where V is the transported vector, highlighting the role of sectional curvature in local deviations.^[35] Beyond pedagogy, Schild's ladder finds applications in computer graphics for interpolating orientations and deformations on curved manifolds, such as in surface parameterization via geodesic splines, where it enables efficient discrete approximations of vector fields without solving differential equations.

Infinitesimal Transport

Infinitesimal parallel transport arises from considering the limiting case of transporting vectors along a curve through a sequence of vanishingly small steps, which leads to a first-order linear ordinary differential equation (ODE) governing the evolution of the vector field along the curve.^[37] For a smooth curve \gamma: [0, T] \to M on a manifold M equipped with a linear connection \nabla, a vector field V along \gamma is parallel if it satisfies the parallel transport equation \frac{DV}{dt} = 0, or in local coordinates, \frac{dV^\lambda}{dt} + \Gamma^\lambda_{\mu\nu}(\gamma(t)) V^\mu \dot{\gamma}^\nu(t) = 0, where \Gamma denotes the connection coefficients.^[37] This ODE can be written compactly as \frac{dV}{dt} = -\Gamma(\gamma(t)) \dot{\gamma}(t) V, treating \Gamma(\gamma(t)) \dot{\gamma}(t) as a matrix acting on the vector components.^[38] The exact solution to this initial value problem provides the parallel transport operator from t=0 to t, given by the path-ordered exponential V(t) = \mathcal{P} \exp\left( -\int_0^t \Gamma(\gamma(s)) \dot{\gamma}(s) \, ds \right) V(0), where the path-ordering \mathcal{P} accounts for the non-commutativity of the connection matrices along the path by arranging infinitesimal factors in the order of integration.^[37] This formulation highlights the infinitesimal nature of the transport, as the exponential arises from compounding infinitesimal displacements \exp(-\varepsilon \Gamma(\gamma(t_i)) \dot{\gamma}(t_i)) at discrete points t_i along \gamma.^[37] For numerical computation, the curve \gamma is discretized into points t_i = i \varepsilon with step size \varepsilon > 0, and the transport is approximated stepwise by solving the ODE iteratively.^[38] Basic methods include the forward Euler scheme, which updates V(t_{i+1}) \approx V(t_i) - \varepsilon \Gamma(\gamma(t_i)) \dot{\gamma}(t_i) V(t_i), or higher-order Runge-Kutta methods, such as the second-order variant that evaluates the right-hand side at intermediate points for improved accuracy.^[38] These approaches converge to the exact path-ordered exponential as \varepsilon \to 0, with the local truncation error for the Euler method being O(\varepsilon^2) per step and the global error O(\varepsilon) over a fixed interval.^[39] In the context of general relativity, this infinitesimal transport relates to Fermi-Walker transport, which generalizes parallel transport for non-geodesic curves by including a rotation term to keep spatial vectors orthogonal to the curve's tangent without twisting.^[40] For geodesics, where acceleration vanishes, Fermi-Walker transport coincides exactly with parallel transport; for twist-free non-geodesic curves, such as circular orbits in stationary spacetimes, it differs by an additional term proportional to the acceleration, ensuring no spurious rotation in the local frame.^[40] When implementing numerical parallel transport on manifolds embedded in Euclidean space, vectors are iteratively projected onto the tangent spaces at each point along the discretized curve, often using orthonormal bases derived from singular value decomposition of local neighborhoods to maintain the embedding constraints.^[41] This projection step ensures the transported vectors remain tangent to the manifold, with the overall scheme converging to the exact transport as the step size \varepsilon \to 0.^[38]

Generalizations and Applications

Principal Bundles and Gauge Fields

A principal G-bundle P \to M over a smooth manifold M consists of a right action of a Lie group G on P that is free and transitive on each fiber, with the projection \pi: P \to M being a submersion.^[42] The connection on such a bundle is defined by a \mathfrak{g}-valued 1-form \omega \in \Omega^1(P, \mathfrak{g}), where \mathfrak{g} is the Lie algebra of G, satisfying two key properties: it reproduces the infinitesimal action via \omega(\xi^\#) = \xi for any \xi \in \mathfrak{g} and the fundamental vector field \xi^\# generated by \xi, and it is equivariant under the right action: R_g^* \omega = \mathrm{Ad}_{g^{-1}} \omega for g \in G.^[43] This \omega decomposes the tangent space TP at each point into horizontal and vertical subbundles, with the vertical bundle being the kernel of \pi_* and the horizontal complement determined by \ker \omega. Parallel transport along a curve \gamma: [0,1] \to M is given by the horizontal lift \tilde{u}: [0,1] \to P starting at some u(0) \in \pi^{-1}(\gamma(0)) such that \pi \circ \tilde{u} = \gamma and \tilde{u}'(t) \in \ker \omega for all t, ensuring \omega(\tilde{u}'(t)) = 0.^[44] The endpoint \tilde{u}(1) lies in the fiber over \gamma(1), and the map from \pi^{-1}(\gamma(0)) to \pi^{-1}(\gamma(1)) induced by such lifts identifies points in the fibers via right multiplication by elements of G, preserving the group structure. This generalizes the notion of parallel transport in vector bundles, where the associated vector bundle E = P \times_G V for a representation \rho: G \to \mathrm{GL}(V) of G on a vector space V is constructed via the quotient (P \times V)/G with the diagonal action (p, v) \cdot g = (p g, \rho(g)^{-1} v); the connection \omega on P induces a linear connection on E, recovering vector transport from principal transport.^[45] The curvature of the connection is captured by the \mathfrak{g}-valued 2-form \Omega = d\omega + \frac{1}{2} [\omega, \omega], where [\omega, \omega] denotes the wedge product with the Lie bracket in \mathfrak{g}.^[43] This form is horizontal and equivariant, making it gauge-invariant under the action of G, and measures the failure of parallel transport around closed loops to be path-independent. A connection is flat if \Omega = 0, in which case the holonomy along loops generates a representation of the fundamental group \pi_1(M) into G up to conjugation, providing a topological invariant.^[44] In the Langlands program, particularly its geometric incarnation over Riemann surfaces, such flat connections on principal bundles correspond to certain automorphic representations via their holonomy, linking differential geometry to number theory.^[46] An illustrative example arises in Riemannian geometry: given a Riemannian manifold (M, g) of dimension n, the orthonormal frame bundle P \to M is a principal O(n)-bundle, where each fiber consists of positively oriented orthonormal bases of the tangent spaces.^[47] The Levi-Civita connection on the tangent bundle corresponds to a unique torsion-free metric-compatible connection on P, defined by the \mathfrak{o}(n)-valued 1-form \omega that satisfies the metric-preserving condition and reproduces the Christoffel symbols in local frames.

Applications in Physics

In general relativity, parallel transport along a worldline defines a non-rotating frame for an observer, ensuring that spatial basis vectors remain orthogonal to the four-velocity and do not rotate relative to distant stars.^[48] For geodesic motion, this coincides with the standard parallel transport of vectors, preserving their components in the local inertial frame. Fermi coordinates extend this concept by constructing a local coordinate system around the worldline using Fermi-Walker transport, a generalization of parallel transport that accounts for acceleration while maintaining non-rotation; these coordinates approximate flat spacetime to second order, with metric deviations encoding gravitational effects. Parallel transport also underpins the geodesic deviation equation, which quantifies tidal forces as the relative acceleration between nearby geodesics.^[6] Specifically, the Riemann curvature tensor governs how a vector parallel-transported along one geodesic deviates when compared to its counterpart on a neighboring geodesic, manifesting as stretching or squeezing due to spacetime curvature; for instance, in the Schwarzschild metric, this describes tidal disruption near black holes. In Yang-Mills gauge theories, the connection is identified with the gauge potential A_\mu, and parallel transport along a path in spacetime yields the path-ordered exponential \mathcal{P} \exp\left(i \oint A\right), known as a Wilson loop for closed paths.^[49] These loops are gauge-invariant observables that probe non-perturbative effects like quark confinement in quantum chromodynamics, where the holonomy around large loops reflects the flux of the gauge field. Originally introduced to study lattice gauge theories, Wilson loops quantify the area-law behavior of the gauge field strength. The Aharonov-Bohm effect illustrates holonomy in electromagnetism, where charged particles acquire a phase shift from parallel transport around a region of zero magnetic field but nonzero vector potential, arising from the nontrivial topology of the U(1) bundle. This phase, \exp(i \oint A \cdot dx), depends solely on the enclosed magnetic flux and demonstrates that gauge potentials encode physically observable effects beyond local fields. In quantum mechanics, the Berry phase emerges as a U(1) holonomy from adiabatic parallel transport of a state vector in parameter space, where the dynamical phase is suppressed, leaving a geometric phase \exp(i \oint \mathbf{A}_B \cdot d\mathbf{R}) with Berry connection \mathbf{A}_B = i \langle n | \nabla_R n \rangle. For a spin-1/2 particle in a slowly varying magnetic field, this phase equals half the solid angle subtended by the parameter path on the Bloch sphere, analogous to monopole holonomy.^[50] Recent developments in the 2020s leverage holonomy for fault-tolerant quantum gates in holonomic quantum computing, where non-Abelian geometric phases enable robust operations immune to local errors via cyclic adiabatic evolution in degenerate subspaces.^[51] For example, continuous measurement protocols generate holonomies on stabilizer codes, supporting universal computation with reduced decoherence.^[52] Additional advances include proposals for holonomic swap and controlled-swap gates using neutral atoms via Rydberg interactions.^[53]

References

[1]
Parallel Transport -- from Wolfram MathWorld
The notion of parallel transport on a manifold M makes precise the idea of translating a vector field V along a differentiable curve to attain a new vector ...
[2]
[PDF] On the history of Levi-Civita's parallel transport - arXiv
Aug 6, 2016 · The role of the principle of virtual works of analytical mechanics, in formulating Levi-Civita's parallel transport of Riemannian geometry, has ...
[3]
parallel transport in nLab
### Summary of Parallel Transport from nLab
[4]
[PDF] Chapter 12 Connections on Manifolds - UPenn CIS
The notion of connection yields the notion of parallel transport. First, we need to define the covariant deriva- tive of a vector field along a curve.
[5]
[PDF] Geometric Interpretation of curvature - UChicago Math
To define parallel transport, we need the notion of affine connections. Definition 2.1. Let M be an n-dimensional manifold. An affine connection, or connection, ...
[6]
Lecture Notes on General Relativity - S. Carroll
We can look at the parallel transport equation as a first-order differential equation defining an initial-value problem: given a tensor at some point along the ...
[7]
[PDF] Extrinsic Differential Geometry
The vector X(t) is called the parallel transport of X(t0) ∈ Tγ(t0)M to Tγ(t) ... b0 : TpM → Em where Em is the vector space of some Euclidean space Em.
[8]
parallel.transport - UCR Math Department
Assuming the earth is a perfect sphere and he doesn't gratuitously swing his ... So do parallel transport that way. But wait --- this "flattening out ...
[9]
[PDF] General Relativity Fall 2019 Homework 5 solution
Exercise 1: Parallel transport on a sphere. The 2-sphere of radius r has line element ds2 = r2dθ2 + r2 sin2 θdϕ2. (i) Compute the Christoffel symbols of this ...
[10]
[PDF] Introduction to Smooth Manifolds - Julian Chaidez
... John M. Lee. Introduction to. Smooth Manifolds. Second Edition. Page 6. John M. Lee. Department of Mathematics. University of Washington. Seattle, WA, USA. ISSN ...
[11]
[PDF] Beyond Riemannian geometry: The affine connection ... - Hal-Inria
May 10, 2020 · The connection defines the parallel transport, and thus a notion of geodesics that extends straight lines: the tangent vector to an auto- ...<|control11|><|separator|>
[12]
affine connection - PlanetMath
Mar 22, 2013 · Perhaps the most important property of affine connections is that they allow one to define geodesics. Because the defining equation of a ...
[13]
[PDF] 3 Parallel transport and geodesics - DSpace@MIT
A well-known example of a geodesic in a curved space is a great circle on a sphere. There are several technical points worth noting about geodesic curves ...
[14]
[PDF] Solving the Geodesic Equation
Dec 12, 2018 · Abstract. We find the general form of the geodesic equation and discuss the closed form relation to find Christoffel symbols.
[15]
https://www.sas.rochester.edu/pas/assets/pdf/undergraduate/solving_geodesic_equation.pdf
[16]
[PDF] Lecture 8. Connections
The parallel transport operators give a convenient way to identify the tangent spaces to M at different points along a smooth curve, and in some ways this is ...
[17]
[PDF] JACOBI FIELDS As we have seen, in the second variational formula ...
A vector field X along a geodesic γ is a Jacobi field if and only if X is the variation field of some geodesic variation of γ. ξ(0) = γ(0), ˙ ξ(0) = Xγ(0). Let ...
[18]
http://staff.ustc.edu.cn/~wangzuoq/Courses/24S-RiemGeom/Notes/Lec17.pdf
[19]
[PDF] what is a connection, and what is it good for? - Cornell Mathematics
manifold under the derivative of all the parallel sections of the vector bundle. ... Differential geometry, volume 166 of Graduate Texts in Mathematics. Springer ...
[20]
[PDF] A Second Course in Differential Geometry
Details can be found in any introductory differential geometry textbook. 1.1.1 Linear algebra. Let V be a real vector space. Its dual space V ∗ is defined ...
[21]
[PDF] Chapter 5 Curvature on Bundles
Indeed, the Poincaré lemma generalizes this result by saying that every closed differential k-form locally is the exterior derivative of some (k − 1)-form. Here ...
[22]
[PDF] Parallel Transport and Curvature
Parallel transport around an infinitesimal closed loop and the Riemann curvature tensor. We now ask the following question. Suppose we parallel transport the ...
[23]
[alg-geom/9404012] Group cohomology construction of the ... - arXiv
These generators are constructed using the de Rham representatives for the cohomology of classifying spaces BK where K is a compact Lie group; such ...
[24]
[PDF] Part III - Differential Geometry (Definitions) - Dexter Chua
parallel transport along γ. The holonomy of ∇ around γ: is the map. H : TpM → TpM given by. H(ξ) = V (1), where V is the parallel transport of ξ along γ. 23.
[25]
[PDF] Lecture 7: Holonomy groups - Spin Geometry
Holonomy groups capture part of Riemannian geometry, specifically the holonomy group of the Levi-Civita connection on a manifold. If M is connected, the ...
[26]
[PDF] Riemannian Holonomy and Algebraic Geometry
So the compact (simply-connected) complex manifolds which admit a metric with holonomy contained in SU(m) are the Calabi-Yau manifolds. This fact has strong ...
[27]
[PDF] Geometry of Calabi-Yau metrics - U.C. Berkeley Mathematics
Oct 31, 2021 · A Riemannian metric g on a differen- tiable manifold M of dimension 2n is Kähler if its holonomy group is contained in U(n). Since. U(n) = GL(n ...
[28]
[PDF] A geometric proof of the Berger Holonomy Theorem
This is an easy consequence of the Bianchi identity, the Ambrose-Singer theorem and a theorem of Cartan on the existence of totally geodesic submanifolds.<|separator|>
[29]
[PDF] Introduction to G2 geometry - Harvard CMSA
Sep 20, 2019 · In 1955, Marcel Berger classified the possible Lie subgroups of O(n) that could occur as Riemannian holonomy groups of a metric g, subject to ...
[30]
[PDF] Holonomy groups in riemannian geometry - arXiv
Jun 14, 2012 · The list of Berger also contains two exceptional examples : G2 and Spin(7). ... ℓ is an irreducible representation of Spin(7) of dimension ℓ.<|separator|>
[31]
[PDF] ASYMPTOTIC EXPANSION OF HOLONOMY 1. Introduction Let G ...
Introduction. Let G → P → M be a principal bundle with some connection ω and let Ω be its curvature form. The realization of curvature as the infinitesimal ...
[32]
ON THE DE RHAM DECOMPOSITION THEOREM - Project Euclid
holonomy theorem of Ambrose-Singer [1], the condition of reducibility of the holonomy group should be reflected in the parallel translation of curvature. Since ...
[33]
[PDF] Angle variable holonomy in adiabatic excursion of an integrable ...
The angle turned through by a vector, and therefore the extra angle for the top, is the integral of the Gaussian curvature over the area enclosed-that is, the ...
[34]
Numerical Accuracy of Ladder Schemes for Parallel Transport on ...
Jul 15, 2020 · We also contribute a new link between Schild's ladder and the Fanning Scheme which explains why the latter naturally converges only linearly.
[35]
[PDF] Schilds Ladder for the parallel transport of deformations in ... - Hal-Inria
Apr 30, 2023 · We can see that for both the Schild's Ladder (BCH scheme) and the Interpolation method, the resulting Jacobian map seems to adapt to the new ...Missing: computer graphics
[36]
The parallel transporter in terms of the connection
... ( v | p ) evaluated on the tangent v|p v | p at successive points p p along C C . The limit of this process is the path-ordered exponential. ∥λμ(C) ...
[37]
[PDF] Parallel transport in shape analysis: a scalable numerical scheme
Nov 23, 2017 · 3.3 Estimating the error associated to a single parallel transport. To study the error in this high-dimensional setting, we compute the parallel.
[38]
[PDF] On the Global Error of Discretization Methods for Ordinary ... - EqWorld
Discretization methods for ordinary differential equations are usually not exact; they commit an error at every step of the algorithm.<|separator|>
[39]
[PDF] arXiv:gr-qc/0202085v2 9 Jan 2003
Finally the Thomas precession fre- quency is shown to be a special case of the simple relationship between the parallel transport and Fermi-Walker transport ...
[40]
[PDF] Parallel Transport Unfolding: A Connection-based Manifold ...
The use of parallel transport on high-dimensional datasets remedies an important limitation of the Isomap approach: unfolding paths between pairs of points ...
[41]
[PDF] Chapter 3 Connections
Conversely, any fiber bundle connection with this property defines G-equivariant parallel transport; one can prove this along the same lines as for vector ...
[42]
[PDF] Connections on Principal Fibre Bundles
Shoshichi Kobayashi and Katsumi Nomizu, Foundations of differential geometry, vol. 1, Interscience, 1963. 6. , Foundations of differential geometry, vol. 2 ...
[43]
[PDF] Connections on principal bundles - Homotopico
Dec 16, 2020 · Given a curve 𝛾 ∶ [0, 1] → 𝑀, if we write 𝑥0 = 𝛾(0) and 𝑥1 = 𝛾(1), then we have a map. PT𝛾 ∶ 𝜋−1(𝑥0) → 𝜋−1(𝑥1), called parallel transport along ...
[44]
[PDF] Chapter 3 - Vector bundles and principal bundles
Providing a vector bundle with a metric, when viewed in terms of the associated principal bundles, is an example of “reduction of the structure group.” We are ...
[45]
[PDF] Mirror symmetry, Langlands duality, and the Hitchin system - arXiv
It is much easier to work with Higgs bundles than flat connections, because of the alge- braic C× -action on the moduli space. So we begin our proof by showing ...
[46]
Isometry types of frame bundles - MSP
The oriented orthonormal frame bundle SO(M) → M is an example of a SO(n)-principal bundle of over M, and it has a natural connection form ω as defined above.
[47]
[gr-qc/0202085] Circular Holonomy, Clock Effects and ... - arXiv
Feb 24, 2002 · The historical origins of Fermi-Walker transport and Fermi coordinates and the construction of Fermi-Walker transported frames in black hole ...
[48]
[PDF] 2. Yang-Mills Theory - DAMTP
In mathematics, this notion is called holonomy. In this case, we can form a gauge invariant object known as the Wilson loop,. W[C] = tr P exp( ...
[49]
[PDF] Lecture notes on Berry phases and topology - SciPost
May 11, 2022 · We will start by first reviewing the adiabatic theorem in some generality, showing how parallel transport and holonomy ... parallel transport, ...
[50]
Holonomic Surface Codes for Fault-Tolerant Quantum Computation
Aug 8, 2017 · Here we propose holonomic surface codes by harnessing the quantum holonomy of the system. In our scheme, the holonomic gates are built via ...<|separator|>