Parallel Transport

A vector living at one point of a curved manifold and a vector living at another point inhabit different tangent spaces, and there is no god-given way to say whether they point "the same way." Their components can be compared only after choosing a basis at each point, and those bases twist and tilt from place to place, so a bare subtraction of components measures the wobble of the coordinate frame far more than any real change in the vector. Parallel transport is the machinery that repairs this defect. It supplies a path-dependent isomorphism between the tangent spaces strung along a curve, declaring precisely which vector at the far end counts as the unchanged image of a given vector at the near end — unchanged, that is, relative to the geometry rather than relative to an arbitrary grid. The rule is read straight off the covariant derivative, itself owned by aq-gr-covariant-derivative: a vector is parallel-transported along a curve when its covariant derivative in the direction of the tangent vanishes at every point. Expanding that one condition in coordinates yields a first-order linear ordinary differential equation whose driving term is the affine connection — the Christoffel symbols of aq-gr-christoffel-symbols. The connection is exactly the data recording how the basis vectors turn, so it is exactly what has to appear to cancel the spurious turning and leave only genuine change. The deep payoff is path dependence. In flat space transport depends only on the endpoints and a closed loop returns your vector untouched; in curved space a loop returns it rotated, by an amount set by the Riemann tensor times the enclosed area. This holonomy is the local, operational fingerprint of curvature, and it is the same physics behind tidal forces, geodesic deviation, and the precession of orbiting gyroscopes.

Key takeaways

Parallel transport carries a vector (or any tensor) along a curve so it stays "as constant as the geometry allows" — formally, its covariant derivative along the curve is set to zero.
The defining ODE $\frac{\mathrm{d}V^\mu}{\mathrm{d}\lambda}+\Gamma^\mu{}_{\nu\rho}\frac{\mathrm{d}x^\nu}{\mathrm{d}\lambda}V^\rho=0$ is first-order and linear; its solution depends on the whole path, not just the endpoints.
Transport around a closed loop in curved space returns a rotated vector; the rotation per unit enclosed area is the Riemann curvature tensor — this is holonomy, the operational meaning of curvature.
A geodesic is precisely the curve that parallel-transports its own tangent vector, which is why the geodesic equation shares the connection term.
With the Levi-Civita connection (metric-compatible and torsion-free) transport preserves lengths and inner products, acting as a Lorentz rotation between tangent spaces.
Key idea: to say a vector is unchanged as it moves, demand that its covariant derivative along the curve vanish, $\frac{D V^\mu}{\mathrm{d}\lambda}=u^\nu\nabla_\nu V^\mu=0$ ; the connection cancels the fake change of the coordinate basis, and curvature is exactly what makes the surviving answer depend on the path.

The core insight

To say a vector is unchanged as it moves along a curve, demand that its covariant derivative along the curve vanish, $\frac{D V^\mu}{\mathrm{d}\lambda}=u^\nu\nabla_\nu V^\mu=0$ . The connection supplies the correction that cancels the spurious change of the coordinate basis, so what remains is genuine geometric change — and curvature is what makes the answer depend on the path.

Where this lives today

The most literal confirmation that parallel transport is physics, not formalism, is Gravity Probe B (launched 2004, results published 2011): four fused-quartz gyroscopes in a $642\ \mathrm{km}$ polar orbit, each spin axis parallel-transported along its worldline through Earth's curved spacetime. The predicted geodetic precession of $6606.1\ \mathrm{mas/yr}$ was measured at $6601.8\pm18.3\ \mathrm{mas/yr}$ , and the far tinier frame-dragging (Lense–Thirring) rotation of $39.2\ \mathrm{mas/yr}$ , from Earth's rotation twisting spacetime, was measured at $37.2\pm7.2\ \mathrm{mas/yr}$ — both the geometric and the "gravitomagnetic" holonomy confirmed. The same mathematics runs far beyond gravity. In gauge theory a connection, its parallel transport, and the holonomy of a closed loop (the Wilson loop) are the language of the Aharonov–Bohm phase and of confinement in lattice QCD; in condensed matter the Berry phase acquired by a quantum state carried adiabatically around a loop in parameter space is holonomy of exactly this kind, and it underlies topological insulators and the quantum Hall effect. Numerical relativists transport polarization vectors and frames along null geodesics to compute gravitational lensing and the polarization of the cosmic microwave background, and every ray-tracer around a black hole is, at bottom, integrating equation (2). Curvature-as-holonomy remains an active open frontier wherever geometry and phase meet.

The physical picture

Stand on the Earth holding an arrow and vow never to turn it — always keep it "pointing the same way" as you stride ahead. Walk a closed circuit: down a meridian to the equator, a quarter of the way around, then back up to the pole. You return to your start and find the arrow swung to a visibly new direction. Nowhere did you rotate it by hand; the rotation was inflicted entirely by the curvature of the surface, harvested over the area your loop enclosed. The connection terms in the transport equation are precisely the bookkeeping that subtracts off the "fake" turning of your latitude-longitude basis, which points in different absolute directions at different places. Parallel transport holds the geometric arrow fixed and lets the components change to compensate — and on a curved surface no amount of component-changing can hide the leftover holonomy.

Figure

Parallel transport around a spherical octant: an arrow carried along the three geodesic edges N → A → B → N comes home rotated by 90°, exactly the enclosed area π/2 divided by R² — on a sphere the holonomy equals the solid angle the loop subtends

Read the transport ODE as a tiny instruction stamped out step by step: at each instant nudge the components by $-\Gamma^\mu{}_{\nu\rho}u^\nu V^\rho\,\mathrm{d}\lambda$ . Integrating these infinitesimal corrections along the path gives the finite transport, and because the correction at each point depends on where you are, the integrated result remembers the entire route — that memory is path dependence. The sharpest way to expose it is to carry a vector around a tiny closed parallelogram two ways, along edge $a$ then edge $b$ versus $b$ then $a$ . The mismatch between the two answers is the commutator of covariant derivatives, $[\nabla_\mu,\nabla_\nu]V^\rho=R^\rho{}_{\sigma\mu\nu}V^\sigma$ , so curvature is defined as the obstruction to transport being path-independent.

Figure

Path dependence in curved space: transporting V first along δa then δb versus first along δb then δa yields two different vectors at the far corner, and their gap ΔV is the Riemann tensor contracted with the enclosed area δa δb — in flat space the two routes agree and ΔV vanishes

The core equations

The definition is deliberately coordinate-free: a vector is parallel-transported along a curve when its rate of change as measured by the connection, in the direction the curve is heading, is zero. Here $u^\nu$ is the tangent to the curve and $\nabla_\nu$ the covariant derivative; the operator $D/\mathrm{d}\lambda$ is their contraction, the directional covariant derivative.

\frac{D V^\mu}{\mathrm{d}\lambda}\equiv u^\nu\nabla_\nu V^\mu = 0,\qquad u^\nu=\frac{\mathrm{d}x^\nu}{\mathrm{d}\lambda} \qquad (1)

where: $V^\mu$ — components of the transported vector; $\lambda$ — parameter along the curve $x^\mu(\lambda)$ ; $u^\nu$ — tangent vector to the curve; $\nabla_\nu$ — covariant derivative carrying the connection; $D/\mathrm{d}\lambda$ — directional covariant derivative along the curve. Physical reading: "no change relative to the geometry, in the direction of motion." Expanding the covariant derivative turns the clean statement (1) into a concrete equation for the components. The ordinary derivative alone is not enough — it would call a component-constant vector "unchanged" even when the basis is spinning — so the connection term is added to subtract that spin.

\frac{\mathrm{d}V^\mu}{\mathrm{d}\lambda}+\Gamma^\mu{}_{\nu\rho}\,\frac{\mathrm{d}x^\nu}{\mathrm{d}\lambda}\,V^\rho = 0 \qquad (2)

where: $\Gamma^\mu{}_{\nu\rho}$ — connection (Christoffel) coefficients; $\mathrm{d}V^\mu/\mathrm{d}\lambda$ — ordinary rate of change of the components along the curve. Physical reading: the components must actively change, at the rate $-\Gamma^\mu{}_{\nu\rho}u^\nu V^\rho$ , just to keep the true vector still. Because (2) is linear in $V$ , the map from the vector at the start of the curve to the vector at any later point is a matrix. That matrix is the parallel propagator, and it packages the whole solution.

V^\mu(\lambda)=P^\mu{}_\nu(\lambda,\lambda_0)\,V^\nu(\lambda_0),\qquad P(\lambda,\lambda_0)=\mathcal{P}\exp\!\Big(-\!\int_{\lambda_0}^{\lambda}\!A(\lambda')\,\mathrm{d}\lambda'\Big) \qquad (3)

where: $P^\mu{}_\nu$ — parallel propagator; $A^\mu{}_\rho(\lambda)=\Gamma^\mu{}_{\nu\rho}\,\mathrm{d}x^\nu/\mathrm{d}\lambda$ — the transport matrix; $\mathcal{P}$ — path ordering; $\lambda_0$ — starting parameter. Physical reading: transport is a linear isomorphism between tangent spaces, and the path ordering (needed because the $A$ 's at different points do not commute) is where path dependence hides. A curve that transports its own tangent vector along itself is the straightest path the geometry allows. Setting $V^\mu=u^\mu$ in (2) turns the transport equation into the geodesic equation, which is why free fall in general relativity is governed by the very same connection.

u^\nu\nabla_\nu u^\mu = 0\ \Longleftrightarrow\ \frac{\mathrm{d}^2 x^\mu}{\mathrm{d}\lambda^2}+\Gamma^\mu{}_{\nu\rho}\frac{\mathrm{d}x^\nu}{\mathrm{d}\lambda}\frac{\mathrm{d}x^\rho}{\mathrm{d}\lambda}=0 \qquad (4)

where: $u^\mu=\mathrm{d}x^\mu/\mathrm{d}\lambda$ — tangent vector, here also the transported object; $\lambda$ — an affine parameter, the one that makes the right-hand side exactly zero. Physical reading: a geodesic never turns relative to itself; it is parallel transport applied to the direction of travel. If the connection is metric-compatible, $\nabla_\alpha g_{\mu\nu}=0$ , then transport does more than move vectors — it moves them rigidly. The inner product of any two transported vectors is conserved along the curve.

\frac{\mathrm{d}}{\mathrm{d}\lambda}\big(g_{\mu\nu}V^\mu W^\nu\big)=g_{\mu\nu}\Big(\frac{DV^\mu}{\mathrm{d}\lambda}W^\nu+V^\mu\frac{DW^\nu}{\mathrm{d}\lambda}\Big)=0 \qquad (5)

where: $g_{\mu\nu}$ — metric tensor; $V^\mu,\,W^\nu$ — two parallel-transported vectors. Physical reading: lengths and angles are preserved, so transport with the Levi-Civita connection is a rotation (in Lorentzian signature, a Lorentz transformation) of tangent spaces — this fixes gyroscope magnitudes and light-cone structure. Finally, the failure of transport to be path-independent is quantified by the commutator of two covariant derivatives, and that commutator is exactly the Riemann tensor (plus torsion, which vanishes for Levi-Civita). Around a small loop the net change is this tensor contracted with the enclosed area.

[\nabla_\mu,\nabla_\nu]V^\rho = R^\rho{}_{\sigma\mu\nu}V^\sigma - T^\lambda{}_{\mu\nu}\nabla_\lambda V^\rho,\qquad \delta V^\rho \approx -R^\rho{}_{\sigma\mu\nu}\,V^\sigma\,\tfrac{1}{2}\oint x^\mu\,\mathrm{d}x^\nu \qquad (6)

where: $R^\rho{}_{\sigma\mu\nu}$ — Riemann curvature tensor; $T^\lambda{}_{\mu\nu}$ — torsion (zero in GR); $\oint x^\mu\,\mathrm{d}x^\nu$ — oriented enclosed area. Physical reading: curvature is holonomy per unit area — the local, measurable answer to "is this spacetime flat?"

Derivation AFrom vanishing covariant derivative to the transport ODE and its propagator

The goal is to convert the compact geometric definition of equation (1), that the directional covariant derivative of the vector is zero, into a first-order differential equation you can actually integrate, and then to solve that equation formally as a matrix propagator so the origin of path dependence is laid bare. We expand the covariant derivative with the connection, collapse the directional part with the chain rule, recognize the result as a linear ODE, and write its solution as a path-ordered exponential — showing at the end exactly why the ordering cannot be dropped.

Step 1. Write out the covariant derivative of a contravariant vector contracted with the tangent: $u^\nu\nabla_\nu V^\mu=u^\nu\big(\partial_\nu V^\mu+\Gamma^\mu{}_{\nu\rho}V^\rho\big)$ . Why: this is the definition of $\nabla$ on a vector — the ordinary derivative plus the connection correction — and contracting with $u^\nu$ projects it onto the direction of travel.

Step 2. Collapse the ordinary-derivative piece with the chain rule: $u^\nu\partial_\nu V^\mu=\frac{\mathrm{d}x^\nu}{\mathrm{d}\lambda}\partial_\nu V^\mu=\frac{\mathrm{d}V^\mu}{\mathrm{d}\lambda}$ . Why: along the curve the only dependence of $V^\mu$ on position is through $\lambda$ , so the directional derivative of a component is just its total $\lambda$ -derivative — no connection needed for that part.

Step 3. Set the sum to zero. Equation (1) becomes $\frac{\mathrm{d}V^\mu}{\mathrm{d}\lambda}+\Gamma^\mu{}_{\nu\rho}\frac{\mathrm{d}x^\nu}{\mathrm{d}\lambda}V^\rho=0$ , which is equation (2). Why: this is the operational transport rule — the $\Gamma$ term is precisely the spurious change caused by the varying basis, and cancelling it is what "keeping the vector fixed" means.

Step 4. Recast as a linear system $\frac{\mathrm{d}V^\mu}{\mathrm{d}\lambda}=-A^\mu{}_\rho(\lambda)V^\rho$ with $A^\mu{}_\rho=\Gamma^\mu{}_{\nu\rho}\,\mathrm{d}x^\nu/\mathrm{d}\lambda$ ; its solution is $V(\lambda)=\mathcal{P}\exp\!\big(-\int_{\lambda_0}^{\lambda}A\,\mathrm{d}\lambda'\big)V(\lambda_0)$ . Why: a linear ODE with a $\lambda$ -dependent matrix integrates to an exponential, but because $A(\lambda_1)$ and $A(\lambda_2)$ generally fail to commute, the naive exponential is wrong and the terms must be path-ordered — and that non-commutativity is the algebraic root of path dependence and holonomy.

Go deeper · Derivation B · 2 minHolonomy of a latitude circle on the unit sphere

Derivation BHolonomy of a latitude circle on the unit sphere

The goal is to compute, from scratch and in closed form, the angle by which a vector rotates after being parallel-transported once around a circle of constant colatitude $\theta_0$ on the unit 2-sphere, so that holonomy stops being a slogan and becomes a number tied to the enclosed solid angle. We take the sphere's Christoffel symbols as given, write the transport ODE along the latitude circle where only $\phi$ varies, decouple the two component equations into a single harmonic oscillator, read off the rotation frequency, and interpret the leftover angle after a full circuit. Metric $\mathrm{d}s^2=\mathrm{d}\theta^2+\sin^2\theta\,\mathrm{d}\phi^2$ ; nonzero symbols $\Gamma^\theta{}_{\phi\phi}=-\sin\theta\cos\theta$ and $\Gamma^\phi{}_{\theta\phi}=\Gamma^\phi{}_{\phi\theta}=\cot\theta$ .

Step 1. Along the circle $\theta=\theta_0$ only $\phi$ advances, so the transport equations reduce to $\frac{\mathrm{d}V^\theta}{\mathrm{d}\phi}=-\Gamma^\theta{}_{\phi\phi}V^\phi=\sin\theta_0\cos\theta_0\,V^\phi$ and $\frac{\mathrm{d}V^\phi}{\mathrm{d}\phi}=-\Gamma^\phi{}_{\theta\phi}V^\theta=-\cot\theta_0\,V^\theta$ . Why: $\mathrm{d}x^\nu/\mathrm{d}\phi$ picks out the $\phi$ direction, and only the two symbols with a $\phi$ index survive, coupling the two components to each other.

Step 2. Differentiate the first equation and substitute the second: $\frac{\mathrm{d}^2V^\theta}{\mathrm{d}\phi^2}=\sin\theta_0\cos\theta_0\frac{\mathrm{d}V^\phi}{\mathrm{d}\phi}=-\cos^2\theta_0\,V^\theta$ . Why: eliminating $V^\phi$ turns the coupled first-order pair into one harmonic-oscillator equation, whose angular frequency $\omega=\cos\theta_0$ is the rotation rate of the components with respect to $\phi$ .

Step 3. Solve: $V^\theta(\phi)=V^\theta(0)\cos(\omega\phi)+\ldots$ with $\omega=\cos\theta_0$ , so after a full circuit $\phi:0\to2\pi$ the frame components have advanced in phase by $\omega\cdot2\pi=2\pi\cos\theta_0$ . Why: the components trace out a rotation at frequency $\cos\theta_0$ ; relative to the fixed background the vector has swung by $2\pi\cos\theta_0$ over the loop.

Step 4. The net rotation relative to the local frame is $2\pi-2\pi\cos\theta_0=2\pi(1-\cos\theta_0)$ , which is exactly the solid angle $\Omega$ of the polar cap the circle bounds. Why: by Gauss–Bonnet the holonomy equals the integral of Gaussian curvature over the enclosed area, and on the unit sphere ( $K=1$ ) that integral is the cap's area, i.e. its solid angle — the number check below confirms it.

Boxed result: $\Delta\phi_{\text{rot}}=2\pi(1-\cos\theta_0)$ (the holonomy angle). Sanity checks: at the equator $\theta_0=\pi/2$ the rotation is $2\pi$ (a great circle is a geodesic and carries the tangent with zero net turn relative to the path); as $\theta_0\to0$ the tiny polar loop gives $\to0$ ; a $30°\text{N}$ circle ( $\theta_0=60°$ ) gives $2\pi(1-\tfrac12)=\pi=180°$ .

Go deeper · Derivation C · 1 minCurvature as the failure of transport to commute

Derivation CCurvature as the failure of transport to commute

The goal is to prove that the very tensor which measures curvature is nothing but the mismatch produced when a vector is parallel-transported around an infinitesimal parallelogram, thereby turning "curvature" into a concrete, operational instruction. We transport around a loop spanned by two small displacements taken in the two possible orders, recognize the difference as a commutator of covariant derivatives, expand that commutator explicitly with the connection, and read off the Riemann tensor as the coefficient acting on the vector.

Step 1. Over one infinitesimal edge $\delta a^\mu$ the transport rule (2) linearizes to $\delta V^\rho=-\Gamma^\rho{}_{\mu\sigma}V^\sigma\,\delta a^\mu$ . Why: for a small step the propagator is the identity minus the transport matrix times the step, which is all that is needed to second order in the loop.

Step 2. Transporting along $a$ then $b$ and subtracting the reverse order is, to second order in the displacements, the action of the commutator $[\nabla_\mu,\nabla_\nu]$ on $V^\rho$ contracted with the oriented area $\delta a^\mu\delta b^\nu$ . Why: the two routes differ only in the order of two covariant derivatives, and the antisymmetrized difference of operators is by definition their commutator times the loop area.

Step 3. Expand the commutator with the connection: $[\nabla_\mu,\nabla_\nu]V^\rho=\big(\partial_\mu\Gamma^\rho{}_{\nu\sigma}-\partial_\nu\Gamma^\rho{}_{\mu\sigma}+\Gamma^\rho{}_{\mu\lambda}\Gamma^\lambda{}_{\nu\sigma}-\Gamma^\rho{}_{\nu\lambda}\Gamma^\lambda{}_{\mu\sigma}\big)V^\sigma$ . Why: for a torsion-free connection the terms containing derivatives of $V$ cancel between the two orderings, leaving a pure tensor that multiplies $V$ alone.

Step 4. Identify the coefficient as the Riemann tensor, $R^\rho{}_{\sigma\mu\nu}=\partial_\mu\Gamma^\rho{}_{\nu\sigma}-\partial_\nu\Gamma^\rho{}_{\mu\sigma}+\Gamma^\rho{}_{\mu\lambda}\Gamma^\lambda{}_{\nu\sigma}-\Gamma^\rho{}_{\nu\lambda}\Gamma^\lambda{}_{\mu\sigma}$ , so $\Delta V^\rho=R^\rho{}_{\sigma\mu\nu}V^\sigma\,\delta a^\mu\delta b^\nu$ . Why: this is the definition of curvature — it is holonomy per unit area, and in flat space every derivative and product of $\Gamma$ cancels, giving $R=0$ and route-independent transport.

Boxed result: $\Delta V^\rho=R^\rho{}_{\sigma\mu\nu}\,V^\sigma\,\delta a^\mu\,\delta b^\nu$ (curvature as holonomy). Consistency: $R=0\Rightarrow\Delta V=0$ recovers flat-space path independence, matching Wald eq. (3.2.3) and MTW §11.4.

Limiting cases

Case	Result	Meaning
Flat space, Cartesian coordinates	$\Gamma^\mu{}_{\nu\rho}=0\Rightarrow \mathrm{d}V^\mu/\mathrm{d}\lambda=0$	components are literally constant; transport is path-independent and loop holonomy is trivial — "the same vector" everywhere is a meaningful statement
Flat space, curvilinear coordinates	$R^\rho{}_{\sigma\mu\nu}=0,\ \Gamma\neq0$	connection nonzero (e.g. polar coordinates) yet transport is still path-independent — curvature, not the connection, controls holonomy
Metric-compatible connection	$\frac{\mathrm{d}}{\mathrm{d}\lambda}(g_{\mu\nu}V^\mu W^\nu)=0$	inner products, lengths and angles are preserved; transport is an orthogonal / Lorentz rotation of tangent spaces
Tangent along a geodesic	$u^\nu\nabla_\nu u^\mu=0$	a geodesic parallel-transports its own tangent — the straightest possible path, with an affine parameter making the right-hand side exactly zero
Great circle on a sphere	$\Delta\phi_{\text{rot}}=2\pi$ (no net turn)	the equatorial case $\theta_0=\pi/2$ : the tangent is carried without turning relative to the path, confirming a geodesic is "straight"

Common trapHolding a vector's components fixed makes it "constant" — so a component-constant vector is parallel-transported

This is the single most common error, and it is false the moment the connection coefficients are nonzero. In polar coordinates on a perfectly flat plane, a vector with constant components $V^r,V^\phi$ is physically rotating, because the basis vectors $\partial_r,\partial_\phi$ point in different absolute directions at different points; the genuinely parallel-transported vector has changing components obeying $\mathrm{d}V^\mu/\mathrm{d}\lambda=-\Gamma^\mu{}_{\nu\rho}u^\nu V^\rho$ . Constant components mean "unchanged" only when $\Gamma^\mu{}_{\nu\rho}=0$ , i.e. Cartesian coordinates in flat space. A companion trap: assuming transport is path-independent so that "the vector at point $B$ " is well defined. On any curved manifold there is no such thing — the result of transport depends on the route, and comparing $V^\mu$ carried along two different curves is meaningless unless you also account for the holonomy between them. Always transport along a specified curve, and never subtract components at different points without one.

Worked examples

1 · Holonomy of a mid-latitude circle (fully numeric)

Parallel-transport a tangent vector once, due east, around the circle of latitude $30°\,\mathrm{N}$ on the unit sphere (colatitude $\theta_0=60°$ ). By what angle does it return rotated, and what solid angle does that equal?

Step 1. From the boxed result of the sphere derivation, the holonomy angle is $\Delta\phi_{\text{rot}}=2\pi(1-\cos\theta_0)$ . With $\theta_0=60°$ , $\cos\theta_0=\tfrac12$ , so $\Delta\phi_{\text{rot}}=2\pi\,(1-\tfrac12)=\pi\ \text{rad}=\mathbf{180°}$ . The returning arrow is anti-parallel to the one you set out with — a dramatic, easily pictured rotation.

Step 2. Check it against the geometry. The polar cap bounded by the $30°\,\mathrm{N}$ circle subtends solid angle $\Omega=2\pi(1-\cos\theta_0)=\pi\ \mathbf{sr}$ , exactly the holonomy angle, because on the unit sphere the Gaussian curvature is $K=1$ and Gauss–Bonnet makes the rotation equal to the enclosed area. Holonomy angle and enclosed solid angle are literally the same number.

Step 3. Interpret. Move the circle toward the pole and $\theta_0\to0$ : the cap shrinks, $\Omega\to0$ , and the rotation vanishes — a tiny loop feels almost no curvature. Push it to the equator, $\theta_0=90°$ , and $\Omega\to2\pi$ , i.e. a full turn, which is the geodesic (great-circle) case with no net turn relative to the path. The $30°$ circle sits between, giving a clean 180° flip enclosing π steradians.

2 · Geodetic precession of a Gravity Probe B gyroscope (numeric estimate)

A gyroscope orbits the Earth in a circle of radius $r=7018\ \mathrm{km}$ (a $642\ \mathrm{km}$ polar orbit). Its spin is parallel-transported along the worldline. Estimate the geodetic (de Sitter) precession of the spin axis per orbit and per year, using $GM_\oplus=3.986\times10^{14}\ \mathrm{m^3/s^2}$ and $c=2.998\times10^8\ \mathrm{m/s}$ .

Step 1. To leading post-Newtonian order the spin rotates, in the orbital plane, by $\Delta\alpha=\dfrac{3\pi GM}{c^2 r}$ per revolution. Evaluate the dimensionless group $\dfrac{GM}{c^2 r}=\dfrac{3.986\times10^{14}}{(8.988\times10^{16})(7.018\times10^{6})}=6.32\times10^{-10}$ , so $\Delta\alpha=3\pi\times6.32\times10^{-10}=\mathbf{5.96\times10^{-9}\ rad/orbit}$ . The tininess is the whole difficulty of the experiment.

Step 2. Count orbits per year. The period is $T=2\pi\sqrt{r^3/GM}=2\pi\sqrt{(7.018\times10^6)^3/3.986\times10^{14}}=\mathbf{5851\ s}\approx97.5\ \mathrm{min}$ , so one year of $3.156\times10^7\ \mathrm{s}$ holds $N=5394$ orbits.

Step 3. Accumulate: $\Delta\alpha_{\text{yr}}=N\,\Delta\alpha=5394\times5.96\times10^{-9}=3.21\times10^{-5}\ \mathrm{rad/yr}$ . Convert with $206265\ \mathrm{arcsec/rad}$ : $\Delta\alpha_{\text{yr}}=\mathbf{6.6\ arcsec/yr}$ . This is holonomy in spacetime — the spin vector carried around the curved orbit comes back tilted, just as the arrow did on the sphere. The leading estimate ≈ 6.6″/yr sits right on Gravity Probe B's measured 6602 ± 18 mas/yr (the full prediction 6606.1 mas/yr includes small oblateness and eccentricity corrections we dropped).

Physics of the portrait

Parallel transport is one of those ideas that had to be invented, not merely discovered: for decades tensor calculus manipulated indices with the connection buried inside the covariant derivative before anyone gave it a stand-alone geometric meaning. The person who did is the one whose name the metric connection now carries.

Origins · 1 minTullio Levi-Civita and how this was discovered

Tullio Levi-Civita

1873–1941 · Italian mathematician who gave the connection its geometric meaning through parallel displacement

Tullio Levi-Civita, working in the Italian school of absolute differential calculus he had built with his teacher Gregorio Ricci-Curbastro, published in 1917 the idea that turned tensor bookkeeping into geometry: parallelismo, parallel displacement of a vector along a curve on a Riemannian manifold. Ricci and Levi-Civita's 1900 memoir Méthodes de calcul différentiel absolu et leurs applications had already handed Einstein the machinery for general relativity — Einstein leaned on it heavily from 1913, and later said that without the "absolute differential calculus" the theory "would perhaps have got no further than its long clothes." But it was Levi-Civita's 1917 notion of parallelism that revealed why the Christoffel symbols appear: they are the instruction for carrying a vector without turning it, and their failure to make transport path-independent is curvature made visible. Within a year Hermann Weyl and Jan Schouten generalized the construction to affine connections divorced from any metric, and in the 1920s Élie Cartan recast the whole subject in moving frames and identified the holonomy group as the record of all loop rotations. Levi-Civita's name survives on the unique torsion-free, metric-compatible connection of general relativity — the Levi-Civita connection — the one for which parallel transport preserves lengths and angles. A committed anti-fascist, he was stripped of his Rome professorship and expelled from Italian academies by the 1938 racial laws, and died isolated in 1941; the geometry he founded now underwrites every gyroscope experiment and every numerical spacetime evolution.

History

1900

— Ricci-Curbastro and Levi-Civita publish the absolute differential calculus, the tensor machinery later used by Einstein

1915

— Einstein completes general relativity, the field equations phrased through the metric connection

1917

— Levi-Civita introduces parallel displacement, giving the connection its geometric meaning

1918

— Weyl and Schouten generalize to affine connections not derived from a metric

1926

— Cartan formalizes the holonomy group as the set of all loop-transport rotations

Going further

Texts. Misner, Thorne & Wheeler, Gravitation (1973), §10.2 and §11.4 · Wald, General Relativity (1984), §3.1 · Carroll, Spacetime and Geometry (2004), §3.1–3.3 · Schutz, A First Course in General Relativity, 2nd ed., §6.3 · Hartle, Gravity (2003), §20.3. Key results to keep in mind: the unit-sphere Christoffel symbols are $\Gamma^\theta{}_{\phi\phi}=-\sin\theta\cos\theta$ and $\Gamma^\phi{}_{\theta\phi}=\cot\theta$ , giving latitude holonomy $2\pi(1-\cos\theta_0)$ — $180°$ at $\theta_0=60°$ ; the Riemann tensor is the torsion-free commutator $[\nabla_\mu,\nabla_\nu]V^\rho=R^\rho{}_{\sigma\mu\nu}V^\sigma$ ; metric compatibility implies constancy of $g_{\mu\nu}V^\mu W^\nu$ under transport; and the geodetic-precession estimate $\Delta\alpha=3\pi GM/c^2r\approx6.6\ \mathrm{arcsec/yr}$ compares with the Gravity Probe B value of $6606.1\ \mathrm{mas/yr}$ . The published 2011 GP-B results are: geodetic precession $6601.8\pm18.3\ \mathrm{mas/yr}$ , frame-dragging $37.2\pm7.2\ \mathrm{mas/yr}$ .

Builds on: Covariant Derivative · The Affine Connection · Christoffel Symbols. Leads to: The Geodesic Equation · Metric Compatibility · Geodesic Deviation Equation · Killing Vectors.

Ask the tutor: ask me to parallel-transport a specific vector around a loop of your choosing — a sphere latitude circle, a cone, or a Schwarzschild orbit — and watch the holonomy fall out as a number.

Isn't "parallel transport" just moving a vector without changing it? Why all the machinery?

The catch is the phrase "without changing it." In a curved space, or even a flat one in curvilinear coordinates, you cannot change a vector's components by nothing and call it unchanged, because the reference frame those components are measured against is itself rotating from point to point. Parallel transport is the precise rule that says: change the components at exactly the rate needed to cancel the frame's rotation, so the geometric vector holds still. That rate is $-\Gamma^\mu{}_{\nu\rho}u^\nu V^\rho$ . All the machinery is there to encode "hold the true arrow fixed while the grid beneath it swings," which no simpler prescription can do.

If the vector is "unchanged," why does it come back rotated after a loop?

Because "unchanged" is defined locally and along a specified path — at each instant the vector does not turn relative to the geometry it is passing through. But those local no-turn instructions do not integrate to a global no-turn statement when the space is curved. Summing infinitesimal corrections that each depend on where you are yields a total that remembers the whole route, and on a closed loop the memory shows up as a net rotation. That rotation is holonomy, and its size per unit enclosed area is the Riemann curvature tensor. Zero rotation for every loop is equivalent to flatness, so the leftover turn is curvature caught in the act.

What is the difference between the connection being nonzero and the space being curved?

They are genuinely different things, and conflating them is a classic mistake. The connection $\Gamma^\mu{}_{\nu\rho}$ is coordinate-dependent: it is nonzero in polar coordinates even on a perfectly flat plane, simply because the basis vectors vary. Curvature is the Riemann tensor $R^\rho{}_{\sigma\mu\nu}$ , built from derivatives and products of $\Gamma$ in a way that is coordinate-independent. In flat space you can always find coordinates (Cartesian) where $\Gamma=0$ everywhere, and there $R=0$ too. Curved space is where no coordinate choice can kill $\Gamma$ over a region, and $R\neq0$ . Parallel transport is path-dependent if and only if $R\neq0$ , never merely because $\Gamma\neq0$ .

Why does the solution to the transport equation need "path ordering"?

The transport equation is linear, $\mathrm{d}V/\mathrm{d}\lambda=-A(\lambda)V$ , and a linear ODE with a constant matrix would solve to a plain matrix exponential. Here the matrix $A^\mu{}_\rho(\lambda)=\Gamma^\mu{}_{\nu\rho}\,\mathrm{d}x^\nu/\mathrm{d}\lambda$ changes along the curve, and the values at different points generally do not commute. When matrices fail to commute you cannot combine their exponentials by adding exponents; you must multiply the infinitesimal transports in the correct order, near end last. That instruction is path ordering, denoted $\mathcal{P}$ . Physically it is the same fact as holonomy: the order in which you accumulate transport matters, and reversing two steps leaves a curvature-sized residue.

How is parallel transport related to the geodesic equation?

A geodesic is the special curve that parallel-transports its own tangent vector. Take the transport equation (2), and instead of an arbitrary vector $V^\mu$ let the transported object be the tangent $u^\mu=\mathrm{d}x^\mu/\mathrm{d}\lambda$ . Substituting turns the first-order transport equation into the second-order geodesic equation, $\ddot x^\mu+\Gamma^\mu{}_{\nu\rho}\dot x^\nu\dot x^\rho=0$ . So "straightest possible path" means "the direction of travel never turns relative to the connection," which is why free-falling particles and light rays, which follow geodesics, are the physical realization of parallel transport applied to their own four-velocity.

Does metric compatibility matter, or is it a technicality?

It matters physically and is what ties the abstract connection to measurable geometry. A general affine connection lets you transport vectors, but their lengths and mutual angles can drift as you go. Metric compatibility, $\nabla_\alpha g_{\mu\nu}=0$ , forbids that drift: the inner product $g_{\mu\nu}V^\mu W^\nu$ of any two transported vectors is conserved, so transport becomes a rigid rotation (a Lorentz transformation in spacetime) rather than a general linear map. Combined with zero torsion this uniquely selects the Levi-Civita connection used in general relativity. Without it, a gyroscope's spin could change magnitude as it orbits, or a light cone could deform under transport — neither of which nature does.

Is this the same "connection" that appears in gauge theories like electromagnetism?

Structurally, yes — and this is one of the deepest unifications in physics. In general relativity the connection tells you how to parallel-transport a spacetime vector; in a gauge theory the connection (the gauge potential $A_\mu$ ) tells you how to parallel-transport an internal charge or colour vector. In both, curvature is the loop holonomy: the Riemann tensor in gravity, the field strength $F_{\mu\nu}$ in Yang–Mills. The Aharonov–Bohm phase is a Wilson-loop holonomy, exactly analogous to the sphere's rotation angle. The mathematics of connections, parallel transport, and holonomy is identical; only what is being transported — a tangent vector versus an internal-symmetry vector — differs.