Parallel Transport and Geodesics

A smooth manifold with a metric knows how to measure a vector at a point, but it has no idea how to compare a vector here with a vector there. The tangent spaces $T_pM$ and $T_qM$ at two different points are separate vector spaces with no built-in identification, so the innocent phrase "the same vector, moved over there" is undefined until you supply extra data. That data is a connection, and the operation it licenses is parallel transport: a rule for dragging a vector along a chosen curve while changing it by only the minimal amount the connection coefficients — the Christoffel symbols — demand. The defining condition could not be simpler to state: the covariant derivative of the transported vector along the curve vanishes. Written out along a path $x^\mu(\lambda)$ , a vector $V^\mu$ is parallel-transported when $\mathrm{d}V^\mu/\mathrm{d}\lambda + \Gamma^\mu{}_{\alpha\beta}\dot x^\alpha V^\beta = 0$ — a first-order linear ODE, so an initial vector propagates uniquely along the curve. When the connection is metric-compatible the transport preserves lengths and angles, acting at each step as a rotation. A geodesic is then the one curve that transports its own tangent vector along itself: the "straightest possible" path, the manifold's version of a straight line. For the Levi-Civita connection this autoparallel curve is simultaneously the extremal-length curve, so "straightest" and "shortest" coincide. And when you carry a vector around a closed loop it generally comes home rotated — the holonomy — with the rotation per unit area set by the Riemann curvature. That single fact binds the local calculus of connections to the global geometry of curvature, and it is the mathematical skeleton of general relativity, where free fall is geodesic motion and tidal forces are geodesic deviation.

Key takeaways

Parallel transport moves a vector along a curve while keeping it "as constant as possible" with respect to a connection, encoded by the vanishing covariant derivative $\mathrm{D}V^\mu/\mathrm{d}\lambda = 0$ .
The result is path-dependent on a curved manifold: transport around a closed loop returns a rotated vector, and that rotation — the holonomy — is the geometric meaning of curvature.
A geodesic is a self-parallel (autoparallel) curve, transporting its own tangent, giving $\ddot x^\mu + \Gamma^\mu{}_{\alpha\beta}\dot x^\alpha\dot x^\beta = 0$ .
For the Levi-Civita connection geodesics also extremize arc length (timelike geodesics extremize proper time), so "straightest" equals "shortest or longest".
Affine parametrization is built in: a non-affine parameter adds a term proportional to $\dot x^\mu$ , so only proper time or arc length keep the right-hand side zero.
Geodesics generalize Newton's first law — free-falling bodies follow timelike geodesics and light follows null geodesics, which is how gravity becomes geometry.
Key idea: A connection defines parallel transport via $\mathrm{D}V^\mu/\mathrm{d}\lambda = 0$ ; a geodesic is a curve that parallel-transports its own tangent, yielding $\ddot x^\mu + \Gamma^\mu{}_{\alpha\beta}\dot x^\alpha\dot x^\beta = 0$ , which for the Levi-Civita connection also extremizes path length.

Where this lives today

Parallel transport and geodesics are not historical furniture — they are the working language of gravitational physics and, increasingly, of quantum and data science. Every global-positioning fix is a geodesic calculation: GPS satellites move on timelike geodesics of the Earth's Schwarzschild-like field, and their clocks must be corrected by tens of microseconds a day or positioning drifts by kilometres. LIGO and Virgo detect gravitational waves through geodesic deviation directly — a passing wave modulates the separation of freely suspended mirrors by a fractional strain $h\sim10^{-21}$ , the tidal term $R^\mu{}_{\alpha\nu\beta}$ of equation (6) made audible. The Event Horizon Telescope images the photon ring of M87 and Sgr A by tracing null geodesics through the Kerr metric, and numerical relativity integrates the geodesic and Einstein equations on supercomputers to predict merger waveforms. The same mathematics escapes gravity entirely: the Berry phase is the holonomy of parallel transport of a quantum state around a loop in parameter space, measured in molecular physics, the quantum Hall effect, and topological insulators; in gauge theory the parallel propagator (6) is the Wilson line, and its loop trace, the Wilson loop, is a fundamental probe of confinement in lattice QCD. Even machine learning borrows it — natural-gradient methods follow geodesics of the Fisher-Rao metric on the manifold of probability distributions.

The physical picture

Carry a spear across the surface of the Earth, holding it "as straight as you can" relative to your motion — never twisting it about the local vertical. Walk a closed triangle: south down a meridian to the equator, a quarter-turn east along the equator, then back up to the pole. The spear arrives home pointing in a visibly different direction, rotated with no local force ever applied. That rotation is curvature made tangible. Parallel transport is path-dependent precisely because the sphere is curved, and the amount of turning is not arbitrary — it equals the solid angle the loop encloses. The first figure pins that down for the cleanest possible loop, the spherical octant, where the answer is exactly a right angle.

Figure

Parallel transport of a vector around a spherical octant on a sphere of radius R: the loop bounds one eighth of the sphere, area A = 4πR²/8 = (π/2)R², so the enclosed solid angle is Ω = A/R² = π/2, and the transported vector returns to the pole rotated by exactly this angle, 90°, with no local force ever applied

A geodesic is the path you trace if you "never turn the steering wheel": your velocity vector is parallel-transported along your own track. On a sphere these autoparallel curves are the great circles; in flat space they are ordinary straight lines. The test is local and sharp — feed the metric into the geodesic equation and the tangent's covariant acceleration must vanish. The equator passes ( $\Gamma^\theta{}_{\phi\phi}=-\sin\theta\cos\theta$ is zero there), but a circle of constant latitude fails, because that same coefficient is nonzero away from the equator, so an unseen sideways push is needed to hold the course. The second figure contrasts a genuine geodesic with an impostor and reads the numbers straight off the connection.

Figure

Great circles are geodesics of the sphere but latitude circles are not: the sphere metric ds² = R²(dθ² + sin²θ dφ²) gives the only relevant Christoffel symbol Γθφφ = −sinθ cosθ, which vanishes on the equator θ = 90° (autoparallel, no turning) but equals −0.433 at colatitude θ = 60° (not autoparallel, a sideways force is required to stay on the latitude circle)

Two notions of "best" curve coincide for the metric connection: the straightest (autoparallel) curve and the extremal-length curve. This is no accident. Metric compatibility plus zero torsion force the connection to be the unique Levi-Civita one, and for it the autoparallel equation is literally the Euler-Lagrange equation of the length functional. Transport with that connection preserves the inner product $g_{\mu\nu}V^\mu W^\nu$ of any two transported vectors, so lengths and angles ride along unchanged and transport acts as an element of the local rotation (or Lorentz) group — a fact we will lean on repeatedly below.

The core equations

The operational starting point is the parallel transport law: a vector is transported when its covariant derivative along the curve vanishes. Everything else — geodesics, holonomy, geodesic deviation — is a specialization or an iteration of this one equation.

\frac{\mathrm{D}V^\mu}{\mathrm{d}\lambda}\equiv\frac{\mathrm{d}V^\mu}{\mathrm{d}\lambda}+\Gamma^\mu{}_{\alpha\beta}\,\frac{\mathrm{d}x^\alpha}{\mathrm{d}\lambda}\,V^\beta=0 \qquad (1)

where: $V^\mu$ — components of the transported vector along the curve; $x^\alpha(\lambda)$ — the curve, parametrized by $\lambda$ ; $\Gamma^\mu{}_{\alpha\beta}$ — connection (Christoffel) coefficients; $\mathrm{D}/\mathrm{d}\lambda$ — the covariant (absolute) derivative along the curve. Being first-order and linear, it propagates any initial vector uniquely. A geodesic is the curve whose own tangent is parallel-transported along itself. Setting $V^\mu = \dot x^\mu$ in (1) gives the autoparallel equation, the geometric definition of "straightest".

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

where: $x^\mu(\lambda)$ — the geodesic; $\lambda$ — an affine parameter (proper time or arc length, up to $\lambda\to a\lambda+b$ ); $\Gamma^\mu{}_{\alpha\beta}$ — connection coefficients. The simple zero on the right holds only for an affine parameter; a general parameter adds a term $\propto\dot x^\mu$ . The connection that makes transport an isometry is the Levi-Civita connection, built from first derivatives of the metric. Its derivation from metric compatibility plus zero torsion is owned by aq-mp-christoffel-symbols-connections and aq-mp-metric-tensor; it is recalled here by name.

\Gamma^\lambda{}_{\mu\nu}=\tfrac{1}{2}g^{\lambda\sigma}\big(\partial_\mu g_{\sigma\nu}+\partial_\nu g_{\sigma\mu}-\partial_\sigma g_{\mu\nu}\big) \qquad (3)

where: $g_{\mu\nu}$ — metric tensor; $g^{\lambda\sigma}$ — inverse metric; $\partial_\mu$ — coordinate partial derivative. It is the unique symmetric ( $\Gamma^\lambda{}_{\mu\nu}=\Gamma^\lambda{}_{\nu\mu}$ ), metric-compatible ( $\nabla g=0$ ) connection, and it vanishes wherever $g$ is constant. Metric compatibility has an immediate payoff for transport: the inner product of any two transported vectors is conserved along the curve, so transport is a rotation, not a distortion.

\frac{\mathrm{d}}{\mathrm{d}\lambda}\big(g_{\mu\nu}V^\mu W^\nu\big)=0 \qquad (4)

where: $V^\mu, W^\nu$ — two parallel-transported vectors; $g_{\mu\nu}$ — metric with $\nabla g=0$ . In particular a geodesic preserves the norm of its own tangent, so timelike, null, and spacelike character never change en route. The metric geodesic follows equally from a variational principle. The cleanest functional is the quadratic "energy" — no square root, and its Euler-Lagrange equations are automatically affine, whereas the arc length $L$ gives the same paths but leaves the parameter free.

S[x]=\tfrac{1}{2}\int g_{\mu\nu}\,\dot x^\mu\dot x^\nu\,\mathrm{d}\lambda,\qquad L[x]=\int\!\sqrt{\,g_{\mu\nu}\,\dot x^\mu\dot x^\nu\,}\;\mathrm{d}\lambda \qquad (5)

where: $S$ — energy action, whose stationary curves are affinely-parametrized geodesics; $L$ — arc length, whose stationary curves are the same paths. Extremizing either recovers equation (2); $S$ is preferred because it fixes the parameter for you. (The formal solution of (1), $V^\mu(\lambda)=P^\mu{}_\nu\,V^\nu(\lambda_0)$ , is the path-ordered exponential $P=\mathcal{P}\exp(-\!\int\Gamma_\alpha\dot x^\alpha\,\mathrm{d}\lambda)$ — the parallel propagator, and the Wilson line of gauge theory in disguise.) Neighbouring geodesics do not stay parallel — their separation accelerates, and the tidal acceleration is the Riemann tensor contracted with the tangent. This is geodesic deviation.

\frac{\mathrm{D}^2\xi^\mu}{\mathrm{d}\tau^2}=-R^\mu{}_{\alpha\nu\beta}\,u^\alpha\xi^\nu u^\beta \qquad (6)

where: $\xi^\mu$ — infinitesimal separation between neighbouring geodesics; $u^\alpha=\dot x^\alpha$ — unit tangent; $R^\mu{}_{\alpha\nu\beta}$ — Riemann curvature tensor. In flat space $R=0$ and neighbours stay parallel; a nonzero $R$ is felt as tidal stretching and squeezing. Finally, transport around an infinitesimal closed loop returns a vector changed by the Riemann tensor times the enclosed area — the holonomy that the two figures visualized.

\Delta V^\rho=-R^\rho{}_{\sigma\mu\nu}\,V^\sigma\,\mathrm{d}a^\mu\,\mathrm{d}b^\nu \qquad (7)

where: $\mathrm{d}a^\mu,\mathrm{d}b^\nu$ — the two edge vectors of the loop; $R^\rho{}_{\sigma\mu\nu}$ — Riemann tensor. This is literally the commutator of covariant derivatives, $[\nabla_\mu,\nabla_\nu]V^\rho=R^\rho{}_{\sigma\mu\nu}V^\sigma$ : curvature is the failure of transport to be path-independent.

Derivation AParallel transport and the autoparallel geodesic

The goal is to build both foundational equations of this article from a single idea — that "moved without change" means "covariant derivative zero along the curve" — and to show that the geodesic is nothing more than the special curve which parallel-transports its own tangent vector. We restrict the covariant derivative to the tangent of a curve, demand it vanish, then feed the tangent back into its own transport law, so that the transport equation (1) and the geodesic equation (2) come from one construction rather than two.

Step 1. Take a vector field $V^\mu$ defined along the curve and act with the directional covariant derivative along the tangent $U^\alpha=\mathrm{d}x^\alpha/\mathrm{d}\lambda$ : $\nabla_U V^\mu = U^\alpha(\partial_\alpha V^\mu + \Gamma^\mu{}_{\alpha\beta}V^\beta)$ . Why: the covariant derivative is the only derivative on a manifold that produces a genuine vector, correcting the ordinary derivative for the twisting of the coordinate basis through $\Gamma$ .

Step 2. Collapse the ordinary-derivative piece with the chain rule: since $V^\mu$ lives only on the curve, $U^\alpha\partial_\alpha V^\mu = (\mathrm{d}x^\alpha/\mathrm{d}\lambda)\,\partial_\alpha V^\mu = \mathrm{d}V^\mu/\mathrm{d}\lambda$ . Why: this turns the field equation into an ordinary differential equation in the single parameter $\lambda$ , which is what makes transport a well-posed initial-value problem.

Step 3. Demand no covariant change along the curve, $\nabla_U V^\mu = 0$ , which is $\mathrm{d}V^\mu/\mathrm{d}\lambda + \Gamma^\mu{}_{\alpha\beta}\dot x^\alpha V^\beta = 0$ — equation (1). Why: "as constant as the connection permits" is exactly the statement that the covariant, not the ordinary, derivative vanishes; the $\Gamma$ term is the unavoidable correction the geometry imposes.

Step 4. Now choose the transported vector to be the curve's own tangent, $V^\mu = \dot x^\mu$ , and substitute: $\mathrm{d}\dot x^\mu/\mathrm{d}\lambda + \Gamma^\mu{}_{\alpha\beta}\dot x^\alpha\dot x^\beta = \ddot x^\mu + \Gamma^\mu{}_{\alpha\beta}\dot x^\alpha\dot x^\beta = 0$ . Why: a geodesic is defined as the curve that carries its velocity forward unchanged, so imposing self-parallel transport on the tangent is the definition, and equation (2) is its content. The construction used only a connection, never the metric — so autoparallels exist for any connection, separating "straight" (affine) from "short" (metric). *Verified: for constant $g$ (Cartesian flat space) all $\Gamma=0$ , giving $\ddot x^\mu=0$ , i.e. straight lines $x^\mu=a^\mu\lambda+b^\mu$ ; the tangent's norm is conserved by equation (4), confirming $\lambda$ is affine.*

Go deeper · Derivation B · 1 minThe geodesic as the extremal-length curve

Derivation BThe geodesic as the extremal-length curve

The aim is to prove that the "straightest" curve of the previous derivation is also the "shortest" (or longest), by extremizing the path length and recovering exactly equation (2) with the Christoffel symbols assembling themselves out of the metric derivatives. We vary the quadratic energy $L=\tfrac12 g_{\mu\nu}\dot x^\mu\dot x^\nu$ rather than the square-root arc length, since the two share the same stationary paths for an affine parameter and the quadratic form is far cleaner to differentiate.

Step 1. Adopt the Lagrangian $L=\tfrac12 g_{\mu\nu}(x)\,\dot x^\mu\dot x^\nu$ and apply the Euler-Lagrange equation $\frac{\mathrm{d}}{\mathrm{d}\lambda}\frac{\partial L}{\partial\dot x^\sigma}-\frac{\partial L}{\partial x^\sigma}=0$ . Why: geodesics are stationary points of length, and Euler-Lagrange is the machine that converts "stationary path" into a differential equation; the quadratic form avoids a square root and delivers an affine parameter automatically.

Step 2. Compute the two pieces: the conjugate momentum $\partial L/\partial\dot x^\sigma = g_{\sigma\nu}\dot x^\nu$ and the position derivative $\partial L/\partial x^\sigma = \tfrac12\,\partial_\sigma g_{\mu\nu}\,\dot x^\mu\dot x^\nu$ . Why: only the metric depends on position, so the entire "force" in the geodesic equation originates in the position-dependence of $g$ — that single term becomes every Christoffel symbol.

Step 3. Differentiate the momentum along the curve, $\frac{\mathrm{d}}{\mathrm{d}\lambda}(g_{\sigma\nu}\dot x^\nu) = g_{\sigma\nu}\ddot x^\nu + \partial_\mu g_{\sigma\nu}\,\dot x^\mu\dot x^\nu$ , and assemble Euler-Lagrange: $g_{\sigma\nu}\ddot x^\nu + \partial_\mu g_{\sigma\nu}\dot x^\mu\dot x^\nu - \tfrac12\partial_\sigma g_{\mu\nu}\dot x^\mu\dot x^\nu = 0$ . Why: the chain rule splits the change of momentum into a genuine acceleration and a piece tracking how the metric varies as the particle moves.

Step 4. Symmetrize the middle term over $\mu\nu$ (legal because it multiplies the symmetric $\dot x^\mu\dot x^\nu$ ), giving $g_{\sigma\nu}\ddot x^\nu + \tfrac12(\partial_\mu g_{\sigma\nu}+\partial_\nu g_{\sigma\mu}-\partial_\sigma g_{\mu\nu})\dot x^\mu\dot x^\nu = 0$ . Why: symmetrization is precisely the move that conjures the three-term Christoffel combination out of a single metric derivative — this is where the connection is born.

Step 5. Contract with the inverse metric $g^{\rho\sigma}$ to free the acceleration, landing on $\ddot x^\rho + \Gamma^\rho{}_{\mu\nu}\dot x^\mu\dot x^\nu = 0$ — the same equation (2) derived from parallel transport. Why: the bracket is exactly $2g_{\sigma\lambda}\Gamma^\lambda{}_{\mu\nu}$ , so "extremal length" and "self-parallel" are literally the same equation, which is the reconciliation of shortest with straightest. *Verified: the conserved quantity $g_{\mu\nu}\dot x^\mu\dot x^\nu=$ const along solutions confirms affine parametrization; the result matches the autoparallel equation of the previous derivation term for term.*

Go deeper · Derivation C · 1 minHolonomy around an infinitesimal loop and the Riemann tensor

Derivation CHolonomy around an infinitesimal loop and the Riemann tensor

The purpose here is to show that the path-dependence of parallel transport — the very effect the octant figure displayed — is controlled quantitatively by the Riemann curvature tensor, so that "transport around a small loop rotates a vector" becomes the precise identity (8). We transport a vector around a coordinate parallelogram with edges $\mathrm{d}a^\mu$ and $\mathrm{d}b^\nu$ , keep terms to second order in the edge lengths, and read off the curvature as the mismatch between going one way round and the other.

Step 1. Transport along a single infinitesimal leg using (1) in the form $\delta V^\rho = -\Gamma^\rho{}_{\mu\sigma}V^\sigma\,\mathrm{d}x^\mu$ . Why: this is the linearized transport law, the elementary increment out of which the whole loop is composed; to leading order transport is just repeated application of $-\Gamma$ .

Step 2. Compose four legs — advance by $\mathrm{d}a$ , then $\mathrm{d}b$ , then back by $\mathrm{d}a$ , then back by $\mathrm{d}b$ — expanding each $\Gamma$ and $V$ to first order in position so the round trip is captured to second order. Why: at first order the loop closes with no change; the rotation is a genuinely second-order effect, which is why curvature is built from derivatives and products of $\Gamma$ .

Step 3. Collect the surviving terms into $\Delta V^\rho = -R^\rho{}_{\sigma\mu\nu}V^\sigma\,\mathrm{d}a^\mu\,\mathrm{d}b^\nu$ with $R^\rho{}_{\sigma\mu\nu}=\partial_\mu\Gamma^\rho{}_{\nu\sigma}-\partial_\nu\Gamma^\rho{}_{\mu\sigma}+\Gamma^\rho{}_{\mu\lambda}\Gamma^\lambda{}_{\nu\sigma}-\Gamma^\rho{}_{\nu\lambda}\Gamma^\lambda{}_{\mu\sigma}$ . Why: the antisymmetric combination of " $\partial\Gamma$ " and " $\Gamma\Gamma$ " terms is exactly the object that fails to cancel between the two traversal orders, and it is coordinate-covariant — a true tensor.

Step 4. Recognize the result as the commutator of covariant derivatives, $[\nabla_\mu,\nabla_\nu]V^\rho = R^\rho{}_{\sigma\mu\nu}V^\sigma$ , and hence the definition of the Riemann tensor. Why: this identifies "transport does not commute" with "covariant derivatives do not commute", tying holonomy, curvature, and the Riemann tensor into one statement. *Verified: in flat space all $\Gamma=0\Rightarrow R=0\Rightarrow$ transport is path-independent, restoring a global notion of parallelism; on the unit sphere the same machinery gives the octant rotation of 90° and $R^\theta{}_{\phi\theta\phi}=\sin^2\theta$ , forwarded to aq-mp-riemann-curvature-tensor.*

Limiting cases

Special cases sharpen the meaning of transport and geodesics, and each recovers a familiar physical statement.

Case	Result	Meaning
Flat space, Cartesian coordinates	$\Gamma^\mu{}_{\alpha\beta}=0\ \Rightarrow\ \ddot x^\mu=0$	Geodesics are straight lines $x^\mu(\lambda)=a^\mu\lambda+b^\mu$ ; transport keeps components constant — Newton's first law.
Sphere $S^2$	geodesics $=$ great circles	Transport around a closed loop rotates a vector by the enclosed solid angle; the octant gives $\Omega=\pi/2$ , a $90°$ turn.
Null geodesics	$g_{\mu\nu}\dot x^\mu\dot x^\nu=0$	Light rays: arc length vanishes, so one cannot parametrize by length — an affine parameter must be chosen instead.
Non-affine parameter	$\ddot x^\mu+\Gamma^\mu{}_{\alpha\beta}\dot x^\alpha\dot x^\beta=f(\lambda)\,\dot x^\mu$	A general parameter adds a tangential source; setting $f=0$ defines affine parameters, unique up to $\lambda\to a\lambda+b$ .
Weak, slow limit of GR	$\ddot x^i\approx-\Gamma^i{}_{00}\approx-\partial_i\Phi$	The geodesic equation reduces to $\ddot{\mathbf{x}}=-\nabla\Phi$ — Newtonian gravity is the leading Christoffel term.

Common trapthe geodesic equation always has a zero on the right

It does not — that simple form holds only for an affine parameter. Parametrize the same geodesic by an arbitrary $\lambda$ (coordinate time, say, or any monotonic relabeling) and the equation becomes $\ddot x^\mu+\Gamma^\mu{}_{\alpha\beta}\dot x^\alpha\dot x^\beta=f(\lambda)\,\dot x^\mu$ , with a nonzero tangential term on the right. Drop that term and you get the wrong trajectory. The trap bites hardest for light: a null geodesic has zero arc length, so you cannot use length as its parameter at all, and must supply an affine parameter by hand (physically, one proportional to the photon's energy-momentum). The safe rule is to extremize the quadratic energy $S=\tfrac12\int g_{\mu\nu}\dot x^\mu\dot x^\nu\,\mathrm{d}\lambda$ , whose Euler-Lagrange equations are affine automatically, rather than the square-root length. A related confusion: parallel transport is only an isometry when the connection is metric-compatible — for a general connection, transported vectors can change length, and the tidy geometric picture of "rotation without distortion" is lost.

Worked examples

1 · Holonomy on the octant, transported by hand (fully worked)

Start at the north pole of a sphere of radius $R$ with a vector tangent to the surface, pointing south along the meridian at longitude $0°$ . Transport it south to the equator, east along the equator by $90°$ , then north back to the pole. By how much has it rotated, and how does that relate to the enclosed area?

Step 1. Down the first meridian, the vector points along the direction of travel (south). Parallel transport along a geodesic keeps a vector at a fixed angle to the geodesic's own tangent — and a meridian is a great circle, hence a geodesic — so at the equator the vector still points due south, now perpendicular to the equator.

Step 2. Along the equator (also a great circle, a geodesic) transport again preserves the angle to the path. The vector was perpendicular to the equator, so it stays perpendicular, pointing south, all the way to longitude $90°$ . It has not turned relative to the local south direction.

Step 3. Up the second meridian to the pole, the vector — still pointing south, i.e. away from the pole along this meridian — is carried home. But the two meridians (longitude $0°$ and $90°$ ) meet at the pole at an angle equal to their longitude difference, $90°$ . So the returned vector makes a $90°$ angle with the original. The holonomy is exactly $90°$ .

Step 4. Compare with area. The loop bounds one octant of the sphere, area $A=\tfrac18(4\pi R^2)=\tfrac{\pi}{2}R^2$ . The enclosed solid angle is $\Omega=A/R^2=\pi/2$ . The rotation angle equals the solid angle: $\Delta\varphi=\Omega=\pi/2=\mathbf{90°}$ — a special case of the Gauss-Bonnet theorem, $\Delta\varphi=\iint K\,\mathrm{d}A$ with Gauss curvature $K=1/R^2$ . Holonomy $=90°=$ enclosed solid angle, independent of $R$ .

2 · Transport around a latitude circle, and the Foucault pendulum (numeric)

A vector is parallel-transported once around a circle of constant colatitude $\theta_0$ on the unit sphere. Show the holonomy is $\Delta\varphi=2\pi(1-\cos\theta_0)$ , evaluate it at $\theta_0=60°$ , and connect it to the precession of a Foucault pendulum.

Step 1. The latitude circle bounds a spherical cap reaching from the pole down to colatitude $\theta_0$ . Its solid angle is $\Omega=\int_0^{\theta_0}\!\int_0^{2\pi}\sin\theta\,\mathrm{d}\phi\,\mathrm{d}\theta = 2\pi(1-\cos\theta_0)$ . By the Gauss-Bonnet result of Example 1 the holonomy equals this enclosed solid angle: $\Delta\varphi=\mathbf{2\pi(1-\cos\theta_0)}$ .

Step 2. At $\theta_0=60°$ : $\cos 60°=0.5$ , so $\Delta\varphi=2\pi(1-0.5)=\pi=\mathbf{180°}$ . A vector carried once around the $60°$ latitude circle comes back exactly reversed. Sanity checks: $\theta_0\to0$ (tiny loop near the pole) gives $\Delta\varphi\to0$ (flat, no rotation), and $\theta_0=90°$ (the equator, a geodesic) gives $\Delta\varphi=2\pi\equiv0$ , consistent with a geodesic transporting its tangent back to itself.

Step 3. The Foucault pendulum is this holonomy in the wild. Earth's rotation carries the pendulum once around its latitude circle each sidereal day, and its swing plane is parallel-transported. Relative to the ground the plane precesses by $2\pi\cos\theta_0=2\pi\sin(\text{latitude})$ per day. At Paris (latitude $48.85°$ ): $2\pi\sin 48.85°=2\pi(0.7528)=4.73$ rad per day $=271°$ per day, i.e. about $11.3°$ per hour — the rate Foucault measured under the Panthéon in 1851. The pendulum is a geometry experiment: its precession is the holonomy of a curved-surface connection.

3 · Deflection of starlight by the Sun from a null geodesic (numeric)

Light grazing the Sun follows a null geodesic of the Schwarzschild metric. The bending angle to leading order is $\delta=4GM/(c^2 b)$ , with $b$ the impact parameter. Evaluate it for a ray skimming the solar limb.

Step 1. Insert solar values: $GM_\odot=1.327\times10^{20}\ \mathrm{m^3\,s^{-2}}$ , $c^2=8.988\times10^{16}\ \mathrm{m^2\,s^{-2}}$ , and impact parameter $b=R_\odot=6.957\times10^{8}\ \mathrm{m}$ .

Step 2. Numerator $4GM_\odot=5.308\times10^{20}$ ; denominator $c^2 b=(8.988\times10^{16})(6.957\times10^{8})=6.254\times10^{25}$ . So $\delta=5.308\times10^{20}/6.254\times10^{25}=\mathbf{8.49\times10^{-6}}$ rad.

Step 3. Convert to arcseconds: $\delta=8.49\times10^{-6}\times206265=\mathbf{1.75''}$ . This is Einstein's 1915 prediction — exactly twice the value a naive Newtonian "photon in a gravitational field" gives, the extra factor of two coming from the curvature of space as well as time. $\delta=1.75''$ , confirmed by Eddington's 1919 eclipse expedition and now measured to better than $0.02\%$ by radio interferometry (VLBI).

Physics of the portrait

Parallel transport is one of those ideas that seems eternal but has a precise birthday and a single author. The Christoffel symbols were on paper for half a century before anyone realized they encoded a rule for moving vectors; the person who saw it was Tullio Levi-Civita, in 1917.

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

Tullio Levi-Civita

1873-1941 · the invention of parallel transport and modern tensor calculus

Tullio Levi-Civita, born in Padua, was already a master of the tensor calculus he had built as a student and collaborator of Gregorio Ricci-Curbastro — their 1900 memoir on the "absolute differential calculus" gave physics the covariant derivative and the machinery Einstein would ransack for general relativity. But his most beautiful single idea came in 1917, in a paper on the "notion of parallelism in any manifold". Einstein's new theory used the Christoffel symbols everywhere, yet no one had a clear geometric picture of what they meant. Levi-Civita supplied it: the symbols are the rule for transporting a vector along a curve so that it stays "as parallel to itself as the curvature allows", and the failure of that transport to close up around a loop is precisely the curvature. This turned Riemann's abstract $\mathrm{d}s^2$ and Christoffel's index gymnastics into a vivid, operational geometry of sliding arrows — and it revealed that curvature is a statement about transport, not just about a metric. The idea was found independently by Jan Schouten a year later and generalized by Hermann Weyl into the notion of an affine connection, seeding gauge theory. Levi-Civita corresponded warmly with Einstein, who called his tensor calculus indispensable. In 1938 the Fascist racial laws stripped him, a Jew, of his Rome professorship and his membership of Italian academies; he died isolated in 1941, but "the Levi-Civita connection" is now the default connection of all of Riemannian geometry and general relativity.

History

1854

— Riemann's habilitation lecture founds the intrinsic geometry of curved manifolds

1869

— Christoffel writes down the three-index connection symbols Γ

1900

— Ricci and Levi-Civita systematize the absolute differential (tensor) calculus

1917

— Levi-Civita defines parallel transport, giving the Christoffel symbols a geometric meaning

1984

— Berry identifies the quantum geometric phase as the holonomy of a connection

Going further

Texts. Stone & Goldbart, Mathematics for Physics, chapter on geometry and connections · Arfken, Weber & Harris, Mathematical Methods for Physicists, tensor-analysis chapter · Hassani, Mathematical Physics, differential-geometry part · Nakahara, Geometry, Topology and Physics, Ch. 7 (connections and holonomy) · Byron & Fuller, Mathematics of Classical and Quantum Physics, tensor chapter · Misner, Thorne & Wheeler, Gravitation, Ch. 10-13 · Wald, General Relativity, Ch. 3.

Provenance. All formulas are standard textbook results, re-verified for this article. A checker should: derive the transport equation (1) by restricting the covariant derivative to a curve's tangent and the geodesic equation (2) by setting $V^\mu=\dot x^\mu$ (Wald §3.1, Stone & Goldbart); re-derive (2) independently by Euler-Lagrange on $L=\tfrac12 g_{\mu\nu}\dot x^\mu\dot x^\nu$ ; confirm the holonomy identity $[\nabla_\mu,\nabla_\nu]V^\rho=R^\rho{}_{\sigma\mu\nu}V^\sigma$ (Arfken & Weber; Wald §3.2). Numerical checks: octant holonomy $=\Omega=A/R^2=\pi/2=90°$ ; latitude-circle holonomy $2\pi(1-\cos\theta_0)$ , giving $180°$ at $\theta_0=60°$ ; Foucault precession $2\pi\sin(\text{lat})$ per day $=11.3°/\mathrm{hr}$ at Paris ( $48.85°$ ); solar light deflection $\delta=4GM_\odot/(c^2R_\odot)=8.49\times10^{-6}$ rad $=1.75''$ ; sphere Christoffel $\Gamma^\theta{}_{\phi\phi}=-\sin\theta\cos\theta=-0.433$ at $\theta=60°$ , zero on the equator.

Prerequisites. Manifolds and Tangent Spaces · The Covariant Derivative · Christoffel Symbols and Affine Connections · The Metric Tensor · Raising and Lowering Indices. Leads to. The Riemann Curvature Tensor · Tensor Contraction and the Trace.

Ask the tutor: Ask me to parallel-transport a vector around any loop you name — a spherical cap, a cone, a Schwarzschild orbit — or to solve the geodesic equation for any metric, step by step from $g_{\mu\nu}$ to the trajectory.

Why isn't a metric enough to define parallel transport — why do I need a connection?

The metric measures vectors at a point, while transport compares vectors at different points, and those are separate jobs. Tangent spaces at two locations are distinct vector spaces with no canonical identification, so "the same vector, over there" is meaningless until a connection supplies the rule $\Gamma^\mu{}_{\alpha\beta}$ for how components change under an infinitesimal step. A metric does select a preferred connection — the Levi-Civita one — so on a Riemannian manifold you get transport for free. But the connection is the extra structure, and connections exist that are not metric-compatible or that carry torsion; those still define transport, just not one that preserves lengths. Keeping the two ideas apart is what makes gauge theory possible, where the connection is a Yang-Mills potential with no metric in sight.

What does it actually mean for a geodesic to be the "straightest" curve?

It means the curve does not turn relative to its own sense of forward: its tangent is parallel-transported along itself, so the covariant acceleration $\mathrm{D}\dot x^\mu/\mathrm{d}\lambda$ vanishes — no sideways change of velocity, only what the geometry unavoidably imposes. On a flat plane that is an ordinary straight line; on a sphere it is a great circle, as straight as anything can be while staying on the surface. "Straightest" is preferable to "shortest" because it is purely affine — it needs only a connection, not a metric. The remarkable fact, proved above, is that for the Levi-Civita connection the straightest curve is also the extremal-length curve, so on any Riemannian or Lorentzian manifold the two meanings coincide.

Why does light need an affine parameter instead of arc length?

Because along a light ray the arc length is identically zero. A null geodesic satisfies $g_{\mu\nu}\dot x^\mu\dot x^\nu=0$ at every point, so $\mathrm{d}s=0$ everywhere and you cannot use $s$ to label points — every point would carry the same "distance" of zero. The geodesic equation still holds, but only with respect to some affine parameter $\lambda$ for which the right-hand side stays zero; physically one takes $\lambda$ proportional to the photon's energy-momentum. For a massive particle there is no such problem: proper time $\tau$ is nonzero and itself affine, which is why timelike geodesics are naturally parametrized by $\tau$ . "Affine" is the primitive notion; "arc length" is merely one affine choice available to non-null curves.

How exactly is the path-dependence of transport the same thing as curvature?

Transport a vector between two points along two different paths; if the results differ, the manifold is curved. Sharpen that to an infinitesimal closed loop and the discrepancy is $\Delta V^\rho=-R^\rho{}_{\sigma\mu\nu}V^\sigma\,\mathrm{d}a^\mu\mathrm{d}b^\nu$ , linear in the enclosed area with the Riemann tensor as the coefficient. Equivalently, curvature is the failure of covariant derivatives to commute, $[\nabla_\mu,\nabla_\nu]V^\rho=R^\rho{}_{\sigma\mu\nu}V^\sigma$ — and $\nabla_\mu,\nabla_\nu$ generate transport in the two coordinate directions. So "transport is path-dependent", "the round trip rotates a vector", and "the Riemann tensor is nonzero" are three phrasings of one fact. When $R=0$ everywhere, transport is path-independent and a global notion of "parallel" exists — that is what makes flat space flat.

Does parallel transport always preserve the length of a vector?

Only if the connection is metric-compatible. For the Levi-Civita connection $\nabla g=0$ , which forces $\frac{\mathrm{d}}{\mathrm{d}\lambda}(g_{\mu\nu}V^\mu W^\nu)=0$ for any two transported vectors — lengths and angles are conserved, and transport acts at each step as a rotation (or, in Lorentzian signature, a Lorentz transformation). This is why a geodesic keeps the norm of its own tangent constant, and why timelike stays timelike and null stays null. For a general connection with $\nabla g\neq0$ , transported vectors can grow or shrink and the "rigid rotation" picture breaks down. Weyl explored exactly such a non-metric connection in 1918 as a unified field theory; the length-change it predicted was unphysical — itself an argument for metric compatibility.

Why is the factor of two in the light-bending formula so important?

It is the cleanest experimental discriminator between Newtonian gravity and general relativity. A photon treated as a fast Newtonian particle deflects by $2GM/(c^2 b)$ ; the full null-geodesic calculation in the Schwarzschild metric gives exactly twice that, $4GM/(c^2 b)$ . The doubling comes from the curvature of space adding an equal contribution to the temporal part Newton alone captures. For a ray grazing the Sun the predictions are $0.87''$ versus $1.75''$ , and Eddington's 1919 eclipse measurement landed on the relativistic value, making Einstein world-famous overnight. Modern radio interferometry (VLBI) confirms $1.75''$ to a few parts in $10^4$ , one of the most precise tests of geodesic motion we have.