CHRISTOFFEL SYMBOLS AND PARALLEL TRANSPORT

In flat space with Cartesian coordinates, the partial derivative of a vector field is perfectly well behaved: differentiate each component, and you get a tensor. The same operation in curved coordinates fails. When you write $\partial_\mu V^\nu$ in polar coordinates on a flat plane — let alone on a curved manifold — the result is not a tensor. It transforms wrongly under coordinate change, picking up spurious terms that come from the fact that the basis vectors $\partial_\mu$ themselves vary from point to point.

The fix is to add a correction term that compensates for the rotation of the basis. On a general manifold the Covariant derivative of a contravariant vector is

where $\Gamma^\nu{}_{\mu\rho}$ are the Christoffel symbols. The extra term subtracts out exactly the coordinate-dependent variation of the basis, leaving a result that transforms as a genuine $(1,1)$ tensor. In flat Cartesian space the $\Gamma$s vanish and $\nabla_\mu = \partial_\mu$. The moment the coordinates curve — or the manifold curves — the $\Gamma$s become non-trivial, and the covariant derivative is the only derivative that tells you something coordinate-invariant.

The correction terms $\Gamma$ are not arbitrary: they are determined uniquely by the Metric tensor. The formula is

Three derivatives of the metric, antisymmetrised in a specific pattern, contracted with the inverse metric $g^{\rho\sigma}$. Every term is computable directly from the components of $g_{\mu\nu}$ and their first derivatives — no additional structure is required, no choice is made. The geometry of the manifold, encoded in the metric, completely determines how vectors are connected from one tangent space to the next.

This connection is called the Levi-Civita connection, discovered by in 1917. It has two special properties that single it out uniquely. First, it is torsion-free: $\Gamma^\rho{}_{\mu\nu} = \Gamma^\rho{}_{\nu\mu}$ — the symbol is symmetric in its two lower indices, which means the connection does not introduce any spurious "twist" in the manifold. Second, it is metric-compatible: $\nabla_\rho g_{\mu\nu} = 0$ — the metric itself is covariantly constant, so the inner product of two parallel-transported vectors does not change. Among all connections on a Riemannian or pseudo-Riemannian manifold, the Levi-Civita connection is the unique one with both properties. This is why it is the connection of general relativity.

A vector is parallel-transported along a curve when it does not rotate or stretch as measured by the local geometry. More precisely, the covariant derivative of the vector along the curve vanishes: $\nabla_{\dot{x}} V = 0$. Written in components, this becomes an ordinary differential equation along the curve $x^\mu(\lambda)$:

$\frac{dV^\mu}{d\lambda} = -\Gamma^\mu{}_{\alpha\beta}\, \dot{x}^\alpha V^\beta$

At each step along the curve, the Christoffel symbols tell you how much to correct the rate of change of $V$ to keep it "parallel" in the intrinsic sense. On flat space, all $\Gamma = 0$, so $dV^\mu/d\lambda = 0$: parallel transport just means "keep the Cartesian components constant," which is exactly what you expect. On a sphere, the $\Gamma$s are non-trivial, and the vector must continuously adjust to stay tangent to the sphere while tracking the curvature of the path. The Parallel transport equation is the relativistic generalisation of "carry a vector without rotating it." A geodesic is the special case where you parallel-transport the tangent vector of the curve itself along the curve: the curve does not accelerate, which is the relativistic version of Newton's first law.

Here is the moment where curvature becomes unmistakable. Take a vector on the unit sphere. Parallel-transport it around a closed loop — three great-circle segments forming a triangle. The vector travels along each side without any local rotation. When it returns to the starting point, it is pointing in a different direction. The angle of rotation is

where $A_{\text{enclosed}}$ is the area of the enclosed region and $R$ is the sphere radius. For the octant triangle above — one eighth of the unit sphere, area $\pi/2$ — the holonomy angle is exactly $\pi/2 = 90°$. The vector arrives pointing 90° away from where it started, without ever having been rotated locally.

This is the Holonomy. It is the global effect of an infinitesimal connection: each infinitesimal step around the loop rotates the vector by a tiny Christoffel-symbol amount, and after a full loop the tiny rotations accumulate into a finite angle that depends only on the enclosed area. In flat space — no matter how complicated the loop — the holonomy is always zero: you get your vector back unchanged. Only a curved manifold can produce a non-zero holonomy. The Christoffel symbols are the local, infinitesimal version of this story; the Riemann curvature tensor (§08) packages the same information in a coordinate-invariant, tensorial form. The spherical-excess formula $\Delta\theta = A/R^2$ is the prototype for the relation between curvature, area, and holonomy that the Riemann tensor will generalise.

The Christoffel symbols are the workhorse of GR calculations, but they are not tensors themselves: their components depend on the choice of coordinates, and they can be made to vanish at any single point by choosing locally flat (normal) coordinates. What they encode is the connection — the rule for identifying tangent spaces at nearby points. The connection is an infrastructure layer that makes tensor calculus possible on a curved manifold.

The real, coordinate-invariant measure of curvature comes one level up: the Riemann curvature tensor $R^\rho{}_{\sigma\mu\nu}$ is built from the Christoffel symbols and their derivatives. It tells you exactly how much a vector rotates when parallel-transported around an infinitesimal loop. When the Riemann tensor vanishes everywhere, the manifold is flat; when it does not, the manifold is genuinely curved in a way that no coordinate transformation can hide. The Riemann tensor and its contractions — the Ricci tensor and the Ricci scalar — are what appear in the Einstein field equations on the left-hand side, telling spacetime how to curve in response to the matter and energy on the right.

The chain runs: metric $g_{\mu\nu}$ → Christoffel symbols $\Gamma^\rho{}_{\mu\nu}$ → Riemann tensor $R^\rho{}_{\sigma\mu\nu}$ → Einstein tensor $G_{\mu\nu}$ → field equations. Every link in that chain can be expressed in terms of derivatives of the metric. GR is, in a precise sense, a theory of the metric and nothing else.