Unpacking Curvature: The Role of Connections in Riemannian Geometry — Epoche B2
The Misconception of Curvature and Parallel Transport Introductory geometry presents curvature as a simple thing: how much a space 'bends'. A sphere is curved, a sheet of paper is not. The picture is useful but it hides the actual definition, and it fails as soon as one asks a question it cannot answer — how would an inhabitant of the surface, with no access to any surrounding space, detect the bending? The answer is that curvature is not about bending at all. It is about whether carrying a vector from one point to another gives an answer that depends on the route taken. This essay builds that claim from the ground up: what a tangent vector is, what a metric measures, what rule allows vectors at different points to be compared, and how the failure of that rule to be route-independent is exactly the Riemann curvature tensor. At the end we compute the effect for a sphere and read the answer off a Foucault pendulum. Manifolds and tangent spaces: the stage for geometry To understand curvature beyond simple bending, we first need the setting: a smooth manifold . A manifold $M$ is a space that locally resembles Euclidean space $\mathbb{R}^n$. The surface of the Earth is a 2-dimensional manifold; globally curved, but any small patch looks flat. At each point $p$ we define the tangent space $T_p M$, an $n$-dimensional vector space containing all the directions, or velocities, one could take starting from $p$. The crucial fact is that $T_pM$ and $T_qM$ are different vector spaces for $p \neq q$: there is no a priori way to say that a vector at $p$ and a vector at $q$ point "the same way". In local coordinates $(x^1, x^2, \dots, x^n)$ around $p$, any tangent vector $V \in T_p M$ is a linear combination of basis vectors: $$ V = V^1 \frac{\partial}{\partial x^1} + V^2 \frac{\partial}{\partial x^2} + \dots + V^n \frac{\partial}{\partial x^n} = V^i \frac{\partial}{\partial x^i}, $$ where the $V^i$ are the components of $V$, and $\partial/\partial x^i$ (abbreviated $\partial_i$) are the coordinate basis vectors. A vector field $X$ is a smooth assignment of a tangent vector $X_p \in T_p M$ to every point $p$ of an open subset of the manifold. Two vector fields also have a Lie bracket $[X,Y]$, itself a vector field, defined by its action on functions as $[X,Y]f = X(Yf) - Y(Xf)$. It measures the extent to which flowing along $X$ and then $Y$ differs from flowing along $Y$ and then $X$, and it vanishes for coordinate fields: $[\partial_i,\partial_j] = 0$. The metric tensor: measuring distances and angles A bare manifold provides smoothness and local coordinates and nothing else. To introduce length and angle we equip it with a Riemannian metric tensor $g$ [1] , a symmetric, positive-definite bilinear form on each tangent space. For any two tangent vectors $V, W \in T_p M$ it returns a real number $g(V,W)$, generalising the Euclidean dot product. In local coordinates the components are $g_{ij} = g(\partial_i, \partial_j)$, and the squared length of $V = V^i \partial_i$ is $$ |V|^2 = g(V, V) = g_{ij} V^i V^j, $$ with summation over repeated indices implied (Einstein's convention). The metric fixes lengths of curves and angles between vectors, and so fixes the intrinsic geometry. We also use the inverse metric $g^{kl}$, defined by $g^{kl}g_{lj} = \delta^k_j$ with $\delta^k_j$ the Kronecker delta. The unit sphere is the example to keep in mind: writing $\theta$ for colatitude (angle from the north pole) and $\varphi$ for longitude, its metric is $g_{\theta\theta} = 1$, $g_{\varphi\varphi} = \sin^2\theta$, $g_{\theta\varphi} = 0$, which simply records that a degree of longitude spans less distance near the poles. The challenge of parallel transport In Euclidean space, comparing vectors at different points is easy: translate one to the other, holding its Cartesian components fixed. This is parallel transport . On a curved manifold there is no globally fixed coordinate system, so 'holding the components fixed' is a statement about the chart, not about the geometry, and different charts give different answers. The sphere shows what goes wrong. Start at the equator with a vector pointing north; carry it up a meridian to the pole, down a second meridian back to the equator, and then along the equator to the start, keeping it as 'unturned' as possible at each step. It comes back pointing in a different direction. Nothing was done to it; the route did it. That is the phenomenon a definition has to capture (Fig. 1). A quadrilateral with corners P, P plus X, P plus Y and P plus X plus Y. An arrow at P is the vector Z. Along the amber route, first in the X direction then the Y direction, Z arrives at the far corner as the arrow Z-one. Along the blue route, first Y then X, it arrives as Z-two. The two arrows at the far corner point in visibly different directions, and the small dashed arc between their tips is the angle by which the two routes disagree. P P + X P + Y P + X + Y X Y Z Z 1 Z 2 route X then Y route Y then X Fig. 1 — Parallel transport around a small quadrilateral. The vector $Z$ at $P$ is carried to the opposite corner along the amber route ($X$ first, then $Y$) and along the blue route ($Y$ first, then $X$). On a flat manifold the two results coincide. On a curved one they do not: the dashed arc is the angle between $Z_1$ and $Z_2$, and to leading order in the size of the loop that discrepancy is $R(X,Y)Z$. Connections and covariant derivatives The missing structure is a connection , a rule for differentiating vector fields on a manifold. For vector fields $X$ and $Y$, the covariant derivative $\nabla_X Y$ measures how $Y$ changes along the direction of $X$, and it is a vector field again. In local coordinates a connection is fixed by what it does to the basis fields, which is recorded by the Christoffel symbols $\Gamma^k_{ij}$: $$ \nabla_{\partial_j} \partial_i = \Gamma^k_{ij} \partial_k. $$ These coefficients say how the coordinate basis twists as one moves across the manifold. Parallel transport of a vect