Book 9A

© 2026 NeckPinch · www.neckpinch.com · All rights reserved.

Course 9Book 9A: Metrics, Connections and CurvatureChapter 2

Connections

How to differentiate a vector field, parallel transport, and the Foucault pendulum.

28 min read · Updated Oct 3, 2026

Read with Lee, Introduction to Riemannian Manifolds (2nd edition), chapter 4 (connections: covariant derivatives of vector fields and tensors, vector fields along curves, parallel transport) and the first half of chapter 5 (the Levi-Civita connection). Petersen's Riemannian Geometry, chapter 2, is the second voice.

In this chapter · 6 sections
  1. 2.1The Foucault pendulum
  2. 2.2Connections
  3. 2.3The Levi-Civita connection
  4. 2.4Parallel transport
  5. 2.5History
  6. 2.6Exercises

To differentiate a function you compare its values at nearby points. To differentiate a vector field you would compare vectors at nearby points, but they live in different tangent spaces, and a manifold gives no way to compare them. In Rn\mathbb{R}^n you differentiate component by component, because all tangent spaces are identified with Rn\mathbb{R}^n. On a manifold, the partial derivatives ∂jXi\partial_jX^i of the components depend on the coordinates and do not form a tensor. The Lie derivative of 8A.6 Flows and the Lie Derivative compares vectors by dragging them along a flow, but it depends on the whole vector field, not just its direction at a point.

A connection is the extra structure that fixes this, and a Riemannian metric determines one: the Levi-Civita connection. It gives a covariant derivative ∇\nabla of every tensor field, the ∇i\nabla_i in every formula of the Ricci flow. It also gives parallel transport, a way of carrying a vector along a curve without "turning" it. On a curved manifold, parallel transport around a loop returns a vector rotated, and the rotation measures the curvature inside the loop. Physics shows this in the Foucault pendulum, in gyroscopes orbiting the Earth, and in light twisting through a coiled optical fibre.

By the end of this chapter you will be able to:

  • state the axioms of a connection and compute with Christoffel symbols;
  • derive the Levi-Civita connection from the Koszul formula, and its Christoffel symbols from the metric;
  • differentiate tensor fields covariantly, in index notation;
  • solve the parallel transport equation along a curve, and compute holonomy around a circle on the sphere;
  • explain why the difference of two connections, and hence the time derivative of the Christoffel symbols under the Ricci flow, is a tensor.

The Foucault pendulum

In the world Data A pendulum that measures the Earth's curvature

In 1851 Léon Foucault hung a heavy pendulum in the Panthéon in Paris and showed that its plane of swing slowly turns relative to the floor. The turning is caused by the Earth's rotation: at latitude φ\varphi the swing plane turns relative to the ground at the rate Ωsin⁡φ\Omega\sin\varphi, where Ω\Omega is the Earth's rate of rotation, one turn per sidereal day of about 23.9323.93 hours. At the poles the plane makes a full turn in a day; at the equator it does not turn at all; in Paris (φ≈48.85°\varphi \approx 48.85°) a full turn takes 23.93/sin⁡48.85°≈31.823.93/\sin 48.85° \approx 31.8 hours.

The geometric reading: over one day, the pendulum is carried once around its circle of latitude, and its swing plane is (to a good approximation) parallel-transported along that circle. Parallel transport around the circle rotates vectors by an angle equal to the area of the polar cap the circle encloses, 2π(1−sin⁡φ)2\pi(1 - \sin\varphi), on the unit sphere. Measured against the local north direction, which is how the floor sees it, that is a rotation by 2πsin⁡φ2\pi\sin\varphi in the opposite sense. You compute this below (Exercise 2.4), and Figure 2.2 shows it by unrolling a cone.

Connections

An affine connection on MM is a map (X,Y)↦∇XY(X, Y) \mapsto \nabla_XY taking two vector fields to a vector field, such that

  • ∇XY\nabla_XY is linear over functions in XX: ∇fX+hX′Y=f∇XY+h∇X′Y\nabla_{fX + hX'}Y = f\nabla_XY + h\nabla_{X'}Y;
  • ∇XY\nabla_XY is linear over R\mathbb{R} in YY and satisfies the product rule ∇X(fY)=(Xf) Y+f ∇XY\nabla_X(fY) = (Xf)\,Y + f\,\nabla_XY.

The first property means (∇XY)(p)(\nabla_XY)(p) depends only on XpX_p, so ∇vY\nabla_vY makes sense for a single vector vv: it is the derivative of YY in the direction vv. The second makes it a derivative. In coordinates, set

∇∂i∂j=Γijk ∂k;\nabla_{\partial_i}\partial_j = \Gamma_{ij}^k\,\partial_k;

the n3n^3 functions Γijk\Gamma_{ij}^k are the Christoffel symbols, and the axioms give

∇XY=(Xi∂iYk+XiYjΓijk)∂k.\nabla_XY = \big(X^i\partial_iY^k + X^iY^j\Gamma_{ij}^k\big)\partial_k.

The Christoffel symbols are the correction to the naive componentwise derivative. They are not the components of a tensor: they depend on the coordinates, and they can be made to vanish at any one point by a choice of coordinates (9A.3 Geodesics and the Exponential Map). But the difference of two connections, ∇XY−∇~XY\nabla_XY - \tilde\nabla_XY, is linear over functions in both XX and YY (the (Xf)Y(Xf)Y terms cancel), so it is a (1,2)(1, 2)-tensor field (8A.7 Tensors and Index Notation).

Covariant derivatives of tensors. A connection extends uniquely to all tensor fields so that it agrees with XfXf on functions, obeys the product rule for tensor products, and commutes with contractions. On a 1-form, (∇Xω)(Y)=X(ω(Y))−ω(∇XY)(\nabla_X\omega)(Y) = X(\omega(Y)) - \omega(\nabla_XY). The total covariant derivative ∇T\nabla T of a (k,l)(k, l)-tensor is a (k,l+1)(k, l + 1)-tensor, with components written ∇iT⋯⋯\nabla_iT^{\cdots}_{\cdots} (8A.7 Tensors and Index Notation):

∇iYk=∂iYk+ΓijkYj,∇iωj=∂iωj−Γijkωk,\nabla_iY^k = \partial_iY^k + \Gamma_{ij}^kY^j, \qquad \nabla_i\omega_j = \partial_i\omega_j - \Gamma_{ij}^k\omega_k,
∇iTjk=∂iTjk−ΓijlTlk−ΓiklTjl,\nabla_iT_{jk} = \partial_iT_{jk} - \Gamma_{ij}^lT_{lk} - \Gamma_{ik}^lT_{jl},

with one +Γ+\Gamma for each upper index and one −Γ-\Gamma for each lower index. Second derivatives are ∇2T=∇(∇T)\nabla^2T = \nabla(\nabla T), with components ∇i∇jT\nabla_i\nabla_jT; for a function, ∇i∇jf=∂i∂jf−Γijk∂kf\nabla_i\nabla_jf = \partial_i\partial_jf - \Gamma_{ij}^k\partial_kf is the Hessian (9A.6 The Laplacian and the Bochner Formula).

The Levi-Civita connection

A connection on a Riemannian manifold is

  • compatible with the metric if X⟨Y,Z⟩=⟨∇XY,Z⟩+⟨Y,∇XZ⟩X\langle Y, Z\rangle = \langle\nabla_XY, Z\rangle + \langle Y, \nabla_XZ\rangle, that is, ∇g=0\nabla g = 0;
  • torsion-free (symmetric) if ∇XY−∇YX=[X,Y]\nabla_XY - \nabla_YX = [X, Y], that is, Γijk=Γjik\Gamma_{ij}^k = \Gamma_{ji}^k in every chart.
Theorem 2.1 The fundamental theorem of Riemannian geometry

On a Riemannian manifold (M,g)(M, g) there is exactly one connection that is compatible with gg and torsion-free, the Levi-Civita connection. It is given by the Koszul formula

2⟨∇XY,Z⟩=X⟨Y,Z⟩+Y⟨Z,X⟩−Z⟨X,Y⟩+⟨[X,Y],Z⟩−⟨[Y,Z],X⟩+⟨[Z,X],Y⟩,2\langle\nabla_XY, Z\rangle = X\langle Y, Z\rangle + Y\langle Z, X\rangle - Z\langle X, Y\rangle + \langle[X, Y], Z\rangle - \langle[Y, Z], X\rangle + \langle[Z, X], Y\rangle,

and its Christoffel symbols are

Γijk=12gkl(∂igjl+∂jgil−∂lgij).\Gamma_{ij}^k = \tfrac12g^{kl}\big(\partial_ig_{jl} + \partial_jg_{il} - \partial_lg_{ij}\big).

Proof. Uniqueness. Suppose ∇\nabla is compatible and torsion-free. Write compatibility three times, cyclically:

X⟨Y,Z⟩=⟨∇XY,Z⟩+⟨Y,∇XZ⟩,X\langle Y, Z\rangle = \langle\nabla_XY, Z\rangle + \langle Y, \nabla_XZ\rangle,
Y⟨Z,X⟩=⟨∇YZ,X⟩+⟨Z,∇YX⟩,Y\langle Z, X\rangle = \langle\nabla_YZ, X\rangle + \langle Z, \nabla_YX\rangle,
Z⟨X,Y⟩=⟨∇ZX,Y⟩+⟨X,∇ZY⟩.Z\langle X, Y\rangle = \langle\nabla_ZX, Y\rangle + \langle X, \nabla_ZY\rangle.

Add the first two and subtract the third, and use torsion-freeness in the form ∇XZ=∇ZX+[X,Z]\nabla_XZ = \nabla_ZX + [X, Z], ∇YX=∇XY−[X,Y]\nabla_YX = \nabla_XY - [X, Y] and ∇ZY=∇YZ−[Y,Z]\nabla_ZY = \nabla_YZ - [Y, Z]. The terms rearrange to the Koszul formula. Its right-hand side involves only the metric and brackets, and since ZZ is arbitrary it determines ∇XY\nabla_XY.

Existence. For fixed X,YX, Y, the right-hand side of the Koszul formula is linear over functions in ZZ (check the fZfZ terms: they cancel in pairs), so it is a 1-form in ZZ, and by the nondegeneracy of gg it equals 2⟨W,Z⟩2\langle W, Z\rangle for a unique vector field WW. Define ∇XY=W\nabla_XY = W; checking the connection axioms, compatibility and torsion-freeness is direct algebra (Lee, chapter 5).

Christoffel symbols. Put X=∂iX = \partial_i, Y=∂jY = \partial_j, Z=∂lZ = \partial_l in the Koszul formula; coordinate fields commute, so 2Γijmgml=∂igjl+∂jgil−∂lgij2\Gamma_{ij}^mg_{ml} = \partial_ig_{jl} + \partial_jg_{il} - \partial_lg_{ij}. Multiply by gklg^{kl}.

From now on ∇\nabla is always the Levi-Civita connection. Isometries preserve it: if ϕ\phi is an isometry, ϕ∗(∇XY)=∇ϕ∗Xϕ∗Y\phi_*(\nabla_XY) = \nabla_{\phi_*X}\phi_*Y, because the Koszul formula involves only the metric and brackets.

The geometric picture. For a submanifold M⊂RNM \subset \mathbb{R}^N with the induced metric, the Levi-Civita connection is

∇XY=(DˉXY)⊤,\nabla_XY = (\bar D_XY)^{\top},

the ordinary derivative of YY in RN\mathbb{R}^N followed by orthogonal projection to TMTM (Exercise 2.5). On the sphere: differentiate in space, then throw away the part normal to the sphere. The thrown-away part is the second fundamental form of 8A.9 The Curvature of Surfaces (9A.8 Submanifolds and Minimal Surfaces).

Examples. In Euclidean coordinates all Γijk=0\Gamma_{ij}^k = 0. In polar coordinates on the plane, g=dr2+r2dθ2g = dr^2 + r^2d\theta^2, the nonzero symbols are Γθθr=−r\Gamma_{\theta\theta}^r = -r and Γrθθ=Γθrθ=1r\Gamma_{r\theta}^\theta = \Gamma_{\theta r}^\theta = \frac1r. For the warped product dr2+φ(r)2dθ2dr^2 + \varphi(r)^2d\theta^2, which includes the sphere (φ=sin⁡r\varphi = \sin r, rr the colatitude),

Γθθr=−φφ′,Γrθθ=Γθrθ=φ′φ,\Gamma_{\theta\theta}^r = -\varphi\varphi', \qquad \Gamma_{r\theta}^\theta = \Gamma_{\theta r}^\theta = \frac{\varphi'}{\varphi},

and all others vanish (Exercise 2.3).

Parallel transport

Along a curve γ:I→M\gamma : I \to M, a vector field along γ\gamma is a smooth assignment V(t)∈Tγ(t)MV(t) \in T_{\gamma(t)}M; the velocity γ′\gamma' is one. A connection gives its covariant derivative along γ\gamma,

DtV=(V˙k+Γijk(γ(t)) γ˙iVj)∂k,D_tV = \big(\dot V^k + \Gamma_{ij}^k(\gamma(t))\,\dot\gamma^iV^j\big)\partial_k,

which equals ∇γ′V~\nabla_{\gamma'}\tilde V for any extension V~\tilde V of VV. VV is parallel along γ\gamma if DtV=0D_tV = 0. This is a linear system of ordinary differential equations for the components Vk(t)V^k(t), so by 2B.10 Ordinary Differential Equations each initial vector V(t0)V(t_0) extends uniquely to a parallel field along the whole curve. The map

Pt0→t1γ:Tγ(t0)M→Tγ(t1)MP_{t_0\to t_1}^\gamma : T_{\gamma(t_0)}M \to T_{\gamma(t_1)}M

taking V(t0)V(t_0) to V(t1)V(t_1) is parallel transport. It is linear, invertible, and, for the Levi-Civita connection, an isometry: if VV and WW are parallel, ddt⟨V,W⟩=⟨DtV,W⟩+⟨V,DtW⟩=0\frac{d}{dt}\langle V, W\rangle = \langle D_tV, W\rangle + \langle V, D_tW\rangle = 0.

Parallel transport recovers the connection: ∇vY=ddt∣t=0Pt→0γY(γ(t))\nabla_vY = \frac{d}{dt}\big|_{t = 0}P^\gamma_{t\to0}Y(\gamma(t)) for any curve with γ′(0)=v\gamma'(0) = v. Covariant differentiation is differentiation after using parallel transport to bring the vectors into one tangent space.

Holonomy. Parallel transport around a closed loop at pp is an isometry of TpMT_pM, the holonomy of the loop. In Rn\mathbb{R}^n it is always the identity. On the sphere it is a rotation: around a geodesic triangle, by the triangle's area (Figure 2.1); around a simple loop on any surface, by ∬DK dA\iint_DK\,dA, the total curvature enclosed. This is Gauss–Bonnet (8A.9 The Curvature of Surfaces) in another form, and it is the first of the meanings of curvature in 9A.4 Curvature and What It Means: curvature is the infinitesimal holonomy.

Figure 2.1. A vector parallel-transported around the octant triangle (three right angles, area π2\frac{\pi}{2}). Along each edge, a geodesic, the vector keeps a constant angle with the edge. It leaves the north pole pointing down the first meridian and returns pointing down the other: rotated by π2\frac{\pi}{2}, the enclosed area times the curvature 11.

The holonomy can be nontrivial even where the curvature vanishes, if the loop surrounds a singular point or a hole. A flat cone with cone angle α\alpha has holonomy rotation 2π−α2\pi - \alpha around its apex. This is how the Foucault pendulum is usually explained: the cone tangent to the Earth along a circle of latitude has the same parallel transport along that circle as the sphere, and the cone can be unrolled flat (Figure 2.2).

Figure 2.2. The Foucault pendulum at the latitude of Paris (φ=48.85°\varphi = 48.85°). Left: the cone tangent to the sphere along the circle of latitude. Right: the cone unrolled. It is a flat sector of angle 2πsin⁡φ≈271°2\pi\sin\varphi \approx 271°, and parallel transport on it is ordinary translation (arrows). When the two cut edges are glued back, the arrow returns making a different angle with the circle: the missing 2π(1−sin⁡φ)≈89°2\pi(1 - \sin\varphi) \approx 89° is the holonomy, the area of the polar cap.
In the world Data Gravity Probe B

In general relativity a freely falling gyroscope's spin axis is parallel-transported (more precisely, Fermi–Walker transported) along its world line in curved spacetime. For a gyroscope in Earth orbit the result, relative to the distant stars, is the geodetic precession predicted by Willem de Sitter in 1916, together with a much smaller frame-dragging effect from the Earth's rotation. NASA and Stanford's Gravity Probe B carried four quartz gyroscopes in a polar orbit at about 642642 km from 2004 to 2005. The final results (Everitt and colleagues, Physical Review Letters, 2011) were a geodetic precession of −6601.8±18.3-6601.8 \pm 18.3 milliarcseconds per year, against a prediction of −6606.1-6606.1, and frame-dragging of −37.2±7.2-37.2 \pm 7.2, against −39.2-39.2: about 6.66.6 arcseconds a year, the holonomy of spacetime curvature measured directly.

In the world Model Light twisting in a coiled fibre

The polarisation of light in an optical fibre is a vector perpendicular to the fibre's direction, and as the light travels it is parallel-transported along the curve that the fibre's unit tangent traces on the sphere of directions. If that curve is a closed loop (the fibre wound into a helix, its tangent sweeping out a cone), the polarisation comes back rotated by the solid angle enclosed: the holonomy of the round sphere. Akira Tomita and Raymond Chiao measured this rotation in a helically wound fibre in 1986, one of the first observations of a geometric phase (Michael Berry, 1984).

Where this goes The connection under the Ricci flow

Under a Ricci flow g(t)g(t), the Levi-Civita connection changes with time. Each ∇g(t)\nabla_{g(t)} is a connection, so the time derivative ∂tΓijk\partial_t\Gamma_{ij}^k, a limit of differences of connections, is a tensor, and that makes it computable by working at a point in coordinates where Γ=0\Gamma = 0. You derive its formula in the last exercise. It is the first step of every evolution equation in 11A.2 How Curvature Evolves.

History

Elwin Bruno Christoffel introduced the symbols named after him in 1869, studying when two quadratic differential forms are equivalent. Gregorio Ricci-Curbastro and his student Tullio Levi-Civita developed the "absolute differential calculus" of tensors in a joint memoir of 1900, the calculus Einstein learned for general relativity. Parallel transport was introduced by Levi-Civita in 1917; Hermann Weyl and Élie Cartan generalised connections in the following years, and Cartan introduced holonomy in the 1920s. The Koszul formula is named after Jean-Louis Koszul. Foucault's pendulum dates from 1851; Berry's geometric phase from 1984.

Recall Where we stand

A connection is a derivative of vector fields with the product rule, given in coordinates by Christoffel symbols; it extends to all tensors, with +Γ+\Gamma for upper and −Γ-\Gamma for lower indices. Christoffel symbols are not a tensor, but the difference of two connections is. A metric determines one connection that is compatible (∇g=0\nabla g = 0) and torsion-free, the Levi-Civita connection, with Γijk=12gkl(∂igjl+∂jgil−∂lgij)\Gamma_{ij}^k = \frac12g^{kl}(\partial_ig_{jl} + \partial_jg_{il} - \partial_lg_{ij}); for a submanifold of RN\mathbb{R}^N it is the tangential part of the ordinary derivative. Parallel transport along curves solves a linear ODE and is an isometry; its holonomy around loops measures the enclosed curvature, as the Foucault pendulum and Gravity Probe B show. 9A.3 Geodesics and the Exponential Map uses the connection to define straight lines: curves whose velocity is parallel.

Exercises

Exercise 2.2 Christoffel symbols are not a tensor

Compute the Christoffel symbols of the Euclidean metric in polar coordinates, dr2+r2dθ2dr^2 + r^2d\theta^2, and confirm Γθθr=−r\Gamma_{\theta\theta}^r = -r and Γrθθ=1r\Gamma_{r\theta}^\theta = \frac1r. Since they vanish in Cartesian coordinates and not in polar ones, explain why they cannot be the components of a tensor.

Solution

grr=1g_{rr} = 1, gθθ=r2g_{\theta\theta} = r^2, and only ∂rgθθ=2r\partial_rg_{\theta\theta} = 2r is nonzero. Γθθr=12grr(−∂rgθθ)=−r\Gamma_{\theta\theta}^r = \frac12g^{rr}(-\partial_rg_{\theta\theta}) = -r, and Γrθθ=12gθθ∂rgθθ=2r2r2=1r\Gamma_{r\theta}^\theta = \frac12g^{\theta\theta}\partial_rg_{\theta\theta} = \frac{2r}{2r^2} = \frac1r. A tensor that vanishes in one coordinate system vanishes in all (its components transform linearly), so the Γ\Gamma cannot be one.

Exercise 2.3 Christoffel symbols of a warped product

For g=dr2+φ(r)2dθ2g = dr^2 + \varphi(r)^2d\theta^2, show that the only nonzero Christoffel symbols are Γθθr=−φφ′\Gamma_{\theta\theta}^r = -\varphi\varphi' and Γrθθ=Γθrθ=φ′φ\Gamma_{r\theta}^\theta = \Gamma_{\theta r}^\theta = \frac{\varphi'}{\varphi}. Specialise to the round sphere, φ=sin⁡r\varphi = \sin r.

Solution

The only nonconstant component is gθθ=φ2g_{\theta\theta} = \varphi^2, with ∂rgθθ=2φφ′\partial_rg_{\theta\theta} = 2\varphi\varphi'. Then Γθθr=−12∂rgθθ=−φφ′\Gamma_{\theta\theta}^r = -\frac12\partial_rg_{\theta\theta} = -\varphi\varphi' and Γrθθ=12φ2⋅2φφ′=φ′φ\Gamma_{r\theta}^\theta = \frac{1}{2\varphi^2}\cdot2\varphi\varphi' = \frac{\varphi'}{\varphi}; every other combination involves a derivative of a constant component. On the sphere: Γθθr=−sin⁡rcos⁡r\Gamma_{\theta\theta}^r = -\sin r\cos r and Γrθθ=cot⁡r\Gamma_{r\theta}^\theta = \cot r.

Exercise 2.4 Holonomy around a circle of latitude

On the unit sphere with g=dr2+sin⁡2r dθ2g = dr^2 + \sin^2r\,d\theta^2, parallel-transport a vector around the circle r=r0r = r_0, parametrised by θ=t∈[0,2π]\theta = t \in [0, 2\pi]. Writing V=a ∂r+c ∂θsin⁡r0V = a\,\partial_r + c\,\frac{\partial_\theta}{\sin r_0} in the orthonormal frame, show that a′=cos⁡r0 ca' = \cos r_0\,c and c′=−cos⁡r0 ac' = -\cos r_0\,a, so VV rotates relative to the frame at the constant rate cos⁡r0\cos r_0. Deduce that after one circuit it has turned by 2πcos⁡r02\pi\cos r_0 relative to the frame, and that this agrees, modulo 2π2\pi, with the area 2π(1−cos⁡r0)2\pi(1 - \cos r_0) of the enclosed cap. With r0=π2−φr_0 = \frac{\pi}{2} - \varphi, recover the Foucault rate 2πsin⁡φ2\pi\sin\varphi per day.

Solution

With V=A ∂r+B ∂θV = A\,\partial_r + B\,\partial_\theta and γ′=∂θ\gamma' = \partial_\theta, DtV=(A′+ΓθθrB)∂r+(B′+ΓθrθA)∂θD_tV = (A' + \Gamma_{\theta\theta}^rB)\partial_r + (B' + \Gamma_{\theta r}^\theta A)\partial_\theta, so A′=sin⁡r0cos⁡r0 BA' = \sin r_0\cos r_0\,B and B′=−cot⁡r0 AB' = -\cot r_0\,A. With a=Aa = A, c=Bsin⁡r0c = B\sin r_0: a′=cos⁡r0 ca' = \cos r_0\,c and c′=−cos⁡r0 ac' = -\cos r_0\,a, a rotation at rate cos⁡r0\cos r_0 (clockwise relative to the frame (∂r,∂θ/sin⁡r0)(\partial_r, \partial_\theta/\sin r_0)). After time 2π2\pi the angle is 2πcos⁡r02\pi\cos r_0 in that sense, and −2πcos⁡r0≡2π−2πcos⁡r0=2π(1−cos⁡r0)-2\pi\cos r_0 \equiv 2\pi - 2\pi\cos r_0 = 2\pi(1 - \cos r_0) modulo 2π2\pi. With cos⁡r0=sin⁡φ\cos r_0 = \sin\varphi: the swing plane turns by 2πsin⁡φ2\pi\sin\varphi relative to the meridian frame, which is the floor's frame, in one day.

Exercise 2.5 The connection of a submanifold

For a submanifold M⊂RNM \subset \mathbb{R}^N with the induced metric, define ∇XY=(DˉXY)⊤\nabla_XY = (\bar D_XY)^\top, where Dˉ\bar D is the ordinary derivative and ⊤\top is orthogonal projection onto TMTM. Show that ∇\nabla is a connection, that it is compatible with the metric, and that it is torsion-free (use DˉXY−DˉYX=[X,Y]\bar D_XY - \bar D_YX = [X, Y], which is tangent to MM). Conclude that it is the Levi-Civita connection.

Solution

The connection axioms are inherited from Dˉ\bar D, since projection is linear over functions. Compatibility: for Y,ZY, Z tangent, X⟨Y,Z⟩=⟨DˉXY,Z⟩+⟨Y,DˉXZ⟩X\langle Y, Z\rangle = \langle\bar D_XY, Z\rangle + \langle Y, \bar D_XZ\rangle, and since ZZ and YY are tangent, only the tangential parts contribute: =⟨∇XY,Z⟩+⟨Y,∇XZ⟩= \langle\nabla_XY, Z\rangle + \langle Y, \nabla_XZ\rangle. Torsion: ∇XY−∇YX=(DˉXY−DˉYX)⊤=[X,Y]⊤=[X,Y]\nabla_XY - \nabla_YX = (\bar D_XY - \bar D_YX)^\top = [X, Y]^\top = [X, Y]. By uniqueness in Theorem 2.1, ∇\nabla is the Levi-Civita connection.

Exercise 2.6 Holonomy of a cone

A flat cone with cone angle α<2π\alpha < 2\pi is made by cutting a sector of angle α\alpha out of the plane and gluing its edges. Explain why parallel transport on the cone is ordinary translation in the sector, and why transport once around the apex rotates vectors by 2π−α2\pi - \alpha. Show that the cone tangent to the unit sphere along the circle of colatitude r0r_0 unrolls to a sector of angle 2πcos⁡r02\pi\cos r_0.

Solution

Away from the apex the cone is flat and the sector is an isometric chart in which the Christoffel symbols vanish, so parallel fields have constant components: translation. Carry a vector around: in the sector it stays fixed, but the two edges are glued by a rotation through α\alpha about the apex, so the vector arrives at the glued edge rotated by α\alpha, that is, by −(2π−α)-(2\pi - \alpha) modulo 2π2\pi, relative to where it would have to be to match. The tangent cone along colatitude r0r_0 has slant distance tan⁡r0\tan r_0 from the circle to the apex and circle circumference 2πsin⁡r02\pi\sin r_0, so its development is a sector of radius tan⁡r0\tan r_0 and angle 2πsin⁡r0tan⁡r0=2πcos⁡r0\frac{2\pi\sin r_0}{\tan r_0} = 2\pi\cos r_0. The holonomy, 2π−2πcos⁡r02\pi - 2\pi\cos r_0, matches Exercise 2.4, because the cone and the sphere have the same tangent planes along the circle, and hence the same parallel transport along it.

Exercise 2.7 The Hessian

Show that ∇i∇jf=∂i∂jf−Γijk∂kf\nabla_i\nabla_jf = \partial_i\partial_jf - \Gamma_{ij}^k\partial_kf, that it is symmetric in i,ji, j, and that at a critical point of ff it equals the ordinary matrix of second partial derivatives in any coordinates.

Solution

∇f=df\nabla f = df has components ∂jf\partial_jf, and the rule for a 1-form gives ∇i(∂jf)=∂i∂jf−Γijk∂kf\nabla_i(\partial_jf) = \partial_i\partial_jf - \Gamma_{ij}^k\partial_kf. Both terms are symmetric in i,ji, j, the second because the Levi-Civita connection is torsion-free. At a critical point, ∂kf=0\partial_kf = 0, so the Γ\Gamma term vanishes.

Exercise 2.8 Rehearsal: the variation of the Christoffel symbols

Let g(t)g(t) be a family of metrics with ∂tgij=hij\partial_tg_{ij} = h_{ij}. Explain why ∂tΓijk\partial_t\Gamma_{ij}^k is a tensor, and show that

∂tΓijk=12gkl(∇ihjl+∇jhil−∇lhij)\partial_t\Gamma_{ij}^k = \tfrac12g^{kl}\big(\nabla_ih_{jl} + \nabla_jh_{il} - \nabla_lh_{ij}\big)

by computing at a point pp in coordinates in which all Γijk(p,t0)=0\Gamma_{ij}^k(p, t_0) = 0 (such coordinates exist by 9A.3 Geodesics and the Exponential Map). Deduce that under the Ricci flow, h=−2Ric⁡h = -2\operatorname{Ric},

∂tΓijk=−gkl(∇iRjl+∇jRil−∇lRij).\partial_t\Gamma_{ij}^k = -g^{kl}\big(\nabla_iR_{jl} + \nabla_jR_{il} - \nabla_lR_{ij}\big).

This is the first evolution equation of 11A.2 How Curvature Evolves.

Solution

∂tΓ=lim⁡s→01s(Γ(t+s)−Γ(t))\partial_t\Gamma = \lim_{s\to0}\frac1s(\Gamma(t + s) - \Gamma(t)) is a limit of differences of connections, which are tensors, so it is a tensor. Differentiate Γijk=12gkl(∂igjl+∂jgil−∂lgij)\Gamma_{ij}^k = \frac12g^{kl}(\partial_ig_{jl} + \partial_jg_{il} - \partial_lg_{ij}) in tt: the term with ∂tgkl\partial_tg^{kl} multiplies Γ\Gamma, which vanishes at pp, leaving 12gkl(∂ihjl+∂jhil−∂lhij)\frac12g^{kl}(\partial_ih_{jl} + \partial_jh_{il} - \partial_lh_{ij}). At pp, where Γ=0\Gamma = 0, partial and covariant derivatives agree, so this is 12gkl(∇ihjl+∇jhil−∇lhij)\frac12g^{kl}(\nabla_ih_{jl} + \nabla_jh_{il} - \nabla_lh_{ij}). Both sides are tensors that agree at pp in one coordinate system, so they agree everywhere. Substituting h=−2Ric⁡h = -2\operatorname{Ric} gives the stated formula.

© 2026 NeckPinch (www.neckpinch.com). All content in the guidebook (text, mathematics, figures and exercises) is protected by copyright. All rights reserved. No part may be copied, republished or redistributed without written permission.