Affine Connection

An affine connection on a smooth manifold MMM is defined as a map ∇:X(M)×X(M)→X(M)\nabla: \mathcal{X}(M) \times \mathcal{X}(M) \to \mathcal{X}(M)∇:X(M)×X(M)→X(M), where X(M)\mathcal{X}(M)X(M) denotes the space of smooth vector fields on MMM, that is R\mathbb{R}R-bilinear and satisfies the Leibniz rule: for smooth vector fields X,Y∈X(M)X, Y \in \mathcal{X}(M)X,Y∈X(M) and smooth functions f∈C∞(M)f \in C^\infty(M)f∈C∞(M),

This structure allows for the extension of differentiation from coordinate vector fields to arbitrary vector fields, providing a way to measure how vector fields change across the manifold.[14]

In local coordinates (x1,…,xn)(x^1, \dots, x^n)(x1,…,xn) on MMM, an affine connection is specified by its action on the coordinate basis vector fields ∂i=∂/∂xi\partial_i = \partial / \partial x^i∂i​=∂/∂xi:

where Γijk\Gamma^k_{ij}Γijk​ are smooth functions known as the Christoffel symbols (or connection coefficients) of the second kind, forming the local components of the connection. These symbols encode the connection's behavior in the given coordinate system and can be used to define covariant derivatives on tensor fields.[14]

Under a change of coordinates from xxx to x′x'x′, where x′l=x′l(x1,…,xn)x'^l = x'^l(x^1, \dots, x^n)x′l=x′l(x1,…,xn), the Christoffel symbols transform according to the non-tensorial law

with the Jacobian matrices ∂x/∂x′\partial x / \partial x'∂x/∂x′ and ∂x′/∂x\partial x' / \partial x∂x′/∂x being inverses of each other; the presence of the second-derivative term demonstrates that the Γijk\Gamma^k_{ij}Γijk​ do not transform as components of a tensor field.[14][15]

A simple example is the trivial flat connection on Euclidean space Rn\mathbb{R}^nRn equipped with the standard coordinates, where all Christoffel symbols vanish, Γijk=0\Gamma^k_{ij} = 0Γijk​=0, corresponding to the usual flat derivative with no curvature or torsion.[14]

Affine connections on a given manifold are not unique: if ∇\nabla∇ and ∇\tilde{\nabla}∇ are two connections, their difference ∇−∇\nabla - \tilde{\nabla}∇−∇ is a tensor field of type (1,2), so the space of all affine connections forms an affine space modeled on the vector space of smooth (1,2)-tensors on MMM. Torsion-freeness, where ∇XY−∇YX=[X,Y]\nabla_X Y - \nabla_Y X = [X, Y]∇X​Y−∇Y​X=[X,Y] for all vector fields X,YX, YX,Y, is an optional property that some connections satisfy but is not required in the general definition.[14]

Covariant Derivative

The covariant derivative induced by an affine connection on a smooth manifold enables the differentiation of sections of tensor bundles in a manner that is compatible with the manifold's geometry, extending the ordinary directional derivative to account for the variation of bases. For vector fields XXX and YYY on the manifold MMM, the covariant derivative ∇XY\nabla_X Y∇X​Y is a vector field that measures the rate of change of YYY along XXX, incorporating connection coefficients to adjust for curvature in coordinate charts. This operator satisfies linearity in both arguments, ∇fX+gYZ=f∇XZ+g∇YZ\nabla_{fX + gY} Z = f \nabla_X Z + g \nabla_Y Z∇fX+gY​Z=f∇X​Z+g∇Y​Z and ∇X(fY)=(Xf)Y+f∇XY\nabla_X (fY) = (X f) Y + f \nabla_X Y∇X​(fY)=(Xf)Y+f∇X​Y for smooth functions f,gf, gf,g, ensuring it behaves like a derivation.[16]

To extend the covariant derivative to general tensor fields, consider a (k,l)(k, l)(k,l)-tensor field TTT on MMM. For a vector field XXX, the covariant derivative ∇XT\nabla_X T∇X​T is defined componentwise in local coordinates (xi)(x^i)(xi) by

where Γijk\Gamma^k_{ij}Γijk​ are the Christoffel symbols of the connection, with positive terms for contravariant indices and negative for covariant ones; the full expression for arbitrary XXX follows by linearity. This extension preserves the tensorial nature, meaning ∇XT\nabla_X T∇X​T is again a (k,l)(k, l)(k,l)-tensor field. Additionally, the operator obeys the Leibniz rule for tensor products: ∇X(T⊗S)=(∇XT)⊗S+T⊗(∇XS)\nabla_X (T \otimes S) = (\nabla_X T) \otimes S + T \otimes (\nabla_X S)∇X​(T⊗S)=(∇X​T)⊗S+T⊗(∇X​S), which facilitates its application to derived constructions like differential forms.[17][18]

A key application of the covariant derivative defines parallel transport along curves. Given a smooth curve γ:I→M\gamma: I \to Mγ:I→M with tangent vector field γ′\gamma'γ′, a vector field VVV along γ\gammaγ is parallel if ∇γ′V=0\nabla_{\gamma'} V = 0∇γ′​V=0 everywhere, meaning VVV does not change relative to the connection when transported along γ\gammaγ; this condition integrates to yield unique parallel vector fields given an initial value at γ(0)\gamma(0)γ(0). For scalar functions fff on MMM, the covariant derivative simplifies to the directional derivative: ∇Xf=X(f)\nabla_X f = X(f)∇X​f=X(f). In the specific case of vector fields, the local expression is ∇XY=X(Yi)∂∂xi+YjXkΓjki∂∂xi\nabla_X Y = X(Y^i) \frac{\partial}{\partial x^i} + Y^j X^k \Gamma^i_{jk} \frac{\partial}{\partial x^i}∇X​Y=X(Yi)∂xi∂​+YjXkΓjki​∂xi∂​, highlighting how the connection corrects the naive pushforward. The covariant derivative also interacts with the Lie bracket of vector fields: ∇XY−∇YX−[X,Y]\nabla_X Y - \nabla_Y X - [X, Y]∇X​Y−∇Y​X−[X,Y] captures the antisymmetric part of the connection, quantifying its deviation from being torsion-free.[19][20]

Torsion Tensor

The torsion tensor TTT of an affine connection ∇\nabla∇ on a smooth manifold is defined by

for vector fields X,YX, YX,Y, where [X,Y][X, Y][X,Y] denotes the Lie bracket.[21] This defines TTT as a tensor of type (1,2), measuring the antisymmetric part of the connection.[22]

In local coordinates, the components of the torsion tensor are given by

where Γijk\Gamma^k_{ij}Γijk​ are the Christoffel symbols of the connection.[23] The torsion tensor is alternating in its lower indices, satisfying Tijk=−TjikT^k_{ij} = -T^k_{ji}Tijk​=−Tjik​, and thus vanishes identically if and only if the connection is torsion-free (i.e., the Christoffel symbols are symmetric in the lower indices).[21] Torsion-free connections are standard in classical Riemannian geometry, where they ensure compatibility with the Lie bracket in the covariant derivative.[24]

Geometrically, the torsion tensor quantifies the "twist" or failure of parallel transport to preserve the Lie bracket, leading to non-closure of infinitesimal parallelograms in the tangent space.[25] This twist is central in teleparallel gravity theories, where curvature vanishes but torsion encodes gravitational effects equivalent to general relativity.[26]

In Cartan geometry, a non-zero torsion tensor models spacetimes incorporating intrinsic spin, such as in Einstein-Cartan theory, where torsion couples to the spin density of fermionic matter.[23]

The torsion tensor admits an irreducible decomposition under the Lorentz group in four dimensions: a trace-free (tensor) part with 16 components, a trace vector part with 4 components, and an axial part with 4 components.[24] For metric-compatible connections, the torsion relates to the contorsion tensor KKK, which decomposes the full connection as Γijk=Γijk+Kijk\Gamma^k_{ij} = \tilde{\Gamma}^k_{ij} + K^k_{ij}Γijk​=Γijk​+Kijk​, where Γ~\tilde{\Gamma}Γ~ is the torsion-free Levi-Civita connection and the lowered contorsion is Kkij=12(Tkij+Tjki−Tkij)K_{kij} = \frac{1}{2} (T_{kij} + T_{jki} - T_{kij})Kkij​=21​(Tkij​+Tjki​−Tkij​), with Tkij=gklTijlT_{kij} = g_{kl} T^l_{ij}Tkij​=gkl​Tijl​.

Curvature Tensor

The curvature tensor of an affine connection ∇\nabla∇ on a smooth manifold is a tensor field of type (1,3), defined for vector fields X,Y,ZX, Y, ZX,Y,Z by

This expression quantifies the extent to which the second covariant derivatives fail to commute, adjusted for the non-tensorial nature of the Lie bracket term. In the presence of torsion, the Lie bracket accounts for first-order antisymmetry, leading to modified curvature formulas that incorporate the torsion tensor.[27][28]

In local coordinates, the components of the curvature tensor are given by

where Γμνρ\Gamma^\rho_{\mu\nu}Γμνρ​ are the Christoffel symbols of the connection and summation over repeated indices is implied. The curvature tensor exhibits several algebraic symmetries: it is antisymmetric in the last two indices, R σμνρ=−R σνμρR^\rho_{\ \sigma\mu\nu} = -R^\rho_{\ \sigma\nu\mu}R σμνρ​=−R σνμρ​, reflecting the skew-symmetry R(X,Y)=−R(Y,X)R(X, Y) = -R(Y, X)R(X,Y)=−R(Y,X); additionally, the first Bianchi identity states that the cyclic sum vanishes in the torsion-free case, R(X,Y)Z+R(Y,Z)X+R(Z,X)Y=0R(X, Y)Z + R(Y, Z)X + R(Z, X)Y = 0R(X,Y)Z+R(Y,Z)X+R(Z,X)Y=0.[28][29]

The second Bianchi identity provides a differential constraint on the curvature: for a torsion-free connection, the covariant exterior derivative satisfies d∇R=0d^\nabla R = 0d∇R=0, or in components,

Geometrically, the curvature tensor measures the obstruction to the local flatness of the connection, capturing how parallel transport around an infinitesimal closed loop deviates from the identity map on tangent spaces. It assesses the integrability of parallel distributions, where vanishing curvature restricted to a ∇\nabla∇-invariant subbundle ensures integrability by a generalized Frobenius theorem. In particular, if the curvature tensor vanishes identically, the connection is locally flat, implying the existence of coordinates in which the Christoffel symbols are zero and the manifold is locally Euclidean. Furthermore, the curvature induces a quadratic form on the space of bivectors, generalizing the notion of sectional curvature in metric settings to quantify bending in arbitrary directions.[29][30][27]

Levi-Civita Connection

The Levi-Civita connection on a Riemannian manifold (M,g)(M, g)(M,g) is the unique affine connection ∇\nabla∇ that is both torsion-free and compatible with the metric ggg. Torsion-freeness means that the torsion tensor vanishes, T(∇)(X,Y)=∇XY−∇YX−[X,Y]=0T(\nabla)(X, Y) = \nabla_X Y - \nabla_Y X - [X, Y] = 0T(∇)(X,Y)=∇X​Y−∇Y​X−[X,Y]=0 for all vector fields X,Y∈Γ(TM)X, Y \in \Gamma(TM)X,Y∈Γ(TM). Metric compatibility requires that the covariant derivative of the metric tensor is zero, ∇g=0\nabla g = 0∇g=0, or equivalently, X(g(Y,Z))=g(∇XY,Z)+g(Y,∇XZ)X(g(Y, Z)) = g(\nabla_X Y, Z) + g(Y, \nabla_X Z)X(g(Y,Z))=g(∇X​Y,Z)+g(Y,∇X​Z) for all X,Y,Z∈Γ(TM)X, Y, Z \in \Gamma(TM)X,Y,Z∈Γ(TM).[31][32]

This connection is constructed explicitly using the Koszul formula, which determines ∇XY\nabla_X Y∇X​Y via its inner product with an arbitrary vector field ZZZ:

The formula arises from applying the metric compatibility condition to the pairs (Y,Z)(Y, Z)(Y,Z), (Z,X)(Z, X)(Z,X), and (X,Y)(X, Y)(X,Y), then combining with torsion-freeness to account for Lie brackets. It ensures that ∇XY\nabla_X Y∇X​Y is uniquely determined pointwise by the metric and its first derivatives.[33][32]

In local coordinates (xi)(x^i)(xi) on MMM, the Levi-Civita connection is specified by its Christoffel symbols of the second kind Γijk\Gamma^k_{ij}Γijk​, defined by ∇∂i∂j=Γijk∂k\nabla_{\partial_i} \partial_j = \Gamma^k_{ij} \partial_k∇∂i​​∂j​=Γijk​∂k​. These are given by

where gij=g(∂i,∂j)g_{ij} = g(\partial_i, \partial_j)gij​=g(∂i​,∂j​) are the components of the metric tensor and gklg^{kl}gkl are those of its inverse. The symmetry Γijk=Γjik\Gamma^k_{ij} = \Gamma^k_{ji}Γijk​=Γjik​ follows directly from torsion-freeness. This formula is obtained by substituting coordinate vector fields into the Koszul formula and evaluating at basis vectors.[31][33]

The uniqueness of the Levi-Civita connection is established by the fundamental theorem of Riemannian geometry. Suppose ∇\nabla∇ is any torsion-free, metric-compatible connection on (M,g)(M, g)(M,g). Applying the compatibility condition to the pairs (Y,Z)(Y, Z)(Y,Z), (Z,X)(Z, X)(Z,X), and (X,Y)(X, Y)(X,Y) yields three equations:

Adding the first two and subtracting the third, then using torsion-freeness to replace ∇YX=∇XY+[X,Y]\nabla_Y X = \nabla_X Y + [X, Y]∇Y​X=∇X​Y+[X,Y] and similarly for other terms, reduces to the Koszul formula. Since the right-hand side depends only on ggg and its derivatives, ∇XY\nabla_X Y∇X​Y is uniquely fixed.[33][32]

For a concrete example, consider the unit 2-sphere S2⊂R3S^2 \subset \mathbb{R}^3S2⊂R3 equipped with the induced metric g=dθ2+sin⁡2θ dϕ2g = d\theta^2 + \sin^2 \theta , d\phi^2g=dθ2+sin2θdϕ2 in spherical coordinates (θ,ϕ)(\theta, \phi)(θ,ϕ), where θ∈(0,π)\theta \in (0, \pi)θ∈(0,π) is the colatitude and ϕ∈[0,2π)\phi \in [0, 2\pi)ϕ∈[0,2π) is the longitude. The non-vanishing Christoffel symbols are Γϕϕθ=−sin⁡θcos⁡θ\Gamma^\theta_{\phi\phi} = -\sin\theta \cos\thetaΓϕϕθ​=−sinθcosθ and Γθϕϕ=Γϕθϕ=cot⁡θ\Gamma^\phi_{\theta\phi} = \Gamma^\phi_{\phi\theta} = \cot\thetaΓθϕϕ​=Γϕθϕ​=cotθ. These symbols arise from substituting the metric components into the general formula for Γijk\Gamma^k_{ij}Γijk​. The associated geodesic equation ∇γ˙γ˙=0\nabla_{\dot\gamma} \dot\gamma = 0∇γ˙​​γ˙​=0 has solutions that are great circles, the shortest paths on S2S^2S2, confirming the connection's role in defining intrinsic geometry.[31][34]

More generally, any metric-compatible connection ∇\nabla∇ (not necessarily torsion-free) on (M,g)(M, g)(M,g) relates to the Levi-Civita connection ∇\tilde{\nabla}∇ via the contorsion tensor KKK, defined such that ∇XY=∇XY+K(X,Y)\nabla_X Y = \tilde{\nabla}X Y + K(X, Y)∇X​Y=∇X​Y+K(X,Y). In components, this is Γijh=Γijh+Kijh\Gamma^h{ij} = \tilde{\Gamma}^h_{ij} + K^h_{ij}Γijh​=Γijh​+Kijh​, where the contorsion is expressed in terms of the torsion tensor TTT of ∇\nabla∇. This decomposition allows general metric connections to be viewed as perturbations of the Levi-Civita connection by a torsion-induced adjustment.

Flat and Projective Connections

A flat connection on a smooth manifold is an affine connection whose curvature tensor vanishes identically, denoted R=0R = 0R=0 everywhere. This condition implies that the connection admits a local trivialization of the associated fiber bundle, where it reduces to the standard flat connection on a product bundle, allowing coordinate charts in which the Christoffel symbols Γijk=0\Gamma^k_{ij} = 0Γijk​=0.[35] Such trivializations enable parallel transport to act as local isomorphisms between fibers, preserving the structure without holonomy obstructions in simply connected regions.[36]

In the context of Ehresmann connections, the vanishing curvature ensures that the horizontal distribution is integrable by the Frobenius theorem, integrating locally to a foliation that yields trivial bundle structures over coordinate neighborhoods.[37] For surfaces, a flat affine connection implies the surface is developable, meaning it can be locally isometrically mapped to the plane without distortion, as the intrinsic geometry aligns with Euclidean flatness.[38]

A classic example is the standard affine flat connection on Rn\mathbb{R}^nRn, where Γ=0\Gamma = 0Γ=0, making straight lines the geodesics and endowing the space with the canonical flat structure.[35]

Weyl connections extend metric-compatible connections to conformal manifolds by allowing a scaling factor in the metric preservation. Specifically, on a manifold equipped with a conformal class of metrics $$, a Weyl connection ∇\nabla∇ satisfies ∇Xg(Y,Z)=Q(X,Y,Z)=−2ω(X)g(Y,Z)\nabla_X g(Y, Z) = Q(X, Y, Z) = -2 \omega(X) g(Y, Z)∇X​g(Y,Z)=Q(X,Y,Z)=−2ω(X)g(Y,Z) for some 1-form ω\omegaω, introducing non-metricity that scales lengths under parallel transport while preserving angles.[39]

Projective connections generalize affine connections by focusing on the projective structure they induce, defined as an equivalence class of torsion-free affine connections that share the same unparametrized geodesics—curves determined up to reparametrization. This equivalence relation groups connections ∇\nabla∇ and ∇^\hat{\nabla}∇^ if ∇^XY=∇XY+α(X)Y+β(Y)X\hat{\nabla}_X Y = \nabla_X Y + \alpha(X) Y + \beta(Y) X∇^X​Y=∇X​Y+α(X)Y+β(Y)X for 1-forms α,β\alpha, \betaα,β, ensuring the geodesic paths coincide projectively.[40] Projective connections preserve this unparametrized path structure, making them invariant under projective transformations.

An illustrative example is the projective structure arising in Klein geometries, where a Cartan connection models the manifold on the projective space RPn\mathbb{RP}^nRPn with the projective linear group acting transitively, capturing the full projective invariance without a preferred metric.[41]

Beyond flat cases, projective connections appear in non-flat settings such as conformal geometry, where they encode projective invariants on manifolds with Weyl structures, allowing non-trivial holonomy while maintaining geodesic projectivity.[42]

Parallel Transport

Parallel transport along a smooth curve γ:I→M\gamma: I \to Mγ:I→M on a manifold MMM equipped with an affine connection ∇\nabla∇ is defined for a vector field VVV along γ\gammaγ that satisfies ∇γ˙(t)V=0\nabla_{\dot{\gamma}(t)} V = 0∇γ˙​(t)​V=0 for all t∈It \in It∈I.[43] In local coordinates, this condition translates to the ordinary differential equation V′(t)+Γ(V(t),γ˙(t))=0V'(t) + \Gamma(V(t), \dot{\gamma}(t)) = 0V′(t)+Γ(V(t),γ˙​(t))=0, where Γ\GammaΓ denotes the Christoffel symbols of the connection.[44]

The existence and uniqueness of such a parallel vector field VVV along γ\gammaγ, given an initial value V(t0)V(t_0)V(t0​), follow from the Picard-Lindelöf theorem, as the Christoffel symbols Γ\GammaΓ are locally Lipschitz continuous on smooth manifolds.[44] This parallel transport map provides a linear isomorphism between the fibers of the pullback bundle γ∗TM\gamma^* TMγ∗TM over the endpoints γ(t0)\gamma(t_0)γ(t0​) and γ(t1)\gamma(t_1)γ(t1​), preserving the vector bundle structure.[45]

For a closed curve γ\gammaγ based at a point p∈Mp \in Mp∈M, the composition of parallel transports around the loop yields an element of the holonomy group Holp(∇)\mathrm{Hol}_p(\nabla)Holp​(∇), which is a Lie subgroup of GL(TpM)≅GL(n,R)\mathrm{GL}(T_p M) \cong \mathrm{GL}(n, \mathbb{R})GL(Tp​M)≅GL(n,R) and encodes the global obstruction to path independence caused by curvature.[18] The holonomy group is influenced by the curvature tensor, as infinitesimal holonomies around small loops approximate the curvature operator.[18]

If the curvature tensor R=0R = 0R=0, the connection is flat, and parallel transport depends only on the endpoints of the curve, independent of the path taken. A classic example occurs on the unit sphere S2S^2S2 with the Levi-Civita connection: parallel transporting a tangent vector around the boundary of a geodesic triangle results in a rotation by the angular excess of the triangle, directly linking to the Gauss-Bonnet theorem.[46]

Geodesics and Riemannian Geometry

In the context of an affine connection on a manifold, geodesics are defined as the auto-parallel curves, meaning curves whose tangent vectors are parallel transported along themselves. This leads to the geodesic equation, which states that for a curve γ:I→M\gamma: I \to Mγ:I→M with velocity γ′\gamma'γ′, the covariant derivative satisfies ∇γ′γ′=0\nabla_{\gamma'} \gamma' = 0∇γ′​γ′=0. In local coordinates xkx^kxk, this equation takes the form d2xkdt2+Γijkdxidtdxjdt=0\frac{d^2 x^k}{dt^2} + \Gamma^k_{ij} \frac{dx^i}{dt} \frac{dx^j}{dt} = 0dt2d2xk​+Γijk​dtdxi​dtdxj​=0, where Γijk\Gamma^k_{ij}Γijk​ are the Christoffel symbols of the connection.

In Riemannian geometry, the Levi-Civita connection, being torsion-free and metric-compatible, defines geodesics that locally minimize the length functional, serving as the shortest paths between points on the manifold. These geodesics parameterize the exponential map exp⁡p:TpM→M\exp_p: T_p M \to Mexpp​:Tp​M→M, where for a vector v∈TpMv \in T_p Mv∈Tp​M, exp⁡p(v)=γ(1)\exp_p(v) = \gamma(1)expp​(v)=γ(1) for the unique geodesic γ\gammaγ satisfying γ(0)=p\gamma(0) = pγ(0)=p and γ′(0)=v\gamma'(0) = vγ′(0)=v, assuming the parameter is affine. This map provides a local diffeomorphism near the zero section, facilitating the study of the manifold's geometry around point ppp.

To analyze the stability and focusing of geodesics, Jacobi fields arise as the first variation fields along geodesic variations. For a Jacobi field JJJ along a geodesic γ\gammaγ, it satisfies the second-order equation ∇2J+R(J,γ′)γ′=0\nabla^2 J + R(J, \gamma') \gamma' = 0∇2J+R(J,γ′)γ′=0, where RRR is the Riemann curvature tensor and ∇2J=∇γ′∇γ′J−∇∇γ′Jγ′\nabla^2 J = \nabla_{\gamma'} \nabla_{\gamma'} J - \nabla_{\nabla_{\gamma'} J} \gamma'∇2J=∇γ′​∇γ′​J−∇∇γ′​J​γ′. Zeros of nontrivial Jacobi fields indicate conjugate points, where nearby geodesics intersect, signaling the boundary of the injectivity radius.

The curvature influences geodesic behavior profoundly, with applications extending to topology and physics. The Gauss-Bonnet theorem relates the integral of the Gaussian curvature over a surface to its Euler characteristic, incorporating geodesic curvature along boundaries to link local geometry to global topology. In general relativity, the Einstein field equations Gμν=8πGc4TμνG_{\mu\nu} = \frac{8\pi G}{c^4} T_{\mu\nu}Gμν​=c48πG​Tμν​ employ the Ricci tensor, the contraction of the Riemann tensor, to describe how matter curves spacetime, with geodesics tracing free-fall trajectories.

A representative example occurs in hyperbolic space Hn\mathbb{H}^nHn, where the constant negative sectional curvature causes nearby geodesics to diverge exponentially: the separation grows proportionally to et−Ke^{t \sqrt{-K}}et−K​ for large ttt, with K<0K < 0K<0 the curvature, contrasting with the convergence in spaces of positive curvature.

Geodesic completeness, the property that all geodesics can be extended indefinitely, is equivalent via the Hopf-Rinow theorem to metric completeness and compactness of closed balls in Riemannian manifolds, ensuring the existence of minimizing geodesics between any two points.

Affine Connection

An affine connection on a smooth manifold MMM is defined as a map ∇:X(M)×X(M)→X(M)\nabla: \mathcal{X}(M) \times \mathcal{X}(M) \to \mathcal{X}(M)∇:X(M)×X(M)→X(M), where X(M)\mathcal{X}(M)X(M) denotes the space of smooth vector fields on MMM, that is R\mathbb{R}R-bilinear and satisfies the Leibniz rule: for smooth vector fields X,Y∈X(M)X, Y \in \mathcal{X}(M)X,Y∈X(M) and smooth functions f∈C∞(M)f \in C^\infty(M)f∈C∞(M),

This structure allows for the extension of differentiation from coordinate vector fields to arbitrary vector fields, providing a way to measure how vector fields change across the manifold.[14]

In local coordinates (x1,…,xn)(x^1, \dots, x^n)(x1,…,xn) on MMM, an affine connection is specified by its action on the coordinate basis vector fields ∂i=∂/∂xi\partial_i = \partial / \partial x^i∂i​=∂/∂xi:

where Γijk\Gamma^k_{ij}Γijk​ are smooth functions known as the Christoffel symbols (or connection coefficients) of the second kind, forming the local components of the connection. These symbols encode the connection's behavior in the given coordinate system and can be used to define covariant derivatives on tensor fields.[14]

Under a change of coordinates from xxx to x′x'x′, where x′l=x′l(x1,…,xn)x'^l = x'^l(x^1, \dots, x^n)x′l=x′l(x1,…,xn), the Christoffel symbols transform according to the non-tensorial law

with the Jacobian matrices ∂x/∂x′\partial x / \partial x'∂x/∂x′ and ∂x′/∂x\partial x' / \partial x∂x′/∂x being inverses of each other; the presence of the second-derivative term demonstrates that the Γijk\Gamma^k_{ij}Γijk​ do not transform as components of a tensor field.[14][15]

A simple example is the trivial flat connection on Euclidean space Rn\mathbb{R}^nRn equipped with the standard coordinates, where all Christoffel symbols vanish, Γijk=0\Gamma^k_{ij} = 0Γijk​=0, corresponding to the usual flat derivative with no curvature or torsion.[14]

Affine connections on a given manifold are not unique: if ∇\nabla∇ and ∇\tilde{\nabla}∇ are two connections, their difference ∇−∇\nabla - \tilde{\nabla}∇−∇ is a tensor field of type (1,2), so the space of all affine connections forms an affine space modeled on the vector space of smooth (1,2)-tensors on MMM. Torsion-freeness, where ∇XY−∇YX=[X,Y]\nabla_X Y - \nabla_Y X = [X, Y]∇X​Y−∇Y​X=[X,Y] for all vector fields X,YX, YX,Y, is an optional property that some connections satisfy but is not required in the general definition.[14]

Covariant Derivative

The covariant derivative induced by an affine connection on a smooth manifold enables the differentiation of sections of tensor bundles in a manner that is compatible with the manifold's geometry, extending the ordinary directional derivative to account for the variation of bases. For vector fields XXX and YYY on the manifold MMM, the covariant derivative ∇XY\nabla_X Y∇X​Y is a vector field that measures the rate of change of YYY along XXX, incorporating connection coefficients to adjust for curvature in coordinate charts. This operator satisfies linearity in both arguments, ∇fX+gYZ=f∇XZ+g∇YZ\nabla_{fX + gY} Z = f \nabla_X Z + g \nabla_Y Z∇fX+gY​Z=f∇X​Z+g∇Y​Z and ∇X(fY)=(Xf)Y+f∇XY\nabla_X (fY) = (X f) Y + f \nabla_X Y∇X​(fY)=(Xf)Y+f∇X​Y for smooth functions f,gf, gf,g, ensuring it behaves like a derivation.[16]

To extend the covariant derivative to general tensor fields, consider a (k,l)(k, l)(k,l)-tensor field TTT on MMM. For a vector field XXX, the covariant derivative ∇XT\nabla_X T∇X​T is defined componentwise in local coordinates (xi)(x^i)(xi) by

where Γijk\Gamma^k_{ij}Γijk​ are the Christoffel symbols of the connection, with positive terms for contravariant indices and negative for covariant ones; the full expression for arbitrary XXX follows by linearity. This extension preserves the tensorial nature, meaning ∇XT\nabla_X T∇X​T is again a (k,l)(k, l)(k,l)-tensor field. Additionally, the operator obeys the Leibniz rule for tensor products: ∇X(T⊗S)=(∇XT)⊗S+T⊗(∇XS)\nabla_X (T \otimes S) = (\nabla_X T) \otimes S + T \otimes (\nabla_X S)∇X​(T⊗S)=(∇X​T)⊗S+T⊗(∇X​S), which facilitates its application to derived constructions like differential forms.[17][18]

A key application of the covariant derivative defines parallel transport along curves. Given a smooth curve γ:I→M\gamma: I \to Mγ:I→M with tangent vector field γ′\gamma'γ′, a vector field VVV along γ\gammaγ is parallel if ∇γ′V=0\nabla_{\gamma'} V = 0∇γ′​V=0 everywhere, meaning VVV does not change relative to the connection when transported along γ\gammaγ; this condition integrates to yield unique parallel vector fields given an initial value at γ(0)\gamma(0)γ(0). For scalar functions fff on MMM, the covariant derivative simplifies to the directional derivative: ∇Xf=X(f)\nabla_X f = X(f)∇X​f=X(f). In the specific case of vector fields, the local expression is ∇XY=X(Yi)∂∂xi+YjXkΓjki∂∂xi\nabla_X Y = X(Y^i) \frac{\partial}{\partial x^i} + Y^j X^k \Gamma^i_{jk} \frac{\partial}{\partial x^i}∇X​Y=X(Yi)∂xi∂​+YjXkΓjki​∂xi∂​, highlighting how the connection corrects the naive pushforward. The covariant derivative also interacts with the Lie bracket of vector fields: ∇XY−∇YX−[X,Y]\nabla_X Y - \nabla_Y X - [X, Y]∇X​Y−∇Y​X−[X,Y] captures the antisymmetric part of the connection, quantifying its deviation from being torsion-free.[19][20]

Torsion Tensor

The torsion tensor TTT of an affine connection ∇\nabla∇ on a smooth manifold is defined by

for vector fields X,YX, YX,Y, where [X,Y][X, Y][X,Y] denotes the Lie bracket.[21] This defines TTT as a tensor of type (1,2), measuring the antisymmetric part of the connection.[22]

In local coordinates, the components of the torsion tensor are given by

where Γijk\Gamma^k_{ij}Γijk​ are the Christoffel symbols of the connection.[23] The torsion tensor is alternating in its lower indices, satisfying Tijk=−TjikT^k_{ij} = -T^k_{ji}Tijk​=−Tjik​, and thus vanishes identically if and only if the connection is torsion-free (i.e., the Christoffel symbols are symmetric in the lower indices).[21] Torsion-free connections are standard in classical Riemannian geometry, where they ensure compatibility with the Lie bracket in the covariant derivative.[24]

Geometrically, the torsion tensor quantifies the "twist" or failure of parallel transport to preserve the Lie bracket, leading to non-closure of infinitesimal parallelograms in the tangent space.[25] This twist is central in teleparallel gravity theories, where curvature vanishes but torsion encodes gravitational effects equivalent to general relativity.[26]

In Cartan geometry, a non-zero torsion tensor models spacetimes incorporating intrinsic spin, such as in Einstein-Cartan theory, where torsion couples to the spin density of fermionic matter.[23]

The torsion tensor admits an irreducible decomposition under the Lorentz group in four dimensions: a trace-free (tensor) part with 16 components, a trace vector part with 4 components, and an axial part with 4 components.[24] For metric-compatible connections, the torsion relates to the contorsion tensor KKK, which decomposes the full connection as Γijk=Γijk+Kijk\Gamma^k_{ij} = \tilde{\Gamma}^k_{ij} + K^k_{ij}Γijk​=Γijk​+Kijk​, where Γ~\tilde{\Gamma}Γ~ is the torsion-free Levi-Civita connection and the lowered contorsion is Kkij=12(Tkij+Tjki−Tkij)K_{kij} = \frac{1}{2} (T_{kij} + T_{jki} - T_{kij})Kkij​=21​(Tkij​+Tjki​−Tkij​), with Tkij=gklTijlT_{kij} = g_{kl} T^l_{ij}Tkij​=gkl​Tijl​.

Curvature Tensor

The curvature tensor of an affine connection ∇\nabla∇ on a smooth manifold is a tensor field of type (1,3), defined for vector fields X,Y,ZX, Y, ZX,Y,Z by

This expression quantifies the extent to which the second covariant derivatives fail to commute, adjusted for the non-tensorial nature of the Lie bracket term. In the presence of torsion, the Lie bracket accounts for first-order antisymmetry, leading to modified curvature formulas that incorporate the torsion tensor.[27][28]

In local coordinates, the components of the curvature tensor are given by

where Γμνρ\Gamma^\rho_{\mu\nu}Γμνρ​ are the Christoffel symbols of the connection and summation over repeated indices is implied. The curvature tensor exhibits several algebraic symmetries: it is antisymmetric in the last two indices, R σμνρ=−R σνμρR^\rho_{\ \sigma\mu\nu} = -R^\rho_{\ \sigma\nu\mu}R σμνρ​=−R σνμρ​, reflecting the skew-symmetry R(X,Y)=−R(Y,X)R(X, Y) = -R(Y, X)R(X,Y)=−R(Y,X); additionally, the first Bianchi identity states that the cyclic sum vanishes in the torsion-free case, R(X,Y)Z+R(Y,Z)X+R(Z,X)Y=0R(X, Y)Z + R(Y, Z)X + R(Z, X)Y = 0R(X,Y)Z+R(Y,Z)X+R(Z,X)Y=0.[28][29]

The second Bianchi identity provides a differential constraint on the curvature: for a torsion-free connection, the covariant exterior derivative satisfies d∇R=0d^\nabla R = 0d∇R=0, or in components,

Geometrically, the curvature tensor measures the obstruction to the local flatness of the connection, capturing how parallel transport around an infinitesimal closed loop deviates from the identity map on tangent spaces. It assesses the integrability of parallel distributions, where vanishing curvature restricted to a ∇\nabla∇-invariant subbundle ensures integrability by a generalized Frobenius theorem. In particular, if the curvature tensor vanishes identically, the connection is locally flat, implying the existence of coordinates in which the Christoffel symbols are zero and the manifold is locally Euclidean. Furthermore, the curvature induces a quadratic form on the space of bivectors, generalizing the notion of sectional curvature in metric settings to quantify bending in arbitrary directions.[29][30][27]

Levi-Civita Connection

The Levi-Civita connection on a Riemannian manifold (M,g)(M, g)(M,g) is the unique affine connection ∇\nabla∇ that is both torsion-free and compatible with the metric ggg. Torsion-freeness means that the torsion tensor vanishes, T(∇)(X,Y)=∇XY−∇YX−[X,Y]=0T(\nabla)(X, Y) = \nabla_X Y - \nabla_Y X - [X, Y] = 0T(∇)(X,Y)=∇X​Y−∇Y​X−[X,Y]=0 for all vector fields X,Y∈Γ(TM)X, Y \in \Gamma(TM)X,Y∈Γ(TM). Metric compatibility requires that the covariant derivative of the metric tensor is zero, ∇g=0\nabla g = 0∇g=0, or equivalently, X(g(Y,Z))=g(∇XY,Z)+g(Y,∇XZ)X(g(Y, Z)) = g(\nabla_X Y, Z) + g(Y, \nabla_X Z)X(g(Y,Z))=g(∇X​Y,Z)+g(Y,∇X​Z) for all X,Y,Z∈Γ(TM)X, Y, Z \in \Gamma(TM)X,Y,Z∈Γ(TM).[31][32]

This connection is constructed explicitly using the Koszul formula, which determines ∇XY\nabla_X Y∇X​Y via its inner product with an arbitrary vector field ZZZ:

The formula arises from applying the metric compatibility condition to the pairs (Y,Z)(Y, Z)(Y,Z), (Z,X)(Z, X)(Z,X), and (X,Y)(X, Y)(X,Y), then combining with torsion-freeness to account for Lie brackets. It ensures that ∇XY\nabla_X Y∇X​Y is uniquely determined pointwise by the metric and its first derivatives.[33][32]

In local coordinates (xi)(x^i)(xi) on MMM, the Levi-Civita connection is specified by its Christoffel symbols of the second kind Γijk\Gamma^k_{ij}Γijk​, defined by ∇∂i∂j=Γijk∂k\nabla_{\partial_i} \partial_j = \Gamma^k_{ij} \partial_k∇∂i​​∂j​=Γijk​∂k​. These are given by

where gij=g(∂i,∂j)g_{ij} = g(\partial_i, \partial_j)gij​=g(∂i​,∂j​) are the components of the metric tensor and gklg^{kl}gkl are those of its inverse. The symmetry Γijk=Γjik\Gamma^k_{ij} = \Gamma^k_{ji}Γijk​=Γjik​ follows directly from torsion-freeness. This formula is obtained by substituting coordinate vector fields into the Koszul formula and evaluating at basis vectors.[31][33]

The uniqueness of the Levi-Civita connection is established by the fundamental theorem of Riemannian geometry. Suppose ∇\nabla∇ is any torsion-free, metric-compatible connection on (M,g)(M, g)(M,g). Applying the compatibility condition to the pairs (Y,Z)(Y, Z)(Y,Z), (Z,X)(Z, X)(Z,X), and (X,Y)(X, Y)(X,Y) yields three equations:

Adding the first two and subtracting the third, then using torsion-freeness to replace ∇YX=∇XY+[X,Y]\nabla_Y X = \nabla_X Y + [X, Y]∇Y​X=∇X​Y+[X,Y] and similarly for other terms, reduces to the Koszul formula. Since the right-hand side depends only on ggg and its derivatives, ∇XY\nabla_X Y∇X​Y is uniquely fixed.[33][32]

For a concrete example, consider the unit 2-sphere S2⊂R3S^2 \subset \mathbb{R}^3S2⊂R3 equipped with the induced metric g=dθ2+sin⁡2θ dϕ2g = d\theta^2 + \sin^2 \theta , d\phi^2g=dθ2+sin2θdϕ2 in spherical coordinates (θ,ϕ)(\theta, \phi)(θ,ϕ), where θ∈(0,π)\theta \in (0, \pi)θ∈(0,π) is the colatitude and ϕ∈[0,2π)\phi \in [0, 2\pi)ϕ∈[0,2π) is the longitude. The non-vanishing Christoffel symbols are Γϕϕθ=−sin⁡θcos⁡θ\Gamma^\theta_{\phi\phi} = -\sin\theta \cos\thetaΓϕϕθ​=−sinθcosθ and Γθϕϕ=Γϕθϕ=cot⁡θ\Gamma^\phi_{\theta\phi} = \Gamma^\phi_{\phi\theta} = \cot\thetaΓθϕϕ​=Γϕθϕ​=cotθ. These symbols arise from substituting the metric components into the general formula for Γijk\Gamma^k_{ij}Γijk​. The associated geodesic equation ∇γ˙γ˙=0\nabla_{\dot\gamma} \dot\gamma = 0∇γ˙​​γ˙​=0 has solutions that are great circles, the shortest paths on S2S^2S2, confirming the connection's role in defining intrinsic geometry.[31][34]

More generally, any metric-compatible connection ∇\nabla∇ (not necessarily torsion-free) on (M,g)(M, g)(M,g) relates to the Levi-Civita connection ∇\tilde{\nabla}∇ via the contorsion tensor KKK, defined such that ∇XY=∇XY+K(X,Y)\nabla_X Y = \tilde{\nabla}X Y + K(X, Y)∇X​Y=∇X​Y+K(X,Y). In components, this is Γijh=Γijh+Kijh\Gamma^h{ij} = \tilde{\Gamma}^h_{ij} + K^h_{ij}Γijh​=Γijh​+Kijh​, where the contorsion is expressed in terms of the torsion tensor TTT of ∇\nabla∇. This decomposition allows general metric connections to be viewed as perturbations of the Levi-Civita connection by a torsion-induced adjustment.

Flat and Projective Connections

A flat connection on a smooth manifold is an affine connection whose curvature tensor vanishes identically, denoted R=0R = 0R=0 everywhere. This condition implies that the connection admits a local trivialization of the associated fiber bundle, where it reduces to the standard flat connection on a product bundle, allowing coordinate charts in which the Christoffel symbols Γijk=0\Gamma^k_{ij} = 0Γijk​=0.[35] Such trivializations enable parallel transport to act as local isomorphisms between fibers, preserving the structure without holonomy obstructions in simply connected regions.[36]

In the context of Ehresmann connections, the vanishing curvature ensures that the horizontal distribution is integrable by the Frobenius theorem, integrating locally to a foliation that yields trivial bundle structures over coordinate neighborhoods.[37] For surfaces, a flat affine connection implies the surface is developable, meaning it can be locally isometrically mapped to the plane without distortion, as the intrinsic geometry aligns with Euclidean flatness.[38]

A classic example is the standard affine flat connection on Rn\mathbb{R}^nRn, where Γ=0\Gamma = 0Γ=0, making straight lines the geodesics and endowing the space with the canonical flat structure.[35]

Weyl connections extend metric-compatible connections to conformal manifolds by allowing a scaling factor in the metric preservation. Specifically, on a manifold equipped with a conformal class of metrics $$, a Weyl connection ∇\nabla∇ satisfies ∇Xg(Y,Z)=Q(X,Y,Z)=−2ω(X)g(Y,Z)\nabla_X g(Y, Z) = Q(X, Y, Z) = -2 \omega(X) g(Y, Z)∇X​g(Y,Z)=Q(X,Y,Z)=−2ω(X)g(Y,Z) for some 1-form ω\omegaω, introducing non-metricity that scales lengths under parallel transport while preserving angles.[39]

Projective connections generalize affine connections by focusing on the projective structure they induce, defined as an equivalence class of torsion-free affine connections that share the same unparametrized geodesics—curves determined up to reparametrization. This equivalence relation groups connections ∇\nabla∇ and ∇^\hat{\nabla}∇^ if ∇^XY=∇XY+α(X)Y+β(Y)X\hat{\nabla}_X Y = \nabla_X Y + \alpha(X) Y + \beta(Y) X∇^X​Y=∇X​Y+α(X)Y+β(Y)X for 1-forms α,β\alpha, \betaα,β, ensuring the geodesic paths coincide projectively.[40] Projective connections preserve this unparametrized path structure, making them invariant under projective transformations.

An illustrative example is the projective structure arising in Klein geometries, where a Cartan connection models the manifold on the projective space RPn\mathbb{RP}^nRPn with the projective linear group acting transitively, capturing the full projective invariance without a preferred metric.[41]

Beyond flat cases, projective connections appear in non-flat settings such as conformal geometry, where they encode projective invariants on manifolds with Weyl structures, allowing non-trivial holonomy while maintaining geodesic projectivity.[42]

Parallel Transport

Parallel transport along a smooth curve γ:I→M\gamma: I \to Mγ:I→M on a manifold MMM equipped with an affine connection ∇\nabla∇ is defined for a vector field VVV along γ\gammaγ that satisfies ∇γ˙(t)V=0\nabla_{\dot{\gamma}(t)} V = 0∇γ˙​(t)​V=0 for all t∈It \in It∈I.[43] In local coordinates, this condition translates to the ordinary differential equation V′(t)+Γ(V(t),γ˙(t))=0V'(t) + \Gamma(V(t), \dot{\gamma}(t)) = 0V′(t)+Γ(V(t),γ˙​(t))=0, where Γ\GammaΓ denotes the Christoffel symbols of the connection.[44]

The existence and uniqueness of such a parallel vector field VVV along γ\gammaγ, given an initial value V(t0)V(t_0)V(t0​), follow from the Picard-Lindelöf theorem, as the Christoffel symbols Γ\GammaΓ are locally Lipschitz continuous on smooth manifolds.[44] This parallel transport map provides a linear isomorphism between the fibers of the pullback bundle γ∗TM\gamma^* TMγ∗TM over the endpoints γ(t0)\gamma(t_0)γ(t0​) and γ(t1)\gamma(t_1)γ(t1​), preserving the vector bundle structure.[45]

For a closed curve γ\gammaγ based at a point p∈Mp \in Mp∈M, the composition of parallel transports around the loop yields an element of the holonomy group Holp(∇)\mathrm{Hol}_p(\nabla)Holp​(∇), which is a Lie subgroup of GL(TpM)≅GL(n,R)\mathrm{GL}(T_p M) \cong \mathrm{GL}(n, \mathbb{R})GL(Tp​M)≅GL(n,R) and encodes the global obstruction to path independence caused by curvature.[18] The holonomy group is influenced by the curvature tensor, as infinitesimal holonomies around small loops approximate the curvature operator.[18]

If the curvature tensor R=0R = 0R=0, the connection is flat, and parallel transport depends only on the endpoints of the curve, independent of the path taken. A classic example occurs on the unit sphere S2S^2S2 with the Levi-Civita connection: parallel transporting a tangent vector around the boundary of a geodesic triangle results in a rotation by the angular excess of the triangle, directly linking to the Gauss-Bonnet theorem.[46]

Geodesics and Riemannian Geometry

In the context of an affine connection on a manifold, geodesics are defined as the auto-parallel curves, meaning curves whose tangent vectors are parallel transported along themselves. This leads to the geodesic equation, which states that for a curve γ:I→M\gamma: I \to Mγ:I→M with velocity γ′\gamma'γ′, the covariant derivative satisfies ∇γ′γ′=0\nabla_{\gamma'} \gamma' = 0∇γ′​γ′=0. In local coordinates xkx^kxk, this equation takes the form d2xkdt2+Γijkdxidtdxjdt=0\frac{d^2 x^k}{dt^2} + \Gamma^k_{ij} \frac{dx^i}{dt} \frac{dx^j}{dt} = 0dt2d2xk​+Γijk​dtdxi​dtdxj​=0, where Γijk\Gamma^k_{ij}Γijk​ are the Christoffel symbols of the connection.

In Riemannian geometry, the Levi-Civita connection, being torsion-free and metric-compatible, defines geodesics that locally minimize the length functional, serving as the shortest paths between points on the manifold. These geodesics parameterize the exponential map exp⁡p:TpM→M\exp_p: T_p M \to Mexpp​:Tp​M→M, where for a vector v∈TpMv \in T_p Mv∈Tp​M, exp⁡p(v)=γ(1)\exp_p(v) = \gamma(1)expp​(v)=γ(1) for the unique geodesic γ\gammaγ satisfying γ(0)=p\gamma(0) = pγ(0)=p and γ′(0)=v\gamma'(0) = vγ′(0)=v, assuming the parameter is affine. This map provides a local diffeomorphism near the zero section, facilitating the study of the manifold's geometry around point ppp.

To analyze the stability and focusing of geodesics, Jacobi fields arise as the first variation fields along geodesic variations. For a Jacobi field JJJ along a geodesic γ\gammaγ, it satisfies the second-order equation ∇2J+R(J,γ′)γ′=0\nabla^2 J + R(J, \gamma') \gamma' = 0∇2J+R(J,γ′)γ′=0, where RRR is the Riemann curvature tensor and ∇2J=∇γ′∇γ′J−∇∇γ′Jγ′\nabla^2 J = \nabla_{\gamma'} \nabla_{\gamma'} J - \nabla_{\nabla_{\gamma'} J} \gamma'∇2J=∇γ′​∇γ′​J−∇∇γ′​J​γ′. Zeros of nontrivial Jacobi fields indicate conjugate points, where nearby geodesics intersect, signaling the boundary of the injectivity radius.

The curvature influences geodesic behavior profoundly, with applications extending to topology and physics. The Gauss-Bonnet theorem relates the integral of the Gaussian curvature over a surface to its Euler characteristic, incorporating geodesic curvature along boundaries to link local geometry to global topology. In general relativity, the Einstein field equations Gμν=8πGc4TμνG_{\mu\nu} = \frac{8\pi G}{c^4} T_{\mu\nu}Gμν​=c48πG​Tμν​ employ the Ricci tensor, the contraction of the Riemann tensor, to describe how matter curves spacetime, with geodesics tracing free-fall trajectories.

A representative example occurs in hyperbolic space Hn\mathbb{H}^nHn, where the constant negative sectional curvature causes nearby geodesics to diverge exponentially: the separation grows proportionally to et−Ke^{t \sqrt{-K}}et−K​ for large ttt, with K<0K < 0K<0 the curvature, contrasting with the convergence in spaces of positive curvature.

Geodesic completeness, the property that all geodesics can be extended indefinitely, is equivalent via the Hopf-Rinow theorem to metric completeness and compactness of closed balls in Riemannian manifolds, ensuring the existence of minimizing geodesics between any two points.