Levi–Civita Connections, Geodesics, and Riemann Curvature

On a smooth manifold with a nondegenerate metric, two conditions select a unique connection on the tangent bundle: parallel transport must preserve the metric, and the connection must have zero torsion. This Levi–Civita connection defines geodesics as curves whose tangent vectors transport parallel to themselves. Its failure to commute in two directions is the Riemann curvature tensor:

$$\begin{aligned} \nabla g=0,\quad T=0 &\quad\Longrightarrow\quad \nabla=\nabla^{\mathrm{LC}}, \\ \nabla_{\dot\gamma}\dot\gamma=0, &\qquad [\nabla_\mu,\nabla_\nu]V^\rho = R^\rho{}_{\sigma\mu\nu}V^\sigma \end{aligned}$$

The first implication is a uniqueness theorem, not an extra choice of coordinates. The geodesic equation describes affinely parametrized autoparallels, not every globally shortest path. Curvature is tensorial even though the connection coefficients are not.

These statements apply to both Riemannian and pseudo-Riemannian metrics; no orientation, positivity, completeness, or field equation is required. When a Lorentzian example is used below, the site’s signature is $(+---)$. This page develops the reusable metric geometry. Gravitational dynamics, curved-spacetime states, horizons, renormalization, and the developed physics of curvature-coupled fields belong to later pages.

Required background. Metrics, Volume Forms, Hodge Star, and Laplace Operators supplies nondegenerate metrics, inverse metrics, metric volume, and the site’s scalar-Laplacian convention.

A connection differentiates tangent fields

Tangent vectors based at different points lie in different vector spaces, so there is no canonical expression such as $V(q)-V(p)$. A connection supplies the missing comparison rule. An affine connection on $TM$ is a map

$$\nabla: \mathfrak X(M)\times\mathfrak X(M) \longrightarrow \mathfrak X(M), \qquad (X,Y)\longmapsto\nabla_XY,$$

that is $C^\infty(M)$-linear in its first argument, real-linear in its second, and satisfies

$$\begin{aligned} \nabla_{fX+gY}Z &= f\nabla_XZ+g\nabla_YZ, \\ \nabla_X(fY+gZ) &= X[f]Y+X[g]Z +f\nabla_XY+g\nabla_XZ . \end{aligned}$$

In coordinates, define its coefficients by

$$\nabla_{\partial_\mu}\partial_\nu = \Gamma^\rho_{\mu\nu}\partial_\rho.$$

The induced covariant derivatives of a vector and a covector are

$$\begin{aligned} \nabla_\mu V^\rho &= \partial_\mu V^\rho +\Gamma^\rho_{\mu\sigma}V^\sigma, \\ \nabla_\mu\omega_\nu &= \partial_\mu\omega_\nu -\Gamma^\rho_{\mu\nu}\omega_\rho. \end{aligned}$$

The opposite signs are forced by differentiating the scalar contraction $\omega_\nu V^\nu$. The same Leibniz rule then extends $\nabla$ to every tensor bundle.

The coefficients $\Gamma^\rho_{\mu\nu}$ are not the components of a tensor. Under a coordinate change $x\mapsto x'$, they obey

$$\begin{aligned} \Gamma'{}^\rho_{\mu\nu} ={}& \frac{\partial x'^\rho}{\partial x^\lambda} \frac{\partial x^\alpha}{\partial x'^\mu} \frac{\partial x^\beta}{\partial x'^\nu} \Gamma^\lambda_{\alpha\beta} \\ &+ \frac{\partial x'^\rho}{\partial x^\lambda} \frac{\partial^2x^\lambda} {\partial x'^\mu\partial x'^\nu}. \end{aligned}$$

The inhomogeneous second-derivative term is precisely what lets $\nabla_\mu V^\rho$ transform tensorially. By contrast, if $\widetilde\nabla$ and $\nabla$ are two connections, then

$$A(X,Y) = \widetilde\nabla_XY-\nabla_XY$$

is $C^\infty(M)$-linear in both arguments and is therefore a $(1,2)$-tensor. This difference test is often the quickest way to distinguish a connection coefficient from a geometric field. Lee 2018, Chapters 4–5 gives a coordinate-free construction of connections, their induced tensor derivatives, and parallel transport.

Here $\nabla$ acts on tangent and tensor indices. An internal gauge connection, conventionally written $D$, acts on a separate vector bundle and is not selected by the spacetime metric. A field carrying both kinds of index may require both connections. Their common bundle language is developed in Bundle Connections, Curvature, Gauge Transformations, and Bianchi Identities; they should not be identified merely because both have local connection coefficients.

The metric selects the Levi–Civita connection

Two tensors measure how an affine connection relates to the metric geometry. Its torsion is

$$T(X,Y) = \nabla_XY-\nabla_YX-[X,Y].$$

In a coordinate frame, whose basis vector fields commute,

$$T^\rho{}_{\mu\nu} = \Gamma^\rho_{\mu\nu} -\Gamma^\rho_{\nu\mu}.$$

Thus zero torsion makes the two lower connection indices symmetric in a coordinate frame. In a noncoordinate frame, the frame commutator also contributes, so symmetry of the displayed coefficients alone is not an invariant definition.

The connection is metric-compatible when

$$(\nabla_Xg)(Y,Z) = X[g(Y,Z)] -g(\nabla_XY,Z) -g(Y,\nabla_XZ) =0.$$

Equivalently, parallel transport preserves all metric pairings. In coordinates this condition is

$$\nabla_\mu g_{\nu\rho} = \partial_\mu g_{\nu\rho} -\Gamma^\lambda_{\mu\nu}g_{\lambda\rho} -\Gamma^\lambda_{\mu\rho}g_{\nu\lambda} =0.$$

The fundamental theorem of pseudo-Riemannian geometry states:

Every smooth nondegenerate symmetric metric has a unique torsion-free, metric-compatible affine connection.

To see why uniqueness is forced, apply metric compatibility to cyclic permutations of $X,Y,Z$ and use $\nabla_XY-\nabla_YX=[X,Y]$. Adding two of the resulting equations and subtracting the third gives the Koszul formula

$$\begin{aligned} 2g(\nabla_XY,Z) ={}& X[g(Y,Z)] +Y[g(Z,X)] -Z[g(X,Y)] \\ &- g(X,[Y,Z]) +g(Y,[Z,X]) +g(Z,[X,Y]). \end{aligned}$$

Its right-hand side depends only on $g$, $X$, $Y$, and $Z$. Nondegeneracy of $g$ therefore determines $\nabla_XY$ uniquely. Conversely, the Koszul formula defines a connection and a direct substitution verifies both required properties, proving existence.

For a coordinate frame, the Lie brackets vanish. Writing the Koszul formula three times, or solving $\nabla_\mu g_{\nu\rho}=0$ together with $\Gamma^\rho_{\mu\nu}=\Gamma^\rho_{\nu\mu}$, gives

$$\boxed{ \Gamma^\rho_{\mu\nu} = \frac12g^{\rho\lambda} \left( \partial_\mu g_{\nu\lambda} +\partial_\nu g_{\mu\lambda} -\partial_\lambda g_{\mu\nu} \right) }$$

for the Levi–Civita coefficients. This formula requires the inverse metric but not an orientation or a preferred signature. Nakahara 2003, §§ 7.2–7.4 independently develops the affine, metric, and Levi–Civita constructions with the same invariant curvature sign used below.

Both hypotheses matter:

• If torsion-freeness is dropped, let $H$ be any three-form and set

  $$\widetilde\nabla_XY = \nabla_XY + \frac12 \bigl(H(X,Y,\mathord\cdot)\bigr)^\sharp .$$

  This new connection is still metric-compatible, but its torsion is $\bigl(H(X,Y,\mathord\cdot)\bigr)^\sharp$.

• If metric compatibility is dropped, adding any symmetric $(1,2)$-tensor $A(X,Y)=A(Y,X)$ preserves zero torsion but generally gives $\widetilde\nabla g\ne0$.

• If the symmetric field called a “metric” is degenerate, there is no inverse musical map and the Koszul right-hand side need not determine $\nabla_XY$. The theorem does not apply.

The metric therefore selects a connection only through the conjunction of nondegeneracy, metric compatibility, and zero torsion.

Parallel transport and geodesics

Let $\gamma(\lambda)$ be a smooth curve and let $W=W^\rho\partial_\rho$ be a vector field along it. Covariant differentiation along the curve is

$$\frac{DW^\rho}{D\lambda} = \frac{\mathrm dW^\rho}{\mathrm d\lambda} + \Gamma^\rho_{\mu\nu} \frac{\mathrm dx^\mu}{\mathrm d\lambda} W^\nu.$$

A field is parallel along $\gamma$ when $DW/D\lambda=0$. This linear ordinary differential equation uniquely transports initial tangent data along the curve wherever the coordinate solution exists. Metric compatibility gives

$$\frac{\mathrm d}{\mathrm d\lambda}g(W,U) = g\left(\frac{DW}{D\lambda},U\right) + g\left(W,\frac{DU}{D\lambda}\right),$$

so parallel fields keep their mutual inner products.

A geodesic of the Levi–Civita connection is a curve whose tangent $\dot\gamma$ is parallel along itself:

$$\boxed{ \nabla_{\dot\gamma}\dot\gamma=0 }$$

or, in coordinates,

$$\frac{\mathrm d^2x^\rho}{\mathrm d\lambda^2} + \Gamma^\rho_{\mu\nu} \frac{\mathrm dx^\mu}{\mathrm d\lambda} \frac{\mathrm dx^\nu}{\mathrm d\lambda} =0.$$

Given initial data $\gamma(0)=p$ and $\dot\gamma(0)=v$, this second-order system has a unique local solution, which extends uniquely to a maximal parameter interval. Along it,

$$\frac{\mathrm d}{\mathrm d\lambda} g(\dot\gamma,\dot\gamma) = 2g(\nabla_{\dot\gamma}\dot\gamma,\dot\gamma) =0.$$

In signature $(+---)$, a nonzero geodesic therefore remains timelike, null, or spacelike according as this constant is positive, zero, or negative.

The displayed coordinate equation uses an affine parameter. Replacing $\lambda$ by $a\lambda+b$, with $a\ne0$, preserves its form. For a general locally invertible parameter change $\lambda=\lambda(u)$, so that $\mathrm d\lambda/\mathrm du\ne0$, the same unparametrized curve satisfies

$$\frac{D}{\mathrm du} \left( \frac{\mathrm d\gamma}{\mathrm du} \right) = \frac{\mathrm d^2\lambda/\mathrm du^2} {\mathrm d\lambda/\mathrm du} \frac{\mathrm d\gamma}{\mathrm du}.$$

The extra tangent term disappears exactly for an affine change of parameter. Proper length or proper time can supply an affine parameter for suitable non-null geodesics, but a null geodesic has zero proper length and needs an affine parameter chosen by another normalization.

“Geodesic” is a local differential condition, not a synonym for “globally shortest path.” In Riemannian geometry, sufficiently short geodesic segments minimize length locally, but longer ones need not: the two semicircles between antipodal points on a sphere already show nonuniqueness. In Lorentzian geometry, suitable timelike geodesic segments locally maximize proper time; null and spacelike curves require different interpretations. Completeness, cut loci, and global causal questions need hypotheses beyond the local equation and are not developed here.

Curvature is the invariant commutator

For any affine connection, define the curvature operator by

$$\boxed{ R(X,Y)Z = \nabla_X\nabla_YZ -\nabla_Y\nabla_XZ -\nabla_{[X,Y]}Z. }$$

The Lie-bracket correction makes this expression $C^\infty(M)$-linear in all three vector fields. Curvature is therefore a tensor even though it is built from non-tensorial connection coefficients. It measures the infinitesimal failure of a vector transported around a small loop to return to its initial value and the failure of covariant derivatives to commute.

This site fixes the convention

$$\boxed{ [\nabla_\mu,\nabla_\nu]V^\rho = R^\rho{}_{\sigma\mu\nu}V^\sigma }$$

and hence

$$\begin{aligned} R^\rho{}_{\sigma\mu\nu} ={}& \partial_\mu\Gamma^\rho_{\nu\sigma} -\partial_\nu\Gamma^\rho_{\mu\sigma} \\ &+ \Gamma^\rho_{\mu\lambda}\Gamma^\lambda_{\nu\sigma} -\Gamma^\rho_{\nu\lambda}\Gamma^\lambda_{\mu\sigma}, \\[2pt] R_{\sigma\nu} ={}& R^\rho{}_{\sigma\rho\nu}. \end{aligned}$$

The metric signature does not determine this sign convention. A source must match both the vector commutator and the Ricci contraction before its component formulas can be imported. On a covector the dual representation introduces a minus sign:

$$[\nabla_\mu,\nabla_\nu]\omega_\rho = -R^\sigma{}_{\rho\mu\nu}\omega_\sigma.$$

For the Levi–Civita connection, lower the first index with the metric,

$$R_{\rho\sigma\mu\nu} = g_{\rho\lambda}R^\lambda{}_{\sigma\mu\nu}.$$

Metric compatibility and zero torsion then imply

$$\begin{aligned} R_{\rho\sigma\mu\nu} &= -R_{\sigma\rho\mu\nu} = -R_{\rho\sigma\nu\mu}, \\ R_{\rho\sigma\mu\nu} &= R_{\mu\nu\rho\sigma}, \\ R^\rho{}_{[\sigma\mu\nu]} &= 0. \end{aligned}$$

The last line is the algebraic, or first, Bianchi identity. The differential, or second, Bianchi identity is

$$\nabla_{[\lambda} R^\rho{}_{|\sigma|\mu\nu]} =0.$$

The Ricci tensor and scalar curvature are

$$R_{\mu\nu} = R^\rho{}_{\mu\rho\nu}, \qquad R = g^{\mu\nu}R_{\mu\nu}.$$

For a Levi–Civita connection, $R_{\mu\nu}$ is symmetric. Contracting the second Bianchi identity yields

$$\nabla^\mu \left( R_{\mu\nu} -\frac12g_{\mu\nu}R \right) =0.$$

This is a geometric identity. It is not an equation of motion and does not select any spacetime metric. Lee 2018, Chapter 7 develops the curvature tensor, its symmetries, and its contractions; Frankel 2012, Chapters 9–11 provides an independent physics-facing development. Frankel uses the same mixed-index curvature definition but a mostly-plus Lorentzian metric, so its Lorentzian scalar contractions require a signature translation.

Two coordinate checks separate coefficients from curvature

Flat spacetime in cylindrical coordinates

Write Minkowski spacetime away from the cylindrical axis as

$$\mathrm ds^2 = \mathrm dt^2-\mathrm dr^2-r^2\mathrm d\theta^2-\mathrm dz^2.$$

The metric components depend on $r$, and the nonzero Levi–Civita coefficients include

$$\Gamma^r_{\theta\theta} = -r, \qquad \Gamma^\theta_{r\theta} = \Gamma^\theta_{\theta r} = \frac1r.$$

Nevertheless,

$$\begin{aligned} R^r{}_{\theta r\theta} &= \partial_r\Gamma^r_{\theta\theta} -\partial_\theta\Gamma^r_{r\theta} +\Gamma^r_{r\lambda}\Gamma^\lambda_{\theta\theta} -\Gamma^r_{\theta\lambda}\Gamma^\lambda_{r\theta} \\ &= -1-(-1) =0. \end{aligned}$$

All other components vanish as well, because this is the flat Minkowski metric in a curvilinear chart. Nonzero connection coefficients can record a changing coordinate frame without recording curvature. The corresponding geodesic equations also need not look like straight coordinate lines; they still describe straight lines in Cartesian coordinates.

The round two-sphere

For a sphere of radius $a$,

$$\mathrm ds^2 = a^2\left( \mathrm d\vartheta^2 +\sin^2\vartheta\,\mathrm d\varphi^2 \right).$$

The independent nonzero coefficients are

$$\Gamma^\vartheta_{\varphi\varphi} = -\sin\vartheta\cos\vartheta, \qquad \Gamma^\varphi_{\vartheta\varphi} = \Gamma^\varphi_{\varphi\vartheta} = \cot\vartheta.$$

Substitution into the site’s component convention gives

$$\begin{aligned} R^\vartheta{}_{\varphi\vartheta\varphi} &= \sin^2\vartheta, \\ R_{\vartheta\varphi\vartheta\varphi} &= a^2\sin^2\vartheta, \\ R_{\mu\nu} &= \frac1{a^2}g_{\mu\nu}, \qquad R = \frac2{a^2}. \end{aligned}$$

Thus the round sphere has positive scalar curvature in the convention used throughout this site. More generally, an $n$-dimensional space of constant sectional curvature $K$ satisfies

$$\begin{aligned} R_{\rho\sigma\mu\nu} &= K\left( g_{\rho\mu}g_{\sigma\nu} -g_{\rho\nu}g_{\sigma\mu} \right), \\ R_{\mu\nu} &= (n-1)K g_{\mu\nu}, \qquad R = n(n-1)K. \end{aligned}$$

The equator $\vartheta=\pi/2$ with constant angular speed satisfies the geodesic equation. A nondegenerate line of latitude with $0<\vartheta<\pi$ and $\vartheta\ne\pi/2$ does not, because its $\vartheta$ equation contains $-\sin\vartheta\cos\vartheta\,\dot\varphi^2$. This supplies a genuinely curved check of both the connection and curvature signs.

At any point of either geometry, normal coordinates can make $\Gamma^\rho_{\mu\nu}=0$ at that point. On the sphere the curvature there remains nonzero because it depends on derivatives and quadratic combinations of the connection. Therefore neither $\Gamma\ne0$ nor $\Gamma=0$ at one point decides whether the geometry is curved.

Connection-based Laplace operators

For a scalar $\phi$,

$$\nabla_\mu\phi = \partial_\mu\phi,$$

but $\partial_\nu\phi$ is a covector. Its next derivative is therefore

$$\nabla_\mu\nabla_\nu\phi = \partial_\mu\partial_\nu\phi -\Gamma^\rho_{\mu\nu}\partial_\rho\phi.$$

Tracing with the inverse metric defines the scalar d’Alembertian or Laplace–Beltrami expression

$$\begin{aligned} \Box_g\phi &= g^{\mu\nu}\nabla_\mu\nabla_\nu\phi \\ &= \frac1{\sqrt{|g|}} \partial_\mu \left( \sqrt{|g|}\, g^{\mu\nu}\partial_\nu\phi \right), \qquad |g|=\left|\det(g_{\mu\nu})\right|. \end{aligned}$$

The second line follows from $\Gamma^\mu_{\mu\nu}=\partial_\nu\log\sqrt{|g|}$. It is the connection-based form promised by the preceding metric and Hodge page. With that page’s conventions,

$$\Delta_{\mathrm H}\phi = -\Box_g\phi.$$

In Riemannian signature this sign makes the Hodge Laplacian nonnegative under the usual compact-support or domain assumptions. In Lorentzian signature, $\Box_g$ is wave-type rather than elliptic.

In the Weitzenböck comparison, an explicit curvature correction first appears when a tensor index is present. For a one-form $\alpha$, the corresponding identity in the site’s conventions is

$$(\Delta_{\mathrm H}\alpha)_\nu = -\nabla^\mu\nabla_\mu\alpha_\nu +R_\nu{}^\rho\alpha_\rho.$$

The Ricci term comes from commuting covariant derivatives. Thus on a curved manifold the Hodge Laplacian is not obtained by applying the scalar formula independently to each coordinate component. Formal adjoints, operator domains, boundary conditions, and spectral theorems remain separate analytic questions.

A bounded QFT bridge: assemble the scalar operator

On a fixed Lorentzian background, the Levi–Civita connection supplies two local geometric ingredients for a curvature-coupled scalar differential expression:

$$P_\xi = \Box_g+m^2+\xi R.$$

This page can assemble and check that expression without developing the field theory. On flat Minkowski spacetime in signature $(+---)$,

$$P_\xi = \partial_t^2-\boldsymbol\nabla^2+m^2,$$

because $R=0$. On an $n$-dimensional fixed background of constant sectional curvature $K$,

$$P_\xi = \Box_g+m^2+\xi n(n-1)K.$$

The connection determines the second-order covariant part, while scalar curvature supplies a coordinate-invariant zeroth-order coefficient. The signs shown here are tied to both the site’s $(+---)$ metric convention and the curvature convention stated above.

Ferguson 2013, pp. 853–892 treats the nonminimally coupled scalar as a locally covariant field theory on curved spacetimes, providing a QFT-facing source for why this geometric operator matters. The action, equation-of-motion derivation, conformal value of $\xi$, stress tensor, boundary variation, mode equations, quantization, states, observables, and renormalization belong to Covariant Scalar Fields and Curvature Coupling.

Common pitfalls

Treating Christoffel symbols as a tensor. Their inhomogeneous transformation term is essential. Use the difference of two connections or the Riemann tensor when a tensorial object is required.

Inferring curvature from the value of the connection. Cylindrical coordinates have nonzero Christoffel symbols in flat spacetime, while normal coordinates make them vanish at one point of a curved space. Curvature requires the full derivative-plus-quadratic combination.

Calling every geodesic a globally shortest path. The invariant definition is $\nabla_{\dot\gamma}\dot\gamma=0$ with an affine parameter. Global minimization, timelike proper-time maximization, uniqueness, and completeness all require additional hypotheses.

Using proper length to parametrize a null geodesic. Null proper length vanishes. A null geodesic still admits affine parameters, but they are not obtained by normalizing its tangent to unit length.

Dropping one Levi–Civita hypothesis. Metric compatibility alone permits connections with torsion; zero torsion alone permits nonmetric connections. Uniqueness uses both conditions and a nondegenerate metric.

Forgetting that a second scalar derivative has a covector index. Although $\nabla_\mu\phi=\partial_\mu\phi$, the correct Hessian contains $-\Gamma^\rho_{\mu\nu}\partial_\rho\phi$.

Importing only one curvature sign. Signature and Riemann-sign conventions are independent. Check the vector commutator, component formula, Ricci contraction, scalar contraction, and every downstream curvature term together.

Reading a Bianchi identity as dynamics. The Bianchi identities follow from the connection and its curvature. They do not impose a gravitational field equation.

Exercises

These checks test the selection theorem, affine parameters, coordinate invariance, and the QFT-facing operator.

Hypothesis check. Starting from the Levi–Civita connection, how can one construct (a) another metric-compatible connection and (b) another torsion-free connection?

Hypothesis answer

For any three-form $H$, define

$$\widetilde\nabla_XY = \nabla_XY + \frac12 \bigl(H(X,Y,\mathord\cdot)\bigr)^\sharp.$$

It remains metric-compatible but has torsion $\bigl(H(X,Y,\mathord\cdot)\bigr)^\sharp$. Alternatively, add any symmetric $(1,2)$-tensor $A(X,Y)=A(Y,X)$. The resulting connection remains torsion-free but is generally not metric-compatible. Hence either condition alone fails to select the Levi–Civita connection.

Parameter check. If $\gamma(\lambda)$ is affinely parametrized and $\lambda=\lambda(u)$ is locally invertible, what equation does the same curve obey in the $u$ parameter? When is $u$ affine?

Parameter answer

The chain rule gives

$$\frac{D}{\mathrm du} \left( \frac{\mathrm d\gamma}{\mathrm du} \right) = \frac{\mathrm d^2\lambda/\mathrm du^2} {\mathrm d\lambda/\mathrm du} \frac{\mathrm d\gamma}{\mathrm du}.$$

The right-hand side vanishes exactly when $\mathrm d^2\lambda/\mathrm du^2=0$, so $\lambda=au+b$ with $a\ne0$. Those and only those reparametrizations preserve the affine geodesic equation.

Curvature check. Use the two cylindrical coefficients $\Gamma^r_{\theta\theta}=-r$ and $\Gamma^\theta_{r\theta}=1/r$ to compute $R^r{}_{\theta r\theta}$.

Curvature answer

Only the radial derivative and one quadratic term survive:

$$\begin{aligned} R^r{}_{\theta r\theta} &= \partial_r(-r) - \Gamma^r_{\theta\theta} \Gamma^\theta_{r\theta} \\ &= -1 - (-r)\frac1r =0. \end{aligned}$$

The nonzero connection coefficients describe the cylindrical frame, not spacetime curvature.

QFT transfer check. What does $P_\xi$ become on an $n$-dimensional background of constant sectional curvature $K$, and what becomes of the curvature term in flat spacetime?

Transfer answer

Since $R=n(n-1)K$,

$$P_\xi = \Box_g+m^2+\xi n(n-1)K.$$

For flat spacetime $K=0$, so the curvature coupling vanishes. With the site’s flat $(+---)$ metric, the remaining differential expression is $\partial_t^2-\boldsymbol\nabla^2+m^2$.

Synthesis and next steps

A connection supplies a derivative that compares tensor fields at nearby points. Metric compatibility and zero torsion select the unique Levi–Civita connection, whose Koszul formula and Christoffel expression are fixed entirely by a nondegenerate metric. Parallel transport along a curve then defines affinely parametrized geodesics. Commuting covariant derivatives produces the Riemann tensor, whose contractions give Ricci and scalar curvature and whose Bianchi identities are geometric rather than dynamical. Flat cylindrical spacetime and the round sphere show why connection coefficients and curvature must be kept distinct.

Continue according to the bundle and physical structure needed:

• Bundle Connections, Curvature, Gauge Transformations, and Bianchi Identities generalizes the connection language to internal gauge bundles;
• Spin Structures and Dirac Operators uses the metric connection after its additional topological and Clifford-algebra prerequisites;
• Covariant Scalar Fields and Curvature Coupling develops the curvature-coupled scalar action, dynamics, and QFT construction.

References

• Matthew Ferguson, “Dynamical Locality of the Nonminimally Coupled Scalar Field and Enlarged Algebra of Wick Polynomials”, Annales Henri Poincaré 14 (2013), 853–892. This article analyzes the role of a nonminimally coupled scalar operator on fixed curved spacetimes.
• Theodore Frankel, The Geometry of Physics: An Introduction, third edition, Cambridge University Press, 2012, Chapters 9–11. This provides an independent treatment of Levi–Civita geometry, geodesics, curvature, and the Bianchi identities. Its Lorentzian examples use the opposite metric signature, so scalar contractions have been translated to the site’s conventions.
• John M. Lee, Introduction to Riemannian Manifolds, second edition, Graduate Texts in Mathematics 176, Springer, 2018, Chapters 4–7. These chapters supply connections, parallel transport, Levi–Civita uniqueness, geodesics, normal coordinates, and curvature.
• Mikio Nakahara, Geometry, Topology and Physics, second edition, Institute of Physics Publishing, 2003, §§ 7.2–7.4. This is the physics-facing source for affine connections, geodesics, torsion, metric compatibility, and Riemann curvature.