The Levi-Civita Connection

⚡ Key Takeaways

One-line statement: The Levi-Civita connection is the unique affine connection on a (pseudo-)Riemannian manifold that is both torsion-free and compatible with the metric. Key result in words: Given any metric tensor, there exists exactly one way to differentiate vector fields that respects lengths/angles and has no intrinsic "twisting," and its coefficients are the Christoffel symbols computed from the metric. Where it applies: Riemannian and pseudo-Riemannian geometry, general relativity, gauge theory backgrounds, geometric mechanics, comparison geometry. Who/when: Tullio Levi-Civita and Gregorio Ricci-Curbastro (parallel transport, ~1900–1917); rooted in Christoffel (1869) and Riemann (1854).

💡 The Core Insight

A metric tells you how to measure lengths and angles at each point, but says nothing a priori about how to compare vectors at different points. A connection supplies that comparison (parallel transport / covariant differentiation). There are infinitely many connections, but if we insist that (i) transporting vectors preserves their inner products and (ii) the connection introduces no "rotation defect" along infinitesimal parallelograms (torsion-free), then exactly one connection survives. Differentiation in metric geometry is therefore not a choice — it is a consequence of the metric.

DefinitionAffine Connection

An affine (linear) connection on a smooth manifold $M$ is an $\mathbb{R}$ -bilinear map

1

\quad \nabla : \Gamma(TM)\times\Gamma(TM)\to\Gamma(TM),\qquad (X,Y)\mapsto \nabla_X Y,

Derivation hint: This is a definition; bilinearity over $\mathbb{R}$ is required before imposing the tensorial/Leibniz axioms below. satisfying $C^\infty(M)$ -linearity in the first slot and the Leibniz rule in the second:

2

\quad \nabla_{fX}Y = f\,\nabla_X Y,\qquad \nabla_X(fY) = (Xf)\,Y + f\,\nabla_X Y,

Derivation hint: The first slot is tensorial (pointwise), so $\nabla_X Y|_p$ depends only on $X_p$ ; the second slot is a derivation, hence carries the $(Xf)Y$ term. for all $X,Y\in\Gamma(TM)$ and $f\in C^\infty(M)$ . Here $\Gamma(TM)$ denotes smooth vector fields.

DefinitionTorsion Tensor

The torsion of a connection $\nabla$ is

3

\quad T(X,Y) := \nabla_X Y - \nabla_Y X - [X,Y],

Derivation hint: Each term is checked to be $C^\infty(M)$ -bilinear in $X,Y$ (the Lie bracket's non-tensorial part cancels the Leibniz terms), so $T$ is a genuine $(1,2)$ -tensor. A connection is torsion-free (symmetric) when $T\equiv 0$ , i.e. $\nabla_X Y - \nabla_Y X = [X,Y]$ .

DefinitionMetric Compatibility

A connection $\nabla$ is compatible with the metric $g$ if parallel transport preserves inner products, equivalently if $\nabla g = 0$ , i.e. for all vector fields $X,Y,Z$ ,

4

\quad X\bigl(g(Y,Z)\bigr) = g(\nabla_X Y, Z) + g(Y, \nabla_X Z).

Derivation hint: Apply the product rule to $g(Y,Z)$ as a scalar function and interpret each factor's "derivative" as the covariant derivative; $\nabla g=0$ is precisely the statement that the metric is parallel.

TheoremFundamental Theorem of Riemannian Geometry

Let $(M,g)$ be a (pseudo-)Riemannian manifold. There exists a unique affine connection $\nabla$ on $M$ that is simultaneously torsion-free and metric-compatible. It is called the Levi-Civita connection and is given explicitly by the Koszul formula:

5

\quad 2\,g(\nabla_X Y, 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).

Derivation hint: Write the metric-compatibility identity [4] three times with cyclic permutations of $(X,Y,Z)$ , add two and subtract one, then use torsion-freeness [3] to convert symmetric covariant-derivative differences into Lie brackets; all $\nabla$ terms collapse except $2g(\nabla_X Y,Z)$ .

DefinitionChristoffel Symbols

In a coordinate chart $(x^1,\dots,x^n)$ with $\partial_i := \partial/\partial x^i$ , write $\nabla_{\partial_i}\partial_j = \Gamma^k_{ij}\,\partial_k$ . The Christoffel symbols of the second kind of the Levi-Civita connection are

6

\quad \Gamma^k_{ij} = \tfrac{1}{2}\,g^{k\ell}\bigl(\partial_i g_{j\ell} + \partial_j g_{i\ell} - \partial_\ell g_{ij}\bigr),

Derivation hint: Specialize the Koszul formula [5] to coordinate vector fields where $[\partial_i,\partial_j]=0$ , then raise the free index with the inverse metric $g^{k\ell}$ . where $g_{ij}=g(\partial_i,\partial_j)$ , $g^{k\ell}$ is the inverse matrix ( $g^{k\ell}g_{\ell j}=\delta^k_j$ ), and the Einstein summation convention is used. Note the symmetry $\Gamma^k_{ij}=\Gamma^k_{ji}$ , reflecting torsion-freeness.

ProofUniqueness and Existence via Koszul

We prove the fundamental theorem constructively.

Step 1 — Necessity (uniqueness). Suppose $\nabla$ is torsion-free and metric-compatible. Apply [4] to the three cyclic orderings:

7

\quad \begin{aligned} X g(Y,Z) &= g(\nabla_X Y,Z) + g(Y,\nabla_X Z),\\ Y g(Z,X) &= g(\nabla_Y Z,X) + g(Z,\nabla_Y X),\\ Z g(X,Y) &= g(\nabla_Z X,Y) + g(X,\nabla_Z Y). \end{aligned}

Derivation hint: This is just [4] with $(X,Y,Z)$ , $(Y,Z,X)$ , $(Z,X,Y)$ substituted in turn.

Step 2 — Combine. Compute (first) + (second) − (third):

8

\quad X g(Y,Z) + Y g(Z,X) - Z g(X,Y) = g(\nabla_X Y + \nabla_Y X, Z) + g(\nabla_X Z - \nabla_Z X, Y) + g(\nabla_Y Z - \nabla_Z Y, X).

Derivation hint: Carefully collect the six terms; pair those sharing a common $Z$ , $Y$ , $X$ slot using the symmetry of $g$ .

Step 3 — Insert torsion-freeness. Using $\nabla_A B - \nabla_B A = [A,B]$ and $\nabla_X Y + \nabla_Y X = 2\nabla_X Y - [X,Y]$ :

9

\quad 2\,g(\nabla_X Y, 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).

Derivation hint: Substitute $\nabla_X Z-\nabla_Z X=[X,Z]=-[Z,X]$ and $\nabla_Y Z-\nabla_Z Y=[Y,Z]$ into [8], then isolate $2g(\nabla_X Y,Z)$ . This is the Koszul formula. Since the right side is fully determined by $g$ and the smooth structure, and since $g$ is non-degenerate, $g(\nabla_X Y, \cdot)$ — hence $\nabla_X Y$ — is uniquely determined. Uniqueness is proved.

Step 4 — Existence. Define $\nabla_X Y$ by declaring [9] to hold for all $Z$ (possible since $g$ is non-degenerate, so the right-hand functional in $Z$ corresponds to a unique vector field). One then verifies directly that this $\nabla$ satisfies tensoriality in $X$ , the Leibniz rule in $Y$ , metric compatibility, and torsion-freeness. Each is a routine algebraic check using bilinearity of brackets and the Leibniz identity. $\blacksquare$

∎

TL

Tullio Levi-Civita

1873–1941

Italian mathematician, student and later collaborator of Gregorio Ricci-Curbastro. In his 1917 paper "Nozione di parallelismo in una varietà qualunque" he introduced the notion of parallel transport on a curved surface, giving a concrete geometric meaning to the absolute differential calculus (tensor calculus) that he and Ricci had developed in their 1900 monograph. Einstein relied on this "Ricci calculus" to formulate general relativity (1915), and the parallel-transport viewpoint clarified what curvature does to vectors moved around loops. Levi-Civita's insight transformed Christoffel's formal symbols into a geometric operation, and the connection now bears his name. He was forced from his Rome professorship in 1938 by the Fascist racial laws and died in 1941.

🧠 Intuition

Think of an ant walking on a curved surface carrying a tiny arrow it tries never to rotate. "Never to rotate" is ill-defined intrinsically — the surface bends — so we declare the rule: keep the arrow's length fixed and turn it as little as the surface permits. That "as little as possible, length-preserving" rule is parallel transport for the Levi-Civita connection.

Metric compatibility = the arrow keeps its length, and two transported arrows keep their angle. Inner products are frozen along the path.
Torsion-free = if you move first along $X$ then $Y$ , versus first $Y$ then $X$ , the only discrepancy in the endpoint comes from the manifold's own non-commuting flows ( $[X,Y]$ ), not from any extra intrinsic twist injected by the connection.

On a sphere, parallel-transporting a vector around a triangle returns it rotated — that residual rotation is curvature, not torsion. The Levi-Civita connection has curvature (it cannot be set to zero on a sphere) but zero torsion: the two are independent failure modes, and Levi-Civita's choice keeps the metric ones and discards the twisting one.

Figure

Parallel transport of a vector around a geodesic triangle on a sphere produces a holonomy rotation, illustrating curvature with zero torsion.

ExampleChristoffel Symbols of the 2-Sphere

Given: The round sphere $S^2$ of radius $R$ with metric in spherical coordinates $(\theta,\phi)$ :

10

\quad g = R^2\, d\theta^2 + R^2\sin^2\!\theta\, d\phi^2,\qquad g_{\theta\theta}=R^2,\ g_{\phi\phi}=R^2\sin^2\theta,\ g_{\theta\phi}=0.

Derivation hint: This is the induced metric from $\mathbb{R}^3$ via $x=R\sin\theta\cos\phi$ , etc.; compute $g_{ij}=\partial_i \mathbf{x}\cdot\partial_j\mathbf{x}$ .

Inverse metric: $g^{\theta\theta}=R^{-2}$ , $g^{\phi\phi}=R^{-2}\sin^{-2}\theta$ .

Step 1 — Nonzero metric derivatives: Only $\partial_\theta g_{\phi\phi} = 2R^2\sin\theta\cos\theta$ is nonzero.

Step 2 — Apply [6]:

11

\quad \Gamma^\theta_{\phi\phi} = \tfrac12 g^{\theta\theta}\bigl(-\partial_\theta g_{\phi\phi}\bigr) = \tfrac12 R^{-2}\bigl(-2R^2\sin\theta\cos\theta\bigr) = -\sin\theta\cos\theta.

Derivation hint: In $\Gamma^\theta_{\phi\phi}$ the relevant term is $-\partial_\theta g_{\phi\phi}$ since the two $\phi$ -indices match and we lower with $g^{\theta\theta}$ .

12

\quad \Gamma^\phi_{\theta\phi} = \Gamma^\phi_{\phi\theta} = \tfrac12 g^{\phi\phi}\,\partial_\theta g_{\phi\phi} = \tfrac12\, \frac{1}{R^2\sin^2\theta}\,(2R^2\sin\theta\cos\theta) = \cot\theta.

Derivation hint: Here one index is $\phi$ and one is $\theta$ ; the surviving term is $\partial_\theta g_{\phi\phi}$ raised by $g^{\phi\phi}$ .

Boxed answer:

\boxed{\ \Gamma^\theta_{\phi\phi} = -\sin\theta\cos\theta,\qquad \Gamma^\phi_{\theta\phi}=\Gamma^\phi_{\phi\theta}=\cot\theta,\qquad \text{all others }=0.\ }

These are dimensionless (connection coefficients have units of inverse length only when one differentiates with a dimensionful coordinate; here $\theta,\phi$ are angles). Note $R$ cancels entirely — consistent with the fact that the shape of geodesics on a sphere is scale-independent.

ExampleGeodesics from the Connection: Great Circles

Given: A geodesic satisfies $\nabla_{\dot\gamma}\dot\gamma=0$ , i.e.

13

\quad \ddot x^k + \Gamma^k_{ij}\,\dot x^i \dot x^j = 0.

Derivation hint: Expand $\nabla_{\dot\gamma}\dot\gamma$ with $\dot\gamma=\dot x^i\partial_i$ using the Leibniz rule [2] and $\nabla_{\partial_i}\partial_j=\Gamma^k_{ij}\partial_k$ . Use the sphere Christoffel symbols from the previous example.

Step 1 — The two equations:

14

\quad \ddot\theta - \sin\theta\cos\theta\,\dot\phi^2 = 0,\qquad \ddot\phi + 2\cot\theta\,\dot\theta\dot\phi = 0.

Derivation hint: For $k=\theta$ only $\Gamma^\theta_{\phi\phi}$ contributes ( $\dot\phi^2$ ); for $k=\phi$ the factor 2 comes from $\Gamma^\phi_{\theta\phi}=\Gamma^\phi_{\phi\theta}$ both contributing.

Step 2 — Test the equator $\theta=\pi/2$ , $\phi(s)=\omega s$ : then $\dot\theta=0$ , $\ddot\theta=0$ , $\cos(\pi/2)=0$ , so the first equation reads $0-0=0$ ✓; second reads $0 + 2\cot(\pi/2)\cdot 0=0$ ✓.

Step 3 — Speed check: $|\dot\gamma|^2 = R^2\sin^2(\pi/2)\,\omega^2 = R^2\omega^2$ , constant, confirming an affinely parametrized geodesic.

Boxed answer:

\boxed{\ \theta(s)=\tfrac{\pi}{2},\quad \phi(s)=\omega s\ \text{ is a geodesic (great circle) of speed } R\omega.\ }

With $R=6371\,\text{km}$ (Earth) and $\omega$ chosen so $R\omega = 800\,\text{km/h}$ , this is the great-circle flight path airlines actually use.

⚠️ Common mistake

"Torsion-free means the connection has no curvature." False. Torsion and curvature are independent tensors. The Levi-Civita connection on a sphere is torsion-free yet has nonzero Riemann curvature — that is exactly why parallel transport around a loop rotates vectors. Torsion measures the failure of infinitesimal parallelograms to close; curvature measures the holonomy (rotation defect) of transport around closed loops. A flat torus has zero curvature; a sphere has zero torsion. Neither implies the other.

" $\Gamma^k_{ij}$ is a tensor." False. Christoffel symbols transform inhomogeneously under coordinate changes (they pick up a second-derivative term). The difference of two connections is a tensor, and the torsion and curvature built from $\Gamma$ are tensors, but $\Gamma$ itself is not — which is why you can make all $\Gamma^k_{ij}(p)=0$ at a single point via normal coordinates, something impossible for a true tensor.

CounterexampleA Metric-Compatible Connection with Torsion

Metric compatibility alone does not single out Levi-Civita. On any $(M,g)$ , take the Levi-Civita connection $\nabla$ and define

15

\quad \tilde\nabla_X Y := \nabla_X Y + \tfrac12\,A(X,Y),

Derivation hint: A new connection differs from $\nabla$ by a $(1,2)$ -tensor $A$ ; metric compatibility of $\tilde\nabla$ constrains the symmetric-in- $g$ part of $A$ . where $A$ is a tensor antisymmetric in the sense $g(A(X,Y),Z)+g(Y,A(X,Z))=0$ (e.g. arising from a non-zero 2-form). Then $\tilde\nabla$ is still metric-compatible ( $\tilde\nabla g = 0$ ) but has torsion $T(X,Y)=\tfrac12(A(X,Y)-A(Y,X))\neq 0$ in general. These "metric connections with torsion" (Cartan connections) are perfectly valid and appear in Einstein–Cartan gravity. The torsion-free condition is what removes this freedom and pins down Levi-Civita uniquely.

connection

The Levi-Civita connection is the keystone object from which the metric geometry edifice is built:

<a class="wiki-link" href="/wiki/riemann-curvature-tensor">Riemann Curvature Tensor</a> is defined as $R(X,Y)Z = \nabla_X\nabla_Y Z - \nabla_Y\nabla_X Z - \nabla_{[X,Y]}Z$ using $\nabla=$ Levi-Civita; its symmetries (e.g. $R_{ijkl}=R_{klij}$ ) follow from metric compatibility and torsion-freeness.
<a class="wiki-link" href="/wiki/geodesics">Geodesics</a> are the curves with $\nabla_{\dot\gamma}\dot\gamma=0$ ; for Levi-Civita these are exactly the locally length-minimizing curves, linking the connection to the Calculus of Variations.
<a class="wiki-link" href="/wiki/parallel-transport">Parallel Transport</a> along a curve is the solution of $\nabla_{\dot\gamma}V=0$ ; metric compatibility makes it an isometry of tangent spaces.
Ricci Curvature and Scalar Curvature are traces of the Levi-Civita Riemann tensor and feed directly into the <a class="wiki-link" href="/wiki/einstein-field-equations">Einstein Field Equations</a>.
The connection is the unique metric connection whose induced Exterior Covariant Derivative on the frame bundle reduces structure group to $O(n)$ (or $O(p,q)$ ) torsion-freely, linking to Principal Bundle Connections.

🔬 Where this appears

⏳ History

- 1854 — Bernhard Riemann's habilitation lecture Über die Hypothesen introduces the concept of a manifold with a metric (Riemannian metric) and the notion of curvature, though without an explicit connection.

- 1869 — Elwin Bruno Christoffel introduces the symbols

\{{}^k_{ij}\}

(Christoffel symbols) while studying the equivalence problem for quadratic differential forms; the "covariant differentiation" formula appears implicitly.

- 1887–1896 — Gregorio Ricci-Curbastro develops the absolute differential calculus (tensor calculus), systematizing covariant derivatives.

- 1900 — Ricci and Levi-Civita publish Méthodes de calcul différentiel absolu et leurs applications in Mathematische Annalen, the foundational treatise on tensor calculus.

- 1915 — Albert Einstein, using the Ricci calculus, completes the general theory of relativity; the metric connection is central though not yet named.

- 1916 — Hermann Weyl and others begin abstracting the notion of "affine connection" beyond the metric case.

- 1917 — Levi-Civita publishes the concept of parallel transport (parallelismo), giving the connection its geometric meaning; nearly simultaneously Jan Schouten develops a similar idea.

- 1918–1923 — Hermann Weyl and Élie Cartan generalize connections (Weyl geometry, Cartan connections with torsion), clarifying by contrast that Levi-Civita's is the unique torsion-free metric one.

- 1963 — Kobayashi and Nomizu's Foundations of Differential Geometry gives the modern bundle-theoretic and Koszul-formula treatment that is standard today.

open-problem

Where the clean theory frays at the edges:

Low regularity. The fundamental theorem assumes a smooth (or at least $C^2$ ) metric. For metrics of low regularity (e.g. $C^{0,1}$ Lipschitz, $W^{1,p}$ , or merely continuous), the Christoffel symbols may exist only distributionally. Defining geodesics, curvature, and the connection for non-smooth / singular spaces (Alexandrov spaces, RCD spaces, gravitational shock waves) is an active program where Levi-Civita's pointwise formula breaks down and must be replaced by synthetic or measure-theoretic notions.

Degenerate metrics. The uniqueness proof crucially uses non-degeneracy of $g$ to invert the Koszul functional. On degenerate (e.g. lightlike/null) hypersurfaces and in Carrollian/Galilean geometries, no unique Levi-Civita connection exists, and one must work with families of compatible connections — a live area in non-relativistic holography and BMS symmetry.

Torsion as physics. Whether spacetime torsion is identically zero is an empirical question, not a mathematical theorem. Einstein–Cartan theory and supergravity admit torsion sourced by spin; current experiments bound but do not exclude it, leaving open whether nature truly selects the torsion-free Levi-Civita connection.

Infinite dimensions. On infinite-dimensional manifolds (loop spaces, spaces of metrics, diffeomorphism groups with weak Riemannian metrics) the Koszul formula need not define a connection — the metric may fail to be "strong," and existence of the Levi-Civita connection can fail. Characterizing when it exists (e.g. for the $L^2$ metric vs. Sobolev metrics on shape spaces) remains subtle.

Formal Statement

Why It Holds / Derivation

Building Intuition

Worked Examples

Common Misconceptions

Connections

Real-World Applications

Historical Development

Subtleties & Limits

See Also