Curvature as the Commutator of Covariant Derivatives

On a flat plane in Cartesian coordinates, partial derivatives commute: $\partial_\mu\partial_\nu f = \partial_\nu\partial_\mu f$ , and the order in which you differentiate never matters. The covariant derivative $\nabla_\mu$ promotes $\partial_\mu$ to curved manifolds and to tensor fields, adding connection terms that compensate for the twisting of the coordinate basis. It is natural to ask whether this upgraded derivative keeps the commutativity of the flat one it replaced. It does not — and the exact amount by which $\nabla_\mu$ and $\nabla_\nu$ fail to commute, when they act on a vector field, is precisely the Riemann curvature tensor. This is the most economical definition of curvature in all of geometry, and one of the most powerful. It is manifestly tensorial in a way that looks almost too good: although $\nabla_\mu V^\rho$ is itself a derivative of $V$ , the antisymmetrized double derivative $[\nabla_\mu,\nabla_\nu]V^\rho$ contains no derivatives of $V$ at all. Every term carrying $\partial V$ cancels, leaving a pointwise multiplicative operator built only from the connection and its first derivatives. Because that operator produces a genuine tensor out of an arbitrary vector, its coefficient $R^\rho{}_{\sigma\mu\nu}$ must be a $(1,3)$ tensor field. The same one line pays for a great deal at once. The antisymmetry of the commutator in $\mu\nu$ hands us $R^\rho{}_{\sigma\mu\nu}=-R^\rho{}_{\sigma\nu\mu}$ for free. Acting on higher-rank tensors gives one curvature term per index with a fixed sign rule; acting on a scalar gives zero for a torsion-free connection. And it is the algebraic twin of a geometric fact — parallel transport around a small closed loop rotates a vector by an amount set by Riemann. The commutator is the infinitesimal generator of that loop. From here the Ricci tensor, the scalar curvature, the Einstein tensor, and the Bianchi identities all follow.

Key takeaways

The Riemann tensor is the failure of second covariant derivatives to commute on a vector: $[\nabla_\mu,\nabla_\nu]V^\rho = R^\rho{}_{\sigma\mu\nu}V^\sigma$ , with all derivatives of $V$ cancelling.
Unlike partial derivatives, covariant derivatives generically do not commute on a curved manifold; the obstruction is local, tensorial, and equal to the curvature.
The commutator picks out curvature, not the connection: in normal coordinates the Christoffel symbols vanish at a point, yet their antisymmetrized derivatives — the Riemann tensor — need not.
Each index of an arbitrary tensor contributes its own term: $+R$ per upper index, $-R$ per lower index; a scalar gives zero, and a torsion term appears only if the connection is not symmetric.
Algebra equals geometry: this is the same object as the holonomy generated by parallel transport around an infinitesimal loop, and it makes Riemann's symmetries and both Bianchi identities transparent.
Key idea: the Riemann tensor is the operator-valued curvature of the connection — $[\nabla_\mu,\nabla_\nu]$ acting on any tensor is purely multiplicative (no surviving derivatives), its coefficient on each index is $\pm R$ , and geometrically it is the holonomy generator for infinitesimal parallel-transport loops.

The core insight

The Riemann tensor is the operator-valued curvature of the connection: $[\nabla_\mu,\nabla_\nu]$ acting on any tensor field is purely multiplicative (no surviving derivatives), and its coefficient on each index is $\pm R$ . Geometrically it is the holonomy generator for infinitesimal parallel-transport loops.

Where this lives today

The commutator-Riemann identity is not a museum piece; it is the working definition of curvature in every modern gravitational calculation. Gravitational-wave detectors read it out directly: a passing wave is, in the transverse-traceless frame, an oscillating set of Riemann components $R_{\hat i\hat 0\hat j\hat 0}$ , and geodesic deviation — the commutator applied along a congruence — is literally what stretches and squeezes the arms of LIGO and Virgo by one part in $10^{21}$ . Numerical relativity evolves the full Riemann tensor (through the metric and connection) to simulate black-hole mergers, and the Weyl part of Riemann encodes the outgoing radiation extracted at large radius. The gauge-theory analogy $[D_\mu,D_\nu]=F_{\mu\nu}$ makes the same structure the organizing principle of the tetrad and Palatini formulations, where gravity is a gauge theory of the Lorentz connection and holonomies of that connection govern how spinors are parallel-transported — the curved-spacetime Dirac equation. The holonomy idea itself has become a computational tool: loop quantum gravity is built from holonomies of the connection around loops, and the same Berry-phase mathematics appears when electrons are transported around loops in a crystal's band structure. Across all of it, the one sentence "curvature is the commutator of covariant derivatives" is the definition being computed.

The physical picture

Carry a vector by parallel transport along two edges of a tiny coordinate parallelogram, then back along the other two edges. On a flat plane the vector returns exactly as it left. On a curved surface it comes back rotated, and the rotation depends only on the loop and the curvature it encloses. The commutator $[\nabla_\mu,\nabla_\nu]$ is precisely the infinitesimal version of "transport around the loop, minus the identity"; its value on a vector is the rotation that loop induces. This is why the order of two covariant derivatives matters — differentiating along $\mu$ then $\nu$ transports along one pair of edges, and the reversed order transports along the other pair. The first figure shows exactly this mismatch, and names the object it defines.

Figure

Order matters: parallel transporting a vector V along δxμ then δxν versus δxν then δxμ leaves a mismatch ΔVρ = −Rρσμν Vσ δaμν, proportional to the enclosed area — the commutator of covariant derivatives is the Riemann curvature

Why must the result be tensorial with no leftover derivatives? Because $\nabla_\mu\nabla_\nu V$ contains a top term $\partial_\mu\partial_\nu V$ that is symmetric in $\mu\nu$ , plus mixed pieces $\Gamma\,\partial V$ that are also symmetric once the dummy indices are relabelled. Antisymmetrizing — subtracting the swapped order — annihilates every one of them. What survives multiplies $V$ with no derivative attached, which is the signature of a curvature. This is exactly the structure of a gauge field strength, $F_{\mu\nu}=[D_\mu,D_\nu]$ : the Riemann tensor is the curvature of the Levi-Civita connection just as $F_{\mu\nu}$ is the curvature of the gauge connection, with the Lorentz group acting on the tangent space in place of the internal symmetry group. The cleanest place to see the geometry become a number is the round sphere. A vector carried around a closed loop on a sphere returns rotated by the solid angle the loop encloses; for a geodesic triangle the rotation equals the angle excess, which equals the enclosed area times the Gaussian curvature. The second figure works the sharpest case — an octant of the unit sphere, three right angles — where the holonomy is exactly a quarter turn.

Figure

Parallel transport around a geodesic triangle — an octant of the unit sphere with three right angles — rotates the vector by the holonomy angle Δθ = π/2 = 90°, equal to the enclosed area (π/2) times the Gaussian curvature K = 1/a² = 1

The core equations

The whole subject rests on a single identity: acting on a contravariant vector, the antisymmetrized pair of covariant derivatives is a multiplication by the Riemann tensor. No derivatives of the vector survive, so the coefficient is a genuine $(1,3)$ tensor. This is the definition of curvature we will unpack.

[\nabla_\mu,\nabla_\nu]\,V^\rho = R^\rho{}_{\sigma\mu\nu}\,V^\sigma \qquad (1)

where: $\nabla_\mu$ — covariant derivative of the Levi-Civita connection; $V^\rho$ — an arbitrary contravariant vector field; $R^\rho{}_{\sigma\mu\nu}$ — the Riemann curvature tensor acting here as a multiplicative operator; $[\,,\,]$ — the commutator $\nabla_\mu\nabla_\nu-\nabla_\nu\nabla_\mu$ . Reading it: the left side looks like two derivatives, the right side has none — that collapse is the whole content. Expanding the commutator explicitly gives the Riemann tensor in terms of the connection and its first derivatives. This is the formula you compute with, and it exposes the two structural pieces: a curl of the connection $\partial\Gamma$ and a quadratic $\Gamma\Gamma$ piece.

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} \qquad (2)

where: $\Gamma^\rho{}_{\mu\nu}$ — Christoffel symbols of the second kind (the Levi-Civita connection); $\partial_\mu$ — ordinary coordinate partial derivative. Reading it: manifest antisymmetry in $\mu\nu$ is built in, and this is Carroll's sign convention, defined with $\nabla_\mu\nabla_\nu$ ordered first. The commutator flips sign on a covariant index. Acting on a one-form, the coefficient is $-R$ with the tensor's free index sitting in the second slot. This opposite sign is exactly what keeps scalars consistent.

[\nabla_\mu,\nabla_\nu]\,\omega_\rho = -\,R^\sigma{}_{\rho\mu\nu}\,\omega_\sigma \qquad (3)

where: $\omega_\rho$ — an arbitrary covariant vector (one-form). Reading it: contract a vector with a one-form into a scalar and the $+R$ and $-R$ terms cancel, so the scalar commutes — as it must. For a tensor of any rank the rule is simply additive: one curvature term per index, $+R$ for each upper slot and $-R$ for each lower slot. Nothing else appears for a torsion-free connection.

[\nabla_\mu,\nabla_\nu]\,T^{\rho_1\cdots}{}_{\sigma_1\cdots} = \sum_i R^{\rho_i}{}_{\lambda\mu\nu}\,T^{\cdots\lambda\cdots}{}_{\sigma\cdots} \;-\; \sum_j R^{\lambda}{}_{\sigma_j\mu\nu}\,T^{\rho\cdots}{}_{\cdots\lambda\cdots} \qquad (4)

where: $T^{\rho\cdots}{}_{\sigma\cdots}$ — a tensor of arbitrary rank; each sum runs over the tensor's indices of that type. Reading it: the metric $g_{\mu\nu}$ gives zero here (two lower indices whose $-R$ terms cancel by Riemann's pair symmetry), which is metric compatibility seen from the commutator. On a scalar the commutator does not automatically vanish — it returns the torsion contracted with the gradient. For the symmetric Levi-Civita connection of general relativity the torsion is zero, so scalars commute; but the general statement is more informative.

[\nabla_\mu,\nabla_\nu]\,f = -\,T^\lambda{}_{\mu\nu}\,\nabla_\lambda f,\qquad T^\lambda{}_{\mu\nu} \equiv \Gamma^\lambda{}_{\mu\nu}-\Gamma^\lambda{}_{\nu\mu} \qquad (5)

where: $f$ — a scalar field; $T^\lambda{}_{\mu\nu}$ — the torsion tensor, the antisymmetric part of the connection. Reading it: curvature lives in $\partial\Gamma$ , torsion in the antisymmetric part of $\Gamma$ itself; they are logically independent obstructions. Restoring torsion on a vector gives the fully general commutator, with a curvature term plus a torsion term proportional to one derivative of $V$ . In standard general relativity the second term is absent.

[\nabla_\mu,\nabla_\nu]\,V^\rho = R^\rho{}_{\sigma\mu\nu}\,V^\sigma - T^\lambda{}_{\mu\nu}\,\nabla_\lambda V^\rho \qquad (6)

where: all symbols as above. Reading it: the curvature term multiplies $V$ , the torsion term differentiates it — the two are distinguishable by whether a derivative of $V$ survives, and GR keeps only the first.

Derivation ADeriving the Riemann tensor as the commutator on a vector field

The goal is to show that the second covariant derivative of a vector, once antisymmetrized in its two derivative indices, loses every term carrying a derivative of the vector itself, leaving a purely multiplicative operator whose coefficient is built from the connection and its first derivatives — the Riemann tensor. We compute $\nabla_\mu\nabla_\nu V^\rho$ term by term, being careful that the inner object $\nabla_\nu V^\rho$ is a $(1,1)$ tensor and must be differentiated as one, then subtract the $\mu\leftrightarrow\nu$ swap and read off what remains.

Step 1. Differentiate once: $\nabla_\nu V^\rho = \partial_\nu V^\rho + \Gamma^\rho{}_{\nu\sigma}V^\sigma$ , a rank- $(1,1)$ tensor with a free upper index $\rho$ and a free lower index $\nu$ . Why: the second derivative cannot treat this as a plain vector — the new lower index $\nu$ demands its own connection term in the next step.

Step 2. Differentiate again, giving every index its connection term: $\nabla_\mu\nabla_\nu V^\rho = \partial_\mu(\nabla_\nu V^\rho) + \Gamma^\rho{}_{\mu\lambda}\nabla_\nu V^\lambda - \Gamma^\lambda{}_{\mu\nu}\nabla_\lambda V^\rho$ . Why: the upper index $\rho$ earns a $+\Gamma$ and the lower index $\nu$ earns a $-\Gamma$ ; this is the covariant derivative of a $(1,1)$ tensor, no more and no less.

Step 3. Expand fully into $\partial_\mu\partial_\nu V^\rho + (\partial_\mu\Gamma^\rho{}_{\nu\sigma})V^\sigma + \Gamma^\rho{}_{\nu\sigma}\partial_\mu V^\sigma + \Gamma^\rho{}_{\mu\lambda}\partial_\nu V^\lambda + \Gamma^\rho{}_{\mu\lambda}\Gamma^\lambda{}_{\nu\sigma}V^\sigma - \Gamma^\lambda{}_{\mu\nu}\nabla_\lambda V^\rho$ . Why: sorting the six terms by whether they carry a derivative of $V$ is the whole trick — three of them do, and those are the ones that must die.

Step 4. Subtract the same expression with $\mu$ and $\nu$ swapped. The pure second derivative $\partial_\mu\partial_\nu V^\rho$ is symmetric in $\mu\nu$ and cancels; the term $\Gamma^\lambda{}_{\mu\nu}\nabla_\lambda V^\rho$ is symmetric in $\mu\nu$ (torsion-free $\Gamma$ ) and cancels; the two cross terms $\Gamma^\rho{}_{\nu\sigma}\partial_\mu V^\sigma$ and $\Gamma^\rho{}_{\mu\lambda}\partial_\nu V^\lambda$ are equal after relabelling the dummy index and cancel. Why: every surviving object must therefore multiply $V$ with no derivative attached — that is what guarantees the result is a tensor times $V$ .

Step 5. Collect the survivors: $[\nabla_\mu,\nabla_\nu]V^\rho = \big(\partial_\mu\Gamma^\rho{}_{\nu\sigma} - \partial_\nu\Gamma^\rho{}_{\mu\sigma} + \Gamma^\rho{}_{\mu\lambda}\Gamma^\lambda{}_{\nu\sigma} - \Gamma^\rho{}_{\nu\lambda}\Gamma^\lambda{}_{\mu\sigma}\big)V^\sigma \equiv R^\rho{}_{\sigma\mu\nu}V^\sigma$ . Why: since $V^\sigma$ is arbitrary and the left side is a tensor, the bracketed coefficient is a $(1,3)$ tensor — the Riemann tensor — with no further assumptions.

Go deeper · Derivation B · 1 minThe scalar commutator and the role of torsion

Derivation BThe scalar commutator and the role of torsion

The goal is to isolate torsion as a separate obstruction from curvature by acting with the commutator on a scalar, where no Riemann term can appear because a scalar has no index for curvature to grab. We compute $[\nabla_\mu,\nabla_\nu]f$ for a scalar $f$ and find that what survives is the antisymmetric part of the connection contracted with the gradient — the torsion — which vanishes identically for the Levi-Civita connection, explaining why curvature in GR only manifests on tensors of rank one or higher.

Step 1. On a scalar the covariant derivative is just the partial: $\nabla_\nu f = \partial_\nu f$ , because a scalar has no index for a connection term to attach to. Why: this makes the first derivative trivially symmetric-looking, so any asymmetry must enter at the second step.

Step 2. The gradient $\nabla_\nu f$ is now a one-form, so the second derivative carries a connection term: $\nabla_\mu\nabla_\nu f = \partial_\mu\partial_\nu f - \Gamma^\lambda{}_{\mu\nu}\,\partial_\lambda f$ . Why: even though we started with a scalar, the first derivative produced a one-form, and one-forms do feel the connection.

Step 3. Antisymmetrize: $[\nabla_\mu,\nabla_\nu]f = -(\Gamma^\lambda{}_{\mu\nu}-\Gamma^\lambda{}_{\nu\mu})\,\partial_\lambda f = -T^\lambda{}_{\mu\nu}\,\partial_\lambda f$ . Why: the symmetric $\partial_\mu\partial_\nu f$ cancels, and only the antisymmetric part of $\Gamma$ — the torsion — remains, with no curvature in sight.

Step 4. For the Levi-Civita connection $\Gamma^\lambda{}_{\mu\nu}=\Gamma^\lambda{}_{\nu\mu}$ , so $T^\lambda{}_{\mu\nu}=0$ and $[\nabla_\mu,\nabla_\nu]f = 0$ . Why: this is why second covariant derivatives of scalars commute in GR, and why the curvature of spacetime can only be detected by transporting a vector or higher tensor, never a scalar.

Go deeper · Derivation C · 1 minParallel transport around an infinitesimal loop

Derivation CParallel transport around an infinitesimal loop

The goal is to connect the algebraic commutator to its geometric twin — the change a vector suffers when parallel transported around a small closed loop — and to show they are the same tensor. We transport $V$ around a coordinate parallelogram spanned by two small displacements, keep terms to second order in the edge lengths (the leading order that survives after the loop closes), and identify the coefficient of the enclosed area with the Riemann tensor of equation (2).

Step 1. Parallel transport along a leg $\delta x^\mu$ shifts the vector by $\delta V^\rho = -\Gamma^\rho{}_{\mu\sigma}V^\sigma\,\delta x^\mu$ , the statement that the covariant derivative along the path is zero. Why: this first-order rule is the definition of "carrying the vector without turning it" relative to the connection.

Step 2. Go around the parallelogram — out along $\delta x^\mu$ , up along $\delta x^\nu$ , back along $-\delta x^\mu$ , down along $-\delta x^\nu$ — accumulating the shift order by order. The first-order pieces cancel because the loop closes; the surviving change is second order, one factor from each pair of opposite edges. Why: a closed loop can only produce an effect at the order where the two directions interfere, which is precisely the antisymmetric combination the commutator computes.

Step 3. The net change is $\delta V^\rho = -R^\rho{}_{\sigma\mu\nu}\,V^\sigma\,\delta a^{\mu\nu}$ , where $\delta a^{\mu\nu}=\delta x^\mu\delta y^\nu-\delta x^\nu\delta y^\mu$ is the oriented area of the loop and the coefficient is exactly the Riemann tensor of equation (2). Why: the same $\partial\Gamma$ and $\Gamma\Gamma$ combination that emerged from the commutator reappears here as the holonomy, so the two constructions define one object.

Step 4. Shrinking the loop, the holonomy per unit area is the commutator: $[\nabla_\mu,\nabla_\nu]V^\rho = R^\rho{}_{\sigma\mu\nu}V^\sigma$ is the infinitesimal generator of loop transport. Why: this is why "curvature" and "non-commuting derivatives" and "holonomy" are three names for a single tensor.

Limiting cases

How the commutator behaves in the important special cases:

Case	Result	Meaning
Flat space	$R^\rho{}_{\sigma\mu\nu}=0 \iff [\nabla_\mu,\nabla_\nu]=0$ on all fields	vanishing curvature is equivalent to all covariant derivatives commuting; the manifold is then locally isometric to flat space even if the chart looks curved
Acting on a scalar	$[\nabla_\mu,\nabla_\nu]f = -T^\lambda{}_{\mu\nu}\nabla_\lambda f$	for the torsion-free Levi-Civita connection this is zero, so second covariant derivatives of scalars are symmetric
Connection with torsion	$[\nabla_\mu,\nabla_\nu]V^\rho = R^\rho{}_{\sigma\mu\nu}V^\sigma - T^\lambda{}_{\mu\nu}\nabla_\lambda V^\rho$	a general affine connection adds a term linear in $\nabla V$ ; GR sets torsion to zero, dropping it
Maximally symmetric space	$R_{\rho\sigma\mu\nu}=\dfrac{R}{n(n-1)}\,(g_{\rho\mu}g_{\sigma\nu}-g_{\rho\nu}g_{\sigma\mu})$	for spheres, hyperbolic space, and (anti-)de Sitter the commutator's coefficient is fixed entirely by the constant scalar curvature $R$ in $n$ dimensions
Unit 2-sphere	$R^\theta{}_{\phi\theta\phi}=\sin^2\theta$ , $\;K=1$ , $\;R=2$	the single independent component; transporting a vector around a loop rotates it by the enclosed solid angle

Common trapThe commutator measures the derivative of the connection, not the connection

The trap is to think that because the Christoffel symbols can be made to vanish at a point (Riemann normal coordinates), the curvature vanishes there too. It does not. The Riemann tensor of equation (2) contains $\partial\Gamma$ , the first derivatives of the connection, which are generally nonzero at that point even when $\Gamma$ itself is zero — only the quadratic $\Gamma\Gamma$ pieces drop out. Curvature is a coordinate-invariant tensor; if it is nonzero in one chart it is nonzero in all of them, and no choice of coordinates can transform it away at a point. A second, independent trap is sign and index conventions: always check whether your source orders the commutator as $\nabla_\mu\nabla_\nu$ or $\nabla_\nu\nabla_\mu$ first, and where it places the derivative indices, because Wald, MTW, and Carroll do not all agree and the difference is an overall sign.

Worked examples

1 · The 2-sphere, from Christoffel symbols to a 90° holonomy (fully worked)

Take the unit 2-sphere with metric $ds^2 = d\theta^2 + \sin^2\theta\,d\phi^2$ . Compute its one independent Riemann component directly from equation (2), confirm the commutator interpretation, and use it to predict the rotation of a vector parallel-transported around an octant.

Step 1. The nonzero Christoffel symbols are $\Gamma^\theta{}_{\phi\phi}=-\sin\theta\cos\theta$ and $\Gamma^\phi{}_{\theta\phi}=\Gamma^\phi{}_{\phi\theta}=\cot\theta$ . These come from the metric in the usual way; everything else vanishes.

Step 2. Evaluate $R^\theta{}_{\phi\theta\phi} = \partial_\theta\Gamma^\theta{}_{\phi\phi} - \partial_\phi\Gamma^\theta{}_{\theta\phi} + \Gamma^\theta{}_{\theta\lambda}\Gamma^\lambda{}_{\phi\phi} - \Gamma^\theta{}_{\phi\lambda}\Gamma^\lambda{}_{\theta\phi}$ . The first term is $\partial_\theta(-\sin\theta\cos\theta)=\sin^2\theta-\cos^2\theta$ ; the second and third vanish; the fourth is $-\Gamma^\theta{}_{\phi\phi}\Gamma^\phi{}_{\theta\phi}=-(-\sin\theta\cos\theta)(\cot\theta)=\cos^2\theta$ . Summing gives $R^\theta{}_{\phi\theta\phi}=\sin^2\theta - \cos^2\theta + \cos^2\theta = \mathbf{\sin^2\theta}$ .

Step 3. Lower the index: $R_{\theta\phi\theta\phi}=g_{\theta\theta}R^\theta{}_{\phi\theta\phi}=\sin^2\theta$ . The Gaussian curvature is $K=R_{\theta\phi\theta\phi}/(g_{\theta\theta}g_{\phi\phi})=\sin^2\theta/(1\cdot\sin^2\theta)=\mathbf{1}$ , and the scalar curvature $R=2K=\mathbf{2}$ — the standard values for the unit sphere.

Step 4. The commutator reading: on the sphere $[\nabla_\theta,\nabla_\phi]V^\theta = R^\theta{}_{\phi\theta\phi}V^\phi = \sin^2\theta\,V^\phi \neq 0$ , so covariant derivatives genuinely fail to commute — the direct algebraic fingerprint of curvature.

Step 5. Geometry as a number: a vector parallel-transported around the boundary of a geodesic triangle rotates by the angle excess, which equals $\int K\,dA = K\times(\text{area})$ . An octant of the unit sphere is a triangle with three right angles, so its excess is $\tfrac{3\pi}{2}-\pi=\tfrac{\pi}{2}$ and its area is $\tfrac{\pi}{2}$ . The holonomy is therefore Δθ = π/2 = 90°, exactly the second figure — and it is the finite integral of the infinitesimal commutator around the loop.

2 · Tidal stretch as a Riemann component (order-of-magnitude)

Geodesic deviation, $\frac{D^2\xi^\rho}{d\tau^2}=-R^\rho{}_{\sigma\mu\nu}u^\sigma\xi^\mu u^\nu$ , is derived by commuting covariant derivatives along a congruence of geodesics — the same algebra as equation (1). Use it to estimate the tidal Riemann component at Earth's surface and the relative acceleration across a standing person.

Step 1. In the Newtonian limit the relevant curvature component in an orthonormal frame is $R^{\hat 1}{}_{\hat 0\hat 1\hat 0} \approx \dfrac{1}{c^2}\partial_{\hat 1}\partial_{\hat 1}\Phi = -\dfrac{2GM}{c^2 r^3}$ for the radial direction outside a spherical mass $M$ , with $\Phi=-GM/r$ .

Step 2. Insert Earth's numbers, $M=5.97\times10^{24}$ kg, $r=6.37\times10^6$ m, $G=6.67\times10^{-11}$ : the combination $2GM/r^3 = \mathbf{3.08\times10^{-6}\ s^{-2}}$ , so the Riemann component is $R^{\hat 1}{}_{\hat 0\hat 1\hat 0}=(2GM/r^3)/c^2 \approx \mathbf{3.4\times10^{-23}\ m^{-2}}$ — the curvature of spacetime you are standing in, in natural geometric units.

Step 3. Geodesic deviation turns this into a tidal acceleration: for a head-to-foot separation $\xi\approx1.8$ m the radial stretch is $\ddot\xi = (2GM/r^3)\,\xi \approx (3.08\times10^{-6})(1.8) \approx \mathbf{5.5\times10^{-6}\ m/s^2}$ — about $6\times10^{-7}\,g$ , utterly imperceptible on Earth but growing as $1/r^3$ until, near a stellar-mass black hole, the same component spaghettifies. The point is structural: a physically measurable tidal force is a component of the tensor that the commutator defines.

Physics of the portrait

The identity that curvature is the commutator of covariant derivatives — and its geometric face, parallel transport around a loop — was not born with the Riemann tensor in 1854. It took the parallel-transport idea of the early twentieth century to turn Riemann's algebraic object into a picture of holonomy, and no one did more to build that bridge than Tullio Levi-Civita.

Origins · 1 minTullio Levi-Civita and how this was discovered

Tullio Levi-Civita

1873–1941 · absolute differential calculus and the geometry of parallel transport

Tullio Levi-Civita was the Italian mathematician who gave curvature its modern geometric meaning. Bernhard Riemann had introduced the curvature tensor in his 1854 Göttingen habilitation lecture as an intrinsic measure of how a manifold departs from flatness, and Elwin Christoffel wrote down the connection symbols in 1869, but the machinery remained formal. Working with his teacher Gregorio Ricci-Curbastro, Levi-Civita systematized these ideas into the absolute differential calculus — tensor analysis — in their 1900 memoir, the toolkit Einstein would seize on to build general relativity in 1915. Levi-Civita's own decisive stroke came in 1917, when he defined parallel transport: a rule for carrying a vector along a curve without turning it relative to the connection. This exposed the meaning hidden in Riemann's formula — the failure of a vector to return unchanged after transport around a closed loop, the holonomy, is exactly the curvature, and the commutator of covariant derivatives is its infinitesimal generator. He corresponded with Einstein, who wrote that he admired Levi-Civita's methods more than the theory of relativity itself. Dismissed from his Rome chair in 1938 under the Fascist racial laws, he died in 1941, but the connection that carries his name — the unique torsion-free, metric-compatible connection whose curvature is the Riemann tensor — is the backbone of every calculation in gravitational physics.

History

1854

— Riemann introduces the curvature tensor in his Göttingen habilitation lecture

1869

— Christoffel writes down the connection symbols and the covariant derivative

1900

— Ricci-Curbastro and Levi-Civita systematize the absolute differential calculus

1915

— Einstein builds general relativity on the Riemann tensor and its contractions

1917

— Levi-Civita defines parallel transport, making curvature the holonomy of a loop

Going further

Texts. Misner, Thorne & Wheeler, Gravitation (1973), §11.3–11.4 · Wald, General Relativity (1984), §3.2 · Carroll, Spacetime and Geometry (2004), §3.6–3.7 · Schutz, A First Course in General Relativity (2009), §6.5 · Hartle, Gravity (2003), Ch. 21–22 · Nakahara, Geometry, Topology and Physics (2003), §7.3. Key results to keep in mind: the full antisymmetrized expansion of $[\nabla_\mu,\nabla_\nu]V^\rho$ has all $\partial V$ terms cancelling; the scalar result is $[\nabla_\mu,\nabla_\nu]f=-T^\lambda{}_{\mu\nu}\partial_\lambda f$ ; the unit-2-sphere component is $R^\theta{}_{\phi\theta\phi}=\sin^2\theta$ , giving $K=1$ , $R=2$ , and octant holonomy $\pi/2$ ; and the Earth tidal estimate gives $2GM/r^3=3.08\times10^{-6}\ \mathrm{s^{-2}}$ . All signs follow the $\nabla_\mu\nabla_\nu$ -first convention; readers using alternate index orderings should adjust accordingly.

Prerequisites. Covariant Derivative and Christoffel Symbols · Riemann Curvature Tensor · Tensors and Transformation Laws. Leads to. Geometric Meaning of Curvature via Parallel Transport · Algebraic Symmetries of the Riemann Tensor · First (Algebraic) Bianchi Identity · Second (Differential) Bianchi Identity · Ricci Tensor.

Ask the tutor: ask me to expand any single line of the vector-commutator cancellation, or to compute the Riemann tensor and holonomy for a metric of your choice.

Why does the commutator have no derivatives of V, even though each ∇ is a derivative?

Because the highest term, $\partial_\mu\partial_\nu V^\rho$ , is symmetric in $\mu\nu$ and cancels when you subtract the swapped order, and the mixed $\Gamma\,\partial V$ cross terms are also symmetric after relabelling the dummy index, so they cancel too. What is left multiplies $V^\sigma$ with no derivative attached. That is precisely why the result can be a tensor times $V$ : if any $\partial V$ survived, the object would depend on how $V$ varies nearby and could not be a pointwise $(1,3)$ tensor. The cancellation is not a coincidence but the defining feature of a curvature.

If the Christoffel symbols vanish at a point, how can the curvature not vanish there?

The connection and its curvature are different objects. In Riemann normal coordinates centred on a point you can always arrange $\Gamma=0$ at that point, which kills the quadratic $\Gamma\Gamma$ terms there. But the Riemann tensor also contains $\partial\Gamma$ , the first derivatives of the connection, and those do not vanish just because $\Gamma$ does — a function can be zero at a point while its derivative is not. Curvature is a genuine tensor, so if it is nonzero it is nonzero in every coordinate system; no clever chart can remove it at a point. This is the mathematical statement of the equivalence principle: you can erase gravity locally (first derivatives), but not tidal forces (second derivatives).

Why does a scalar commute but a vector does not?

A scalar carries no index for the curvature to act on. When you feed the commutator a scalar, the only thing that can survive is the antisymmetric part of the connection contracted with the gradient — the torsion — and for the Levi-Civita connection of GR the torsion is zero, so scalars commute. A vector, by contrast, has an index that the curvature grabs: the Riemann tensor rotates the vector's components, and that rotation is exactly the holonomy of transporting it around a loop. So the difference is geometric — a scalar has no direction to be rotated, while a vector does.

How is this the same as the field strength Fμν of a gauge theory?

Very literally. In a gauge theory the covariant derivative is $D_\mu=\partial_\mu+A_\mu$ , and its commutator is the field strength: $[D_\mu,D_\nu]=F_{\mu\nu}$ . In gravity the covariant derivative is $\nabla_\mu=\partial_\mu+\Gamma_\mu$ , and its commutator is the Riemann tensor. Both are curvatures of a connection; the only difference is the symmetry group — an internal group like $SU(3)$ for the gauge case, the Lorentz group acting on the tangent space for gravity. This is why gravity can be written as a gauge theory of the Lorentz connection (the tetrad and Palatini formulations), and why the same holonomy language describes Aharonov-Bohm phases and parallel transport of vectors on a manifold.

Why does each index get its own ±R term, with plus for upper and minus for lower?

Because the covariant derivative treats upper and lower indices with opposite-sign connection terms ( $+\Gamma$ for upper, $-\Gamma$ for lower), and the commutator inherits that pattern. Running the vector derivation for each index independently gives a separate Riemann term per index, additive because the second derivative acts linearly on the product of index structures. The signs are fixed by consistency: contract an upper index against a lower one to form a scalar, and the $+R$ and $-R$ must cancel so the scalar commutes. The same bookkeeping shows the metric $g_{\mu\nu}$ gives zero under the commutator, which is metric compatibility read off from curvature.

What is the difference between curvature and torsion, since both come from the connection?

They are independent obstructions living in different parts of the connection. Torsion is the antisymmetric part of $\Gamma$ itself, $T^\lambda{}_{\mu\nu}=\Gamma^\lambda{}_{\mu\nu}-\Gamma^\lambda{}_{\nu\mu}$ ; it measures whether infinitesimal parallelograms fail to close. Curvature is built from the derivatives of $\Gamma$ (plus a quadratic piece); it measures whether parallel transport around a closed loop rotates a vector. You can have one without the other: a space can be flat but have torsion (teleparallel gravity), or curved but torsion-free (standard GR). The commutator separates them cleanly — the torsion term always carries a derivative of the transported field, while the curvature term multiplies it.

How does the infinitesimal commutator relate to a finite loop like the sphere octant?

The commutator is the holonomy per unit area in the limit of a vanishingly small loop: transport a vector around an infinitesimal parallelogram of area $\delta a^{\mu\nu}$ and it changes by $-R^\rho{}_{\sigma\mu\nu}V^\sigma\delta a^{\mu\nu}$ . To get a finite loop you integrate this over the enclosed surface — a path-ordered exponential in general, but for the round sphere it collapses to a clean result. Around an octant of the unit sphere the accumulated rotation is exactly $\pi/2$ , equal to the enclosed area times the constant curvature $K=1$ . So the 90° holonomy of the second figure is the finite integral of the same tensor the commutator defines pointwise.