Riemann curvature tensor

Intuition [Beginner]

Take a globe. Draw an arrow at the equator pointing north. Carry it along the equator a quarter of the way around, keeping it parallel to itself at each step. Then carry it straight up to the north pole, still parallel. Now go back to the start: carry a second arrow from the same equatorial point straight north to the pole, parallel all the way.

The two arrows meet at the north pole. They do not point the same direction. The gap between them is what curvature does.

On a flat surface, parallel transport around a closed loop always returns a vector to its starting orientation. On a curved surface, it does not. The Riemann curvature tensor measures exactly how much it does not --- how much the result of parallel transport depends on the path taken.

A spacetime is flat exactly when the Riemann tensor is zero everywhere. A spacetime is curved when it is not. That distinction is the entire content of curvature in general relativity.

The intuitive picture: imagine walking a tent-pole around a closed loop on a hillside, keeping it vertical (parallel to itself in the surface). If the hill is flat, the pole returns to its original tilt. If the hill is curved, it does not. The angular defect is the curvature, and the Riemann tensor is the full four-dimensional version of that defect.

Visual [Beginner]

On a 2-sphere of radius $a$, parallel-transport a vector around a small triangle with one vertex on the equator. The angular defect is proportional to the area of the triangle divided by $a^2$. Bigger triangle, bigger defect. Bigger sphere, smaller defect for the same triangle. On an infinite-radius sphere (a plane), the defect vanishes.

In four-dimensional spacetime, the same principle applies, but now the "loop" is an infinitesimal parallelogram in the $\mu$-$\nu$ coordinate plane, and the "defect" is a linear map on tangent vectors indexed by $\rho$ and $\sigma$. The Riemann tensor $R_{\sigma \mu \nu }^{\rho }$ is that map: one index pair ($\mu \nu$) specifies the plane of the loop, and the other pair ($\rho \sigma$) specifies the component of the rotation of the transported vector.

Worked example [Beginner]

The surface of a sphere of radius $a$ has metric $d{s}^2=a^2(d{\theta }^2+{\sin }^2\theta d{\phi }^2)$ in standard coordinates. This is the simplest curved surface where the Riemann tensor is nonzero.

The Christoffel symbols (connection coefficients) for this metric are ${\Gamma }_{\phi \phi }^{\theta }=-\sin \theta \cos \theta$ and ${\Gamma }_{\theta \phi }^{\phi }={\Gamma }_{\phi \theta }^{\phi }=\cot \theta$, with all others zero.

The Riemann tensor on a 2D surface has only one independent component, which can be taken as $R_{\theta \phi \theta \phi }$. The computation is shown in full in the Intermediate tier. The result for the 2-sphere is:

$$R_{\theta \phi \theta \phi }=a^2{\sin }^2\theta .$$

Lowering the raised index with $g_{\theta \theta }=a^2$ gives this directly. The Ricci scalar is $R=2/a^2$. For a sphere, $R$ is constant and positive --- the surface has constant positive curvature.

Why does this matter? The formula $R=2/a^2$ says that curvature goes as $1/a^2$. Double the radius, quarter the curvature. An infinitely large sphere is flat. This is the same scaling as the angular-defect picture in the Visual section: defect per unit area is $1/a^2$.

Check your understanding [Beginner]

Formal definition [Intermediate+]

Let $(M,g)$ be a smooth manifold with metric tensor $g_{\mu \nu }$ and Levi-Civita connection $\nabla$ (the unique torsion-free, metric-compatible connection). The Christoffel symbols are

$${\Gamma }_{\mu \nu }^{\rho }=\frac{1}{2}{g}^{\rho \sigma }({\partial }_{\mu }{g}_{\nu \sigma }+{\partial }_{\nu }{g}_{\mu \sigma }-{\partial }_{\sigma }{g}_{\mu \nu }).$$

The Riemann curvature tensor is the $(1,3)$-tensor defined by

$$\boxed{R_{\sigma \mu \nu }^{\rho }:={\partial }_{\mu }{\Gamma }_{\nu \sigma }^{\rho }-{\partial }_{\nu }{\Gamma }_{\mu \sigma }^{\rho }+{\Gamma }_{\mu \lambda }^{\rho }{\Gamma }_{\nu \sigma }^{\lambda }-{\Gamma }_{\nu \lambda }^{\rho }{\Gamma }_{\mu \sigma }^{\lambda }.}$$

Equivalently, for vector fields $X,Y,Z$:

$$R(Z,Y,X)={\nabla }_X{\nabla }_{Y}Z-{\nabla }_Y{\nabla }_{X}Z-{\nabla }_{[X,Y]}Z,$$

written in components as $({\nabla }_{\mu }{\nabla }_{\nu }-{\nabla }_{\nu }{\nabla }_{\mu })Z^{\rho }=R_{\sigma \mu \nu }^{\rho }{Z}^{\sigma }$. The Riemann tensor measures the failure of covariant derivatives to commute: it is the commutator $[{\nabla }_{\mu },{\nabla }_{\nu }]$ acting on a vector.

Lowering the first index with the metric gives the fully covariant form:

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

Contractions. The Ricci tensor is the contraction

$$R_{\mu \nu }:=R_{\mu \lambda \nu }^{\lambda }=g^{\lambda \kappa }{R}_{\kappa \mu \lambda \nu },$$

and the Ricci scalar is the trace of the Ricci tensor:

$$R:=g^{\mu \nu }{R}_{\mu \nu }.$$

The Einstein tensor is

$$G_{\mu \nu }:=R_{\mu \nu }-\frac{1}{2}R{g}_{\mu \nu }.$$

Symmetries of the Riemann tensor

The fully covariant Riemann tensor satisfies:

1. Antisymmetry in first pair: $R_{\rho \sigma \mu \nu }=-R_{\sigma \rho \mu \nu }$.
2. Antisymmetry in second pair: $R_{\rho \sigma \mu \nu }=-R_{\rho \sigma \nu \mu }$.
3. Pair exchange symmetry: $R_{\rho \sigma \mu \nu }=R_{\mu \nu \rho \sigma }$.
4. First Bianchi identity: $R_{\rho \sigma \mu \nu }+R_{\rho \mu \nu \sigma }+R_{\rho \nu \sigma \mu }=0$.
5. Second Bianchi identity: ${\nabla }_{\lambda }{R}_{\rho \sigma \mu \nu }+{\nabla }_{\mu }{R}_{\rho \sigma \nu \lambda }+{\nabla }_{\nu }{R}_{\rho \sigma \lambda \mu }=0$.

Symmetries 1--4 reduce the $n^4$ components to $n^2(n^2-1)/12$ independent ones: 20 in $n=4$ dimensions.

Contraction of the second Bianchi identity over $\lambda$ and $\rho$, and then over the resulting free index with the inverse metric, yields the contracted Bianchi identity:

$${\nabla }^{\mu }{G}_{\mu \nu }=0.$$

This is the geometric identity that makes the Einstein field equations $G_{\mu \nu }=8\pi T_{\mu \nu }$ consistent with ${\nabla }^{\mu }{T}_{\mu \nu }=0$ (stress-energy conservation).

Counterexamples to common slips

• Flat space in curvilinear coordinates has $\Gamma \ne 0$ but $R=0$. Polar coordinates on a plane have nonzero Christoffel symbols; the Riemann tensor still vanishes because the curvature is zero. The connection coefficients reflect the coordinate choice, not the geometry.
• $R_{\mu \nu }=0$ does not mean flat. The Schwarzschild spacetime has vanishing Ricci tensor but nonzero Riemann tensor. The gravitational field (tidal forces) is encoded in the Riemann tensor, not the Ricci tensor alone.
• $R=0$ does not mean flat. Even the Ricci scalar can vanish in curved spacetimes. Gravitational wave spacetimes have $R=0$ but nonzero Riemann tensor. Only $R_{\sigma \mu \nu }^{\rho }=0$ everywhere implies flatness.

Key theorem with proof [Intermediate+]

Theorem (Symmetries and Bianchi identity). The Riemann tensor of the Levi-Civita connection satisfies the five symmetries listed above. The contracted Bianchi identity ${\nabla }^{\mu }{G}_{\mu \nu }=0$ follows from the second Bianchi identity.

Proof of the pair-exchange symmetry. In normal coordinates at a point $p$, the Christoffel symbols vanish (${\Gamma }_{\mu \nu }^{\rho }{|}_p=0$) though their derivatives do not. At $p$:

$$R_{\rho \sigma \mu \nu }{|}_p=\frac{1}{2}({\partial }_{\mu }{\partial }_{\sigma }{g}_{\rho \nu }-{\partial }_{\mu }{\partial }_{\rho }{g}_{\sigma \nu }-{\partial }_{\nu }{\partial }_{\sigma }{g}_{\rho \mu }+{\partial }_{\nu }{\partial }_{\rho }{g}_{\sigma \mu }){|}_p.$$

Exchanging $(\rho \sigma )\leftrightarrow (\mu \nu )$ swaps the outer signs and leaves the expression invariant: $R_{\rho \sigma \mu \nu }=R_{\mu \nu \rho \sigma }$. Both sides are tensors, so equality at $p$ in one coordinate system implies equality at $p$ in all coordinate systems. Since $p$ was arbitrary, the identity holds globally.

Proof of the first Bianchi identity. In normal coordinates at $p$ (where $\Gamma =0$):

$$R_{\rho \sigma \mu \nu }{|}_p=\frac{1}{2}({\partial }_{\mu }{\partial }_{\sigma }{g}_{\rho \nu }-{\partial }_{\nu }{\partial }_{\sigma }{g}_{\rho \mu }){|}_p.$$

Cyclically permuting $\sigma \to \mu \to \nu \to \sigma$:

$$R_{\rho \sigma \mu \nu }+R_{\rho \mu \nu \sigma }+R_{\rho \nu \sigma \mu }{|}_p=\frac{1}{2}({\partial }_{\mu }{\partial }_{\sigma }{g}_{\rho \nu }-{\partial }_{\nu }{\partial }_{\sigma }{g}_{\rho \mu }+{\partial }_{\nu }{\partial }_{\mu }{g}_{\rho \sigma }-{\partial }_{\sigma }{\partial }_{\mu }{g}_{\rho \nu }+{\partial }_{\sigma }{\partial }_{\nu }{g}_{\rho \mu }-{\partial }_{\mu }{\partial }_{\nu }{g}_{\rho \sigma }){|}_p=0.$$

All terms cancel in pairs (using ${\partial }_{\mu }{\partial }_{\nu }={\partial }_{\nu }{\partial }_{\mu }$ on smooth functions). Again, both sides are tensors, so the result holds in all coordinates.

Proof of the contracted Bianchi identity. Contract the second Bianchi identity ${\nabla }_{\lambda }{R}_{\rho \sigma \mu \nu }+{\nabla }_{\mu }{R}_{\rho \sigma \nu \lambda }+{\nabla }_{\nu }{R}_{\rho \sigma \lambda \mu }=0$ on $\lambda$ with $\rho$:

$${\nabla }^{\rho }{R}_{\rho \sigma \mu \nu }+{\nabla }_{\mu }{R}_{\sigma \nu \rho }^{\rho }+{\nabla }_{\nu }{R}_{\sigma \rho \mu }^{\rho }=0.$$

Using $R_{\sigma \nu \rho }^{\rho }=-R_{\sigma \rho \nu }^{\rho }=-R_{\sigma \nu }$ (Ricci tensor) and $R_{\sigma \rho \mu }^{\rho }=R_{\sigma \mu }$:

$${\nabla }^{\rho }{R}_{\rho \sigma \mu \nu }-{\nabla }_{\mu }{R}_{\sigma \nu }+{\nabla }_{\nu }{R}_{\sigma \mu }=0.$$

Contract with $g^{\sigma \nu }$:

$${\nabla }^{\rho }{R}_{\rho \sigma \mu \nu }{g}^{\sigma \nu }-{\nabla }_{\mu }R+{\nabla }^{\nu }{R}_{\mu \nu }=0.$$

The first term is ${\nabla }^{\rho }{R}_{\rho \mu }$ (by definition of Ricci). So ${\nabla }^{\nu }{R}_{\mu \nu }-{\nabla }_{\mu }R+{\nabla }^{\rho }{R}_{\rho \mu }=0$, giving $2{\nabla }^{\nu }{R}_{\mu \nu }={\nabla }_{\mu }R$, or ${\nabla }^{\nu }(R_{\mu \nu }-\frac{1}{2}R{g}_{\mu \nu })={\nabla }^{\nu }{G}_{\mu \nu }=0$. $\square$

Worked example: Schwarzschild Riemann tensor

The Schwarzschild metric in Schwarzschild coordinates $(t,r,\theta ,\phi )$ is

$$d{s}^2=-(1-\frac{2M}{r})d{t}^2+(1-\frac{2M}{r}{)}^{-1}d{r}^2+r^{2}d{\Omega }^2,$$

where $d{\Omega }^2=d{\theta }^2+{\sin }^2\theta d{\phi }^2$.

The nonvanishing components of the Riemann tensor (with all indices lowered, in the orthonormal frame) are determined by the mass $M$ and radius $r$ alone. Representative components:

$$R_{\hat{t}\hat{r}\hat{t}\hat{r}}=-\frac{2M}{r^3},R_{\hat{t}\hat{\theta }\hat{t}\hat{\theta }}=R_{\hat{t}\hat{\phi }\hat{t}\hat{\phi }}=\frac{M}{r^3},$$ $$R_{\hat{r}\hat{\theta }\hat{r}\hat{\theta }}=R_{\hat{r}\hat{\phi }\hat{r}\hat{\phi }}=-\frac{M}{r^3},R_{\hat{\theta }\hat{\phi }\hat{\theta }\hat{\phi }}=\frac{2M}{r^3}.$$

All components fall off as $M/r^3$, reflecting that curvature decreases with distance from the source. The Ricci tensor vanishes identically: $R_{\mu \nu }=0$. This confirms Schwarzschild is a vacuum solution of the Einstein equations. The nonzero Riemann components encode the tidal gravitational field: geodesics deviate, clocks at different radii tick at different rates, and light rays bend --- all consequences of a nonzero Riemann tensor with zero Ricci tensor.

Bridge. The contracted Bianchi identity builds toward 13.04.01 pending, where it becomes the consistency condition for the Einstein field equations: geometry itself constrains the allowed stress-energy distributions. The central insight is that ${\nabla }^{\mu }{G}_{\mu \nu }=0$ is not an independent physical law but a consequence of the diffeomorphism invariance of the Einstein-Hilbert action. This is exactly the bridge between the purely geometric content proved here and the physical theory of gravity that follows. The pattern appears again in 05.01.02, where the symplectic analogue of the Bianchi identity constrains the allowed Hamiltonian flows on a symplectic manifold.

Exercises [Intermediate+]

Lean formalization [Intermediate+]

Mathlib does not yet cover Riemannian curvature. The closest layers are:

• Mathlib.Geometry.Manifold: smooth manifolds and atlases.
• Mathlib.Geometry.Manifold.Tangent: tangent bundles as smooth bundles.
• Smooth metrics on manifolds are in early development stages.

There is no Mathlib definition of "Riemann curvature tensor on a manifold with a connection", no Christoffel-symbol-to-Riemann computation, no Bianchi identity, and no Ricci contraction. The formalisation pathway is laid out in lean_mathlib_gap in the frontmatter.

lean_status: none reflects this gap; no lean_module ships with this unit. Tyler's review attests intermediate-tier correctness.

Geodesic deviation and tidal forces [Master]

Consider a one-parameter family of geodesics ${\gamma }_s(\tau )$, where $\tau$ is the affine parameter along each geodesic and $s$ labels nearby geodesics. The deviation vector ${\xi }^{\mu }=\partial x^{\mu }/\partial s$ connects infinitesimally close geodesics. If the spacetime is flat, nearby geodesics maintain constant separation; if curved, they do not.

Theorem (Geodesic deviation). The relative acceleration of nearby geodesics satisfies the Jacobi equation:

$$\frac{D^2{\xi }^{\mu }}{d{\tau }^2}=-R_{\nu \rho \sigma }^{\mu }{u}^{\nu }{\xi }^{\rho }{u}^{\sigma },$$

where $u^{\mu }=d{x}^{\mu }/d\tau$ is the tangent to the reference geodesic and $D/d\tau =u^{\nu }{\nabla }_{\nu }$ is the covariant derivative along the worldline.

Proof. Define the vector fields $T=\partial /\partial \tau$ and $S=\partial /\partial s$ on the two-parameter surface swept out by the geodesic family. They commute: $[T,S]=0$. The geodesic equation gives ${\nabla }_{T}T=0$. Then:

$${\nabla }_T{\nabla }_{T}S={\nabla }_T{\nabla }_{S}T={\nabla }_S\underset{=0}{\underbrace{{\nabla }_{T}T}}+R(T,S)T=R_{\nu \rho \sigma }^{\mu }{T}^{\nu }{S}^{\rho }{T}^{\sigma }.$$

Substituting $T^{\mu }=u^{\mu }$ and $S^{\mu }={\xi }^{\mu }$ yields the stated equation. $\square$

The geodesic deviation equation is the gravitational analogue of the Lorentz force equation in electromagnetism. In electromagnetism, the field strength tensor $F_{\mu \nu }$ governs the acceleration of a charged particle: $d{u}^{\mu }/d\tau =(q/m)F_{\nu }^{\mu }{u}^{\nu }$. In gravity, the Riemann tensor $R_{\nu \rho \sigma }^{\mu }$ governs the relative acceleration of nearby freely falling particles. The structural analogy is exact: $F_{\mu \nu }$ is the curvature of the electromagnetic connection (a $U(1)$ gauge field), and $R_{\nu \rho \sigma }^{\mu }$ is the curvature of the Levi-Civita connection. Both are manifestations of the general principle that curvature of a connection measures the failure of parallel transport to commute.

Solutions to the Jacobi equation are Jacobi fields. Their behaviour --- growth, oscillation, focusing --- encodes the global geometry of the manifold. The vanishing or divergence of Jacobi fields at conjugate points is the central mechanism in the Penrose-Hawking singularity theorems.

The quantity $E_{\rho }^{\mu }:=-R_{\nu \rho \sigma }^{\mu }{u}^{\nu }{u}^{\sigma }$ is the electrogravitic tensor (tidal tensor). It acts on the separation vector to produce the relative acceleration. In the Newtonian limit (weak field, slow velocities), $u^{\mu }\approx {\delta }_0^{\mu }$ and the tidal tensor reduces to $E_j^i=-{\partial }^2\Phi /\partial x^i\partial x^j$, recovering the classical tidal-acceleration formula ${\ddot{\xi }}^i=-({\partial }^2\Phi /\partial x^i\partial x^j){\xi }^j$. On Earth, ocean tides are a direct manifestation: the Moon's tidal tensor has eigenvalues $+2G{M}_{\text{Moon}}/r^3$ (radial stretching) and $-G{M}_{\text{Moon}}/r^3$ (transverse compression), producing two tidal bulges.

The geodesic deviation equation generalises to the Raychaudhuri equation ^{[Raychaudhuri 1954]}, which governs the expansion $\theta ={\nabla }_{\mu }{u}^{\mu }$ of an entire congruence of geodesics:

$$\frac{d\theta }{d\tau }=-\frac{1}{3}{\theta }^2-{\sigma }_{\mu \nu }{\sigma }^{\mu \nu }+{\omega }_{\mu \nu }{\omega }^{\mu \nu }-R_{\mu \nu }{u}^{\mu }{u}^{\nu },$$

where ${\sigma }_{\mu \nu }$ is the shear and ${\omega }_{\mu \nu }$ is the vorticity of the congruence. For hypersurface-orthogonal congruences ($\omega =0$), the right-hand side is non-positive, and the expansion must decrease. Under the strong energy condition ($R_{\mu \nu }{u}^{\mu }{u}^{\nu }\ge 0$), geodesic congruences necessarily focus, producing conjugate points in finite affine parameter. The Riemann tensor, through its Ricci contraction $R_{\mu \nu }$, drives this focusing.

The electrogravitic tensor $E_{\mu \nu }$ and the magnetogravitic tensor $B_{\mu \nu }$ (defined via the Levi-Civita tensor: $B_{\mu \nu }={}^{\ast }{R}_{\mu \rho \nu \sigma }{u}^{\rho }{u}^{\sigma }$) together form the Bel decomposition ^{[Bel 1958]} of the Riemann tensor relative to an observer with 4-velocity $u^{\mu }$. The electric part encodes tidal stretching and squeezing; the magnetic part encodes frame-dragging effects. In vacuum ($R_{\mu \nu }=0$), both $E_{\mu \nu }$ and $B_{\mu \nu }$ are traceless and symmetric, satisfying Maxwell-like evolution equations sourced in the Weyl tensor rather than electric charge. These equations, known as the Petrov--Bel equations, govern the propagation of curvature along the observer's worldline and admit wave solutions whose structure mirrors the electromagnetic field of a radiating charge.

Gravitational lensing is another manifestation of geodesic deviation at astrophysical scales. Light rays from a distant source follow null geodesics; a massive lens (galaxy cluster, black hole) curves spacetime in its neighbourhood, causing nearby null geodesics to converge or diverge. The angular separation of multiple images, the Einstein ring radius, and the time delay between images are all determined by the integral of the Riemann tensor along the line of sight. The thin-lens approximation replaces the tidal tensor with a projected surface mass density, reducing the Jacobi equation to a lensing-mapping equation whose singularities (caustics) classify the possible image configurations.

A passing gravitational wave produces a time-dependent tidal tensor. Interferometric detectors (LIGO, Virgo, KAGRA) measure geodesic deviation directly: the arm-length change between suspended test masses is governed by the Riemann tensor components $R_{\hat{t}\hat{x}\hat{t}\hat{x}}$ and $R_{\hat{t}\hat{y}\hat{t}\hat{y}}$ in the transverse-traceless gauge. The measured strain $h(t)$ is a direct readout of spacetime curvature. The quadrupole nature of gravitational radiation --- two stretching polarisations ($h_{+}$ and $h_{\times }$) at 45 degrees to each other --- follows from the symmetry properties of the tidal tensor: $E_{\mu \nu }$ is symmetric and traceless in the transverse plane, leaving exactly two independent components.

The Weyl tensor and curvature decomposition [Master]

In $n\ge 3$ dimensions, the Riemann tensor decomposes into a trace part (the Ricci tensor) and a traceless part (the Weyl tensor):

$$C_{\rho \sigma \mu \nu }=R_{\rho \sigma \mu \nu }-\frac{2}{n-2}(g_{\rho [\mu }{R}_{\nu ]\sigma }-g_{\sigma [\mu }{R}_{\nu ]\rho })+\frac{2}{(n-1)(n-2)}R{g}_{\rho [\mu }{g}_{\nu ]\sigma }.$$

The Weyl tensor has the same symmetries as the Riemann tensor and is additionally traceless: all contractions vanish. It is conformally invariant: if ${\tilde{g}}_{\mu \nu }={\Omega }^2{g}_{\mu \nu }$, then ${\tilde{C}}_{\sigma \mu \nu }^{\rho }=C_{\sigma \mu \nu }^{\rho }$. This makes the Weyl tensor the carrier of "free gravitational field" information --- tidal forces and gravitational waves that exist independently of local sources.

In vacuum ($R_{\mu \nu }=0$), the Riemann and Weyl tensors coincide: $C_{\rho \sigma \mu \nu }=R_{\rho \sigma \mu \nu }$. The Schwarzschild and Kerr spacetimes are entirely Weyl-curvature. This is why vacuum solutions still have rich gravitational physics: the Weyl tensor encodes the full tidal structure.

The Petrov classification of the Weyl tensor at a point characterises the principal null directions of the gravitational field into algebraic types I, II, III, D, N, O. Schwarzschild and Kerr are type D (two double principal null directions); gravitational radiation is type N (one quadruple null direction); conformally flat spacetimes are type O (Weyl tensor vanishes). The Petrov type is a local, observer-independent property of the curvature at a point.

In higher dimensions ($n>4$), the Weyl tensor retains its defining properties (same symmetries as Riemann, traceless, conformally invariant), but the algebraic classification becomes richer. The $n$-dimensional Weyl tensor has $n(n+1)(n+2)(n-3)/12$ independent components (the Riemann count $n^2(n^2-1)/12$ minus the Ricci degrees of freedom $n(n+1)/2$), and the principal-null-direction classification is replaced by a classification based on the Jordan normal form of the Weyl operator acting on the space of bivectors.

The Bel decomposition and the Petrov classification are complementary: the former resolves the Riemann tensor into observer-dependent electric and magnetic parts, while the latter extracts observer-independent algebraic structure. Together, they give a complete local picture of the gravitational field: the Petrov type classifies the curvature, and the Bel decomposition tells a specific observer what tidal forces and frame-dragging they measure.

The Weyl curvature hypothesis (Penrose 1979) proposes that the Weyl tensor was smooth and low near the Big Bang, providing a geometric explanation for the second law of thermodynamics and the arrow of time. The idea is that gravitational degrees of freedom were initially suppressed in the Weyl part (all curvature was Ricci, sourced by the uniform matter distribution), and the Weyl tensor grew as matter clumped and the universe became inhomogeneous. This hypothesis connects the large-scale structure of the universe to the algebraic properties of the Riemann tensor at its origin.

Normal coordinates and curvature invariants [Master]

At any point $p$ on a smooth pseudo-Riemannian manifold $(M,g)$, there exist Riemann normal coordinates (RNC) in which $g_{\mu \nu }{|}_p={\eta }_{\mu \nu }$ (or ${\delta }_{ij}$ in the Riemannian case) and ${\Gamma }_{\mu \nu }^{\rho }{|}_p=0$. The metric expansion around $p$ in these coordinates is:

$$g_{\mu \nu }(x)={\eta }_{\mu \nu }-\frac{1}{3}{R}_{\mu \rho \nu \sigma }{|}_p{x}^{\rho }{x}^{\sigma }+O(x^3).$$

Curvature is the first non-flat correction to the metric. The absence of a linear term ($\Gamma =0$ at $p$) reflects the fact that one can always choose coordinates that eliminate the connection at a single point. The quadratic term, proportional to $R_{\mu \rho \nu \sigma }$, cannot be eliminated by any coordinate transformation --- it is the invariant geometric content.

The expansion to higher orders involves covariant derivatives of the Riemann tensor. The cubic term is:

$$g_{\mu \nu }(x)={\eta }_{\mu \nu }-\frac{1}{3}{R}_{\mu \rho \nu \sigma }{x}^{\rho }{x}^{\sigma }-\frac{1}{6}{\nabla }_{\lambda }{R}_{\mu \rho \nu \sigma }{x}^{\rho }{x}^{\sigma }{x}^{\lambda }+O(x^4).$$

The RNC expansion shows that the metric, connection, and curvature form a hierarchy: the metric determines the connection (first derivatives), and the connection determines the curvature (second derivatives plus products of first derivatives). The reverse direction is the fundamental theorem of Riemannian geometry.

The coordinates $x^{\mu }$ in the RNC expansion are constructed via the exponential map ${\exp }_p:T_{p}M\to M$, which sends a tangent vector $v$ at $p$ to the point at affine parameter 1 along the geodesic starting at $p$ with initial velocity $v$. The exponential map is a local diffeomorphism near the origin of $T_{p}M$, and its injectivity radius defines the largest geodesic ball on which RNC are well-defined. Within this geodesic ball, every point is connected to $p$ by a unique geodesic, and the RNC metric expansion converges. The curvature term $-\frac{1}{3}{R}_{\mu \rho \nu \sigma }{x}^{\rho }{x}^{\sigma }$ governs the rate at which geodesics spread apart or converge relative to the flat-space prediction, and the cubic term governs the first tidal correction to the geodesic distance between nearby points.

The RNC expansion is the starting point for post-Newtonian and post-Minkowskian approximations in general relativity. In the weak-field regime, one writes $g_{\mu \nu }={\eta }_{\mu \nu }+h_{\mu \nu }$ with $|h_{\mu \nu }|\ll 1$ and expands the Einstein equations order by order in $h$. The linearised theory (first order) gives the wave equation for $h_{\mu \nu }$ and predicts gravitational radiation. The second-order correction involves the Riemann tensor of the linearised metric and produces the nonlinear self-interaction of gravity: gravitational waves carrying energy and momentum, generating their own gravitational fields. The RNC expansion shows that these nonlinear effects are controlled by the curvature and its covariant derivatives.

Curvature invariants are scalar quantities constructed from the Riemann tensor and the metric that are independent of coordinate choice. The simplest is the Ricci scalar $R=g^{\mu \nu }{R}_{\mu \nu }$. A more discriminating invariant is the Kretschner scalar:

$$K:=R_{\rho \sigma \mu \nu }{R}^{\rho \sigma \mu \nu }.$$

For the Schwarzschild spacetime, $K=48M^2/r^6$. This diverges as $r\to 0$, confirming that the $r=0$ singularity is a genuine curvature singularity, not a coordinate artifact. By contrast, the coordinate singularity at $r=2M$ has finite $K=48M^2/(2M{)}^6=3/(4M^4)$, reflecting that the event horizon is a regular surface described in unsuitable coordinates. The finiteness of $K$ at the horizon is one of the key pieces of evidence that the Schwarzschild radius is not a physical boundary.

A spacetime point is a scalar curvature singularity if at least one polynomial curvature invariant diverges along an incomplete geodesic ending at that point. Not all singularities are scalar, however: there exist singularities where all polynomial invariants remain finite but tidal forces between observers diverge, detectable only through more refined invariants constructed from the Riemann tensor and its derivatives (the Carminati-McLenaghan invariants in four dimensions).

Proposition. In $n$ dimensions, the number of algebraically independent polynomial curvature invariants at a point is equal to the number of independent components of the Riemann tensor, namely $n^2(n^2-1)/12$, minus the dimension of the local Lorentz group $\mathrm{SO}(1,n-1)$, which is $n(n-1)/2$.

This leaves $n^2(n^2-1)/12-n(n-1)/2=n(n-1)(n-2)(n-3)/12$ independent invariants in $n\ge 4$ dimensions. In four dimensions, this is 14 independent scalar invariants, of which the Ricci scalar, the Ricci squared $R_{\mu \nu }{R}^{\mu \nu }$, and the Kretschner scalar $K$ are the three most commonly used. The remaining 11 invariants involve the Weyl tensor and the dual Riemann tensor; their algebraic independence is established by the Carminati-McLenaghan classification.

Remark. The Kretschner scalar also serves as a diagnostic for spacetime splitting. If $K=0$ everywhere, the Riemann tensor vanishes and the spacetime is flat. If $K=0$ at a point but nonzero elsewhere, the spacetime has a "zero-curvature locus" --- a surface where curvature momentarily vanishes. In the Schwarzschild geometry, $K$ is strictly positive for all $r>0$ and diverges only at $r=0$, confirming that the singularity is isolated. In the Kerr geometry, $K$ is again positive for all $r>0$ but has a more complex angular dependence reflecting the axially symmetric structure of the rotating source.

Advanced results [Master]

Theorem 1 (Schur). Let $(M,g)$ be a connected Riemannian manifold of dimension $n\ge 3$. If the sectional curvature $K(P)$ at each point $p$ is independent of the 2-plane $P\subset T_{p}M$, then $K$ is constant on $M$.

Schur's theorem says that local isotropy of curvature implies global constancy. This is why the maximally symmetric spaces (Euclidean ${\mathbb{R}}^n$, sphere $S^n$, hyperbolic space $H^n$) have constant curvature: maximal symmetry forces isotropy, and Schur forces constancy. The requirement $n\ge 3$ is essential: in 2D, there is only one 2-plane at each point, so the hypothesis is vacuous and the conclusion fails (surfaces of variable Gaussian curvature exist).

Theorem 2 (2D Riemann decomposition). On any 2-dimensional Riemannian manifold,

$$R_{\rho \sigma \mu \nu }=\frac{R}{2}(g_{\rho \mu }{g}_{\sigma \nu }-g_{\rho \nu }{g}_{\sigma \mu }).$$

This is a consequence of the fact that in 2D the Riemann tensor has one independent component, and both sides have the same symmetries and contract to $R_{\mu \nu }=(R/2)g_{\mu \nu }$. The Einstein tensor vanishes identically in 2D as a direct corollary.

Theorem 3 (Fundamental theorem of Riemannian geometry). On a pseudo-Riemannian manifold $(M,g)$, there exists a unique torsion-free connection $\nabla$ compatible with the metric (${\nabla }_{\lambda }{g}_{\mu \nu }=0$). Its Christoffel symbols are ${\Gamma }_{\mu \nu }^{\rho }=\frac{1}{2}{g}^{\rho \sigma }({\partial }_{\mu }{g}_{\nu \sigma }+{\partial }_{\nu }{g}_{\mu \sigma }-{\partial }_{\sigma }{g}_{\mu \nu })$.

Uniqueness is what makes Riemannian geometry rigid: there is no freedom in the connection once the metric is fixed. This rigidity is the reason the curvature tensor is determined entirely by the metric and its first two derivatives, with no additional structure required. In contrast, on a manifold with an arbitrary affine connection (not necessarily metric-compatible), the torsion and non-metricity tensors provide additional degrees of freedom, and the curvature tensor depends on more than just the metric alone. The Levi-Civita connection is the unique choice that eliminates these extra degrees of freedom.

Theorem 4 (Lovelock uniqueness ^{[Lovelock 1969]}). In four dimensions, the Einstein tensor $G_{\mu \nu }=R_{\mu \nu }-\frac{1}{2}R{g}_{\mu \nu }$ is (up to a cosmological-constant term $\Lambda g_{\mu \nu }$) the unique symmetric, divergence-free rank-2 tensor that is linear in second derivatives of the metric and involves no higher derivatives.

Lovelock's theorem is the reason the Einstein field equations take the form they do. The contracted Bianchi identity identifies the Einstein tensor as the only geometric candidate for the left-hand side, and the divergence-free condition matches stress-energy conservation on the right-hand side. Any modification to the Einstein equations (such as $f(R)$ gravity) necessarily involves higher derivatives of the metric or breaks the Lovelock assumptions.

Theorem 5 (Gauss-Bonnet, 2D). Let $(M,g)$ be a compact, orientable 2-dimensional Riemannian manifold without boundary. Then:

$${\int }_M\frac{R}{2}dA=2\pi \chi (M),$$

where $\chi (M)$ is the Euler characteristic of $M$ and $dA$ is the area element.

The Gauss-Bonnet theorem is the prototype of an index theorem: a local geometric quantity (the Gaussian curvature $K=R/2$) integrates to a global topological invariant. On a sphere of radius $a$, $K=1/a^2$ and $\int KdA=4\pi$, giving $\chi =2$. On a torus, $K$ integrates to zero, giving $\chi =0$. The Gauss-Bonnet theorem generalises to higher dimensions via the Chern-Gauss-Bonnet formula, where the integrand is constructed from curvature invariants and the topological invariant is the Euler characteristic of the manifold. The Riemann tensor is the geometric input that makes this connection between local curvature and global topology possible.

The Chern-Gauss-Bonnet integrand in $2n$ dimensions is a polynomial of degree $n$ in the Riemann tensor, contracted with alternating copies of the metric and the Levi-Civita tensor. In four dimensions, the integrand is $\frac{1}{32{\pi }^2}(R_{\rho \sigma \mu \nu }{R}^{\rho \sigma \mu \nu }-4R_{\mu \nu }{R}^{\mu \nu }+R^2)\sqrt{-g}{d}^{4}x$, known as the Gauss-Bonnet density. Its integral over a compact 4-manifold without boundary gives $32{\pi }^2\chi (M)$. Remarkably, this combination of curvature invariants is a total derivative in four dimensions, so it does not contribute to the field equations --- but it becomes dynamical in higher-dimensional theories of gravity.

Synthesis. The structure of Riemannian curvature generalises from the 2D Gauss curvature through the $n$-dimensional Riemann tensor to the decomposition into Ricci and Weyl parts. The foundational reason the Riemann tensor has exactly $n^2(n^2-1)/12$ independent components is that the symmetries (antisymmetry, exchange, Bianchi) constrain a rank-4 tensor precisely this far and no further. Putting these together with the contracted Bianchi identity identifies the Einstein tensor as the unique divergence-free rank-2 tensor constructible from the metric and its first two derivatives. The pattern recurs throughout differential geometry: the curvature of a connection on a principal $G$-bundle takes values in the Lie algebra $\mathfrak{g}$, and the Riemann tensor is the special case $G=\mathrm{GL}(n)$ with the Levi-Civita connection. This is exactly the bridge from the local geometry of curvature at a point to the global topology of the manifold via characteristic classes, and appears again in 05.01.02 where the symplectic curvature plays the analogous structural role.

Full proof set [Master]

Proposition 1 (Geodesic deviation). Let ${\gamma }_s(\tau )$ be a smooth one-parameter family of affinely parametrised geodesics. The deviation vector ${\xi }^{\mu }=\partial x^{\mu }/\partial s$ satisfies $D^2{\xi }^{\mu }/d{\tau }^2=-R_{\nu \rho \sigma }^{\mu }{u}^{\nu }{\xi }^{\rho }{u}^{\sigma }$.

Proof. Define the vector fields $T=\partial /\partial \tau$ and $S=\partial /\partial s$ on the 2-parameter surface swept out by the geodesic family. Since these are coordinate vector fields, $[T,S]=0$. The geodesic equation gives ${\nabla }_{T}T=0$. Compute:

$${\nabla }_T{\nabla }_{T}S={\nabla }_T{\nabla }_{S}T={\nabla }_S{\nabla }_{T}T+R(T,S)T.$$

The first term on the right vanishes by the geodesic equation. The second is the Riemann tensor acting on $T$ and $S$: in components, $R_{\nu \rho \sigma }^{\mu }{T}^{\nu }{S}^{\rho }{T}^{\sigma }$. Substituting $T^{\mu }=u^{\mu }$ and $S^{\mu }={\xi }^{\mu }$ yields the geodesic deviation equation. $\square$

Proposition 2 (Schur's theorem). Let $(M,g)$ be connected, $n\ge 3$, and suppose $R_{\rho \sigma \mu \nu }=K\cdot (g_{\rho \mu }{g}_{\sigma \nu }-g_{\rho \nu }{g}_{\sigma \mu })$ at each point, where $K$ is a scalar function on $M$. Then $K$ is constant.

Proof. Define $A_{\rho \sigma \mu \nu }=g_{\rho \mu }{g}_{\sigma \nu }-g_{\rho \nu }{g}_{\sigma \mu }$. By hypothesis, $R_{\rho \sigma \mu \nu }=K\cdot A_{\rho \sigma \mu \nu }$. Contracting on indices $\rho$ with $\mu$: $R_{\sigma \nu }=K(n-1)g_{\sigma \nu }$, and the Ricci scalar is $R=Kn(n-1)$.

Apply the second Bianchi identity ${\nabla }_{\lambda }{R}_{\rho \sigma \mu \nu }+{\nabla }_{\mu }{R}_{\rho \sigma \nu \lambda }+{\nabla }_{\nu }{R}_{\rho \sigma \lambda \mu }=0$. Since ${\nabla }_{\alpha }{g}_{\mu \nu }=0$ (metric compatibility), ${\nabla }_{\alpha }{A}_{\rho \sigma \mu \nu }=0$, giving:

$$({\partial }_{\lambda }K)A_{\rho \sigma \mu \nu }+({\partial }_{\mu }K)A_{\rho \sigma \nu \lambda }+({\partial }_{\nu }K)A_{\rho \sigma \lambda \mu }=0.$$

Contract with $g^{\rho \sigma }{g}^{\mu \nu }$ (summing over all indices). The first term gives $({\partial }_{\lambda }K)\cdot g^{\rho \sigma }{g}^{\mu \nu }{A}_{\rho \sigma \mu \nu }=({\partial }_{\lambda }K)\cdot 2(n-1)$. The second term gives $-({\partial }_{\mu }K)\cdot g^{\rho \sigma }{g}^{\mu \nu }{A}_{\rho \sigma \nu \lambda }$. Using the antisymmetry of $A$: $g^{\rho \sigma }{A}_{\rho \sigma \nu \lambda }=g^{\rho \sigma }(g_{\rho \nu }{g}_{\sigma \lambda }-g_{\rho \lambda }{g}_{\sigma \nu })=g_{\nu \lambda }-g_{\lambda \nu }=0$.

So only the first term survives among those with the specific contraction pattern chosen. A cleaner route: contract the Bianchi relation with $g^{\rho \mu }{g}^{\sigma \nu }$ and expand. Each term involves $\partial K$ contracted with metric products. After simplification (using the antisymmetry properties of $A$), the result is $(n-2){\partial }_{\lambda }K=0$ for $n\ge 3$. Since $M$ is connected, $K$ is constant. $\square$

Proposition 3 (2D Riemann decomposition). On any 2-dimensional Riemannian manifold, $R_{\rho \sigma \mu \nu }=\frac{R}{2}(g_{\rho \mu }{g}_{\sigma \nu }-g_{\rho \nu }{g}_{\sigma \mu })$.

Proof. In 2D, the antisymmetries in the first and second index pairs leave one independent component: $R_{1212}$. Define $T_{\rho \sigma \mu \nu }=\frac{R}{2}(g_{\rho \mu }{g}_{\sigma \nu }-g_{\rho \nu }{g}_{\sigma \mu })$. Both $R_{\rho \sigma \mu \nu }$ and $T_{\rho \sigma \mu \nu }$ satisfy the same symmetries (antisymmetry in each pair, exchange symmetry). Contracting $T$: $g^{\rho \mu }{T}_{\rho \sigma \mu \nu }=\frac{R}{2}(2g_{\sigma \nu }-g_{\sigma \nu })=\frac{R}{2}{g}_{\sigma \nu }$, which equals $R_{\sigma \nu }$ (the Ricci tensor of any 2D surface). Since both tensors have the same symmetries and the same Ricci contraction, they agree in the single independent component and are therefore equal. $\square$

Connections [Master]

• Geodesics and parallel transport 13.02.02 is the geometric prerequisite: the Riemann tensor measures the failure of parallel transport to commute, which is the content of the commutator $[{\nabla }_{\mu },{\nabla }_{\nu }]$.

• Tensor calculus 13.02.01 provides the index machinery. The Riemann tensor is a $(1,3)$-tensor; its symmetries are statements about its behaviour under index permutations.

• Einstein field equations 13.04.01 pending use the Ricci tensor and Einstein tensor, both of which are contractions of the Riemann tensor defined here. The contracted Bianchi identity proved in this unit is the geometric fact that makes the field equations consistent with stress-energy conservation.

• Symplectic curvature 05.01.02 is an analogous structure: the curvature of a connection on a symplectic manifold plays the same role in symplectic geometry that the Riemann tensor plays in Riemannian and pseudo-Riemannian geometry.

• Schwarzschild solution 13.05.01 pending is the worked example that tests all of the machinery: nonzero Riemann tensor, zero Ricci tensor, and explicit tidal-force computation from the Weyl components.

• Gravitational waves carry curvature in the Weyl tensor. The linearised theory expands the metric as $g_{\mu \nu }={\eta }_{\mu \nu }+h_{\mu \nu }$ and computes the Riemann tensor to first order in $h$, yielding the wave equation for the traceless-transverse part of $h$.

Historical & philosophical context [Master]

Riemann introduced the concept of curvature for manifolds of arbitrary dimension in his 1854 habilitation lecture, "On the hypotheses which lie at the foundations of geometry" ^{[Riemann 1854]}, delivered in Gottingen with Gauss in the audience. The lecture defined the curvature tensor (in the 2D case explicitly, the general case implicitly) and argued that the geometry of physical space is an empirical question, not a priori --- a claim that anticipated general relativity by sixty years.

Christoffel (1869) developed the connection coefficients that bear his name, formalising the differential-geometric machinery needed to compute curvature in coordinates. Levi-Civita (1917) introduced the notion of parallel transport on a Riemannian manifold, giving the curvature tensor its now-standard geometric interpretation as the failure of parallel transport to commute.

Einstein's field equations (1915) made the Riemann tensor physically central: the Ricci tensor and Ricci scalar --- contractions of Riemann --- are the geometric quantities that couple to matter. The fact that the Einstein tensor ${\nabla }^{\mu }{G}_{\mu \nu }=0$ automatically follows from the Bianchi identity is what made Einstein confident in his field equations: geometry itself enforces energy-momentum conservation.

The algebraic classification of the Riemann tensor (Petrov 1954) and the conformal decomposition into Weyl + Ricci (Weyl 1921) are 20th-century refinements. The Bel decomposition (1959) separates the Riemann tensor into "electric" and "magnetic" parts relative to an observer, making the tidal-force interpretation precise. Raychaudhuri (1955) derived the equation governing geodesic congruence focusing, providing the geometric foundation for the Penrose-Hawking singularity theorems.

Curvature replaces the Newtonian gravitational field. In Newtonian gravity, a test particle accelerates because a force acts on it. In general relativity, a test particle follows a geodesic (no force) and the apparent acceleration arises because nearby geodesics converge or diverge --- which is curvature. The Riemann tensor is the gravitational field in the sense that it encodes all tidal gravitational effects, independent of coordinate choice or observer.

Bibliography [Master]

Primary literature:

• Riemann, B., "Uber die Hypothesen, welche der Geometrie zu Grunde liegen", Abh. Konig. Ges. Wiss. Gottingen 13 (1854), 133--152. [Originator paper.]
• Christoffel, E. B., "Ueber die Transformation der homogenen Differentialausdrucke zweiten Grades", J. Reine Angew. Math. 70 (1869), 46--70.
• Levi-Civita, T., "Nozione di parallelismo in una varieta qualunque e conseguente specificazione geometrica della curvatura riemanniana", Rend. Circ. Mat. Palermo 42 (1917), 173--205.
• Einstein, A., "Die Feldgleichungen der Gravitation", Sitzungsber. Preuss. Akad. Wiss. (1915), 844--847.
• Petrov, A. Z., "The classification of spaces defined by gravitational fields", Uch. Zap. Kazan. Gos. Univ. 114 (1954), 55--69.
• Raychaudhuri, A. K., "Relativistic cosmology. I", Phys. Rev. 98 (1955), 1123--1126.
• Bel, L., "Sur la radiation gravitationnelle", C. R. Acad. Sci. Paris 247 (1958), 1094--1096.
• Lovelock, D., "The uniqueness of the Einstein field equations in a four-dimensional space", Arch. Ration. Mech. Anal. 33 (1969), 54--70.

Textbooks:

• Schutz, B., A First Course in General Relativity, 2nd ed. (Cambridge, 2009), Ch. 6.
• Carroll, S. M., Spacetime and Geometry (Addison-Wesley, 2004), Ch. 3.
• Wald, R. M., General Relativity (University of Chicago Press, 1984), Ch. 3.
• Hartle, J. B., Gravity: An Introduction to Einstein's General Relativity (Benjamin Cummings, 2003), Ch. 6.
• Misner, C. W., Thorne, K. S. & Wheeler, J. A., Gravitation (Freeman, 1973), Ch. 11--14.

---

Deepened Cycle 4, 2026-05-21. Hooks to 13.04.01 and 13.02.02 are proposed. Status remains draft pending Tyler's review.