Geodesic Deviation Equation

Stand inside a freely falling elevator and, by the equivalence principle, you can find local inertial coordinates in which the Christoffel symbols vanish and every test particle obeys the law of inertia: gravity, locally, has been switched off. What no coordinate choice can switch off is the relative motion of neighbouring free-fallers. Release two marbles side by side above the Earth and they drift toward each other, because each aims at the planet's centre; release one above the other and they drift apart, because the lower falls in a stronger field. These tidal effects are the true, coordinate-independent signature of gravity — the thing that survives when the uniform field has been transformed away, and the geodesic deviation equation is their exact relativistic expression. It takes a smooth family of geodesics, measures the second covariant derivative of the vector connecting one to its neighbour, and finds the relative acceleration set entirely by the Riemann curvature tensor: $D^2 S^\mu/\mathrm{d}\tau^2 = -R^\mu{}_{\nu\rho\sigma}\,u^\nu S^\rho u^\sigma$ . The equation is linear in the separation, so curvature acts as a tidal operator — feed it a connecting vector, get back a relative acceleration — whose eigenvalues stretch space along some axes and squeeze it along others. Because only the Riemann tensor appears, never the connection alone, the effect cannot be removed by any change of coordinates: this is exactly what it means for curvature to be intrinsic and for a gravitational field to be physically present. From the Newtonian tidal tensor to the oscillating arms of a gravitational-wave detector to the focusing theorems behind the singularity theorems, this single line is the bridge between the geometry of spacetime and the forces we measure.

Key takeaways

The geodesic deviation equation gives the relative acceleration of two infinitesimally close geodesics as $\frac{D^2 S^\mu}{\mathrm{d}\tau^2} = -R^\mu{}_{\nu\rho\sigma}\,u^\nu S^\rho u^\sigma$ , with $S^\mu$ the connecting vector and $u^\mu$ the tangent.
It is the precise statement that spacetime curvature is observable: free-fallers converge or diverge purely because of the Riemann tensor, with no force acting in any local frame.
It follows from the geodesic equation and a torsion-free connection by taking two covariant derivatives of $S^\mu$ and using the commutator $[\nabla_\rho,\nabla_\sigma]V^\mu = R^\mu{}_{\nu\rho\sigma}V^\nu$ — the curvature enters as the failure of covariant derivatives to commute.
In the Newtonian limit it reduces to the classical tidal equation $\ddot S^i = -\partial_i\partial_j\Phi\,S^j$ , identifying the tidal components $R^i{}_{0j0}\approx\partial_i\partial_j\Phi$ with second derivatives of the potential.
In transverse-traceless gauge a gravitational wave drives $\ddot S^i = \tfrac12\ddot h_{ij}S^j$ , the operating principle of every interferometric detector; only the transverse part of $S^\mu$ carries tidal content.
In flat spacetime $R=0$ , so initially parallel geodesics stay parallel forever — the operational meaning of "no gravitational field".
Key idea: curvature is measured by the failure of nearby free-falling worldlines to stay parallel, and the Riemann tensor is exactly the operator that turns their separation into a relative acceleration.

The core insight

Curvature is measured by the failure of nearby freely falling worldlines to stay parallel; the Riemann tensor is precisely the operator that converts the separation vector between them into a relative acceleration.

Where this lives today

Geodesic deviation is the working principle of an entire branch of astronomy. Every gravitational-wave observation — LIGO's GW150914 in 2015, the multi-messenger neutron-star merger GW170817 in 2017, and the growing catalogue from the LIGO–Virgo–KAGRA network — is a measurement of the relative acceleration of free test masses driven by $\ddot S^i=\tfrac12\ddot h_{ij}S^j$ , exactly Pirani's prescription realised at the scale of a proton's width. The space mission LISA, targeted for the 2030s, will read the same equation across arms of 2.5 million kilometres, opening the millihertz band; and pulsar-timing arrays such as NANOGrav, which in 2023 reported evidence for a nanohertz gravitational-wave background, treat the whole Galaxy as a geodesic-deviation detector, watching pulsar arrival times wobble as low-frequency curvature stretches the space between us and each pulsar. The equation's other face, spaghettification, is now caught in the act: tidal disruption events, where a star is torn apart by the tidal eigenvalues computed above, flare across the sky and are monitored by surveys like ZTF and the Vera Rubin Observatory. And in mathematical relativity, tracing the deviation equation along a congruence yields the Raychaudhuri equation, whose focusing term $R_{\mu\nu}u^\mu u^\nu\ge0$ under the energy conditions drives the Penrose–Hawking singularity theorems — so the same tensor that stretches LIGO's mirrors also proves that black holes and the Big Bang must harbour singularities.

The physical picture

A single geodesic tells you nothing about gravity — ride along it and you feel weightless, your local frame indistinguishable from deep space. Gravity announces itself only when you compare two neighbouring geodesics and watch the vector between them accelerate. Drop a small spherical cloud of dust from rest above a planet and the sphere does not stay a sphere: it elongates along the line to the centre, because the near side falls faster, and narrows across that line, because every grain aims at the same point. Released from rest in vacuum the cloud's volume is unchanged to second order in time, but its shape is not, and that deformation — pure shape change with no force felt by any grain — is geodesic deviation, its rate the Riemann tensor read out as a tidal operator.

Figure

A small ring of freely falling test particles deforms under tidal curvature: it stretches along the radial direction toward the mass and squeezes transverse to it, while the cloud's full three-dimensional volume is unchanged to second order in vacuum. The relative acceleration of each particle is D²Sμ/dτ² = −Rμ νρσ uν Sρ uσ, linear in the separation S

Think, too, of two ships setting off due north from the equator, each steering perfectly straight along a great circle. On a flat map they run parallel forever; on the round Earth they converge and meet at the pole. Neither captain turns the wheel, yet their separation accelerates toward zero at a rate set by the Gaussian curvature $K=1/a^2$ — the two-dimensional ancestor of the Riemann tensor. Geodesic deviation is this picture promoted to four dimensions: "steering straight" becomes following a geodesic, "curvature" becomes $R^\mu{}_{\nu\rho\sigma}$ . The tidal operator has real eigenvalues that stretch along some axes and compress along others, and a passing gravitational wave makes those axes rotate — exactly the pattern a kilometre-scale interferometer is built to register.

Figure

A gravitational wave with plus polarization drives geodesic deviation between the free-hanging mirrors of an interferometer: one arm lengthens while the perpendicular arm shortens, oscillating as ΔL slash L = h slash 2. For LIGO with arm length L = 4 km and peak strain h = 1×10⁻²¹, the arm length change is ΔL = 2×10⁻¹⁸ m — about one thousandth of a proton diameter

The core equations

The central result packages tidal gravity into one tensor equation: the relative acceleration of two neighbouring free-fallers is a linear function of their separation, the operator being the Riemann tensor contracted twice with the four-velocity. Everything else here either derives this equation, limits it, or reads a number out of it.

\frac{D^2 S^\mu}{\mathrm{d}\tau^2} = -R^\mu{}_{\nu\rho\sigma}\, u^\nu S^\rho u^\sigma \qquad (1)

where: $S^\mu = \partial x^\mu/\partial s$ — the deviation (connecting) vector to the neighbouring geodesic; $u^\mu = \partial x^\mu/\partial\tau$ — the unit tangent; $R^\mu{}_{\nu\rho\sigma}$ — the Riemann curvature tensor; $D/\mathrm{d}\tau = u^\alpha\nabla_\alpha$ — the covariant derivative along the geodesic; $\tau$ — the affine parameter (proper time for timelike geodesics). Reading: the left side is a frame-independent acceleration; the right side vanishes if and only if spacetime is flat. The whole construction rests on a smooth two-parameter map: a congruence of geodesics labelled by $s$ , each running in its own affine parameter $\tau$ . The tangent points along the flow; the deviation vector points across it.

u^\mu = \frac{\partial x^\mu}{\partial \tau}, \qquad S^\mu = \frac{\partial x^\mu}{\partial s} \qquad (2)

where: $x^\mu(\tau,s)$ — the smooth congruence; $s$ — the label distinguishing neighbouring geodesics; $\tau$ — the affine parameter along each one. Reading: because $u$ and $S$ are coordinate basis vectors of the same map, their mixed partials commute and their Lie bracket vanishes, $[u,S]=0$ — the single fact that makes the derivation work. Each curve is a geodesic: its tangent is parallel-transported along itself. This affine form of the geodesic equation is what kills every non-curvature term in the derivation.

u^\nu\nabla_\nu u^\mu = 0 \qquad (3)

where: $\nabla_\nu$ — the Levi-Civita (torsion-free, metric-compatible) covariant derivative. Reading: the tangent does not accelerate covariantly, so any surviving relative acceleration must come from the geometry between geodesics, not from the motion of any one. The covariant derivative along the curve is what " $D/\mathrm{d}\tau$ " means, and the vanishing Lie bracket promotes to a clean interchange identity for a torsion-free connection.

\frac{D V^\mu}{\mathrm{d}\tau} \equiv u^\alpha\nabla_\alpha V^\mu = \frac{\mathrm{d}V^\mu}{\mathrm{d}\tau} + \Gamma^\mu{}_{\alpha\beta}u^\alpha V^\beta, \qquad u^\nu\nabla_\nu S^\mu = S^\nu\nabla_\nu u^\mu \qquad (4)

where: $V^\mu$ — any vector transported along the geodesic; $\Gamma^\mu{}_{\alpha\beta}$ — the Christoffel symbols. Reading: the second identity says the change of $S$ along $u$ equals the change of $u$ along $S$ ; it is where torsion-freeness enters and what lets the geodesic equation act on the deviation. Curvature enters as the obstruction to commuting two covariant derivatives. This commutator defines the Riemann tensor, and substituting it is the only step that introduces $R$ .

[\nabla_\rho,\nabla_\sigma]\,V^\mu = R^\mu{}_{\nu\rho\sigma}\,V^\nu, \qquad R^\mu{}_{\nu\rho\sigma} = \partial_\rho\Gamma^\mu{}_{\sigma\nu} - \partial_\sigma\Gamma^\mu{}_{\rho\nu} + \Gamma^\mu{}_{\rho\lambda}\Gamma^\lambda{}_{\sigma\nu} - \Gamma^\mu{}_{\sigma\lambda}\Gamma^\lambda{}_{\rho\nu} \qquad (5)

where: $V^\nu$ — an arbitrary vector field. Reading: in local inertial coordinates the $\Gamma\Gamma$ products vanish but the derivatives $\partial\Gamma$ need not, so $R$ cannot be transformed away — this is why the connection alone never appears in equation (1). In the Newtonian limit — slow particles, weak static field, $u^\mu\approx(1,\mathbf 0)$ — equation (1) collapses onto the classical tidal equation governed by second derivatives of the potential.

\frac{\mathrm{d}^2 S^i}{\mathrm{d}t^2} = -\,\partial_i\partial_j\Phi\;S^j, \qquad R^i{}_{0j0}\approx\partial_i\partial_j\Phi \qquad (6)

where: $S^i$ — the spatial separation of two nearby falling bodies; $\Phi$ — the Newtonian potential with $g_{00}\approx-(1+2\Phi)$ ; $\partial_i\partial_j\Phi$ — the tidal tensor, whose trace obeys $\nabla^2\Phi=4\pi G\rho$ . Reading: the "electric" components $R^i{}_{0j0}$ of curvature are literally the Newtonian tidal field — a Riemann tensor is what a tide is, made exact. Finally, only the transverse part of $S^\mu$ is physical. The projection along the motion is conserved and merely reparametrises which geodesic is "next", carrying no tidal content.

u_\mu S^\mu = \text{const}, \qquad \frac{D(u_\mu S^\mu)}{\mathrm{d}\tau} = 0 \qquad (7)

where: $u_\mu S^\mu$ — the longitudinal component of the deviation. Reading: if $S$ starts orthogonal to $u$ it stays orthogonal, so one conventionally works with purely transverse deviation vectors, the ones the tidal operator can actually stretch or squeeze.

Derivation ADeriving the geodesic deviation equation from a family of geodesics

The goal is to start from nothing but a smooth congruence of geodesics and the two structural facts that define it — each curve obeys $u^\nu\nabla_\nu u^\mu=0$ , and the tangent and deviation vectors have vanishing Lie bracket — and to show that the second covariant derivative of the deviation vector reduces, after every non-curvature term cancels, to a single contraction of the Riemann tensor. We work throughout with the torsion-free, metric-compatible Levi-Civita connection, tracking each dummy index so the final cancellation is transparent.

Step 1. Write the data of the family: $u^\mu=\partial_\tau x^\mu$ , $S^\mu=\partial_s x^\mu$ , and the geodesic condition $u^\nu\nabla_\nu u^\mu=0$ holding on every member. Because $u$ and $S$ are coordinate basis vectors of the map $x^\mu(\tau,s)$ , mixed partials commute and $[u,S]^\mu=0$ . Why: these two facts power everything — geodesy kills the tangent's own acceleration, the commuting frame lets us trade derivatives of $S$ for derivatives of $u$ .

Step 2. Convert the vanishing bracket into a covariant identity. For a torsion-free connection $[u,S]^\mu = u^\nu\nabla_\nu S^\mu - S^\nu\nabla_\nu u^\mu$ , so $u^\nu\nabla_\nu S^\mu = S^\nu\nabla_\nu u^\mu$ . Why: this is the only place torsion-freeness is used; with torsion an extra $T^\mu{}_{\nu\rho}u^\nu S^\rho$ term would survive and spoil the clean curvature-only result.

Step 3. Take the first covariant derivative of the deviation along the geodesic, $\frac{DS^\mu}{\mathrm{d}\tau}=u^\rho\nabla_\rho S^\mu = S^\rho\nabla_\rho u^\mu$ using the identity from step 2. Why: this recasts the relative velocity of neighbouring geodesics as the gradient of the tangent field contracted with the separation, so the second derivative can be attacked with the geodesic equation.

Step 4. Differentiate once more and expand with the product rule: $\frac{D^2 S^\mu}{\mathrm{d}\tau^2} = u^\rho\nabla_\rho\big(S^\nu\nabla_\nu u^\mu\big) = (u^\rho\nabla_\rho S^\nu)\nabla_\nu u^\mu + u^\rho S^\nu\nabla_\rho\nabla_\nu u^\mu$ . Why: the second covariant derivative is what "relative acceleration" means; splitting it into these two pieces isolates the term that will become curvature.

Step 5. Rewrite the first piece with the step-2 identity, $u^\rho\nabla_\rho S^\nu = S^\rho\nabla_\rho u^\nu$ , giving $(S^\rho\nabla_\rho u^\nu)\nabla_\nu u^\mu$ . Why: this puts the first piece into a form that will cancel exactly against a term generated when we commute derivatives in the second piece.

Step 6. In the second piece swap the order of covariant derivatives using the Riemann commutator, $u^\rho\nabla_\rho\nabla_\nu u^\mu = u^\rho\nabla_\nu\nabla_\rho u^\mu + u^\rho R^\mu{}_{\lambda\rho\nu}u^\lambda$ . Why: interchanging $\nabla_\rho$ and $\nabla_\nu$ is exactly where the geometry announces itself — the price of the swap is one Riemann term, the seed of the whole result.

Step 7. Handle the reordered term with the geodesic equation: $S^\nu u^\rho\nabla_\nu\nabla_\rho u^\mu = S^\nu\nabla_\nu(u^\rho\nabla_\rho u^\mu) - S^\nu(\nabla_\nu u^\rho)\nabla_\rho u^\mu$ . The first term vanishes because $u^\rho\nabla_\rho u^\mu=0$ , and the second cancels the surviving piece from step 5 after relabelling $\rho\leftrightarrow\nu$ . Why: this cancellation makes the equation tidal — every term in first derivatives of $u$ disappears, leaving no dependence on the connection alone.

Step 8. Collect what remains, only the curvature term: $\frac{D^2 S^\mu}{\mathrm{d}\tau^2} = R^\mu{}_{\lambda\rho\nu}\,u^\lambda u^\rho S^\nu$ . Relabel indices and use the antisymmetry $R^\mu{}_{\lambda\rho\nu}=-R^\mu{}_{\lambda\nu\rho}$ to bring it to the conventional form $\frac{D^2 S^\mu}{\mathrm{d}\tau^2} = -R^\mu{}_{\nu\rho\sigma}\,u^\nu S^\rho u^\sigma$ . Why: the overall sign is convention-dependent on index ordering, but the content is fixed — relative acceleration is a linear, curvature-only functional of the separation.

Go deeper · Derivation B · 1 minRecovering the Newtonian tidal tensor in the weak-field limit

Derivation BRecovering the Newtonian tidal tensor in the weak-field limit

The goal is to show that equation (1), which looks purely geometric, contains ordinary Newtonian tides as its slow-and-weak limit, and to pin down the exact dictionary between the curvature components and the second derivatives of the potential. We take a static weak field with $g_{00}=-(1+2\Phi)$ and $|\Phi|\ll1$ , particles slow enough that $u^\mu\approx(1,0,0,0)$ and $\tau\approx t$ , and keep only terms linear in $\Phi$ , since the tidal effect is already first order.

Step 1. Restrict equation (1) to its dominant components. With $u^\mu\approx\delta^\mu_0$ only $\nu=\sigma=0$ survives, so $\frac{D^2 S^i}{\mathrm{d}\tau^2}\approx -R^i{}_{0j0}\,S^j$ , and the spatial separation $S^j$ dominates. Why: slow motion means the four-velocity is almost purely temporal, which projects the four-index tidal operator down to the familiar three-by-three tidal matrix.

Step 2. Compute the relevant Riemann components from equation (5). To linear order the $\Gamma\Gamma$ products are second order and drop, and for a static field time derivatives vanish, leaving $R^i{}_{0j0}\approx\partial_j\Gamma^i{}_{00}-\partial_0\Gamma^i{}_{j0}\approx\partial_j\Gamma^i{}_{00}$ . Why: in the weak-field, static regime the Riemann tensor is just a single spatial derivative of the one Christoffel symbol that carries the field.

Step 3. Evaluate that Christoffel symbol: $\Gamma^i{}_{00}\approx\tfrac12 g^{ij}(2\partial_0 g_{j0}-\partial_j g_{00})\approx\partial_i\Phi$ , using $g_{00}=-(1+2\Phi)$ , $g^{ij}\approx\delta^{ij}$ , and $\partial_0 g_{j0}=0$ for a static field. Why: this is precisely the term that reproduces Newton's law $\ddot x^i=-\partial_i\Phi$ for a single particle; here we need its gradient, not the term itself.

Step 4. Differentiate to assemble the tidal tensor: $R^i{}_{0j0}\approx\partial_j\partial_i\Phi$ , and hence $\frac{\mathrm{d}^2 S^i}{\mathrm{d}t^2}=-\partial_i\partial_j\Phi\,S^j$ . Why: this is the exact Newtonian tidal equation — stretching along the radial direction, compression transverse to it, with trace $-\nabla^2\Phi=-4\pi G\rho$ that vanishes in vacuum, so a free-falling cloud changes shape but not volume outside matter. *The identification $R^i{}_{0j0}\approx\partial_i\partial_j\Phi$ is the operational content of "curvature is the tidal field".*

Go deeper · Derivation C · 1 minThe gravitational-wave form: geodesic deviation in transverse-traceless gauge

Derivation CThe gravitational-wave form: geodesic deviation in transverse-traceless gauge

The goal is to specialise the deviation equation to a weak plane gravitational wave and obtain the relation an interferometer exploits, $\ddot S^i=\tfrac12\ddot h_{ij}S^j$ , showing that a passing wave is a time-varying tidal field whose principal axes rotate with the polarization. We write $g_{\mu\nu}=\eta_{\mu\nu}+h_{\mu\nu}$ with $|h|\ll1$ in transverse-traceless gauge, where $h_{0\mu}=0$ , $h_{ij}$ is transverse and traceless, and the test masses start at rest so $u^\mu\approx(1,\mathbf 0)$ .

Step 1. Linearise the Riemann tensor. To first order $R_{\mu\nu\rho\sigma}=\tfrac12(\partial_\rho\partial_\nu h_{\mu\sigma}+\partial_\sigma\partial_\mu h_{\nu\rho}-\partial_\sigma\partial_\nu h_{\mu\rho}-\partial_\rho\partial_\mu h_{\nu\sigma})$ , exact enough because the wave amplitude is tiny. Why: at strains of order $10^{-21}$ every quadratic-in- $h$ correction is utterly negligible, so the linearised curvature is all we ever need.

Step 2. Extract the tidal component $R_{i0j0}$ . Setting the indices and dropping every term containing a $0$ -component of $h$ leaves $R_{i0j0}=-\tfrac12\partial_0\partial_0 h_{ij}=-\tfrac12\ddot h_{ij}$ , so $R^i{}_{0j0}=-\tfrac12\ddot h_{ij}$ . Why: the TT gauge is engineered so that only the second time derivative of the spatial metric perturbation survives in the tidal slot — the wave's curvature is its acceleration in time.

Step 3. Insert into equation (1) with $u^\mu\approx\delta^\mu_0$ : $\frac{\mathrm d^2 S^i}{\mathrm dt^2}=-R^i{}_{0j0}S^j=+\tfrac12\ddot h_{ij}S^j$ . Why: this says the proper separation of two free masses oscillates in step with $\ddot h$ , so the fractional length change tracks the strain, $\delta L/L\sim\tfrac12 h$ — the quantity a detector measures.

Step 4. Read off the polarization pattern. The $+$ mode has $h_{xx}=-h_{yy}=h_+(t)$ , stretching $x$ while squeezing $y$ ; the $\times$ mode has $h_{xy}=h_{yx}=h_\times(t)$ , doing the same along axes at $45^\circ$ . Why: two orthogonal arms therefore beat against each other in quadrature, which is precisely why an L-shaped interferometer with perpendicular arms is the natural antenna for a gravitational wave. *Note that this result is consistent with equation (6) once $h_{ij}$ is promoted to a full spacetime-dependent wave rather than a static potential.*

Limiting cases

Pushing the equation into special regimes shows both what it contains and how it connects to older physics.

Case	Result	Meaning
Flat spacetime	$R^\mu{}_{\nu\rho\sigma}=0 \Rightarrow \frac{D^2 S^\mu}{\mathrm{d}\tau^2}=0$	Initially parallel geodesics stay parallel; the deviation vector is parallel-transported with zero relative acceleration. This is the operational definition of "no gravitational field".
Newtonian / weak-field	$\frac{\mathrm{d}^2 S^i}{\mathrm{d}t^2}\approx-\partial_i\partial_j\Phi\,S^j$	With $R^i{}_{0j0}\approx\partial_i\partial_j\Phi$ and $u^\mu\approx(1,\mathbf 0)$ , the equation is exactly the classical tidal tensor of Newtonian gravity.
Plane gravitational wave (TT gauge)	$\ddot S^i\approx\tfrac12\ddot h_{ij}S^j$	A passing wave drives oscillating geodesic deviation between free test masses; the fractional length change is $\delta L/L\sim\tfrac12 h$ — the principle of interferometric detection.
Two-sphere of radius $a$	$\frac{D^2 S}{\mathrm{d}\lambda^2}=-\frac{1}{a^2}S$	Great-circle geodesics converge sinusoidally with wavelength $2\pi a$ ; the Gaussian curvature $K=1/a^2$ sets the focusing rate — the two-dimensional prototype.
Schwarzschild radial infall	$\frac{D^2 S^{\hat r}}{\mathrm{d}\tau^2}=+\frac{2GM}{r^3}S^{\hat r}$	The radial tidal eigenvalue stretches; the two transverse eigenvalues are $-GM/r^3$ and squeeze. The radial value equals the Newtonian tide exactly, all the way to the horizon.

Common trapThe relative acceleration is the ordinary second derivative of the separation

It is not — it is the covariant second derivative $D^2 S^\mu/\mathrm{d}\tau^2$ , and the distinction is the whole point. The ordinary coordinate acceleration $\mathrm{d}^2 S^\mu/\mathrm{d}\tau^2$ of a separation vector is not a tensor: it mixes in the Christoffel symbols and can be made to take almost any value at a point by a bad choice of chart. Only the covariant version transforms as a genuine vector and describes something physical. This is exactly why the final equation contains the Riemann tensor and never the connection alone: at any point you can adopt local inertial coordinates where $\Gamma=0$ , killing the ordinary acceleration of any single geodesic, yet the covariant relative acceleration of neighbours persists because it is built from $\partial\Gamma$ , which no coordinate change can annihilate. Confuse the two and you will "discover" tidal forces in flat spacetime that are pure coordinate artefacts.

Worked examples

1 · Spaghettification: why a stellar black hole shreds you but a giant one does not (fully worked)

Model a falling astronaut of height $L=1.8$ m as two free particles separated radially by $L$ . Using the Schwarzschild radial tidal eigenvalue $2GM/r^3$ , compute the differential acceleration across the body at the event horizon of a $10\,M_\odot$ stellar black hole and of the $4.3\times10^6\,M_\odot$ hole Sgr A at the Galactic centre.*

Step 1. From the Schwarzschild tidal tensor in the free-faller's orthonormal frame, the radial component of geodesic deviation is $\frac{D^2 S^{\hat r}}{\mathrm{d}\tau^2}=\frac{2GM}{r^3}S^{\hat r}$ — a stretch, since the coefficient is positive. Remarkably this equals the Newtonian tidal gradient exactly, so we may evaluate it with $G=6.674\times10^{-11}$ , $M_\odot=1.989\times10^{30}$ kg. The horizon sits at $r_s=2GM/c^2$ .

Step 2. Stellar hole, $M=10\,M_\odot$ . Then $r_s=2GM/c^2=\mathbf{2.95\times10^{4}\ m}$ (about 30 km), and the tidal gradient at the horizon is $2GM/r_s^3=\mathbf{1.03\times10^{8}\ s^{-2}}$ . Across the body this is a differential acceleration $\Delta a=(1.03\times10^8)(1.8)=\mathbf{1.9\times10^{8}\ m/s^2}\approx\mathbf{1.9\times10^{7}\,g}$ — twenty million gravities pulling head from feet. Lethal long before the horizon.

Step 3. Supermassive hole, $M=4.3\times10^6\,M_\odot$ . Now $r_s=\mathbf{1.27\times10^{10}\ m}$ (about 0.08 AU), and the gradient is $2GM/r_s^3=\mathbf{5.6\times10^{-4}\ s^{-2}}$ , giving $\Delta a=(5.6\times10^{-4})(1.8)=\mathbf{1.0\times10^{-3}\ m/s^2}\approx\mathbf{1.0\times10^{-4}\,g}$ — utterly imperceptible.

Step 4. The scaling explains the paradox: at its own horizon $r_s\propto M$ , so the tidal gradient $2GM/r_s^3\propto M/M^3=1/M^2$ . A hole a million times heavier has a horizon tide a trillion times gentler. You would be spaghettified outside a stellar black hole but cross the horizon of a supermassive one intact — the tidal stretch there is $10^{-4}\,g$ .

2 · The strain LIGO must measure

A gravitational wave of peak strain $h=1\times10^{-21}$ passes a detector whose arms have length $L=4$ km, with one arm aligned to the $+$ -polarization stretch axis. How far does that arm move?

Step 1. From the TT-gauge result $\ddot S^i=\tfrac12\ddot h_{ij}S^j$ , integrating over the wave gives a proper length change $\delta L/L=\tfrac12 h$ for an arm along a principal axis — the strain directly modulates the arm length.

Step 2. Insert the numbers: $\Delta L=\tfrac12 h L=\tfrac12(1\times10^{-21})(4000\ \mathrm{m})=\mathbf{2\times10^{-18}\ m}$ .

Step 3. Scale it: a proton is about $1.7\times10^{-15}$ m across, so $\Delta L$ is roughly $10^{-3}$ of a proton diameter. The perpendicular arm shrinks by the same amount and in quadrature, so the differential signal is $2\Delta L$ — which is what lets an interferometer register a length change thousands of times smaller than a nucleus. The wave moves each 4 km arm by 2×10⁻¹⁸ m — about one thousandth of a proton — precisely the geodesic deviation LIGO detected in 2015.

3 · Great circles on the Earth: geodesic deviation you can walk (analytic)

Two hikers stand on the equator 100 km apart and both set off due north along great circles (meridians). Treating the Earth as a sphere of radius $a=6371$ km, find their separation at latitude $60^\circ$ and at the pole, and identify the curvature that drives their convergence.

Step 1. On a two-sphere the deviation equation reduces to $\frac{\mathrm{d}^2 S}{\mathrm{d}\lambda^2}=-\frac{1}{a^2}S$ , with $\lambda$ the arc length along the geodesic and $K=1/a^2$ the Gaussian curvature. Parametrising by colatitude $\theta$ (so $\theta=90^\circ$ at the equator, $0^\circ$ at the pole), the separation of two meridians a fixed longitude $\Delta\phi$ apart is $S(\theta)=a\sin\theta\,\Delta\phi$ , the sinusoidal solution.

Step 2. Fix the constant from the start line: at the equator $\sin\theta=1$ , so $S=a\Delta\phi=100$ km, giving $\Delta\phi=100/6371=0.0157$ rad.

Step 3. Evaluate along the way. At latitude $60^\circ$ ( $\theta=30^\circ$ ): $S=6371\sin30^\circ\times0.0157=6371\times0.5\times0.0157=\mathbf{50.0\ km}$ . At the pole ( $\theta=0$ ): $S=6371\times0\times0.0157=\mathbf{0\ km}$ — the hikers meet, though neither ever turned.

Step 4. No force acted; the convergence is pure positive curvature, the two-dimensional face of the Riemann tensor. Starting 100 km apart, the "straight-walking" hikers are 50 km apart at 60°N and collide at the pole, their relative acceleration set entirely by $K=1/a^2$ .

Physics of the portrait

The geodesic deviation equation had been written down as pure geometry by Levi-Civita and cast in relativistic form by Synge, but it took a young physicist in the 1950s to insist on what it meant: that the Riemann tensor is not a bookkeeping device but something an experimenter can measure, by watching free particles drift.

Origins · 2 minFelix Pirani and how this was discovered

Felix Pirani

1928–2015 · argued that the Riemann tensor is physically measured through geodesic deviation, giving gravitational waves an observable, coordinate-free reality

Felix Arnold Edward Pirani, born in London in 1928 and raised largely in Canada, arrived at general relativity when the field was a backwater and one of its central questions was embarrassingly unsettled: were gravitational waves real, or mere ripples in the coordinates? He took a doctorate under Alfred Schild at the Carnegie Institute of Technology and a second under Hermann Bondi at Cambridge, and in 1956 published "On the Physical Significance of the Riemann Tensor," which cut through the confusion with a single idea — to give curvature an operational meaning, watch a cloud of freely falling test particles and measure their relative accelerations. Those accelerations are governed by the geodesic deviation equation, so the components of the Riemann tensor are directly readable from the tidal drift of free masses, a statement no coordinate change can undo. Applied to a passing wave, Pirani's criterion said unambiguously that the wave would shove free particles back and forth and could therefore be detected — the theoretical seed of the interferometer. He carried the argument to the pivotal 1957 Chapel Hill conference, where Bondi and Feynman sharpened it into the "sticky bead" thought experiment: a wave passing through a rod slides beads along it and deposits heat, so it carries energy and is physically real. With Bondi at King's College London, Pirani built one of the world's leading relativity groups and, with Robinson, studied exact plane-wave solutions carrying genuine energy. His 1956 insight remains the conceptual charter of gravitational-wave astronomy: LIGO's mirrors are exactly the free test particles whose geodesic deviation he told us to measure.

History

1837

— Jacobi analyses the second-variation equation for geodesics, defining what are now called Jacobi fields and conjugate points

1917

— Levi-Civita introduces parallel transport, giving the covariant derivative its geometric meaning

1926

— Levi-Civita names and formulates geodesic deviation in The Absolute Differential Calculus

1934

— Synge casts the deviation equation in its relativistic tensor form

1956

— Pirani argues the Riemann tensor is measured physically through geodesic deviation of free particles

2015

— LIGO detects GW150914 as oscillating geodesic deviation between freely suspended mirrors

Going further

Texts. Misner, Thorne and Wheeler, Gravitation, §1.6 and §§11.1–11.4 · Wald, General Relativity, §3.3 · Carroll, Spacetime and Geometry, §3.10 (Eq. 3.198) · Hartle, Gravity, §21.5 · Schutz, A First Course in General Relativity, 2nd ed., §6.5 and §9. The overall sign in equation (1) is convention-dependent on index ordering; the sign conventions here follow the MTW/Carroll Riemann convention. Key checkpoints in the derivations: the cancellation of non-curvature terms in the main derivation uses only the torsion-free commutator $[\nabla_\rho,\nabla_\sigma]V^\mu=R^\mu{}_{\nu\rho\sigma}V^\nu$ , metric compatibility, and the geodesic equation; the Newtonian identification $R^i{}_{0j0}\approx\partial_i\partial_j\Phi$ follows from $\Gamma^i{}_{00}\approx\partial_i\Phi$ ; the TT-gauge result $R^i{}_{0j0}=-\tfrac12\ddot h_{ij}$ gives $\ddot S^i=\tfrac12\ddot h_{ij}S^j$ ; the Schwarzschild radial tidal eigenvalue $2GM/r^3$ yields $1.9\times10^7\,g$ across 1.8 m at a $10\,M_\odot$ horizon and $1.0\times10^{-4}\,g$ at Sgr A*; LIGO gives $\Delta L=\tfrac12 hL=2\times10^{-18}$ m for $h=10^{-21}$ , $L=4$ km; and the sphere convergence $S(\theta)=a\sin\theta\,\Delta\phi$ gives 50 km at 60°N from 100 km at the equator.

Builds on: The Geodesic Equation · Covariant Derivative · Parallel Transport · Christoffel Symbols · The Levi-Civita Connection. Leads to: Tidal Forces in General Relativity · Killing Vectors.

Ask the tutor: Ask me to work the tidal operator for any spacetime you like — Schwarzschild infall, a plane gravitational wave, or the round sphere — and watch the Riemann tensor turn a separation vector into a relative acceleration.

Why can't a single freely falling particle detect gravity, but two can?

A single free particle obeys $u^\nu\nabla_\nu u^\mu=0$ : in its own local inertial frame the Christoffel symbols vanish and it floats weightless, indistinguishable from an inertial observer in empty space. That is the equivalence principle, and it means the uniform part of a gravitational field is a coordinate effect. What no coordinate choice can erase is the difference between neighbouring free-fallers, governed by the Riemann tensor through the geodesic deviation equation. Two particles let you measure a relative acceleration, and a nonzero relative acceleration is the invariant, physical signature of curvature — the true gravitational field.

What exactly is the deviation vector, and why must it be transverse?

The deviation vector $S^\mu=\partial x^\mu/\partial s$ points from a chosen geodesic to its infinitesimal neighbour, so it measures their separation. Its component along the four-velocity, $u_\mu S^\mu$ , is conserved along the motion and corresponds merely to a shift in affine parameter — a relabelling of which point on the neighbour is "next" — so it carries no tidal information. Only the transverse part, orthogonal to $u^\mu$ , is genuine physical separation that the tidal operator can stretch or squeeze. Because the longitudinal part is preserved, one conventionally sets $u_\mu S^\mu=0$ and works entirely with transverse deviation.

Where does the Riemann tensor actually come from in the derivation?

From one move: commuting two covariant derivatives. When you take the second covariant derivative of $S^\mu$ and simplify with the geodesic equation, you reach a term $u^\rho\nabla_\rho\nabla_\nu u^\mu$ with the derivatives in the "wrong" order. Swapping them is not free — for a torsion-free connection the price is exactly $[\nabla_\rho,\nabla_\nu]u^\mu=R^\mu{}_{\lambda\rho\nu}u^\lambda$ , and that single Riemann term survives after every other piece cancels against the geodesic equation. Curvature enters as the failure of covariant derivatives to commute, which is the definition of the Riemann tensor.

Can two geodesics that start parallel and at rest still fly apart?

Yes — this is how tides build up from nothing. The deviation equation is second order in the affine parameter: even if the relative velocity $DS^\mu/\mathrm{d}\tau$ is zero initially, a nonzero right-hand side $-R^\mu{}_{\nu\rho\sigma}u^\nu S^\rho u^\sigma$ produces a nonzero relative acceleration, so relative velocity develops from rest. This is exactly what happens to a cloud released at rest in a tidal field: it starts perfectly still yet immediately deforms, stretching along one axis and compressing along another, purely because the Riemann tensor is nonzero.

Is the geodesic deviation equation the same as the Jacobi equation?

They are one equation seen from two traditions. In Riemannian geometry, solutions $S^\mu$ of $\frac{D^2 S^\mu}{\mathrm{d}\lambda^2}+R^\mu{}_{\nu\rho\sigma}u^\nu S^\rho u^\sigma=0$ are Jacobi fields, after Jacobi's study of the second variation of arc length; points where a nontrivial Jacobi field vanishes at both ends are conjugate points, beyond which a geodesic stops locally extremising length. In relativity the identical equation is the geodesic deviation equation and its solutions describe tidal drift. One tradition asks "when does a geodesic stop being the shortest path?", the other "how do free particles fall relative to each other?" — but the mathematics is the same object.

How does a gravitational-wave detector use this equation?

Directly and literally. The suspended mirrors of an interferometer are, to excellent approximation, free test masses on geodesics, so the vector between two of them obeys geodesic deviation. In transverse-traceless gauge a passing wave gives $\ddot S^i=\tfrac12\ddot h_{ij}S^j$ , so the proper distance between mirrors oscillates with fractional amplitude $\delta L/L\sim\tfrac12 h$ . The $+$ polarization lengthens one arm while shortening the perpendicular one, and $\times$ does the same at $45^\circ$ , so an L-shaped detector reads the strain as a differential arm-length change. For LIGO's 4 km arms and $h\sim10^{-21}$ that motion is about $2\times10^{-18}$ m — the geodesic deviation the instrument is engineered to resolve.