Killing Vectors

Every conservation law in physics has a symmetry hiding behind it, and in general relativity that symmetry is a geometric object rather than a statement about a Lagrangian. Noether's theorem — established for field theories acting on a matter action — tells you a continuous symmetry yields a conserved current; in curved spacetime the symmetry is a symmetry of the geometry itself, a transformation that slides the whole manifold along and leaves every distance and angle unchanged. When such a rigid motion is built up continuously from infinitesimal steps, the vector field that generates it is a Killing vector field, after Wilhelm Killing, who wrote down the defining equation while studying the infinitesimal motions of space forms. The precise condition is that the metric is Lie-dragged invariantly along the flow of $\xi^\mu$ , $\mathcal{L}_\xi g_{\mu\nu}=0$ . Expanded with the metric-compatible, torsion-free Levi-Civita connection this collapses to the compact, first-order, overdetermined system $\nabla_{(\mu}\xi_{\nu)}=0$ — Killing's equation. Solutions need not exist: a generic bumpy metric admits none. Their number measures how symmetric a spacetime is, from zero for a lumpy cosmology up to the maximal $n(n+1)/2$ for constant-curvature spaces, and the full set closes into a finite-dimensional Lie algebra, the isometry algebra of the manifold. From this one equation flow the conserved energy and angular momentum of orbits, the redshift between static observers, Komar's mass and angular momentum, the surface gravity of a horizon, and the thermodynamic laws of black holes — all developed below from the ground up.

Key takeaways

A Killing vector field $\xi^\mu$ generates a one-parameter family of isometries — a continuous rigid motion that leaves the metric invariant, $\mathcal{L}_\xi g_{\mu\nu}=0$ .
Its defining condition is Killing's equation $\nabla_\mu\xi_\nu+\nabla_\nu\xi_\mu=0$ : the symmetric part of the gradient of $\xi$ vanishes, so the flow has no expansion and no shear.
Each Killing vector hands a geodesic a conserved quantity $\xi_\mu u^\mu$ — energy from $\partial_t$ , angular momentum from $\partial_\phi$ — reducing orbits to a one-dimensional energy equation.
The integrability identity $\nabla_\mu\nabla_\nu\xi_\rho=R_{\rho\nu\mu}{}^\sigma\xi_\sigma$ closes the system, fixing a Killing vector everywhere from $\xi^\mu$ and $\nabla_{[\mu}\xi_{\nu]}$ at one point and bounding their number by $n(n+1)/2$ (ten in 4D).
They define what stationary, static, and axisymmetric mean, build the Komar charges, and give a Killing horizon its constant surface gravity and Hawking temperature.
Key idea: a Killing vector $\xi$ solves $\nabla_{(\mu}\xi_{\nu)}=0$ ; it generates an isometry, gives a conserved quantity $\xi_\mu p^\mu$ on every geodesic, and is determined globally by $\xi^\mu$ and $\nabla_{[\mu}\xi_{\nu]}$ at one point.

Where this lives today

Killing vectors are working tools at the frontier, not historical furniture. The Event Horizon Telescope's images of M87\ and Sgr A\ are read through exactly the Killing structure worked above: the ring and its inner shadow depend on the photon sphere ( $r=3GM$ ) and on gravitational redshift, both fixed by the timelike Killing vector, while the ISCO sets the inner edge of the accretion flow. In gravitational-wave astronomy the ringdown after two black holes merge is a spectrum of quasinormal modes governed by the surface gravity $\kappa$ of the final Killing horizon — LIGO and Virgo measure these to test whether the remnant is truly Kerr. And Kerr's two Killing vectors cannot integrate its geodesics; the fourth constant, Carter's, comes from a Killing tensor $\nabla_{(\mu}K_{\nu\rho)}=0$ encoding a hidden symmetry with no vector field behind it, and the search for such tensors probes whether astrophysical black holes deviate from Kerr.

The physical picture

Picture sliding the entire metric field a small amount along the integral curves of a vector field $\xi$ . If, after sliding by any amount, every measured distance and angle is exactly what it was — the geometry is indistinguishable from its original — then the flow is an isometry and $\xi$ is a Killing vector. It is the curved-space version of picking up a flat sheet and translating or rotating it: the sheet lands on a congruent copy of itself. On a round sphere, spinning about any axis carries each point along a circle of latitude while leaving all distances fixed; those rotations are the sphere's Killing flows. Deform the sphere into a lumpy potato and no rigid motion survives. Killing's equation $\nabla_{(\mu}\xi_{\nu)}=0$ is the infinitesimal bookkeeping of that statement. The gradient $\nabla_\mu\xi_\nu$ splits into a symmetric part — the deformation tensor of expansion and shear, which stretch and squash distances — and an antisymmetric rotation part, which does not. A rigid motion is allowed the rotation but forbidden the stretching, so its symmetric part must vanish. That is exactly Killing's equation.

Figure

On the round 2-sphere the rotation ξ = ∂φ carries each point along a circle of latitude while every distance is preserved, so ξ is Killing and the sphere has n(n+1)/2 = 3 independent Killing vectors forming SO(3). On a lumpy surface a generic flow stretches a test circle into an ellipse, distances change, and ∇(μ ξν) ≠ 0 — no Killing vector survives

Because the second derivatives of a Killing vector are locked to curvature, $\xi$ everywhere is reconstructed from just its value (a translation) and its curl (a rotation) at any one point — which is why the maximum is $n$ translations plus $n(n-1)/2$ rotations. Each independent Killing vector hands every orbit and light ray a label it cannot shed. In Schwarzschild the time-translation $\partial_t$ and axial rotation $\partial_\phi$ give the conserved $E$ and $L$ that flatten geodesic motion into $\tfrac12\dot r^2+V_{\rm eff}(r)=\tfrac12 E^2$ — and every landmark of the orbit sits in the shape of that one potential.

Figure

Schwarzschild effective potential V_eff = ½(1−2GM/r)(1+L²/r²) built from the conserved E and L that the two Killing vectors supply. The blue curve (L = 4GM) has a barrier and a well — an unstable and a stable circular orbit. The amber curve (L = √12 GM) is marginal: its saddle sits exactly at the innermost stable circular orbit r = 6GM, and the photon sphere is at r = 3GM

The core equations

The starting point is a demand: the geometry must look identical after the flow. The cleanest way to say "a tensor is unchanged by the flow of $\xi$ " is that its Lie derivative — the connection-free rate of change along the flow — vanishes. Applied to the metric this is the isometry condition.

\mathcal{L}_\xi g_{\mu\nu} = 0 \qquad (1)

where: $\mathcal{L}_\xi$ — Lie derivative along $\xi$ ; $g_{\mu\nu}$ — spacetime metric; $\xi^\mu$ — the Killing vector field generating the isometry. Physical reading: dragging the metric along $\xi$ returns the same metric, so the flow is a rigid motion of spacetime. Equation (1) is geometric but awkward, hiding Christoffel symbols inside the Lie derivative. Expanding it with the Levi-Civita connection and using metric compatibility ( $\nabla_\rho g_{\mu\nu}=0$ ) turns it into a clean, covariant, first-order equation — the working form.

\nabla_\mu \xi_\nu + \nabla_\nu \xi_\mu = 0 \qquad (2)

where: $\nabla_\mu$ — Levi-Civita covariant derivative; $\xi_\nu=g_{\nu\rho}\xi^\rho$ — the Killing covector. Physical reading: the symmetric part of $\nabla_\mu\xi_\nu$ vanishes; only an antisymmetric rotation part survives, so the flow neither expands nor shears — the signature of a distance-preserving motion. The single most useful consequence follows at once. Contract $\xi_\mu$ with the tangent $u^\mu$ of an affine geodesic and differentiate: the geodesic equation kills one term and Killing's equation the other, leaving a constant of motion.

\frac{\mathrm{d}}{\mathrm{d}\lambda}\big(\xi_\mu u^\mu\big) = 0 \qquad (3)

where: $u^\mu$ — tangent to an affinely parametrized geodesic, $u^\nu\nabla_\nu u^\mu=0$ ; $\lambda$ — affine parameter. Physical reading: every Killing vector supplies one conserved quantity per geodesic — $-\xi_{(t)\mu}u^\mu=E$ and $\xi_{(\phi)\mu}u^\mu=L$ in a stationary, axisymmetric spacetime. Combining the Ricci identity with the first Bianchi identity ties the second derivative of $\xi$ directly to curvature — the structural heart of the subject, making Killing's equation a closed first-order system.

\nabla_\mu \nabla_\nu \xi_\rho = R_{\rho\nu\mu}{}^{\sigma}\,\xi_\sigma \qquad (4)

where: $R_{\rho\nu\mu}{}^{\sigma}$ — Riemann curvature tensor (Wald sign convention). Physical reading: the second derivatives are curvature times $\xi$ , not free data, so specifying $\xi_\rho$ and the antisymmetric $\nabla_{[\mu}\xi_{\nu]}$ at one point determines $\xi$ on the whole connected manifold. Taking the trace of Killing's equation with the inverse metric shows the flow is incompressible: it moves the geometry without changing any volume.

\nabla_\mu \xi^\mu = 0 \qquad (5)

where: $\nabla_\mu\xi^\mu$ — divergence of the Killing field. Physical reading: the flow preserves the invariant volume element $\sqrt{-g}\,\mathrm{d}^n x$ , consistent with a rigid motion, and this is what makes the Komar surface integrals well-defined. Finally, because the stress-energy tensor is symmetric and conserved, a Killing vector converts it into a conserved four-current — the field-theoretic face of Noether's theorem, valid for continuous matter, not just point particles on geodesics.

\nabla_\mu\big(T^{\mu\nu}\xi_\nu\big)=0 \qquad (6)

where: $T^{\mu\nu}$ — stress-energy tensor with $\nabla_\mu T^{\mu\nu}=0$ . Physical reading: $J^\mu=T^{\mu\nu}\xi_\nu$ is a conserved current whose flux through a spatial slice is the total conserved charge carried by the matter.

Derivation AFrom metric invariance to Killing's equation

The goal is to show that the geometric demand "the flow of $\xi$ is an isometry", written compactly as $\mathcal{L}_\xi g_{\mu\nu}=0$ , is identical to the covariant tensor equation $\nabla_{(\mu}\xi_{\nu)}=0$ . The route is to write the Lie derivative of the metric in ordinary partial derivatives, promote each to a covariant one, and let metric compatibility do the rest. The payoff is that a coordinate-dependent-looking condition on the metric becomes a manifestly covariant equation solvable in any frame, whose flat-space limit reproduces the ten Poincaré generators exactly.

Step 1. Write the Lie derivative of the $(0,2)$ metric tensor from its definition, $\mathcal{L}_\xi g_{\mu\nu}=\xi^\rho\partial_\rho g_{\mu\nu}+g_{\rho\nu}\partial_\mu\xi^\rho+g_{\mu\rho}\partial_\nu\xi^\rho$ . Why: the Lie derivative measures the change of $g$ along the flow of $\xi$ and is defined with plain partial derivatives, needing no connection — it is a pure statement about how the flow reshuffles points.

Step 2. Replace each $\partial$ by the covariant $\nabla$ of the Levi-Civita connection: $\mathcal{L}_\xi g_{\mu\nu}=\xi^\rho\nabla_\rho g_{\mu\nu}+g_{\rho\nu}\nabla_\mu\xi^\rho+g_{\mu\rho}\nabla_\nu\xi^\rho$ . Why: the extra Christoffel terms introduced by the three substitutions cancel identically because the connection is torsion-free, so the Lie derivative is unchanged — this is legitimate precisely for a symmetric connection.

Step 3. Invoke metric compatibility, $\nabla_\rho g_{\mu\nu}=0$ , which deletes the first term and lets the metric slide through the remaining covariant derivatives: $\mathcal{L}_\xi g_{\mu\nu}=g_{\rho\nu}\nabla_\mu\xi^\rho+g_{\mu\rho}\nabla_\nu\xi^\rho=\nabla_\mu\xi_\nu+\nabla_\nu\xi_\mu$ . Why: the Levi-Civita connection annihilates $g$ , so lowering the index on $\xi^\rho$ is free and the messy first term simply disappears.

Step 4. Impose the isometry condition $\mathcal{L}_\xi g_{\mu\nu}=0$ to read off $\nabla_\mu\xi_\nu+\nabla_\nu\xi_\mu=0$ , i.e. $\nabla_{(\mu}\xi_{\nu)}=0$ . Why: setting the Lie derivative to zero is exactly the demand that the geometry be invariant, and the result is Killing's equation — the symmetric part of the gradient of $\xi$ vanishes. Cross-check: in flat space with Cartesian coordinates $\Gamma=0$ , the equation becomes $\partial_\mu\xi_\nu+\partial_\nu\xi_\mu=0$ , solved by $\xi_\mu=a_\mu+\omega_{\mu\nu}x^\nu$ with $\omega_{\mu\nu}=-\omega_{\nu\mu}$ — four $a_\mu$ (translations) plus six $\omega_{\mu\nu}$ (three rotations, three boosts), the ten Poincaré generators (Carroll §3.8).

Go deeper · Derivation B · 2 minConserved charges and the effective potential of Schwarzschild orbits

Derivation BConserved charges and the effective potential of Schwarzschild orbits

The goal is to turn the two obvious symmetries of the Schwarzschild metric into the conserved energy $E$ and angular momentum $L$ of a test particle, then collapse the four geodesic equations into a single one-dimensional energy equation. The strategy: read the two Killing vectors off the coordinate independence of the metric, lower their indices, contract with the four-velocity to build the constants, and feed them into the mass-shell normalization $g_{\mu\nu}u^\mu u^\nu=-1$ . What emerges, $\tfrac12\dot r^2+V_{\rm eff}=\tfrac12 E^2$ , is the master equation from which the ISCO and photon sphere fall out immediately.

Step 1. In Schwarzschild coordinates $g_{\mu\nu}$ depends on neither $t$ nor $\phi$ , so the coordinate basis vectors $\xi^\mu_{(t)}=(1,0,0,0)$ and $\xi^\mu_{(\phi)}=(0,0,0,1)$ satisfy $\mathcal{L}_\xi g=0$ and are Killing. Why: $\mathcal{L}_{\partial_a}g_{\mu\nu}=\partial_a g_{\mu\nu}$ for a coordinate vector, which vanishes exactly when the metric components are independent of $x^a$ — the manifest symmetries of the line element.

Step 2. Lower the indices with $g_{tt}=-(1-2GM/r)$ and $g_{\phi\phi}=r^2\sin^2\theta$ to get $\xi^{(t)}_\mu=(-(1-2GM/r),0,0,0)$ and $\xi^{(\phi)}_\mu=(0,0,0,r^2\sin^2\theta)$ . Why: the conserved quantity is $\xi_\mu u^\mu$ with a lowered index, so the metric factors that will become the redshift function and the moment arm must be brought in now.

Step 3. Contract with the four-velocity to define $E\equiv-\xi^{(t)}_\mu u^\mu=(1-2GM/r)\dot t$ and $L\equiv\xi^{(\phi)}_\mu u^\mu=r^2\sin^2\theta\,\dot\phi$ , both constant along any geodesic by equation (3). Why: the minus sign makes $E$ the physically positive energy per unit mass (rest plus kinetic at infinity), while $L$ is the angular momentum per unit mass — the two labels the orbit cannot lose.

Step 4. Restrict to the equatorial plane $\theta=\pi/2$ and substitute $E$ , $L$ into the normalization $g_{\mu\nu}u^\mu u^\nu=-1$ ; algebra collapses it to $\tfrac12\dot r^2+V_{\rm eff}(r)=\tfrac12 E^2$ with $V_{\rm eff}=\tfrac12(1-2GM/r)(1+L^2/r^2)$ . Why: the two conservation laws eliminate $\dot t$ and $\dot\phi$ , leaving a single Newtonian-looking radial equation whose potential carries the full relativistic content — the $L^2/r^2$ centrifugal term multiplied by the redshift factor is what creates an innermost stable orbit. Verified (Hartle §9, Carroll §5.4): circular orbits sit at extrema of $V_{\rm eff}$ , existing as a stable/unstable pair only for $L^2>12(GM)^2$ ; the two merge at the ISCO $r=6GM$ when $L^2=12(GM)^2$ , and the null version of the potential gives the photon sphere at $r=3GM$ .

Go deeper · Derivation C · 1 minThe maximum number of Killing vectors

Derivation CThe maximum number of Killing vectors

The goal is to derive the ceiling $N_{\max}=n(n+1)/2$ on the number of independent Killing vectors an $n$ -dimensional (pseudo-)Riemannian manifold can carry, and to see why only constant-curvature spaces saturate it. The engine is the integrability identity (4): once the second derivative of $\xi$ is fixed by curvature and $\xi$ itself, Killing's equation is a closed first-order transport system, and the solution space is just the free data at a single point — which separates into translation-type and rotation-type pieces, giving the formula and the ten symmetries of four dimensions.

Step 1. Introduce $A_{\mu\nu}\equiv\nabla_\mu\xi_\nu$ ; Killing's equation makes it antisymmetric, $A_{\mu\nu}=A_{[\mu\nu]}$ , and identity (4) gives its derivative, $\nabla_\mu A_{\nu\rho}=R_{\rho\nu\mu}{}^\sigma\xi_\sigma$ . Why: this pair, $\nabla_\mu\xi_\nu=A_{\mu\nu}$ and $\nabla_\mu A_{\nu\rho}=(\text{curvature})\,\xi$ , is a closed first-order system — the derivatives of the unknowns are expressed back in terms of the unknowns, so a solution is a transport problem, not a genuine PDE with functional freedom.

Step 2. Count the free data at one point: $\xi_\mu$ has $n$ components and the antisymmetric $A_{\mu\nu}$ has $n(n-1)/2$ . Why: because the system transports $(\xi_\mu,A_{\mu\nu})$ uniquely along any curve, a solution is fixed once these values are chosen at a single point — the dimension of the space of Killing vectors cannot exceed the number of independent initial values.

Step 3. Add the contributions, $N_{\max}=n+\tfrac{n(n-1)}{2}=\tfrac{n(n+1)}{2}$ , with equality only when the transport is integrable for every choice of initial data, which forces constant curvature. Why: the $n$ counts translation generators and the $n(n-1)/2$ counts rotation/boost generators; the bound is reached solely by maximally symmetric spaces — Minkowski, de Sitter, anti-de Sitter — where no integrability obstruction ever appears. In $n=4$ this gives $4\cdot5/2=10$ , matching the Poincaré group of Minkowski and the $SO(1,4)$ or $SO(2,3)$ isometry groups of (anti-)de Sitter space (Wald §C.3, Weinberg ch. 13).

Limiting cases

How many isometries a spacetime carries — and which algebra they form — is a direct measure of its symmetry class:

Case	Result	Meaning
Minkowski (flat, maximally symmetric)	$\dim=\tfrac{n(n+1)}{2}=10$	4 translations, 3 rotations, 3 boosts — the Poincaré algebra in 4D
de Sitter / anti-de Sitter (curved, maximally symmetric)	$\dim=\tfrac{n(n+1)}{2}=10$	same maximal count, but the algebra is $SO(1,4)$ or $SO(2,3)$ , not Poincaré
Kerr (rotating black hole, stationary axisymmetric)	$\dim=2$	one timelike $\partial_t$ plus one axial $\partial_\phi$ ; the fourth constant needs a Killing tensor, not a vector
Schwarzschild (static, spherical)	$\dim=4$	one timelike $\partial_t$ plus the three $SO(3)$ rotation generators of the 2-sphere
Generic spacetime	$\dim=0$	no Killing vectors at all — symmetry is non-generic and must be imposed by hand

Common trapThe conserved quantity is ξ·p on any geodesic, so any coordinate vector must be a symmetry — both halves are traps

Two errors travel together. First, the conserved quantity is $\xi_\mu p^\mu$ built with the momentum $p^\mu$ (or four-velocity), and conservation holds only along a geodesic — the derivation uses $u^\nu\nabla_\nu u^\mu=0$ to drop a term, and off a geodesic that term survives so $\xi_\mu p^\mu$ drifts. For continuous matter the honest conserved object is the current $\nabla_\mu(T^{\mu\nu}\xi_\nu)=0$ , needing only that $T^{\mu\nu}$ is symmetric and conserved. Second, a coordinate basis vector $\partial_a$ is Killing only if the metric is actually independent of $x^a$ , $\partial_a g_{\mu\nu}=0$ ; the Minkowski metric in spherical or Rindler coordinates does not make every $\partial_a$ a symmetry, and bad coordinates can hide a genuine Killing vector behind messy components. Being a coordinate vector is neither necessary nor sufficient — only $\mathcal{L}_\xi g=0$ is.

Worked examples

1 · Redshift, the innermost stable orbit, and accretion efficiency (numeric)

The two Killing vectors of Schwarzschild fix both the gravitational redshift and the innermost stable circular orbit (ISCO). Take Sgr A\ with $M=4.3\times10^6\,M_\odot$ . Find (a) the redshift of light climbing from the photon sphere to infinity, (b) the ISCO radius in km and AU, and (c) the maximum energy a thin accretion disk can radiate as a fraction of rest mass.*

Step 1. For observers at rest along the timelike Killing vector $\xi_{(t)}$ , the conserved photon energy $\xi_\mu p^\mu$ ties measured frequencies to the norm $-\xi^2=(1-2GM/r)$ , giving $\nu_\infty/\nu_{\rm em}=\sqrt{1-2GM/r_{\rm em}}$ . From the photon sphere $r=3GM$ the factor is $\sqrt{1/3}=0.577$ , a redshift $z=\sqrt{3}-1=\mathbf{0.73}$ . A conservation law made visible as a frequency shift.

Step 2. The ISCO sits at $r_{\rm ISCO}=6GM/c^2$ . With $GM_\odot/c^2=1.476$ km, $GM/c^2=1.476\times4.3\times10^6=6.35\times10^6$ km, so $r_{\rm ISCO}=6\times6.35\times10^6=\mathbf{3.8\times10^7\ km}$ . Dividing by $1\ \mathrm{AU}=1.496\times10^8$ km gives $r_{\rm ISCO}=\mathbf{0.25\ AU}$ — the inner edge of any accretion flow around Sgr A\* lies at a quarter of Earth's orbital radius.

Step 3. The specific energy of a circular orbit is $E=(1-2GM/r)/\sqrt{1-3GM/r}$ ; at $r=6GM$ this is $E=(2/3)/\sqrt{1/2}=\sqrt{8/9}=\mathbf{0.9428}$ . A particle spiralling from rest at infinity ( $E=1$ ) to the ISCO radiates the difference, so the maximum efficiency of a Schwarzschild thin disk is $1-\sqrt{8/9}=\mathbf{5.7\%}$ of rest mass — several times the $\sim0.7\%$ of hydrogen fusion, which is why accreting black holes are the most luminous steady engines in the universe.

2 · The three Killing vectors of the round 2-sphere (full derivation)

Solve Killing's equation on the unit 2-sphere, $\mathrm{d}s^2=\mathrm{d}\theta^2+\sin^2\theta\,\mathrm{d}\phi^2$ , and confirm the count $n(n+1)/2=3$ .

Step 1. The only nonzero Christoffel symbols are $\Gamma^\theta_{\phi\phi}=-\sin\theta\cos\theta$ and $\Gamma^\phi_{\theta\phi}=\cot\theta$ . Writing $\nabla_{(\mu}\xi_{\nu)}=0$ gives three components: $\partial_\theta\xi_\theta=0$ ; $\partial_\phi\xi_\phi+\sin\theta\cos\theta\,\xi_\theta=0$ ; and $\partial_\theta\xi_\phi+\partial_\phi\xi_\theta-2\cot\theta\,\xi_\phi=0$ . Three scalar equations, matching the three components of a symmetric $2\times2$ tensor.

Step 2. One solution is immediate: $\xi_{(3)}=\partial_\phi$ , i.e. $\xi_\phi=\sin^2\theta$ , $\xi_\theta=0$ , since the metric is $\phi$ -independent. This is rotation about the polar axis and carries the conserved $L_z$ . The manifest azimuthal symmetry supplies the first generator.

Step 3. Solving the remaining coupled equations yields two more, $\xi_{(1)}=\sin\phi\,\partial_\theta+\cot\theta\cos\phi\,\partial_\phi$ and $\xi_{(2)}=\cos\phi\,\partial_\theta-\cot\theta\sin\phi\,\partial_\phi$ — rotations about the two equatorial axes, exactly $L_x$ and $L_y$ written on the sphere. The full $SO(3)$ appears.

Step 4. The three fields close under the bracket, $[\xi_{(1)},\xi_{(2)}]=-\xi_{(3)}$ and cyclic, reproducing the $\mathfrak{so}(3)$ algebra. The count is 3, saturating $n(n+1)/2=2\cdot3/2=3$ — the 2-sphere is maximally symmetric, as constant positive curvature demands. A maximally symmetric space carries the full complement of Killing vectors.

Physics of the portrait

Killing vectors are one of those rare places where nineteenth-century pure mathematics, invented with no physics in mind, turned out to be the exact language a later physical theory needed.

Origins · 1 minWilhelm Killing and how this was discovered

Wilhelm Killing

1847–1923 · the geometer who wrote down the equation of infinitesimal isometry

Wilhelm Karl Joseph Killing was a German mathematician who worked far from the research centres of his day, teaching at Braunsberg and later as professor at Münster, yet he shaped two fields at once. Studying space forms and the "infinitesimal motions" a geometry can admit, he introduced in the early 1890s the differential equation that now bears his name — the condition that a vector field generate a continuous family of distance-preserving transformations. Earlier, between 1888 and 1890, he carried out the astonishing classification of the simple Lie algebras through what are now called root systems, discovering the exceptional algebras $G_2$ , $F_4$ , $E_6$ , $E_7$ , $E_8$ ; his rough arguments were later made rigorous by Élie Cartan, who credited Killing with the essential ideas. For decades his geometric equation was a curiosity in Riemannian geometry. Then general relativity arrived, and the infinitesimal isometries he had studied became precisely the symmetries of spacetime — the objects that produce conserved energy and angular momentum, define stationary and static geometries, and give a horizon its temperature. A quiet provincial professor had, twice over, written down structures twentieth-century physics could not do without.

History

1847

— Wilhelm Killing is born in Burbach, Westphalia

1888

— Killing classifies the simple Lie algebras via root systems, later made rigorous by Cartan

1892

— Killing formulates the equation for infinitesimal isometries of a space form

1918

— Noether's theorem ties every continuous symmetry to a conserved quantity

1963

— Kerr finds the rotating vacuum solution, carrying a timelike and an axial Killing vector

1970

— Walker and Penrose trace Carter's fourth constant of Kerr motion to a Killing tensor

Going further

Texts. Misner, Thorne & Wheeler, Gravitation, §25.2 · Wald, General Relativity, §C.3 and §11.1–11.2 · Carroll, Spacetime and Geometry, §3.8 and §5.2–5.4 · Schutz, A First Course in General Relativity, §6.4–6.5 · Hartle, Gravity, §8.2, §9, §22.4 · Weinberg, Gravitation and Cosmology, ch. 13 (maximally symmetric spaces). Key results collected here for reference: the integrability identity $\nabla_\mu\nabla_\nu\xi_\rho=R_{\rho\nu\mu}{}^\sigma\xi_\sigma$ and the $n(n+1)/2$ maximum count; the Schwarzschild effective potential $V_{\rm eff}=\tfrac12(1-2GM/r)(1+L^2/r^2)$ , ISCO $r=6GM$ , photon sphere $r=3GM$ , circular-orbit energy $E=(1-2GM/r)/\sqrt{1-3GM/r}$ , and ISCO value $E=\sqrt{8/9}=0.9428$ (thin-disk efficiency $5.7\%$ ); the Sgr A\* figures use $M=4.3\times10^6\,M_\odot$ and $GM_\odot/c^2=1.476$ km.

Builds on: Covariant Derivative · Metric Compatibility · The Levi-Civita Connection · Christoffel Symbols · The Geodesic Equation. Leads to: Geodesics as Extremal Paths · Tidal Forces in General Relativity · Geodesic Deviation Equation.

Ask the tutor: Want to see how the same two Killing vectors that fix Schwarzschild's ISCO get generalized in Kerr, where a hidden Killing tensor rescues the fourth constant of motion?

What is the difference between a Killing vector and an ordinary coordinate basis vector?

A coordinate basis vector $\partial_a$ is just an axis of whatever chart you use; it exists in every coordinate system. It becomes Killing only when the metric genuinely does not depend on that coordinate, $\partial_a g_{\mu\nu}=0$ , because then $\mathcal{L}_{\partial_a}g=0$ . So " $\partial_t$ is Killing" is a statement about the geometry, not the coordinates: it says the spacetime is time-translation invariant. In bad coordinates a real symmetry can be hidden behind complicated components, or a $\partial_a$ can fail to be Killing even in flat space — spherical coordinates on Minkowski make $\partial_r$ non-Killing. The invariant test is always Killing's equation, never the look of the basis.

Why must the orbit be a geodesic for ξ·u to be conserved?

Differentiating $\xi_\mu u^\mu$ along the curve splits it into two pieces: one proportional to $\nabla_{(\mu}\xi_{\nu)}$ , which vanishes by Killing's equation, and one proportional to the acceleration $u^\nu\nabla_\nu u^\mu$ , which vanishes only on an unforced geodesic. Put a rocket on the particle and that acceleration term survives, so $\xi_\mu u^\mu$ drifts — the symmetry no longer protects it. The universal version needs no geodesics: for continuous matter $J^\mu=T^{\mu\nu}\xi_\nu$ is conserved, $\nabla_\mu J^\mu=0$ , because $T^{\mu\nu}$ is both symmetric (meeting the symmetric Killing equation) and divergence-free.

Does every spacetime have Killing vectors?

No — and that is the generic situation. A random metric admits none, because Killing's equation is heavily overdetermined: it imposes $n(n+1)/2$ equations on the $n$ components of $\xi$ , so solutions exist only when the geometry is specially arranged. The count runs from zero, for a lumpy cosmology or a generic gravitational-wave spacetime, up to the maximum $n(n+1)/2$ for constant-curvature spaces. Realistic solutions sit between: Schwarzschild four, Kerr two. The number is a coordinate-invariant measure of symmetry, and it is precisely the symmetries that make a metric solvable — which is why almost every exact solution we know is highly symmetric.

What distinguishes a stationary spacetime from a static one?

Both are defined through a timelike Killing vector $\xi$ . A spacetime is stationary if it simply admits one — the geometry is unchanging in time, though there may be a preferred sense of rotation. It is static if, in addition, $\xi$ is hypersurface-orthogonal, $\xi_{[\mu}\nabla_\nu\xi_{\rho]}=0$ , so $\xi$ is everywhere perpendicular to a family of spatial slices; this rules out rotation and makes the geometry symmetric under $t\to-t$ . Schwarzschild is static: nothing swirls, and the metric has no $\mathrm{d}t\,\mathrm{d}\phi$ cross term. Kerr is only stationary: the rotating hole drags spacetime around it, so its timelike Killing vector is not hypersurface-orthogonal and time reversal fails. The distinction is the geometric fingerprint of rotation.

What is a Killing horizon, and how does it give a black hole a temperature?

A Killing horizon is a null surface on which some Killing vector $\chi$ becomes null, $\chi^\mu\chi_\mu=0$ . For Schwarzschild the horizon at $r=2GM$ is where $\partial_t$ turns null; for Kerr it is a corotating combination $\partial_t+\Omega_H\partial_\phi$ . There $\chi^\nu\nabla_\nu\chi^\mu=\kappa\,\chi^\mu$ defines the surface gravity $\kappa$ , which the zeroth law shows is constant over the horizon. That constancy makes $\kappa$ act like a temperature: the first law reads $\mathrm{d}M=\tfrac{\kappa}{8\pi G}\mathrm{d}A+\Omega_H\,\mathrm{d}J$ , and Hawking's calculation fixes $T=\kappa/2\pi$ . The whole thermodynamics of black holes hangs on the Killing structure of the horizon.

If two Killing vectors are not enough to solve Kerr orbits, what fills the gap?

Integrating geodesics in four dimensions needs four constants. Kerr supplies only two from its Killing vectors — energy from $\partial_t$ , axial angular momentum from $\partial_\phi$ — plus the mass-shell norm $g_{\mu\nu}u^\mu u^\nu=-1$ , leaving the system one short. The missing constant is Carter's, and it comes not from a Killing vector but from a Killing tensor $K_{\mu\nu}$ , a symmetric object satisfying $\nabla_{(\mu}K_{\nu\rho)}=0$ whose contraction $K_{\mu\nu}u^\mu u^\nu$ is conserved along geodesics. Such a tensor encodes a hidden symmetry with no continuous transformation behind it. Its existence makes Kerr geodesics completely integrable and separable, and testing for it probes whether astrophysical black holes are exactly Kerr.