Killing Tensors and Hidden Symmetries

A continuous symmetry of spacetime — a direction you can slide along without the metric noticing — is encoded by a Killing vector, and Noether's theorem turns each one into a quantity conserved along every geodesic, linear in the momentum. That is the whole story of energy and angular momentum in a symmetric field, but not the whole story of what is conserved. When Brandon Carter set out in 1968 to integrate the orbits of the rotating Kerr black hole, the two obvious symmetries — time-translation and axial rotation — plus the mass shell gave only three constants, one short of the four a four-dimensional system needs to be solvable by quadratures. A fourth constant existed anyway: quadratic in the momenta, descending from no isometry of the metric, its geometric source a symmetric rank-2 tensor obeying a higher-rank cousin of the Killing equation — a Killing tensor. This article develops the object behind that hidden symmetry. A Killing tensor is a totally symmetric tensor whose fully symmetrized covariant derivative vanishes, and its contraction with the right number of geodesic tangents is automatically constant. The metric is the trivial example, products of Killing vectors give reducible ones that carry nothing new, and the interesting case is an irreducible Killing tensor — present only in special spacetimes, its charge independent of the manifest symmetries. Kerr is the canonical home of one, and its charge is the Carter constant. Beneath it sits an even sharper object, the Killing-Yano tensor, an antisymmetric form whose "square" is the Killing tensor and which additionally separates the Dirac equation — together they explain why the geodesics, scalar wave, and spinor field of a rotating black hole all come apart into ordinary differential equations.

Key takeaways

A Killing tensor is a totally symmetric rank-n tensor with $\nabla_{(\lambda}K_{\mu_1\cdots\mu_n)}=0$ , the higher-rank generalization of the Killing vector equation $\nabla_{(\mu}\xi_{\nu)}=0$ .
Each gives a conserved charge $K_{\mu_1\cdots\mu_n}\dot x^{\mu_1}\cdots\dot x^{\mu_n}=\text{const}$ along geodesics — a polynomial in momentum that need not come from any spacetime isometry, hence a hidden symmetry.
The metric is the trivial rank-2 Killing tensor; reducible ones are symmetrized products of Killing vectors; only irreducible ones carry new information, and generic spacetimes admit none.
The Carter constant of Kerr is the charge of an irreducible rank-2 Killing tensor — the missing fourth constant that makes Kerr geodesics Liouville-integrable and separates the Hamilton-Jacobi and Klein-Gordon equations.
The Kerr Killing tensor is the square $K_{\mu\nu}=Y_{\mu\lambda}Y^{\lambda}{}_{\nu}$ of a Killing-Yano two-form, the deeper object that additionally separates the Dirac equation and generates the symmetry tower.
Key idea: a symmetric tensor whose fully symmetrized covariant derivative vanishes produces a polynomial-in-momentum constant of geodesic motion; when irreducible it is a hidden symmetry with no underlying continuous isometry — the Carter constant of Kerr being the canonical case.

Where this lives today

Hidden symmetry is load-bearing in current gravitational-wave science. The Laser Interferometer Space Antenna (LISA), due to fly in the mid-2030s, will detect extreme-mass-ratio inspirals — a small compact object spiraling through tens of thousands of orbits around a supermassive Kerr black hole — and extracting those signals demands waveform templates of exquisite phase accuracy. That is only tractable because the geodesics are integrable: the Carter constant labels the orbit by action variables evolved adiabatically as radiation slowly changes $E$ , $L_z$ , and $Q$ . The same integrability underlies the analytic ray-tracing behind the Event Horizon Telescope's shadow images, whose photon-ring structure is set by spherical photon orbits classified through their Carter constant. The Killing-Yano tensor beneath it separates the Teukolsky equation, the workhorse of quasinormal-mode "ringdown" spectroscopy now tested against LIGO-Virgo-KAGRA data. And because irreducible Killing tensors are non-generic, their presence is a sharp diagnostic: searching pulsar-timing and orbital data for a Carter-like fourth constant tests whether a compact object really is Kerr — a direct probe of the no-hair expectation. In higher dimensions a single principal conformal Killing-Yano tensor in Kerr-NUT-AdS spacetimes generates a whole tower of Killing tensors, guaranteeing integrability and separability in every dimension.

The physical picture

A Killing vector makes a promise about spacetime itself: slide everything along its flow and the metric is unchanged, and the reward is a charge $p_\mu\xi^\mu$ linear in momentum and conserved along every geodesic. A Killing tensor makes a weaker but still rigid promise one level up — not in spacetime but in phase space. It says a particular quadratic combination of momenta is conserved even though no single direction of spacetime is symmetric. There is no flow to slide along, no picture of "the metric looks the same" — only a function on phase space that Poisson-commutes with the geodesic Hamiltonian, and that is enough to constrain the motion. The cleanest way to feel where such a thing comes from is to watch spherical symmetry break. In Schwarzschild the three rotational Killing vectors give charges whose sum of squares $L^2=L_x^2+L_y^2+L_z^2$ is conserved. Spin the hole up to Kerr and two of those rotations are destroyed — only $\partial_\phi$ survives, so only $L_z$ remains a Killing-vector charge, and naively $L^2$ should be lost. Yet a single quadratic invariant, a deformed remnant of $L^2$ , survives and is repackaged into an irreducible Killing tensor $K_{\mu\nu}$ . The Carter constant is the ghost of the rotational symmetries the spin took away — a conservation law that outlived the symmetry that once explained it.

Figure

The four constants that make a Kerr geodesic integrable: the metric gives m², the Killing vectors ∂t and ∂φ give E and Lz, and the irreducible Killing tensor K gives the Carter constant Q — four commuting constants on a 4-dimensional configuration space, which is exactly Liouville integrability

Geometrically, that $K_{\mu\nu}\dot x^\mu\dot x^\nu$ is constant means the Killing tensor pins each geodesic to a fixed quadric surface in velocity space — the quadratic analogue of the way a Killing vector pins the momentum to a hyperplane. For a Kerr orbit the most visible consequence is in the polar angle: the Carter constant alone decides the band of latitudes the orbit can reach, turning $\theta$ -motion into a one-dimensional oscillation between two turning points. The second figure is that potential made concrete for a specific orbit.

Figure

The polar potential Θ(θ) for a Kerr geodesic with a = 0.9M, E = 0.95, Lz = 2.0, m = 1 and Carter constant Q = 3.0: motion is allowed only where Θ ≥ 0, confining the orbit to a latitude band between turning points θ = 49.3° and θ = 130.7° fixed entirely by Q

The core equations

The definition is a single differential equation, the exact higher-rank echo of the Killing vector equation. Demanding that the fully symmetrized covariant derivative vanish is what makes the tensor's contraction with geodesic tangents conserved — nothing weaker will do.

\nabla_{(\lambda}K_{\mu_1\mu_2\cdots\mu_n)}=0 \qquad (1)

where: $K_{\mu_1\cdots\mu_n}$ — totally symmetric rank- $n$ tensor field; $\nabla_\lambda$ — Levi-Civita covariant derivative; $(\cdots)$ — full symmetrization over all $n+1$ indices with weight $1/(n+1)!$ . Reading: the object has $n$ tensor slots, and the derivative index is symmetrized in with them all. For $n=1$ equation (1) is nothing but the Killing vector equation, so a Killing tensor is a strict generalization: rank 1 is the manifest isometries, rank $\ge 2$ opens the door to hidden ones.

\nabla_{(\mu}\xi_{\nu)}=0 \qquad (2)

where: $\xi_\mu$ — Killing covector generating a one-parameter isometry. Reading: the symmetric part of $\nabla_\mu\xi_\nu$ vanishes, so $\xi$ 's flow drags the metric onto itself. The payoff is one line: contract the tensor with $n$ copies of the geodesic tangent and the result is a constant of the motion — a polynomial of degree $n$ in the momenta.

C=K_{\mu_1\cdots\mu_n}\,\dot x^{\mu_1}\cdots\dot x^{\mu_n}=\text{const} \qquad (3)

where: $\dot x^\mu=\mathrm{d}x^\mu/\mathrm{d}\tau$ — tangent with $\tau$ affine; $C$ — degree- $n$ polynomial in momentum, conserved along the geodesic. Rank 1 gives the linear charge $\xi_\mu\dot x^\mu$ ; rank 2 gives a quadratic one. The same fact reads most cleanly in phase space, where "conserved along geodesics" becomes "Poisson-commutes with the geodesic Hamiltonian." This is the language in which integrability is stated.

\{H,C\}=0,\qquad H=\tfrac12\,g^{\mu\nu}p_\mu p_\nu \qquad (4)

where: $H$ — geodesic Hamiltonian with $p_\mu=g_{\mu\nu}\dot x^\nu$ ; $C=K^{\mu_1\cdots\mu_n}p_{\mu_1}\cdots p_{\mu_n}$ , which brackets to zero with $H$ iff $K$ is a Killing tensor; $\{\,,\}$ — canonical Poisson bracket on the cotangent bundle. Reading: the Killing tensor equation is exactly the condition $\{H,C\}=0$ . Beneath the Kerr Killing tensor sits a sharper object: a Killing-Yano tensor, an antisymmetric two-form obeying a different symmetrization — only the derivative index and the first tensor index, leaving the form's antisymmetry intact.

\nabla_{(\lambda}Y_{\mu)\nu}=0,\qquad Y_{\mu\nu}=-Y_{\nu\mu} \qquad (5)

where: $Y_{\mu\nu}$ — antisymmetric Killing-Yano two-form; $(\lambda\mu)$ — symmetrization on the derivative index and the first tensor index only. Reading: this is not the Killing tensor equation — the index structure is genuinely different. The link between the two is algebraic: the "square" of a Killing-Yano tensor is automatically a Killing tensor. This is why $Y$ is regarded as the more fundamental "square root."

K_{\mu\nu}=Y_{\mu\lambda}\,Y^{\lambda}{}_{\nu} \qquad (6)

where: $K_{\mu\nu}$ — symmetric rank-2 Killing tensor built from $Y$ . Reading: antisymmetry of $Y$ makes $Y_{\mu\lambda}Y^\lambda{}_\nu$ symmetric in $\mu\nu$ , and Derivation B shows it solves equation (1). In the Kerr metric this machinery produces the Carter constant. The rank-2 charge splits into a piece built from the Killing-vector charges $E, L_z$ and an irreducible remainder $Q$ — that remainder is the genuinely new constant.

K_{\mu\nu}\dot x^\mu\dot x^\nu = Q + (L_z-aE)^2 \qquad (7)

where: $Q$ — Carter constant, the irreducible part of the rank-2 charge; $E$ — energy from $\partial_t$ ; $L_z$ — axial angular momentum from $\partial_\phi$ ; $a=J/M$ — Kerr spin parameter. Reading: subtract the reducible Killing-vector piece and what is left, $Q$ , is the hidden-symmetry charge.

Derivation AWhy a Killing tensor charge is conserved along geodesics

The goal is to prove that the scalar $C=K_{\mu_1\cdots\mu_n}\dot x^{\mu_1}\cdots\dot x^{\mu_n}$ is constant along any affinely parametrized geodesic, and to see why the definition demands full symmetrization rather than the weaker divergence condition. The strategy generalizes the Killing-vector argument: differentiate the charge along the curve, expand with the Leibniz rule, kill every tangent-derivative term with the geodesic equation, and dispatch the survivor with the Killing tensor equation, using that the tangents form a totally symmetric product.

Step 1. Define the candidate charge $C=K_{\mu_1\cdots\mu_n}\dot x^{\mu_1}\cdots\dot x^{\mu_n}$ along a geodesic with affine parameter $\tau$ , and note that because $C$ is a scalar its ordinary rate of change equals its directional covariant derivative, $\mathrm{d}C/\mathrm{d}\tau=\dot x^\lambda\nabla_\lambda C$ . Why: promoting $\mathrm{d}/\mathrm{d}\tau$ to $\dot x^\lambda\nabla_\lambda$ lets us use the covariant Leibniz rule and the geodesic equation, both written with $\nabla$ .

Step 2. Apply the Leibniz rule: one term where $\nabla$ hits the tensor and $n$ terms where it hits a tangent factor, $\mathrm{d}C/\mathrm{d}\tau=(\nabla_\lambda K_{\mu_1\cdots\mu_n})\dot x^\lambda\dot x^{\mu_1}\cdots\dot x^{\mu_n}+\sum_i K_{\cdots}\,\dot x^\lambda\nabla_\lambda\dot x^{\mu_i}\prod_{j\neq i}\dot x^{\mu_j}$ . Why: separating the "derivative of $K$ " term from the "derivative of a tangent" terms isolates the two physical inputs.

Step 3. Kill the tangent-derivative terms. Along a geodesic $\dot x^\lambda\nabla_\lambda\dot x^{\mu_i}=0$ , so all $n$ terms in the sum vanish, leaving only $(\nabla_\lambda K_{\mu_1\cdots\mu_n})\dot x^\lambda\dot x^{\mu_1}\cdots\dot x^{\mu_n}$ . Why: this is the sole place the geodesic condition enters — off a geodesic these terms survive and nothing is conserved.

Step 4. Symmetrize the survivor. The product $\dot x^\lambda\dot x^{\mu_1}\cdots\dot x^{\mu_n}$ is totally symmetric in all $n+1$ indices, so it picks out only the totally symmetric part of the derivative: the term equals $(\nabla_{(\lambda}K_{\mu_1\cdots\mu_n)})\dot x^\lambda\dot x^{\mu_1}\cdots\dot x^{\mu_n}$ . Why: this is the crux — a symmetric contractor sees only a symmetric tensor, which is why the condition is the fully symmetrized derivative and not the bare divergence $\nabla^\lambda K_{\lambda\mu_2\cdots}$ .

Step 5. Invoke the definition. By equation (1) the last term vanishes, so $\mathrm{d}C/\mathrm{d}\tau=0$ ; setting $n=1$ recovers the Killing-vector result $\mathrm{d}(\xi_\mu\dot x^\mu)/\mathrm{d}\tau=0$ as the base case. Why: the chain closes — geodesic equation for the tangent terms, Killing tensor equation for the last one — giving an exactly conserved charge at every rank.

Go deeper · Derivation B · 1 minThe square of a Killing-Yano tensor is a Killing tensor

Derivation BThe square of a Killing-Yano tensor is a Killing tensor

The goal is to show that if $Y_{\mu\nu}$ is an antisymmetric Killing-Yano tensor obeying $\nabla_{(\lambda}Y_{\mu)\nu}=0$ , then the contracted square $K_{\mu\nu}=Y_{\mu\lambda}Y^{\lambda}{}_{\nu}$ is a symmetric rank-2 Killing tensor obeying $\nabla_{(\rho}K_{\mu\nu)}=0$ . This is the identity that makes the Killing-Yano tensor the "square root" of the Kerr Carter constant — $Y$ is the deeper object and $K=Y\cdot Y$ inherits its conservation law — and the proof is a short exercise in the Leibniz rule plus the special index symmetry of the Killing-Yano equation.

Step 1. Check symmetry. Because $Y$ is antisymmetric, $K_{\nu\mu}=Y_{\nu\lambda}Y^\lambda{}_\mu=Y_{\mu}{}^{\lambda}Y_{\lambda\nu}=K_{\mu\nu}$ , so $K$ is a legitimate symmetric rank-2 tensor and a candidate Killing tensor. Why: only a symmetric tensor can satisfy the Killing tensor equation and give a well-defined quadratic charge.

Step 2. Differentiate with the plain Leibniz rule: $\nabla_\rho K_{\mu\nu}=(\nabla_\rho Y_{\mu\lambda})Y^\lambda{}_\nu+Y_{\mu\lambda}(\nabla_\rho Y^\lambda{}_\nu)$ . Why: the Killing tensor condition is a statement about $\nabla K$ , so we need it in terms of $\nabla Y$ , where the Killing-Yano property applies.

Step 3. Symmetrize over $\rho\mu\nu$ . Grouping the terms and using antisymmetry of $Y$ collapses the expression to $\nabla_{(\rho}K_{\mu\nu)}=2\,Y^\lambda{}_{(\nu}\nabla_\rho Y_{\mu)\lambda}$ , each piece carrying the Killing-Yano combination $\nabla_{(\rho}Y_{\mu)\lambda}$ . Why: the point is to reorganize the symmetrized derivative terms so the object we know vanishes appears explicitly.

Step 4. Apply the Killing-Yano equation: every surviving term contains $\nabla_{(\rho}Y_{\mu)\lambda}=0$ , so $\nabla_{(\rho}K_{\mu\nu)}=0$ and $K$ is a Killing tensor. The converse fails — a generic Killing tensor has no Killing-Yano root, so $Y$ carries strictly more information. Why: this asymmetry is why the modern viewpoint treats the Killing-Yano tensor, not the Killing tensor, as the fundamental generator of hidden symmetry.

Go deeper · Derivation C · 2 minSeparating the Kerr Hamilton-Jacobi equation and reading off the Carter constant

Derivation CSeparating the Kerr Hamilton-Jacobi equation and reading off the Carter constant

The goal is to show explicitly how the Kerr geodesic problem, seemingly coupled in $r$ and $\theta$ , comes apart into two independent one-dimensional problems, and to identify the separation constant that is the Carter constant $Q$ — the charge of the irreducible Killing tensor. We write the Hamilton-Jacobi equation in Boyer-Lindquist coordinates, use the two cyclic coordinates to peel off $E$ and $L_z$ , multiply by $\Sigma$ , and watch the equation split into a pure- $r$ side and a pure- $\theta$ side that can only be equal if each is constant.

Step 1. Write the geodesic Hamilton-Jacobi equation $g^{\mu\nu}\partial_\mu S\,\partial_\nu S=-m^2$ , with $p_\mu=\partial_\mu S$ and $m$ the rest mass ( $m^2=0$ for null geodesics). Why: Hamilton-Jacobi turns the conserved-momentum problem into one PDE for $S$ , and additive separability of $S$ is exactly equivalent to having enough commuting constants.

Step 2. Record the Kerr structure functions $\Sigma=r^2+a^2\cos^2\theta$ and $\Delta=r^2-2Mr+a^2$ ; the Boyer-Lindquist inverse metric organizes into $1/\Sigma$ times a clean sum of an $r$ -sector and a $\theta$ -sector. Why: this $\Sigma$ -weighted additive structure is the geometric fingerprint of the hidden symmetry — what lets the coordinates decouple after one multiplication.

Step 3. Use the cyclic coordinates. Since $t,\phi$ are absent from the metric ( $\partial_t,\partial_\phi$ are Killing vectors), their momenta are constants $E=-p_t$ , $L_z=p_\phi$ , and the action is additive, $S=-Et+L_z\phi+S_r(r)+S_\theta(\theta)$ . Why: the two manifest isometries supply two constants for free and reduce the PDE to $r$ and $\theta$ only.

Step 4. Substitute and multiply by $\Sigma$ ; the equation regroups into a function of $r$ alone equal to a function of $\theta$ alone,

\Delta\Big(\frac{\mathrm{d}S_r}{\mathrm{d}r}\Big)^2-\frac{[(r^2+a^2)E-aL_z]^2}{\Delta}+m^2r^2=-\Big(\frac{\mathrm{d}S_\theta}{\mathrm{d}\theta}\Big)^2-\cos^2\theta\Big[a^2(m^2-E^2)+\frac{L_z^2}{\sin^2\theta}\Big].

Why: an equality between a function of $r$ only and a function of $\theta$ only can hold for all $r,\theta$ only if both sides equal the same constant — this is the mechanism of separation.

Step 5. Setting each side equal to a constant defines the Carter constant; the $\theta$ -side gives the clean form

Q=\Big(\frac{\mathrm{d}S_\theta}{\mathrm{d}\theta}\Big)^2+\cos^2\theta\Big[a^2(m^2-E^2)+\frac{L_z^2}{\sin^2\theta}\Big].

Why: $Q$ is not tied to either coordinate — it is the shared value the two decoupled sectors must agree on, which is why no single symmetry direction explains it.

Step 6. This same $Q$ is the charge of the irreducible Kerr Killing tensor, $Q+(L_z-aE)^2=K_{\mu\nu}\dot x^\mu\dot x^\nu$ , so the algebraic separation constant and the geometric hidden-symmetry charge are one object. Why: separability and the existence of a Killing tensor are two faces of one structure; conventions for $Q$ differ between texts by how much of the reducible $(L_z-aE)^2$ piece is absorbed.

Limiting cases

Degenerations and special cases sharpen what a Killing tensor is — and, just as usefully, what it is not.

Case	Result	Meaning
Rank 1	$\nabla_{(\mu}\xi_{\nu)}=0$	A rank-1 Killing tensor is exactly a Killing vector; its charge $\xi_\mu\dot x^\mu$ is linear in momentum and tied to a manifest isometry.
The metric	$K_{\mu\nu}=g_{\mu\nu}\Rightarrow\nabla_{(\lambda}g_{\mu\nu)}=0$	The metric is the universal trivial rank-2 Killing tensor; its charge $g_{\mu\nu}\dot x^\mu\dot x^\nu=-m^2$ is the mass shell, already guaranteed by affine parametrization.
Reducible	$K_{\mu\nu}=c_1\,\xi^{(1)}_{(\mu}\xi^{(2)}_{\nu)}+c_2\,g_{\mu\nu}$	Symmetrized products of Killing vectors plus metric multiples — no information beyond the isometries already present.
Schwarzschild ( $a\to0$ )	$Q\to L^2-L_z^2$	The Kerr Killing tensor degenerates to a reducible one; the "hidden" constant becomes ordinary total angular momentum.
Killing-Yano square	$K_{\mu\nu}=Y_{\mu\lambda}Y^{\lambda}{}_{\nu},\ \nabla_{(\lambda}Y_{\mu)\nu}=0$	Any Killing-Yano two-form's square is a Killing tensor; the converse fails.

Common trapConfusing the Killing tensor equation with the Killing-Yano equation

The two equations look alike and are constantly conflated, but they are distinct PDEs with distinct objects. The Killing tensor equation $\nabla_{(\lambda}K_{\mu\nu)}=0$ acts on a symmetric tensor and symmetrizes over all indices, the derivative index included. The Killing-Yano equation $\nabla_{(\lambda}Y_{\mu)\nu}=0$ acts on an antisymmetric two-form and symmetrizes only the derivative index with the first tensor index, leaving the second free. The relationship is one-directional: the Killing tensor is the square $K_{\mu\nu}=Y_{\mu\lambda}Y^\lambda{}_\nu$ , never the reverse, and a Killing tensor need not descend from any Killing-Yano tensor. A subtler trap: conservation requires the fully symmetrized derivative to vanish, a genuinely stronger condition than the divergence-free $\nabla^\mu K_{\mu\nu}=0$ — do not substitute one for the other.

Worked examples

1 · The latitude band of a Kerr orbit (fully worked)

A test particle orbits a Kerr black hole with spin $a=0.9M$ (take $M=1$ ), energy $E=0.95$ , axial angular momentum $L_z=2.0$ , rest mass $m=1$ , and Carter constant $Q=3.0$ . Find the range of polar angle $\theta$ the orbit can reach.

Step 1. The polar motion obeys $(\mathrm{d}\theta/\mathrm{d}\tau)^2\propto\Theta(\theta)$ with $\Theta(\theta)=Q-\cos^2\theta\big[a^2(m^2-E^2)+L_z^2/\sin^2\theta\big]$ ; the orbit is allowed only where $\Theta\ge0$ , and the turning points ( $\dot\theta=0$ ) are the roots of $\Theta=0$ .

Step 2. Evaluate the constant coefficient: $a^2(m^2-E^2)=0.81\times(1-0.9025)=\mathbf{0.07898}$ , so with $L_z^2=4.0$ , $\Theta(\theta)=3.0-\cos^2\theta\,[\,0.07898+4.0/\sin^2\theta\,]$ .

Step 3. At the equator $\theta=90^\circ$ , $\cos^2\theta=0$ and $\Theta=Q=\mathbf{3.0}$ — the maximum, so the orbit certainly crosses the equatorial plane. Toward a pole the $L_z^2/\sin^2\theta$ term drives $\Theta$ negative: $\Theta(55^\circ)=+1.01$ , $\Theta(49^\circ)=-0.06$ , so the orbit is forbidden beyond the sign change.

Step 4. Solving $\Theta=0$ gives $\theta_{\min}=\mathbf{49.3^\circ}$ , and since $\Theta$ depends only on $\cos^2\theta$ the other turning point is $\theta_{\max}=180^\circ-49.3^\circ=\mathbf{130.7^\circ}$ . The orbit oscillates in latitude between them, in a band symmetric about the equator, never reaching a pole. The Carter constant alone fixes the band: θ oscillates between 49.3° and 130.7°.

Step 5. Sanity check the role of $Q$ : a larger $Q$ raises the curve and widens the band, while $Q\to0$ pinches it to the equatorial plane — the hidden symmetry doing concrete work, turning coupled $(r,\theta)$ motion into decoupled one-dimensional oscillations. This is exactly the potential drawn in the second figure.

2 · The Schwarzschild limit: a hidden constant becomes an ordinary one

Show that as the spin $a\to0$ the Carter constant reduces to a reducible charge built from the rotational Killing vectors, and evaluate it for an orbit with total angular momentum $L=4$ and axial component $L_z=2$ .

Step 1. Take $a\to0$ in the polar potential: the $a^2(m^2-E^2)$ term vanishes and $Q\to p_\theta^2+\cos^2\theta\,L_z^2/\sin^2\theta$ , with $p_\theta=\mathrm{d}S_\theta/\mathrm{d}\theta$ — a pure function of the angular momenta, independent of $E$ and $m$ .

Step 2. Compare with total angular momentum $L^2=p_\theta^2+L_z^2/\sin^2\theta$ . Subtracting, $L^2-L_z^2=p_\theta^2+\cos^2\theta\,L_z^2/\sin^2\theta=Q$ , so $\mathbf{Q=L^2-L_z^2}$ in the spherically symmetric limit.

Step 3. This is manifestly reducible: $L^2=L_x^2+L_y^2+L_z^2$ is built from the three rotational Killing vectors, so in Schwarzschild the "fourth constant" is nothing hidden — it is total angular momentum minus its axial part. The Kerr Killing tensor is the irreducible deformation that survives when spin destroys two of those rotations.

Step 4. Numbers: with $L=4$ and $L_z=2$ , $Q=L^2-L_z^2=16-4=\mathbf{12}$ — twelve units of non-axial angular momentum squared, all recoverable from manifest symmetries, which is why the Schwarzschild Killing tensor is reducible and the Kerr one is not. In the spherical limit the Carter constant collapses to Q = L² − L_z² = 12, a reducible charge.

Physics of the portrait

The Carter constant arrived as a computational surprise and only afterward acquired a geometric name: Carter was looking for orbits, found a fourth integral, and it took Walker and Penrose two more years to recognize the tensor behind it.

Origins · 1 minBrandon Carter and how this was discovered

Brandon Carter

b. 1942 · Australian-British relativist who found the fourth constant of Kerr geodesic motion and the tensor structure behind it

Brandon Carter, born in England in 1942 and educated at Cambridge under Dennis Sciama alongside a cohort that included Stephen Hawking and Martin Rees, spent the late 1960s on the global structure and orbital dynamics of the Kerr solution — the rotating black hole Roy Kerr had written down in 1963. In 1968, studying the Hamilton-Jacobi equation for test particles, he found that the Kerr geodesic problem separates additively, yielding a fourth conserved quantity beyond energy, axial angular momentum, and the mass shell. That quantity, now universally the Carter constant, is quadratic in the momenta and descends from no isometry; it renders Kerr geodesics completely integrable and reduces them to quadratures — the analytic foundation on which modern black-hole orbit computations, extreme-mass-ratio waveform templates, and ray-traced shadow images all rest. Two years later Walker and Penrose identified the object responsible: an irreducible rank-2 Killing tensor, itself the square of a Killing-Yano tensor allowed by Kerr's Petrov type-D structure. Carter's reach went further still — the conformal diagrams that carry his and Penrose's names, key theorems toward the no-hair conjecture, the four laws of black-hole mechanics with Bardeen and Hawking in 1973, and the phrase "anthropic principle." But it is the hidden symmetry of Kerr — a conservation law with no visible symmetry to explain it — that most directly seeded this article.

History

1892

— Wilhelm Killing studies the infinitesimal isometries later named Killing vectors

1952

— Kentaro Yano introduces the antisymmetric tensor fields now called Killing-Yano tensors

1963

— Roy Kerr discovers the metric of a rotating black hole

1968

— Brandon Carter finds the fourth constant separating Kerr geodesic motion

1970

— Walker and Penrose identify the Killing-Yano and Killing tensor structure of type-D spacetimes

Going further

Texts. Misner, Thorne & Wheeler, Gravitation (1973), §33.5 on Kerr orbits and the Carter constant · Wald, General Relativity (1984), Ch. 12 and App. C on symmetries · Carroll, Spacetime and Geometry (2004), §6 on Killing vectors and conservation laws · Hartle, Gravity (2003), Ch. 15 on orbits in Kerr · Schutz, A First Course in General Relativity (2009), Ch. 11 on Schwarzschild orbits.

Primary and review literature. Carter, Phys. Rev. 174, 1559 (1968), "Global Structure of the Kerr Family of Gravitational Fields" · Walker & Penrose, Commun. Math. Phys. 18, 265 (1970), on the Killing-Yano structure of type-D spacetimes · Frolov, Krtouš & Kubizňák, Living Rev. Relativity 20, 6 (2017), "Black holes, hidden symmetries, and complete integrability."

Provenance. The conservation theorem and reducibility follow Wald (1984) and Carroll (2004); the Kerr Carter constant and Hamilton-Jacobi separation follow Carter (1968) with MTW §33.5 and Hartle Ch. 15; the Killing-Yano origin follows Walker & Penrose (1970); the higher-dimensional principal-tensor framework follows Frolov-Krtouš-Kubizňák (2017). A verifier should recompute: (i) the $n=1$ reduction of Derivation A; (ii) $\nabla_{(\lambda}g_{\mu\nu)}=0$ from metric compatibility; (iii) the Killing-Yano-square identity of Derivation B; (iv) the polar turning points $\theta=49.3^\circ,130.7^\circ$ for $a=0.9,E=0.95,L_z=2.0,m=1,Q=3.0$ ; (v) the Schwarzschild limit $Q=L^2-L_z^2=12$ for $L=4,L_z=2$ . Conventions for $Q$ differ between texts by terms in $(L_z-aE)^2$ .

Prerequisites. The Covariant Derivative · The Levi-Civita Connection · Metric Compatibility · The Geodesic Equation · Parallel Transport · Affine Parameters. Leads to. Geodesics as Extremal Paths · Tidal Forces in General Relativity · The Geodesic Deviation Equation.

Ask the tutor: Ask me to walk through how the Carter constant separates the Kerr Hamilton-Jacobi equation, or why an irreducible Killing tensor exists in Kerr but not in a generic spacetime.

What is the difference between a Killing vector and a Killing tensor, in one sentence?

A Killing vector is a symmetry of spacetime itself — a direction along which the metric is unchanged — and produces a conserved quantity linear in momentum, $\xi_\mu\dot x^\mu$ . A Killing tensor is a symmetry of the geodesic dynamics — a function on phase space that Poisson-commutes with the Hamiltonian — and produces a polynomial (quadratic for rank 2) conserved quantity. The crucial difference is that a Killing tensor need not correspond to any visualizable motion of the spacetime; its symmetry is hidden in phase space rather than manifest in the geometry, which is exactly why the Carter constant surprised everyone.

Why is the metric always a Killing tensor, and does it ever tell us anything new?

By metric compatibility, $\nabla_\lambda g_{\mu\nu}=0$ identically, so certainly $\nabla_{(\lambda}g_{\mu\nu)}=0$ — the metric solves the rank-2 Killing tensor equation in every spacetime. Its charge $g_{\mu\nu}\dot x^\mu\dot x^\nu$ is, for an affine geodesic, just the constant norm of the tangent, $-m^2$ for a massive particle or $0$ for a photon: the mass-shell condition, already guaranteed by the affine parameter, so it carries no new information. It is the reference point against which "irreducible" is defined — a genuinely new Killing tensor is one you cannot build from the metric and Killing vectors.

What exactly does "irreducible" mean, and why do generic spacetimes have none?

A reducible Killing tensor is a linear combination of the metric and symmetrized products of Killing vectors, such as $c_1\,\xi^{(1)}_{(\mu}\xi^{(2)}_{\nu)}+c_2\,g_{\mu\nu}$ ; its charge factorizes into charges you already had from manifest isometries, so it is bookkeeping, not new physics. An irreducible one cannot be assembled that way — its charge is functionally independent of the isometry charges. The Killing tensor equation is overdetermined, and for a generic metric the only solutions are the metric and reducible combinations; irreducible solutions require special algebraic structure (Kerr is Petrov type D). That non-genericity is what makes a Carter-like constant a sharp test that a metric really is Kerr.

Where does the Carter constant come from physically if there is no symmetry behind it?

There is a structure behind it — just not an isometry of the metric. Kerr admits an irreducible rank-2 Killing tensor $K_{\mu\nu}$ (the square of a Killing-Yano tensor), and the Carter constant is its charge, $K_{\mu\nu}\dot x^\mu\dot x^\nu$ minus a reducible piece. Operationally it "comes from" the additive separability of the Hamilton-Jacobi equation: because the equation splits into a pure- $r$ part and a pure- $\theta$ part, there must be a separation constant both parts share, and that constant is $Q$ . The deep fact is that separability and the existence of a Killing tensor are equivalent — the algebraic accident and the geometric object are one thing seen two ways.

Does every Killing tensor come from a Killing-Yano tensor?

No, and this asymmetry matters. Every Killing-Yano two-form $Y$ produces a Killing tensor via the square $K_{\mu\nu}=Y_{\mu\lambda}Y^\lambda{}_\nu$ (Derivation B), so Killing-Yano tensors are a source of Killing tensors — but the converse fails: some Killing tensors have no Killing-Yano root at all. This is why the modern viewpoint treats the Killing-Yano tensor as more fundamental: when it exists it carries strictly more information. In Kerr it does extra work the Killing tensor cannot — it builds a first-order operator commuting with the Dirac operator, which separates the Dirac equation for a spin-½ field.

How does a hidden symmetry lead to "integrability," and why exactly four constants?

A geodesic in a four-dimensional spacetime lives in an eight-dimensional phase space; Liouville's theorem calls the motion integrable — solvable by quadratures — given four constants that are functionally independent and mutually Poisson-commute ("in involution"), matching the four configuration-space dimensions. In Kerr those four are $m^2$ (metric), $E$ (from $\partial_t$ ), $L_z$ (from $\partial_\phi$ ), and $Q$ (irreducible Killing tensor). With only three the $r$ - $\theta$ motion stays coupled; the hidden symmetry supplies the fourth, decoupling the radial and polar equations into independent first-order ODEs.