Symmetry Operators for Field Equations

A classical constant of motion becomes useful for a wave equation only when it lifts to a differential operator that maps solutions to solutions. On Kerr, the stationary and axial Killing vectors give first-order operators, while the rank-two Killing tensor gives a second-order Carter operator. For the scalar wave equation these operators commute and their common eigenvalues are exactly the separation labels. Higher-spin equations require field-specific lifts; formal commutation alone does not establish a globally complete mode basis.

Required background. Killing Fields and Isometry Groups supplies Lie derivatives and Killing tensors; Events, Intervals, and Minkowski Geometry supplies local Lorentz and field-component conventions.

Helpful background. Black-Hole Geometry and Mechanics fixes Kerr; PDEs and Green Functions for Gravitation supplies operator domains and spectral language; Formulations of General Relativity distinguishes equation symmetries from gauge; Frames, Coframes, and Anholonomy prepares the Killing–Yano and spinor constructions.

Scalar Carter-operator theorem

A differential symmetry is a way of producing another solution without solving the equation again. If $\mathcal L\Phi=0$ and $[\mathcal L,\mathcal S]=0$, then

$$\mathcal L(\mathcal S\Phi)=\mathcal S(\mathcal L\Phi)=0.$$

Here $[A,B]=AB-BA$ denotes composition, including differentiation of the coefficients of the operator on the right. More generally, $\mathcal L\mathcal S=\widetilde{\mathcal S}\mathcal L$ suffices to map solutions to solutions; this is an intertwining identity, not necessarily a vanishing commutator.

Assume a smooth four-dimensional vacuum Kerr exterior. Write $\mu=GM>0$, $a=J/M$, $|a|\leq\mu$, and

$$\Delta=r^2-2\mu r+a^2,\qquad \Sigma=r^2+a^2\cos^2\theta,\qquad B=a^2\cos^2\theta.$$

The calculation uses $r>r_+=\mu+\sqrt{\mu^2-a^2}$ and $0<\theta<\pi$. The axes and horizon require regular charts; these coordinate restrictions are not extra physical boundaries. Let $K^{\alpha\beta}$ be the Carter tensor with the normalization in Killing Tensors and Carter Separability. For smooth scalar fields and a constant scalar mass $m_s$, define

$$\mathcal C\Phi=\nabla_\alpha(K^{\alpha\beta}\nabla_\beta\Phi), \qquad \mathcal L=\Box-m_s^2.$$

Theorem. The four operators $\mathcal L$, $i\partial_t$, $-i\partial_\phi$, and $\mathcal C$ commute pairwise as differential operators. This is a local identity on smooth scalars. It does not yet specify an operator domain with boundary conditions, a Hilbert-space realization, or a complete spectral expansion.

A direct proof on arbitrary fields

The radial and angular pieces are

$$\begin{aligned} \mathscr R&=\partial_r(\Delta\partial_r) -\frac{\big[(r^2+a^2)\partial_t+a\partial_\phi\big]^2}{\Delta},\\ \mathscr T&=\frac1{\sin\theta}\partial_\theta(\sin\theta\,\partial_\theta) +\frac{\big[\partial_\phi+a\sin^2\theta\,\partial_t\big]^2}{\sin^2\theta}. \end{aligned}$$

Their coefficients separate in $r$ and $\theta$, and the shared derivatives $\partial_t,\partial_\phi$ commute. Thus $[\mathscr R,\mathscr T]=0$. The metric density $\sqrt{-g}=\Sigma\sin\theta$ and the divergence ordering give

$$\Box=\Sigma^{-1}(\mathscr R+\mathscr T), \qquad \mathcal C=\mathscr T-B\Box.$$

For clarity, the tensor underlying the second formula is

$$K^{\alpha\beta} =(\partial_\theta)^\alpha(\partial_\theta)^\beta +\frac{v^\alpha v^\beta}{\sin^2\theta} -Bg^{\alpha\beta}, \qquad v=\partial_\phi+a\sin^2\theta\,\partial_t.$$

Here $v$ is a convenient vector in the formula, not an additional Killing vector. In the divergence, differentiating $\Sigma$ in the angular term produces $(\partial_\theta B/\Sigma)\partial_\theta$; differentiating $-Bg^{\alpha\beta}$ produces its negative. Keeping these terms is essential to obtain $\mathscr T-B\Box$.

Since $\mathscr T$ commutes with multiplication by $r^2$,

$$\begin{aligned} [\Box,\mathscr T]&=\Sigma^{-1}[\mathscr T,B]\Box,\\ [\Box,B]&=\Sigma^{-1}[\mathscr T,B]. \end{aligned}$$

For example, the first line follows from $[\Sigma^{-1},\mathscr T] =\Sigma^{-1}[\mathscr T,B]\Sigma^{-1}$. Consequently

$$[\Box,\mathcal C] =[\Box,\mathscr T]-[\Box,B]\Box=0.$$

The coefficients are independent of $t$ and $\phi$, and a constant mass commutes with everything. This proves the complete scalar theorem without restricting to a separated test field.

The geometric reason is that the Killing-tensor equation cancels the highest-derivative terms; a remaining curvature obstruction must also vanish. Ricci flatness supplies the sufficient condition here. A Killing tensor on an arbitrary curved background does not alone prove scalar commutation (Carter 1977, § VI, pp. 3409–3410). The principal-tensor scalar tower is a broader example with the required lower-order cancellations (Frolov, Krtouš, and Kubizňák 2017, § 6.3, pp. 119–120, Eqs. (6.41)–(6.49)).

Common eigenvalues and the separation constant

Take simultaneous eigenfunctions satisfying

$$\begin{gathered} i\partial_t\Phi=\omega\Phi,\qquad -i\partial_\phi\Phi=m\Phi,\\ \mathcal L\Phi=0,\qquad \mathcal C\Phi=\kappa\Phi. \end{gathered}$$

Single-valued scalar fields have integer $m$. The first two equations fix the time and azimuthal dependence. The other two become

$$(\mathscr T-m_s^2B)\Phi=\kappa\Phi, \qquad (\mathscr R-m_s^2r^2)\Phi=-\kappa\Phi.$$

Each now involves only one remaining coordinate. Product solutions $\Phi=e^{-i\omega t}e^{im\phi}R(r)S(\theta)$ therefore follow from ordinary differential equations, rather than from commutation as an abstract slogan. The angular equation is

$$\frac1{\sin\theta}(\sin\theta\,S')' +\left[a^2(\omega^2-m_s^2)\cos^2\theta -\frac{m^2}{\sin^2\theta}+A\right]S=0.$$

The precise dictionary is

$$\kappa=-\lambda, \qquad \lambda=A+a^2\omega^2-2am\omega.$$

Indeed, expanding $\mathscr T$ on the Fourier factor gives $-a^2\omega^2\sin^2\theta+2am\omega$ in addition to the angular derivatives. This proves the displayed shift and sign. The positive-divergence operator $\mathcal C$ has eigenvalue $-\lambda$; $-\mathcal C$ has eigenvalue $+\lambda$. Both are dimensionless. The angular parameter squared is $a^2(\omega^2-m_s^2)$, whereas the radial shift contains $\omega^2$, not $\omega^2-m_s^2$.

At $a=0$, $\mathcal C=\Delta_{S^2}$, the unit-sphere Laplacian, so $\kappa=-\ell(\ell+1)$ and $\lambda=\ell(\ell+1)$. This is a useful sign check. Regular angular boundary conditions select $A$; radial ingoing and outgoing conditions impose a further problem and are not consequences of the local operator theorem. The massless dictionary connects to Spin-Weighted Spheroidal Harmonics.

Boundary conditions can break the symmetry

There are three different questions: does an operator preserve the differential equation, does it preserve the admissible solutions, and do its eigenfunctions give a complete expansion? Only the first has been proved. Even formal symmetry under integration by parts is not self-adjointness: boundary terms, domains, and the chosen inner product still matter.

For a boundary trace condition $\mathcal B\Phi=0$, one must additionally check $\mathcal B\mathcal C\Phi=0$ on the solutions of interest. An exact counterexample exists already in Schwarzschild, where $a=0$ and $\mathcal C=\Delta_{S^2}$. Choose a length $L>\mu$ and set

$$\Phi=\frac{r-\mu}{L}\cos\theta-1.$$

This static field solves the massless wave equation because

$$\partial_r\!\left[\Delta\,\partial_r(r-\mu)\right]=2(r-\mu), \qquad \Delta_{S^2}\cos\theta=-2\cos\theta.$$

Impose a reflecting Dirichlet wall $(r-\mu)\cos\theta=L$. It lies entirely in $r>2\mu$ and is timelike: the gradient of its defining function is nonzero and spacelike in this static exterior. On that wall,

$$\Phi\big|_{\rm wall}=0, \qquad \mathcal C\Phi\big|_{\rm wall}=-2.$$

Thus $\mathcal C\Phi$ still solves the bulk equation but violates the same Dirichlet condition. The example is a smooth local solution, not a claim about finite total energy on an unbounded region. As $\mu\to0$, the wall becomes the plane $z=L$, which also makes the geometry transparent.

Failure of this operator does not prove that no other variables or operators could separate the boundary problem. In the flat limit, Cartesian coordinates do separate the plane problem. Likewise, a constant-$r$ wall is compatible with the scalar Carter operator on solutions: its tangential trace vanishes together with the Dirichlet field. The wall must be checked, not merely called reflecting.

For black-hole radiation problems, local joint eigenfunctions likewise do not automatically span finite-energy initial data. A quasinormal expansion requires the retarded operator realization and its analytic continuation, including any continuous contribution or branch cut; see Retarded Green Functions on Black-Hole Exteriors.

Potentials: an obstruction and its exact correction

Keep the Kerr metric fixed but replace the scalar equation by

$$\mathcal L_V=\Box-V(x).$$

The product rule gives, for an arbitrary smooth field,

$$[\mathcal L_V,\mathcal C]\Phi =\mathcal C(V)\Phi +2K^{\alpha\beta}(\nabla_\alpha V)(\nabla_\beta\Phi).$$

Thus the unmodified Carter operator generally fails even though the geodesic Killing tensor survives. Failure of this particular operator, however, does not rule out a corrected one.

When a multiplication term restores commutation

Seek $\mathcal C_V=\mathcal C+U$, where $U$ multiplies the field and does not change the second-order part. Equating the first-derivative coefficient in $[\mathcal L_V,\mathcal C+U]$ gives the necessary condition

$$\nabla^\alpha U=-K^{\alpha\beta}\nabla_\beta V.$$

It is also sufficient: taking the divergence gives $\Box U=-\mathcal C(V)$, canceling the remaining zeroth-order coefficient. This is a local integrability problem for $U$, not a condition on geodesics.

Assume now that $V$ and $U$ are stationary and axisymmetric. In the regular Kerr exterior the radial and polar equations reduce to

$$\partial_rU=B\,\partial_rV,\qquad \partial_\theta U=-r^2\partial_\theta V.$$

Equality of their mixed derivatives is precisely

$$\partial_r\partial_\theta(\Sigma V)=0.$$

On a connected coordinate rectangle this holds if and only if

$$\Sigma V=F(r)+G(\theta).$$

A compatible correction and the resulting operator are

$$U=BV-G,\qquad \mathcal C_V=\mathcal C+BV-G =\mathscr T-G-B\mathcal L_V.$$

For a direct second check, set $\mathscr R_V=\mathscr R-F$ and $\mathscr T_V=\mathscr T-G$. They commute, while $\mathcal L_V=\Sigma^{-1}(\mathscr R_V+\mathscr T_V)$. The preceding scalar proof then gives $[\mathcal L_V,\mathcal C_V]=0$ unchanged. The split has the harmless freedom $F\mapsto F+c$, $G\mapsto G-c$; it shifts $\mathcal C_V$ by the constant $c$ and relabels its eigenvalue.

Here $V$ has dimensions of inverse length squared, and $F,G,U$ are dimensionless. A constant mass gives $F=m_s^2r^2$, $G=m_s^2B$, hence $U=0$ as required. A nonconstant example is $V=F(r)/\Sigma$: it separates, but normally requires $U=BV$. By contrast, for nonzero dimensionless $\eta$,

$$V=\frac{\eta r\cos\theta}{\mu\Sigma}, \qquad \partial_r\partial_\theta(\Sigma V) =-\frac{\eta}{\mu}\sin\theta\ne0.$$

No stationary, axisymmetric multiplication correction of the specified Carter operator exists on this rectangle. This is not a no-go theorem for every differential order, field redefinition, or coordinate system. Mode Coupling and Broken Separability explains what a named deformation does to an angular mode expansion.

Spinor and higher-spin operators

The scalar proof cannot be applied component by component to spinors or tensor fields: their covariant derivatives act on different bundles and have different curvature commutators. A useful explicit spinor lift uses the smooth principal closed conformal Killing–Yano two-form $h$ on Kerr. In abstract indices its normalization obeys

$$\nabla_c h_{ab}=g_{ca}\xi_b-g_{cb}\xi_a,\qquad \xi_a=\tfrac13\nabla^b h_{ba}.$$

On the spin bundle, let $\nabla_a$ now be the spin connection, $\{\gamma^a,\gamma^b\}=2g^{ab}I$, and $\gamma^{abc}=\gamma^{[a}\gamma^b\gamma^{c]}$, with unit-weight antisymmetrization. For $D=\gamma^a\nabla_a$, the named lift is

$$Q_h=\gamma^{abc}h_{bc}\nabla_a +\frac23\gamma^a\nabla^b h_{ba}.$$

The local theorem is $[D,Q_h]=0$, and therefore also $[D+m_s,Q_h]=0$ for constant $m_s$ in this Dirac convention. The derivative-of-$h$ term is indispensable. The Clifford algebra cancels the second-derivative coefficient; the conformal Killing–Yano equation and its curvature integrability identity cancel the remaining terms. This spinor curvature calculation is not the scalar proof above. The explicit operator and its commuting spinorial Killing-vector partners are given in Frolov, Krtouš, and Kubizňák 2017, § 3.5, p. 59, Eqs. (3.86)–(3.87). A component-level derivation of the spinor separation equations lies beyond this page; the displayed operator identifies exactly which theorem is used.

For Maxwell and linearized gravity on four-dimensional vacuum Kerr, use the extreme spin components in the declared principal null tetrad and the field-specific Teukolsky equation. Its weighted operators are not scalar Carter operators acting on arbitrary components. Turning separated curvature fields into vector or metric perturbations additionally invokes constraints, gauge, and reconstruction and completion. These links specify the fields and tetrad conventions; no general higher-dimensional or arbitrarily coupled higher-spin theorem is asserted.

Exercises

A wall that does preserve the scalar symmetry

For a smooth massive scalar solution, impose Dirichlet data on $r=r_0>r_+$. Prove that $\mathcal C\Phi$ also satisfies those data. Does this prove completeness of separated modes?

Solution

On solutions, $\mathcal C\Phi=(\mathscr T-m_s^2B)\Phi$. The operator on the right differentiates only tangentially to the wall. A smooth trace that vanishes identically has vanishing tangential derivatives, so $(\mathcal C\Phi)|_{r_0}=0$. This proves preservation of this boundary condition on smooth solutions. It supplies neither a self-adjoint realization nor a complete mode basis; horizon, asymptotic, and regularity conditions still need their own analysis.

A radial potential need not be separable

Let $V=W(r)$, with no polar dependence. Determine when the specified multiplication correction exists for rotating Kerr. Compare $a=0$.

Solution

The mixed derivative is

$$\partial_r\partial_\theta(\Sigma W) =-2a^2\sin\theta\cos\theta\,W'(r).$$

For $a\ne0$, vanishing on an open coordinate rectangle requires $W'=0$: a generic radial $W$ does not satisfy the condition. At $a=0$, $\Sigma=r^2$, so any smooth radial $W$ has $F=r^2W$, $G=0$, and $U=0$. The ordinary angular Laplacian commutes with a radial potential on a spherically symmetric background. The equatorial line alone is not an open rectangle and cannot establish the rotating result.

Recover the sign from a short-wavelength field

Let $\Phi=b(x)e^{iW(x)/\varepsilon}$, where $b$ varies slowly and $p_\alpha=\nabla_\alpha W$. Find the leading term of $\mathcal C\Phi$ as $\varepsilon\to0$.

Solution

Two derivatives acting on the phase give

$$\mathcal C\Phi =-\varepsilon^{-2}K^{\alpha\beta}p_\alpha p_\beta\,\Phi +O(\varepsilon^{-1})e^{iW/\varepsilon}.$$

Thus the semiclassical principal symbol of $-\varepsilon^2\mathcal C$, under $p_\alpha\leftrightarrow-i\varepsilon\nabla_\alpha$, is the quadratic geodesic integral with no additional minus sign. This explains the minus sign in $\kappa=-\lambda$ at short wavelengths. It does not remove lower-order curvature obstructions on a general metric.

References

• Carter, Brandon. “Killing Tensor Quantum Numbers and Conserved Currents in Curved Space.” Physical Review D 16 (1977): 3395–3414. DOI.
• Frolov, Valeri P., Pavel Krtouš, and David Kubizňák. “Black Holes, Hidden Symmetries, and Complete Integrability.” Living Reviews in Relativity 20 (2017): 6. DOI. Open PDF, arXiv version 3.