Quadratic Killing Tensor Fields

• Quadratic Killing tensor fields are symmetric rank-2 tensors that satisfy the Killing equation, yielding conserved quadratic integrals in geodesic flows.
• They play a crucial role in Riemannian geometry, gauge-covariant Hamiltonian systems, and black-hole spacetimes by revealing hidden symmetries and integrability.
• Advanced methods like prolongation, decomposability analysis, and Finsler generalizations provide robust frameworks for classifying and constructing these tensors.

Searching arXiv for recent and foundational papers on quadratic Killing tensor fields. Using the arXiv search tool to gather relevant sources. A quadratic Killing tensor field is a symmetric rank-$2$ tensor field $K_{ab}=K_{(ab)}$ satisfying the Killing equation

$\nabla_{(a}K_{bc)}=0,$

equivalently $K^{(ij;k)}=0$ in contravariant notation. Its defining geometric significance is that it generates a conserved quantity quadratic in the geodesic momenta or velocities, such as

$$F=K^{\mu\nu}p_\mu p_\nu \quad\text{or}\quad I=K_{ab}\dot x^a\dot x^b,$$

and therefore encodes a hidden symmetry of the geodesic flow (Cariglia et al., 2014). In contemporary work, the quadratic case appears simultaneously as a classical object in Riemannian and pseudo-Riemannian geometry, as the leading term of completed constants of motion in gauge-covariant Hamiltonian systems, as a representation-theoretic object on symmetric spaces, and as a diagnostic for integrability or non-integrability in black-hole spacetimes and other Hamiltonian models (Igata et al., 2010, Matveev et al., 30 Apr 2026).

1. Definition and basic geometric role

For a symmetric covariant tensor field of degree $p$, the Killing equation is

$L_{(i_1\dots i_p,j)}=0.$

In the quadratic case $p=2$, this specializes to

$$L_{(ij,k)}=0, \qquad\text{equivalently}\qquad \nabla_{(k}L_{ij)}=0,$$

or, in contravariant form,

$\nabla^{(\lambda}K^{\mu\nu)}=0.$

These are the standard definitions of a quadratic Killing tensor field (Igata et al., 2010, Matveev et al., 30 Apr 2026).

The corresponding conserved quantity is quadratic in momenta or velocities. For a free relativistic particle one has

$K_{ab}=K_{(ab)}$0

and for geodesic motion this is often written as

$K_{ab}=K_{(ab)}$1

On a Riemannian manifold, the same statement appears as constancy of

$K_{ab}=K_{(ab)}$2

along geodesics (Igata et al., 2010, Heil et al., 2016). This is the standard reason quadratic Killing tensors are treated as hidden symmetries: rank-$K_{ab}=K_{(ab)}$3 Killing tensors are Killing vectors and generate linear integrals, whereas rank-$K_{ab}=K_{(ab)}$4 tensors generate genuinely quadratic first integrals and need not correspond to simple transformations of configuration space (Cariglia et al., 2014).

The metric itself is always a quadratic Killing tensor, since $K_{ab}=K_{(ab)}$5. Much of the theory therefore distinguishes between quadratic Killing tensors that are built from the metric and Killing vectors, and those that are not. On symmetric spaces this distinction is formalized by calling a Killing tensor decomposable if it is a polynomial in Killing vector fields under the symmetric product, and indecomposable otherwise (Matveev et al., 30 Apr 2026).

2. Quadratic first integrals in Hamiltonian systems

In canonical geometry, the quadratic Killing tensor is the highest-order coefficient in a polynomial constant of motion. For a Hamiltonian system on a manifold with metric $K_{ab}=K_{(ab)}$6, equipped with gauge-covariant momenta

$K_{ab}=K_{(ab)}$7

the Hamiltonian is written

$K_{ab}=K_{(ab)}$8

The gauge-covariant Poisson brackets are

$K_{ab}=K_{(ab)}$9

with

$\nabla_{(a}K_{bc)}=0,$0

In this framework, the natural quadratic ansatz is

$\nabla_{(a}K_{bc)}=0,$1

Requiring $\nabla_{(a}K_{bc)}=0,$2 yields the chain

$\nabla_{(a}K_{bc)}=0,$3

$\nabla_{(a}K_{bc)}=0,$4

$\nabla_{(a}K_{bc)}=0,$5

together with

$\nabla_{(a}K_{bc)}=0,$6

Thus the quadratic Killing tensor controls the highest-order part of the conserved quantity, while lower-order momentum terms are generally required in the presence of gauge and scalar couplings (Cariglia et al., 2014).

A closely related generalization appears for relativistic particles in external fields. There one starts from the constrained Hamiltonian

$\nabla_{(a}K_{bc)}=0,$7

and weakens exact conservation to

$\nabla_{(a}K_{bc)}=0,$8

equivalently

$\nabla_{(a}K_{bc)}=0,$9

For a quadratic ansatz

$K^{(ij;k)}=0$0

the resulting hierarchy has as its top equation

$K^{(ij;k)}=0$1

If the $K^{(ij;k)}=0$2-terms vanish, this reduces to the ordinary rank-$K^{(ij;k)}=0$3 Killing tensor equation; if not, the highest member is conformal Killing rather than strictly Killing (Igata et al., 2010). This places the ordinary quadratic Killing tensor field inside a larger constrained-Hamiltonian hierarchy.

In second-order dynamical systems written as

$K^{(ij;k)}=0$4

with quadratic ansatz

$K^{(ij;k)}=0$5

the determining equations again begin with

$K^{(ij;k)}=0$6

The quadratic part of every quadratic first integral is therefore governed by a rank-$K^{(ij;k)}=0$7 Killing tensor of the kinetic metric, while the lower-order terms are fixed by compatibility equations involving the generalized forces (Karpathopoulos et al., 2018).

3. Prolongation, integrability, and constructive methods

The Killing equation is an overdetermined linear PDE, and one modern approach is prolongation. For rank $K^{(ij;k)}=0$8, the prolonged variables are $K^{(ij;k)}=0$9, $$F=K^{\mu\nu}p_\mu p_\nu \quad\text{or}\quad I=K_{ab}\dot x^a\dot x^b,$$0, and $$F=K^{\mu\nu}p_\mu p_\nu \quad\text{or}\quad I=K_{ab}\dot x^a\dot x^b,$$1, and the prolonged system is

$$F=K^{\mu\nu}p_\mu p_\nu \quad\text{or}\quad I=K_{ab}\dot x^a\dot x^b,$$2

$$F=K^{\mu\nu}p_\mu p_\nu \quad\text{or}\quad I=K_{ab}\dot x^a\dot x^b,$$3

followed by a closed equation for $$F=K^{\mu\nu}p_\mu p_\nu \quad\text{or}\quad I=K_{ab}\dot x^a\dot x^b,$$4 involving $$F=K^{\mu\nu}p_\mu p_\nu \quad\text{or}\quad I=K_{ab}\dot x^a\dot x^b,$$5, $$F=K^{\mu\nu}p_\mu p_\nu \quad\text{or}\quad I=K_{ab}\dot x^a\dot x^b,$$6, curvature, and $$F=K^{\mu\nu}p_\mu p_\nu \quad\text{or}\quad I=K_{ab}\dot x^a\dot x^b,$$7 (Houri et al., 2017). The first nontrivial integrability condition occurs only at the top stage $$F=K^{\mu\nu}p_\mu p_\nu \quad\text{or}\quad I=K_{ab}\dot x^a\dot x^b,$$8; for $$F=K^{\mu\nu}p_\mu p_\nu \quad\text{or}\quad I=K_{ab}\dot x^a\dot x^b,$$9 the lower ones vanish,

$p$0

while $p$1 is a nontrivial projected expression involving $p$2, $p$3, $p$4, $p$5, and $p$6 (Houri et al., 2017). This gives explicit algebraic restrictions on possible quadratic Killing tensors.

A different prolongation program, tailored specifically to Killing two-tensors, starts from

$p$7

and rewrites it as

$p$8

with $p$9 in the hook bundle. In the locally symmetric case $L_{(i_1\dots i_p,j)}=0.$0, a second prolongation yields a connection on

$L_{(i_1\dots i_p,j)}=0.$1

whose parallel sections are exactly Killing 2-tensors. The same paper shows that the canonical quadratic map

$L_{(i_1\dots i_p,j)}=0.$2

is compatible with prolongation, and that its kernel is controlled by Killing–Yano 3-forms via an exact sequence of bundles (Eastwood et al., 20 Apr 2026).

There is also a constructive geometric method that avoids solving $L_{(i_1\dots i_p,j)}=0.$3 directly. In a spacetime with a codimension-one foliation $L_{(i_1\dots i_p,j)}=0.$4, unit normal $L_{(i_1\dots i_p,j)}=0.$5, lapse $L_{(i_1\dots i_p,j)}=0.$6, and slice second fundamental form $L_{(i_1\dots i_p,j)}=0.$7, one starts from a trivial Killing tensor on each leaf $L_{(i_1\dots i_p,j)}=0.$8, built from the induced metric and slice-tangent Killing vectors $L_{(i_1\dots i_p,j)}=0.$9, and then lifts it to a spacetime Killing tensor. Under explicit compatibility and integrability conditions,

$p=2$0

one obtains

$p=2$1

which can be nontrivial even though the slice tensor is trivial (Kobialko et al., 2021). This construction is purely geometric and algebraic in the sense emphasized in that work.

4. Decomposability, rigidity, and symmetric-space classifications

Several classification results sharply separate decomposable from indecomposable quadratic Killing tensors. On the conformally flat $p=2$2-torus

$p=2$3

every quadratic Killing tensor is reducible: $p=2$4 Equivalently, there are no irreducible quadratic Killing tensors on this class of tori (Heil et al., 2016).

On $p=2$5 with the Fubini–Study metric, all Killing tensors of arbitrary rank are generated by Killing fields, and for rank $p=2$6 the space of quadratic Killing tensors has dimension

$p=2$7

Thus every quadratic Killing tensor on $p=2$8 is generated by Killing vector fields (Eastwood, 2023). This continues the classical positive picture known for spaces of constant curvature.

That optimistic picture fails on other symmetric spaces. Explicit indecomposable quadratic Killing tensors exist on quaternionic projective spaces $p=2$9 for $$L_{(ij,k)}=0, \qquad\text{equivalently}\qquad \nabla_{(k}L_{ij)}=0,$$0 and on the Cayley projective plane $$L_{(ij,k)}=0, \qquad\text{equivalently}\qquad \nabla_{(k}L_{ij)}=0,$$1. In $$L_{(ij,k)}=0, \qquad\text{equivalently}\qquad \nabla_{(k}L_{ij)}=0,$$2, $$L_{(ij,k)}=0, \qquad\text{equivalently}\qquad \nabla_{(k}L_{ij)}=0,$$3, the indecomposable subspace has dimension

$$L_{(ij,k)}=0, \qquad\text{equivalently}\qquad \nabla_{(k}L_{ij)}=0,$$4

while on $$L_{(ij,k)}=0, \qquad\text{equivalently}\qquad \nabla_{(k}L_{ij)}=0,$$5 there is a $$L_{(ij,k)}=0, \qquad\text{equivalently}\qquad \nabla_{(k}L_{ij)}=0,$$6-dimensional irreducible $$L_{(ij,k)}=0, \qquad\text{equivalently}\qquad \nabla_{(k}L_{ij)}=0,$$7-submodule all of whose nonzero elements are indecomposable (Matveev et al., 2023).

A 2026 structural advance reframes the symmetric-space problem in terms of top slot Killing tensors. On compact irreducible symmetric spaces, every quadratic Killing tensor is spanned by top-slot quadratic tensors, and in normal coordinates such a tensor has the form

$$L_{(ij,k)}=0, \qquad\text{equivalently}\qquad \nabla_{(k}L_{ij)}=0,$$8

where $$L_{(ij,k)}=0, \qquad\text{equivalently}\qquad \nabla_{(k}L_{ij)}=0,$$9 is a constant $\nabla^{(\lambda}K^{\mu\nu)}=0.$0-tensor satisfying explicit curvature identities: $\nabla^{(\lambda}K^{\mu\nu)}=0.$1

$\nabla^{(\lambda}K^{\mu\nu)}=0.$2

$\nabla^{(\lambda}K^{\mu\nu)}=0.$3

The same paper proves that quadratic Killing tensor fields on $\nabla^{(\lambda}K^{\mu\nu)}=0.$4 are decomposable, while for $\nabla^{(\lambda}K^{\mu\nu)}=0.$5, $\nabla^{(\lambda}K^{\mu\nu)}=0.$6, and $\nabla^{(\lambda}K^{\mu\nu)}=0.$7, the full quadratic space is spanned by decomposable tensors together with the previously constructed indecomposable families (Matveev et al., 30 Apr 2026). This completes the rank-one classification on Riemannian symmetric spaces of rank one.

5. Hidden symmetry, separability, and explicit physical examples

Quadratic Killing tensors are central in hidden symmetry and separability. In the gauge-covariant Hamiltonian framework, the reduced quantum-dot system with metric

$\nabla^{(\lambda}K^{\mu\nu)}=0.$8

admits explicit rank-$\nabla^{(\lambda}K^{\mu\nu)}=0.$9 Killing tensors generating quadratic constants of motion. One example yields a Runge–Lenz-type invariant

$K_{ab}=K_{(ab)}$00

under the condition

$K_{ab}=K_{(ab)}$01

Another rank-$K_{ab}=K_{(ab)}$02 tensor yields

$K_{ab}=K_{(ab)}$03

provided

$K_{ab}=K_{(ab)}$04

The same paper also gives an explicit rank-$K_{ab}=K_{(ab)}$05 tensor that fails to extend to a conserved quantity unless $K_{ab}=K_{(ab)}$06, showing that the existence of a quadratic Killing tensor is only the first step; lower-order compatibility conditions with $K_{ab}=K_{(ab)}$07 and $K_{ab}=K_{(ab)}$08 are decisive (Cariglia et al., 2014).

In four-dimensional gravity and supergravity, the Carter-type picture persists in more general matter-coupled settings. A recent integration of stationary axisymmetric Einstein–Maxwell–dilaton–axion theory starts from a metric class with separable conformal factor

$K_{ab}=K_{(ab)}$09

which guarantees an exact rank-$K_{ab}=K_{(ab)}$10 Killing tensor

$K_{ab}=K_{(ab)}$11

The resulting canonical metric,

$K_{ab}=K_{(ab)}$12

supports Carter-type hidden symmetry beyond Petrov type D, and the paper states that the general solutions are typically Petrov type $K_{ab}=K_{(ab)}$13 and may have no Killing–Yano structures (Gal'tsov et al., 9 Mar 2025). This suggests that quadratic Killing tensors are more robust than type-D algebraic speciality or Killing–Yano structures in these supergravity families.

The converse phenomenon is equally important. In slowly rotating black holes of dynamical Chern–Simons and scalar Gauss–Bonnet gravity, a nontrivial rank-$K_{ab}=K_{(ab)}$14 Killing tensor exists through $K_{ab}=K_{(ab)}$15, but cannot be extended to $K_{ab}=K_{(ab)}$16. In dynamical Chern–Simons gravity the failure persists through all searched ranks up to $K_{ab}=K_{(ab)}$17, and both families become Petrov type I at the same perturbative order (Owen et al., 2021). A plausible implication is that the exact rotating quadratic-gravity black holes do not possess a Carter-like fourth constant of motion.

6. Euclidean integrable systems, Finsler variants, and related generalizations

In Euclidean space, every valency-two Killing tensor has the general form

$K_{ab}=K_{(ab)}$18

where

$K_{ab}=K_{(ab)}$19

For quartic Hamiltonians related to symmetric spaces, the paper on second-order Killing tensors related to symmetric spaces shows that the quadratic integrals come from special combinations of rotations and translations, but that full Liouville integrability generally requires additional integrals of fourth, sixth, and higher order in the momenta (Porubov et al., 2023). In the special $K_{ab}=K_{(ab)}$20 Garnier case the associated quadratic Killing tensors fit the classical separable picture, while in the higher-dimensional examples the relevant Killing tensors have nonzero Haantjes torsion (Porubov et al., 2023).

A further reformulation appears in Finsler geometry. There one introduces a Killing non-linear 1-form $K_{ab}=K_{(ab)}$21 and the spray operator

$K_{ab}=K_{(ab)}$22

For the rank-$K_{ab}=K_{(ab)}$23 ansatz

$K_{ab}=K_{(ab)}$24

the condition

$K_{ab}=K_{(ab)}$25

implies

$K_{ab}=K_{(ab)}$26

For $K_{ab}=K_{(ab)}$27, this is the Finsler analogue of the quadratic Killing tensor equation, and the associated conserved quantity is

$K_{ab}=K_{(ab)}$28

The same framework presents the Carter constant on Kerr spacetime as a rank-$K_{ab}=K_{(ab)}$29 example and gives a Finslerized construction of Runge–Lenz-type hidden conserved quantities in Newtonian gravity (Ootsuka et al., 2016).

Across these settings, the recurring structural point is stable. A quadratic Killing tensor field is the geometric source of a quadratic first integral, but its actual dynamical role depends strongly on context: in geodesic problems it is already a complete hidden symmetry; in externally coupled Hamiltonian systems it is the highest-order term in a completed conserved quantity; on symmetric spaces it may be decomposable or genuinely new; and in concrete physical models its existence or obstruction sharply tracks separability, photon-surface geometry, or the loss of integrability (Cariglia et al., 2014, Kobialko et al., 2021, Matveev et al., 30 Apr 2026).