Which pseudo-Riemannian H-type nilmanifolds are geodesic orbit manifolds?

Geodesic orbit means every geodesic is an isometry orbit

A pseudo-Riemannian manifold is called geodesic orbit (GO) when every geodesic is the orbit of a one-parameter subgroup of its full isometry group. Such a manifold is necessarily homogeneous, so the question becomes algebraic: for each initial vector, does the isometry algebra contain a skew-symmetric derivation that reproduces the geodesic?

The pseudo-Riemannian setting changes the problem in a way that has no Riemannian counterpart. A geodesic may be null, and for null geodesics the classical geodesic lemma must be modified: the generator of the one-parameter group only has to reproduce the geodesic up to an affine reparametrization. That extra freedom is encoded in a constant k, which vanishes unless the initial vector is null. Because of this, many Riemannian results do not carry over directly.

The paper studies this question for pseudo H-type groups N_{r,s}: 2-step nilpotent Lie groups whose Lie algebra n_{r,s} = z ⊕ v is built from an admissible module v of the Clifford algebra Cl(R^{r,s}), with a left-invariant metric whose restriction to the centre z is indefinite, meaning s ≥ 1. Here z is identified with R^{r,s}, and v carries a neutral scalar product.

The Riemannian benchmark allows larger centres and modules

For Riemannian H-type groups, the GO problem was solved completely by Kaplan and Riehm. With z an m-dimensional centre and v an n-dimensional orthogonal complement, N is geodesic orbit if and only if one of three conditions holds: m = 1, 2, 3 with n arbitrary; m = 5, 6 with n = 8; or m = 7 with n = 8, 16, 24 and v an isotypic Clifford module. In that last case the condition is equivalent to the linear transformation X ↦ J_{Z1}J_{Z2}···J_{Z7}(X) of v being either Id or −Id for an orthonormal basis Z_1, …, Z_7 of z.

The same Riemannian result also states that N is naturally reductive if and only if m = 1 or m = 3. So in the Riemannian case, a seven-dimensional centre still admits GO metrics on isotypic modules of dimensions 8, 16 and 24.

This is the comparison the paper uses to frame its own result. The pseudo-Riemannian case turns out to be far more rigid: the split analogue of that seven-dimensional-centre situation is the 15-dimensional group N_{3,4}(v_min), and it loses the GO property as soon as v is enlarged, whether by an equivalent or by a non-equivalent minimal module.

Only two signatures are naturally reductive

The first main result identifies the signatures in which N_{r,s} is naturally reductive: this happens if and only if (r, s) is (0,1) or (1,2). In both of those cases the manifold is also geodesic orbit, and this holds for every admissible Clifford module, not only for minimal ones.

The two cases are checked explicitly. For N_{0,1}, the Lie algebra is isometric to R^{0,1} ⊕ R^{l,l}, and the Clifford module splits as a direct sum of l minimal admissible modules v_min ≅ R^{1,1}. In a suitable orthonormal basis the operator J_Z spans a one-dimensional subspace V = J(z) ⊂ so(l,l) that is an abelian subalgebra, so the bracket condition [J(Z_1), J(Z_2)] = J(τ_{Z_1}(Z_2)) is satisfied with τ ≡ 0.

For N_{1,2}, the Lie algebra is isometric to R^{1,2} ⊕ R^{2l,2l} and the group has dimension 4l + 3 with l ≥ 1. The Clifford algebra Cl(R^{1,2}) has two irreducible non-equivalent modules V_+ and V_− of dimension 2, distinguished by whether J_{Z1}J_{Z2}J_{Z3} acts as +Id or −Id. A minimal admissible module is v_min = V_+ × V_− with a neutral metric, isometric to R^{2,2}, and any admissible module is an orthogonal direct sum of l such minimal modules.

The natural reductivity in these two signatures is established through Ovando's criterion together with the structure of the Lie algebra generated by J(z). The criterion requires that V = J(z) be a Lie subalgebra of so(v) and that [J(Z_1), J(Z_2)] = J(τ_{Z_1}(Z_2)) for some τ_{Z_1} ∈ so(z).

N_{3,4} is geodesic orbit only for the minimal module

The exceptional pseudo-Riemannian case is N_{3,4}. The paper proves that N_{3,4} is geodesic orbit but not naturally reductive exactly when v is a minimal dimensional admissible Clifford module. This is the 15-dimensional group N_{3,4}(v_min), and it is the split analogue of the Riemannian seven-dimensional-centre case.

For this group the authors explicitly construct the one-parameter isometry groups that generate all geodesics, including the null ones, and they describe the affine family of their generators. The construction uses a distinguished basis of v_min adapted to the Clifford relations, and the analysis yields all generators of homogeneous geodesics on v_min.

The contrast with the Riemannian situation is sharp. In the Riemannian case a seven-dimensional centre admits GO metrics on isotypic modules of dimensions 8, 16 and 24. In the pseudo-Riemannian case the exceptional signature is GO only for the smallest possible module. Enlarging v, even by an equivalent minimal module, destroys the property.

Enlarging the module destroys the geodesic orbit property

The paper proves that N_{3,4} with v not a minimal dimensional admissible Clifford module is not geodesic orbit. This covers both v_min ⊕ v_min and the non-equivalent combination v_min^− ⊕ v_min^+.

For v_min ⊕ v_min, the admissible scalar product on v_min is unique only up to a non-zero factor, so the sum may carry ⟨·,·⟩ ⊕ σ⟨·,·⟩ with σ = ±1. The contradiction derived in the proof is the same for both values of σ, so the non-GO result holds for both metrics.

For v_min^− ⊕ v_min^+, the module is isometric to R^{8,8}. With Ω = Z_1···Z_7 the volume form of Cl(R^{3,4}), one has Ω² = 1 and J_Ω acts as +Id on v_min^+ and as −Id on v_min^−. Taking the initial vector (z_0, x_0) = (Z_1, v^− ⊕ v^+), the proof shows there is no D ∈ h satisfying the required equation: the system reduces to α_1 + ᾱ_1 = 0 and simultaneously α_1 + ᾱ_1 = 2, a contradiction.

The general statement is Theorem 10: the pseudo-Riemannian H-type nilmanifold N_{3,4}(⊕^p v_min^+ ⊕ ⊕^q v_min^−) with p + q > 1 is not a geodesic orbit manifold. This follows from the totally geodesic subgroup argument together with the two explicit non-GO results.

Totally geodesic subgroups carry non-GO results upward

The proof strategy explains why the rigidity is so strong. An admissible submodule v′ ⊂ v gives a totally geodesic subgroup N_{r,s}(v′) ⊂ N_{r,s}(v). Since complete totally geodesic submanifolds of GO manifolds are GO, every non-GO result for a module passes to all modules containing it.

This is the mechanism that turns the explicit failures at N_{3,4}(v_min ⊕ v_min) and N_{3,4}(v_min^− ⊕ v_min^+) into a statement about arbitrary larger modules. It also explains why the GO property at (3,4) cannot survive any enlargement of v.

The proof combines three ingredients: the totally geodesic subgroup argument for propagation; Ovando's criterion together with the structure of the Lie algebra generated by J(z) for the naturally reductive cases; and an explicit analysis based on a distinguished basis of v_min adapted to the Clifford relations for N_{3,4}. On v_min this analysis yields all generators of homogeneous geodesics, including the null ones, while on v_min ⊕ v_min and on v_min^− ⊕ v_min^+ it yields explicit initial vectors whose geodesics are not homogeneous.

The signature split is the paper's decisive threshold

The complete answer for arbitrary admissible modules is a four-part classification. N_{r,s} is naturally reductive, hence geodesic orbit, if and only if (r, s) ∈ {(0,1), (1,2)}. N_{3,4} with v a minimal dimensional admissible Clifford module is geodesic orbit but not naturally reductive. N_{3,4} with v not minimal dimensional is not geodesic orbit. And N_{r,s} with (r, s) ∉ {(0,1), (1,2), (3,4)} is not geodesic orbit for any admissible Clifford module.

The practical threshold is therefore the pair (r, s) together with the size of the module. If the signature is (0,1) or (1,2), the GO property is unconditional across admissible modules. If the signature is (3,4), the GO property depends entirely on whether v is minimal: the smallest possible module is the only one that works, and the property is lost as soon as v is enlarged by an equivalent or a non-equivalent minimal module. Every other signature fails outright.

The most decision-relevant point is that the pseudo-Riemannian case is not a mild deformation of the Riemannian one. Where the Riemannian seven-dimensional centre admits GO metrics on isotypic modules of dimensions 8, 16 and 24, the split analogue admits GO only for the smallest possible module. Readers comparing the two settings should treat the module dimension, not just the signature, as the variable that decides the answer.

Sources used in this answer