Generator Completeness and Horizon Components

Horizon generators can be complete as symmetry orbits, complete as affinely parametrized geodesics, or merely inextendible while remaining on the horizon. These are different claims. For a nondegenerate Killing horizon, a complete Killing orbit covers only a half-infinite affine interval on one branch; for a general event horizon, absence of future endpoints does not by itself imply infinite affine length. Global causal and extension hypotheses determine which stronger statement is available.

Required background. Killing Fields and Isometry Groups supplies complete flows and their orbits.

Helpful background. Exact Solutions of Einstein’s Equations supplies Kruskal and collapse examples; Causality and Global Structure supplies achronal boundaries and inextendible curves; The Einstein Equation enters focusing and area theorems; Black-Hole Geometry and Mechanics distinguishes event and Killing horizons.

Three completeness questions

Let a smooth null generator be written as $x^a(v)$ with tangent $\chi^a=dx^a/dv$. Ask separately:

1. Does $v$ range over all real numbers, so that $\chi^a$ generates a global one-parameter group on the relevant spacetime?
2. Does an affine parameter $\lambda$ range without a finite endpoint?
3. Does the generator have an endpoint as a curve contained in the horizon?

A local Killing field need not be a complete vector field. A null geodesic can be inextendible in an incomplete spacetime while having finite affine length. And a generator can have a past endpoint at a smoothness failure of the horizon even though the spacetime is completely regular there. None of these questions is answered by a Gaussian null chart alone.

For a future horizon understood as a closed achronal, future-null-geodesically ruled topological hypersurface, its generators are future inextendible within the horizon and may have past endpoints, but no future endpoints. This precise definition and its low-regularity setting are given in Chruściel, Delay, Galloway, and Howard 2001, arXiv v2, §2, pp. 6–8. It is a statement about the generator’s membership in the boundary, not yet about affine completeness.

Killing time versus affine time

Suppose $\mathcal H$ is a Killing horizon, its generator obeys

$$\chi^b\nabla_b\chi^a=\kappa\chi^a,$$

and $\kappa$ is constant along the chosen generator. Let $k^a=dx^a/d\lambda$ be an affine tangent and write $\chi^a=q k^a$, where $q=d\lambda/dv$. Substitution gives

$$\frac{d}{dv}\ln q=\kappa.$$

For $\kappa\ne0$,

$$q=C e^{\kappa v}, \qquad \lambda-\lambda_0=\frac{C}{\kappa}e^{\kappa v}.$$

Take a future outer horizon with $\kappa\gt0$ and a complete Killing orbit $-∞\lt v\lt∞$. As $v\to-∞$, the affine parameter approaches the finite value $\lambda_0$; as $v\to+∞$, it diverges. The isolated future branch is therefore past incomplete and future complete as an affine geodesic. In a bifurcate extension, $\lambda_0$ labels the bifurcation surface, where $\chi^a$ vanishes even though the affine geodesic can continue in the larger spacetime.

When $\kappa=0$, $q$ is constant and $\lambda=Cv+\lambda_0$. If the Killing orbit is complete, it is affine-complete in both directions. This elementary relation is a decisive check on the different global behavior of degenerate and nondegenerate horizons.

A theorem for stationary final-state horizons

Consider the class used by Rácz and Wald: assume a smooth, globally hyperbolic, stationary spacetime with one controlled asymptotic end, a black-hole region but no white-hole region, and a connected event horizon $\mathcal N$ that is a Killing horizon with compact cross-sections. Their global argument shows that the horizon Killing field has no zero in the original spacetime, that a smooth Cauchy surface meets $\mathcal N$ in a global cross-section, and that its horizon Killing orbits are diffeomorphic to $\mathbb R$. See Rácz and Wald 1996, §§2–3, pp. 545–547.

If, in addition, $\kappa$ is a nonzero constant, their Theorem 4.2 supplies a global bifurcate extension Rácz and Wald 1996, §4, pp. 548–550. The proof architecture is important: global hyperbolicity and the no-white-hole condition control the orbit space; compactness provides a global section; constant nonzero $\kappa$ supplies regular boost-like coordinates; a patching argument then produces the extension. Dropping those hypotheses can leave a local extension while obstructing a global one.

The conclusion is componentwise if disconnected horizon components are allowed, but one must check compact sections, normalization, and causal separation for each component. “The horizon is complete” is not a substitute for those statements.

Event-horizon endpoints and crease sets

An event horizon $\partial J^-(\mathscr I^+)$ is an achronal boundary. Where it is differentiable, it has one null generator direction. Generators may enter the horizon at past endpoints, often forming a crease set at which multiple generators meet and the horizon fails to be differentiable. They do not split and leave through a future endpoint: that would violate the future-horizon property.

This one-way behavior permits topology change in spatial cuts. In a binary merger, two components on an early cut can become one on a later cut because new generators join at past endpoints. Reading the process backward makes the single horizon appear to branch. The null hypersurface itself need not be smooth at the joining set, so a theorem restricted to a smooth Killing horizon does not describe that merger region.

No-future-endpoint still does not imply future affine completeness. Chruściel and collaborators explicitly treat future completeness as one sufficient hypothesis in their area theorem; alternatives use global hyperbolicity plus regular conformal completion. Their exact alternatives are Chruściel, Delay, Galloway, and Howard 2001, Theorem 1.1, pp. 109–112. Under the completeness route, negative expansion plus null convergence produces a focal point at finite affine parameter, while achronality forces the generator to leave the horizon, a contradiction; the proof is Chruściel, Delay, Galloway, and Howard 2001, §4.2, Proposition 4.17, pp. 135–137.

Thus energy conditions constrain focusing only after global hypotheses keep the generator available long enough. They do not manufacture completeness. Singular boundaries, excised regions, or a truncated coordinate patch are simple limitations.

Distinguish failure modes

Suppose a connected Killing horizon has constant $\kappa\gt0$, and every Killing orbit has parameter range $v\in\mathbb R$. A student concludes that each generator is affine-complete to both past and future. Locate the error and state what further structure resolves it.

Solution

The Killing parameter is not affine. The relation $\lambda-\lambda_0=(C/\kappa)e^{\kappa v}$ shows that the entire range $v\in\mathbb R$ covers only $\lambda\in(\lambda_0,∞)$ for a future branch. The generator is future affine-complete but reaches a finite affine value to the past. A regular bifurcate extension can continue the geodesic through the bifurcation surface, but the existence of that global extension requires the causal, compactness, and stationarity hypotheses of an extension theorem; it does not follow from constant nonzero $\kappa$ alone.

Nondegenerate and Degenerate Horizon Geometry next synthesizes this affine-parameter distinction with the different near-horizon expansions and bifurcation behavior.

References

• Chruściel, Piotr T., Erwann Delay, Gregory J. Galloway, and Ralph Howard. “Regularity of Horizons and the Area Theorem.” Annales Henri Poincaré 2 (2001): 109–178. DOI. Open PDF.
• Rácz, István, and Robert M. Wald. “Global Extensions of Spacetimes Describing Asymptotic Final States of Black Holes.” Classical and Quantum Gravity 13 (1996): 539–553. DOI. Open PDF.