Focusing and Geodesic Incompleteness

A trapped surface is a compact spacelike two-surface from which both future null normal families initially converge. Penrose’s singularity theorem turns that local optical statement into future null-geodesic incompleteness, but only after adding an energy inequality and global causal hypotheses. The theorem is therefore a chain of distinct arguments: Raychaudhuri focusing, loss of achronality after a focal point, compactness of a future boundary, and a topological contradiction. Its conclusion is an incomplete null geodesic—not, by itself, curvature blowup or the existence of an event horizon.

Required background. Analysis for Mathematical Relativity supplies the differential inequalities and compactness arguments; timelike congruences and spatial projection supplies expansion, shear, and Raychaudhuri evolution; and the Einstein equation converts an energy condition into a Ricci-curvature inequality.

Helpful background. Causality and Global Structure explains achronal boundaries and Cauchy surfaces, while PDEs and Green Functions for Gravitation gives the broader evolution context.

Null focusing from a trapped surface

Let $S$ be a smooth, compact, spacelike codimension-two surface in a four-dimensional spacetime. Choose future null normals $k_+$ and $k_-$; their positive rescalings do not change the signs of their expansions. With affine parameter $\lambda$ along one hypersurface-orthogonal null congruence, the Raychaudhuri equation is

$$\frac{d\theta}{d\lambda} =-\frac12\theta^2 -\sigma_{ab}\sigma^{ab} -R_{\mu\nu}k^\mu k^\nu.$$

There is no twist term because the generators are normal to a surface. If the null convergence condition

$$R_{\mu\nu}k^\mu k^\nu\ge 0$$

holds, then $\sigma_{ab}\sigma^{ab}\ge0$ gives

$$\frac{d\theta}{d\lambda}\le-\frac12\theta^2.$$

For initial expansion $\theta_0<0$, divide by $\theta^2$ and integrate:

$$\frac{1}{\theta(\lambda)} \ge \frac{1}{\theta_0}+\frac{\lambda}{2}.$$

Thus the cross-sectional area must focus by affine parameter at most $2/|\theta_0|$. More precisely, a focal point to $S$ occurs by that value if the null generator continues that far. The compactness of $S$ makes the negative upper bound on both initial expansions uniform.

The affine normalization must be common across the initial surface: for example, choose a smooth future unit timelike field $U$ along $S$ and impose $g(k_\pm,U)=-1$ there. Smoothness and compactness then give a finite uniform focusing bound. Rescaling $k$ by a positive constant rescales $\theta_0$ by that constant and the affine parameter inversely; the bound transforms consistently and does not define an observer-independent numerical distance.

A flat-space check separates a caustic from a singularity. At $t=0$, a round Minkowski sphere of radius $R$ has expansions $2/R$ and $-2/R$ for the null normals $\partial_t\pm\partial_r$. Its future ingoing rays meet at $t=R$, saturating $2/|\theta_0|=R$, and continue through the center as geodesics. The congruence ceases to be a regular family there; spacetime does not. Only one expansion is negative, so this sphere is not trapped.

For Einstein’s equation, contraction with a null vector removes both the scalar-curvature and cosmological-constant terms:

$$R_{\mu\nu}k^\mu k^\nu =8\pi G\,T_{\mu\nu}k^\mu k^\nu.$$

Consequently the null energy condition implies null convergence. The proof needs only the geometric inequality; Einstein’s equation is the bridge from a matter assumption to it.

Penrose’s incompleteness theorem

A standard four-dimensional formulation is the following.

Theorem (Penrose, classical smooth trapped-surface form). Let $(M,g)$ be a smooth, time-oriented Lorentzian spacetime with a noncompact Cauchy hypersurface. Suppose $M$ contains a smooth, compact, future-trapped surface $S$ and the null convergence condition holds. Then $(M,g)$ is future null-geodesically incomplete.

The Cauchy-surface assumption is much stronger than causality at one point: it implies global hyperbolicity and excludes closed causal curves. “Noncompact” is also essential to the particular contradiction. Modern treatments can weaken smoothness, but the causal and curvature assumptions must then be reformulated carefully; the classical statement is documented in Penrose 1965, pp. 57–59, and a regularity-aware proof is reviewed in Steinbauer 2022, §§3.1–3.5 and §§4.4–4.10.

Proof architecture

Assume, for contradiction, that every future null geodesic is complete.

1. The boundary $E^+(S)=J^+(S)\setminus I^+(S)$ is generated by future null geodesics orthogonal to $S$. A generator can remain on this achronal boundary only until its first focal point; beyond it, a timelike variation connects $S$ to a later point.
2. Both null expansions are negative. Raychaudhuri therefore gives a uniform finite upper bound on the affine length of every generator that can stay in $E^+(S)$.
3. Completeness lets one exponentiate the compact set of normalized future null normals over only that bounded parameter interval. Its continuous image is compact, so $E^+(S)$ is compact.
4. Choose a smooth future timelike vector field and flow $E^+(S)$ back to a Cauchy hypersurface $\Sigma$. Achronality makes this map injective, while invariance of domain makes its image open; compactness makes the image closed. The image is therefore a union of connected components of $\Sigma$. Under the usual connected-spacetime formulation it must be all of $\Sigma$, forcing the noncompact Cauchy hypersurface to be compact—a contradiction.

Some presentations package the last step as a covering or projection lemma. The logical work is the same: focusing supplies compactness, and global hyperbolicity turns compactness into a contradiction with the topology of the Cauchy surface.

What the theorem does and does not establish

An incomplete affine parameter means that some inextendible null geodesic has finite affine length in the spacetime under consideration. The theorem does not locate a boundary point, choose a preferred extension, or say why extension fails. Three stronger-sounding statements require additional arguments:

• Curvature divergence. An incomplete spacetime may have bounded scalar invariants, and even a complete list of bounded scalar invariants need not control parallelly propagated tidal components. See curvature blowup criteria for the distinct tests.
• Black-hole formation. A trapped surface is powerful evidence for a black hole in an asymptotically predictable setting, but Penrose’s theorem itself neither assumes nor constructs future null infinity or an event horizon.
• Formation from regular data. The theorem begins after a trapped surface exists. Producing one dynamically from initially untrapped data is the hard estimate developed on the next page.

Compact cosmologies also explain the noncompactness hypothesis. A closed FLRW model can have compact Cauchy surfaces, and compact achronal boundaries do not contradict that topology. This does not refute focusing; it blocks this particular global step.

Check the argument

Suppose one future null expansion on $S$ satisfies $\theta_0\le-4/L$ and null convergence holds. Show that a generator remaining normal and achronal must develop a focal point within affine length $L/2$. Does that calculation alone prove geodesic incompleteness?

Solution

Raychaudhuri gives $1/\theta(\lambda)\ge1/\theta_0+\lambda/2$. Since $1/\theta_0\ge-L/4$ for the least negative permitted initial value, the right-hand side reaches zero no later than $\lambda=L/2$; the congruence must focus by then. This is only a local optical conclusion. Geodesic incompleteness also needs the global causal argument: compact $S$, global hyperbolicity through a Cauchy surface, and the noncompactness contradiction. A complete generator may simply leave the achronal boundary after its focal point.

References

• Penrose, Roger. “Gravitational Collapse and Space-Time Singularities.” Physical Review Letters 14 (1965): 57–59. DOI.
• Steinbauer, Roland. “The Singularity Theorems of General Relativity and Their Low Regularity Extensions.” arXiv:2206.05939v2 (2022). arXiv. Journal DOI.